Understanding Jitter Measurements 2BW-18906-0 - CPC06

Understanding
Jitter Measurement for Serial Digital Video Signals
A Tektronix Video Primer

Primer
Contents
1.0 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3
2.0 Fundamental Concepts and Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4
2.1. Encoding method, unit interval, SDI signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4
2.2. Decoding process, clock recovery, bit scrambling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5
2.3. Time interval error, jitter, jitter waveform, jitter spectrum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5
2.4. Decoding errors, normalized jitter amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6
2.5. Wander, timing jitter, alignment jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6
2.6. Random jitter, deterministic jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7
2.7. Intersymbol interference, equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8
2.8. Pathological signals, SDI checkfield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9
2.9. Decoding decision threshold, AC-coupling effects, symmetric signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10
2.10. Jitter input tolerance, jitter transfer, intrinsic jitter, output jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .11
2.11. Eye diagram, equalized Eye diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .12
2.12. Equivalent-time Eye, Real-time Eye . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14
2.13. Bit error ratio, Bathtub curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .15
3.0 Specifications on Video Jitter Performance and Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .17
3.1. Standards documents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .17
3.2. Specifications on jitter frequency bandpass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .18
3.3. Specifications on signal voltage levels and transition times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
3.4. Specifications on connecting cables and other system elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .19
3.5. Specifications on peak-to-peak jitter amplitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .20
3.6. Specifications on measurement time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .20
3.7. Specifications on data patterns . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21
3.8. Summary of jitter specifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .21
4.0 The Functions Comprising Jitter Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .22
4.1. Equalization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .22
4.2. Transition detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .23
4.3. Phase detection/demodulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25
4.3.1. Phase detection/demodulation: Equivalent-time Eye method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25
4.3.2. Phase detection/demodulation: Phase Demodulation method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .28
4.3.3. Phase detection/demodulation: Real-time Acquisition method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .30
4.3.4. Phase detection/demodulation: Summary of methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .32
4.4. Measurement filters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33
4.4.1. Filter realization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .33
4.4.2. Filter accuracy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .35
4.5. Peak-to-Peak measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36
4.5.1. Peak-to-peak detection methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36
4.5.2. Independent jitter samples and normalized measurement time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .36
4.5.3. Measuring the peak-to-peak amplitude of random jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37
4.5.4. Measurement times . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
4.5.5. Dynamic range and jitter value quantization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
4.6. Jitter noise floor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .40
4.7. Comparing jitter measurement methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .41
5.0 Data Error Rates and Jitter Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .44
5.1. Random jitter and BER . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .44
5.2. Jitter measurement and standards compliance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .45
5.3. BER and jitter measurement time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .46
5.4. Jitter budget . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .47
6.0 Jitter Measurement with Tektronix Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
6.1. Jitter measurement with the Tektronix WFM700M . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
6.2. Jitter measurement with other Tektronix video instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
6.2.1. Wander rejection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49
6.2.2. Measurement of random jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .49
6.2.3. Measurement of deterministic jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
6.3. Jitter measurement with Tektronix real-time oscilloscopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
7.0 Recommendations for Measuring Jitter in SDI Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
7.1. Video system monitoring, maintenance and troubleshooting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
7.2. Video equipment qualification and installation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
7.3. Video equipment design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .52
8.0 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .53
9.0 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54
10.0 Acknowledgement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54
Appendix A: Impact of bandwidth limitation in video jitter measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
Appendix B: Peak-to-Peak and RMS measurement of typical video jitter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .57
Appendix C: Limits to clock recovery bandwidth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .58
2 www.tektronix.com/video
Primer
1.0 Introduction
In this technical guide, we describe the different techniques for measuring jitter in serial digital
video signals and how they can lead to different measurement results. We further identify areas
where the standards should supply additional specifications and guidance to help ensure more
consistent jitter measurements.
This guide focuses on video jitter measurement techniques used to characterize jitter in video systems. It will not cover
typically found in video-specific instruments, e.g., waveform some topics often mentioned in other discussions of jitter
monitors and video measurement sets. General-purpose measurement, e.g., techniques for separating random and
measurement instruments, e.g., sampling and real-time deterministic jitter components.
oscilloscopes, are also used to measure jitter in serial digital The material assumes some understanding of serial digital
video signals. These instruments can offer more extensive transmission theory and practice, the design and implemen-
jitter analysis capabilities based on sophisticated signal tation of signal acquisition systems, the mathematical tech-
processing. niques used in characterizing signal transmission, and the
We will briefly touch on some very basic aspects of video properties of random processes.
jitter measurement using general-purpose instruments in This guide contains the following major sections:
this guide, specifically related to comparing results with
Fundamental Concepts and Terminology: Reviews the
measurements made on video-specific instruments. We
key concepts and terminology we will use to describe
will not explore the range of jitter measurement capabilities
jitter measurement.
available on sampling or real-time oscilloscopes, or on other
general-purpose instruments. Specifications on Video Jitter Performance and
Measurement: Surveys relevant standards and
For the most part, this guide describes jitter measurement
specifications.
methods broadly. It does not give details on specific
implementations in particular instruments. It does describe The Functions Comprising Jitter Measurement:
some aspects of jitter measurement on Tektronix video- Examines the steps involved in measuring peak-to-peak
specific instruments to illustrate some of the key concepts jitter amplitude, the different ways to implement these
discussed in the guide. steps, and the impacts these differences have on
measurement results.
Timing variation in serial digital signals and the measure-
ment of these timing variations are complex technical top- Data Error Rates and Jitter Measurements: Explores
ics. To explain how and why jitter measurements differ, this the relationship between data error rates in video sys-
guide gives a technical overview of jitter measurement tems and the requirements for measuring the jitter per-
techniques and includes technical descriptions of several formance of video equipment used in these systems.
key concepts. Although we examine jitter measurement in Jitter Measurement with Tektronix Instruments:
some detail, we do not comprehensively cover all aspects Describes implementations of jitter measurement
of this topic nor do we explore jitter measurement in methods in Tektronix instruments and explains
extensive technical depth. differences in measurement results.
Rather, this guide focuses on describing common reasons Recommendations for Measuring Jitter in SDI
for differences in measuring jitter in serial digital video signals: Recommends tactics for effectively using jitter
signals. In particular, it examines differences associated measurement methods and tools.
with the jitter frequencies in the video signal and with the
duration of the peak-to-peak amplitude measurements
www.tektronix.com/video 3
Primer
2.0 Fundamental Concepts and

Terminology
In this section we review some fundamental concepts and
terminology needed to describe jitter measurement. This
review will briefly touch on several concepts. It does not
cover these concepts in any depth.
Those experienced in digital communications will be familiar
with many of the concepts reviewed in this section. They
may wish to skip this part of the guide, or scan the material
Figure 1. Unit interval and encoding method for SDI signals.
to review any less familiar terminology or concepts.
2.1 Encoding method, unit interval, SDI signals
The SMPTE standards specify that the clock frequency
Distributing digital video over any significant distance
used to create these SDI signals will equal the signal bit
requires converting the digital content into a serial digital
rate. As a result, SDI signals encode one bit in one clock
video signal. Creating these signals involves converting the
cycle, i.e. the unit interval equals the clock period. So, the
original digital content into a sequence of individual bits and
unit interval of a 270 Mb/s SD-SDI signal equals one period
representing these bits by voltage or light waveforms. A
of a 270 MHz clock or 3.7 ns. Similarly, the unit interval of
clock signal determines the time interval used to encode
a 1.485 Gb/s HD-SDI signal equals 673 ps or one period
a bit in the sequence and an encoding method determines
of a 1.485 GHz clock.1
the signal characteristics that represent a ‘0’ or a ‘1’ bit
value, e.g., Manchester encoding or NRZ encoding. The The SMPTE standards also specify that SDI signals encode
time interval corresponding to one bit in these serial data the serialized data-bit values using the NRZI method (Non-
signals is called the unit interval (UI). return to Zero Inverted). In this method, ‘0’ bit values are
encoded as no change in the signal level, while ‘1’ bit
The Society of Motion Picture and Television Engineers
values are encoded as a change in the current signal level.
(SMPTE) has approved standards that define a serial digital
If the current signal is high, a ‘1’ bit value causes a transi-
interface (SDI) for digital video equipment. SMPTE 259M
tion to the low signal level. If the current signal level is low,
defines the interface for standard-definition (SD) digital video
a ‘1’ bit value causes a transition to the high signal level
formats and SMPTE 292M deals with high-definition (HD)
(Figure 1).
video formats. We will refer to serial digital video signals
conforming to these standards as SDI signals.
The SMPTE standards define serial digital interfaces for sev-
eral different video formats. The information on jitter meas-
urement given in this technical guide applies to SDI signals
conforming to any of these specifications. In this guide, we
will reference two very common types of SDI signals:
The 270 Mb/s signals conforming to SMPTE 259M
specifications for standard-definition, 4:2:2 component
video with either a 4x3 or 16x9 aspect ratio as defined in
ITU-R BT.601- 5 (SD-SDI signals)
The 1.485 Gb/s signals conforming to SMPTE 292M
specifications for various high-definition video formats
(HD-SDI signals)
1 SMPTE 292M also defines an HD format with a data rate of 1.485 GHz/1.001. This SDI signal has a unit interval of 674 ps.
Primer
2.2 Decoding process, clock recovery,

bit scrambling
To extract the digital content from an SDI signal, video
equipment samples the SDI signal at the midpoint of the
time intervals containing data bits (see Figure 1) and
converts these sampled levels to the corresponding bit
values. The sampling process uses a clock with the same
frequency as the encoding clock, and aligned in time to
ensure sampling occurs at the midpoint of the unit interval.
Typically, video equipment does not have direct access to
the clock used to create the serial data signal. Instead,
equipment implements a clock recovery process that uses
a phase-lock loop (PLL) to extract the appropriate sampling Figure 2. SDI signal with sinusoidal edge variation (ideal positions
shown in darker lines).
clock from the received signal. For reliable clock recovery,
the edges in the SDI signal, i.e. transitions between the
As a simple example, suppose the edges in an SDI signal
signal levels, must occur at an adequate rate. Long periods
have a sinusoidal variation around their ideal positions
of a constant signal level can cause the sampling clock to
relative to a reference clock. If we viewed this SDI signal on
drift out of synchronization.
an oscilloscope triggered on the reference clock, the actual
Because of the NRZI encoding, long sequences of ‘1’ bit
edges would appear as a blur around the ideal positions as
values in the serialized data sequence will have edges at
illustrated in Figure 2. We can fully define this simple sinu-
each bit in the sequence. However, serialized digital video
soidal jitter with two parameters, the frequency of the
content can easily contain extended sequences of ‘0’ bit
variation and its peak-to-peak amplitude.
values. This could create SDI signals with long periods at
In actual SDI signals, jitter will rarely have the simple sine
a constant signal level. To avoid this, the SMPTE standards
wave characteristics shown in this example. In real systems,
specify that SDI signal sources will randomize the data
a wide variety of factors influence the timing of signal transi-
before applying the NRZI encoding, using a process
tions. These different sources introduce variations over a
known as scrambling.
range of frequencies and amplitudes. The peak amounts
The scrambling process in an SDI signal source converts
that any particular edge leads or lags its ideal position may
the serialized data bits into a pseudo-random bit sequence.
differ and there may be long time intervals between edges
SDI receivers implement the inverse of this scrambling
with large peak-to-peak variation.
process to extract the original data bit sequence from the
The jitter waveform is the amount of variation in a signal’s
pseudo-random bit sequence. In most cases, this scram-
transitions as a function of time, and the jitter spectrum
bling process ensures a fairly large number of bit transitions,
is the frequency-domain representation of the time-domain
although long sequences of ‘0’ bits can infrequently occur.
jitter waveform. In actual signals, the jitter waveform
2.3. Time interval error, jitter, jitter waveform,
typically has a complex shape created by the combined
jitter spectrum
effects of various sources, and the jitter spectrum contains
Ideally, the time interval between transitions in an SDI signal a wide range of spectral components at different
should equal an integer multiple of the unit interval. In real frequencies and amplitudes.
systems, however, the transitions in an SDI signal can vary
from their ideal locations in time. This variation is called time
interval error (TIE), commonly referred to as jitter. This timing
variation can be induced by a variety of frequency, ampli-
tude, and phase-related effects. In this guide, we will view
jitter as essentially a phase variation in a signal’s transitions,
i.e. a phase modulation of the serial data signal.
Primer
HD-SDI signal a variation of 340 ps is more than one-half

the 673 ps unit interval, while for a 270 Mb/s SD-SDI signal
this same variation is less than one-tenth of this signal’s 3.7
ns unit interval.
In order to describe these timing variations without referring
to specific signal data rates, amplitudes are typically
expressed using unit intervals. In these normalized units,
the variation shown in Figure 3 for clock period n+1 has an
amplitude value of slightly more than 0.5 UI. An amplitude
value of 0.5 UI would equal 1.85 ns in an SD-SDI signal
and 337 ps in an HD-SDI signal.
2.5. Wander, timing jitter, alignment jitter
Figure 3. Decoding error caused by a large amplitude variation
in edge position.
In the preceding examples, we have described variations
in the position of signal transitions with respect to an ideal,
2.4. Decoding errors, normalized jitter jitter-free reference clock, i.e. a clock signal in which all
amplitude edges occur at their ideal locations in time. Actual reference
In the decoding process, SDI receivers use a reference clocks used in decoding are not jitter free.
clock to determine when to sample the input SDI signal. As noted in section 2.2, the decoding process typically
Ideally, the transitions in the input SDI signal occur at uses a recovered clock extracted from the received SDI
appropriate clock edges and sampling occurs at the mid- signal. The clock recovery process “locks” the recovered
point of the unit interval. In the ideal situation shown in clock to the input signal and the clock will follow timing
Figure 1, the signal transitions align with the clock’s falling variations in the input signal that fall within the bandwidth
edges and sampling occurs at the clock’s rising edges. of the recovery process. Hence, the timing variations in
Real SDI signals, however, have some amount of jitter in the SDI signal introduce variations in the transitions of the
their edges. Jitter of sufficiently large amplitude will cause recovered clock.
sampling errors. Figure 3 illustrates this situation. It shows Since the transitions in the recovered clock determine when
an SDI signal that encodes two ‘1’ bit values during two the decoder samples the SDI signal, using a recovered
clock periods, n and n+1. In the ideal situation, the sam- clock actually reduces the number of decoding errors
pling process would capture a high signal value in clock associated with low frequency variations. The sampling
period n and a low signal value in clock period n+1. time “tracks” these variations and samples at the correct
In the actual signal, the transitions vary significantly from location inside the unit interval.
their ideal locations relative to the reference clock. During The recovered clock does not track variations in signal
clock period n, the actual edge varies by less than one-half transitions if the frequency of the variation lies above the
the reference clock period. At the sampling time determined bandwidth of the clock extraction process. At these higher
by the reference clock, the sampling process captures a frequencies, the position of signal transitions can vary
high signal value, as it would in the ideal situation. relative to the edges of the recovered clock and these
During clock period n+1, however, the actual transition variations can create decoding errors.
occurs more than one-half the clock period from its ideal As noted in section 2.3, the jitter spectrum in actual SDI
position relative to the reference clock. Since the actual signals generally contains a range of spectral components.
edge occurs after the sampling time determined by the ref- The recovered clock will generally track spectral compo-
erence clock, the sampling process captures a high signal nents below the clock recovery bandwidth, but will not
value instead of the low signal value it would have sampled track spectral components above this bandwidth. Hence,
in the ideal situation. the impact of jitter on decoding depends on both the jitter’s
When expressed in seconds, the amount of timing variation amplitude and its frequency components. This has led to
needed to generate a decoding error depends on the clock a frequency-based classification of jitter.
period, i.e. the size of the unit interval. For a 1.485 Gb/s
Primer
Conventionally, the term “jitter” refers to short-term time However, reclocking may not significantly reduce wander or
interval error, i.e. spectral components above some low low-frequency timing jitter since the recovered clock tracks
frequency threshold. For SDI signals, the SMPTE standards these variations. Hence, low-frequency variations can build
set this threshold at 10 Hz and refer to spectral compo- through a video system. Amplitudes can eventually grow
nents above this frequency as timing jitter. beyond the tracking capability of clock recovery processes.
The term wander refers to long-term time interval error. For At this point, decoding errors will appear and the clock
SDI signals, components in the jitter spectrum below 10 Hz recovery hardware might not remain locked to the
are classified as wander. Since video equipment can gener- input signal.
ally track these long-term variations, characterizing wander This guide examines techniques for measuring timing and
in terms of actual edge positions relative to their ideal posi- alignment jitter. We will not examine wander and wander
tions does not give meaningful information. Instead, wander measurement techniques. However, wander does impact
is measured in terms of frequency offset and frequency drift jitter measurements since these measurements must
rate. These parameters characterize the deviation from exclude contributions from spectral components below
expected clock rates in normalized units of parts per million 10 Hz. Differences in wander rejection can lead to different
(ppm and ppm/sec) or parts per billion (ppb and ppb/s) measurement results, and we will examine these effects
rather than UI. in later sections.
Alignment jitter refers to components in the jitter spectrum 2.6. Random jitter, deterministic jitter
above a specified frequency threshold related to typical To fully understand the impact of jitter in video systems,
bandwidths of the clock recovery processes. In other we need to consider its statistical properties in addition
words, alignment jitter is a subset of timing jitter that to its amplitude and spectral content. Commonly used
excludes spectral components the clock recovery process approaches to characterizing and modeling these proper-
can track. The specified frequency threshold differs for ties distinguish between two basic jitter types. Random
SD-SDI and HD-SDI signals and is defined in the relevant jitter has essentially no discernable pattern. It is best
SMPTE standard (see section 3.2). For SD-SDI signals, characterized by a probability distribution and statistical
alignment jitter refers to spectral components above 1 kHz. properties like mean and variance. Deterministic jitter is
For HD-SDI signals, spectral components above 100 kHz more predictable (determinable) and is often characterized
are classified as alignment jitter. by some definable periodic or repeatable pattern with a
In general, video equipment does not track alignment jitter, determinable peak-to-peak extent.
though some equipment may track some low frequency Random jitter
alignment jitter. Thus, high amplitude alignment jitter gener-
Random processes, e.g., thermal or shot noise, introduce
ally introduces decoding errors. Since video equipment can
random jitter into an SDI signal. We typically use a Gaussian
track wander and low frequency timing jitter, these spectral
probability distribution to model this jitter behavior, and we
components often have less impact on signal decoding.
can use the standard deviation of this distribution (equiva-
While low frequency variations may have less impact on lent to the RMS value) as a measure of the jitter amplitude.
signal decoding, they can have significant impact in other However, the peak-to-peak jitter amplitude and the RMS
areas. Other processes, e.g., digital-to-analog conversion jitter amplitude are not the same. In particular, the peak-to-
stages, use this recovered clock, or a sub-multiple of this peak amplitude value depends on the observation time.
clock. Since this clock tracks the low frequency jitter in the
In the Gaussian distribution used to model random jitter,
input SDI signal, its edges vary from their ideal positions.
small amplitude variations in edge position are most
This jitter in the clock signal can introduce errors, e.g.,
probable, but very large amplitude variations may infre-
non-linearity in D-to-A conversion.
quently occur. A record of amplitudes made over a short
Clock recovery also affects the way jitter and wander accu- observation time could include a large amplitude value, but
mulate in a video system. Reclocking video equipment uses probably will not. By contrast, a record of amplitudes made
the recovered clock to regenerate the SDI signal. Since over a long observation time might not contain any large
the recovered clock does not track alignment jitter well, amplitude values, but probably will contain at least one. So,
reclocking can substantially reduce alignment jitter. on average, we would expect that a peak-to-peak ampli-
Primer
tude measurement made over a long observation time

would have a larger value than a peak-to-peak amplitude
measurement made over a short observation time.
The “tails” of a Gaussian distribution can reach arbitrarily
large amplitudes. Hence, by observing over a sufficiently
large time interval, we could theoretically measure arbitrarily
large peak-to-peak jitter amplitude. We describe this prop-
erty by saying that random jitter has “unbounded” peak-
to-peak amplitude.
Technically, this description applies only to the mathematical
model for random jitter. For all practical purposes, however,
a Gaussian distribution adequately models random jitter
in real systems. Thus, we can say that over any region of Figure 4. Frequency response for 300 m of cable and typical
response of a compensating equalizer.
interest, random jitter in actual SDI signals has unbounded
peak-to-peak amplitude.
amplitude. Thus, for all practical purposes, deterministic
Deterministic jitter jitter has bounded peak-to-peak amplitude and random
A wide range of sources can introduce deterministic jitter jitter has unbounded peak-to-peak jitter amplitude.
into an SDI signal. For example: 2.7. Intersymbol interference, equalization
Noise in a switching power supply can introduce periodic In real serial digital signals, transitions from one voltage level
deterministic jitter. to another do not occur instantaneously. They have finite
The frequency response of cables or devices can intro- rise and fall times. Further, the frequency-dependent
duce data-dependent jitter that is correlated to the bit response of devices and communication channels will
sequence in the SDI signal (see section 2.7). cause temporal spreading in these transitions. Intersymbol
Differences in the rise and fall times of transition can interference (ISI) occurs when the spreading of transitions
introduce duty-cycle dependent jitter (see section 4.2). in earlier bits affect transitions in later bits.
In addition to these general sources of deterministic jitter, These effects cause transitions to vary from their ideal
SDI signals can contain deterministic jitter correlated with shapes and locations. In other words, ISI introduces jitter
video properties. For example: in the signal. Specifically, it produces predictable and
repeatable jitter whose magnitude depends on the
The line and field structure of video data can introduce
frequency responses of devices and channels, and on
a periodic deterministic jitter that we will call raster-
the data patterns in the signal. Hence, ISI produces
dependent jitter.
deterministic, data-dependent jitter.
Converting the 10-bit words used in digital video to and
In particular, cable attenuation greater than 1 dB can
from a serial bit sequence can introduce high frequency
introduce significant intersymbol interference. To avoid
data-dependent jitter at 1/10 the clock rate, typically
data errors due to this ISI, receivers typically have cable
called word-correlated jitter.
equalizers that compensate for the 1/√ƒ frequency
Deterministic jitter attains some maximum peak-to-peak response of the cable. Figure 4 shows the typical frequency
amplitude within a determinable time interval. Increasing the responses of a cable and equalizer.
observation time beyond this time interval will not increase
the peak-to-peak jitter amplitude measurement. Unlike ran-
dom jitter, repeatable deterministic jitter has a determinable
upper bound on its peak-to-peak jitter amplitude.
Even if deterministic jitter has infrequent long-term deter-
minable behavior, this jitter can be adequately modeled
with a predictable pattern that has bounded peak-to-peak
Primer
Once initiated, the stress patterns in pathological SDI sig-

nals occur only to the end of an active video line. SDI signal
formats insert information between lines of active video
content, e.g., the SAV (start-of-active-video) and EAV
(end-of-active-video) synchronization words. This added
information disrupts the special shift register state and bit
sequences that create these long constant-voltage inter-
vals. Even if the next active line contains the same special
bit sequence, the shift register will generally not have the
appropriate initial state and the SDI signal will not contain
the stress patterns.
Figure 5. SDI signals for stress testing clock recovery and
cable equalization. Repeating the special bit sequence on multiple video lines
will cause the stress pattern to reappear. Eventually, the
2.8. Pathological signals, SDI checkfield shift register in the scrambler will enter an appropriate initial
As noted in section 2.2, clock recovery requires frequent state at the right position in the bit sequence. When this
signal transitions, i.e. the signal must have a sufficient occurs, the stress pattern will reappear and will continue to
transition density. Cable equalization algorithms also need the end of the active video line. The conditions needed to
many edges in the signal to determine and maintain the initiate a stress pattern happen infrequently. So, pathologi-
frequency-dependent gain that compensates for the cal SDI signals consist of occasional occurrences of video
1/√ƒ frequency response of the cable. Long intervals of lines containing stress patterns statistically interspersed
constant signal level stress these processes and can lead among many video lines that look like typical SDI signals.
to decoding errors or synchronization problems. Further, Some video test signal generators can create these signals
AC-coupling can reduce noise margins in decoding if the in either full-field formats, or in split-field formats that com-
input signal remains at the same voltage level for a signifi- bine the clock recovery and equalizer stress patterns. For
cant percentage of time (see section 2.9). testing recovery and equalization in video equipment with
In most cases, scrambling and NZRI encoding ensures that SDI inputs, SMPTE developed two recommended practices
SDI signals have many transitions. Typical SDI signals do (RP 178 and RP 198) that define a specific split-field format
not have long intervals of constant voltage that stress clock called an SDI checkfield.
recovery, equalization, or decoding processes.
However, particular word patterns in digital video content
can produce SDI signals with long constant-voltage inter-
vals. If the shift register used in the scrambling process has
a particular state and the scrambler receives one of several
special input bit sequences, the resulting SDI signal after
NRZI encoding will have one of the patterns shown in
Figure 5. The paper by Takeo Eguchi listed in the
References provides additional information.
SDI signals containing these patterns are called pathological
SDI signals. Video semiconductor and equipment designers
can use these signals to “stress test” clock recovery and
equalization processes and to verify the correct operation
of clamping or DC-restoration circuits that compensate for
AC-coupling effects. As shown, the patterns needed for
testing equalization differs from the pattern needed to stress
test the clock recovery PLL.
Primer
Figure 6a shows a segment of an AC-coupled signal

derived from an input SDI signal that does not contain any
long duration at the same voltage level. In this case, the
high signal level in the AC-coupled signal equals +0.5·Vpp
and the low signal level equals -0.5·Vpp, where Vpp is the
peak-to-peak amplitude of the input SDI signal. The fixed
decision threshold falls at an optimal position midway
between the two levels.
Figure 6b shows a segment of an AC-coupled signal
derived from an input SDI signal that stays at the low signal
Figure 6. Shift in AC-coupled signals relative to fixed decision level.
level for long periods of time, e.g., one of the equalizer
stress patterns described in section 2.8. In this example,
2.9. Decoding decision threshold, AC- the signal remains at the low signal level 95% of the time.
coupling effects, symmetric signals To maintain an average signal level of zero volts, the low
To determine whether a sampled signal voltage corre- signal level in the AC-coupled signal must equal -0.05·Vpp,
while the high signal level must equal +0.95·Vpp. The low
sponds to a “high” or “low” signal level, decoders compare
signal level is very close to the fixed decision threshold for
the sampled voltage against a particular voltage level called
decoding, which eliminates the noise margins for this signal
the decision threshold or decision level. An optimally
level and will lead to decoding errors.
chosen decision threshold will equally protect against errors
generated by noise on either signal level. If each signal level In effect, the AC-coupling has generated intersymbol inter-
has the same amount of noise, the optimal decision thresh- ference. The values of earlier bits (long strings of ‘0’ bit
old equals the average of the two signal voltage levels. values after scrambling) have impacted the decoding of
later bits.
SDI receivers generally use fixed decision thresholds in the
decoding process. For optimal performance, signal levels The amount of shift depends on the coupling time constant.
must keep the same relative relationship to this fixed volt- For example, with a coupling constant of 10 µsec, an
age level. A shift in the signal relative to the decision thresh- equalizer stress pattern will shift almost 78% closer to the
old reduces the noise margin for one of the signal levels, fixed decision level over one-half of an HD video line. With
which can lead to decoding errors. a coupling time constant of 75 µsec, the stress pattern will
shift by less than 33% over an entire HD video line.
SDI receivers typically have AC-coupled inputs that remove
DC-offsets in the input SDI signal and maintain a constant To compensate for this AC-coupling effect, SDI receivers
average voltage in the AC-coupled signal. In many imple- typically clamp or DC-restore the AC-coupled signal to
mentations, this average signal level equals zero volts, maintain the relationship between the signal levels and
although biasing circuitry in the receiver could set the the fixed decision threshold.
average signal level of the AC-coupled signal to a non-zero Due to scrambling and NRZI encoding, SDI signals are
value. Typically, the fixed decision threshold equals the symmetric, i.e. they spend nearly the same amount of time
average voltage of the AC-coupled signal. However, the at each signal level. More specifically, typical SDI signals are
optimal decision threshold may differ from the average symmetric when signal levels are averaged over many unit
signal voltage if one signal level can have more noise intervals. Shorter-term, SDI signals can have several periods
than the other. of constant signal level, with pathological SDI signals as the
While AC-coupling filters out DC offsets in the input SDI extreme case.
signal that could lead to decoding errors, it can also shift
the signal levels in the AC-coupled signal relative to a fixed
decision threshold. Figure 6 illustrates this situation for an
implementation of AC-coupling that maintains the average
signal level in the AC-coupled signal at zero volts. In this
example, the decoding process also uses zero volts as the
fixed decision threshold.
Primer
Even in SDI signals with frequent transitions, AC-coupling Some video equipment, e.g., a distribution amplifier, pro-
can introduce a shift in the signal relative to the fixed deci- duces an SDI output from an SDI signal applied at an input.
sion level. If the rise and fall times of signal transitions differ Typically, jitter in the input SDI signal does not directly trans-
significantly, the signal will spend more time at one of the late to jitter in the corresponding output. In particular, clock
signal levels. For example, if the signal has fast rise times recovery can filter out high frequency jitter, or may amplify
and slow fall times, it will spend more time in the high signal some jitter in the input signal. Jitter transfer is the jitter on
state. AC-coupling will then shift the high signal level closer an SDI output resulting from jitter in an input SDI signal,
to the fixed decision threshold, reducing noise margin. and the jitter transfer function is the ratio of output jitter to
Typically, SDI signals have symmetric rise and fall times, but applied input jitter as a function of frequency.
asymmetric line drivers and optical signal sources (lasers) Like receivers, source and regeneration equipment also has
can introduce non-symmetric transitions. While significant, internal jitter. This internal jitter will appear on an SDI output
these source asymmetries do not have especially large signal even if the associated SDI input has no jitter. Intrinsic
impacts on signal rise and fall times. In particular, cable jitter is the amount of jitter at an SDI output in the absence
attenuation will generally have a much larger impact on of input jitter. Output jitter is the total amount of jitter at an
signal transition times. SDI output resulting from intrinsic jitter and the transfer of
Without appropriate compensation or other adjustments, jitter in any associated SDI input.
asymmetries in SDI signals can reduce noise margins with
respect to the decision threshold used in decoding and
can lead to decoding errors. As we explore in section 4.2,
these same asymmetric conditions can also impact jitter
measurements.
2.10. Jitter input tolerance, jitter transfer,
intrinsic jitter, output jitter
SDI signal receivers can differ in their implementations of
the processes described in the preceding sections. A par-
ticular receiver’s clock recovery process may not track jitter
as well as others, or it may not sample the SDI signal near
the midpoint of the unit interval. The design and hardware a
receiver uses to implement equalization, clock recovery and
decoding processes may introduce a significant amount of
additional jitter into the signal. Thus, a particular receiver
may have multiple data errors when decoding SDI signals
that other receivers can decode without error. Such a
receiver has a lower jitter input tolerance.
A receiver’s jitter input tolerance depends on the jitter fre-
quencies in the SDI signal. As noted in section 2.5, clock
recovery can track low frequency jitter, so receivers typically
have a higher tolerance for low frequency jitter. The jitter
input tolerance drops significantly for jitter frequencies
above the clock recovery bandwidth.
Primer
Figure 7. Eye diagram for signal with very small amplitude jitter. Figure 8. Nearly closed Eye caused by large amplitude jitter.
2.11. Eye diagram, equalized Eye diagram For signals with a small amount of jitter, the edges in the
Engineers commonly use Eye diagrams to analyze serial aligned segments occur in nearly the same location. The
data signals and diagnose problems. Measurement instru- small variations in edge position create only a slight “blur”
ments create Eye diagrams by superimposing short around the nominal edge positions (see Figure 7). Most
segments of the serial data signal. The finite rise and fall of the space between the crossover points is free of
times of these transitions create the characteristic ‘X’ transitions. In this situation, we say the Eye is “open.”
patterns in the Eye diagram (see Figure 7). The eye-shaped As the amplitude of the jitter increases, more transitions
area without transitions gives the display its name. We will move into the open space between crossover points, i.e.
call the point where the rising and falling edge transitions the Eye starts to “close” (see Figure 8).
intersect the crossover points in the Eye diagram.
Using Eye diagrams, engineers can quickly form a qualita-
The time interval between the crossover points in the Eye tive impression of the jitter in a signal and the potential
equals the unit interval. In the ideal case, the decoding for decoding errors. Overall, a signal that forms a large,
process samples the signal at the mid-point between the wide-open Eye is less likely to produce decoding errors
crossover points and the decision threshold corresponds than a signal that forms a small or closed Eye. However, in
to the widest part of the Eye opening (Figure 7). making this qualitative assessment, one of the key factors
To make the Eye diagram, the instrument aligns the seg- engineers need to consider is the difference between the
ments using a reference clock signal. Typically this reference Eye clock recovery bandwidth and the bandwidth of the
clock is extracted from the data signal, but may be a sepa- receiver’s clock recovery process.
rate reference clock signal. It can be externally supplied, If the clock recovery bandwidth in the receiver equals the
e.g., through the trigger input on an oscilloscope, or Eye clock recovery bandwidth, the size of the Eye opening
extracted within the measurement instrument. correlates reasonably well with the potential for decoding
If the transitions in the input signal align with the edges in errors. If the input signal forms a large, “wide-open” Eye,
this reference clock they will lie on top of each other in the the decoding process will most likely sample the signal
Eye diagram. Any transitions that vary from the nominal before the transition to the next bit.
positions determined by this reference clock will appear in If the clock recovery bandwidth in the receiver is less than
different locations. If the instrument uses a recovered clock the Eye clock recovery bandwidth, the signal may contain
to form the Eye diagram, the reference clock will track jitter jitter frequencies below the Eye clock recovery bandwidth
below the loop bandwidth of this clock recovery process. that impact the decoding process but do not appear in the
Thus, the Eye diagram will only show jitter components with Eye diagram. The decoding process may generate errors
frequencies above this bandwidth threshold, called the Eye even though the Eye diagram has a large Eye opening.
clock recovery bandwidth.
Primer
While equalization can compensate for cable effects, the

equalized signal can still contain signal jitter or amplitude
noise that reduces or closes the Eye opening. To qualita-
tively assess the remaining potential for decoding errors
after equalization, engineers can use an equalized Eye
diagram constructed from the equalized version of the
input signal.
Eye diagrams can also show AC-coupling effects. Signal
level shifts due to AC-coupling (section 2.9) causes a
Figure 9. Eye diagram for SD-SDI signal showing a nearly closed
Eye due to attenuation in a 100 m cable. corresponding shift in the superimposed segments that
. form the Eye diagram. This can occur even if the measure-
On the other hand, if the clock recovery bandwidth in the ment instrument forming the Eye-diagram has a DC-
receiver is greater than the Eye clock recovery bandwidth, coupled input. Other equipment in the system may have
the Eye diagram may show jitter that does not impact the AC-coupled inputs, causing shifts in the SDI signal before
decoding process. The receiver may decode the signal it reaches the measurement instrument.
without errors even though the Eye diagram has a small
Eye opening or is completely closed.
Other factors also influence the qualitative assessment
of signal jitter using Eye diagrams. If receivers introduce
a significant amount of internal jitter or do not consistently
sample near the middle of the unit interval, they may
generate more decoding errors than suggested by the
size of the Eye opening.
Thus, in using an Eye diagram to assess the potential for
data errors, engineers need to consider the combined
effects of the receiver’s clock recovery, equalization and
decoding processes. In other words, they need to consider
the receiver’s jitter input tolerance (see section 2.10). A
receiver with low jitter input tolerance can generate errors
in decoding a signal that forms a wide-open Eye diagram,
while a receiver with high jitter input tolerance may correctly
decode a signal that forms a closed Eye diagram.
As noted in section 2.7, frequency-dependent cable attenu-
ation “spreads” transitions in SDI signals. This intersymbol
interference can significantly reduce or completely close the
Eye opening in an Eye diagram constructed from a signal at
the end of a long cable (see Figure 9).
However, a small or closed Eye opening in the Eye diagram
of a non-equalized signal at the end of a long cable does
not necessarily indicate a high potential for decoding errors.
The cable equalization used in receivers will restore the
signal’s transitions and “re-open” the Eye. With adequate
equalization, the ISI from cable attenuation will not signifi-
cantly impact the decoding process. Without adequate
equalization, the data-dependent jitter introduced by cable
effects can lead to decoding errors.
Primer
Figure 10. AC-coupling effects.
Figure 10a shows a pathological SDI signal containing diagram represent the composite effect of many separated
an equalizer stress pattern in an Eye display set to a sweep edges in the actual signal, possibly widely-separated edges.
rate equal to several video lines. At this slow sweep rate, The sampling rate used to construct the Eye can strongly
the resulting waveform contains thousands of individual influence the results of peak-to-peak jitter amplitude
Eyes per division. This display clearly shows that the signal measurements (section 4.3.1).
levels shift higher because of AC-coupling effects due Real-time digital oscilloscopes can construct an Equivalent-
to the long intervals at a low signal level (top pattern in time Eye diagram using the equivalent-time technique
Figure 5b). mentioned above. They can also construct Eye diagrams
Figure 10b shows an Eye display for the same signal using using real-time sampling techniques that over-sample the
a much lower sweep rate that displays a full video field. This input signal. These instruments use software-based clock
display demonstrates the effects from the two different recovery techniques in creating the Eye.
equalizer stress patterns shown in Figure 5b. We will refer to an Eye diagram constructed using this
2.12. Equivalent-time Eye, Real-time Eye real-time sampling technique as a Real-time Eye. In this
The instruments most commonly used to monitor and technique, the edges in the Eye diagram are actual edges
measure signal jitter construct Eye diagrams by sampling in the input signal. The amount of acquisition storage and
the input signal. They acquire these samples using two the sampling rate will influence the results of peak-to-peak
different methods. jitter amplitude measurements (section 4.3.3).
Many instruments, including waveform monitors and other

video-specific measurement instruments, use equivalent-
time sampling techniques to create the Eye diagram. These
techniques use under-sampling to approximate an over-
sampled acquisition. We will refer to an Eye diagram
constructed in this manner as an Equivalent-time Eye.
Measurement instruments use different equivalent-time
sampling techniques with a wide range of capabilities and
characteristics. Broadly speaking, because of the under-
sampling used in this approach, the “edges” in the Eye
Primer
Figure 11. Bathtub curve – BER vs. location in Eye Diagram.
2.13. Bit error ratio, Bathtub curve In random jitter, smaller amplitude variations happen more
All SDI signals contain some amount of random jitter. frequently than larger amplitude variations. Thus, as the
As noted in section 2.6, random jitter has no discernible sampling location moves towards a crossover point,
pattern. Thus, decoding errors due to random jitter in a random jitter can more frequently shift transitions into
signal will not occur at determinable times or rates. In place the wrong positions relative to the sampling location. This
of error rates, the combined impact of deterministic and leads to an increased number of decoding errors and an
random jitter on decoding can be usefully characterized by increased BER.
a bit error ratio (BER), the ratio of the number of incorrectly The sketch in Figure 11b shows this relationship between
decoded bits to the total number of bits decoded. the BER due to jitter in a signal and the sampling location
For example, consider an HD-SDI signal with a small in the unit interval. This is called a Bathtub plot or a Bathtub
amount of random jitter and a receiver that always samples curve because the shape looks like a cross-section of
at the midpoint of the unit interval. Suppose that the total a bathtub.
jitter in this signal, i.e. the combined effects of deterministic
and random jitter, causes sampling errors in this receiver
at an average rate of 1 per minute. In one minute, the
1.485 Gb/s HD-SDI signal transmits 8.91 x 1010 bits. So,
the total jitter in the signal corresponds to a BER of at least
1.12 x 10-11 in this ideal receiver. For a 270 Mb/s SD-SDI
signal, an average of one decoding error per minute
corresponds to a BER of at least 6.17 x 10-11. Due to
error propagation effects in the SDI receiver associated
with bit scrambling and NRZI to NRZ conversion, one
sampling error can lead to multiple bit errors, producing
a higher BER.
Now imagine moving the sampling location away from
the midpoint of the unit interval and towards one of the
crossover points in the Eye diagram. Figure 11a illustrates
this process with a sketch of an Eye diagram that has
accumulated edges long enough for large amplitude ran-
dom jitter to nearly close the Eye. As the sampling location
moves closer to a crossover point, smaller jitter amplitudes
can now cause transitions to occur in the wrong position
relative to the sample location.
Primer
Figure 12. Bathtub curves at the receiver input for SDI signals from two sources.
Bathtub curves are useful in assessing whether a video signal anywhere inside a 0.5 UI region centered in the
system can achieve a target BER. For example, suppose unit interval.
an operation wants the BER in a video system to stay As described in section 2.5, receivers track signal jitter at
below 10-10. Consider two different sources within the frequencies below the bandwidth of their clock recovery
system whose output SDI signals contain different amounts process and can adjust the sampling location to compen-
of random and deterministic jitter. At a particular receiver’s sate for this variation. However, clock recovery cannot track
input, suppose the total jitter in the two signals has the these variations perfectly.
same RMS amplitude and generates the Bathtub curves
Suppose that timing errors in the clock recovery process
shown in Figure 12.
could cause the sampling location of the receiver used in
The shapes of the Bathtub curves offer insight into the this example to fall anywhere within a 0.4 UI region cen-
signal jitter. The steeper curve on the signal from Source A tered in the unit interval. Then, signals from source B will
indicates a lower amount of random jitter compared to the most likely not meet the 10-10 BER requirement due to
signal from source B. As the number of bits observed the larger random jitter component in these signals.
increases, the peak-to-peak jitter amplitude increases less
Signals from Source A can more easily meet this BER
in the signal from source A than in the signal from Source
requirement. Except for 1 transition in 1010 bits, the Eye
B. Since the total jitter in the signals have equal RMS jitter
opening in the SDI signals from Source A is greater than
amplitudes, the ratio of deterministic jitter to random jitter
the potential variation in the receiver’s sampling location.
is greater in the signal from Source A compared to the
This includes some margin to allow for small, occasional
signal from Source B.
increases in internal jitter or variation in sampling location.
For a BER of 10-10, the sides of the Bathtub curve for the
The primer “Understanding and Characterizing Timing
SDI signal from Source A define a 0.5 UI region centered in
Jitter” listed in the References contains additional informa-
the unit interval. Presuming that any signal transition in this
tion about Bathtub plots and the impact of random and
region causes a decoding error, we can say that the Eye
deterministic jitter.
opening for this signal equals 0.5 UI except for 1 transition
in 1010 bits. By contrast, the Eye opening for the SDI signal
from Source B equals 0.33 UI except for 1 transition in
1010 bits.
To meet the 10-10 BER target, the receiver must sample the
SDI signal from Source B inside a 0.33 UI region around the
midpoint of the unit interval. The receiver has greater margin
in sampling the signal from Source A. It can sample this
Primer
3.0 Specifications on Video Jitter Performance and Measurement
Document Title Content description

SMPTE RP 184 Specification of Jitter in Bit-Serial Methods and performance templates for specifying
Digital Systems jitter input tolerance, jitter transfer, and output jitter.
SMPTE RP 192 Jitter Measurement Procedures in Methods for carrying out the jitter performance
Bit-Serial Digital Interfaces measurements identified in RP 184.
SMPTE 259M 10-Bit 4:2:2 Component and 4ƒsc Specifications on performance limits for jitter at
Composite Digital Signals—Serial the SDI outputs of SD-SDI signal sources.
Digital Interface
SMPTE 292M Bit-Serial Digital Interface for High- Specifications on performance limits for jitter at
Definition Television Systems the SDI outputs of HD-SDI signal sources.
SMPTE EG 33 Jitter Characteristics and Guidance on jitter measurement and minimizing
Measurements jitter in video systems.
IEEE Std 1521 IEEE Trial-Use Standard for Specifications for output jitter and wander only,
Measurement of Video Jitter and including performance templates, and methods
Wander for measuring jitter, including jitter measurement
frequency response.
Table 1. Standards and other documents that apply to video jitter.
Consistency in jitter measurements necessarily starts with RP 184 gives the framework for specifying jitter perform-
the standards. The industry develops and adopts these ance, including jitter input tolerance, jitter transfer, and
standards to ensure that equipment will perform satis- output jitter. This includes methods for specifying the jitter
factorily in video production, distribution, and transmission frequencies included in peak-to-peak amplitude measure-
systems. Video equipment manufacturers must design ments. This recommended practice only describes the form
and deliver products that meet these standards. Video test of jitter specifications. All parameters are in symbols without
equipment manufacturers must fully understand the stan- specific performance limits.
dards, implement test procedures that conform to the In particular, RP 184 does not give values for measurement
requirements documented in these standards, and make bandpass cutoff frequencies or peak-to-peak jitter limits.
these implementations as accurate as possible within the These measurement parameters depend on the particular
constraints of their specific implementations. SDI format and are listed in the standard defining the
However, implementing test procedures in conformance format. Also, RP 184 defers specification on measurement
to the relevant video standards does not ensure consistent time to other standards or recommended practices.
measurements. In particular, the current video standards RP 192 gives examples of jitter measurement techniques
allow for significantly different jitter measurement methods that conform to RP 184 and describes these particular
that can yield noticeably different results. Hence, any techniques in detail. However, RP 192 does not preclude
discussion of jitter measurement and variability in measure- other techniques that conform to RP 184. This recom-
ment results must begin by looking at the relevant stan- mended practice does not specify particular measurement
dards and specifications. times, but does describe a procedure for determining the
3.1. Standards documents minimum measurement time for oscilloscope-based jitter
SMPTE publishes standards, recommended practices (RP), measurement.
and engineering guidelines (EG) for the video industry. The SMPTE 259M, section 3.5, deals with jitter in SDI signals
Institute of Electrical and Electronics Engineers (IEEE) also carrying standard-definition digital video content. SMPTE
publishes video standards. Table 1 lists the standards and 292M, section 8.1.8, deals with jitter in SDI signals carrying
recommended practices that apply to video jitter and briefly high-definition digital video content. For their respective
describes their jitter-related content.2 formats, these standards specify the performance limits
2 The ITU also publishes video standards containing specifications on jitter performance, e.g., ITU-R BT.656, ITU R-BT.799, and ITU-R BT. 1363. In Japan, the ARIB standards
contain specifications in this area. To a significant extent, the guidelines and specifications in these documents agree with those in the SMPTE and IEEE documents described
in this guide.
Primer
on jitter from “the serial output of a source derived from a

parallel domain signal whose timing and other characteris-
tics meet good studio practices.”
Hence, these standards define performance limits only on
output jitter. In particular, they assign specific values for the
parameters identified in RP 184 for measuring output jitter.
These include the measurement bandpass corner frequen-
cies, peak-to-peak jitter limits, and the test signal to use
in making jitter measurements. These standards do not Figure 13. Frequency bandpass for measuring timing jitter in an
HD-SDI signal.
specify a peak-to-peak amplitude measurement time.
EG 33 gives engineers more detailed information on jitter in RP 184 recommends at least a 20 dB/decade slope on the
SDI signals and guidance on jitter measurement techniques. highpass filtering at ƒ1. RP 192 recommends at least a 40
It describes some of the impacts jitter can have on system dB/decade slope, and IEEE Std. 1521 recommends at least
operations and suggests design approaches that minimize a 60 dB/decade slope. The IEEE standard requires the
or mitigate these impacts. steeper 60 dB/decade slope to separate jitter from wander
IEEE Standard 1521 describes requirements for specifying (See Figure 1 in IEEE Std. 1521). To conform to the IEEE
the measurement of jitter and wander for both analog and standard and both recommended practices, jitter measure-
digital video. As with RP 184, it gives only the form of the ment equipment must use at least a 60 dB/decade high-
specification. It does not give values for measurement filter pass ramp, as shown in Figure 13.3
corner frequencies or peak-to-peak jitter limits. It also RP 184 recommends at least a –20 dB/decade slope on
describes three methods for making jitter and wander the lowpass filtering at ƒ4. It also recommends an in-band
measurements. ripple less than ±1 dB, but does not give any guidance on
In this technical guide, we consider only the measurement the accuracy of the highpass corner frequency, ƒ1. The
of output jitter. The video standards specify performance lowpass corner frequency, ƒ4, can have any value above
limits only on this type of jitter. These are the most com- 1/10 the clock rate.
monly performed measurements, and they have generated To determine the amount of alignment jitter we also need to
the greatest confusion. measure over a range of frequencies, but one with different
3.2. Specifications on jitter frequency highpass corner frequency and slope (Figure 14).
bandpass SMPTE 292M specifies the low-frequency cutoff for meas-
As described in section 2.5, video jitter is classified based uring alignment jitter as ƒ3 = 100 kHz. In agreement with
on frequency. To measure the amount of jitter in these dif- the timing jitter specification, it requires that the high-fre-
ferent classes, measurements must be restricted to specific quency cutoff, ƒ4, shall be at least 1/10 the clock rate.
frequency ranges. RP184, RP192, and IEEE Std. 1521 all Specifications of the highpass corner frequency in the
contribute specifications on bandpass shapes. The relevant bandpass for measuring alignment jitter reflect expectations
SDI specification gives the bandpass corner frequencies. about the bandwidths of clock recovery processes that
As an example, Figure 13 shows the bandpass for measur- track low frequency jitter.
ing timing jitter in an HD-SDI signal. The values shown
in the figure combine the specifications from all relevant
standards and recommended practices.
SMPTE 292M specifies the low-frequency cutoff at ƒ1 = 10
Hz, consistent with the definition of timing jitter. It specifies
that the high-frequency cutoff, ƒ4, shall be > 1/10 the clock
rate, which equals 148.5 MHz for HD-SDI signals.
3 Based on discussions currently underway, the recommendation in RP 184 and RP 192 for the highpass slope will likely change to at least 60 dB/decade.
Primer
3.4. Specifications on connecting cables and

other system elements
For SD-SDI signals, SMPTE 259M specifies that measure-
ments of source output signal characteristics shall be
made across a resistive load connected by a “short coaxial
cable.” For HD-SDI signals, SMPTE 292M specifies a “1-m
coaxial cable.” Hence, for both SD- and HD-SDI signals,
the standards only specify jitter performance near the
Figure 14. Frequency bandpass for measuring alignment jitter in an source output as measured over a short cable length.
HD-SDI signal.
For SDI signal receivers, the standards place some require-
ments on the SDI inputs, including impedance and return
SMPTE selected the value for ƒ3 shown in Figure 14 with
loss. They do not, however, define any performance limits
the expectation that equipment handling HD-SDI signals
on the jitter input tolerance of an SDI receiver. Also, the
will have clock recovery bandwidths of at least 100 kHz.
standards do not define performance limits on jitter transfer
Equipment handling SD-SDI signals may have smaller clock
in system elements.
recovery bandwidths, especially legacy equipment. Hence,
SMPTE 259M specifies that ƒ3 = 1 kHz in the bandpass for The standards do not specify particular cable types, but
measuring alignment jitter in SD-SDI signals. both do require that coaxial connections have the 1/√ƒ
frequency response needed for the correct operation of
RP 184 recommends at least a 20 dB/decade slope on the
cable equalizers. For HD-SDI signals, SMPTE 292M goes
highpass filtering at ƒ3, while RP 192 recommends at least
somewhat further and specifies the cable return loss.
a 40 dB/decade slope. To conform to both recommended
practices, jitter measurement equipment must use at least Neither standard places performance limits on the data-
a 40 dB/decade high-pass slope, as shown in Figure 14.4 dependent jitter introduced by ISI on long cables. They do
RP 184 recommends at least a –20 dB/decade slope in the say the receivers should nominally operate with cable loss-
low-pass filtering at ƒ4. As with timing jitter, RP 184 recom- es up to 20 dB at one-half the clock frequency. This is not a
mends an in-band ripple less than ±1 dB and does not performance limit, however, as they also say that “receivers
give any guidance on the accuracy of the highpass corner that operate with greater or lesser signal attenuation are
frequency, ƒ3. The lowpass corner frequency in the band- acceptable.” The standards do not specify performance
pass for measuring alignment jitter, ƒ4, can have any value characteristics on other sources of ISI in video systems,
greater than 1/10 the clock rate. including reflections on connectors in patch panels.
3.3. Specifications on signal voltage levels

and transition times
For SD-SDI outputs, SMPTE 259M specifies a peak-to-
peak signal amplitude of 800 mV ± 10% with a DC offset
equal to 0.0 V ± 0.5V. The transition between voltage levels
can take no less than 0.4 ns and no more than 1.5 ns, and
the rise and fall times cannot differ by more than 0.5 ns.
For HD-SDI outputs, SMPTE 292M specifies the same
signal amplitude conditions. The transition between voltage
levels can take no more than 270 ps, and the rise and fall
times cannot differ by more than 100 ps.
Hence, the SMPTE standards allow asymmetric rise and fall
times in SDI signals and significant DC offsets. As noted in
section 2.9, these signal characteristics can impact decod-
ing. They can also impact jitter measurement, as we
describe in section 4.2.
4 Based on discussions currently underway, the recommendation in RP 184 for the high-pass slope will likely change to at least 40 dB/decade.
Primer
3.5. Specifications on peak-to-peak jitter

amplitude
SMTPE 292M specifies that the timing jitter in the HD-SDI
output of a source derived from a parallel domain signal will
have peak-to-peak amplitude less than 1.0 UI (673 ps). It
also specifies that the alignment jitter in the SDI output will
have peak-to-peak amplitude less than 0.2 UI, which equals
135 ps (0.2 x 673 ps).
SMTPE 259M specifies that both the timing and the align-
ment jitter in the SD-SDI output of a source derived from a
parallel domain signal will have peak-to-peak amplitude less
than 0.2 UI, which equals 740 ps (0.2 x 3.7 ns).
Note that these two standards only specify the maximum Figure 15. Peak-to-peak measurement value depends on
measurement time.
peak-to-peak amplitude of output jitter allowed in the SDI
signal at the output of a source that derives this signal from
the peak-to-peak measurement increases the probability
a parallel domain input. They do not specify the maximum
that some larger amplitude variations will occur during the
peak-to-peak amplitude of output jitter allowed in SDI sig-
measurement period, which increases the measured peak-
nals at the output of devices that derive the output signal
to-peak jitter amplitude. We examine these effects in more
directly from an SDI input.
detail in section 4.5.3.
3.6. Specifications on measurement time
As noted in section 3.1, the standards offer very limited
The measured peak-to-peak jitter amplitude depends on guidance on peak-to-peak measurement time. Thus, differ-
the time interval used to make the measurement. Section ent manufacturers of video jitter measurement equipment
2.6 describes this dependency for random jitter. It also can, and do, measure the peak-to-peak jitter amplitude
applies to peak-to-peak jitter amplitude measurements over different time intervals. Variations in measurement time
made on signals containing deterministic jitter. Figure 15 typically lead to discrepancies in the measured values.
gives a simple illustration of this dependency. To enable greater consistency in measuring peak-to-peak
In this example, the SDI signal contains periodic, determin- jitter amplitude, the standards will need to specify
istic alignment jitter that consists of well-separated pulses. measurement times.
One advances transitions from their ideal positions; the
other delays transitions. An instrument that makes the
peak-to-peak measurement over a 50 ms observation
window will only measure a single jitter peak and will
indicate that the signal has 0.15 UI of alignment jitter
peak-to-peak. This amount of jitter is within the specified
performance limit. However, an instrument that makes the
peak-to-peak measurement over 150 ms will detect both
the advance and delay peaks. This instrument will indicate
that the signal has 0.3 UI of alignment jitter peak-to-peak,
above the specified performance limits.
While SDI signals can have deterministic jitter behavior of
the kind shown in Figure 15, it is not a typical pattern.
However, all SDI signals have some amount of random jitter.
As noted in section 2.6, random jitter can be modeled by a
Gaussian probability distribution of jitter amplitudes and, for
all practical purposes, does not have an upper bound on
peak-to-peak jitter amplitude. Extending the time interval for
Primer
Specification Signal format

SD-SDI HD-SDI
Highpass Corner frequency 10 Hz 10 Hz

Bandpass for characteristics Slope At least 60 db/decade At least 60 db/decade
measuring
timing jitter Lowpass Corner frequency > 1/10 clock rate > 1/10 clock rate
characteristics Slope At least –20 db/decade At least –20 db/decade
Bandpass ripple ± 1 dB ± 1 dB
Highpass Corner frequency 1 kHz 100 kHz

Bandpass for characteristics Slope At least 40 db/decade At least 40 db/decade
measuring
alignment jitter Lowpass Corner frequency > 1/10 clock rate > 1/10 clock rate
characteristics Slope At least –20 db/decade At least –20 db/decade
Bandpass ripple ± 1 dB ± 1 dB
Peak-to-peak signal amplitude 800 mV ± 10 % 800 mV ± 10 %
DC offset 0.0V ± 0.5 V 0.0V ± 0.5 V
Voltage and Maximum transition time 0.4 ns Not specified
transition times Minimum transition time 1.5 ns 270 ps
Maximum difference between
0.5 ns 100 ps
rise and fall times
Peak-to-peak Timing jitter 0.2 UI 1.0 UI
jitter amplitude Alignment jitter 0.2 UI 0.2 UI
Recommended test signal Color bars Color bars
Table 2. Summary of jitter specifications.
3.7. Specifications on data patterns While pathological signals are quite valuable in stress test-
RP184 also recommends that jitter specifications identify ing, tests to verify that a signal source conforms to SMPTE
the test signal used for the jitter measurement. Both jitter specifications should not use these signals. Methods
SMPTE 259M and SMPTE 292M specify color bars as that use an external reference rather than clock extraction
a non-stressing test signal for jitter measurement. They could successfully measure jitter on pathological signals.
caution that using a stressing signal with a long run of However, as noted in SMPTE RP 192, these methods only
zeros can give misleading results. give a “coarse survey of jitter in an SDI signal.” The meas-
urement result “depends on the stability of the reference
In particular, the SDI checkfield defined in SMPTE RP 198
signal” and “does not allow bandwidth restriction as gener-
will generate pathological signals for stress testing hard-
ally required in jitter specification.”
ware-based equalization and clock recovery processes that
contain long intervals of constant signal voltage. Suppose 3.8. Summary of jitter specifications
that the method used to measure jitter on a source output Table 2 summarizes the specifications relevant to measuring
includes such clock extraction or equalization processes. and characterizing SDI signal jitter.
While these processes always contribute some small level
of internal jitter, the pathological signals can increase this
internal jitter significantly, which can increase the measured
peak-to-peak amplitude value relative to SDI signals with
more typical characteristics.
Primer
4.0 The Functions Comprising Jitter Measurement
Figure 16. The functions involved in jitter measurement (Causes of measurement discrepancies are indicated in parentheses).
To understand how different instruments can yield different To address this issue, some waveform monitors and other
jitter measurements, we have to enumerate the functions video measurement instruments include an equalization
involved in jitter measurement, i.e. all the places where step in the jitter measurement process as shown in Figure
differences can occur. Figure 16 shows a generic block 16. The peak-to-peak jitter amplitude measured after equal-
diagram of a jitter measurement process. ization will better reflect jitter components that can affect
This diagram is a reasonable representation of hardware- the receiver’s performance. The measurement will not
based jitter measurement processes implemented in many include jitter components that the receiver’s equalizer
video-specific measurement instruments. However, it will remove.
does not fully represent any specific method. Also, this Differences in the cable equalizer used in the equalization
diagram does not correspond to the jitter measurement stage can introduce differences in jitter measurement
processes used in general-purpose oscilloscopes, results. Specifically, equalizers differ in their ability to com-
especially software-based approaches. Primarily, we pensate for cable-related ISI. When measuring jitter in SDI
use it as a convenient organizing structure for describing signals at the end of long cables, data-dependent jitter due
the jitter measurement process. to imperfect equalization can increase the peak-to-peak
4.1. Equalization jitter amplitude measurement compared to a measurement
made with a better equalizer. While noticeable, the differ-
As described in section 3.4, the SMPTE standards specify
ences in peak-to-peak jitter amplitude measurements due
that measurements of a source output should be made
to imperfect equalization are much smaller than the differ-
over short cable lengths. At these lengths, cable attenuation
ences between equalized and non-equalized measurements.
will not impact the jitter measurement.
When measuring jitter in an SDI source output over a short
However, this is not the case when making measurements
cable, an equalization stage can affect peak-to-peak ampli-
to diagnose jitter-related problems in a video system. In this
tude measurements. Noise in the equalization process can
case, engineers are typically measuring jitter in SDI signals
add jitter to the SDI signal that can increase the measure-
at the end of a long cable. Due to frequency-dependent
ment result. Typically, equalization noise adds only a small
cable attenuation, these SDI signals will have intersymbol
amount of jitter, though it can make a noticeable contribu-
interference that will appear as data-dependent jitter if the
tion to the jitter noise floor (see section 4.6).
signal is not equalized.
Equalization can also affect jitter measurements when the
If the method used to measure the jitter in this SDI signal
SDI signal has long intervals at constant voltage, e.g.,
does not have an equalization stage, the measured peak-
pathological signals. As noted in section 2.8, these signal
to-peak value will include the jitter introduced by ISI due to
characteristics stress the equalization process. In this case,
cable length. However, because SDI receivers have cable
equalization-related effects can impact peak-to-peak jitter
equalizers (see section 2.7), this jitter will typically not result
amplitude measurements made over cables of any length.
in decoding errors. In effect, the non-equalized measure-
ment result will include a jitter component that does not
lead to decoding errors.
Primer
Figure 17. Eye diagram of signal with equal rise and fall times Figure 18. Non-optimal decision threshold introduces duty cycle
showing optimum decision threshold for transition dependent jitter.
detection at the 50% point.
Because the standards focus on jitter performance at the signal whose transitions have equal rise and fall times. A
serial output of an SDI source, they do not give any guid- transition detector using the optimal decision threshold
ance on the use of equalization or any specifications for located midway between the signal levels (blue line) would
equalization methods. While jitter measurement has a key locate both rising and falling edges at the Eye crossover
role in evaluating video source equipment, it has an equally points. The time between any rising edge detection and the
critical role in deploying and maintaining highly reliable detection of the subsequent falling edge will always equal
production, distribution and broadcast systems. Hence, some multiple of the unit interval. The time between any
the standards will need to address equalization in jitter falling edge detection and the detection of the subsequent
measurement methods to ensure measurement accuracy rising edge will also equal a multiple of the unit interval.
and consistency in this application. Now consider a transition detector using a non-optimal
4.2. Transition detection decision threshold located closer to the high signal level
As noted in section 2.3, jitter is the variation of signal transi- (red line). When the signal had a rising edge followed by a
tions from their ideal positions in time. To measure these falling edge, this transition detector would locate the rising
variations, the measurement process needs to determine edge after the crossover point and the corresponding falling
the point in time when an actual signal transition occurs. edge before the crossover point. The time between these
Like signal decoding, it uses a decision threshold in this two edges would be less than the appropriate multiple of
transition detection process. the unit interval.
The time separation between ideal positions equals a multi- When the signal had a falling edge followed by a rising
ple of the unit interval. Hence, the optimal decision level for edge, this transition detector would locate the falling edge
transition detection would ensure that the time separation before the crossover point and the corresponding rising
of actual transitions in a jitter-free signal would also equal edge after the crossover point. The time between these
multiples of a unit interval. two edges would be greater than the appropriate multiple
of the unit interval.
For transitions with equal rise and fall times, this optimal
decision level equals the 50% point in the transition. This Because of the non-optimal decision threshold, the detect-
threshold falls on the crossover points in the Eye diagram ed transitions vary from their ideal positions. This non-opti-
for these signals and equals the level where the Eye has mal transition detection process has introduced a determin-
the maximum width (Figure 17). istic jitter component called duty-cycle dependent jitter.
Using a non-optimal decision threshold in transition detec-

tion introduces jitter. Figure 18 illustrates this effect using a
schematic representation of the Eye diagram for a jitter-free
Primer
Using the optimal decision threshold in transition detection

will yield a smaller jitter measurement since the measured
value does not include duty-cycle dependent jitter arising
from a non-optimal level. Hence, we can say that the opti-
mal decision threshold for transition detection is the level
that minimizes jitter and maximizes Eye width.
Most measurement instruments have AC-coupled inputs
and set the decision threshold for transition detection to the
average signal voltage of the AC-coupled signal. In most Figure 19. Acceptable SDI signal with slow rise time and a fast fall
time shows the 50% point does not always equal the
cases, this approach results in near-optimal transition optimal decision level for transition detection.
detection because typical SDI signals are symmetric “in the
in rise and fall times. Figure 19 shows an SDI signal that
long term” (see section 2.9). Over durations equal to many
conforms to the standard but has a slow rise time and a
unit intervals, the signal spends nearly the same amount of
fast fall time. The Eye crossover points appear well below
time at each voltage level. The average signal voltage over
the 50% point in the transition.
these durations lies close to optimal position for transition
detection at the midpoint of the Eye height (Figure 17). If the transition detection stage in a jitter measurement
process used a decision level equal to the 50% point in the
The AC-coupling effects described in section 2.9 can affect
transition, the measurement results would include a signifi-
jitter measurement as well as signal decoding. In particular,
cant amount of duty-cycle dependent jitter. A jitter measure-
long constant-voltage intervals in an SDI signal can shift the
ment method that could align the decision threshold with
signal relative to a fixed decision threshold (see Figure 6).
the Eye crossover points (maximum Eye width) would give a
The near-optimal decision threshold for a symmetric SDI
smaller result (minimal jitter).
signal with many transitions becomes a non-optimal deci-
sion threshold for a shifted signal with long constant voltage The standards do not give any guidance regarding the deci-
intervals. In this case, the transition detection stage in the sion threshold for the transition detection stage in the jitter
jitter measurement introduces duty-cycle dependent jitter. measurement process. In particular, they do not have speci-
fications on compensating for AC-coupling effects or
In the short term, typical SDI signals can spend several unit
accommodating non-symmetric signal transitions.
intervals at the same signal level. Generally, measurement
instruments adequately compensate for AC-coupling effects While differences in transition detection are not a primary
related to this short-term behavior. So, for most SDI signals, source of difference in jitter measurement, additional guid-
the transition detection stage does not introduce enough ance will help ensure more consistent results. Appropriate
duty-cycle dependent jitter to affect the peak-to-peak specification in this area reduces the potential for inconsis-
amplitude measurement. tent measurement results due to duty-cycle dependent jitter
arising from non-optimal transition detection.
Equalizer stress patterns in pathological SDI signals can
introduce significant shifts in the AC-coupled signal (see
Figure 10). In this case, duty-cycle dependent jitter intro-
duced in the transition detection stage could increase the
peak-to-peak jitter amplitude measurement.
These effects support SMPTE’s caution against using stress
patterns in measuring output jitter in SDI signal sources.
The recommended color bar pattern generates an SDI sig-
nal with frequent transitions. After AC-coupling, the average
voltage level in this signal will closely match the optimal
decision threshold for transition detection.
Non-symmetric transitions can also impact jitter measure-
ment. Although most SDI signals have nearly symmetric
transitions times, the standards allow a significant difference
Primer
4.3. Phase detection/demodulation

As noted in section 2.3, we will view jitter as a phase
modulation of the serial data stream. The phase detection/
demodulation stage in the jitter measurement process sep-
arates this phase modulation from the input signal using
one of two basic approaches:
Sampling the phase modulation in the input signal by Figure 20. Equivalent-time Eye method.
collecting individual amplitude measurements of the

phase variation in signal transitions. 4.3.1. Phase detection/demodulation: Equivalent-time
Using demodulation techniques to generate a Eye method
demodulated jitter signal that corresponds to the Figure 20 illustrates the Equivalent-time Eye method.
phase modulation in the input signal. While this diagram reasonably represents implementation
In the following sections, we will describe the characteristics of this method in video-specific measurement instruments,
of three methods used in this stage of the jitter measure- it does not represent jitter measurement in sampling oscillo-
ment process: scopes or other general-purpose instruments that use
The Equivalent-time Eye method constructs an equivalent-time sampling. Current-generation sampling
Equivalent-time Eye diagram of the signal and measures oscilloscopes and signal analyzers use significantly more
the amount the edge samples in the Eye vary from their sophisticated techniques in implementing jitter measure-
ideal positions. ment processes, often with extensive software-based
signal processing. Hence, the following description does
The Phase Demodulation method applies two appro-
not reflect the capabilities and performance of these
priately filtered clock signals to a phase detector. The
instruments.
output from the phase detector is the demodulated
jitter signal.
The Real-time Acquisition method applies signal pro-
cessing algorithms to one or more acquisition records
captured in real-time from single trigger events to
measure the amount each signal edge in the acquisition
record varies from its ideal position.
Video-specific measurement instruments commonly use
the first two methods listed. We include a description of
the third method to briefly touch on some aspects of using
real-time oscilloscopes to measure jitter in SDI signals.
These general descriptions will cover some key characteris-
tics of these methods that can introduce differences in jitter
measurements. They are not in-depth reviews of measure-
ment techniques or the capabilities and performance of
specific instruments. Rather, they identify some key factors
to consider when comparing jitter measurement made by
different instruments.
Primer
Operation and frequency response

This method uses a recovered clock to form an Equivalent-
time Eye diagram of the input SDI signal. This recovered
clock tracks jitter frequencies below the loop bandwidth
of the narrow-band phase-lock loop (PLL), ƒnb.
This low frequency jitter in the recovered clock compen-
sates for the corresponding low frequency jitter in the input
signal. As a result, the Eye diagram only contains jitter with
spectral components above ƒnb. With the proper configura-
tion (see section 4.4.1), the narrow-band PLL can realize
the highpass characteristics specified for measuring timing
jitter (Figure 13) or alignment jitter (Figure 14).
Figure 21. Histogram measurement of peak-to-peak jitter.
To make an automated jitter measurement, the instrument
detects signal samples in the Equivalent-time Eye that lie
close to the decision threshold used for transition detection. For symmetric SDI signals, vertically centering the his-
For each sample, it then computes the time difference togram window in the middle of the Eye diagram will gener-
between the signal sample and the Eye crossover point. ally produce optimal transition detection. For SDI signals
The signal sample corresponds to an actual edge position with non-symmetric rise and fall times, the crossover points
and the Eye crossover point corresponds to an ideal edge are not vertically centered in the Eye (Figure 19). For these
position. The computed difference equals the amount of signals, a peak-to-peak jitter amplitude measurement made
phase variation in an actual signal edge, or equivalently, a with a histogram window centered in the middle of the Eye
sample of the signal jitter. The instrument builds a histogram will contain some duty-cycle dependent jitter. The measured
of these jitter amplitude measurements over some number value will be larger than a measurement made with a his-
of edge samples and reports the width of this histogram as togram window centered on the optimal decision level that
the peak-to-peak jitter amplitude. passes through the crossover point of the Eye.
Figure 21 shows an Equivalent-time Eye pattern created by Unless the measurement process compensates for AC-
using this method on an HD-SDI signal. It also shows a coupling effects, the Eye diagram can shift from its nominal
typical histogram of jitter amplitude measurements and the position (see section 2.11). With a histogram collection win-
associated collection window. In this example, the his- dow fixed in the middle of the nominal Eye diagram, these
togram width equals 187 ps. The unit interval for an HD-SDI shifts will introduce duty-cycle dependent jitter. Even with
signal equals 673 ps. So, the measured peak-to-peak jitter compensation, pathological SDI signals with equalizer
amplitude for this signal is 187/673 = 0.278 UI. stress patterns can significantly shift the Eye diagram and
Transition detection add duty-cycle dependent jitter that increases the peak-to-
In the Equivalent-time Eye method, transition detection peak jitter amplitude measurement. In Equivalent-time Eye
occurs when the instrument selects the Eye samples used diagrams, the presence of “flyer” points outside the main
in the jitter amplitude measurements. To avoid adding duty- Eye diagram indicates that AC-coupling effects have shifted
cycle dependent jitter to the peak-to-peak jitter amplitude the signal levels.
measurement, the histogram collection window should align
with the optimal decision threshold for transition detection
(see section 4.2). Further, the height of the histogram win-
dow should not exceed 5% of the Eye height. Otherwise,
the measurement process will include edge samples from
the upper and lower portions of the Eye where the finite rise
and fall times of signal transitions introduce duty-cycle
dependent jitter (see section 4.2).
Primer
Dynamic range Manual measurements

The Equivalent-time Eye method computes variations in To make a manual jitter measurement, vertical cursors are
transition times with respect to the nearest Eye crossover placed on the signal samples that are:
point, not with respect to the actual transition in the signal. Lying on, or very close to, a line connecting the Eye
Thus, the absolute value of these variations cannot exceed crossover points (optimal decision threshold)
0.5 UI, even if some of the samples actually came from
Farthest to the left and right of a crossover point during
transitions that varied more than 0.5 UI from their ideal
a given time period
positions. Hence, the method shown in Figure 20
cannot quantify jitter with peak-to-peak amplitude The difference between the cursor readings divided by the
greater than 1 UI. unit interval equals the peak-to-peak jitter amplitude in UIs.
Because of the low density or visibility of samples with large
SMPTE RP 192 describes an alternative to the method
jitter amplitudes, it may be difficult to place cursors at the
shown in Figure 20 that can measure peak-to-peak jitter
full extent of the signal jitter.
amplitudes above 1 UI. However, this method does not use
the input signal to create the Eye diagram. Instead, it
replaces the input signal with a second extracted clock
signal. We will briefly describe this method in section
4.3.2 since it shares some characteristics with the Phase
Demodulation method.
Sampling, sampling rate, and coverage
The collection rate for the histogram of jitter amplitude
measurements falls well below the Equivalent-time Eye
sampling rate. Since the measurement process needs to
use a small histogram window to avoid adding duty-cycle
dependent jitter to the peak-to-peak amplitude measure-
ment, only a small percentage of the Eye samples will fall
within the collection window.
For example, suppose that only 2.5% of the Eye samples
fall inside the histogram window. If the instrument adds
samples to the Equivalent-time Eye diagram at a 10 MS/s
rate, the measurement process would collect histogram
values at a rate of 250 kS/s. Thus, the histogram collection
process corresponds to a low-rate sampling of the signal
jitter. As a result, long observations times may be necessary
to fill out a histogram that adequately captures the peak-to-
peak jitter amplitude.
Signal displays
A single histogram collected as described in this section
contains samples of jitter from widely separated signal
edges. Jitter waveform or spectrum displays cannot be
meaningfully constructed from these histogram values.
The Eye display allows some qualitative assessment of jitter
behaviors. The Eye closure indicates the peak-to-peak jitter
amplitude, and patterns in the Eye display may indicate the
presence of deterministic jitter.
Primer
Figure 22. Phase Demodulation method.
4.3.2. Phase detection/demodulation: Phase Transition detection and dynamic range

Demodulation method
In the Phase Demodulation method, transition detection
Figure 22 illustrates the Phase Demodulation method. occurs during clock recovery. The issues around non-sym-
Operation and frequency response metric signal inputs described in section 4.2 could also
affect jitter measurements made with this method.
A wide-band clock recovery circuit extracts clock x from the
digital signal. This clock tracks the jitter in the input signal Although the phase detector has a usable dynamic range
up to the clock recovery bandwidth, ƒcr. A narrow-band of no more than one clock cycle, the Phase Demodulation
PLL derives a reference clock y from clock x. Clock y con- method can measure peak-to-peak jitter amplitudes greater
tains only the jitter components with frequencies below ƒnb. than 1 UI by dividing the frequency of the x and y clock
The phase detector creates a signal z proportional to the signals. Frequency division increases the clock period, i.e.
phase difference between x and y, i.e. the demodulated the size of the unit interval, but the amplitude of jitter in the
jitter signal. This signal contains all jitter components with clock edges remains the same. With an appropriately
frequencies between ƒnb and ƒcr. A subsequent stage selected division ratio, jitter amplitudes much greater than
measures the peak-to-peak value of signal z. 1 UI with respect to the original clock frequency will corre-
spond to amplitudes below 1 UI with respect to the new
With proper frequency response the narrow-band PLL will
clock frequency. The phase detector can now successfully
realize the low frequency characteristics of the bandpass
separate the jitter signal from these lower frequency clocks.
defined for measuring timing jitter (Figure 13) or alignment
Rescaling the demodulated jitter signal converts these
jitter (Figure 14). Alternatively, a high-pass filter inserted
amplitudes back to the appropriate jitter amplitudes with
between the phase detector and the peak-to-peak meas-
respect to the actual unit interval.
urement stage can implement these required restrictions.
Sampling, sampling rate and coverage
To realize the high frequency bandpass characteristics
specified in the SMPTE standards the loop bandwidth of Earlier-generation implementations of the Phase
the clock recovery hardware, ƒcr, would need to be at least Demodulation method used analog peak detection to
1/10 the clock rate of the SDI signal input. In practice, clock measure the peak-to-peak amplitude of signal z in Figure
recovery hardware cannot achieve this high loop bandwidth 22. Later-generation implementations measure the peak-to-
for SDI signals. The loop bandwidth depends on the num- peak jitter amplitude of a digital version of the phase detec-
ber of edges in the signal, and typical SDI signals do not tor output (see section 4.5.1). Sampling rates meet the
have a sufficient number of edges. Reasonably achievable Nyquist criterion of 2 x ƒcr for this band-limited demodulat-
loop bandwidths fall well below 1/10 the data clock rate. ed jitter signal. At typical sampling rates, this method can
collect a large number of jitter samples in a short amount
Thus, the Phase Demodulation method cannot realize the
of time.
high frequency bandpass characteristics specified in the
standards. If the input SDI signal contains jitter components The Phase Demodulation method continuously monitors the
with frequencies between ƒcr and 1/10 the clock rate, the output of the phase detector. The peak-to-peak amplitude
peak-to-peak amplitude measured with this method could measurement will include any intermittent jitter spikes and
be less than a measurement made with a method that fully other occasional or non-periodic jitter behaviors that occur
realized the lowpass characteristics. in this signal within the measurement window.
Primer
floor from the measured peak-to-peak jitter amplitude (see

section 4.6). In this case, the automated measurement can
be noticeably smaller than its corresponding manual meas-
urement. This difference can arise with any manual peak-to-
peak jitter amplitude measurement, not just measurements
made on the jitter waveform display available with the
Phase Demodulation method.
Measuring jitter amplitudes greater than 1 UI with the
Equivalent-time Eye method
As mentioned in section 4.3.1, SMPTE RP 192 describes
Figure 23. Jitter Waveform Display on a Tektronix WFM700M. an alternative Equivalent-time Eye method that can meas-
ure peak-to-peak jitter amplitude above 1 UI. Like the
Phase Demodulation method, this approach extracts clocks
Signal displays x and y from the input signal and divides their frequency.
As shown in Figure 23, the Phase Demodulation method Instead of using these signals as inputs to a phase detec-
can present the demodulated jitter signal at the output of tor, this method uses the y clock to trigger an oscilloscope
the phase detector as a jitter waveform display. In this that forms an Eye diagram from the x clock signal. This
example, the vertical scale shows the jitter amplitude in alternative Equivalent-Eye method can measure peak-to-
normalized units (UIs). The sweep rate equals two video peak amplitude greater than 1 UI for jitter frequencies
fields, and the display clearly shows jitter correlated to the between the bandwidth of the narrow-band PLL and the
video field rate. bandwidth of the wide-band clock recovery circuit.
This direct, in-depth view of the jitter signal is especially
useful in characterizing jitter performance in video equip-
ment and in diagnosing jitter-related problems in video
distribution systems.
Converting the jitter waveform to the frequency domain
will yield a jitter spectrum display, another valuable tool for
understanding jitter. The jitter waveform and spectrum dis-
plays created from the phase detector output will contain
information on jitter frequencies between ƒnb and ƒcr.
Manual measurement
To make a manual jitter measurement, horizontal cursors
are placed at the positive and negative peaks of the jitter
waveform. The difference between the cursor readings
equals the peak-to-peak jitter amplitude in UIs.
As shown in Figure 22, the phase detector output signal is
used to form the jitter waveform display and the jitter signal
output. This signal contains any internal jitter added by pre-
vious stages in the jitter measurement process. Manual
peak-to-peak jitter amplitude measurements made using
the jitter waveform display will contain this internal jitter. Any
displays or measurements made on the jitter signal output
will also contain this internal jitter contribution.
When making automated jitter measurements, many instru-
ments subtract a conservative estimate of the jitter noise
Primer
Figure 24. Real-time Acquisition method.
4.3.3. Phase detection/demodulation: Real-time With currently available acquisition record sizes, a single
Acquisition method acquisition record can capture the jitter frequencies within
Figure 24 shows some of the main processes in the the specified bandpass for alignment jitter measurement
Real-time Acquisition method for jitter measurement. (Figure 14). Measuring spectral components down to the
10 Hz low-frequency corner specified for timing jitter meas-
Overview of method and frequency response
urement (Figure 13) requires TIE measurements collected
An instrument first captures an acquisition record from a
over multiple acquisitions. These TIE measurements will not
single trigger event. Signal processing software detects
contain any information about signal jitter in the gaps
transitions and extracts a reference clock that defines the
between acquisitions.
ideal positions for transitions in the data signal.
Transition detection and dynamic range
After establishing this reference, the instrument measures
Typically, the parameters used in transition detection and
the time interval error (TIE) for each transition in the data
clock recovery can be adjusted. In particular, changing the
signal, i.e. the difference in time between the actual and
decision level used in the edge finding process can adjust
ideal positions. The Statistics stage analyzes the collection
jitter measurements for non-symmetric SDI signals. Also,
of TIE measurements and determines various properties. In
the Real-time Acquisition method does not require any
particular, this stage computes the difference between the
special configuration, e.g., clock division, to measure peak-
maximum and minimum TIE values, which equals the peak-
to-peak jitter amplitudes greater than 1 UI.
to-peak amplitude of the jitter in the acquisition record with
respect to the recovered reference clock. Sampling, sampling rate and coverage
The signal processing used in extracting the reference clock The TIE measurements correspond to samples of the
can implement different clock recovery algorithms. These demodulated jitter signal, although they are not equally-
clock recovery algorithms differ in their ability to exclude spaced samples of the jitter signal. Figure 25 illustrates
wander from timing and alignment jitter measurements. the correspondence between TIE measurement and
Additional filtering may be needed to realize the highpass jitter samples.
filter characteristics shown in Figure 13 and Figure 14.
Acquisition record size also affects this method’s frequency
response. In particular, it affects timing jitter measurement.
For example, suppose an instrument samples an SDI signal
at 10 GS/s and stores these samples in a 64 MB acquisi-
tion record. This acquisition corresponds to a time interval
of 6.4 ms, or slightly more than one period of 160 Hz jitter.
A measurement of the peak-to-peak timing jitter amplitude
using this record will not include a full cycle of any spectral
components in the jitter with frequencies below 160 Hz.
Primer
The jitter waveform and spectrum displays contain informa-

tion on the jitter frequencies that can be adequately
captured in a single acquisition record. Because of the
gaps between acquisition records, these displays cannot
accurately represent low frequency jitter components that
cannot be captured in a single record.
Due to the high sampling rates used in data acquisition, the
Real-time Acquisition method can determine the shape of
the input SDI signal. In particular, it can show actual edge
transitions in a Real-time Eye diagram or in a signal wave-
form display. Engineers can “zoom in” on various displays
to examine the signal data in different parts of the acquisi-
Figure 25. TIE measurements and jitter samples. tion record at various scales. They can use this capability to
correlate specific jitter behavior with data patterns or other
signal characteristics in the SDI input signal.
The figure shows a reference clock, a data signal with
Vertical cursors can be used to make manual jitter meas-
sinusoidal jitter, and samples of the jitter signal taken at the
urement on either a Real-time Eye or Equivalent-time Eye
clock rate. TIE measurements correspond to the samples
using the procedure described in section 4.3.1. Many
shown in red. The samples shown in black, dashed lines
instruments also offer Eye mask testing and other analysis
have no corresponding TIE measurement because the data
tools that can help characterize both amplitude noise and
signal does not have a transition at this sample point.
jitter in SDI signals.
The value of these “missing” samples can be interpolated
from the actual TIE measurements. The combined set of
actual and interpolated TIE measurements correspond to
sampling a demodulated jitter signal at the data clock rate.
This “effective” sampling rate is well above the Nyquist
criteria for spectral components in the jitter signal with
frequencies less than 1/10 the data clock rate, i.e. the
minimum high-frequency corner point for measuring timing
or alignment jitter. At these sampling rates, the Real-time
Acquisition method can collect a large number of TIE
measurements in a short duration.
The Real-time Acquisition method cannot continuously
monitor the jitter signal over durations longer than the time
span of a single acquisition record. It can detect any inter-
mittent jitter spikes and other occasional or non-periodic
jitter behaviors that occur during an acquisition, but will
not capture behaviors that occur during the gaps between
acquisitions.
Signal displays and manual measurement
The plotting subsystem shown in Figure 24 can generate
several graphical views of the TIE measurements. In partic-
ular, this section can produce a time trend plot of the TIE
measurements that is equivalent to the jitter waveform
display available with the Phase Demodulation method.
Applying a Fourier transform to the TIE measurements will
yield a jitter spectrum display.
Primer
Phase Detection/Demodulation Method

Equivalent-time Eye Phase Demodulation Real-time Acquisition
Measures individual edge Demodulates baseband jit- Measures individual edge
variation using samples ter signal from input signal variation on all edges
Operation from widely separated using recovered clocks, captured in one or more
edges captured in an band-limited by clock acquisitions
Equivalent-time Eye recovery bandwidth
Frequency response No frequency-related Clock recovery bandwidth Acquisition record size
limitations affects measurement of affects measurement of
high frequency jitter low frequency jitter
Dynamic range Typically less than 1 UI Typically greater than 1 UI Greater than 1 UI
Sampling of Style Equivalent-time Real-time Real-time

baseband
jitter Typically between 25%
modulation Typical rates < 1.5 MS/sec 5-20 MS/sec
and 50% of data rate
Coverage Depending on observation Continuously monitors Full coverage within acqui-

time, low sample density band- limited demodulated sition record, no coverage
can miss contribution from jitter signal (phase detector in gaps between records
intermittent jitter behaviors output)
Signal displays Cannot construct jitter Can produce jitter Can produce jitter wave-
waveform or spectrum dis- waveform and spectrum form and spectrum displays
play from histogram data displays of band-limited from TIE measurements
demodulated jitter signal within a single acquisition
record
Table 3. Summary of jitter specifications.
4.3.4. Phase detection/demodulation: Summary

of methods
For the three phase detection/demodulation methods
described in this section, Table 3 summarizes the key
characteristics that most affect peak-to-peak jitter
measurement.
Primer
For this analysis, let the s-domain representation of the

lowpass transfer functions of the narrow-band PLL and
the clock recovery be:
Where:
Figure 26. Mathematical flow graph of Phase a3 = 2πƒ3 = 2π · (0.1 MHz), a4 = 2πƒ4 = 2π · (148.5 MHz).
Demodulation method.
H3(s) is the transfer function for a 3rd-order narrow-band
4.4. Measurement filters PLL and H4(s) is the transfer function for a 1st-order wide-
band clock recovery process. The values selected for a3
Differences in the phase detection/demodulation methods
and a4 will realize the bandpass specification for measuring
lead to differences in realizing the bandpass restrictions
alignment jitter on an HD-SDI signal.
specified for measuring timing and alignment jitter (Figure
13, Figure 14). In this section, we present a short mathe- The total transfer function is:
matical analysis of these differences. We also describe the
impact of filter accuracy on jitter measurement. H(s) = H4(s)[1 – H3(s)] =
4.4.1. Filter realization
We start with an analysis for the Phase Demodulation
method. As shown in Figure 22, this method uses two
phase-locked loops, i.e. the wide-band clock recovery
PLL and a narrow-band PLL. This makes the analysis
slightly more complex.
Figure 26 is a mathematical flow graph of the phase pro-
cessing corresponding to Figure 22. The phase of the input
signal is designated θin, and the phases of clocks x and y
are designated θx and θy, respectively. The lowpass transfer
function of the clock recovery is H4(s), and the lowpass
transfer function of the narrow-band PLL is H3(s). The com-
bination of these two transfer functions realize a bandpass
function H(s) that satisfies the frequency restrictions for
measuring timing and alignment jitter.
Primer
Figure 27. Frequency responses of H3(s), H4(s), and H(s) in Figure 26.
The frequency responses of these transfer functions are Removing H4(s) from the model in Figure 26 produces the
plotted in Figure 27. Note that the lowpass function H3(s), mathematical flow graph of the phase processing for the
when subtracted from unity, yields the highpass function Equivalent-time Eye method (Figure 20). To conform to IEEE
in H(s). Compare the frequency response of |H(s)| with the Std. 1521, the narrow-band PLL in this method would need
frequency response in Figure 14. the same lowpass transfer function:
The choice of a4 realizes the SMPTE 292M specification
that ƒ4 be at least 1/10 of the 1.485 GHz clock rate of the
HD-SDI signal. As noted in section 4.3.2, clock recovery
This leads to a high-pass transfer function for the total
hardware cannot achieve this loop bandwidth because SDI
transfer function H(s) of:
signals do not have sufficient edges. For an actual imple-
mentation of the Phase Demodulation method, a4 would be
a smaller value determined by the clock recovery bandwidth. H(s) = 1 – H3(s) =
This situation does not necessarily imply that measurements
made with the Phase Demodulation method will be lower
than measurements made with the other methods. The jitter
spectrum of many SDI signals does not contain significant
energy above commonly available clock recovery band-
widths. In this case, measurements made with the Phase
Demodulation method can agree closely with the measure-
ments from other methods (Appendix A).
With a 3rd-order, narrow-band PLL, the highpass slope
of H(s) rises faster than the minimum 40 dB/decade
specification shown in Figure 14. Instead, it realizes the 60
dB/decade highpass slope shown in Figure 13 and speci-
fied by IEEE Std. 1521 for proper rejection of wander. Using
this approach, the same implementation can measure either
timing or alignment jitter.
Primer
4.4.2. Filter accuracy

To correctly reject wander components in SDI signals, IEEE
Std. 1521 specifies that the highpass corner of the band-
pass for timing jitter measurements, ƒ1, should equal 10 Hz
± 20%. The standards do not specify accuracy for ƒ3 and
ƒ4. As described in section 3.2, the standards specify mini-
mum values for the bandpass slopes. The slopes shown in
Figure 13 and Figure 14 represent the combination of mini-
mal specifications from all relevant standards.
Using different values for the frequency cutoffs and slopes
Figure 28. Frequency response of total transfer function H(s) for can produce different peak-to-peak jitter amplitude meas-
Equivalent-Eye Method.
urements. The amount of variation depends on the spectral
characteristics of the jitter signal.
The result is a total transfer function H(s) that effectively has
no roll-off at high frequencies, as shown in Figure 28. This Filter accuracy has an especially strong impact on timing
response is consistent with the specification that ƒ4 be at jitter measurement because these measurements need to
least 1/10 the clock rate. If a3 equal 2π · (0.1 MHz), the reject wander components. Some video devices, e.g.,
response shown in the figure has a low-frequency cutoff MPEG decoders, can produce outputs with large wander
that meets SMPTE 292M requirements for measuring components. Further, wander can build as a signal flows
alignment jitter. through a video system.
In the Real-time Acquisition method, proper configuration In these cases, differences in realizing the bandpass cutoff
of the measurement filter stage could realize a frequency frequency and slope specified for timing jitter measure-
response like Figure 28. Proper configuration of PLL-based ments can lead to significantly different values for peak-to-
clock recovery software could also realize this transfer peak amplitude. In particular, measurements made with a
function if the clock recovery algorithm could implement bandpass that conforms to the SMPTE RP 192 specifica-
a 3rd-order PLL. Otherwise, additional filtering would be tion for at least 40 dB/decade wander rejection can be
required to realize the appropriate highpass slope. This much larger than measurements made with a bandpass
method could also realize the frequency response in Figure that conforms to the IEEE Std. 1521 specification for 60
27 with proper configuration of the filtering and clock dB/decade wander rejection (see section 6.2.1).
recovery algorithms.
For all methods, setting a3 = 2π · (10 Hz) in the narrow-
band PLL produces a system transfer function that realizes
the bandpass restriction for measuring timing jitter
(Figure 13).
Primer
4.5. Peak-to-Peak measurement bandpass shown in Figure 28, the effective jitter signal
The last stage of the jitter measurement process determines bandwidth equals the highest jitter frequency in the
the peak-to-peak jitter amplitude. While the standards input signal.
specify that output jitter shall be specified and measured as A set of samples collected at a rate greater than the
a peak-to-peak quantity, they give little guidance on making Nyquist rate will have the information needed to reconstruct
the measurement (see section 3.6). Much of the variation the jitter signal. In fact, this sample set contains redundant
seen in jitter measurement results arises from differences information about the jitter signal. Specifically, adjacent
in implementing this stage in the process. samples are not independent samples of the demodulated
4.5.1. Peak-to-peak detection methods jitter signal. The larger sample set can be constructed
from a smaller subset of independent samples. Adjacent
The Equivalent-time Eye method and the Real-time
samples in the full sample set have some degree of
Acquisition method measure peak-to-peak amplitude by
time-correlation.5
calculating the difference between the minimum and maxi-
mum values in a collection of jitter amplitude values. Any set of jitter samples collected at a rate lower than the
Nyquist rate cannot represent the complete jitter signal.
The Phase Demodulation method measures the peak-to-
However, this set may be sufficient to determine several
peak amplitude of a band-limited demodulated jitter signal
statistical properties of the phase modulation, e.g., mean,
(phase detector output). Earlier-generation implementations
variance, and RMS. For best results, this sub-Nyquist sam-
of the Phase Demodulation method use analog peak detec-
pling should produce a set of independent jitter samples.
tion. Later-generation implementations digitally sample the
With a sufficient number of independent jitter samples, this
signal from the phase detector and measure the peak-to-
set can yield acceptable estimates of statistical properties.
peak amplitude of the jitter samples.
As we show in the following sections, both the sampling
In analog peak detection, the attack time of the peak detec-
method and the number of independent jitter samples used
tors strongly affects the accuracy of the peak-to-peak
in a peak-to-peak amplitude measurement can significantly
amplitude measurements. Short attack times let the peak
impact the measurement result. The different jitter measure-
detector more accurately track rapid changes in jitter signal
ment methods collect independent jitter samples in different
amplitudes. Long attack times cannot track these changes
ways and at different rates. Thus, we cannot easily com-
and produce lower amplitude measurements. Generally,
pare peak-to-peak measurements made over identical time
digital implementations can more accurately measure the
intervals because these measurements do not involve the
true peak-to-peak value.
same number of independent jitter samples.
4.5.2. Independent jitter samples and normalized meas-
Instead, we will use the number of independent jitter sam-
urement time
ples as a method-independent way to describe how the
In section 3.6, we briefly described the relationship between
duration of a peak-to-peak amplitude measurement affects
measured peak-to-peak jitter amplitude and measurement
the measurement result. We will call the number of inde-
time. In exploring this relationship further, we will use two
pendent jitter samples collected during measurement time
concepts connected with sampling the jitter signal:
T the normalized measurement time. We can then compare
Nyquist-rate sampling and independent samples.
the actual measurement times each method requires to col-
To collect all the available information about a jitter signal, lect the number of independent jitter samples correspon-
including the jitter waveform shape, the sampling rate ding to a particular normalized measurement time.
must be at least the Nyquist rate. The Nyquist rate equals
2· ƒJBW where ƒJBW is the “effective” jitter signal bandwidth.
In the case of the Phase Demodulation method, the effec-
tive jitter signal bandwidth is less than or equal to the loop
bandwidth of the wide-band clock recovery process. To the
degree that the other two methods realize the measurement
5 In mathematical terms, the first zero in the autocorrelation function associated with the demodulated jitter signal occurs at ± 1/ (2·ƒ
BW), where ƒBW is the bandwidth of the
signal’s power spectrum. Samples separated by this time interval will be independent, i.e. will not be time-correlated. Samples spaced at less than this interval will have some
degree of correlation. A sample spacing of 1/ (2·ƒBW) corresponds to a sample rate of 2·ƒBW, i.e. the Nyquist rate.
Primer
Figure 29. Effect of normalized measurement time on measured Figure 30. Effect of normalized measurement time on the
peak-to-peak jitter amplitude (J/rms). consistency (standard deviation σ ) of a peak-to-peak
jitter measurement.
The normalized measurement time, N, corresponding As expected, the ratio increases. Increasing the number of
to an actual measurement time, T, is given by the equation independent jitter samples increases the probability that the
N = T · min(S, 2·ƒJBW). In this equation, S is the actual rate set of jitter samples will include some of the high amplitude
that the jitter measurement process collects samples of the values in the tails of the Gaussian distribution. Hence,
signal jitter; ƒJBW is the effective jitter signal bandwidth; J increases with respect to the RMS value. Since the
and the function min(x,y) equals the minimum value of the Gaussian distribution has unbounded peak-to-peak
two arguments. amplitude, this ratio will continue to grow with the number
4.5.3. Measuring the peak-to-peak amplitude of of independent jitter samples.
random jitter However, using the same number of independent jitter sam-
We first consider the impact of normalized measurement ples in multiple measurements of peak-to-peak amplitude
time on the measured peak-to-peak amplitude of random will not yield identical results because these measurements
jitter. As noted in section 2.6, random jitter is generally are sampling a random process. Different sample sets will
modeled by a Gaussian amplitude distribution. In practice, contain amplitudes from different points in the Gaussian dis-
the peak-to-peak amplitude of random jitter has the tribution. Thus, multiple peak-to-peak amplitude measure-
“unbounded” property associated with this probability ment will also have random variation.
distribution, i.e. the measured peak to peak amplitude The red line in Figure 30 shows this variation for Gaussian
increases as the measurement time increases. random jitter by plotting the ratio of the standard deviation
With any random process, measurement results depend on in a set of peak-to-peak measurements (σ) to the average
the number of independent samples used in the measure- peak-to-peak measurement (J) as a function of normalized
ment. A more precise statement of the unbounded property measurement time. As the number of independent jitter
would say that the measured peak-to-peak amplitude of samples increases, this ratio decreases. The larger sample
random jitter increases as the number of independent sets more fully characterize the random jitter and produce
samples of the random jitter increases. In other words, the more consistent results over multiple measurements.
peak-to-peak amplitude measurement increases monotoni- Typical video signals include both deterministic jitter and
cally with increased normalized measurement time. random jitter. The J/rms and σ /J ratios for signals contain-
Figure 29 shows the relationship between peak-to-peak ing both deterministic and random jitter will not fall on the
amplitude and normalized measurement time for Gaussian red lines in Figure 29 and Figure 30.
random jitter (red line). The graph plots the ratio of the
peak-to-peak measurement (J) to the RMS jitter amplitude
as a function of normalized measurement time.
Primer
To see why, consider two signals, A and B, where signal

A contains only Gaussian-like random jitter and signal B
contains both Gaussian-like random jitter and deterministic
jitter. Suppose the RMS amplitude of the random jitter in
signal A, (RMSA) equals the RMS amplitude of the total
jitter in signal B from both random and deterministic com-
ponents (RMSB). In this case, the random jitter in signal
B (RMSBran) must have lower RMS amplitude than the
random jitter in signal A (RMSA).
The sketch in Figure 31 illustrates the relationship between
peak-to-peak jitter amplitude measurements and normal-
ized measurement times for these two signals. In signal
A (green line) random jitter determines the peak-to-peak Figure 31. Peak-to-peak jitter amplitude for signals A and B.
amplitude measurement for all normalized measurement
times. Consistent with Figure 29, the peak-to-peak ampli-
tude measurement increases as the normalized measure- decreases the J/rms and σ /J ratios relative to the values for
ment time increases. signals containing only random jitter.
For signal B (blue line), bounded deterministic jitter domi- As these plots illustrate, we cannot define a “correct”
nates the peak-to-peak jitter amplitude measurement for normalized measurement time based on the distribution of
normalized measurement time below some value, N. For jitter values, either Gaussian or mixed. The measured peak-
normalized measurement times larger than N, the unbound- to-peak amplitude continues to increase as the normalized
ed Gaussian-like random jitter in B determines the value of measurement time increases. It does not “level off” above
the peak-to-peak jitter amplitude measurement. Since some specific number of independent jitter samples.
RMSBran is less than RMSA the peak-to-peak jitter ampli- Thus, to enable greater consistency in the peak-to-peak
tude in signal B does not grow as quickly as the peak-to- jitter amplitude measurements made by different measure-
peak jitter amplitude of signal A when the normalized ment methods, the standards need to specify the number
measurement time increases. of independent jitter samples used in these measurements.
Suppose JA is the peak-to-peak amplitude of 1 x 104 For example, if each method measured the peak-to-peak
independent samples of the jitter in signal A, and JB is the amplitude of 5 x 105 independent jitter samples from a
peak-to-peak jitter amplitude of signal B over the same nor- typical video signal, Figure 30 indicates that the standard
malized measurement time. Since signal A has only random deviation of the measured values would fall between
jitter, the ratio JA / RMSA will fall on the red line in Figure 29. 2% and 2.5%.
For signal B, this same ratio will fall below the red line. The
ratio JB / RMSB is less than the ratio JA / RMSA because
RMSA = RMSB and JB < JA.
So, for the same normalized measurement time, the J/rms
ratio for a signal with only random jitter (signal A) will fall on
the red line in Figure 29, while the J/rms ratio for a signal
with both deterministic and random jitter will fall below this
red line. The σ /J ratios will behave in the same way.
Appendix B describes an experiment that measures peak-
to-peak jitter amplitudes for one example of a typical video
signal. The blue lines in Figure 29 and Figure 30 show a
plot of these measurements. The results agree with the
preceding analysis that the presence of deterministic jitter
Primer
4.5.4. Measurement times Thus, peak-to-peak amplitude measurements made over

The different methods can collect jitter samples at different the same number of jitter samples do not necessarily corre-
rates. As a result, peak-to-peak amplitude measurements spond to measurements made over the same number of
made over the same duration can involve significantly differ- independent jitter samples, i.e. over the same normalized
ent numbers of jitter samples. As shown in section 4.5.3, measurement time. If low frequency jitter dominates the
measurements made over different normalized measure- jitter spectrum, a larger number of closely-spaced samples
ment times can yield significantly different results, primarily may be needed to correspond to the same normalized
due to the unbounded nature of random jitter. measurement time as a set of more widely-spaced samples.
To produce comparable results, the different jitter measure- 4.5.5. Dynamic range and jitter value quantization
ments need to measure the peak-to-peak jitter amplitude Implementation of the three jitter measurement methods
over the same normalized measurement time, i.e. the same can differ in the dynamic range and quantization of the
number of independent jitter samples. This will require peak-to-peak amplitude measurement. In particular, some
different actual measurement times. As a simple example, implementations may measure peak-to-peak amplitude
we consider the actual time each method requires to collect greater than 1 UI, while others may measure only peak-to-
1 x 106 jitter samples. peak jitter amplitudes up to 1 UI. If the two implementations
As described in section 4.3.1, the Equivalent-time Eye capture peak-to-peak amplitude measurements in digital
method collects histogram values at a rate determined by words with the same number of bits, values in an imple-
the sampling rate used in forming the Eye and the height of mentation with a dynamic range greater than 1 UI will have
the histogram window. If the measurement process collects larger quantization steps than values in an implementation
histogram values at a 5 kS/s rate, it will take 3.33 minutes with a 1 UI dynamic range.
to collect 1 x 106 jitter samples. At a 250 kS/s rate, this
measurement time decreases to 4 seconds.
A digital implementation of the Phase Demodulation
method directly samples the output of the phase detector
at or above the Nyquist rate for this band-limited signal. If
the measurement process samples at 10 MS/s it can col-
lect 1 x 106 jitter samples of this band-limited demodulated
jitter signal in 100 ms.
As discussed in section 4.3.3, the Real-time Acquisition
method samples the jitter signal on each transition in the
data signal, so the number of samples in an acquisition
depends on the number of signal edges that occurred
during the acquisition. If transitions occurred on 50% of
the unit intervals in a 1.485 Gb/sec HD-SDI signal, then
acquisitions covering 1.35 ms of this signal would acquire
1 x 106 jitter samples.
The degree of time correlation between adjacent jitter
samples depends on the spacing between samples and
the jitter spectral components (see section 4.5.2). Broadly
speaking, widely-separated samples, e.g., like samples
collected in the Equivalent-time Eye method, will have less
time correlation than more closely-spaced samples, e.g.,
like samples collected by the Real-time Acquisition method.
Primer
4.6. Jitter noise floor Trigger jitter: If a jitter measurement process collects
Every implementation of any jitter measurement method has samples from acquisitions taken over multiple triggers,
internal distortions, noise, and variations. These can arise variations in this triggering process can introduce varia-
from unavoidable physical properties or from engineering tions in the timing of the sampling process. These timing
choices made when designing and implementing a jitter variations can contribute to the jitter noise floor.
measurement process. While careful design can reduce the
impact of these inherent behaviors, it cannot eliminate them
completely. They impose a lower bound on jitter amplitude
measurements, called the jitter noise floor.
Any step in the measurement process can contribute to the
jitter noise floor. Primary contributors include:
Timebase jitter: Various processes, e.g., sampling and
clock recovery, require a stable timing reference. Many
implementations use crystal or dielectric resonator oscil-
lators to create a timing reference signal. Phase noise
or other distortions in these oscillators contribute to
the jitter noise floor. Silicon oscillators can contribute
significant timebase jitter to the jitter noise floor.
Clock recovery jitter: Distortions and noise in the hard-
ware-based clock recovery process can also contribute
to the jitter noise floor. Phase noise can contribute ran-
dom jitter and long strings of identical bits can contribute
deterministic, data-dependent jitter. The peak-to-peak
amplitude of this data-dependent jitter is proportional to
the clock recovery bandwidth. In Appendix C, we show
that clock recovery bandwidths large enough to realize
SMPTE measurement bandpass specifications introduce
significant data-dependent jitter.
Equalizer jitter: Hardware-based equalization processes
can also contribute random and deterministic jitter com-
ponents to the jitter noise floor. This includes determinis-
tic data-dependent ISI due to imperfect equalization and
duty-cycle dependent jitter from non-optimal threshold
detection (section 4.1).
The frequency-dependent gain used in equalization (see
Figure 4 in section 2.7) can also introduce jitter. This gain
can increase amplitude noise present in the pre-equal-
ized signal. In the transition detection stage, amplitude
noise near the decision threshold can cause noticeable
phase noise (jitter) in the detected signal edges. This
AM-to-PM effect contributes high frequency jitter to the
noise floor.
Primer
Method Key Characteristics

Measures peak-to-peak jitter amplitude using a histogram of jitter values collected from
equivalent-time samples of an SDI signal.
Can measure any jitter frequencies within the SMPTE-specified bandpass restrictions.
Equivalent-time Eye
Dynamic range equals 1 UI in typical implementations, though an alternative version can
measure peak-to-peak jitter amplitudes greater than 1 UI.
Requires long measurement times to collect a large number of independent jitter samples.
Measures peak-to-peak jitter amplitude by continuously sampling a band-limited
demodulated jitter signal formed by detecting phase differences in two clock signals
recovered from the SDI signal
Typically has an equalization stage for measuring jitter across long cables, which can
increase jitter noise floor.
Can measure any jitter frequencies within the SMPTE-specified bandpass restrictions
Phase Demodulation up to the loop bandwidth of the clock recovery circuit, where maximum loop bandwidths
fall well below 1/10 the data clock rate.
Dynamic range greater than 1 UI in typical implementations.
Can collect a large number of independent jitter samples over a short measurement time.
Can display a jitter waveform and jitter spectrum of the phase detector output, i.e. of a
band-limited demodulated jitter signal.
Can supply a continuous, band-limited demodulated jitter signal output from the
phase detector.
Measures peak-to-peak jitter amplitude by computing time interval errors for each signal
transition in one or more fixed-size records acquired by sampling an SDI signal in real-time.
Can measure any jitter frequencies within the SMPTE-specified bandpass restrictions,
requiring multiple acquisitions to measure down to the low-frequency limit.
Real-time Acquisition Can measure peak-to-peak jitter amplitudes greater than 1 UI.
Can collect a large number of independent jitter samples over a short measurement time,
may require multiple acquisitions to achieve normalized measurement times.
Can generate a jitter waveform and jitter spectrum from TIE measurements.
Can correlate jitter to signal data.
Table 4. Comparison of jitter measurement methods.
4.7. Comparing jitter measurement methods Comparing jitter measurements: sinusoidal jitter
Table 4 summarizes the key characteristics of the different Suppose an SDI signal contains random jitter with very low
jitter measurement methods. RMS amplitude and 1 MHz sinusoidal jitter with peak-to-
Depending on the characteristics of the jitter in an SDI peak amplitude less than 1 UI. The different methods will
signal, differences among the described methods can measure similar peak-to-peak jitter amplitude for this
produce different peak-to-peak jitter amplitude measure- signal. Some variation may occur due to differences in
ments. The following examples illustrate several factors the quantization steps used in the measurement, or in
to consider when comparing jitter measurements made the jitter noise floor.
with these methods.
Primer
Comparing jitter measurements: high amplitude jitter Comparing jitter measurements: isolated jitter spikes
Now consider the same SDI signal, but with peak-to-peak Instead of sinusoidal jitter, suppose that the deterministic
amplitude greater than 1 UI. Typical implementations of the jitter in the SDI signal consists of alternating jitter spikes
Phase Demodulation and the Real-time Acquisition method where these spikes are 300 µs in width, have amplitudes
will detect this high amplitude jitter and will measure a simi- slightly above 0.1 UI and are separated by 32 ms.
lar value. Typical implementations of the Equivalent-time Eye As noted in section 3.6, because the SMPTE standards do
method (Figure 20) cannot detect jitter amplitudes greater not specify a measurement time, it is not clear whether the
than 1 UI. “correct” peak-to-peak amplitude measurement for this jitter
Comparing jitter measurements: high frequency jitter should include both jitter spikes. For this example, we will
For this comparison, suppose the sinusoidal jitter in the SDI interpret the standards to mean that the peak-to-peak jitter
signal is now 25 MHz sinusoidal jitter with peak-to-peak amplitude measured in this situation should be near 0.2 UI.
amplitude less than 1 UI. Typical implementations of the An implementation of the Phase Demodulation method
Equivalent-time Eye and Real-time Acquisition method can that measured the peak-to-peak jitter amplitude over a
measure the peak-to-peak jitter amplitude for this signal. time interval greater than 64 ms would produce a result
Implementations of the Equivalent-time Eye method may near 0.2 UI.
have a higher jitter noise floor, including trigger jitter. Hence, With the Equivalent-time Eye method, measurements made
they may produce slightly higher values. with histograms collected and reset over short observation
Typical implementations of the Phase Demodulation method times may not contain samples from both peaks, which
have clock recovery bandwidths below 25 MHz. The peak- would produce a peak-to-peak amplitude value below 0.2
to-peak amplitude measurement will not include the contri- UI. Collecting more samples over longer observation times
bution from this high frequency jitter and will produce a will eventually yield a histogram that contains samples from
smaller peak-to-peak amplitude measurement. both jitter peaks, which would yield a result near 0.2 UI.
Comparing jitter measurements: low frequency jitter At current acquisition record sizes, an implementation of the
Next consider an SDI signal containing 10 Hz sinusoidal Real-time Acquisition method cannot capture both jitter
jitter with peak-to-peak amplitude less than 1 UI. The three peaks in a single acquisition. A sufficient number of multiple
methods can measure similar peak-to-peak jitter amplitude acquisitions will likely contain samples from both jitter peaks
for this signal as long as they correctly implement the high- and would yield a result near 0.2 UI. Measurements made
pass characteristics of the specified bandpass for timing over a small number of acquisitions may not contain sam-
jitter measurements. Since a single acquisition in the Real- ples from both peaks, which would produce a lower peak-
time Acquisition method will capture only a segment of this to-peak amplitude value. The actual time needed to acquire
variation, this method will need to acquire TIE measure- and process a sufficient number of acquisitions depends on
ments over multiple acquisitions. the combined durations of the individual acquisitions and
the processing gaps between acquisitions.
Comparing jitter measurements: wander
Comparing jitter measurements: intersymbol interference
Now suppose an SDI signal does not contain any determin-
due to cable attenuation
istic jitter above 10 Hz, but does have a high amplitude
wander component, i.e. a timing variation at a frequency Suppose that:
below 10 Hz. This wander component can impact peak-to- (1) The output of an SDI signal source is almost
peak amplitude methods made with any method. Suppose jitter-free;
one implementation (any method) realizes a bandpass filter (2) This output is routed through long cables and several
with a 40 dB/decade attenuation of frequencies below 10 pieces of non-reclocking video equipment to an SDI
Hz, while another implementation (any method) realizes the receiver; and
60 dB/decade slope specified in IEEE Std. 1521. The first
(3) We want to measure the jitter in the signal at the
implementation will not reject the high amplitude wander
receiver input.
component as well as the second implementation and will
produce higher peak-to-peak amplitude measurements
(see section 6.2.1).
Primer
SDI receivers have cable equalizers that compensate for With observation times that appropriately account for these
cable attenuation effects. To assess signal jitter that will considerations, any of the methods described in this guide
affect signal decoding, the measurement process should can measure peak-to-peak jitter amplitudes over similar
exclude the data-dependent jitter due to ISI from cable normalized measurement times and produce consistent and
attenuation. They can do this by implementing an equaliza- comparable results. Without considering these factors, the
tion process as shown in Figure 16, or by using signal random jitter in SDI signals can lead to significantly different
processing algorithms to separate and remove the data- values for peak-to-peak jitter amplitude measurements.
dependent jitter component. Measurement processes that
do not exclude this jitter will produce larger peak-to-peak
amplitude measurements.
Comparing jitter measurements: jitter measurement over
a short cable
Now suppose we use the same measurement processes to
directly measure the output of the SDI signal source in the
previous example over a short cable. Since, by supposition,
the source output is nearly jitter-free, the jitter noise floor
of the measurement process will determine the result of a
peak-to-peak jitter amplitude measurement. Processes with
a hardware-based equalization stage may produce a larger
result due to contributions that the cable equalizer makes
to the jitter noise floor.
Comparing jitter measurements: random jitter
All SDI signals contain some amount of random jitter, and
random jitter is a primary contributor to variation in jitter
measurements. Sections 4.5.2 to 4.5.4 describe several
key considerations that influence peak-to-peak amplitude
measurements of this essentially unbounded jitter compo-
nent. In particular:
The measured peak-to-peak jitter amplitude depends
on the number of independent jitter samples used in
the measurement.
Peak-to-peak jitter amplitude measurements made over
the same number of independent jitter samples will pro-
duce consistent results. Using larger normalized meas-
urement times, i.e. more independent jitter samples, will
lead to lower variation in measurement results.
The different measurement methods collect different
numbers of jitter samples over equal observations times.
Depending on the spacing of the jitter samples and the
spectral components in the jitter, measurements made
of the same number of jitter samples do not necessarily
correspond to the measurements over the same normal-
ized measurement time, i.e. the same number of inde-
pendent jitter samples.
Primer
5.0 Data Error Rates and Jitter Measurements

In section 4.0 we showed how differences in measurement the same number of samples from Source B’s output would
processes and technology can lead to differences in peak- equal approximately 0.13 UI. With the additional alignment
to-peak jitter amplitude measurements. Measurements jitter from routing equipment, the alignment jitter in the sig-
made by two different methods on the same SDI signal nal from Source A has peak-to-peak amplitude of 0.28 UI
can differ substantially. In fact, measurements made on the at the receiver’s input. The alignment jitter in the signal from
same SDI signal by different implementations of the same Source B has peak-to-peak amplitude of 0.33 UI at the
jitter measurement method can vary significantly. receiver’s input. So, for signals from either SDI source, the
Each method has strengths and limitations. The compar- signal jitter at the receiver’s input will not exceed 0.4 UI at a
isons in section 4.7 showed that no one jitter measurement rate greater than 1 in 103 bits, i.e. the BER for this system
method can fully capture the wide variation in jitter charac- lies below 1 x 10-3.
teristics. Further, because of random jitter, the measured The situation changes for a larger number of samples.
peak-to-peak jitter amplitude depends on measurement Over 1 x 1010 samples, the random jitter in Source B’s
time, or more accurately, the number of independent jitter output has a peak-to-peak amplitude of J = 13 · rms =
samples used to determine the peak-to-peak value. 13 · 0.020 = 0.26 UI, while Source A’s output has a peak-
As a result, we cannot determine the jitter in a video system to-peak amplitude near 0.16 UI. With the additional align-
and its impact on system operation by making a single ment jitter from routing equipment, the alignment jitter in the
peak-to-peak amplitude measurement with one jitter meas- signal from Source B now has peak-to-peak amplitude of
urement method. Engineers may need to use a combination 0.46 UI at the receiver’s input. The alignment jitter in the
of jitter measurement methods to fully characterize jitter. signal from Source A has peak-to-peak amplitude of 0.36
In fact, the variation in measurements made with different UI at the receiver’s input. Over 1010 bits, the SDI receiver
methods can offer useful information on jitter characteris- can still tolerate the jitter in signals from Source A since the
tics. We will discuss this topic further in section 7.0. peak-to-peak amplitude falls below 0.4 UI. Signals from
Source B will produce bit errors because the peak-to-peak
Because of the relationship between random jitter and data
amplitude lies above the receiver’s alignment jitter tolerance.
error rates, engineers also need to consider the number of
Due to the random jitter in Source B, the video system can-
independent jitter samples used in measuring peak-to-peak
not sustain a BER of 1 x 10-10.
jitter amplitude. In this section, we show that measurements
made over a small number of independent samples will not
assess the potential for infrequent data errors due to high-
amplitude random jitter.
5.1. Random jitter and BER
As an example, consider two SDI signal sources whose
outputs contain Gaussian, spectrally flat, random jitter.
Suppose that the jitter in the output from Source A has
RMS amplitude of 0.012 UI, while the jitter in Source B’s
output has higher RMS amplitude of 0.020 UI. The output
from either source can appear at the input of an SDI receiv-
er that can tolerate alignment jitter up to a peak-to-peak
amplitude value of 0.4 UI. Routing equipment between the
sources and this receiver adds approximately 0.2 UI peak-
to-peak of additional alignment jitter.
From Figure 29, we see that 1 x 103 independent samples
of the random jitter in Source A’s output would have a
peak-to-peak amplitude of approximately J = 6.5 · rms =
6.5 · 0.012 = 0.08 UI, while the peak-to-peak amplitude of
Primer
Generally, these differences arise from the random jitter in

the source output. As noted in section 2.6, deterministic jit-
ter has bounded peak-to-peak amplitude. If the peak-to-
peak amplitude of deterministic alignment jitter in a source
output lies below 0.2 UI, increasing the normalized meas-
urement times will not produce peak-to-peak amplitude
measurements above 0.2 UI for this deterministic jitter
component.
Figure 32. Bathtub curves for sources with random jitter at Random jitter does not have this bounded property. Even if
different RMS amplitude.
the random alignment jitter in a source output has a low
RMS amplitude, jitter amplitudes above 0.2 UI can occa-
Bathtub curves of SDI signal source outputs can help
sionally occur. Longer normalized measurement times will
determine the potential BER for video systems using these
detect these higher amplitude variations and yield a peak-
sources. Figure 32 illustrates this technique using sketches
to-peak amplitude measurement above 0.2 UI.
of Bathtub curves of the output signal from two SDI
sources, X and Y. Both output signals contain only The SMPTE standards do not specify how long the peak-
Gaussian-like random jitter. The random jitter in Source Y to-peak jitter amplitude in a source output must remain
has the larger RMS amplitude. below the specified limits, which creates ambiguity in
assessing source compliance. For example, consider the
In a video system with the SDI receiver and routing equip-
following statements about the output signal from Source Y
ment described above, transitions in the source output that
in Figure 32.
occur within ±0.3 UI of the Eye opening’s center point can
cause a bit error in the receiver. The Bathtub curve for Source Y would comply with a specification that said
Source X shows that less than 1 in 1012 transitions would the peak-to-peak alignment jitter amplitude in the output
enter this region in this source’s output signal. In the output signal must not exceed 0.2 UI except for 1 in 106
signal from Source Y, transitions would enter the region at a transitions.
rate of more than 1 in 107 transitions. In other words, with Source Y would not comply with a specification that said
Source X, the system BER lies below 1 x 10-12, while with the peak-to-peak alignment jitter amplitude in the output
Source Y the BER falls between 1 x 10-6 and 1 x 10-7. signal must not exceed 0.2 UI except for 1 in 1010
For an HD-SDI signal, 1012 bits corresponds to 11.2 min- transitions.
utes of video. So, a BER below 1 x 10-12 corresponds to a In this example, Source Y complies or does not comply
data error rate of less than 1 error in 11 minutes. On the with the SMPTE-specified limits on the peak-to-peak align-
other hand, 107 bits corresponds to 6.73 ms of video. A ment jitter amplitude, depending on the number of transi-
BER above 1 x 10-7 corresponds to a data error rate of tions considered in making the assessment. Since the
more than 1 error per video frame. standards do not specify this value, Source Y’s compliance
5.2. Jitter measurement and standards com- cannot be unambiguously determined.
pliance
For sources A and B described in the section 5.1, a peak-
to-peak amplitude measurement of alignment jitter made
over 1 x 103 independent jitter samples would indicate
that both sources comply with SMPTE specifications on
alignment jitter, i.e. less than 0.2 UI. A peak-to-peak ampli-
tude measurement of alignment jitter made over 1 x 107
independent jitter samples would also show that Source A
is in compliance, but would indicate that Source B is out of
compliance. Thus, differences in the number of independent
jitter samples in a measurement can lead to different
assessments about compliance.
Primer
Bathtub curves can be used to assess the relative compli-

ance of two sources. For the sources illustrated in Figure
32, the Bathtub curves indicate that Source X will stay in
compliance with SMPTE specifications on peak-to-peak
alignment jitter amplitude longer than Source Y.
However, without a specification on the number of transi-
tions that can exceed the specified limits, neither source
unambiguously complies with the standards. In particular,
the Bathtub curve indicates that Source X would not com-
ply with the specification if the peak-to-peak alignment jitter
amplitude in the output signal could not exceed 0.2 UI
except for 1 in 1014 transitions.
The Japanese video standard ARIB RT-B24 does give a
specification on data error rates in video systems. It speci-
fies that a video system should have an average data error Figure 33. BER vs. peak-to-peak jitter amplitude for a typical
rate of less than one error every 3 minutes. In 3 minutes, an video signal.
HD-SDI signal typically has more than 1 x 1011 transitions.

So, the ARIB specification corresponds to a BER below nent. The measured source output in this experiment had
1 x 10-11. peak-to-peak jitter amplitude equal to 0.19 UI when meas-
ured over 3 x 103 independent jitter samples. If peak-to-
Suppose that the two sources shown in Figure 32 were
peak jitter amplitude of 0.19 UI in a source output does not
HD-SDI sources and were used in video systems that
cause bit errors in a system, this source can ensure that bit
would have bit errors if the peak-to-peak jitter amplitude in
errors occur at a rate of less than 1 error in 3 x 103 bits.
a source output exceeded 0.2 UI. Then, systems that used
This corresponds to a BER of 1 / 3 x 103 = 3.3 x 10-4 as
Source X would comply with the ARIB standard, but sys-
shown on the plot.
tems that used Source Y would not comply.
For 1.6 x 104 independent jitter samples, the measured
5.3. BER and jitter measurement time
peak-to-peak amplitude equaled 0.21 UI. If jitter in the
Fortunately, verifying that signal sources can meet low BER source output can reach peak-to-peak amplitude of 0.21 UI
targets does not require making peak-to-peak amplitude before a bit error occurs in a system, then this source can
measurements over a very large number of independent jit- ensure a BER of 6.25 x 10-5 = 1 / 1.6 x 104 as shown.
ter samples. Various methods exist that use fewer inde-
The red line in the plot shows a theoretical curve for
pendent jitter samples collected over shorter observation
Gaussian random jitter that falls slightly to the right of the
times, to estimate the peak-to-peak jitter amplitude over a
measured peak-to-peak jitter amplitude values taken over
larger number of independent jitter samples.
different values of N independent jitter samples. This curve
Figure 33 illustrates one procedure for using peak-to-peak is essentially the same as the red line in Figure 29. The BER
amplitude measurements made over smaller normalized values on the vertical axis in Figure 33 correspond to the
measurement times to estimate the results of measure- normalized measurement times on the horizontal axis in
ments made with a large number of independent jitter sam- Figure 29. The horizontal axis in Figure 33 shows peak-to-
ples. This Mathcad plot shows the relationship between the peak amplitude measurements, not the J/rms ratio. Hence,
value of a peak-to-peak amplitude measurement and a the red line in Figure 33 shows the theoretical peak-to-peak
BER associated with the number of independent jitter sam- jitter amplitude of Gaussian random jitter with a particular
ples used in making the measurement. RMS amplitude of 0.028 UI (18.8 ps).
The blue diamonds on the graph show this relationship for We can use the curve in Figure 33 to make a “worst-case”
the measurements made in Appendix B. These measure- estimate of the peak-to-peak jitter amplitude in a source
ments were made on an HD SDI source output containing a that has random jitter in the output signal with RMS ampli-
random jitter component with an RMS amplitude value of tude of 0.028 UI. In particular, the curve shows that over
about 18.8 ps and a negligible deterministic jitter compo-
Primer
almost 1012 jitter samples, the worst-case peak-to-peak As a second example, suppose a facility adopted the ARIB
jitter amplitude in this signal equals 0.4 UI (black circle on requirements for acceptable data error rates. The engineer-
diagram). Suppose a video system that had to meet the ing staff confirmed that Source A has less than 0.2 UI and
ARIB specification included this source. This system would they included this source in the system. The SDI receivers
have to tolerate peak-to-peak jitter amplitude of around 0.4 in the system all have a jitter input tolerance around
UI in a source output. Otherwise, the random jitter in the 0.4 UI and the routing equipment does not add more than
output from this source, with this particular RMS amplitude 0.15 UI jitter.
of 0.028 UI, will create a BER above the ARIB specification. Now suppose a routine equipment replacement introduces
This procedure estimates BER values based on the a receiver with a jitter input tolerance of 0.35 UI and new
behavior of random jitter. Due to omnipresent thermal routing equipment that adds more than 0.15 UI of jitter.
noise, all SDI signals will have some level of random jitter, On average, the peak-to-peak jitter amplitude in Source
and this “unbounded” random jitter will determine the A’s output will exceed 0.16 UI once in every 1010 bits. This
peak-to-peak amplitude value measured over a large amplitude will generate data errors in the new system and
number of independent jitter samples, i.e. a long normalized the system BER will fall below the ARIB specification. Thus,
measurement time. measured peak-to-peak jitter amplitude values below 0.2 UI
For smaller normalized measurement times, bounded deter- in a source output do not ensure low data error rates
ministic jitter can determine the peak-to-peak amplitude. in all systems.
The normalized measurement time at which the random Currently, the video standards impose a conservative
component will begin to dominate the total peak-to-peak requirement on signal sources to ensure their output signals
amplitude value depends on the ratio of the RMS amplitude have a wide-open Eye most of the time. Since the source
of the random jitter to the peak-to-peak amplitude of the output does not completely determine the data error rate in
deterministic jitter. a video system, the current situation presents SDI source
So, the procedure described in this section must include designers with a substantial challenge. Without specifica-
an initial set of measurements to determine the maximum tions or guidance on acceptable data error rates, minimal
peak-to-peak amplitude of any deterministic jitter in the jitter input tolerance, and maximum jitter transfer, they can-
signal. This establishes a lower bound on the number of not determine how long the jitter in the source output must
independent jitter samples used for the measurements in remain below the specified thresholds.
the plot. SDI source designers need information on the jitter budget
5.4. Jitter budget for a video system, i.e. the data error rate and the jitter per-
formance each element must achieve to ensure acceptable
As noted in section 3.5, the SMPTE standards specify the
operation. This information will determine how long jitter in
allowed jitter in a source output. However, jitter in the
the source output must remain below specified thresholds.
source output does not completely determine the data error
rate in a video system. With an understanding of jitter budgets, the properties of
signal jitter, the jitter performance of the equipment installed
In the example used in section 5.1, Source B generated
in video systems, and the characteristics of various jitter
bit errors because the video equipment between Source B
measurement methods, engineers can more effectively and
and the SDI receiver added 0.2 UI of jitter. This pushed the
efficiently deploy jitter measurement instruments in equip-
signal jitter above the receiver’s jitter input tolerance.
ment design, system integration, equipment qualification
Now suppose that the video equipment between Source B and system maintenance.
and the receiver only added 0.1 UI of jitter. This system
would have a BER well below 1 x 10-10. Thus, measured
peak-to-peak jitter amplitude values above 0.2 UI in a
source output do not necessarily correlate with high data
error rates.
Primer
Figure 34. Phase Demodulation method in WFM700M.
6.0 Jitter Measurement with Tektronix Instruments

Tektronix offers several video and general-purpose instru- The ADC samples the demodulated jitter signal at a rate
ments that make automated peak-to-peak jitter amplitude greater than the 10 MS/s Nyquist rate for the 5 MHz
measurements. In this section, we use these products to bandwidth of the demodulated jitter signal. The final stage
illustrate particular implementations of each of the jitter measures the peak-to-peak value of these samples over
measurement methods examined in this guide. T = 500 ms. This measurement duration corresponds
6.1. Jitter measurement with the to a normalized measurement time of N = 5 x 106
Tektronix WFM700M independent samples.
The Tektronix WFM700M Waveform Monitor uses the 6.2. Jitter measurement with other Tektronix
Phase Demodulation method to automatically measure video instruments
peak-to-peak video jitter on HD and SD-SDI signals (Figure The WFM601M Waveform Monitor also uses the Phase
34). The instrument uses a commercially-available SDI Demodulation method with a 5 MHz clock recovery band-
receiver IC with a cable equalizer. Clock x in Figure 34 is the width to measure peak-to-peak jitter amplitude in equalized
clock output from the receiver IC. Thus, the WFM700M SD-SDI signals. The peak-to-peak measurement stage uses
measures the jitter in an equalized SDI signal that corre- analog peak detection.
sponds closely to the signal that SDI receivers decode. The VM700T Option 1S can measure jitter amplitude in
The phase detector generates a demodulated jitter signal equalized SD-SDI signals with either the Phase
in real-time. The instrument can display this jitter waveform Demodulation method or the Equivalent-time Eye method.
and can route the signal to an output BNC. Connecting The clock recovery bandwidth for the Phase Demodulation
this output to a spectrum analyzer will generate a jitter method is approximately 7 MHz, larger than the WFM601M
spectrum display. or WFM700M. Like the WFM601M, the peak-to-peak
The wide-band clock recovery circuit used in the SDI measurement stage uses analog peak detection. The
receiver IC has a 5 MHz bandwidth, which establishes the Equivalent-time Eye method samples the input SD-SDI sig-
high frequency cutoff of the measurement bandpass filter nal at approximately 3 MS/s. The VM700T can also meas-
(ƒ4). A measurement filter after the phase detector supple- ure wander in SD-SDI signals, and both wander and jitter in
ments the high-pass filtering of the narrow-band PLL band- analog video signals.
width. This filter establishes the low frequency cutoffs for
the bandpass filters used for measuring timing and align-
ment jitter either (ƒ1 or ƒ3). The bandpass slopes comply
with SMPTE recommendations and the wander rejection of
the timing jitter filter complies with IEEE Std. 1521.
Primer
Tektronix developed and introduced the WFM601M, VM700 Thus, implementations of jitter measurement methods
Option 1S, and WFM700M at different times. Consequently, that realize only a 40 dB/decade slope in the high-pass
they have different implementations of the jitter measure- characteristics of the timing jitter filter (e.g., the WFM601M)
ment methods and can give different results when measur- can overestimate the peak-to-peak jitter amplitude. In the
ing jitter on the same video signal. We can use the informa- presence of a wander component commonly seen in the
tion in the preceding sections to understand these differ- SDI output from an MPEG decoder, the WFM601M
ences. overestimated jitter amplitude by more than 500% (1.6 UI
6.2.1. Wander rejection compared to 0.3 UI). Implementations with a 60 db/decade
slope better reject the wander component. In this experi-
Tektronix launched the WFM601M before IEEE Std. 1521
ment, wander that passed through the measurement filters
was proposed. Hence, the implemented measurement filter
implemented in the WFM700M and VM700T Option 1S
has a 40 dB/decade high-pass response at 10Hz rather
contributed only about 33% to the timing jitter measure-
than the 60 dB/decade slope shown in Figure 13.
ment (0.4 UI compared to 0.3 UI).
Tektronix developed the VM700 Option 1S before IEEE
6.2.2. Measurement of random jitter
drafted Std. 1521, but the design anticipated this specifica-
tion. Hence, the Phase Demodulation method used in the These products use different technology in making the
VM700 Option 1S complies with the 60 dB/decade slope peak-to-peak measurements and consequently have differ-
required for wander rejection. The Equivalent-time Eye ent measurement times. As we have shown, measurement
method implemented in the VM700T has the 40 dB/decade time has a large effect on peak-to-peak jitter values for SDI
slope commonly used in forming equivalent-time Eye pat- signals with significant random jitter (Figure 29).
tern, which does not comply with IEEE Std. 1521 specifica- The Phase Demodulation method implemented in the
tion for wander rejection. WFM700M has the longest detection time (500 ms) with
To determine how differences in wander rejection affect SDI a digital, highly accurate peak-to-peak detector.
jitter measurement, measurements were made of the peak- Implementations of the Phase Demodulation method in
to-peak jitter amplitude in the SD-SDI output of an MPEG the VM700 Option 1S and WFM601M use analog peak
decoder with the WFM700M, WFM601M and the Phase detection technology with an effective measurement time
Demodulation method implemented in VM700 Option 1S. of approximately 2.5 ms.
The MPEG decoder received its input signal from a QPSK On an SDI signal with only random jitter, experiments have
satellite receiver tuned to a signal from a satellite DTV verified that the analog-peak detection used in the
provider. Due to MPEG data buffering, the SD-SDI output WFM601M and VM700T Option 1S led to peak-to-peak
commonly has a significant wander component. amplitude measurements approximately 30% lower than
Measurements made with the Wander application on the the digital peak detection in the WFM700M.
VM700T confirmed a frequency-offset variation in steps of The peak-to-peak amplitude measurements on the
1.75 ppm with a peak frequency-offset of 2.5 ppm. The SDI WFM601M and VM700T Option 1S were lower than the
output from the MPEG decoder complies with the 2.8 ppm WFM700M because they measured the random jitter over
frequency-offset limit for a studio-quality video reference but a short duration. They are all valid measurements of this
exceeds the 0.028 ppm/sec frequency drift-rate limit by 20 peak-to-peak amplitude since the standards do not specify
to 30 times. However, this SDI output would not normally a measurement time or the number of independent jitter
appear as a studio reference and consumer video equip- samples to use in the measurement.
ment can track this wander. Although all these instruments validly measure the peak-to-
We then measured the jitter on the SDI output of the MPEG peak jitter amplitude, the digital implementation in the
decoder. The WFM601M measured 1.6 UI of timing jitter WFM700M does offer an improved measurement compared
compared to the 0.4 UI measured with the Phase to the analog-based implementations in the earlier-genera-
Demodulation method on the VM700 Option 1S and tion instruments. The longer measurement time available
WFM700M. With the frequency offset removed, the meas- with the digital implementation can more consistently meas-
ured jitter drops to around 0.3 UI for all instruments. ure contributions from infrequent, high-amplitude random
jitter that can produce data errors in video systems.
Primer
6.2.3. Measurement of deterministic jitter Peak-to-peak jitter amplitude measurements made with
Some SDI test signal generators can introduce sinusoidal these software applications can vary from measurements
jitter at different amplitudes and frequencies into their SDI made with the video-specific instruments due to differences
outputs. These three video instruments will measure similar in several areas.
peak-to-peak amplitudes for this sinusoidal jitter, although The video-specific products have built-in support for the
variations can occur due to differences in frequency SMPTE-specified measurement bandpass restrictions.
response, the amount of random jitter in the signal, and Realizing these measurement filters in the TDSRT-EYE
differences in jitter noise floor compensation. and TDSJIT3 application requires proper configuration
Larger differences can arise with rapidly varying or “spike- of clock recovery and filtering algorithms. Differences
like” deterministic jitter. If the deterministic jitter has rapid in measurement filtering can generate different
variations, the WFM601M and VM700T Option 1S may give measurement results.
lower results than the WFM700M because the attack time Accurately measuring jitter frequencies down to 10 Hz
in the analog implementation of the phase detector may not or widely-separated jitter peaks requires an adequate
fully track these variations. With narrow, separated jitter number of multiple acquisitions on a real-time oscillo-
peaks or intermittent jitter, the Equivalent-time Eye method scope. Using a smaller number of acquisitions in
in the VM700T may give lower results than either the these measurements could lead to differences in
WFM700M or the WFM601M because the equivalent-time measurement results.
sampling used in forming the Eye may not sample the true TDSRT-EYE and TDSJIT3 software can measure jitter
peaks during the measurement period. frequencies up to and beyond the SMPTE specification
Differences can also arise if a signal has high-frequency of 1/10 clock rate. Depending on the actual high fre-
deterministic jitter. A peak-to-peak jitter amplitude measure- quency jitter components in the signal, they can produce
ment made with the Equivalent-time Eye method can higher peak-to-peak jitter amplitude measurements
include contributions from any spectral component within compared to measurements made with the Phase
the measurement bandpass. Measurements made with the Demodulation method in the video-specific instruments.
Phase Demodulation method do not include contributions Both these software applications can compute jitter sta-
from spectral components beyond the demodulated jitter tistics over a specified population of TIE measurements.
signal’s bandwidth. With appropriately-sized populations of TIE measure-
As a result: ments that account for the effect of random jitter, these
A peak-to-peak amplitude measurement made with the applications can produce peak-to-peak amplitude
Equivalent-time Eye method on the VM700T may have a measurements similar to the video-specific instruments.
larger value than measurements made on the same sig- Differing numbers of independent jitter samples can lead
nal with the WFM601M and WFM700M if a signal’s jitter to different measurement results.
spectrum contains spectral components above 5 MHz. The video-specific instruments use hardware-based
A peak-to-peak amplitude measurement made with the equalization and clock recovery and have a higher jitter
Equivalent-time Eye method on the VM700T Option 1S noise floor. This can produce noticeable differences
may have a larger value than a measurement made on in jitter measurement made over short cables. For
the same signal and instrument with the Phase measurements made over long cables, the equalization
Demodulation method if a signal’s jitter spectrum con- stage in the video-specific instruments excludes jitter
tains spectral components above 7 MHz. related to cable attenuation. To yield comparable results,
these software applications would need to similarly
6.3. Jitter measurement with Tektronix
exclude this jitter component.
real-time oscilloscopes
Tektronix real-time oscilloscopes implement versions of
the Real-time Acquisition method in application software
for jitter measurement. The TDSRT-EYE® and TDSJIT3
software measure peak-to-peak jitter amplitude from TIE
measurements based on a recovered reference clock.
Primer
7.0 Recommendations for Measuring Jitter in SDI Signals
As noted in section 5.0, we cannot determine the jitter in Comparing measurements from both methods can help
a video system and its impact on system operation by determine a signal’s jitter characteristics. Typically, measure-
making a single peak-to-peak amplitude measurement with ments made with the Equivalent-time Eye method use
one jitter measurement method. Assessing jitter perform- fewer independent jitter samples than measurements made
ance in video systems requires the effective use of multiple with the Phase Demodulation method. Thus, if both meth-
jitter measurement methods and techniques. ods measured only random jitter, the Phase Demodulation
In this section, we make a few recommendations for method would usually produce a larger measurement.
measuring jitter in three applications: So, if measurements made with the two methods agree,
Video system monitoring, maintenance and the signal most likely has highly regular deterministic jitter
troubleshooting (e.g., sinusoidal jitter) and a small amount of random jitter.
If the Equivalent-time Eye method consistently produces
Video equipment qualification and installation
the larger measurement, the signal contains regular, high
Video equipment design frequency deterministic jitter. If the Phase Demodulation
In these recommendations, we show how the jitter meas- method produces a much larger value, the signal contains
urements made with different methods can give additional some combination of narrow jitter spikes, intermittent
insight into the jitter characteristics of a video system. deterministic jitter or a significant level of random jitter.
7.1. Video system monitoring, maintenance 7.2. Video equipment qualification
and troubleshooting and installation
Assessing jitter in a video system and diagnosing jitter- Measuring jitter with more than one method also helps
related problems may require measuring jitter with both in qualifying and installing new video equipment.
the Phase Demodulation and Equivalent-time Eye methods. The benefits of the Phase Demodulation method identified
Since the Phase Demodulation method continuously moni- above for video system monitoring, maintenance, and trou-
tors a demodulated jitter signal, it can detect and generate bleshooting also apply in this application. With continuous,
alarms on a wide range of signal jitter, including peak-to- Nyquist-rate sampling of the demodulated jitter signal, this
peak jitter amplitude greater than 1 UI. By including an method can detect and log the effects of narrow jitter
equalization stage, instruments can monitor jitter throughout spikes, intermittent deterministic jitter, or significant levels
a video system and detect jitter in the equalized signal the of random jitter.
receiver decodes. By sampling the demodulated jitter signal at the Nyquist
Typical implementations of the Phase Demodulation method rate, this method can collect a large number of independent
measure jitter over several frames of the video signal. This jitter samples quickly. These long normalized measurement
can capture raster-dependent deterministic jitter related to times can help assess the impact of random jitter on data
line and field rates and can catch sporadic jitter. The jitter error rates.
waveform and spectrum displays available with this method The jitter signal output available in some implementations
can help better characterize and diagnose jitter-related of the Phase Demodulation method complements the
problems. internally-generated jitter signal displays. Routing this output
Measurements made with the Equivalent-time Eye method to an oscilloscope or spectrum analyzer can reveal more
can complement measurements made by the Phase details on the temporal and frequency characteristics of
Demodulation method. In particular, the Equivalent-time Eye signal jitter.
method can measure the peak-to-peak amplitude of deter- An equalization stage also helps in equipment installation.
ministic jitter at frequencies above the bandwidth of the When this stage is present, installers and engineers can
clock recovery process used to implement the Phase measure jitter at the end of long cables. With this capability,
Demodulation method. This high-frequency jitter may not they can evaluate jitter at the receiver’s input as well as the
propagate through a video system (see section 2.5), but it source output. This helps detect and diagnose problems
may affect individual link operation. due to jitter introduced by equipment or connections
Primer
between the source and receiver, e.g., patch panels and With an implementation of the Phase Demodulation
non-reclocking distribution amplifiers. method, designers can functionally verify a design and
The jitter noise floor in implementations of the Phase make initial assessments of jitter performance. Most of the
Demodulation method may be larger than implementation capabilities applicable to monitoring jitter in video systems,
of other methods, primarily due to contributions from qualifying and installing video equipment, and diagnosing
equalization and clock recovery. If the jitter in a source jitter-related problems apply in this design application. Using
output signal is close to the SMPTE-specified limits, the a common measurement method across these varied
peak-to-peak amplitude measurement may not have the applications can help correlate design parameters with
resolution needed to separate sources during a precision behaviors in actual video systems. The jitter waveform
qualification process. display of the demodulation jitter signal available with this
method is especially useful for detecting raster-correlated
In this case, peak-to-peak amplitude measurements made
deterministic jitter.
with an implementation of the Real-time Acquisition method
can complement measurements made with the Phase An implementation of the Real-time Acquisition method
Demodulation method. These implementations can have offers additional measurement precision and in-depth analy-
a very low noise floor. Measurements made on a single sis capabilities that help characterize jitter behaviors. The
acquisition do not have contributions from equalization, stored acquisition record that contains a highly-sampled
clock recovery, or trigger jitter. They offer more than ade- version of the input SDI signal offers unique benefits.
quate resolution for precision screening of video equipment Designers can isolate and examine individual edges in the
or components. SDI signal and correlate jitter behaviors with particular data
patterns.
Implementations of the Real-time Acquisition method can
also measure contributions from high-frequency determinis- Both methods can collect a large number of independent
tic jitter in a source output that the Phase Demodulation jitter samples over reasonably short measurement times.
method would miss. By making a TIE measurement at each This supports a thorough examination of random jitter and
edge, the method samples this deterministic jitter above the its impact on data error rates. The Phase Demodulation
Nyquist rate. method can continuously monitor the demodulated jitter
signal to detect jitter spikes or other intermittent determinis-
With multiple acquisitions, the Real-time Acquisition method
tic jitter. Finally, both methods measure peak-to-peak jitter
can collect a large population of independent jitter samples.
amplitudes greater than 1 UI.
Peak-to-peak amplitude measurements of random jitter
made with this method can complement measurements The combined capabilities of these two jitter measurement
made using the Phase Demodulation method. These imple- methods offer the breadth and depth needed in video
mentations also offer a variety of jitter displays and analysis equipment design applications.
algorithms that can help in qualifying video equipment.
In many cases, general-purpose measurement instruments
that use equivalent-time sampling can substitute for an
implementation of the Real-time Acquisition method in this
application. Typical implementations of the Equivalent-time
Eye method in video-specific equipment do not have similar
capabilities and can only partially substitute for these more
powerful instruments.
7.3. Video equipment design
Jitter measurements with the Real-time Acquisition and the
Phase Demodulation methods best address needs in video
equipment design.
Primer
8.0 Conclusion
In this technical guide, we described the three commonly

used methods for automated jitter measurement and
how differences in these methods can generate dissimilar
peak-to-peak jitter amplitude measurements. Variation
in the number of independent jitter samples used in the
peak-to-peak measurement was a common contributor
to this divergence. Several other factors can bring about
discrepancies and in some cases can generate significant
differences.
We explored the complex nature of jitter in digital video
signals and the challenges involved in measuring jitter. By
gaining a better understanding of different jitter characteris-
tics and the key elements of jitter measurement, engineers
can more quickly resolve problems with signal jitter, and
more fully use the varied jitter measurement and analysis
solutions.
Standards bodies play a key role in evolving best
practices in jitter measurement. Additional specification
and guidance on jitter budgets and jitter measurement
techniques are needed to ensure greater consistency in
measurement results.
Additional information from video equipment manufacturers
on jitter input tolerance and jitter transfer will also help in
designing and qualifying video equipment that ensures
low data error rates in video systems.
Tektronix remains committed to delivering high quality tools
for monitoring and measuring jitter in video systems and
to working with standards bodies, industry groups, video
equipment manufacturers, and video network operators to
address the recommendations made in this technical guide.
Primer
9.0 References
1. SMPTE RP 184-1996, Specification of Jitter in Bit-Serial
Digital Systems
2. SMPTE RP 192-2003, Jitter Measurement Procedures
in Bit-Serial Digital Interfaces
3. SMPTE EG 33-1998, Jitter Characteristics and
Measurements
4. ANSI/SMPTE 259M-1997, 10-Bit 4:2:2 Component and
4fsc Composite Digital Signals — Serial Digital Interface
5. SMPTE 292M-1998, Bit-Serial Digital Interface for
High-Definition Television Systems
6. IEEE Std 1521™-2003, IEEE Trial-Use Standard for
Measurement of Video Jitter and Wander
7. Understanding and Characterizing Timing Jitter,
Tektronix Primer, 2003
8. Takeo Eguchi, Pathological Check Codes for Serial
Digital Interface Systems, SMPTE Journal, August, 1992,
pages 553-558
10.0 Acknowledgement
Tektronix wishes to gratefully acknowledge the extensive
contribution of Dr. Dan H. Wolaver in the development of
this technical guide.
Primer
Figure A-1. Test setup for comparing jitter measurement methods. Figure A-2. CSA803A histogram window.
Appendix A: Impact of Bandwidth Limitation in Video Jitter

Measurement
In the Phase Demodulation method, the bandwidth of the We used the Tektronix CSA803A oscilloscope to measure
clock recovery process used in the implementation deter- jitter using the Equivalent-time Eye method. We configured
mines the upper limit for jitter frequencies included in the the CSA803A to determine the peak-to-peak jitter
peak-to-peak jitter amplitude measurement. The bandwidth amplitude using a histogram of 5 x 106 edge variation
of commonly used clock recovery hardware falls well below measurements collected within the histogram window
the 1/10 clock rate high-frequency cutoff that SMPTE spec- shown in Figure A-2. The reading from the CSA803A
ifies for measuring timing and alignment jitter. In particular, is designated x1.
the Phase Demodulation method implemented in the
Tektronix WFM700M has a 5 MHz high-frequency cutoff.
This bandwidth limitation will not significantly affect meas-
urement results if the bandpass covers most of the jitter’s
spectral content. In this appendix, we describe an experi-
ment to test the assertion that the video jitter spectrum
in “typical” video signals has negligible content above
common bandwidths for clock recovery.
Figure A-1 shows the test setup used for the experiment.
We used the internally-generated color bar test pattern from
an evaluation board for a commercially-available video IC
as the test signal containing “typical” jitter. An analysis of
the output test pattern from the evaluation board indicated
that this signal contained predominately random jitter. The
evaluation board also provided a 74.25-MHz crystal clock
for triggering the oscilloscope.
We used the WFM700M to measure jitter with the Phase
Demodulation method. The WFM700M determined the
peak-to-peak amplitude of 5 x 106 independent jitter sam-
ples. For comparison, we made separate measurements
with two WFM700Ms. The peak-to-peak jitter measure-
ments are designated y1 and y2.
Primer
Equipment raw measurement intrinsic jitter final measurement

CSA803A x1 = 178 ps p-p x0 = 45 ps p-p xact = 172 ps p-p
WFM700M #1 y1 = 164 ps p-p y1 = 164 ps p-p
WFM700M #2 y2 = 170 ps p-p y2 = 170 ps p-p
Table A-1. Peak-to-peak jitter amplitude measurements.
Table A-1 shows the peak-to-peak jitter amplitudes meas-

ured by the three instruments. The three raw measurements
of peak-to-peak jitter amplitude shown in this table agree
within 8.5%.
Some of this difference in measurement values is due to
compensation for internal jitter. The WFM700M compen-
sates for internal jitter. The reading from the CSA803A
Figure A-3. Test setup for calibrating the intrinsic noise x0 of the
includes uncompensated jitter noise, primarily trigger jitter,
CSA803A.
which contributes to the higher x1 reading. This internal
jitter in the CSA803A is also essentially random jitter, and Thus:
is uncorrelated with the predominately random jitter in the
test pattern output.
Figure A-3 shows the setup used to measure the peak-to-
peak amplitude, x0, of this internal jitter in the CSA803A.
In order to compensate for this internal jitter (x0) in the x1
measurement made in this experiment, we viewed the
measured jitter as the sum of two uncorrelated random jitter
components, the actual random jitter in the test signal and We apply this relationship between the average peak-
the internal random jitter from the CSA803A. In this case, to-peak jitter amplitude measurements to the values we
the RMS value (standard deviation) of the measured signal, measured in this experiment6, i.e.
σ 1, equals the root sum of squares of the actual RMS value
of the random jitter in test signal, σ act, and the RMS value Using this equation to compensate for internal jitter on the
of the random internal jitter in the CSA803A, σ 0. CSA803A, the final values xact, y1, and y2 agree within 3.5%.
The agreement among the measured values indicates that
For random jitter, the average value of the peak-to-peak the test output from the evaluation board does not have
jitter amplitude (J) and the RMS value ( σ ) are related by the significant jitter above the 5 MHz loop bandwidth of the
equation J = N σ , where N is the normalized measurement clock recovery process used in the WFM700M.
time (section 4.5.3). We ran this experiment on a particular signal that we
considered representative of typical video signals. Others
could perform the same experiment on signals of interest
to determine if the bandwidth limitation in the WFM700M
affects the results of a peak-to-peak amplitude
measurement.
6In general, the peak-to-peak amplitudes of jitter sources do not add as root sums of squares. Using this approach to compensate for the internal jitter in the CSA803A pro-
duces a reasonable result in this case because the two jitter sources are uncorrelated with predominately random jitter.
Primer
Appendix B: Peak-to-Peak and RMS Measurement of Typical

Video Jitter
N Jrms (ps) Jp-p (ps) σp-p (ps) Jp-p/Jrms σp-p/Jp-p

3x 103 18.8 129 5.3 6.867 4.1%
1.6 x 104 18.8 142 4.5 7.534 3.2%
5x 105 18.8 168 3.9 8.937 2.3%
1.3 x 107 18.8 196 — 10.396 —
Table B-1. Video jitter measurement versus jitter sample number N.
Figure B-1. Test setup for measuring “typical” video jitter.
To illustrate the relationship between peak-to-peak Table B-1 gives the RMS jitter amplitude Jrms, the average
amplitude measurements for purely random jitter and peak-to-peak jitter amplitude Jp-p, and the standard devia-
those for “typical” video jitter, we measured jitter on the tion σ p-p of the peak-to-peak jitter amplitude for each N.
output signal from an evaluation board for a commercially- Figure 29 shows a plot of the Jp-p/Jrms and Figure 30
available video IC. shows a plot of σ p-p/Jp-p.
Figure B-1 shows the measurement setup. We set the SDI
signal output to a 1080i/60 color bar output and measured
the output Eye diagram on a Tektronix CSA803A oscillo-
scope triggered on the 74.25-MHz crystal clock from the
evaluation board. We determined that the output had negli-
gible jitter components below 1 kHz and so did not need
to implement any high-pass filtering.
We measured the RMS and peak-to-peak jitter amplitudes
using the Equivalent-time Eye method (see Figure 21) with
four different histograms containing N = 3 x 103,
N = 1.6 x 104, N = 5 x 105, and N = 1.3 x 107 measure-
ments of edge variation. For each of the first three values of
N, we made 16 measurements to find the standard
deviation between measurements. Because of the meas-
urement time required for N = 1.3 x 107, we only made
one measurement.
Primer
Figure C-1. Data pattern d and corresponding pattern-dependent jitter ƒout .
Appendix C: Limits to Clock Recovery Bandwidth

SMPTE 259M and SMPTE 292M specify that the high In general, we can approximate the peak-to-peak value
frequency cutoff of the bandpass filters used to measure of this data-dependent jitter by the formula:
timing and alignment jitter should be at least 1/10 of the
data clock rate. This specification corresponds to a jitter
measurement bandwidth of 148.5 MHz for HD-SDI signals Where:
and 27 MHz for SD-SDI signals.
BW = the loop bandwidth of the clock recovery PLL
In theory, the Phase Demodulation method could realize this
ƒb = SDI input signal bit rate
cutoff if the clock recovery process used in the measure-
ment had a bandwidth of one-tenth the bit rate. In practice, Nmax = the maximum period without transitions
however, such a high bandwidth leads to unacceptable ƒoff = the phase detector offset
phase noise in the recovered clock. For a loop bandwidth of one-tenth the bit rate, this
Unavoidable offset in the phase detector of the phase- formula becomes
locked loop (PLL) used in clock recovery causes this noise. øoutpp = 0.67·Nmax·ƒoff,
The PLL uses the signal data transitions to apply a non-
In Figure C-1, the longest period without data transition
zero phase at the input to cancel this offset voltage on
is Nmax = 14 UI, making the peak-to-peak jitter in this
average. In lock, then, the phase detector output has an
example equal to 0.132 UI.
average voltage of zero.
An offset of øoff = 0.015 UI is close to the lowest value we
However, when the data pattern has no transitions (a long
could reasonably expect in a well-designed clock recovery
sequence of either ‘0’s or ‘1’s), the phase detector output
circuit. Further, these circuits should allow for periods of
equals the offset voltage. This offset voltage causes a fre-
Nmax = 20 without data transitions. For BW = 0.1·ƒb these
quency offset at the PLL output, corresponding to a ramp
parameters result in øoutpp = 0.2 UI.
of the output phase øout lasting as long as the period with
no transitions. This generates data-dependent jitter that Hence, we can see that the SMPTE requirement for the
corrupts the jitter measurement. high frequency cutoff of the timing and alignment jitter
filters will lead to an excessive amount of internally
Figure C-1 illustrates this data-dependent jitter. In this
generated clock recovery jitter in the Phase Demodulation
example, the clock recovery bandwidth is one-tenth the bit
method, especially when trying to verify conformance
rate and the phase detector has a 0.015 UI offset. Note
to a 0.2 UI limit.
that the phase ramps upward when the signal has no
transitions, and it descends exponentially when the signal
has transitions.
Primer
From the general formula, we see that the peak-to-peak

jitter amplitude decreases with decreasing loop bandwidth
of the clock recovery PLL. At BW/ƒb = 0.0036 the general
formula becomes øoutpp ≈ 0.05·Nmax·øoff, which gives
øoutpp = 0.015 UI for øoff = 0.015 UI and Nmax = 20.
We can accept this small amount of internally generated
data-dependent jitter when measuring against a 0.2 UI
acceptance limit.
In the Phase Demodulation method implemented in the
Tektronix WFM700M, the clock recovery PLL has a 5 MHz
loop bandwidth. For 1.485 Gb/s HD-SDI signals, this
corresponds to BW/ƒb = 0.0034. Hence, this approach
will generate less than 0.015 UI of internal jitter during long
periods of no data transitions. For 270 Mb/s SD-SDI
signals, BW/ƒb = 0.0185 and this component of internal
jitter will fall below 0.05 UI.
Contact Tektronix:
ASEAN / Australasia / Pakistan (65) 6356 3900
Austria +41 52 675 3777
Balkan, Israel, South Africa and other ISE Countries +41 52 675 3777
Belgium 07 81 60166
Brazil & South America 55 (11) 3741-8360
Canada 1 (800) 661-5625
Central East Europe, Ukraine and the Baltics +41 52 675 3777
Central Europe & Greece +41 52 675 3777
Denmark +45 80 88 1401
Finland +41 52 675 3777
France & North Africa +33 (0) 1 69 86 81 81
Germany +49 (221) 94 77 400
Hong Kong (852) 2585-6688
India (91) 80-22275577
Italy +39 (02) 25086 1
Japan 81 (3) 6714-3010
Luxembourg +44 (0) 1344 392400
Mexico, Central America & Caribbean 52 (55) 56666-333
Middle East, Asia and North Africa +41 52 675 3777
The Netherlands 090 02 021797
Norway 800 16098
People’s Republic of China 86 (10) 6235 1230
Poland +41 52 675 3777
Portugal 80 08 12370
Republic of Korea 82 (2) 528-5299
Russia & CIS 7 095 775 1064
South Africa +27 11 254 8360
Spain (+34) 901 988 054
Sweden 020 08 80371
Switzerland +41 52 675 3777
Taiwan 886 (2) 2722-9622
United Kingdom & Eire +44 (0) 1344 392400
USA 1 (800) 426-2200
For other areas contact Tektronix, Inc. at: 1 (503) 627-7111
Updated 15 June 2005
For Further Information

Tektronix maintains a comprehensive, constantly expanding collection of
application notes, technical briefs and other resources to help engineers
working on the cutting edge of technology. Please visit www.tektronix.com
Copyright © 2005, Tektronix, Inc. All rights reserved. Tektronix products are covered by U.S. and foreign
patents, issued and pending. Information in this publication supersedes that in all previously
published material. Specification and price change privileges reserved. TEKTRONIX and TEK are
registered trademarks of Tektronix, Inc. All other trade names referenced are the service marks,
trademarks or registered trademarks of their respective companies.
10/05 EA/WOW 2BW-18906-0

Understanding Jitter Measurements 2BW-18906-0 - CPC06

Uploaded by

Copyright:

Available Formats

Understanding Jitter Measurements 2BW-18906-0 - CPC06

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Understanding Jitter Measurements 2BW-18906-0 - CPC06

Uploaded by

Copyright:

Available Formats

Understanding

Jitter Measurement for Serial Digital Video Signals

A Tektronix Video Primer

2.0 Fundamental Concepts and

2.2 Decoding process, clock recovery,

HD-SDI signal a variation of 340 ps is more than one-half

tude measurement made over a long observation time

Once initiated, the stress patterns in pathological SDI sig-

Figure 6a shows a segment of an AC-coupled signal

While equalization can compensate for cable effects, the

Figure 10. AC-coupling effects.

Many instruments, including waveform monitors and other

Figure 11. Bathtub curve – BER vs. location in Eye Diagram.

3.0 Specifications on Video Jitter Performance and Measurement

Document Title Content description

Table 1. Standards and other documents that apply to video jitter.

on jitter from “the serial output of a source derived from a

3.4. Specifications on connecting cables and

3.3. Specifications on signal voltage levels

3.5. Specifications on peak-to-peak jitter

Specification Signal format

Highpass Corner frequency 10 Hz 10 Hz

Highpass Corner frequency 1 kHz 100 kHz

Table 2. Summary of jitter specifications.

4.0 The Functions Comprising Jitter Measurement

Using a non-optimal decision threshold in transition detec-

Using the optimal decision threshold in transition detection

4.3. Phase detection/demodulation

collecting individual amplitude measurements of the

Operation and frequency response

Dynamic range Manual measurements

Figure 22. Phase Demodulation method.

4.3.2. Phase detection/demodulation: Phase Transition detection and dynamic range

floor from the measured peak-to-peak jitter amplitude (see

Figure 24. Real-time Acquisition method.

The jitter waveform and spectrum displays contain informa-

Phase Detection/Demodulation Method

Sampling of Style Equivalent-time Real-time Real-time

Coverage Depending on observation Continuously monitors Full coverage within acqui-

Table 3. Summary of jitter specifications.

4.3.4. Phase detection/demodulation: Summary

For this analysis, let the s-domain representation of the

4.4.2. Filter accuracy

To see why, consider two signals, A and B, where signal

4.5.4. Measurement times Thus, peak-to-peak amplitude measurements made over

Method Key Characteristics

Table 4. Comparison of jitter measurement methods.

5.0 Data Error Rates and Jitter Measurements

Generally, these differences arise from the random jitter in

Bathtub curves can be used to assess the relative compli-

HD-SDI signal typically has more than 1 x 1011 transitions.

Figure 34. Phase Demodulation method in WFM700M.

6.0 Jitter Measurement with Tektronix Instruments

7.0 Recommendations for Measuring Jitter in SDI Signals

In this technical guide, we described the three commonly

Appendix A: Impact of Bandwidth Limitation in Video Jitter

Equipment raw measurement intrinsic jitter final measurement

Table A-1. Peak-to-peak jitter amplitude measurements.

Table A-1 shows the peak-to-peak jitter amplitudes meas-