Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper

TUPM5-5

Correction of Time Domain Measurement Data of Vector

Network Analyzers

Bertalan Eged, Assistant Professor (C-EGED@...)

Péter Bajor, Postgraduate Student (PETER@...)
István Novák, Associate Professor, SIEEE (T_NOVAK@...)
Department of Microwave Telecommunications, Technical University of Budapest
1111 Budapest, Goldmann tér 3, HUNGARY
Phone/FAX: +36-1-181-2968, E-mail: ...NOV.MHT.BME.HU

Abstract - Vector Network Analyzers (VNA) measure tested circuits in the frequency domain but their
frequency range does not include DC. When VNAs are used for time-domain measurements, the proper
transformation requires the measured DC value as well. Under certain conditions extrapolation of the DC
response can be incorrect, what causes incorrect time domain response, too. This paper presents a solution
to this problem.

I. INTRODUCTION

Time-Domain Reflectometry (TDR) and Time-Domain Transmission (TDT) measurements are often
necessary to characterize the electrical performance of components and systems. Time-domain
measurements can be easily performed in the time domain by connecting a known waveform to the input of
the Device Under Test (DUT) and by measuring its response by an oscilloscope. With the advent of more
sophisticated instruments with post-processing capability, the measurement of time-domain response
through frequency-domain data and Inverse Fourier Transform has become possible (see Fig. 1). For TDR
measurements, instead of measuring the sum of reflected and incident waveforms in the time domain, the
reflected wave as a function of frequency is measured, while the amplitude of incident wave is constant.

Assuming continuous sampling of measured values, the time-domain impulse response can be expressed
as:

k (t ) = F −1[Γin (ω )] (1)

Time-domain TDR instruments usually perform step-response measurements:

h(t ) = ∫ k (t )dt (2)

t
The integral can be performed in the time or frequency domain:

h(t ) = ∫ F −1[Γin (ω )]dt (3)

t

⎧1 ⎫
h(t ) = F −1 ⎨ Γin (ω )⎬ (4)
⎩ω ⎭
The above equivalency is valid only for linear systems; therefore measurement of heavily nonlinear (e.g.
active) devices in the frequency domain will generate new problems.
Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper
TUPM5-5

a
DUT VNA DUT
b

Γin = b
a

Fig. 1: Block schematic of TDR measurements in the time and frequency domain.

Modern instruments work with sampled signals. The measured waveform is digitized at discrete intervals,
and one frame (N pieces) of samples is processed. Finite sample number and discretized time or frequency
points will result in granularity and periodicity of the data.

N
1
k (t ) = ∑
2 N + 1 n=− N
Γ(2πnΔf ) e j 2π n Δf t (5)

N
Γ(2π n Δf ) e j 2π nΔf t
h(t ) = Γ(ω = 0) t + ∑
n=− N j 2πn
(6)
n ≠0

With finite-bandwidth VNA measurements, the equivalent time-domain data will also be limited. The rise-
time of the equivalent input stimulus is determined by the upper band-limit of VNA. Consequently, the
spatial resolution of the measurement is also limited. In some applications the task is to locate
discontinuities along transmission lines. For such applications, resolution of the TDR measurements can be
increased by suitable postprocessing [3], [4].

The time-span of the equivalent TDR response is determined by the lower frequency limit of the VNA data.

Even though measuring TDR response through frequency-domain data may seem very attractive, a
significant drawback of the method stems from the fact that typical vector network analyzers cannot
measure the frequency response at zero frequency. In (5) and (6) this means that the data block to be
transformed will miss the zero-frequency elements. The missing element would determine the constant
term in the k(t) function, hence the average slope of the step response. If extrapolation of the DC response
from the measured values gives the proper value, the resulting step response is correct. If, however, the
extrapolation of DC response is incorrect, the resulting TDR response will also be incorrect, as it is
illustrated in Fig. 2. The figure shows the measured TDR impulse and step responses of a 3-meter long
coaxial cable. For both measurements, the instrument setting was exactly the same, except the upper
frequency limit was 818 MHz, and 820 MHz, respectively.

The instrument selects the measuring frequencies as:

m
stop _ freq m = 1...N
N
Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper
TUPM5-5

RFL Re 200 mU/REF 0 U RFL Re 200 mU/REF 0 U

-43.869 μ U -37.827 mU

20 ns 20 ns

START 0 s STOP 50 ns START 0 s STOP 50 ns

Fig. 2: Correct and incorrect TDR responses shown by an HP 8752A Vector Network Analyzer. For both
cases the DUT was a 3-meter long 50-ohm coax cable with open termination, measured at 51 frequency
points. (a, left) Correct TDR response is obtained with a stop frequency of 818 MHz. (b, right) Incorrect
TDR response is obtained with a stop frequency of 820 MHz.

II. DESCRIPTION OF PROBLEM

Let us examine a simple arrangement. If the DUT is a lossless transmission line with open termination at
its end, we know that the magnitude of the voltage reflection coefficient at the input is |Γ|=1. The phase is
directly proportional to the frequency. Although the VNA cannot measure at DC, we know that Γ(ω=0) =
+1 (assuming that the input DC resistance of the DUT is infinite). The instrument must extrapolate this
value from the measured lower frequency data. We obtain the correct TDR result only if this extrapolation
is successful. As it is shown in Fig. 2, increasing the stop frequency just slightly can significantly
deteriorate the step response. What is its reason? Let us have a closer look at the frequency response in Fig.
3. Measurement (b) differs from measurement (a) only in the frequency points, all other settings and the
DUT itself was the same. The VNA needs harmonically related frequency-domain data points to generate
time domain data, therefore the start frequency must be equal to the stop frequency divided by the number
of points. This is why increasing the stop frequency (and leaving the number of points unchanged) will
increase the start frequency. The start frequency at the same time equals the frequency increment, Δf.
Because of the increased start frequency, in case of (b), the electrical length of the transmission line
becomes greater than the quarter wavelength corresponding to the start frequency. Note that in Fig. 3., the
input impedance at the start frequency is capacitive for case (a), and inductive for case (b), as shown by the
marker readouts on the graphs.

The figure shows the measured frequency-domain data for the same 3-meter long coax in exactly the same
setting of VNA that was used to generate Fig. 2.
Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper
TUPM5-5

RFL 1 U FS RFL 1 U FS

16.039 215 MHz 2 16.078 431 MHz 2

1 1
0 0

956.15mOhm 924.44mOhm
106.19nF 1.0497nH
START 16.039 215 MHz START 16.078 431 MHz
STOP 817.999 965 MHz STOP 819.999 981 MHz

Fig. 3: Illustration of wrong extrapolation of the zero-frequency data point. (a, left) If the difference
between the phases of the zero-frequency element (which is 0 or π) and first measured element is less than
π, the extrapolation is correct for Γ(ω=0), see Fig. 2a. (b, right) If the phase difference is greater than π,
the extrapolation will fail on the sign of Γ(ω=0), see Fig. 2b. On both graphs 3 points are labeled; 0
denotes the (missing) zero-frequency data, 1 and 2 denote the first and second measured values.

III. SUGGESTED CORRECTION ALGORITHM

Based on the investigation of several test situations, it was realized that the VNA properly extrapolates the
magnitude of the missing zero-frequency response, but under certain conditions fails on the extrapolation of
phase.

In the proposed procedure, first the measured voltage-reflection coefficient data is read from the
instrument. It is than transformed into the time domain by Inverse Fourier Transform, substituting zero
value to the zero-frequency response. In this way we obtain an approximation of the impulse response. The
step response can be obtained if before the transformation we divide Γ(nω) by (j2πn) for every n = 1..N.

To smooth the time domain result, windowing may be applied either in the frequency or in the time
domain. Because we have data only at discrete frequency points, the responses in the time domain will be
periodical with a period of 1/Δf. This means that a time domain data at a small negative time τ is equal to
the data at the end of the time window minus τ. If this window is wide enough (Δf is small enough) so that
the transients of the impulse response die out before the end of the interval, we can determine the DC
response so that the impulse response at τ will be zero. If we cannot be sure that the impulse response is
sufficiently close to zero at τ, we can calculate the DC response from the frequency domain data. Absolute
value and phase also have to be calculated, but it is known that the phase is either 0 or π.

If we could read the extrapolated DC value, it would be simple to produce the corrected impulse and step
responses, but only measured data is readable from the VNA. Therefore we have to calculate the time
domain response with the DC-response equals zero assumption, at those time points where the instrument
does its calculation, and the DC value can be calculated from the difference between the two results. The
main steps of the computation are as follows:

Read the step response from the VNA to the vector Tm (windowing must be turned off). Since the number
of time domain points is the same as the number of frequency domain points, m runs from 0 to (N-1). The
corresponding tm time points within the block Tm are:
Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper
TUPM5-5

stop _ time − start _ time

t m = start _ time + m (7)
N −1

Read the complex frequency response data block into the complex vector Γn (n = 1..N). Compute the step
response:

N
⎧ Γ e j 2π nΔ f tm ⎫
Tm* = ∑ Im⎨ n ⎬ (8)
n =1 ⎩ nπ ⎭

Now we can calculate the zero-frequency response Γ0:

( )
N −1 N −1
2 ∑ Tm* − Tm − ( N − 1)∑ m(Tm* − Tm )
Γ0 = 6 m =0 m =1
(9)
N ( N − 1) Δf Δt
2

where Δt is the time step between tm and tm+1. We have to know a τ time where the step response can be
assumed to be zero. Calculate (8) with τ

N
⎧ Γn e j 2π nΔf τ ⎫
T *
(τ ) = ∑ Im⎨ ⎬ (10)
n =1 ⎩ nπ ⎭

And now we can tell the k(t) impulse response and h(t) step response at any t time:

{ }
N
Γ0 + 2∑ Re Γn e j 2π nΔf t
k (t ) = n =1
(11)
1 + 2N
N
⎧ Γn e j 2π nΔf t ⎫
h(t ) = ∑ Im⎨ ⎬ − T(τ ) + Δf Γ0 (t − τ )
*
(12)
n =1 ⎩ nπ ⎭
Manuscript for the IEEE Instrumentation and Measurement Technology Conference, May 10-12, 1994, Shizuoka, Japan, paper
TUPM5-5

RFL Re &M 200mU/ REF 0U

START 0 s STOP 50 ns

Fig 4: Incorrect (lower trace) and corrected (upper trace) TDR step response. Compare with Fig. 2.

IV. MEASUREMENT RESULTS

The corrected and uncorrected impulse responses differ only in a DC shift, which is very small: Γ0/(2N+1).
However, the integral of this small DC value can cause a significant slope error in the step response as we
see in Fig. 2. It is possible to reload the corrected data into the analyzer and compare with the incorrect one
stored in the memory of the instrument. Fig. 4 shows the result of the correction algorithm on the simple
test arrangement of Fig. 2.

V. CONCLUSIONS
We have shown how the VNA can be used for time domain measurements. If the lowest measurement
frequency is too high, the extrapolated DC response and time domain response will be incorrect. A simple
method was introduced to correct this error.

ACKNOWLEDGEMENT
This research project has been partially funded by the National Committee for Technical Development
(OMFB) in connection with the COST project #229, and by the National Fund for Scientific Research
(OTKA) under project T007233, and by the Fund for the Hungarian Science.

REFERENCES
[1] Norbert Hesselmann, Digitale Signalverabeitung. Würzburg, Germany: Vogel-Verlag, 1983.
[2] L. Schnell, Technology of Electrical Measurements. England: John Wiley & Sons Ltd., 1993.
[3] Z. A. Maricevic, T. K .Sarkar, Y. Hua, A. R. Djordjevic, “Time-Domain Measurements with the
Hewlett-Packard Network Analyzer HP 8510 Using the Matrix Pencil Method,” IEEE
Transactions on Microwave Theory and Techniques, vol. 39, No. 3, pp. 538-547, March 1991.
[4] Y. Hua, T. K. Sarkar, “Generalized pencil-of-function method for extracting poles of an EM
system from its transient response,” IEEE Trans. Antennas Propagat., vol. 37, pp. 229-234, Feb.
1989.
[5] “TDR Tools in Modeling Interconnets and Packages,” Tektronix Product Sheet, 1993.
[6] Alberto Biondello, Viscardo Costa, “A High-Performance Enviroment for Modelling and
Simulation of Digital Systems,” Hewlett Packard Digital Design Seminar, 1993.