Rxqual and Voice Quality

a Technical White Paper Series
RXQUAL and
Voice Quality
RXQUAL and Voice Quality
Ascom White Paper Series Issue No. 102/96
© Copyright 1999 by Ascom Infrasys AG
All rights reserved

The information contained herein is for the personal use of the reader
Library of Congress Catalog-in-Publication Data
Brechtmann, Christian
ISBN 3-9521195-0-4
All inquiries and requests for titles of other White Papers in the Networking Series
should be addressed to the publisher
Ascom Infrasys AG
P.O. Box
CH-4503 Solothurn
Switzerland
Phone +41 32 624 21 21
Fax +41 32 624 21 43
E-mail qvoice@infrasys.ascom.ch
http://www.ascom.ch/qos
Limits of Liability and Disclaimer of Warranty

The author and publisher of this White Paper have used their best efforts in preparing the
material presented and make no warranty with regard to its content.
Price: US$ 20.00
2nd revised Edition, 1999

Printed in Switzerland
Ascom Infrasys AG Ascom Technical White Paper Series No. 103/98
Contents
Introduction ......................................................................... 5
QoS Criteria and Speech Quality ......................................... 7
Assessment of speech quality ............................................................... 8
Subjective versus measured speech quality ........................................... 9
Overview of GSM Speech Transmission .......................... 11
and Radio Channel Signal Processing
Speech transmission .......................................................................... 11
Bit error rate measurements in GSM ................................................... 12
Simulation system .............................................................................. 13
Radio channel signal processing ......................................................... 14
Argument No. 1: ................................................................ 16
RXQUAL Does Not Consider the Varying Efficiency
of Interleaving and Bit Error Correction Under
Different Environmental Conditions
The benefits of interleaving and channel coding ................................. 16
Fast mobile phone ............................................................................. 16
Slow mobile phone ............................................................................ 17
Test Results ........................................................................................ 18
The benefit of frequency hopping ...................................................... 20
Fading environment ........................................................................... 20
Argument No. 2: ................................................................ 22
RXQUAL Does Not Consider Quality Degradation
Caused by Stolen Speech Frames
Argument No. 3: ................................................................ 23
RXQUAL Does Not Detect Echo or Other PSTN
Quality Impairing Effects
Argument No. 4: ................................................................ 25
Speech Quality is the Only Parameter to Detect
Defects Within Voice Processing Circuits
1
Argument No. 5: ................................................................ 26

RXQUAL Interpretation Suffers from
Coarse Quantisation
Argument No. 6: ................................................................ 27
RXQUAL Does Not Recognise Whether Bad
Speech Frame Indication Failed
Argument No. 7: ................................................................ 28
RXQUAL Does Not Correctly Describe Speech Quality
after Activating Frequency Hopping
Background on frequency planning .................................................... 28
Interference on individual channels .................................................... 29
Frequency hopping creates interferer diversity .................................... 30
Measuring the benefits of frequency hopping – case study ................. 31
Quality testing system in BSS parameter tuning .................................. 33
Argument No. 8: ................................................................ 34
In Case of EHR Speech Codec Introduction
RXQUALEFR and RXQUALEHR Would Have a
Different Mapping of Speech Quality
Argument No. 9: ................................................................ 35
Speech Quality is the Only Parameter to Compare
Networks of Different Technologies
2
Argument No. 10: .............................................................. 36

Downlink RXQUAL May Fail in Detecting
QoS Degradation in Uplink Direction
System related causes unbalanced power budget ............................... 37
Different equalisation depths within base station and mobile phone ... 38
Causes related to interference on the radio channel ........................... 39
Different levels of radio transmission activity ....................................... 39
Intermodulation or receiver blocking conditions .................................. 40
Third party unauthorised use of the allocated GSM band .................... 41
Spurious emissions ............................................................................. 41
Receiver blocking ............................................................................... 42
Intermodulation ................................................................................. 43
Different radio link topology in uplink and downlink direction ............. 44
The near-far effect ............................................................................. 45
Propagation characteristics ................................................................. 45
Different locations of Rayleigh fades .................................................. 46
Conclusions ....................................................................................... 46
Argument No. 11: .............................................................. 47
RXQUAL Does Not Consider Acoustical
Power Level Variations
Importance of Speech Assessment in QoS ........................ 48
QVoice Introduction ........................................................... 49
References .......................................................................... 52
3
4
Introduction
It has only been a few years since GSM technology found its way from devel-
opment laboratories into the field of commercial service. The complexity and
multiple variations of all operational situations in the real world could never
be simulated and tested in the laboratory, therefore many bugs and system
configuration problems had to be discovered and resolved in the field. Con-
sequently, the accepted “culture” of performing GSM field measurements
slowly evolved, over a period of time, into a suite of test protocols using
standard engineering and trouble shooting tools. This evolution still contin-
ues and is still being driven by the needs of technicians and engineers who
are responsible for network implementation and operational stability.
Accordingly, the traditional and common approach used to investigate GSM
radio link control procedures is with protocol analysers and first generation
cellular verification systems. These engineering and trouble shooting tools
extract diagnostic information from the GSM system, concentrating on
parameters like RXQUAL, RXLEV and layer 3 signalling messages. This data is
useful in network maintenance but is limited when addressing the important
issue of Quality of Service (QoS) – particularly as viewed by the mobile phone
user.
The subscriber’s perception of “quality” – specifically voice quality – is still
rarely taken into consideration. The primary reason being that radio techni-
cians believe that a “human interpretation” of quality would not be a relia-
ble, objective and reproducible measurement.
However, the subject of QoS – especially in cellular networks – is now being
recognised as “critical to success” by more and more network operators. In
today’s competitive environment, it is important for GSM operators to be
able to accurately assess performance and voice quality of their networks in
order to maintain or extend market position.
5
From a subscriber’s point of view, the relevant performance measure is the

call success rate and received speech quality. These two very important QoS
criteria have, however, been difficult to assess, quantify and analyse. The
simpler approach, the RXQUAL or bit error rate (BER) method to assess speech
quality has been preferred for ease of access and economic reasons. BER is
readily available on the downlink – and it is commonly believed to correspond
well to speech quality1. This idea – stemming from the limited agreement of
BER and speech quality at very low and very high BER values – has generated
some serious misunderstandings.
RXQUAL, unfortunately, does not represent “real life” conditions. This
White Paper discusses inherent problems of RXQUAL and demonstrates
that the measurement cannot be related to perceived speech quality
in an accurate and reliable way.
6
QoS Criteria and Speech Quality
The subject of QoS in cellular networks is very broad and covers a wide variety
of topics. For our purposes, it can best be defined by two groups of criteria:
Call Procedure and Speech Quality.
Call procedure related QoS criteria:

Before call attempt:
• probability of no network service availability
If service available:
• probability of call attempt failure (uplink + downlink)
If call attempt successful:
• call set-up duration (uplink + downlink)
• probability for call drop
Speech quality related QoS criteria:

If call held successfully until desired release:
• actual perceived speech quality over call duration (uplink)
• actual perceived speech quality over call duration (downlink)
• statistical distribution of different quality levels (e.g. percentage of call
duration suffering from lost syllables, echo, robot-voice, ping-pong)
It is commonly believed that call procedure and speech quality can be derived
from engineering measurements. The call procedure related items would simply
be taken from the signalling messages while the more subtle criteria of speech
quality is derived from the downlink RXQUAL (this is also suggested by the
name of this parameter).
7
Experienced engineers – responsible for quality assurance of cellular net-

works – know how difficult it is to categorise a call attempt into classes of call
success by correctly interpreting signalling messages. The major problem in
this regard is that the majority of available tools do not provide suitable post-
processing capabilities. And, call success analysis is only the beginning of
understanding QoS issues. A more serious matter, arises when downlink
RXQUAL is used to draw conclusions about speech quality. Results are
always ambiguous and can be very misleading. This White Paper will
demonstrate the necessity of real speech quality measurement and explain in
detail why downlink RXQUAL is not the suitable parameter for this purpose.
For a better understanding of the various points presented, some basic infor-
mation about methods of speech quality measurement, speech transmission
in GSM and the meaning of the RXQUAL parameter is provided.
Assessment of It is extremely difficult to define a term as vague as speech quality; everybody

speech quality has a different view of the subject. A pragmatic approach is as follows:
If many people assess the “quality” of a particular speech sample, the
average of all votes is taken as a measure of the speech quality of that
sample.
This average is called the mean opinion score (MOS). In fact, it is a reasonable
performance measure for GSM: as speech quality has a different meaning for
every mobile phone user, the MOS (ideally) reflects the average perception of
quality among the subscribers. ITU-T recommends 5 levels of quality:
“excellent” – “good” – “fair” – “poor” – “bad” and describes a procedural
approach to determine MOS by listening tests under laboratory conditions.
These tests had been applied in the selection process for the best suitable
GSM speech codec.
8
Subjective versus In practice – in an operational network – it is not feasible to assess speech

measured speech quality quality at different locations by listening tests because of the enormous effort
that such tests would require. Instead, one resorts to automated measuring
methods that aim to approximate the MOS. There exists a wide variety of
schemes, ranging from simply measuring the bit error rate to modelling the
human hearing system. As a rule of thumb, more complex schemes yield
more accurate results.
Any reasonable measurement method will agree well with speech quality if
there are no transmission errors (quality must be good) and if there are many
transmission errors (quality must be bad). However, the dividing line between
acceptable and unacceptable speech quality lies somewhere in between.
Therefore, it is important to employ a measurement method that meets these
two requirements:
1. Yields accurate measurements in the intermediate quality range
2. Does not suffer from systematic errors.
The first requirement is illustrated in Figure 1. Scheme A yields accurate meas-
urements for all levels of speech quality, whereas scheme B can deviate heavily
from the target value at intermediate speech quality. Only scheme A is reliable
around the level of “just acceptable” speech quality.
The second requirement is equally important. For example, a systematic error
could result in always overestimating speech quality under specific circum-
stances; as a consequence, this would prevent detection of quality problems
in that case.
9
best
Speech quality derived from measurement
Scheme A
Scheme B
bad
Figure1
Accurate (A) and inaccurate (B)
measurement patterns of speech quality bad Subjective speech quality excellent
Tests with the QVoice 3) system demonstrate that speech quality measure-
ment by automatic procedures provide sufficient accuracy to fully substitute
listening tests in operational cellular networks or similar applications.
10
Overview of GSM Speech Transmission

and Radio Channel Signal Processing
Speech transmission The GSM speech transmission path is sketched in Figure 2. First, speech is
compressed to (approximately) 20% of the input data. The resulting data
stream is then protected by an error-control coding scheme. On the radio
transmission path, various sources of errors can disturb the transmitted data.
At the receiver, the channel decoder attempts to recover from these errors
and delivers a “cleaned up” version of the received data. Finally, speech is
reconstructed in the speech decompression block.
Speech
Transmission
Speech compression Data reduction
Channel coding Error protection
Errors Radio channel
Reception
Channel decoding Error recovery
Speech decompression Speech reconstruction
Figure 2:
GSM speech transmission Speech
11
Bit error rate Two points are noteworthy. First, the speech compression algorithm is lossy.
measurements in GSM Consequently, some degradation of speech quality is unavoidable – even
without any transmission errors. Second, no error-control scheme can cope
with all possible error events. Bad transmission conditions usually result in
unrecoverable errors, which, in turn, cause a degradation in speech quality.
This White Paper considers only degradation effects resulting from transmis-
sion errors.
In GSM, bit error rate (BER) measurements are used for two purposes. On the
one hand, they help to decide whether transmitter power should be changed
(power control). On the other hand, BER assists in deciding whether a call
should be attached to another base station (handover). In both cases, BER is
only one among several criteria; moreover, moderately precise measurements
suffice for both purposes.
A single BER measurement in GSM extends over a period of approximately
0.5 seconds and is reported as one of eight quality levels (RXQUAL_0...7).
RXQUAL – estimated by backward coding of the decoded bit sequence and
comparing it to the received bit sequence – is a measure of the raw bit error
rate, and does not take into consideration channel coding.
12
Simulation system Several of the arguments, particularly 1, 4 and 5 were proven by intensive
simulation studies of the GSM transmission path 2).
The actual system used included the following modules:
• Speech compression (GSM 06.10)
• Channel coding and interleaving (GSM 05.03)
• A simplified flat fading channel model
• Channel decoding and de-interleaving (GSM 05.03)
• Bad frame substitution (GSM 06.12)
• Speech decompression (GSM 06.10)
The speech samples used were produced under a wide variety of conditions.
13
Radio channel signal For better understanding the following arguments, the most important steps
processing of the radio signal processing and transmission sequence has been outlined
in Table A. It is not intended to be a complete listing but to show the “influ-
ential” segments in terms of QoS discussions. Table A is the backbone to
several of the arguments and will be referred to throughout the White Paper.
Processing Entity Transmitted Information Remarks
1 Human speaker or QVoice Original voice Start of signal

processing QVoice
reference sample
2 Transmission system and Voice coded in 64 Kbit /s Potential source of

switches (ISDN and PSTN) A-law echo and distortion
3 Voice activity detection and Voice coded in 64 Kbit /s Potential source of

discontinous transmission A-law or Silence indication voice clipping
4 Speech coder Speech frame:
20 ms duration, 260 bits
182 class 1 bits
78 class 2 bits
Transmission
5 Convolutional coding Protected speech frame:

(only for class 1 and 3 20 ms duration, 456 bits
parity bits) 378 class 1 bits (protected)
78 class 2 bits (unprotected)
6 Interleaving Partitioning of speech

frame: 8 subsequent bits
are spread into 8 different
time-slots (each 57 bits)
7 Burst formatter 2 x 57 data bits

26 bit training sequence
guard bits
8 GMSK modulator Original GMSK modulated

baseband signal
9 RF mixer Original GMSK modulated

Table A: - fixed carrier frequency radio signal
Processing sequence of a voice signal - frequency hopping
trough a GSM network
14
Processing Entity Transmitted Information Remarks
10 Radio channel Distorted

- path loss GMSK modulated
Radio Channel
- slow lognormal fading Radio channel

- multipath propagation
- fast rayleigh fading
- intersymbol interference
- noise
- interference
- doppler shift
11 Receiver front end Distorted

- antenna diversity reception GMSK modulated
- single carrier of FH Basband signal
12 Receiver: 8 received bursts: Benefiting from

- adaptive matched filter 26 bit distorted training antenna and path
and equaliser sequence 2 x 57 distorted diversity
data bits
13 De-interleaver Received protected

speech frame:
20 ms duration, 456 bits
378 class 1 bits (protected)
78 class 2 bits (unprotected)
14 Channel decoder Received speech frame: RXQUAL estimation

error correction 20 ms duration, 260 bits different efficiency
Reception
(only on 378 class 1 bits) 182 class 1 bits of error correction

78 class 2 bits depending on
CRC parity check result speed, fading
pattern, frequency
hopping
15 If bad frame indication Reconstructed BFI may fail with a

based on CRC test voice in 64 Kbit /s 12.5% probability
or stealing flag for due to 3 bit CRC
FACCH use check creating
or otherwise invalid ping-pong sound
voice data, speech
frame substituion
else if silence indication
flag comfort noise
generation
else speech decoder
16 Human ear or QVoice 3) Evaluator of

speech quality
15
Argument No. 1:
RXQUAL Does Not Consider the
Varying Efficiency of Interleaving and
Bit Error Correction Under Different
Environmental Conditions.
The benefits of As shown in step 4 of Table A, the speech coder converts segments of 20 ms
interleaving and of speech into 182 class 1 bits and 78 class 2 bits – a total of 260 bits.
Class 1 bits are more important for reliable coding of speech. Therefore, only
channel coding class 1 bits are protected by 2:1 convolutional coding. See step 5 of Table A.
Class 2 bits are left unprotected because they have only minor significance
for speech quality. This yields 456 coded bits for every 20 ms of speech.
Within the interleaver – step 6 of Table A – subsequent bits are separated so
that they are transmitted within 8 different time-slots. This calculates to be
up to 40 ms in delayed transmission.
This procedure is necessary in order to handle short term Rayleigh fading on
the radio channel. This is step 10 of Table A. The process, however, leaves
some interesting effects demanding careful consideration, such as shown in
the following examples which compare the efficiency of interleaving in com-
bination with error correction coding of class 1 bits for different speeds of the
mobile phone.
Fast mobile phone Figure 3 shows a typical stationary Rayleigh fading pattern that is caused by
multi path propagation such as in urban areas.
If a signal breakdown lasts 10 ms then every 4 th bit of the original non-
interleaved bit sequence may be damaged. This can be repaired efficiently
and effectively by the 2:1 redundancy of the channel coding taking place in
step 14. Assuming that a deep fading notch has a width of λ /5 = 6 cm (being
the wavelength of the 900 MHz RF signal), then a speed of 120 km/h would
be sufficient to reduce the time of fading to 1.8 ms. This means that the
speech reference although damaged during a signal breakdown would arrive
in good condition at the receiving end, step 16 of Table A.
16
Slow mobile phone A more critical situation arises when the mobile phone is stationary or mov-
ing very slowly such as pedestrian speed, approximately 7 km/h. In this case
deep fading of 6 cm width would last 30 ms – or in other terms nearly the full
interleaving width. Under these conditions, GSM signal processing would
generate an uninterrupted sequence of damaged bits and the error correc-
tion taking place in step 14 of Table A would fail. Therefore, class 1 bits would
suffer from fading and the transmitted speech would arrive in a degraded
state at the receiving end, step 16 of Table A.
The crucial question with respect to the theme of this White Paper is: Does
RXQUAL report different speech qualities for the two situations described
above? The answer is no.
RXQUAL is completely insensitive to the differences of these realistic situa-
tions of fast and slow moving mobile phones. This is because the measure-
ment is only an estimate of the raw bit error ratio and therefore complete
ignores the benefits of channel coding and interleaving as depicted in steps
5, 6, 13 and 14 of Table A. RXQUAL is the estimate for the BER of unprotected
class 2 bits rather than for class 1 bits.
17
Test Results The effect of the Rayleigh fading channel is illustrated in Figure 3 and Figure
4 for the Rayleigh fading channel. An English speech sample was transmitted
(by simulation) to a slow mobile (7.2 km/h, 900 MHz) and to a fast-moving
mobile (120 km/h, 900 MHz).
(best) 0
1
Figure 3: Rayleigh fading, fast mobile (120 km /h, 900 MHz)
2
RXQUAL
RXQUAL vs. time [s] for Rayleigh fading Speech quality: GOOD to FAIR
for two mobiles of different speed. Note 3
that RXQUAL of the slow mobile is at least 4
as good most of time.
5
Sample of fast moving mobile typical 6
of a vehicle installed phone (worst) 7
0 1 2 3 4 5
Time [sec]
(best) 0
1
Rayleigh fading, slow mobile (7.2 km /h, 900 MHz)
2
RXQUAL
Speech quality: FAIR to POOR

3
4
5
Sample of slow moving mobile typical 6
of a hand-held pedestrian phone (worst) 7
0 1 2 3 4 5
Time [sec]
18
Power
λ = 30 cm
width of fades < λ

λ
λ
5 interleaving depth
at 120 km/h
Figure 4: at 7.2 km/h
Stationary fading pattern
0 1 2 3 4 5 ... meters
30 60 90 120 150 ... ms at v = 120 km/h
500 1000 1500 2000 2500 ... ms at v = 7.2 km/h
Fades last much longer for the slow mobile, thus resulting in much longer
error bursts. In spite of that, the raw BER measurements described by RXQUAL
are very similar due to their averaging effect; the slow mobile RXQUAL was
as good as or better than that of the fast mobile most of the time. Neverthe-
less, speech quality was clearly better for the fast mobile.
The conclusion from this analysis is that RXQUAL measurements can not take
into account the speed of a mobile phone that has a significant impact on
speech quality.
19
The benefit of Frequency hopping is the GSM technology to combat Rayleigh fading and to
frequency hopping mitigate co-channel and adjacent channel interference.
Fading environment Frequency hopping would be effective when applied within the fading sce-
nario described above for a slow moving mobile phone. The reason for this is
that frequencies which are sufficiently spaced do not generate correlated
fading patterns. Figure 4 shows a typical fading pattern. This means that
statistical characteristics – like average field strength and field strength prob-
ability distribution – would be identical while the individual location of fading
is not correlated. Switching, therefore, from one frequency to another shifts
fading notches when mobile phones are slowly moving or stationary. The
result for signal processing efficiency would then be equivalent to the fast
moving mobile example presented earlier.
Figure 5 shows BER (class 1) for a slow moving mobile – with 5 km/h for 4
different cases:
a) without error correction (class 2, RXQUAL), FH applied
b) without error correction (class 2, RXQUAL), no FH
c) with error correction (class 1), FH applied
d) with error correction (class 1), no FH
Cases c) and d) of Figure 5 show that frequency hopping reduces class 1 BER
by a factor of more than 10. Cases a) and b) of Figure 5, in comparison,
cannot be separated (same curve).
One can conclude that RXQUAL cannot distinguish the obvious bene-
fits of frequency hopping when considering voice quality in fading
environments.
20
10 0
BER
10 -1
a) without error correction, FH

b) without error correction, no FH
10 -2
Figure 5: c) with error correction, no FH

Depending of BER (class) on Frequency
Hopping and Error Correction Coding
d) with error correction, FH
flat fading without diversity
v = 5 km/h
10 -3
Source: Lucent Technologies 0 5 10 15
Nürnberg, Germany Eb/No [dB]
21
Argument No. 2:
RXQUAL Does Not Consider
Quality Degradation Caused
by Stolen Speech Frames
Stealing speech frames is a common procedure in GSM when fast signalling

is required during the call period.
Normally, the traffic channel for full rate speech (TCH/FR) is accompanied by
a slow associated control channel (SACCH) which is used to transfer non-
time critical information. SACCH has a recurrence interval of 480 ms and an
interleaving delay of 360 ms. It is used to transfer system information mes-
sages in a regular way in downlink or to transfer MS measurement result
messages in uplink.
However, if handover is required and executed, an immediate response is
necessary and therefore the traffic channel capacity is used. This is achieved
by marking a speech frame of 20 ms duration as stolen (i.e. invalid for speech)
and inserting signalling messages instead. The flag is called Stealing Flag and
the stolen TCH becomes FACCH (fast associated control channel).
When a speech frame is lost in a controlled way (setting the stealing flag), it
is replaced by another type of speech data – mostly a repetition of the last
valid speech frame, which is muted more and more for subsequent stolen
frames. This occurs at step 15 of Table A).
Fast signalling is a required standard event during a call at each handover
execution. Interruption of the valid speech transmission for a successful hand-
over is about 100 to 200 ms, actual time being dependent mostly on how
well the old and new cells are synchronised. Handover failures could how-
ever cause significant speech interruptions and should be monitored.
The actual degradation due to stolen frames and the efficiency of speech
frame substitution can be monitored with advanced quality testing and anal-
ysis systems such as QVoice 3).
RXQUAL is not a suitable test parameter in determining quality effects
of erased frames. It does not consider stolen speech frames nor result-
ing speech quality degradation in any way.
22
Argument No. 3:
RXQUAL Does Not Detect Echo or
Other PSTN Quality Impairing Effects
BER measurements only take into account errors on the radio transmission
path (Figure 2). But the transmitted bits have no meaning to the measure-
ment unit – they are just bits. Thus, distortions prior to the speech compres-
sion unit remain undetected by BER measurements. This includes echo, cross-
talk and noise in the analogue part of the network.
Echoes originate within the analogue part of a network during speech trans-
mission. A fixed network telephone, for example, might have a design flaw
or construction error that allows an acoustical feedback loop to form from
the receiver back into the transmitting end.
Similar effects occur in the electrical circuitry of Public Switched Telephone
Network (PSTN) switches or analogue transmission devices. Within pure an-
alogue networks these effects do not cause echo, in the sense of human
perception, because the signals are travelling over wires or airways without
built-in delays like buffering or coding. The resulting time delay between the
main signal and the echo is too short for human perception.
When a PSTN is connected to a digital GSM network the situation is different.
GSM has several processing steps where signals are buffered and transmitted
with certain time delays. Speech coding which is taking place in step 4 of
Table A and interleaving, at step 6 of Table A causes a total one way delay of
38 ms. That means it takes 38 ms after a message is spoken to reach the
PSTN network. The path of the acoustical echo and the way back into the
GSM mobile phone – steps 13 and 15 of Table A – will add up to another
delay of at least 38 ms. Delays of this order of magnitude are perceived as
annoying echo and turn out to be very irritating to the speaker as well as the
listener on the receiving end of the link.
23
As an example, consider the case where a speech signal is severely impaired

by crosstalk in the wired part of the network. Referring to Figure 2, the dis-
torted speech signal would be compressed, error protection added, and the
resulting bit stream transmitted. At the receiving end, BER measurements
would only detect errors on the radio link (i.e. in the bit stream), but not
distortions of the input speech signal. The speech signal at the output side of
Figure 2 would remain impaired by crosstalk (poor speech quality), even if the
radio link transmission was perfect – and thus RXQUAL also measuring per-
fect.
The conclusion is that RXQUAL cannot detect echo or other disturbing
PSTN effects. It is important, however, for an operator to continuously mon-
itor the network for these effects before a high voice quality can be assured.
24
Argument No. 4:
Speech Quality is the Only
Parameter to Detect Defects
Within Voice Processing Circuits
Argument No 3 discussed several effects of the PSTN network which might

degrade speech quality and which are not detectable by surveying the radio
path which is provided by the RXQUAL parameter. Under certain circum-
stances the non-radio related GSM network infrastructure may also turn out
to be a source of speech quality degradation and even outage of the speech
in uplink, downlink or both. The following functional units have to be consid-
ered:
• Voice activity detection and discontinuous transmission
(step 3 of Table A)
• Voice coding done by TRAU (transcoder / rate adaptation unit)
(step 4 of Table A)
• Voice decoding done by TRAU
(step 15 of Table A)
The TRAU is placed between BSC and MSC and therefore not within the
scope of RXQUAL measurements. In case of malfunction one of the proc-
essor boards, assigned to a given traffic channel, the speech quality,
can be heavily degraded. This may happen without notice of the op-
eration and maintenance centre because self-testing capabilities of
individual boards are not always implemented.
25
Argument No. 5:
RXQUAL Interpretation Suffers
from Coarse Quantisation
In GSM, the BER measurements are quantified to eight RXQUAL levels. The
coarse quantification makes it possible to have different speech qualities with
different BERs that are quantified to the same RXQUAL level. This means that
it would be impossible to distinguish different levels of speech quality from
looking at just RXQUAL measurements.
An example is shown in Figure 6, in which an English speech sample was
transmitted (by simulation) over two AWGN (Additive White Gaussian Noise)
channels with different Bit Error Rates (different BERs can arise for several
reasons such as different distances of the mobiles to the base station). As
both BER values belong to the same RXQUAL level – the result of the coarse
quantification algorithm – the RXQUAL measurements are identical. In spite
of that, speech quality on the channel with the lower BER was clearly better.
Therefore, coarse RXQUAL quantification may yield ambiguities in
speech quality.
(best) 0
1
2 AWGN channel, lower end of RXQUAL level (low BER)
RXQUAL
Speech quality: GOOD

3
4
5
6
(worst) 7
0 1 2 3 4 5
Time [sec]
(best) 0
1
AWGN channel, upper end of RXQUAL level (high BER)
2
RXQUAL
Speech quality: FAIR to POOR

3
4
5
Figure 6: 6
RXQUAL vs time(s) for AWGN (worst) 7
with different BER, but identical RXQUAL 0 1 2 3 4 5
Time [sec]
26
Argument No. 6:
RXQUAL Does Not Recognise Whether
Bad Speech Frame Indication Failed
Every GSM user has been annoyed by the characteristic sound which resem-
bles a bouncing Ping-Pong ball. Sometimes, this noise is quite useful because
it is an early announcement before loosing the connection. However, the
designers of GSM did not really plan to include such noise in the system. The
“Ping-Pong” sound is caused by bad speech frames which the GSM system
itself is unable to detect.
The designers of GSM included a 3 bit cyclic redundancy (parity) check at step
14 in Table A. The design goal was to recognise bad speech frames in order
to avoid the creation of sounds which are generated when feeding the speech
decoder with spurious bit sequences. The recognition is only based on a 3 bit
CRC check; and, if the test reports a failure the Bad Frame Indication flag is
set. As a consequence of a BFI flag set, a substitution is made, replacing the
bad frame with the last valid frame. In cases of multiple subsequent BFI the
acoustical signal is muted gradually to suppress annoying or strange sounds.
After a valid frame is detected muting is cancelled.
Because only 3 bits have been reserved for BFI the probability for a non-
detected bad frame and the potential creation of “dropping bottle” sound is
12.5%.
RXQUAL is not suitable to distinguish “Ping-Pong” sound from mod-
erate degradation of speech quality caused by frame substitution. By
contrast, state-of-the-art quality testing and analysis systems can detect this
and other annoying effects. Being able to accurately detect these quality
degrading effects is of particular importance because future half rate speech
coders – or improved full rate coders – will most probable be more stable
against erroneous sound – and a poor RXQUAL does not necessarily imply
presence of artificial sounds.
27
Argument No. 7:
RXQUAL Does Not Correctly Describe
Speech Quality after Activating
Frequency Hopping
Operators who decide to operate in switch on mode of frequency hopping

are seeking improved speech quality for following purposes:
• To combat fading caused by Rayleigh fading for slow moving mobiles
• Allow mitigation of co-channel or adjacent channel interference for
creation of additional capacity
The first item was treated extensively in Argument 1 The second is treated
herein:
Background on The frequency spectrum is an extremely valuable resource. One fact derived
frequency planning from the US auctions was that for every US citizen under radio coverage, a
license for mobile operation carried a value of 20 to 50 US dollars. Network
planning engineers, not surprisingly, seek to optimise this resource by reusing
frequencies within short distances in order to obtain the maximum number
of frequency channels per cell. This leads to a non-diminishing probability of
mutual interference Pint.
The Pint function states that for a given location and time – within a service
area of a cell – the ratio of the wanted signal strength (C) to the sum (I) of
all interfering signals from the same or adjacent channels is lower than the
system threshold.
In GSM, the system threshold for a “good” and “realistic practical” quality is
defined as the following:
(C/I)thr = 9 dB
The planning goal for a network is usually in the order of:
Pint = 1%
During peak traffic periods, a network with lower levels of interference can
only be realised with exponentially increasing costs.
28
Interference on Frequencies are specifically assigned to different cells according to traffic re-
individual channels quirements and not according to fixed frequency groups of equal size. The
reason being is that cellular traffic density – as measured in Erlang/km2 can
vary greatly between urban and ruraI areas – and the number of TRX per cell
has a typical range between 1 and 6. A consequence of this – “allocation by
cell traffic need” – is that each frequency in a given cell may find its co-
channel and adjacent channel interferers in different cells with individual
propagation path losses.
Furthermore, the actual activity and strength of individual interferers is de-
pendent on random variables indicating the following:
• Whether the interfering TRX (on the relevant time-slots) has a traffic
channel TCH assigned
• In case of TCH being assigned – whether it is currently in an active
time of discontinuous transmission (DTX)
• Which level of power control is actually applied
These complex conditions lead to individual interference situations for each
TRX at a specific time and location. By way of a case study, Table B depicts a
typical instantaneous interference situation in a cell with 4 TRX at a random
time, time-slot and location:
29
Table B:
Case study TRX No C/I RXQUAL Speech qual. RXQUAL Speech qual.
Instantaneous interference situation case of case of case of case of
in case of Hopping /No-Hopping no FH no FH FH FH
1 30 dB 0 excellent 5 good
2 11 dB 2 good 5 good
3 5 dB 7 bad 5 good
4 13 dB 1 excellent 5 good
Table B shows a typical situation of various C/I values over different frequen-
cies. In the “no FH case” TRX 3 – with highest interference load of C/I = 5 dB
– turns out to be unusable. The probability of handover failure, and subse-
quent dropped calls would increase dramatically. Call attempts would also
result in many connect failures on TRX 3. The effective usable capacity of the
cell would be only 75% of the installed capacity.
Frequency hopping Collisions on single traffic channels with co-channel and adjacent channel
creates interferer diversity frequencies that lead to intolerable interference – under some circumstances
– are unavoidable. Frequency hopping was included in the GSM design to
smooth out these interference collisions by creating Interferer Diversity. The
process also allows for the repair of spurious bits – on a single channel – by
using the good transmission characteristics of the other channels. Because
the GSM signal processing was designed with a high level of efficiency, good
reception quality is possible even if one of the four frequencies of a hopping
sequence is completely blocked (resulting in a bit error rate for that frequency
of 50%).
Frequency hopping is a key feature in economic operation of a network be-
cause frequency plans no longer have to be designed for worst case (worst
frequency) scenarios. With frequency hopping, the targeted QoS can be
reached using a smoothed average scenario. The potential gain is an increase
of traffic capacity with revenues per cell from 10% up to 30%. Frequency
hopping makes colliding frequencies again usable. Discontinuous Transmis-
sion (DTX), Power Control and temporarily unused channels all contribute to
a reduced interference level. This becomes usable “space” to frequency hop-
ping and as a consequence QoS or capacity can be improved.
30
Measuring the After activation of frequency hopping, one GSM network operator reported
benefits of frequency the ambiguous experience of QoS improvement on the one hand with indica-
tions of degradation by RXQUAL on the other hand! Described below are the
hopping – case study observations followed by the author’s explanation.
QoS improvements: Measurements of call attempt success rates and hand-
over success rates showed some improvement while the number of dropped
calls decreased (in a medium loaded network, frequency hopping may re-
duce the number of dropped calls about 20% of the previous rate).
Indications for QoS degradation: Contradictory to the observed quality
improvements, the number of “handovers caused by bad quality” had in-
creased steeply. In fact, statistical measurements of RXQUAL reported an
average increase of about one unit. Under normal circumstances this means
a serious degradation of transmission quality.
Interpretation of the results: The RXQUAL parameter does not increase
linearly with the error rate of unprotected bits but with its logarithm. (RXQUAL
increases by one unit if the bit error rate is doubled or decreases by 2 units if
the bit error rate is divided by 4). This explains the RXQUAL values shown in
Table B above.
In frequency hopping the bit error rates (for unprotected bits) for the differ-
ent hopping sequence frequencies are averaged – and mapped into a
RXQUAL FH value for the hopping channel. This RXQUAL FH value is not calcu-
lated as the arithmetical average of the RXQUAL TRXi values of the individual
transceivers TRX i in the non hopping mode:
RXQUAL FH ≥ nint [Σ i =1,n RXQUAL TRXi / n]
where n is the number of TRXs and nint [ ] is the nearest integer function.
Examples are given in the Table C.
31
Table C:
Examples of RXQUAL calculations RXQUAL FH RXQUAL TRXi average potential speech
for hopping channels i=1, n RXQUAL TRXi quality
5 0, 2, 7, 1 2.50 good
5 5, 5, 5, 5 5.00 fair
4 0, 6, 0, 0 1.50 excellent
2 0, 0, 4, 1 1.25 excellent
1 3, 0, 0, 0 0.75 excellent
1 0, 0, 0, 2 0.50 excellent
0 0, 1, 0, 0 0.25 excellent
The potential speech quality is included in the table, but there is not a unique
mapping or correlation to RXQUAL. The reason being that speech quality is
closer related to the protected class 1 bits, than to the unprotected class 2
bits – which are related to RXQUAL.
RXQUAL FH is a very questionable parameter as a quality figure for
two main reasons:
• The same value of RXQUALFH can yield significantly different
speech qualities as depicted in rows 1 and 2 of Table C
• RXQUALFH has a different mapping to speech quality than
RXQUALnoFH. This is caused by the dominance of bad frequen-
cies in the RXQUALFH calculation and by the correction of these
errors in the GSM signal processing.
32
Quality testing system Even when frequency hopping does not have a direct effect on class 2 bit
in BSS parameter tuning error rates, a degradation of the average RXQUAL is reported as an artefact
in the calculation. This explains the case for increased rate of “handovers due
to bad quality”.
On the other hand, the average resulting speech quality along with other
QoS parameters in the network can be demonstrated to have improved using
sophisticated quality testing equipment. Consequently, parameter threshold
settings for power control and handover algorithms must be modified upon
switch-on of frequency hopping.
It can be seen that RXQUAL values, again, give only a vague and am-
biguities picture for assessing FH benefits in interference mitigation.
Furthermore, speech quality measurements are required to find suita-
ble RXQUAL threshold settings in power control and handover algo-
rithms after introduction of frequency hopping.
The last chapter of this White Paper presents QVoice 3). Dedicated to quality
test and analysis, the QVoice system can measure resulting improvements of
speech quality. Furthermore, the system helps in setting handover and power
control parameters, because it provides stable and reproducible criteria. With-
out relying on objective measurements, there is a danger of misinterpretation
of observations and incorrect tuning.
33
Argument No. 8:
In Case of EHR Speech Codec
Introduction RXQUAL EFR Would Have
a Different Mapping of Speech Quality
The initial goal for the introduction of enhanced full rate coders is to improve
speech quality. This should be possible by use of higher processing power
expected from the latest generation of GSM chips.
The RXQUAL parameter – as the measure of unprotected bit error rate will
survive the introduction of enhanced full rate. However, the different levels
of RXQUAL and corresponding speech quality for EFR codecs will show signif-
icant differences from what is known from full rate codec. Therefore, the
interpretation of RXQUAL measurements has to be revised after
introduction of EFR, by means of independent voice quality measure-
ments.
34
Argument No. 9:
Speech Quality is the Only Parameter
to Compare Networks of Different
Technologies
RXQUAL has been specifically defined in combination with the full rate speech
coding according by GSM recommendations. Therefore, RXQUAL might be a
helpful supplementary parameter in addition to speech quality in comparing
the quality of GSM 900, DCS1800 and PCS1900 networks with one another.
This is possible because the algorithm of speech coding is identical.
Significant constraints in using RXQUAL for comparisons have to be consid-
ered following the introduction of half rate speech coding as mentioned
under Argument No. 8; or, implementation of improved full rate speech coding
which is currently under consideration within the GSM community. RXQUAL
is still available but it has a different mapping to quality.
For competitive analysis between completely different technologies, QoS
parameters as shown in the introduction of this White Paper are the only
way. In the US market, there is for example, CDMA with variable rate voice
coder up to 13 kbit/s, TDMA with 6.5 kbit/s, PCS at 1900 (=GSM) with a
different 13 kbit/s coder, AMPS with 30 kHz FM and NAMPS with 10 kHz FM.
These will all be in strong technological competition against each other. Only
the subscriber’s ear or voice quality analysis systems can establish a
ranking between different cellular and PCS technologies. RXQUAL will
not find a corresponding parameter in other technologies with the
same meaning to support pan-technology comparisons.
35
Argument No. 10:

Downlink RXQUAL May Fail in
Detecting QoS Degradation in
Uplink Direction
The asymmetry of uplink and downlink quality may have different causes
summarised below:
System related causes:

• Unbalanced power budget
• Different equalisation depths within base station and mobile phone
Causes related to interference on the radio channel:

• Different levels of radio activity up/down
• Inter-modulation or receiver blocking caused by 3rd party interference
• Different topology of co-channel and adjacent channel interference for
uplink and downlink
• Near/far effect
Propagation characteristics:
• Different location of Rayleigh fades
Most of the asymmetric effects indicate severe malfunction. Therefore, the

monitoring of uplink and downlink with the same method is recommended.
Failure to do so may leave QoS degradation undetected.
36
System related causes The Theorem of Reciprocity tells that the path loss over the physical propaga-
unbalanced power budget tion environment is the same for uplink and downlink direction. Therefore,
the transmission and reception characteristics of the GSM system shall allow
for the same path loss as well. The small difference in frequency of about 5%
is negligible below 1 dB. The two system paths are as follows:
The path loss is defined as
L := EIRP - Prec
where
EIRP Effective Isotropic Radiated Power
Prec Received power with (hypothetical) isotropic antenna
The system gain is as EIRPmax - Precmin:
for downlink:
EIRPBTSmax = maximum BTS power

– combiner and filter loss
– antenna cable loss
+ antenna gain
PMSrecmin = GSM reference sensitivity for mobile phone class i

+ antenna cable loss (if existing)
– antenna gain
for uplink:
EIRPMSmax = maximum transmission power of MS class i

– antenna cable loss (if existent)
+ antenna gain
PBTSrecmin = GSM reference sensitivity for BTS

+ gain of tower mounted antenna preamplifier
(if existing)
+ antenna diversity gain (if existing)
+ antenna cable loss
– antenna gain
37
A balanced power budget is achieved if and only if
system_gainup = system_gaindown
which is equivalent to
EIRPBTSmax – PMSrecmin = EIRPMSmax – PBTSrecmin
In reality, this relationship is only achieved with an accuracy of ±5 dB, de-

pending primarily on the employed mobile phone class and the initial net-
work design goals. Therefore, uplink and downlink may have different power
ranges which may lead to significant performance differences on the cell
periphery or at indoor locations.
Downlink measurements of RXQUAL may not be used to draw conclusions

about uplink quality. Major quality differences between uplink and downIink
might be an indication for BTS or antenna maIfunction.
Different equalisation Another minor difference between uplink and downlink might appear in
depths within base station propagation environments with long multipath delays. The GSM specifica-
tion is to equalise delays of up to 16 µs (corresponding to 4 bits duration) –
and mobile phone technically, more is possible, based on the length of the training sequence
which is 26 bits.
Mobile phone manufacturers will be forced by price competition to stay with
the minimum specification and because multipath delay capability is not a
consumer feature. But base station manufacturers might invest in more
processing capability to achieve longer delay compensation up to perhaps
22 µs. This could also contribute to different quality in uplink and downlink
direction.
38
Causes related to inter- Co-channel and adjacent channel interference depends on the activity of
ference on the radio channel frequencies in the network since levels of radio activity in uplink and down-
link are, in general, different. This is summarised in Table D.
Different levels of radio MS to BTS BTS to MS

transmission activity Duty cycle on BCCH Pactive = Tload / NTCH 100%
frequency
Duty cycle on non-BCCH Pactive = Tload / NTCH Pactive = Tload / NTCH
frequency
Power control mandatory to cope with not for BCCH-freq.
near /far effect optional for non-BCCH-freq.
Table D:
DTX enabled mostly used because of not for BCCH-freq.
Factors with Impact on radio
longer talk time optional for non-BCCH-freq.
transmission activity
The probability that a traffic channel is activated is described by the following

equation:
Pactive = Tload / NTCH
The average traffic load (Tload) of the radio cell is given in Erlang (the average
number of calls done at the same time within the cell). Tload is heavily depend-
ent on time of day and on weekly and seasonal variations. NTCH is the number
of traffic channels configured within the cell.
After launch of a new network Pactive will be close to zero due to the small
number of subscribers. Therefore, interference within young networks can
be dominated by downlink interference on BCCH frequencies. However, if
the frequency assignment was not restricted to select BCCH frequencies –
from a small subset of the total spectrum – that type of interference (in
young networks) would be negligible.
39
But even in a mature network Pactive will rarely exceed 60% to 75%. This is
caused by the requirement that blocking probabilities (probability to find only
busy channels at call set-up or incoming handover) have been specified by
most operators to figures of 0.5% to 5%. The number of traffic channels
NTCH must exceed the average traffic in order to match the blocking require-
ment. The relation is given in Table E.
NTRX NTRX Tloadi(2% blk) Pactive

1 7 2.9 erl < 0.42
2 14 8.2 erl < 0.59
3 22 14.9 erl < 0.69

Table E:
Relationship of traffic channels
to average traffic
4 29 21.0 erl < 0.73
Mature networks also have an unbalanced load between the uplink and down-
link direction. This unbalance can be even more biased due to different DTX-
and power control configurations in uplink and downlink.
Intermodulation or When designing a GSM network, not only the self-made co-channel and
receiver blocking conditions adjacent channel interference should be considered as potential sources of
degradation. There are also other systems operating in neighbouring bands
of the spectrum that should be taken into consideration. The mechanisms of
degradation are many fold.
40
Third party In a number of countries the supervision of the spectrum by a state’s author-
unauthorised use of ity is underdeveloped and unauthorised users of the spectrum are not identi-
fied by regular spectrum surveillance and prosecuted. As a consequence a
the allocated GSM band GSM network could suffer from 3rd party interference. Interference could
occur in uplink and downlink direction. Downlink interference is mostly lim-
ited to a certain area in the environment of the interfering transmitter, while
uplink interference effectively reduces the radius of the cell.
Surveys using spectrum analysers connected to the base station antennas
before network launch are strongly recommendable. In the running network,
QVoice best in combination with OMC performance can provide hints about
location and frequency of a potential interference. Further measurements
should be done by use of high sensitivity spectrum analysers.
Spurious emissions Sometimes, sites are shared by the GSM operator and other users. In partic-
ular, PTTs who own an old analogue network like to use the same sites when
introducing GSM. Operating frequencies of the co-sited equipment will be
different. However, the attenuation curves from the transmitters have a lim-
ited steepness and therefore, some power is radiated in certain distance from
the transmitter frequency. A GSM base station receiver’s performance is de-
graded if interfering power density of -114 dBm /200 kHz is exceeded. A
typical transmitter of an analogue AMPS or TACS network has a power den-
sity of 40 dBm/30 kHz. The attenuation required for operation is 162 dB!
41
Normally, antenna decoupling accounts for about 50 dB. For large band sep-
aration of several 10 MHz the transmitter power density has decreased
sufficiently. A very critical case is given when AMPS systems are sharing sites
with GSM networks:
AMPS uplink: 825 - 845 MHz downlink: 870 - 890 MHz
GSM uplink: 890 - 915 MHz downlink: 935 - 960 MHz
The strong AMPS downlink is immediately adjacent to the GSM uplink (e.g.
used in Taiwan). This leads to the necessity that a guard band has to be
reserved. This is a waste of spectrum resources, and therefore, the guard
band will be so small that a certain degradation by spurious emissions might
not be excluded. These degradation will become obvious as a worse quality
in uplink direction which is immediately measurable.
Receiver blocking Another undesired characteristics of a radio receiver is the so-called blocking.
Blocking is a desensitisation of the receiver for the wanted signal. It occurs if
a strong signal arrives at the receiver even if it has zero power density at the
frequency of the wanted signal. The blocking specification, for example, for
a GSM receiver (i.e. less than 3 dB desensitisation) for an out-of-band signal
is 0 dBm according to GSM recommendation 05.07..
Receiver blocking can be a problem in all cases if a strong transmitter is
present. At the base station site it can be removed by additional filter equip-
ment. This is not possible if mobile phones are affected (see also next para-
graph “Intermodulation”).
42
Intermodulation Intermodulation is another effect by which interfering signals within the used
GSM band can be created from adjacent foreign signals out of that band. If
strong foreign signals at frequencies f1 and f2 are arriving at a receiver other
signals at frequencies 2f1 - f2 and f2 - f1 are created within the receiver. They
are called 3rd order intermodulation products. They can interfere with the
wanted signal like any other signal. A multiply increase of 3rd order and higher
order products are created if a large number of signals are in the adjacent
bands. The strength of intermodulation products is a function of the source
signal strength and the receiver linearity.
Intermodulation can be a problem when many adjacent signals are present.
An example is the coexistence of a GSM with a TACS network. Both network
technologies use the same uplink and downlink frequency ranges. If a TACS
base station with about 20 active channels (each of 10 W TX power) is locat-
ed at the periphery of a GSM cell, the resulting intermodulation products can
cause serious quality degradation in downlink while the uplink is completely
unaffected.
Typically, all 3rd party interference effects will have impact – either on down-
link or uplink – but not in both directions. Therefore, uplink and downlink
measurements are required which is difficult to achieve with RXQUAL with-
out significant effort.
43
Different radio link Within traditional networks and network usage with all base stations above
topology in uplink and roof top level, and all mobile phones on street level (as for car telephone)
simulations have shown that uplink and downlink interference levels are quite
downlink direction similar.
Today, many base stations are installed at microcellular sites below roof top.
On the other hand, mobile phones are used as hand-held units in any situa-
tion from deep indoor at ground level up to elevated positions in high-rise
buildings above the level of most of the base stations. This created a variety
of scenarios and only one of these is investigated as an example:
Suffered downlink interference:

Downlink interference suffered by the elevated mobile phone is typically within
acceptable order of magnitude due to extremely good serving cell signal.
However, signals are not longer ranked by distance r according to 1/r 4 but to
1/r 3 ... 1/r 2 due to virtually free space propagation conditions to the sur-
rounding base stations. This may cause unexpected problems if a co-channel
base station is within visibility.
Caused uplink interference:
Significant uplink interference may be caused by the elevated mobile phone.
Its uplink (which is similar to free space propagation) competes with the
Table F: uplink of other mobiles in co-channel cells. Those mobiles are potentially in
Case Study indoor (-15 dB penetration loss) or deep indoor (-30 dB penetration loss)
Interference suffered and caused by a
single mobile phone above roof level
locations so that the range of power control (-20 dB for a 2W hand-held) may
(e.g. hand-held in high rise building turn out to be insufficient.
or topographical location)
44
Many other scenarios are possible and in most of these cases, uplink and
downlink interference mechanisms would be different. Therefore, investiga-
tion of differences in uplink and downlink quality in different usage situations
(street level, indoor ground level, elevated positions) by use of quality testing
tools such as QVoice3) is a method to understand the interference mecha-
nisms and to derive conclusions for network tuning.
The near-far effect A mobile phone at a cell border and another mobile phone in close proximity
to the base station are using channels with a small separation, say 2 channel
widths. GSM specifies that a C/I for the 2nd adjacent channel of -41 dB is
tolerated. On the other hand, the difference in propagation loss between the
near and the far mobile phone to the BTS may easily exceed this threshold.
Therefore, uplink power control with an additional 20 dB dynamic range for
2 W hand-held units is a mandatory configuration. On the downlink direc-
tion, there is no such problem.
In very specific situations, even more than 60 dB propagation loss difference
may occur so that the near-far effect could give a small contribution to unbal-
anced quality in uplink and downlink.
Propagation characteristics Due to the reciprocity law of electromagnetic wave propagation the physical
path loss measured between mobile and base station antennas is identical.
This means that the local average of the propagation loss taken over several
wave lengths does not vary significantly.
45
Different locations Only the location of individual Rayleigh fades is not correlated mainly due to
of Rayleigh fades 45 MHz difference between uplink and downlink frequencies. For a station-
ary mobile without frequency hopping a different quality for the duration of
a call may be measurable. However, this effect does not have a general pref-
erence to uplink and downlink and therefore, will average out even over
small data sets.
Conclusion In many cases quality testing equipment have reported different speech
quality in uplink and downlink direction, and many problems identi-
fied that would not have been detected at all while monitoring the
downlink only. These would include unbalanced power budget, inter-
nal co-channel or adjacent channel interference or 3rd party interfer-
ence causing blocking or intermodulation.
46
Argument No. 11:

RXQUAL Does Not Consider
Acoustical Power Level Variations
A further capability of QVoice3) beside of the evaluation of speech quality is

the measurement of acoustical power level. Each speech sample of QVoice is
initiated with a synchronisation sweep which is transmitted with a constant
and well defined power level. The signal may suffer from acoustical power
variations when travelling through analogue transmission lines, or when the
64 kbit/s PCM A-law is converted by the speech transcoder into the 13 kbit/s
stream. Normally, a conservation of acoustical power is achieved. However,
certain GSM operators may decide to modify the power level within the
TRAU (Transcoder/Rate Adaptation Units) to cope with the different environ-
mental conditions of mobile phones in comparison with standard phones.
The effect of this tuning measures and the stability of acoustical pow-
er level for the duration of a call cannot be monitored with RXQUAL.
The measurement is only possible with inband signal strength measurements
as provided by QVoice.
47
Importance of
Speech Assessment in QoS
The ultimate criterion for QoS provided by a network operator is the satisfac-
tion of the mobile phone user. Users are sensitive to “network search” periods
as well as the time it takes to set up a call. They remember dropped calls
especially when the phone’s display indicates “full field strength” and the
frustration of poor speech quality – where conversations consisted of repeat-
ing broken sentences and tolerating an echo. Surveys have shown that listen-
ing effort in both directions is the single most important factor for sustained
subscriber satisfaction. The key then to high QoS, comes by consistently and
accurately evaluating voice quality from the end user’s perspective.
Assessing speech quality is a complex subject involving the human percep-
tion process and only partly depends on technical parameters which give
limited or often misleading information such as discussed in this White Paper
on RXQUAL.
There are a number of alternative ways of assessing voice quality, but most
evaluation methods were developed to assess quality of speech coders and
speech transmission systems under laboratory conditions. QVoice 3) however,
is a unique system – that tests and analyses speech quality as well as network
performance – on a real-time basis – at any location within a cellular network
service area.
48
QVoice Introduction
The engineering and consulting group of Ascom lnfrasys AG, Solothurn,

Switzerland – who are supporting this White Paper Series on cellular network
– also developed QVoice – an advanced testing and analysis system designed
exclusively to evaluate speech quality on the communication link in cellular
networks.
All the shortcomings of RXQUAL presented in the White Paper are resolved
with QVoice – including the measurement of each of the Quality of Service
parameters listed in the chapter entitled QoS Criteria.
To highlight a few benefits, QVoice can provide the following information
unavailable with RXQUAL measurements:
• Report quality improvements for slow moving mobile phones in fading
environments after activating frequency hopping – Argument No. 1
• Detect the actual quality degradation due to stolen frames and measure
the efficiency of speech frame substitution – Argument No. 2
• Detect echo and other speech degradation due to PSTN analogue effects
such as crosstalk or erroneous noise. This is achieved by using specific
DSP-based algorithms – Argument No. 3
• Detect ‘Ping-Pong’ sound effects – Argument No. 6
• Measure the quality improvement in interference limited networks after
activating frequency hopping; and, assist in necessary BSS parameter revi-
sion for handover and power control – Argument No. 7
• Measure actual speech quality following enhanced full rate speech codec
introduction without need to re-evaluate meaning of RXQUAL – Argu-
ment No. 8
• Compare QoS in networks of different technologies such as analog and
digital – Argument No. 9
• Detect uplink quality degradation by 3rd party intermodulation or simple
co-channel interference which are invisible on the downlink
– Argument No. 10
• Supervise acoustical power level stability for the duration of a call
– Argument No. 11
49
In most situations RXQUAL and Speech quality measurements done by QVoice

are in reasonable correlation. However, it is just the special case of discrepancy
which need special attention. Two examples are provided to demonstrate
that significant inconsistencies between RXQUAL and actual speech quality
on the downlink have been observed.
Figure 7a provides a typical case when RXQUAL from 0 to 1 suggests an

excellent service. However, the speech quality measurement done by QVoice
reports only a “fair” speech quality. Potential causes may be found within the
transcoding devices or within the PSTN and further investigation may be
suggested after such an observation.
But also the opposite case might be observed. Figure 7 b demonstrates a call
where RXQUAL from 3 to 5 suggests marginal quality. However, speech qual-
ity measurement done by QVoice reports an ‘excellent’ speech quality. Those
situations are typical for travelling with normal car speed through the Rayleigh
fading environment or with frequency hopping as described in Argument
No. 1. The network optimization engineer will use this kind of measurements
to adjust RXQUAL thresholds which are used in power control and handover
algorithms.
Figure 7a:
RXQUAL is not sensitive to
speech quality degradation
50
Figure 7b:
RXQUAL does not reflect the high
protection of speech transmission by
convolutional coding and interleaving
Speech Quality The secret of the system’s power is simple: QVoice – in addition to recording
Measurement with PACE conventional air-interface signals – measures speech quality accurately from
the user’s perspective. It therefore takes into account all degrading effects
generated by lack of field strength coverage, co-channel interference, adja-
cent channel interference, radio multipath delays, blocking due to lack of
radio channels, lines or switch capacity, echo, or any other causes. It has
many clear advantages over alternative methods.
Since Ascom first introduced QVoice to the market in 1993, based on a then-
revolutionary application of patented neural networking technology, it has
continued to invest millions of dollars into R&D and into even more advanced
techniques for optimising network quality. The result is the Ascom speech
evaluation algorithm called PACE.
PACE, which evaluates end-to-end speech transmission quality with real speech
samples, and is now being used by Ascom as the basis for both the QVoice
production line and its fixed-line network equivalent, QNet. PACE was exten-
sively tested by the International Telecommunications Union (ITU) in Septem-
ber 1998 and found to be the only algorithm that produced excellent results
in all experiments, accurately predicted speech quality in the presence of
transmission errors and performed equally well for all languages used.
For further detailes on PACE, please refer to the Ascom Technical White Pa-
per “Speech quality and its Objective Evaluation with PACE”.
51
References
1) BER exists also in the uplink but is measureable only in the base sta-
tions
2) Felix Tarköy, Endora Tech AG, Basel “Bit error rate and speech quality”
study conducted for Ascom Infrasys, June 1996
3) An introduction on QVoice is given on Page 49
Literature:
[1] Wolfgang Koch, Jürgen èetersen, Diversity und Frequenzsprungverfahren

im D-Netz, PKI Techn. Mitt. 2/1990
[2] M. Mouly, M.B. Pautet, The GSM System for Mobile Communications,
Palaiseau / France
[3] GSM recommendation 05.08: Radio sub-system link control,
Version 4.5.0, ETSI, Sofia Anti polis / France
52
Please write for titles of other White Papers in the
Networking Series or for additional information
and consulting services to:
Ascom Infrasys AG
P.O. Box
CH-4503 Solothurn
Switzerland
Phone +41 32 624 21 21
Fax +41 32 624 21 43
E-mail qvoice@infrasys.ascom.ch

Rxqual and Voice Quality

Uploaded by

Copyright:

Available Formats

Rxqual and Voice Quality

Uploaded by

Document Information

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

Rxqual and Voice Quality

Uploaded by

Copyright:

Available Formats

a Technical White Paper Series

All rights reserved

Library of Congress Catalog-in-Publication Data

Limits of Liability and Disclaimer of Warranty

Price: US$ 20.00

2nd revised Edition, 1999

Argument No. 5: ................................................................ 26

Argument No. 10: .............................................................. 36

From a subscriber’s point of view, the relevant performance measure is the

QoS Criteria and Speech Quality

Call procedure related QoS criteria:

Speech quality related QoS criteria:

Experienced engineers – responsible for quality assurance of cellular net-

Assessment of It is extremely difficult to define a term as vague as speech quality; everybody

Subjective versus In practice – in an operational network – it is not feasible to assess speech

Overview of GSM Speech Transmission

Channel coding Error protection

Errors Radio channel

Speech decompression Speech reconstruction

Processing Entity Transmitted Information Remarks

1 Human speaker or QVoice Original voice Start of signal

2 Transmission system and Voice coded in 64 Kbit /s Potential source of

3 Voice activity detection and Voice coded in 64 Kbit /s Potential source of

5 Convolutional coding Protected speech frame:

6 Interleaving Partitioning of speech

7 Burst formatter 2 x 57 data bits

8 GMSK modulator Original GMSK modulated

9 RF mixer Original GMSK modulated

Processing Entity Transmitted Information Remarks

10 Radio channel Distorted

- slow lognormal fading Radio channel

11 Receiver front end Distorted

12 Receiver: 8 received bursts: Benefiting from

13 De-interleaver Received protected

14 Channel decoder Received speech frame: RXQUAL estimation

(only on 378 class 1 bits) 182 class 1 bits of error correction

15 If bad frame indication Reconstructed BFI may fail with a

16 Human ear or QVoice 3) Evaluator of

Speech quality: FAIR to POOR

width of fades < λ

a) without error correction, FH

Figure 5: c) with error correction, no FH

Stealing speech frames is a common procedure in GSM when fast signalling

As an example, consider the case where a speech signal is severely impaired

Argument No 3 discussed several effects of the PSTN network which might

Speech quality: GOOD

Speech quality: FAIR to POOR

Operators who decide to operate in switch on mode of frequency hopping

Argument No. 10:

System related causes:

Causes related to interference on the radio channel:

Most of the asymmetric effects indicate severe malfunction. Therefore, the

The path loss is defined as

The system gain is as EIRPmax - Precmin:

EIRPBTSmax = maximum BTS power

PMSrecmin = GSM reference sensitivity for mobile phone class i

EIRPMSmax = maximum transmission power of MS class i

PBTSrecmin = GSM reference sensitivity for BTS

A balanced power budget is achieved if and only if

EIRPBTSmax – PMSrecmin = EIRPMSmax – PBTSrecmin