The information element SQI (Speech Quality Index) estimates the speech quality in the cellular network as perceived by a human listener. This chapter explains how SQI works.
Why SQI?
Traditionally, speech quality in GSM networks is measured by means of the RxQual parameter (which is also available in TEMS Investigation). RxQual, however, suffers from a number of drawbacks which make it an unreliable indicator of speech quality. SQI is a more sophisticated measure which is dedicated to reflecting the quality of the speech (as opposed to radio environment conditions). This means that when optimizing the speech quality in your network, SQI is the best criterion to use.
The rest of this chapter discusses SQI and speech quality issues in some more detail. Specifically, it deals with
- RxQual and its deficiencies (section 15.2)
- the design of SQI (section 15.3)
- the relation between SQI and RxQual (section 15.4).
RxQual and Its Limitations
RxQual is obtained by transforming the bit error rate (BER) into a scale from 0 to 7 (see GSM 05.08). In other words, RxQual is a very basic measure: it simply reflects the average BER over a certain period of time (0.5 s). By contrast, a listener's assessment of speech quality is a complex process which is influenced by many factors. Some of these, all of which RxQual fails to take into account, are the following:
- The distribution of bit errors over time. For a given BER, if the BER fluctuates very much, the perceived quality is lower than if the BER remains rather constant most of the time. Different channel conditions give rise to radically different BER distributions. However, since RxQual just measures the average BER, it cannot capture this. (In fact, the logarithmic scale of RxQual gives rise to the opposite effect: a high BER variance gives a better RxQual than a low variance does. This is completely misleading from a speech quality point of view.)
- Frame erasures. When entire speech frames are lost, this affects the perceived quality in a very negative way.
- Handovers. Handovers always cause some frames to be lost, which generally gives rise to audible disturbances. This does not show at all in RxQual, however, since during handovers BER measurements are suppressed.
- The choice of speech codec. The general quality level and the highest attainable quality vary widely between speech codecs. Moreover, each codec has its own strengths and weaknesses as regards types of input and channel conditions.
In short, RxQual fails to capture many phenomena that have a decisive influence on a listener's judgment of speech quality. Using RxQual for optimization of speech quality in the network thus leads to suboptimal results.
Design of SQI
SQI has been designed to take into consideration all the phenomena discussed in the preceding section. This ensures that it will produce an unbiased prediction of the speech quality, independently of channel conditions and other circumstances. Somewhat roughly, the computation of SQI involves
- the bit error rate (BER)
- the frame erasure rate (FER)
- data on handover events
- statistics on the distributions of each of these parameters.
Furthermore, for each speech codec, SQI is computed by a separate algorithm which is tuned to the characteristics of that codec.
Relation between SQI and RxQual
How does SQI relate to RxQual? As is clear from section 15.2, this question cannot be answered once and for all, since it is not possible to state a relation between the parameters that is generally valid. For a given RxQual, the real quality indicated by SQI will vary depending on the channel conditions and the speech codec used. Conversely, for a given real quality level, a wide range of RxQual values is possible.
Still, to give some examples of what the relation may look like, the graph below is included. It shows SQI as a function of RxQual for the EFR codec and a number of channel conditions. (It must be kept in mind that the curves represent time-averaged RxQual-to-SQI relations; individual segments of speech may of course deviate from these.)
C/I
TEMS Investigation includes a set of information elements containing C/I, the carrier-over-interference ratio. This chapter explains in some detail how the C/I measurements are made and why they are useful.
Why C/I?
The carrier-over-interference ratio is the ratio between the signal strength of the current serving cell and the signal strength of undesired (interfering) signal components. The C/I measurement function built into TEMS Investigation enables the identification of frequencies that are exposed to particularly high levels of interference, something which comes in useful in the verification and optimization of frequency plans.
The rest of this chapter (sections 16.2-16.5) deals with C/I measurement in dedicated mode. It is however also possible to measure C/I in idle mode. This is handy when interacting with a test transmitter (such as TEMS Transmitter) which simulates a base station but is not capable of setting up calls. It should be pointed out that the sampling rate and hence the quality of idle mode C/I values is critically dependent on the settings governing quality measurement in idle mode: see section
Requirements on a Robust C/I Measure
Downlink quality in a radio network can be monitored using the TEMS Speech Quality Index, SQI (see chapter 15). In this way, areas with inadequate speech quality can be identified. However, if frequency hopping is used in the network, it is difficult to pinpoint the frequencies that are affected by the degradation. To help resolve such ambiguities, TEMS Investigation offers the possibility of measuring average C/I for each of the frequencies used in a call.
To obtain a correct C/I estimate, one must take into account the possible use of power control and/or discontinuous transmission (DTX). In the past, rough C/I measurements have sometimes been carried out by comparing the BCCH signal power of the serving cell with that of neighboring cells using the same traffic channels (but different BCCHs). Since such a scheme fails to allow for power control and DTX on the TCHs, it may produce misleading results. By contrast, TEMS Investigation does consider these network functions and is thus able to indicate the actual C/I experienced by the mobile station
Details on C/I Measurements
In dedicated mode, average C/I is presented twice a second, which is equal to the ordinary measurement interval. If frequency hopping is employed, the average C/I for each frequency is presented.
The measurement range extends from -5 dB to +25 dB. A C/I below -5 dB can be regarded as highly unlikely; in addition, if the number of hopping frequencies is low, C/I values below this limit would normally result in a dropped call. Beyond the upper limit, the speech quality is not further improved. Hence, the limitation of the measurement range is not a restriction.
If downlink DTX is used, the number of bursts transmitted from the base station to the mobile station may be lower than the maximum, depending on the speech activity level on the transmitting side. TEMS Investigation makes measurements only on the bursts actually sent from the base station and disregards bursts not transmitted.
Accuracy
The number of hopping frequencies determines the number of bursts used for the C/I measurement on each frequency. For example, if four frequencies are used, 25 bursts (on average) per frequency are received in each 0.5 s interval. With more frequencies, there are fewer bursts for each frequency. This implies that the accuracy of the measurements is better for small sets of hopping frequencies.
If true C/I is within the range 0 to 15 dB, and four frequencies are used for transmission, and there are no DTX interruptions, the measurement error is typically smaller than 1 dB
An Example
To illustrate the use of C/I, data from a test drive is depicted in the figure below. The test drive lasts 40 seconds. EFR speech coding and cyclic frequency hopping with four frequencies are employed throughout. The upper part of the graph shows SQI and RxLev, while the lower part shows C/I for each of the four frequencies:
As appears from the upper graph, SQI dips sharply towards the end of the test drive (after 35 s), indicating poor speech quality. On the other hand, RxLev stays about 50 dB above -110 dBm the whole time. This means that the dip in quality is not due to low signal power level, that is, the quality problem is to do with interference rather than coverage. In fact, and interestingly, RxLev increases during the SQI dip, probably because the power of the interferer increases. Now, looking at the C/I graph, one sees that two of the four frequencies (the thick lines) have a C/I worse than 10 dB during the SQI dip. This explains the poor speech quality, identifying precisely which channels are disturbed. Such information can then be utilized in the process of optimizing the frequency plan for the area.
No comments:
Post a Comment