A.1 Performance of the RPE‑LTP
3GPP46.010Full rate speechRelease 17TranscodingTS
A.1.1 Introduction
The purpose of this annex is to give a broad outline of the performance of the RPE‑LTP codec with other parts of the digital network. Some general guidance is also offered on non‑voice services.
A.1.2 Speech performance
Planning rules for digital processes are defined in terms of quantizing distortion units (qdu) which can be realized from the following formula (reference 1) using the assumption that the formula accuracy represents the determination of qdus from QN measurements:
QN = 37 ‑ 15 log10(n) ,where n is the qdu (A1.1)
By definition 1 qdu is the quantization distortion arising from one commercial PCM codec.
NOTE: The subjective testing methodology to determine QN for the RPE‑LTP codec was consistent with current ITU‑T methods (reference 2).
A.1.2.1 Single encoding
Under error‑free transmission conditions the perceived quality of the RPE‑LTP codec (see figure A.1.1) is lower than both codecs conforming to recommendations ITU‑T G.711 and ITU‑T G.721 (superseded by G.726). Table A.1.1 indicates the relative performance of the codec and can be compared with codecs conforming to recommendations ITU‑T G.711 and ITU‑T G.721 (superseded by G.726).
The performance of the RPE‑LTP codec has been found to be substantially unaffected down to a carrier to interference (C/I) ratio of 10 dB, but may be considered to have acceptable performance down to 7 dB. Smaller C/I ratios produce unacceptable degradation of speech performance and should be avoided.
NOTE 1: It should be noted that there are doubts as to whether the simulations which generated the error pattern properly represent real operating conditions. The C/I values quoted should therefore only be considered as parameters of this simulation. They may not correspond to real radio interference conditions. Results from early GSM validation hardware show that the C/I values which give the performance quoted may be several dBs higher. Some error statistics of the simulations are shown in table A.1.2.
NOTE 2: The real condition C/I = 10 dB is believed to correspond to about 90 % coverage.
Table A.1.1: Relative levels of speech performance under error‑free conditions
|
Codec |
QN (dB) |
qdu |
|
G.711 |
37 |
1 |
|
(64 kbit/s, A- or -law (PCS 1900) PCM) |
||
|
G.721 (superseded by G.726) |
||
|
(32 kbit/s, ADPCM) |
29 |
3.5 (*) |
|
RPE-LTP |
23‑25 |
7‑8 (*) |
|
(*) Commercial A- or -law (PCS 1900) PCM input and output circuitry included. |
||
|
NOTE: The qdu value for the RPE‑LTP codec is a conservative estimate. At present there are no specific CCITT rules for determining qdus for encoding below 32 kbit/s. |
||
Table A1.2: Bit error statistics for C/I test conditions
|
Simulated C/I ratio: |
10 dB |
7 dB |
4 dB |
|
Total number of errors in class I (182 bits |
|||
|
protected by a 1/2 rate code) |
0.016% |
0.61% |
4.1% |
|
Total number of errors in class II (78 bits |
|||
|
unprotected) |
4.5% |
8.3% |
13.0% |
|
Number of "frame erasure" |
|||
|
indications by CRC |
1 |
15 |
95 |
|
Number of "frame erasures" |
|||
|
not detected by CRC |
1 |
14 |
76 |
|
NOTE: The total number of frames was 750. CRC means Cyclic Redundancy Check. |
|||
A.1.2.2 Speech performance when interconnected with coding systems on an analogue basis
A.1.2.2.1 Performance with 32 kbit/s ADPCM (G.721, superseded by G.726)
The speech performance of the RPE‑LTP codec when interconnected with encoding at 32 kbit/s (see figure A.1.3 and A.1.4) decreases in accordance with the formula in clause A.1.2, and appears to obey the law of additivity when qdus have been determined for the individual codecs.
A.1.2.2.2 Performance with another RPE‑LTP codec
The speech performance of the RPE‑LTP codec when interconnected with another codec of the same type (see figure A1.2) is lower than that of A1.2.2.1. It again appears to obey the law of additivity when qdus have been determined for the individual codecs.
A.1.2.2.3 Performance with encoding other than RPE‑LTP and 32 kbit/s ADPCM (G.721, superseded by G.726)
No information is available on this point, so great care shall be exercised when interconnection is made to codecs with encoding different from that of A.1.2.2.1 and A.1.2.2.2.
A.1.3 Non‑speech performance
It should be noted that the RPE‑LTP speech codec is an adaptive system which has been optimized for speech inputs. Great care shall be taken when making measurements with non‑speech signals because the normal assumptions of time invariance and linearity cannot be made.
A.1.3.1 Performance with single sine waves
Detailed experiments have shown that the RPE‑LTP codec will pass sine waves with segmental signal to noise ratios generally in excess of 20 dB in the frequency range of 100 ‑ 2000 Hz. However, in some cases reproduction above 2000 Hz is not as good.
It should be noted that sine waves above 1300 Hz may be reproduced with significant fluctuations in amplitude and frequency due to the adaptive sub‑sampling technique employed. This results in irregularities in the measured frequency response.
A typical frequency response measured with A‑law PCM input circuitry is shown in figure A.1.5. If 13 bit linear PCM input circuitry is used, the irregularity is less.
A.1.3.2 Performance with DTMF tones
It has been shown that the RPE‑LTP codec transfers DTMF signals of 80 ms duration. However, questions like minimum allowable signal duration, pause duration and the behaviour in the presence of transmission errors have not been investigated.
A.1.3.3 Performance with information tones
Experiments have shown that network originated signalling tones, conforming to recommendation ITU‑T Q.35, are easily recognizable when passed through the RPE‑LTP codec.
A.1.3.4 Performance with voice‑band data
Tests have shown that voice‑band data transmission does not work satisfactorily with 1200 bit/s modems according to recommendation ITU‑T V.23. Voice‑band data according to recommendation ITU‑T V.21 (300 bit/s) will not be subject to any significant degradation.
This behaviour has been tested for one RPE‑LTP link (encoder‑decoder). The effect of transmission errors has not been tested.
A.1.4 Delay
The theoretical minimum delay of the RPE‑LTP codec is 20 ms. However, practical realizations may have an additional processing time in the order of 3 ‑ 8 ms.
Figure A.1.1: One ‑ transcoding scheme (A-law)
Figure A.1.2: Two ‑ transcodings scheme (A-law)
Figure A.1.3: Mixed transcodings ‑ scheme 1 (A-law)
Figure A 1.4: Mixed transcodings ‑ scheme 2 (A-law)
Figure A.1.5: Frequency response for RPE‑LTP codec
(with commercial A‑law PCM input and output circuitry)
A.1.5 Bibliography
1) ITU‑T: "Subjective performance assessment of digital processes using the Modulated Noise Reference Unit (MNRU)", annex C, Supplement no 14, Red book, volume V, 1985.
2) ITU‑T: "Subjective performance assessment of digital processes using the Modulated Noise Reference Unit (MNRU)", annex A, Supplement no 14, Red book, volume V, 1985.
3) ITU‑T: "Technical characteristics of tones for the telephone service", recommendation Q.35, Red book, volume VI.1, 1985.