1、 ETSI TR 102 582 V1.1.1 (2007-07)Technical Report Terrestrial Trunked Radio (TETRA);Evaluation of low rate (2,4 kbit/s) speech codecETSI ETSI TR 102 582 V1.1.1 (2007-07) 2 Reference DTR/TETRA-05131 Keywords CODEC, radio, TETRA, voice ETSI 650 Route des Lucioles F-06921 Sophia Antipolis Cedex - FRANC
2、E Tel.: +33 4 92 94 42 00 Fax: +33 4 93 65 47 16 Siret N 348 623 562 00017 - NAF 742 C Association but non lucratif enregistre la Sous-Prfecture de Grasse (06) N 7803/88 Important notice Individual copies of the present document can be downloaded from: http:/www.etsi.org The present document may be
3、made available in more than one electronic version or in print. In any case of existing or perceived difference in contents between such versions, the reference version is the Portable Document Format (PDF). In case of dispute, the reference shall be the printing on ETSI printers of the PDF version
4、kept on a specific network drive within ETSI Secretariat. Users of the present document should be aware that the document may be subject to revision or change of status. Information on the current status of this and other ETSI documents is available at http:/portal.etsi.org/tb/status/status.asp If y
5、ou find errors in the present document, please send your comment to one of the following services: http:/portal.etsi.org/chaircor/ETSI_support.asp Copyright Notification Reproduction is only permitted for the purpose of standardization work undertaken within ETSI. The copyright and the foregoing res
6、trictions extend to reproduction in all media. European Telecommunications Standards Institute 2007. All rights reserved. DECTTM, PLUGTESTSTM and UMTSTM are Trade Marks of ETSI registered for the benefit of its Members. TIPHONTMand the TIPHON logo are Trade Marks currently being registered by ETSI f
7、or the benefit of its Members. 3GPPTM is a Trade Mark of ETSI registered for the benefit of its Members and of the 3GPP Organizational Partners. ETSI ETSI TR 102 582 V1.1.1 (2007-07) 3 Contents Intellectual Property Rights5 Foreword.5 1 Scope 6 2 References 6 3 Definitions and abbreviations.6 3.1 De
8、finitions6 3.2 Abbreviations .6 4 General .7 4.1 Work requirements.7 4.2 Tasks 7 5 Initial study of the TETRA speech Codec7 5.1 Introduction 7 5.2 Polynomial search for mother code rate .8 5.3 Bit classification.12 5.3.1 Bit distribution constraints12 5.3.2 Average Protection Level (APL) metric .13
9、5.4 Puncturing Patterns.15 5.5 Integration and Testing.17 5.5.1 Speech encoder .17 5.5.2 Channel encoder .17 5.5.3 Channel decoder .19 5.5.4 Speech decoder .19 5.5.5 Frame Stealing Mode - CRC Test.19 6 Performance Evaluation .20 6.1 Evaluation Criteria .20 6.2 Results 23 6.3 Additional TU 50 results
10、 28 7 Summary 29 8 Conclusions 30 9 Further Work30 Annex A: Complete simulation data.31 A.1 Distribution 2-44-6-231 A.1.1 Polynomial 17 (1+ X2+ X3+X4)31 A.1.2 Polynomial 1E (1+ X + X2+X3).32 A.1.3 Polynomial 1D (1+ X + X2+X4) 34 A.1.4 Polynomial 0F (X + X2+ X3+X4)36 A.2 Distribution 12-28-9-537 A.2.
11、1 Polynomial 1E (1+ X + X2+X3).37 A.2.2 Polynomial 1D (1+ X + X2+X4) 39 A.2.3 Polynomial 17 (1+ X2+ X3+X4).40 A.2.4 Polynomial 0F (X + X2+ X3+X4)42 A.3 Distribution 20-12-17-544 A.3.1 Polynomial 1E (1+ X + X2+X3).44 A.3.2 Polynomial 1D (1+ X + X2+X4) 45 A.3.3 Polynomial 17 (1+ X2+ X3+X4)47 A.3.4 Pol
12、ynomial 0F (X + X2+ X3+X4)49 A.4 Distribution 30-4-6-1450 ETSI ETSI TR 102 582 V1.1.1 (2007-07) 4 A.4.1 Polynomial 1E (1+ X + X2+X3).50 A.4.2 Polynomial 1D (1+ X + X2+X4) 52 A.4.3 Polynomial 17 (1+ X2+ X3+X4)53 A.4.4 Polynomial 0F (X + X2+ X3+X4)55 History 57 ETSI ETSI TR 102 582 V1.1.1 (2007-07) 5
13、Intellectual Property Rights IPRs essential or potentially essential to the present document may have been declared to ETSI. The information pertaining to these essential IPRs, if any, is publicly available for ETSI members and non-members, and can be found in ETSI SR 000 314: “Intellectual Property
14、 Rights (IPRs); Essential, or potentially Essential, IPRs notified to ETSI in respect of ETSI standards“, which is available from the ETSI Secretariat. Latest updates are available on the ETSI Web server (http:/webapp.etsi.org/IPR/home.asp). Pursuant to the ETSI IPR Policy, no investigation, includi
15、ng IPR searches, has been carried out by ETSI. No guarantee can be given as to the existence of other IPRs not referenced in ETSI SR 000 314 (or the updates on the ETSI Web server) which are, or may be, or may become, essential to the present document. Foreword This Technical Report (TR) has been pr
16、oduced by ETSI Technical Committee Terrestrial Trunked Radio (TETRA). The present document provides the performance results of an investigation into the suitability of NATOs STANAG 4591 MELP speech codec for use in TETRA. ETSI ETSI TR 102 582 V1.1.1 (2007-07) 6 1 Scope The present document presents
17、the study carried out to evaluate the feasibility of using the 2,4 kbit/s MELP codec (i.e. STANAG 4591 codec) over TETRA channels. 2 References For the purposes of this Technical Report (TR), the following references apply: NOTE: While any hyperlinks included in this clause were valid at the time of
18、 publication ETSI cannot guarantee their long term validity. 1 ITU-T Recommendation P.861: “Objective quality measurement of telephone-band (300-3 400 Hz) speech codecs“. 2 ETSI ETS 300 395-2: “Terrestrial Trunked Radio (TETRA); Speech codec for full-rate traffic channel; Part 2: TETRA codec“. 3 ITU
19、-T Recommendation G.191: “Software tools for speech and audio coding standardization“. 4 Dr Michael Street, CIS Division NATO C3 Agency, The NATO Post-2000 Narrow Band Coder: Test and Selection of STANAG 4591. 5 North Atlantic Treaty Organization, Standardization Agreement (STANAG). 6 U.S. Departmen
20、t of Defense, Multi-Excited Linear Predictive Coder (MELP) Bit Stream Study, 15 February 2000. 3 Definitions and abbreviations 3.1 Definitions For the purpose of the present document, the following terms and definitions apply: Adaptive Multi-Rate (AMR) codec: speech and channel codec capable of oper
21、ating at various combinations of speech and channel coding (codec mode) bit-rates Average Protection Level (APL): metric for assessing the effectiveness of error protection applied to bits within codec frames. APL is dependent on bit distribution within codec frames codec mode adaptation: control an
22、d selection of the codec mode bit-rates 3.2 Abbreviations For the purpose of the present document, the following abbreviations apply: ACELP Algebraic Code Excited Linear Prediction AMR Adaptive Multi-Rate APL Average Protection Level CRC Cyclic Redundancy Check FEC Forward Error Correction (Coding)
23、FS Frame Stealing LSF Line Spectral Frequency MELP Minimum Excitation Linear Prediction MELPe Minimum Excitation Linear Prediction enhancement MOS Mean Opinion Score ETSI ETSI TR 102 582 V1.1.1 (2007-07) 7 MSB Most Significant Bit PESQ Perceptual Speech Quality Measure RCPC Rate Compatible Punctured
24、 Convolutional (Coding) SNR Signal to Noise Ratio STANAG Standardisation Agreement TDMA Time Division Multiple Access TETRA Terrestrial Trunked RAdio 4 General 4.1 Work requirements It has been decided to use the 2,4 kbit/s mode of the STANAG 4591 codec. In order to make assessments across the cover
25、age area, rather than in error-free conditions, it is necessary to provide a representative FEC scheme and inject soft channel bit errors with a TETRA modem and radio channel simulation. In order to assess the performance of the codec, the PESQ tool has been used as it reflects the perceived user sp
26、eech quality of the speech accurately. 4.2 Tasks As part of this study the following tasks have been carried out: 1) Polynomial search for reducing the mother code rate to . 2) Bit classification. 3) Puncturing investigations for achieving the required code rates. 4) Frame stealing investigations. 5
27、) Performance evaluation using the PESQ tool. 5 Initial study of the TETRA speech Codec 5.1 Introduction The testbench used is shown in figure 5.1 Note that the highlighted blocks in the figure 5.1 are irrelevant to the measurements mentioned in the present document. ETSI ETSI TR 102 582 V1.1.1 (200
28、7-07) 8 SourceSamplespeech fileSTANAG4591EncoderTETRAFECEncoderSoft ChannelError InjectionBlockTETRAFECDecoderSTANAG4591DecoderDecodedSamplespeech filePESQBlockFigure 5.1: Testbed block diagram 5.2 Polynomial search for mother code rate From an initial convolutional code used in the original TETRA c
29、odec which has a constraint length K=5 and a mother code rate of 1/3, the purpose was to find the best possible fourth polynomial to obtain a new mother code rate of 1/4 with acceptable performance. Indeed, adding a new polynomial will normally increase the error-correction capability of the convolu
30、tional code if it is well chosen. In the present clause, the selection criteria used will be explained. First, let us summarize the properties of the original TETRA convolutional code. It is defined by the following three polynomials, a constraint length K=5 (4 shift registers). As there is no punct
31、uring, its rate is 1/3, which is also known as the “mother code rate“. G1(D)= 1 +D +D2+D3+ D4G2(D) = 1 +D+D3+ D4G3(D) = 1 +D +D2+ D4G4(D)= ? The objective is to find the fourth polynomial. D D D DUnr1nr2nr3na b c dFigure 5.2: Original convolutional encoder structure ETSI ETSI TR 102 582 V1.1.1 (2007
32、-07) 9 Before proceeding further, the notations introduced thus far will be explained first. Unrepresents the input bit at time n. Snis the state represented by “abcd“ at time n. In other words, Snrepresents the bits Un-1,Un-2,Un-3,Un-4. Rnis the “r1r2r3“ codeword at time n of the branch leading fro
33、m state Snto Sn+1 (represented by output r1n, r2n, r3n). The outputs defined by the generator polynomials are given by the following relationships: r1n = Un Un-1 Un-2 Un-3 Un-4 r2n = Un Un-1 Un-3 Un-4 r3n = Un Un-1 Un-2 Un-4 where is the exclusive OR operator. Table 5.1 shows all the states transiti
34、ons of the original convolutional code for all possible information bit inputs: Table 5.1: Original Convolutional Code State Transitions UnSnSn+1Rn0 0000 0000 000 0 0001 0000 111 0 0010 0001 110 0 0011 0001 001 0 0100 0010 101 0 0101 0010 010 0 0110 0011 011 0 0111 0011 100 0 1000 0100 110 0 1001 01
35、00 001 0 1010 0101 000 0 1011 0101 111 0 1100 0110 011 0 1101 0110 100 0 1110 0111 101 0 1111 0111 010 1 0000 1000 111 1 0001 1000 000 1 0010 1001 001 1 0011 1001 110 1 0100 1010 010 1 0101 1010 101 1 0110 1011 100 1 0111 1011 011 1 1000 1100 001 1 1001 1100 110 1 1010 1101 111 1 1011 1101 000 1 110
36、0 1110 100 1 1101 1110 011 1 1110 1111 010 1 1111 1111 101 The corresponding trellis structure is given in figure 5.2. There are 2K-1=24=16 states in the trellis. One can see that the minimum free distance (dmin) is equal to 12. As we can describe a convolutional code by its trellis diagram, what we
37、 call the free distance (or minimum free distance), is the Hamming weight on the branches of the shortest path which diverges from the 0000 state and re-emerges with it. In general, the higher the minimum free distance is for a convolutional code, the better its error performance will be. Adding a n
38、ew generator polynomial will not change the number of states, but the mother code rate will drop to . Consequently, the branch values will change with every bit added to each branch. Hence, the minimum free distance will, on average increase, allowing the encoder to have improved error performance.
39、ETSI ETSI TR 102 582 V1.1.1 (2007-07) 1000011111000110101001110011000100011101110010101011100000111001010110001100111001010110001101011010111011110100000001001000110100010101100111100010011010101111001101111011110100001000010000Un=0Un=1Figure 5.3: Trellis of Original Convolutional Code In the remain
40、ing part of the present clause, the addition of an extra polynomial and the search criteria used to do this will be explained. As there are 4 shift registers, the degree of polynomials used in this code is 4 or less. Also, we have to consider 31 possibilities (32 minus the all-zero polynomial which
41、is irrelevant). Note that, reuse of any of the existing polynomials is not considered. As a result, there are 28 candidate polynomials to choose from. Our polynomial suitability criteria is based on maximizing the free distance. A very important property is that the addition of a new generator polyn
42、omial will not change the path on which the minimum free distance is calculated. Therefore, to calculate the new free distance, we only need to know the new output values corresponding to the branches of the free distance path indicated on the trellis diagram presented earlier. For each polynomial t
43、ested, we have to calculate the values of the new outputs introduced on the free distance path. This is illustrated in the table 5.2. Table 5.2: Free distance problem in polynomial addition UnSn-1Snr4n1 0000 1000 ? 0 1000 0100 ? 0 0100 0010 ? 0 0010 0001 ? 0 0001 0000 ? The new minimal free distance
44、 will be 12 plus the Hamming weight of the five parity bits. Table 5.3 shows the results obtained. The second column (G4(D) lists all the candidate polynomials where a binary codeword 10011 represents G4(D)= 1 + D +D4 (the MSB of the codeword corresponds to the coefficient of D4). ETSI ETSI TR 102 5
45、82 V1.1.1 (2007-07) 11The third column lists the r4output described earlier. EXAMPLE: 10100 means that the output value is 1 on the first trellis depth, 0 on the second, etc. Then the last column contains the minimum free distance provided by each polynomial. Table 5.3: Free distance profile of cand
46、idate polynomials G4(D) Outputs r4n,r4n+1 dmin1 00001 10000 13 2 00010 01000 13 3 00011 11000 14 4 00100 00100 13 5 00101 10100 14 6 00110 01100 14 7 00111 11100 15 8 01000 00010 13 9 01001 10010 14 10 01010 01010 14 11 01011 11010 15 12 01100 00110 14 13 01101 10110 15 14 01110 01110 15 15 01111 11
47、110 16 16 10000 00001 13 17 10001 10001 14 18 10010 01001 14 19 10011 11001 15 20 10100 00101 14 21 10101 10101 15 22 10110 01101 15 23 10111 11101 16 24 11000 00011 14 25 11001 10011 15 26 11010 01011 15 27 11011 11011 16 28 11100 00111 15 29 11101 10111 16 30 11110 01111 16 31 11111 11111 17 The a
48、rrays 21, 27 and 31 (which are highlighted in grey in the table 5.3) are not considered, as those polynomials are identical to one of the original ones. Therefore, the maximum free distance value that cn be achieved is 16, provided by the following four polynomials: G4,1(D)= 1 +D +D2+D3G4,2(D)= 1 +D
49、 +D2+D4G4,3(D)= 1 +D2+D3+D4G4,4(D)= D +D2+D3+D4Also, in order to determine which ones provide the best error performance, simulation data are needed. ETSI ETSI TR 102 582 V1.1.1 (2007-07) 125.3 Bit classification The output bits from the STANAG 4591 Encoder are classified into 4 classes according to their sensitivity which is related to the importance of the information they contain. Each speech bit is classified as either Class 0 (minimum protection, code rate=2/3), Class 1 (code rate=4/9), Class 2 (code rate =1/3) and Class 3 (maximum prot
