3rd Generation Partnership Project; Technical Specification Group TSG-SA4: Codec; Performance Characterization of the AM Speech Codec !3G T "#$%&' (ersion )$)$*+ The present document has been developed within the 3 rd Generation Partnership Project (3GPP TM ) and may be further elaborated for the purposes of 3GPP. The present document has not been subject to any approval process by the 3GPP
Oranisational Partners and shall not be implemented. This !pecification is provided for future development wor" within 3GPP
only. The Oranisational Partners accept no liability for any use of this !pecification. !pecifications and reports for implementation of the 3GPP TM system should be obtained via the 3GPP Oranisational Partners# Publications Offices. 3GPP $eference DTS/TSG<name abbv>-0<WG no><se! no> " %eywords <#e$%o&'() #e$%o&'*> 3GPP Postal address 3GPP support office address 650 Ro+,e 'es -+!.o/es - So0.a 1n,.o/.s Va/bonne - 2R1345 Te/.6 733 8 92 98 82 00 2a96 733 8 93 65 87 16 &nternet 0,,6//%%%.3:.o&: Copyright Notification 'o part may be reproduced e(cept as authori)ed by written permission. The copyriht and the foreoin restriction e(tend to reproduction in all media. * +,,,- 3GPP Orani)ational Partners (.$&/- 01T!- 2T!&- T+- TT.-TT0). .ll rihts reserved. 3G T "#$%&' ,)$)$* !"***-*)+ " 4on,en,s 3oreword........................................................................................................................................................... + !cope...................................................................................................................................................... 4 $eferences............................................................................................................................................... 3 5efinitions and abbreviations.................................................................................................................. 3.+ 5efinitions........................................................................................................................................................... 3.4 .bbreviations....................................................................................................................................................... 6 General................................................................................................................................................... 6.+ Project 7istory..................................................................................................................................................... 6.4 Overview of the .M$ 0oncept........................................................................................................................... 6.3 3unctional 5escription........................................................................................................................................ 6.6 Presentation of the followin sections.............................................................................................................. 8 9uality in 0lean !peech and 2rror 0onditions...................................................................................... : 9uality under bac"round noise and 2rrors 0onditions........................................................................ ; Performances in Tandemin and with variation of the input speech level............................................... < Performances with the 0odec .daptation turned on.............................................................................. , =.5>5T? Performances...................................................................................................................... +@ Performances with 5TM3 tones........................................................................................................... ++ Transparency to !inalin tones............................................................................................................ +4 Performances with special input sinals................................................................................................ +3 Aanuae 5ependency.......................................................................................................................... +6 Transmission 5elay............................................................................................................................... +8 3reBuency $esponse.............................................................................................................................. +: 0omple(ity........................................................................................................................................... 7istory............................................................................................................................................................ 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3 2o&e%o&' This Technical !pecification has been produced by the 3GPP. The contents of the present document are subject to continuin wor" within the T!G and may chane followin formal T!G approval. !hould the T!G modify the contents of this T!- it will be reCreleased by the T!G with an identifyin chane of release date and an increase in version number as followsD =ersion 3.y.) whereD ( the first diitD + presented to T!G for informationE 4 presented to T!G for approvalE 3 &ndicates T!G approved document under chane control. y the second diit is incremented for all chanes of substance- i.e. technical enhancements- corrections- updates- etc. ) the third diit is incremented when editorial only chanes have been incorporated in the specificationE 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4 1 S!oe This technical report provides bac"round information on the performances of the .daptive MultiC$ate (.M$) speech codec. 2(perimental test results from the =erification and 0haracteri)ation phases of testin are reported to illustrate the behavior of .M$ in multiple operational conditions. 2 Re;e&en!es The followin documents contain provisions that- throuh reference in this te(t- constitute provisions of the present document. $eferences are either specific (identified by date of publication- edition number- version number- etc.) or nonCspecific. 3or a specific reference- subseBuent revisions do not apply. 3or a nonCspecific reference- the latest version applies. F+G G!M @+.@6D H5iital cellular telecommunications system (Phase 4I)E .bbreviations and acronymsH. F4G G!M @3.8@D HTransmission plannin aspects of the speech service in the G!M Public Aand Mobile 'etwor" (PAM') systemH. F3G G!M @:.@<D H5iital cellular telecommunications systemE 7alf rate speechE Performance of the G!M half rate speech codecH. F6G G!M @:.88D H5iital cellular telecommunications systemE Performance 0haracteri)ation of the G!M 2nhanced 3ull $ate (23$) speech codecH. F8G G!M @<.:@D H5iital cellular telecommunications systemE &nCband control of remote transcoders and rate adapters for 3ull $ate traffic channelsH. F:G G!M @<.:+D H5iital cellular telecommunications systemE &nCband control of remote transcoders and rate adapters for 7alf $ate traffic channels H. 3 De;.n.,.ons an' abb&ev.a,.ons 3.1 De;.n.,.ons The followin terminoloy is used throuhout this report. Adaptive Multi-Rate (AMR) codec; !peech and channel codec capable of operatin at ross bitCrates of ++.6 "bit>s (Jhalf-rateK) and 44.< "bit>s (Jfull-rateK). &n addition- the codec may operate at various combinations of speech and channel codin (codec mode) bitCrates for each channel mode. Bit-rate change; 0hane of the codec mode bitCrates for a iven (7$>3$) channel mode. Channel mode; G!M 7alfCrate or fullCrate operation Channel mode adaptation; The control and selection of the (3$ or 7$) channel mode. Codec mode; 3or a iven channel mode- the bit partitionin between the speech and channel codecs. Codec mode adaptation; The control and selection of the codec mode bitCrates. Error Patterns 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ ' Error Insertion Device; $esult of offline simulations stored on files. To be used by the H2rror &nsertion 5eviceH to model the radio transmission from the output of the channel decoder and interleaver to the input of the deinterleaver and channel decoder. ull-rate (R); G!M 3ullCrate channel or G!M channel mode !ross "it-rate; The bitCrate of the channel mode selected (44.< "bit>s or ++.6 "bit>s in G!M). #al$-rate (#R); G!M 7alfCrate channel or G!M channel mode In-Band %ignaling; !inalin for codec mode indication and modification carried within the traffic channel. &ut-o$-Band %ignaling; !inalin on the G!M control channels to support lin" control 'oll (ualit); !peech Buality normally achieved on modern wireline telephones. !ynonym with H&!5' BualityH in most western countries. *ireline +ualit); !peech Buality provided by modern wireline networ"s. 'ormally ta"en to imply Buality at least as ood as that of 34"bit>s G.;4: or G.;4< +: "bit>s codecs. 3.2 1bb&ev.a,.ons 3or the purposes of the present document- the followin abbreviations applyD .>5 .naloue to 5iital .0$ .bsolute 0ateory $atin .5P0M .daptive 5ifferential Pulse 0ode Modulation .M$ .daptive MultiC$ate /!0 /ase !tation 0ontroller /T! /ase Transceiver !tation 0>& 0arrierCtoC&nterfere ratio 0& 0onfidence &nterval 0'& 0omfort 'oise &nsertion 0$0 0yclic $edundancy 0hec" 5>. 5iital to .naloue 5.T 5iital .udio Tape 50$ 5eradation 0ateory $atin 5!P 5iital !inal Processor 5TM3 5ual Tone Multi 3reBuency 5T? 5iscontinuous Transmission for power consumption and interference reduction 23$ 2nhanced 3ull $ate 2!P Product of 2 (2fficiency)- ! (!peed) and P (Percentae of Power) of the 5!P 3$ 3ull $ate (also G!M 3$) 37 3reBuency 7oppin G.;4: &TL +:>46>34"bit>s .5P0M codec G.;4< &TL +:"bit>s A5C02AP codec G.;4, &TL <>:.6>++.< "bit>s speech codec G/2$ .verae ross bit error rate G!M Global !ystem for Mobile communications 7$ 7alf $ate (also G!M 7$) &$! &ntermediate $eference !ystem &TLCT &nternational Telecommunication Lnion C Telecommunications !tandardi)ation !ector M'$L Modulated 'oise $eference Lnit Mod. &$! Modified &$! MOP! Million of Operation per !econds MO! Mean Opinion !core M! Mobile !tation M!0 Mobile !witchin 0enter P0M Pulse 0ode Modulation P!T' Public !witched Telecommunications 'etwor" 9 !peechCtoCspeech correlated noise power ratio in d/ !5 !tandard 5eviation 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ # !&5 !ilence 5escriptor !MG !pecial Mobile Group !'$ !inal To 'oise $atio T07C.3! Traffic 07annel .daptive 3ull rate !peech T07C.7! Traffic 07annel .daptive 7alf rate !peech T5M. Time 5ivision Multiple .ccess T3O Tandem 3ree Operation tMOP! true Million of Operations per !econds TL( Typical Lrban at multipath propaation profile at ( "m>s =.5 =oice .ctivity 5etector wMOP! weihted Million of Operations per !econds Multiple 2rror Patterns were used durin the 0haracteri)ation tests. They are identified by the propaation 2rror 0onditions from which they are derived. The followin conventions are usedD 20( 2rror 0onditions at ( d/ 0>& simulatin a radio channel under static 0>& usin ideal 3reBuency 7oppin in a TL3 multipath propaation profile 520i 5ynamic 2rror 0ondition Mi simulatin a radio channel with a slowly varyin 0>& representative of slow fadin conditions- under ideal 3reBuency 7oppin in a TL3 multipath propaation profile unless otherwise stated. (, different 5ynamic 2rror 0onditions were used in the .M$ 0haracteri)ation Phase) 3or abbreviations not iven in this subCclause- see G!M @+.@6 F+G. 8 Gene&a/ 8.1 <&o=e!, >.s,o&$ 3ollowin the standardi)ation of the 23$ speech codec- the !MG4 !peech 2(pert Group (!2G) and especially the !9!G (!peech 9uality !tratey Group) were tas"ed by !MG to study possible strateies for the continuous improvement of the end to end performances of the speech service in G!M networ"s. !2G was specifically as"ed to evaluate the opportunity to desin a robust 3ull $ate mode and>or an 2nhanced 7alf $ate mode. The !9!G report- presented to !MG in +,,:- recommended to start a oneCyear feasibility study of a MultiC$ate speech codec capable to offer at the same time a $obust 3ull $ate mode and an 2nhanced 7alf $ate mode providin wireline Buality under low propaation error conditions , . The feasibility study was completed in 39,; and the results presented to !MGM43. /ased on the feasibility report- !MG approved a new $,< 1or" &tem for the development of the .daptive MultiC$ate (.M$) !peech 0odec. . 9ualification Phase was completed by the end of 49,< with the preCselection of 8 candidates amon the ++ proposals received by !MG++. The selection tests too" place in the summer of +,,< and the results analy)ed in !MG++M; in !eptember +,,<. !MG++ reached a consensus on one solution and recommended to !MG to select the 2'!+ solution proposed by 2ricsson- 'o"ia and !iemens as the basis of the .M$ standard. This proposal was approved by !MGM4;. The completion of the .M$ development included a short optimi)ation phase restricted to the codec proponents followed by an e(haustive =erification and 0haracteri)ation Phase whose results are reported in this Technical $eport. !MG later approved two additional 1or" &tems for the selection of a 'oise !uppresser and the development of a 1ideband e(tension of the .M$ speech codec. The outcome of these 1or" &tems is not included in this Technical $eport. 8.2 ?ve&v.e% o; ,0e 1@R 4on!e, Lnli"e previous G!M speech codecs (3$- 23$- and 7$) which operate at a fi(ed rate and constant error protection level- the .M$ speech codec adapts its error protection level to the local radio channel and traffic conditions. .M$ , The !2G report also proposed to evaluate and standardi)e the Tandem 3ree Operation of the G!M codecs and proposed the creation of a new !T0- later called !MG++- responsible for the end to end Buality of the speech service in G!M 'etwor"s. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ & selects the optimum channel (half or full rate) and codec mode (speech and channel bit rates) to deliver the best combination of speech Buality and system capacity. This fle(ibility provides a number of important benefitsD C &mproved speech Buality in both halfCrate and fullCrate modes by means of codec mode adaptation i.e. by varyin the balance between speech and channel codin for the same ross bitCrateE C The ability to trade speech Buality and capacity smoothly and fle(ibly by a combination of channel and codec mode adaptationE this can be controlled by the networ" operator on a cell by cell basisE C &mproved robustness to channel errors under marinal radio sinal conditions in fullCrate mode. This increased robustness to errors and hence to interference may be used to increase capacity by operatin a tihter freBuency reCuse patternE C .bility to tailor .M$ operation to meet the different needs of operatorsE C Potential for improved handover and power control resultin from additional sinalin transmitted rapidly inCband. The .M$ codec concept is adaptable not only in terms of its ability to respond to chanin radio and traffic conditions but also to be customi)ed to the specific needs of networ" operators. This allows the codec to be operated in many ways of which three important e(amples areD C 3ullCrate only for ma(imum robustness to channel errors. This additional robustness may be used to e(tend the coverae in marinal sinal conditions- or to improve the capacity by usin a tihter freBuency reCuse- assumin hih .M$ M! penetration. C 7alfCrate only for ma(imum capacity advantaeE more than +@@N capacity increase achievable relative to 3$ or 23$ (i.e. same as e(istin 7$). !inificant Buality improvements relative to the e(istin 7$ will be iven for a lare proportion of mobiles as a result of the codec mode adaptation to the channel conditions and e(cellent (wireline li"e) speech Buality in half rate mode for low error conditions. C Mi(ed half>full rate operation allowin a tradeCoff between Buality and capacity enhancements accordin to the radio and traffic conditions and operator priorities. 8.3 2+n!,.ona/ Des!&.,.on The .M$ speech codec includes a set of fi(ed rate speech codecs modes for half rate and full rate operation- with the possibility to switch between the different modes as a function of the propaation error conditions. 2ach codec mode provides a different level of error protection throuh a dedicated distribution of the available ross bit rate (44.< "bit>s in 3ull $ate and ++.6 "bit>s in 7alf rate) between source codin and channel codin. The actual speech rate used for each speech frame depends on the e(istin radio channel conditions. . codec adaptation alorithm selects the optimi)ed speech rate (or codec mode) as a function of the channel Buality. The most robust codec mode is selected in bad propaation conditions. The codec mode providin the best Buality is selected in ood propaation conditions. The codec adaptation relies on channel Buality measurements performed in the M! and the networ" and on in band information sent over the .ir &nterface toether with the speech data. The followin diaram shows the main information flows over the "ey system interfacesD 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ - @S ATS "/.n# See!0 Da,a 4o'e! @o'e Bn'.!a,.on (for uplink) S+::es,e' 4o'e! @o'e (for downlink) Do%n/.n# See!0 Da,a 4o'e! @o'e Bn'.!a,.on (for downlink) 4o'e! @o'e 4omman' (for uplink) 4o'e! 1'a,a,.on 4o'e! 1'a,a,.on S<D S<D S<5 S<5 4>5 4>5 4>D 4>D 4>56 40anne/ 5n!o'e& 4>D6 40anne/ De!o'e& S<56 See!0 5n!o'e& S<D6 See!0 De!o'e& TR1" &n both directions- the speech data frames are associated with a 0odec Mode &ndication used by the receivin end to select the correct channel and source decoders. &n the networ"- the 0odec Mode &ndication must also be sent to the Transcoder Lnits so that the correct source decodin is selected. 3or the adaptation of the uplin" codec mode- the networ" must estimate the channel Buality- identify the best codec for the e(istin propaation conditions and send this information to the M! over the .ir &nterface (0odec Mode 0ommand 5ata field). 3or the downlin" codec adaptation- the M! must estimate the downlin" channel Buality and send to the networ" a Buality information- which can be mapped in the networ" to a OsuestedP codec mode. &n theory- the codec mode can be chaned every speech frame. &n practice- because of the propaation delays and necessary filterin in the codec adaptation functions- the codec mode should be adapted at a lower rate. 2ach lin" may use a different codec mode but it is mandatory for both lin"s to use the same channel mode (either full rate or half rate). The channel mode is selected by the $adio $esource manaement function in the networ". &t is done at call set up or after a handover. The channel type can further be chaned durin a call as a function of the channel conditions. The "ey characteristics of the selected .M$ solution areD C < codec modes in 3ull $ate mode includin the G!M 23$ and &!+3: 23$. C : codec modes in 7alf $ate mode (also supported in 3ull $ate)- includin the &!+3: 23$. C Possibility to operate on a set of up to 6 codec modes selected at call set up or handover. C 0odec Mode &ndications multiple(ed with the Lplin" 0odec Mode 0ommand and !uested 5ownlin" 0odec Mode every other frame. Q &n band sinalin based on a 4 bits information field sent every other bloc" coded over the .ir &nterface. The full set of codec modes is listed in the followin tableD 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ % Channel Source codec .it-rate 12.2 #b.,/s (GS@ 52R) 10.2 #b.,/s 7.95 #b.,/s TC/01S0AM 7.80 #b.,/s (BS136 52R) !TC/0A1S+ 6.70 #b.,/s 5.90 #b.,/s 5.15 #b.,/s 8.75 #b.,/s 7.95 #b.,/s 7.80 #b.,/s (BS136 52R) TC/0/S0AM 6.70 #b.,/s !TC/0A/S+ 5.90 #b.,/s 5.15 #b.,/s 8.75 #b.,/s Ta.le 4$"$): AM Speech Codec Modes 8.8 <&esen,a,.on o; ,0e ;o//o%.n: se!,.ons The followin sections provide a summary of the 0haracteri)ation Phase test results and bac"round information on the codec performances analy)ed durin the =erification Phase. !ections 8 to , summari)e the codec subjective Buality performances under different representative environmental conditions as measured durin the 0haracteri)ation Phase of the project. .n overview of the 0haracteri)ation Phase is included in .nne( .. .dditional test results are also provided in .nne( 0 and 5. !ections +@ to +: provide information on the codec characteristics as reported durin the =erification Phase includinD C The transparence to 5TM3 tones- C The transparence to networ" sinalin tones C The performances special input sinals C The lanuae and tal"er dependency C The freBuency response C The transmission delay C The comple(ity .nne( / lists the reference contributions used in these sections. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )* 5 C+a/.,$ .n 4/ean See!0 an' 5&&o& 4on'.,.ons The codec performances in clean speech and error conditions were measured in 2(periment +a (3ull $ate) and +b (7alf $ate) of the 0haracteri)ation phase of testin. The clean speech performance reBuirements were set for the best codec mode in each error condition as defined in the followin tableD C-I ull Rate Best Codec per$ormance re+uirement #al$ Rate Best Codec per$ormance (re+uirement) 'o 2rrors 23$ 'o 2rrors G.;4< no errors +, d/ 23$ 'o 2rrors G.;4< no errors +: d/ 23$ 'o 2rrors G.;4< no errors +3 d/ 23$ 'o 2rrors 3$ at +3 d/ +@ d/ G.;4< 'o 2rrors 3$ at +@ d/ ; d/ G.;4< 'o 2rrors 3$ at ; d/ 6 d/ 23$ at +@ d/ 3$ at 6 d/ Ta.le '$): 2est Codec Performance e3uirements in Clean Speech and 4rror Conditions . summary of the essential test results is provided below. .dditional results are included in .nne( 0. The followin fiures provide a raphical representation (in Mean Opinion !cores) of the .M$ performances in clean speech in 3ull $ate mode , . 3iure 8.+ compares the performance recorded for the best .M$ full rate codec mode for each impairment condition- with the correspondin performance of 23$ and the related .M$ project performance reBuirement. Experiment 1a - Test Results 1.0 2.0 3.0 8.0 5.0 Condi ti ons M! Se/. ReD+.&. 1@R-2R 52R Se/. ReD+.&. 8.01 8.01 8.01 3.65 1@R-2R 8.06 8.06 8.13 8.0E 3.96 3.59 2.66 52R 8.01 8.01 3.65 3.05 1.53 3o 5&&o&s 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 4/BF 1 'A 1i5ure '$): AM full rate0clean speech performances cur(e !2est AM Codec (s$ 41 (s$ Performance e3uirements+ 3iure 8.4 shows the performances recorded for all < .M$ full rate codec modes in clean speech and error conditions. Important .ote/ MO! values are provided in these fiures $or in$ormation onl). Mean Opinion !cores can only be representative of the test conditions in which they were recorded (speech material- speech processin- listenin conditions- lanuae- and cultural bac"round of the listenin subjectsR). Aistenin tests performed with other conditions than those used in the .M$ 0haracteri)ation phase of testin could lead to a different set of MO! results. On the other hand- the relative performances of different codec under tests is considered more reliable and less impacted by cultural difference between listenin subjects. 3inally- it should be noted that a difference of @.4 MO! between two test results was usually found not statistically sinificant. , &n these fiures- the performance of 23$ at +3 d/ was arbitrarily set to the performance of 23$ in 'o 2rrors conditions. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )) Experiment 1a - Test Results 1.0 2.0 3.0 8.0 5.0 Condi tions M! 52R 12.2 10.2 7.95 7.8 6.7 5.9 5.15 8.75 52R 8.01 8.01 3.65 3.05 1.53 12.2 8.01 8.06 8.13 3.93 3.88 1.86 10.2 8.06 3.96 8.05 3.E0 2.08 7.95 3.91 8.01 8.0E 3.96 3.26 1.83 7.8 3.E3 3.98 3.9E 3.E8 3.11 1.39 6.7 3.77 3.E0 3.E6 3.29 1.E7 5.9 3.72 3.69 3.59 2.20 5.15 3.50 3.5E 3.88 2.83 8.75 3.50 3.52 3.83 2.66 3o 5&&o&s 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 4/BF 1 'A 1i5ure '$": 1amil6 of cur(es for 47periment )a !Clean speech in 1ull ate+ The .M$ 0haracteri)ation test results showed that the selected solution satisfies the .M$ reBuirements in clean speech in 3ull $ate 0hannel. The previous results demonstrate that the combination of all < speech codec modes provide a robust 3ull $ate speech codec down to 6 d/ 0>&. The results also showed that the four hihest codec modes (+4.4- +@.4- ;.,8 S ;.6) are eBuivalent to 23$ in error free conditions and barely affected by propaation errors over a wide rane 0hannel conditions (down to +@C; 0>&). The four lowest codec modes (:.;- 8.,- 8.+8 S 6.;8) are all juded in error free conditions to be eBuivalent to 23$ at +@ d/ 0>&. The three lowest codec modes are statistically unaffected by propaation errors down to 6 d/ 0>&. The followin fiures provide a raphical representation (in Mean Opinion !cores) of the .M$ performances in clean speech in 7alf $ate mode 0 . 3iure 8.3 compares the performance recorded for the best .M$ half rate codec mode for each impairment condition- with the correspondin performance of the 23$- G!M 3$ and G!M 7$ speech codecs and the related .M$ project performance reBuirement. 3iure 8.3 shows the performances recorded for all : .M$ half rate codec modes in clean speech and error conditions. Important .ote/ Once aain- MO! values are provided in these fiures $or in$ormation onl). Mean Opinion !cores can only be representative of the test conditions in which they were recorded (speech material- speech processin- listenin conditions- lanuae- and cultural bac"round of the listenin subjectsR). Aistenin tests performed with other conditions than those used in the .M$ 0haracteri)ation phase of testin could lead to a different set of MO! results. On the other hand- the relative performances of different codec under tests is considered more reliable and less impacted by cultural difference between listenin subjects. 3inally- it should be noted that a difference of @.4 MO! between two test results was usually found not statistically sinificant. 0 &n these fiures- the performances of 23$ at +3 d/ were arbitrarily set to the performances of 23$ in 'o 2rrors conditions. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )" Experiment 1" - Test Results 1.0 2.0 3.0 8.0 5.0 Condi ti ons M! Se/. ReD+.&. 1@R->R 52R 2R >R Se/. ReD+.&. 3.99 3.99 3.99 3.18 2.78 1.50 1@R->R 8.11 8.08 3.96 3.72 3.3E 3.10 2.00 52R 8.21 8.21 3.78 3.38 1.5E 2R 3.50 3.50 3.18 2.78 1.50 >R 3.35 3.28 2.E0 1.92 3o 5&&o&s 4/BF19 'A 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 1i5ure '$3: AM half rate0clean speech performances cur(e !2est AM Codec (s$ 41 (s$ GSM 1 (s$ GSM 1 (s$ Performance e3uirements+ Experiment 1" - Test Results 1.0 2.0 3.0 8.0 5.0 Condi ti ons M! 52R 7.95 7.8 6.7 5.9 5.15 8.75 2R >R 52R 8.21 8.21 3.78 3.38 1.5E 7.95 8.11 8.08 3.96 3.37 2.53 1.60 7.8 3.93 3.93 3.95 3.52 2.78 1.7E 6.7 3.98 3.90 3.53 3.10 2.22 1.21 5.9 3.6E 3.E2 3.72 3.19 2.57 1.33 5.15 3.70 3.60 3.60 3.3E 2.E5 1.E8 8.75 3.59 3.86 3.82 3.30 3.10 2.00 2R 3.50 3.50 3.18 2.78 1.50 >R 3.35 3.28 2.E0 1.92 3o 5&&o&s 4/BF19 'A 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 1i5ure '$4: 1amil6 of cur(es for 47periment ). !Clean Speech in /alf ate+ The .M$ 0haracteri)ation test results showed that the selected solution complies with the .M$ reBuirements in clean speech in 7alf $ate 0hannel. The results demonstrate that the combination of all : speech codec modes provide a 7alf $ate speech codec eBuivalent to the &TL G.;4< (+: "bit>s) speech codec down to +: d/ 0>&. 3urthermore- the results show that .M$ can provide sinificantly better performances than G!M 3$ in the full rane of test conditions- and sinificantly better performances than the G!M 7$ codec down to ; d/ 0>&. The four hihest codec modes (;.,8- ;.6- :.; and 8.,) were found sinificantly better than the G!M 3$ in error free conditions down to +3 d/ 0>& and at least eBuivalent to the 23$ at +@ d/ 0>& down to +: d/ 0>&. The three hihest modes (;.,8- ;.6 and :.;) are eBuivalent to the error free 23$ in very low error conditions. The two lowest modes were found at least eBuivalent to the G!M 3$ over the full rane of test conditions. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )3 6 C+a/.,$ +n'e& ba!#:&o+n' no.se an' 5&&o&s 4on'.,.ons The codec performances under bac"round noise and error conditions were measured in : different 2(periments of the 0haracteri)ation phase of testinD 2(p. 3a- 3b and 3c (3ull $ate) and 2(p. 3d- 3e and 3f (7alf $ate). The followin bac"round noise types were included in the testsD !treet 'oise at +8 d/ !'$ (3a S 3d)- 0ar noise at +8 d/ !'$ (3b S 3e) and Office noise at 4@ d/ !'$ (3c S 3f). The correspondin performance reBuirements were set for the best codec mode in each error condition as defined in the followin tableD C-I ull Rate Best Codec per$ormance re+uirement #al$ Rate Best Codec per$ormance (re+uirement) 'o 2rrors 23$ 'o 2rrors 23$ 'o 2rrors +, d/ 23$ 'o 2rrors G.;4,>3$ 'o 2rrors +: d/ 23$ 'o 2rrors G.;4,>3$ 'o 2rrors +3 d/ 23$ 'o 2rrors 3$ at +3 d/ +@ d/ G.;4,>3$ 'o 2rrors 3$ at +@ d/ ; d/ G.;4,>3$ 'o 2rrors 3$ at ; d/ 6 d/ 3$ at +@ d/ 3$ at 6 d/ Ta.le #$): 2est Codec Performance e3uirements under .ac85round noise and 4rror Conditions . summary of the essential test results is provided below. .dditional results are included in .nne( 0. The followin fiures provide a raphical representation (in Mean Opinion !cores) of the performances recorded in 3ull $ate in 2(periments 3a- 3b S 3c , . Experiment 3a - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 4/BF 1 'A Condi ti ons #M! Se/. ReD+.&. 1@R-2R 52R 2R G.729 1i5ure #$): AM performance cur(es for 47periment 3a !1ull rate 9ith Street :oise+ Experiment 3" - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 4/BF 1 'A Condi ti ons #M! Se/. ReD+.&. 1@R-2R 52R 2R G.729 , &n these fiures- the performances of 23$ at +3 d/ were arbitrarily set to the performances of 23$ in 'o 2rrors conditions. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )4 1i5ure #$": AM performance cur(es for 47periment 3. !1ull rate 9ith Car :oise+ Experiment 3c - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A 4/BF 1 'A Conditi ons #M! Se/. ReD+.&. 1@R-2R 52R 2R G.729 1i5ure #$3: AM performance cur(es for 47periment 3c !1ull rate 9ith ;ffice :oise+ The .M$ 0haracteri)ation test results showed that the selected solution complies with the .M$ reBuirements under bac"round noise in 3ull $ate 0hannel. The results demonstrate that the combination of the : hihest speech codec modes provide a robust 3ull $ate speech codec down to 6 d/ 0>&. .t hih 0>& (down to +3 d/) the three hihest codec modes (+4.4- +@.4 and ;.,8) were found eBuivalent to 23$ in error free condition. .ll codecs modes down to the .M$ 8., performed better than the G!M 3$ across all test conditions. . couple of codecs (:.;- 8.,) still provide at 6 d/ 0>& a Buality eBuivalent to the G!M 3$ at +@ d/ 0>&.. The two lowest modes (8.+8 and 6.;8) were usually found worse than the G!M 3$ at +@ d/ 0>& across the rane of test conditions 0 . 0 The support of the two lowest modes in 3ull $ate is reBuired to allow Tandem 3ree Operation between a 7alf $ate M! and a 3ull $ate M!. They should not be the primary choice for operation in 3ull $ate mode only 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )' The followin fiures provide a raphical representation (in Mean Opinion !cores) of the performances recorded in 7alf $ate in 2(periments 3d- 3e S 3f 1 . Experiment 3d - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF19 'A 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A Condi ti ons #M! Se/. ReD+.&. 1@R->R 52R 2R >R 1i5ure #$4: AM performance cur(es for 47periment 3d !/alf rate 9ith Street :oise+ Experiment 3e - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF19 'A 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A Condi ti ons #M! Se/. ReD+.&. 1@R->R 52R 2R >R 1i5ure #$': AM performance cur(es for 47periment 3e !/alf rate 9ith Car :oise+ Experiment 3f - Test Results 1.0 2.0 3.0 8.0 5.0 3o 5&&o&s 4/BF19 'A 4/BF16 'A 4/BF13 'A 4/BF10 'A 4/BF 7 'A 4/BF 8 'A Condi ti ons #M! Se/. ReD+.&. 1@R->R 52R 2R >R 1i5ure #$#: AM performance cur(es for 47periment 3f !/alf rate 9ith ;ffice :oise+ These results show that the hihest .M$ modes perform well under bac"round noise conditions in half rate channel 1 &n these fiures- the performance of 23$ at +3 d/ was arbitrarily set to the performances of 23$ in 'o 2rrors conditions. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )# down to +: d/ 0>&. &n these conditions- the .M$ performances are almost eBuivalent to 23$ and sinificantly better than the G!M 3$ or G!M 7$ in the same test conditions. 'one of the codec modes is able to meet the initial project reBuirement at +@ d/ 0>&. .ll codec modes are found worse than the taret 3$ at +@ d/ 0>& in these conditions. This is the only critical failure recorded in the characteri)ation phase. .t ; d/ 0>& and below the two lowest codec modes match or e(ceed the performances of the G!M 3$ and G!M 7$. 7 <e&;o&man!es .n Tan'em.n: an' %.,0 va&.a,.on o; ,0e .n+, see!0 /eve/ 2(periment 4 and 2(periment : of the 0haracteri)ation Test plan were intended to evaluate the performances of the .M$ 0odec modes in selfCtandemin and crossCtandemin and with variation of the input speech level. .n overview of the correspondin results is provided in the followin fiuresD Experiment $ - Test Results 1.00 1.50 2.00 2.50 3.00 3.50 8.00 8.50 12.2 10.20 7.95 7.80 6.70 5.9 5.15 8.75 M! D.&e!, Tan'. %/12.2-52R Tan'. %/10.2 Tan'. %/7.95 Tan'. %/7.8 Tan'. %/6.7 Tan'. %/5.9 Tan'. %/5.15 Tan'. %/8.75 Tan'. %/2R Tan'. %/>R 1i5ure &$): 47periment " Test esults !cross-codec tandemin5+ Experiment $ - Test Results 1.00 1.50 2.00 2.50 3.00 3.50 8.00 8.50 52R 2R >R M! D.&e!, Tan'. %/12.2-52R Tan'em Tan'. %/10.2 Tan'. %/7.95 Tan'. %/7.8 Tan'. %/6.7 Tan'. %/5.9 Tan'. %/5.15 Tan'. %/8.75 1i5ure &$": AM Codec Tandemin5 performances 9ith e7istin5 GSM Codecs 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )& Experiment % - Test Results 1.00 1.50 2.00 2.50 3.00 3.50 8.00 8.50 52R 12.2 10.20 7.95 7.80 6.70 5.9 5.15 8.75 Codec M;S -16 'Aov/ -26 'Aov/ -36 'Aov/ Tan'em 1i5ure &$3: Com.ined results for 47periment # !<nfluence of input speech le(el and Tandemin5+ Experiment % - Test Results 1 1.5 2 2.5 3 3.5 8 8.5 3om.na/ 710'A 3om.na/ 3om.na/ -10 'A Tan'em Conditi ons M! G.729-11.8 52R 12.2 10.20 G.729-E 7.95 7.80 6.70 G.729-6.8 5.9 5.15 8.75 2R >R 1i5ure &$4: Com.ined results for 47periment # !<nfluence of input speech le(el and Tandemin5+ ordered .6 impairment t6pe The "ey performances demonstrated by 2(periment 4 test results areD C Tandemin with the clean speech error free +4.4 and +@.4 modes of .M$ do not sinificantly derade the sinle encodin performances of any of the .M$ codec or e(istin G!M codecs. C .ny other tandemin confiuration involvin any two other .M$ codecs introduce a sinificant deradation when compared to the sinle encodin performances of any of the two codecs involved in the tandem confiuration. This deradation is however less sinificant than a tandem confiuration involvin either the G!M 3$ or the G!M 7$. C .ll tandemin confiurations between two .M$ speech codecs (e(cept the worst confiuration 8.+8C 6.;8) are sinificantly better than the G!M 3$ or G!M 7$ in Tandem 2(periment : test results show that the different .M$ speech codec were not sinificantly more impacted by the input speech level than 23$. The hihest codec modes (+4.4 down to ;.6) were enerally found eBuivalent for each impairment type (with variation of the input level or in tandem). The lowest codec modes were always found as least as ood as the G!M 3$. &n tandem conditions- the hihest modes (down to ;.6 "bit>s) do not present a sinificant deradation compare to the sinle encodin condition. The lowest modes are at least as ood as the G!M 3$ in tandemin and always better than the G!M 7$. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )- E <e&;o&man!es %.,0 ,0e 4o'e! 1'a,a,.on ,+&ne' on 2(periments 6a (3ull $ate) and 6b (7alf $ate) of the 0haracteri)ation phase of testin were desined to evaluate the .M$ performances with the adaptation turned on in lon dynamic 0>& profiles representative of operational propaation conditions. Multiple 0>& profile were enerated simulatin different behavior of the radio channel and different slow fadin effects. One profile was used to enerate multiple 2rror Patterns representative of different 3reBuency 7oppin operation modeD &deal freBuency hoppin- nonCideal freBuency hoppin limited to 6 freBuencies and no freBuency hopin. Three different sets of codec modes were used in these 2(periments. They are defined in the followin tableD Codec Modes for 47periment 4a Codec Modes for 47periment 4. Se, G1 12.2) 7.95) 5.9 7.95) 6.7) 5.9) 5.15 Se, G2 12.2) 7.95 6.7) 5.9) 8.75 Se, G3 12.2) 7.80) 6.7) 5.15 7.80) 5.15 Ta.le -$): Sets of codec modes for 47periment 4a = 4. The thresholds and 7ysteresis used for the codec adaptation in the different confiurations are listed in the followin tableD Adaptation Thresholds and /6steresis for 47periment 4a Threshold ) /6steresis ) Threshold " /6steresis " Threshold 3 /6steresis 3 Se, G1 11.5 'A 2.0 'A 6.5 'A 2.0 'A Se, G2 11.5 'A 2.0 'A Se, G3 11.5 'A 2.0 'A 7.0 'A 2.0 'A 5.5 'A 2.0 'A Adaptation Thresholds and /6steresis for 47periment 4. Threshold ) /6steresis ) Threshold " /6steresis " Threshold 3 /6steresis 3 Se, G1 15.0 'A 2.0 'A 12.5 'A 2.5 'A 11.0 'A 2.0 'A Se, G2 12.5 'A 2.0 'A 11.0 'A 2.0 'A Se, G3 13.5 'A 2.0 'A Ta.le -$": Codec Mode Adaptation thresholds and /6steresis used in 47periment 4a = 4. The results of 2(periments 6a and6b are presented in the followin fiuresD Experiment &a - Test Results 1.0 2.0 3.0 8.0 5.0 D541 D5417DTH D542 D543 D548 D545 (B'ea/ 2>* D546 (3on- B'ea/ 2>* D546 (3on B'ea/ 2>*7DTH D547 (3o 2>* Condi ti ons M! 52R 1@R-2R Se, 1 1@R-2R Se, 2 1@R-2R Se, 3 3o,e6 D545) D546 I D547 +se' ,0e same 4/B <&o; ./e D546 on 8 !a&&.e&s 1i5ure -$): 47periment 4a Test esults !>6namic 4rror conditions in 1ull ate+ 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ )% Experiment &" - Test Results 1.0 2.0 3.0 8.0 5.0 D541 D5417DTH D542 D543 D548 D545 (B'ea/ 2>* D546 (3on- B'ea/ 2>* D546 (3on B'ea/ 2>*7DTH D547 (3o 2>* Condi tions M! 2R 1@R->R Se, 1 1@R->R Se, 2 1@R->R Se, 3 3o,e6 D545) D546 I D547 +se' ,0e same 4/B <&o; ./e) D546 on 8 !a&&.e&s D541) D543) D545) D546) D547 %.,0 3 'A 0.:0e& 4/B .n >R 40anne/s D542) D548 %.,0 6 'A 0.:0e& 4/B on >R 40anne/s 1i5ure -$": 47periment 4. Test esults !>6namic 4rror conditions in /alf ate+ The results of 2(periments 6a and 6b can be summari)ed as followsD C &n 3ull $ate- the three tested .M$ codec sets were found sinificantly better than 23$ in ideal or nonC ideal freBuency hoppin cases. &n some cases- the benefit was hiher than + point MO!. C &n 7alf $ate- the three codec sets were found sinificantly better than the G!M 3$ codec (tested at 3 d/ or : d/ lower averae 0>&) in most cases with ideal or nonCideal freBuency hoppin activated. C The performances with nonCideal freBuency hoppin were usually found eBuivalent to the performances with ideal freBuency hoppin for the .M$ codec. The 23$ codec seemed slihtly more impacted in this case. C 'o sinificant improvement compared to the references was identified in nonCfreBuency hoppin cases and low mobile speed in either full rate or half rate channels. The performances of all codecs without freBuency hoppin activated were always found sinificantly worse than their performances when ideal or nonCideal freBuency hoppin was used. C 'o sinificant difference was found when 5T? was activated in the return lin" in either full rate or half rate mode. C There was no sinificant difference between the three codec sets used in full rate or half rate modes- even when the set was limited to two codec modes. 9 V1D/DTH <e&;o&man!es The objective of 2(periment ; of the characteri)ation test plan was to evaluate the deradation induced by the activation of the voice activity detection and discontinuous transmission on the lin" under test , . The e(periment was divided in 6 subCe(periments to separately test the effect on the 3ull $ate and 7alf $ate channel operation and then the performances of each =.5 alorithm (2'! solution and Motorola solution). The tests used a ;Cpoint 0omparison 0ateory $atin to amplify any possible deradation. They consisted in comparin a speech sample for which the =.5>5T? has been applied with the same speech sample without =.5>5T? but in the same channel error>impairment condition. The ;Cpoint scale (0MO!TC3 to I3) corresponded to Buality deradation defined asD OMuch worseP- O1orseP- O!lihtly worseP- O.bout the sameP- O!lihtly betterP- O/etterP and OMuch betterP. The followin impairment type were included in each e(periment and tested for multiple error conditions (6- +@ S +: d/ 0>& in 3ull $ate- ;- +3 S +, d/ in 7alf $ate)D C !inle encodin in clean speech at nominal input level C !inle encodin in clean speech +@ d/ below the nominal input level C !inle encodin in clean speech +@ d/ above the nominal input C !inle encodin in street noise at +8 d/ !'$ C Tandemin in street noise at +8 d/ C !inle encodin in car noise at +8 d/ C !inle encodin in office noise at 4@ The tests were performed with the adaptation turned on- usin the sets of codec modes M+ of Table :.+. 'evertheless- a static 0>& profile was used for all test conditions involvin propaation errors. , The influence of discontinuous transmission on the in band sinalin (mode command and Buality reportin) was tested in 2(periment 6a S 6b. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "* The tests also included a set of references usin the 23$ codec with the oriinal 23$ =.5 and the new .M$ =.5 alorithms in a subset of the impairment conditions- and the 3$ codec in clean speech with the oriinal 3$ =.5. . null condition was also included in the test. .ll test results with one e(ception showed that the activation of the .M$ =.5>5T? do not introduce any sinificant deradation to the performances of .M$. The difference between the scores obtained by the different conditions were below their respective ,8N confidence interval indicatin that the deradation is not sinificantly different for either impairment type. The same results were found for both =.5 solutions. . direct comparison between the two =.5 options in paired e(periments (2(periments ;a and ;c in 3ull $ate and 2(periments ;b and ;d in 7alf $ate) did not allow to differentiate their respective performances. The only condition showin a sinificantly hiher deradation level in all tests performed was for the G!M 3$ codec with its own =.5 alorithm. 2ven then- the score obtained by the 3$>=.5 codec association was not as bad as a bein Bualified as O!lihtly worseP (first deradation level in the ;Cpoint 0MO! scale). &t was in the order of the deradation of a M'$L at 3@ d/ !>' compared with the oriinal speech sample. 10 <e&;o&man!es %.,0 DT@2 ,ones Twelve e(periments were performed durin the verification phase to evaluate the transparency of the .M$ codec modes to 5TM3 tones. The correspondin test conditions are listed in Table +@.+. The e(periments were limited to error free conditions only. The freBuency deviation was set for the duration of a diit- and was randomly chosen between C+.8 and I+.8N. The rane of tone levels was chosen to avoid clippin in the diital domain and to e(ceed the minimum acceptable input level for the Ainemaster unit used for the detection of 5TM3 tones. . set of ten codecs was tested in each e(periment- comprisin the eiht .M$ modes- the fullCrate G!M speech codec and the .Claw codecs alone (direct condition). E2periment 3o4 tone level (,) #igh tone level (,) '4ist Digit duration re+uenc) deviation + C: d/m C: d/m @ d/ 8@ ms none 4 C+: d/m C+: d/m @ d/ 8@ ms none 3 C4: d/m C4: d/m @ d/ 8@ ms none 6 C+: d/m C+: d/m @ d/ 8@ ms I>C +.8N 8 C+, d/m C+3 d/m C: d/ 8@ ms none : C+3 d/m C+, d/m : d/ 8@ ms none ; C: d/m C: d/m @ d/ <@ ms none < C+: d/m C+: d/m @ d/ <@ ms none , C4: d/m C4: d/m @ d/ <@ ms none +@ C+: d/m C+: d/m @ d/ <@ ms I>C +.8N ++ C+, d/m C+3 d/m C: d/ <@ ms none +4 C+3 d/m C+, d/m : d/ <@ ms none Ta.le )*$): 47perimental conditions for the e(aluation of the AM Codecs Transparenc6 to >TM1 Tones 'ote +D The levels are iven as measured at the input to the 5TM3 detector- however- since the 5.0 is calibrated accordin to &TLCT $ec. G.;++- @d/m in the analoue section is eBuivalent to C:.+8d/ov in the diital section. 'est se+uences/ 3or each e(periment- 4@ test seBuences were processed per codec under test. 2ach test seBuence was produced by the 5TM3 enerator- and comprised a header of x ms followed by each of the +: 5TM3 diits as defined in &TLCT $ec. 9.43. The ap between adjacent 5TM3 diits was eBual to the duration of the diits (see Table +). The lenth of the header in seBuence number n- was set to xT4@@In milliseconds E where nT@..+,. This approach was ta"en to e(ercise the speech codecs over the complete rane of possible phase relationships between the start of a 5TM3 diit and a speech codec frame (4@ms in lenth). Thus each codec mode was subjected to 34@ separate diits per e(periment. 'est Procedure/ 3or each test seBuence- the number of diits undetected by the 5TM3 detector was recorded. 'o attempt to identify 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ ") misdetected diits was made- althouh there were no out of seBuence diits observed. Results/ The percentae of undetected diits measured for each codec mode is iven in Table +@.4a for 2(periments + to : (8@ms diits)- and in Table +@.4b for 2(periments ; to +4 (<@ms diits). Codec mode Rate (5"it-s) E2p6 , E2p6 0 E2p6 1 E2p6 7 E2p6 8 E2p6 9 Mean .M$ mode @ 6.;8 38.3N 6@.,N 3<.+N 6+.3N 8@.@N 63.<N 6+.:N .M$ mode + 8.+8 34.<N 3<.6N 36.;N 3<.<N 84.8N 3;.8N 3,.+N .M$ mode 4 8.,@ +,.;N 4@.3N 48.@N 48.3N 3;.<N +,.+N 46.8N .M$ mode 3 :.;@ ;.<N ;.<N +@.:N <.<N 43.6N :.3N +@.<N .M$ mode 6 ;.6@ 3.<N 8.@N 6.;N 6.+N +3.+N 4.4N 8.8N .M$ mode 8 ;.,8 @.3N +.3N +.3N 4.4N ,.;N @.:N 4.:N .M$ mode : +@.4@ @.@N @.@N @.3N @.@N @.3N @.@N @.+N .M$ mode ; +4.4@ @.@N @.@N @.@N @.@N @.@N @.@N @.@N 3$ G!M +3.@@ @.@N @.@N @.3N @.@N @.:N @.@N @.4N 5irect (.Claw) C @.@N @.@N @.@N @.@N @.@N @.@N @.@N Ta.le )*$"a: Percenta5e of >TM1 di5its undetected 9hen passed throu5h different codecs 9ith '*ms >TM1 di5its$ The mean (alue is calculated o(er all si7 e7periments$ Codec mode Rate (5"it-s) E2p6 : E2p6 ; E2p6 < E2p6 ,= E2p6 ,, E2p6 ,0 Mean .M$ mode @ 6.;8 4+.3N 46.;N 4;.8N 4:.,N 38.,N 4:.:N 4;.+N .M$ mode + 8.+8 +<.+N 4+.3N 48.,N 44.<N 33.6N 4<.+N 46.,N .M$ mode 4 8.,@ <.<N ++.:N ++.:N ;.<N 46.+N ,.6N +4.4N .M$ mode 3 :.;@ +.:N +.:N 4.8N 4.8N 8.,N 3.<N 3.@N .M$ mode 6 ;.6@ @.@N @.@N @.3N @.:N 4.4N @.3N @.:N .M$ mode 8 ;.,8 @.@N @.@N @.@N @.@N +.,N @.3N @.6N .M$ mode : +@.4@ @.@N @.@N @.@N @.@N @.@N @.@N @.@N .M$ mode ; +4.4@ @.@N @.@N @.@N @.@N @.@N @.@N @.@N 3$ G!M +3.@@ @.@N @.@N @.@N @.@N @.@N @.@N @.@N 5irect (.Claw) C @.@N @.@N @.@N @.@N @.@N @.@N @.@N Ta.le )*$".: Percenta5e of >TM1 di5its undetected 9hen passed throu5h different codecs 9ith -*ms >TM1 di5its$ The mean (alue is calculated o(er all si7 e7periments$ urther o"servations/ &nspection of the results for the .M$ speech codecs reveals notably worse performance for 5TM3 sinals enerated with neative twist. To eliminate the 5TM3 detector as the cause of this effect- subsets of 2(periments 8 and : were repeated usin a proprietary networ" based 5TM3 detection alorithm. These additional e(periments also showed substantially worse performance in the presence of neative twist. .n analysis of the processed files revealed that for 5TM3 diits enerated with neative or )ero twist- the .M$ speech codecs have a tendency to add additional neative twist to the sinal. This effect is more pronounced for the lower rate speech codecs. Conclusions/ The results for the fullCrate G!M speech codec appear to be consistent with results from previous tests. 'o detection errors were measured for the reference .Claw condition. 3or 8@ms 5TM3 diits- the +@.4 and +4.4 "bit>s .M$ modes appear to be essentially transparent to 5TM3 sinals under error free conditions- whereas the lower rate modes do not appear to be transparent. 3or <@ms 5TM3 diits the ;.6- ;.,8- +@.4 and +4.4 "bit>s modes appear to be essentially transparent to 5TM3 sinals under error free conditions- whereas the lower rate modes do not appear to be transparent. The .M$ codecs seem to have a tendency to add neative twist to 5MT3 sinals- and are therefore less transparent to diits with neative twist than positive twist. &t is noted that 5TM3 sinals are often enerated by P!T' telephones with neative twist- e.. C4d/- to account for the characteristics of the local loop. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "" 11 T&ansa&en!$ ,o S.:na/.n: ,ones The transparency to networ" sinalin tones was tested for all < codec modes usin typical 3rench and German sinalin tones. rench %ignaling 'ones 3ive different types of 3rench networ" sinalin tones were testedD Two different dial tones- one rinin tone- a busy tone and a special information tone. The description of the different tones is iven belowD 0ontinuous 5&.A TO'2 number + at 66@ 7)- +@s duration 0ontinuous 5&.A TO'2 number 4 at 33@I66@ 7)- +@s duration $&'G&'G TO'2 at 66@ 7) with +.8Q3.8s form factor and a total sinal duration of +4.8s /L!U TO'2 at 66@7) with @.8Q@.8s form factor and a total sinal duration of +4.8s !P20&.A &'3O$M.T&O' TO'2 at ,8@>+6@@>+<@@ 7) and duration of (3V@.3Q4(@.@3)Q+.@s and a total sinal duration of +4.8 s The tone amplitude was set to C+@ d/m@. !erman %ignaling 'ones !i( different types of German networ" sinalin tones were testedD Two dial tones- two rinin tones- a busy tone and a special information tone. The description of the different tones is iven below. 0ontinuous 5&.A TO'2 number + at 648 7)- +8s duration 0ontinuous 5&.A TO'2 number 4 at 68@ 7)- +8s duration $&'G&'G TO'2 number + at 648 7) with @.48Q6.@Q+.@Q6.@Q+.@Q6.;8s form factor- +8s total duration $&'G&'G TO'2 number 4 at 68@ 7) with @.48Q6.@Q+.@Q6.@Q+.@Q6.;8s form factor- +8s total duration /L!U TO'2 at 6487) with @.6<Q@.6<s form factor and a total duration of +@s !P20&.A &'3O$M.T&O' TO'2 at ,8@>+6@@>+<@@ 7) and 3V@.33Q+.@s form factor and a total duration of +@s The tone amplitude was set to C+@ d/m@. .dditionally- a set of sinalin tones was enerated at Q+8 d/m@- which is the lowest level recommended in &TLCT 2.+<@. 'est conditions The sinalin tones at a level of C+@ d/m@ were tested under clean error conditions with no adaptation activated and fi(in the codec mode to the < different possible modes. The sinalin tones were also tested with adaptation on- under static errors with 0>& T ; d/. This was tested for 5T? off and 5T? on. The German sinalin tones at a level of C+8 d/m@ were only tested under clear channel conditions with 5T? activated. This was done to ensure that the artifact identified for the 3$ speech codec with low level sinalin tones and 5T? did not appear in the case of .M$. 'est results The testin has been performed by informal listenin involvin trained listeners- their main concern bein to reconi)e the sinalin tones. The test results can be summari)ed as followsD +. 'o sinificant difference was perceived between the tests performed with 5T? O' and those performed with 5T? O33 4. 3or the error free conditionsD the decoded tones were always easily reconi)ed. Uet the perceived Buality was found to decrease when the codec rate decreases and for the two lowest bit rates (6.;8 and 8.+8) the Buality was rather poor. 3. &n presence of channel errors in 7alf $ate mode- the result was rather poor for the whole set of tones. &n 3ull $ate mode- the Buality was found acceptable with a sliht deradation for the two dial tones. 'ote that the effect of errors was perceived for both channel modes- but more limited and clustered in some parts of the sinal in 3ull $ate mode. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "3 Conclusion .lthouh the Buality of networ" sinalin tones is audibly decreasin for lower bit rates and especially in presence of channel errors in 7alf $ate mode- the sinalin tones were always easily reconi)ed under all testin conditions. .dditionally- 5T? activation did not create any deradation of the transparency of the .M$ codec towards sinalin tones. This conclusion is still valid for low amplitude sinalin tones.. 12 <e&;o&man!es %.,0 se!.a/ .n+, s.:na/s The behavior of the .M$ speech codec in presence of multiple Jspecial input sinalsK was tested durin the =erification Phase. These tests includedD Overload conditions .dditional bac"round 'oises and Tal"ers Music sinals &dle channel behavior &n informal e(pert listenin tests- coverin a wide rane of overload levels and error conditions- there was no evidence to suest that the .M$ speech channel e(hibits any sinificant problems- such as ross instability- in the presence of overload sinals. !imilarly- tests in presence of multiple types of bac"round noises or with a hiher number of tal"ers did not e(hibit any problem with any of the .M$ speech codec modes. The tests in presence of Music indicated that the .M$ speech codec did not e(hibit any problem when compared to the behavior of other wellC"nown speech codecs (23$- &!C:6+- G.;4,). 3inally- no sinificant problem was identified when testin the codec with sinals at very low sinal levels representative of an idle channel. 13 -an:+a:e Deen'en!$ The selection and characteri)ation tests were performed by a lare number of laboratories worldwide usin different lanuaes (see .nne( .). Tests were performed inD 2nlish (L! S L%)- 3rench- German- &talian- Mandarin- !panish The results reported by the different laboratories were consistent. 'o sinificant Buality difference was identified between the results reported by the different listenin laboratories for the different .M$ 0odec Modes. 18 T&ansm.ss.on De/a$ The transmission delay of a communication usin .M$ has been evaluated usin the same method as for the previous G!M speech codecs F4- 3 S 6G. The reference system delay distribution for the downlin" and uplin" directions are provided in fiures +6.+ and +6.4 respectively. The speech transcoders are assumed to be remote located from the /T! (+: "bit>s or < "bit>s subCmultiple(in on the .bis S .ter &nterfaces). Techo Tmsc @a&:.n Tbsc Tsample Tsps Tabisd @a&:.n Tbuff Tencode @a&:.n Trftx Trxproc Tproc @a&:.n Td/a @S4 AS4 ATS @S 1i5ure )4$): eference >o9nlin8 dela6 distri.ution Tmsc @a&:.n Tbsc Tproc @a&:.n Tabisu Trxproc @a&:.n Trftx Tencode Ttransc Tsample @a&:.n Ta/' @S4 AS4 ATS @S 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "4 1i5ure )4$": eference ?plin8 dela6 distri.ution The definition of the different delay parameters is iven in the followin table. The table also provides the value used for the parameter when not dependent of the type of speech codec or subCmultiple(in scheme over the .bis S .ter interfaces. Tabisd Time reBuired to transmit the minimum number of speech data bits over the downlin" .bis interface to start encodin a radio speech frame. 5epends on the speech codec mode- the T$.L frame format and the .bis>.ter subCmultiple(in scheme. 'ote that most T$.L frame synchroni)ation bits can ideally be transmitted by anticipation and are usually not included in this parameter. Tabisu Time reBuired to transmit the minimum number of speech data bits over the uplin" .bis interface to start decodin a speech frame. 5epends on the speech codec mode- the T$.L frame format and the .bis>.ter subC multiple(in scheme. 'ote that the T$.L frame synchroni)ation bits can ideally be transmitted by anticipation and are usually not included in this parameter. Ta/d 5elay in the analoue to diital converter in the uplin" (implementation dependent). !et to +ms F6G. Tbsc !witchin delay in the /!0 (implementation dependent). !et to @.8ms F4 S 6G. Tbuff /ufferin time reBuired for the time alinment procedure for the inCband control of the remote transcoder. !et to +.48 ms F4 S 6G. Td/a 5elay in the diital to analoue converter in the downlin" (implementation dependent). !et to +ms F4 S 6G. Techo 5elay induced by the echo canceller (implementation dependent). !et to +ms F4 S 6G. TencodeD Processin delay reBuired to perform the channel encodin (implementation dependent). 5epends on the channel codin comple(ity of each codec mode. Tmsc !witchin delay in the M!0 (implementation dependent). !et to @.8ms F4 S 6G. Tproc Processin delay reBuired to perform the speech decodin (implementation dependent). 5epends on the speech decodin comple(ity of each codec mode. Trftx Time reBuired for the transmission of a speech frame over the air interface. 5erived from the radio framin structure and the interleavin scheme. 1orst case is 3;.8 ms in 3ull $ate mode and 34.8 ms in 7alf $ate mode F4 S 6G. Trxproc Processin delay reBuired to perform the channel eBuali)ation- the channel decodin and !&5Cframe detection (implementation dependent). The channel decodin depends on the codec mode. The channel eBuali)ation part was set to :.<6 ms in 3ull $ate mode and 3.8 ms in 7alf $ate mode F6G. Tsample 5uration of the sement of P0M speech samples operated on by the speech transcoderD 48 ms in all cases correspondin to 4@ ms for the processed speech frame and 8 ms of loo" ahead. Tsps 1orst case processin delay reBuired by the downlin" speech encoder before an encoded bit can be sent over the .ter>.bis interface ta"in into account the speed on the .ter>.bis interface (implementation dependent). 5epends on the speech codin comple(ity of each codec mode and on the subCmultiple(in rate on the .ter>.bis interface. /ecause of the priority iven to the decodin- Tproc is also added to the overall downlin" transmission delay. Ttransc M! speech encoder processin delay- from input of the last P0M sample to output of the final encoded bit (implementation dependent). 3or the evaluation of the transmission delay- it was assumed that the speech decodin has a hiher priority than the speech encodin- i.e. this delay is artificially increased by the speech decodin delay. Marin &mplementation dependent marins in the different system components. !et as followsD M!0 MarinD @.8 ms F4 S 6G /!0 MarinD @.8 ms F4 S 6G /T! MarinD @.68 ms downlin"- @.3 ms uplin" F4 S 6G M! MarinD 4 ms in 3ull $ate- +., ms in 7alf $ate F4 S 6G. The processin delays were estimated usin comple(ity fiures for each codec mode. &n addition- to ta"e into account the dependence on the 5!P implementation- the computation was based on the same methodoloy used for the previous G!M speech codecs F6G. The 5!Ps runnin the speech and channel codec are modeled with the 3 followin parametersD E represents the 5!P 2fficiency. This corresponds to the ratio tMOP!>wMOP! of the codec implementation on the 5!P. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "' % represents for the speed of the 5!PD Ma(imum 'umber of Operations that the 5!P can run in + second. This number is e(pressed in MOP!. P represents the percentae of 5!P processin power assined to the codec. The processin delay of a tas" of comple(ity ? (in wMOP!) can then be computed usin the eBuationD ESP X D 4@ = ms 3or compatibility reasons- the same 2!P parameter used for the 23$ processin delays computation F6G was usedD 2!PT48 , . The followin tables provide the overall transmission delay parameters for each codec mode. The desin objective for the .lorithmic $ound Trip Transmission 5elay (.$T5 T 4Tsample I 4Trftx I Tabisu I Tabisd) was set to the 23$ .$T5 increased by +@ ms in 3ull $ate mode- and the G!M 7$ .$T5 increased by +@ ms in 7alf $ate mode. Tables +6.+ and +6.4 define the parameters impactin the computation of the transmission delays over the .bis>.ter interfaces (Tabisu S Tabisd) for the +: "bit>s and <"bit>s subCmultiple(in schemes respectively. The definition of different parameters is provided below. They are derived from the .M$ T$.L frame format provided in F8 S :G. Min > o$ "its/ Minimum number of speech bits reBuired to start the ne(t operation (speech decodin in uplin" or channel encodin in downlin"). %)nc6 "its/ .dditional synchroni)ation bits in the T$.L frame (synchroni)ation header not included) before reachin the last reBuired bit. Min > Data/ $an" of the last reBuired bit in the T$.L frame. > Anticip6/ 'umber of bits that can be sent by anticipation. > Re+uir6/ $esultin number of bits that must be received (Min M5ata C M .nticip.). Min @ S6nc$ Min @ Mode of .its .its >ata @ anticip$ @ e3uir$ Ta.isu Full )"$" 6 183 3E 105 6.625 Rate )*$" 6 188 3E 106 6.625 16k &$%' 6 188 3E 106 6.625 Upl &$4 6 188 3E 106 6.625 #$& 6 188 3E 106 6.625 '$% 6 188 3E 106 6.625 '$)' 6 188 3E 106 6.625 4$&' 6 188 3E 106 6.625
where inp(") and out(") are the input (oriinal) and output (processed) sinals and M is the total number of processed samples. The freBuency response was computed for all < codec modes (+4.4- +@.4- ;.,8- ;.6- :.;- 8.,- 8.+8 and 6.;8 "bit>s)- in errorCfree condition- with 5T? disabled. Tone sinals were enerated and processed in the rane 8@C3,,< 7) with a freBuency step of 4+ 7). 2ach tone lasted < seconds at a level of Q4: d/ovl. &n order to discard potential transition effects of the codec- the first 8+4 samples (:6 ms at fcT< "7)) of the input S output sinals were not ta"en into account in the computation. 3iure +8.+ provides the freBuency responses measured for the < .M$ speech codecs. Table +8.+ lists the attenuation measured for each codec at the edes of the telephone bandwidth. The usual definition of 3Cd/ bandwidth can be applied to the 6 hihest bitCrates leadin to a bandwidth eBual or wider than the telephone band. !ome limitations appear for the 6 lower bitCrates. Input 3evel dependenc)/ The same computation was repeated with different input levelsD C+: d/ovl and Q3: d/ovl to chec" the dependency of the freBuency response to the input sinal level. !imilar curves were found in both curves. 'ransition "ehavior/ &n order to chec" if the potential transition behavior of the codec influences the shape of the curves- the computations were repeated without discardin the first 8+4 samples and usin tones with a shorter lenth (8@@ ms). Once aain- very similar curves were found in these conditions. 'MR Codec (re)uency Response -35 -30 -25 -20 -15 -10 -5 0 5 50 870 E90 1310 1730 2150 2570 2990 3810 3E30 (re)uency *+,- *d.- 12.2 #b.,/s 10.2 #b.,/s 7.95 #b.,/s 7.80 #b.,/s 6.70 #b.,/s 5.90 #b.,/s 5.15 #b.,/s 8.75 #b.,/s 1i5ure )'$): AM Speech Codec 1re3uenc6 esponses 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ "% AMR Codec Modes ?5"it-s@ Attenuation A $re+B1=0 #C ?dB@ Attenuation A $re+B17,= #C ?dB@ ,060 C@.4< C@.;: ,=60 C@.+< C+.@4 :6<8 C@.++ C3.<; :67 C@.43 C3.34 96: C@.34 C6.:: 86< C@.68 C;.3< 86,8 C@.3@ C<.:8 76:8 C@.46 C<.++ Ta.le )'$): Attenuation at the telephone .and limits 16 4om/e9.,$ Editors Note 2: Section based on the content of Tdoc S!"" "#$/%%& This document 'as produced before a final a(reement 'as reached on the format of the )bis and )ter T*)+ frames& The final format could ha,e an impact on the dela- fi(ures presented belo'& The .M$ speech codec modes comple(ity were evaluated usin the methodoloy previously areed for the standardi)ation of the G!M 7$ and G!M 23$ speech codec. 3or each codec mode- the comple(ity is characteri)ed by the followin itemsD 'umber of cyclesE 5ata memory si)eE Proram memory si)e. The actual values for these items will eventually depend on the final 5!P implementation. The methodoloy adopted for the standardi)ation of previous G!M speech codecs provides a way to overcome this difficulty. &n this methodoloy- the speech and channel codin functions are coded usin a set of basic arithmetic operations. 2ach operation is allocated a weiht representative of the number of instruction cycles reBuired to perform that operation on a typical 5!P device. The Theoretical 1orst 0ase comple(ity (wMOP!) is then computed by a detailed countin of the worst case number of basic operations reBuired to process a speech frame. The wMOP! fiure Buoted is a weihted sum of all operations reBuired to perform the speech and>or channel codin. 'ote that in the course of the codec selection- the 1orst Observed 3rame comple(ity was also measured by recordin the worst case comple(ity fiure over the full set of speech samples used for the selection of the .M$ codec. &n the case of .M$- the comple(ity was further divided in the followin itemsD !peech codin comple(ity in terms of wMOP!- $.M- $OM Tables and Proram $OM 3ull $ate and 7alf $ate channel comple(ity in terms of wMOP!- $.M- $OM Tables and Proram $OM The separation of the speech and channel comple(ity was motivated by the fact that these functions were enerally handled by different system components in the networ" (speech transcodin functions in the T$.L and channel codin>decodin in the /T!). Table +:.+ presents the Theoretical 1orst 0ase (T10) comple(ity (wMOP!) for the different .M$ speech codecs in addition to the 1orst Observed 3rame (1O3) reported durin the selection phase. Tables +:.4 and +:.4 provide the same parameters for the 3ull $ate and 7alf $ate channel codecs. Table +:.6- +:.8 and +:.: provide the $.M- $OM Tables and Proram $OM comple(ity fiures for the different speech and channel codecs. 3or reference- the correspondin .M$ project objectives are also provided in these tables. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3* Mode 12.2 10.2 7.95 7.4 6.7 5.9 5.15 4.75 TWC WOF Speech encoder 14.05 13.66 14.18 13.03 14.03 11.35 9.65 11.63 14.18 13.14 Speech decoder 2.31 2.33 2.53 2.24 2.49 2.57 2.57 2.57 2.57 2.19 EFR Total Speech 16.36 15.99 16.71 15.27 16.52 13.92 12.22 14.20 16.75 15.33 15.21 24 ~ 1.6 EFR Ta.le )#$): AM Speech Codec Theoretical Aorst Case Comple7it6 !in 9M;PS+ Mode 12.2 10.2 7.95 7.4 6.7 5.9 5.15 4.75 Constrant !en"th 5 5 7 5 5 7 5 7 TWC WOF FR Channel Coder 0.34 0.36 0.31 0.29 0.32 0.3 0.29 0.29 0.36 0.33 FR Channel #ecoder 2.42 2.18 4.85 1.66 1.61 3.85 1.34 3.2 4.85 4.45 $R O%& ect Total Channel FR 2.76 2.54 5.16 1.95 1.93 4.15 1.63 3.49 5.21 4.78 2.69 5.7 ~ 2.1 HR Ta.le )#$": AM 1ull ate Channel Codec Theoretical Aorst Case Comple7it6 !in 9M;PS+ Mode 7.95 7.4 6.7 5.9 5.15 4.75 Constrant !en"th 5 5 5 5 5 7 TWC WOF $R Channel Coder 0.19 0.19 0.18 0.17 0.18 0.17 0.19 0.19 $R Channel #ecoder 1.38 1.35 1.25 1.18 1.14 2.66 2.66 2.64 $R O% &ec Total Channel $R 1.57 1.54 1.43 1.35 1.32 2.83 2.85 2.83 2.69 3 ~ 1.1 HR Ta.le )#$3: AM /alf ate Channel Codec Theoretical Aorst Case Comple7it6 !in 9M;PS+ Statc #'na(c Total Speech encoder 1429 3039 4468 Speech decoder 812 946 1758 EFR O%&ect )e Total speech 2241 3039 5280 4711 10000 ~ 2.1 EFR Channel encoder FR 271 1843 2114 Channel decoder FR 280 1915 2195 $R O%&ect )e Total channel FR 551 1915 2466 3154 6600 ~ 2.1 HR Channel encoder $R 385 1317 1702 Channel decoder $R 394 1420 1814 $R O%&ect )e Total channel $R 779 1420 2199 3154 3500 ~ 1.1 HR Channel Meas * Control 107 66 173 Ta.le )#$4: AM Codec AM e3uirements !in )# .its 9ords+ ROM EFR O%&ec t)e Speech 14571 5267 17000 ~ 3.2 EFR $R O%&ec t)e Channel 5049 900 5000 ~ 5.6 HR Channel Meas * Control 187 Total 19807 Ta.le )#$': AM Codec ;M Ta.les e3uirements !in )# .its 9ords+ + o, operators Speech 4851 Channel 1279 Channel Meas * Control 63 Total 6193 Ta.le )#$': AM Codec Pro5ram ;M !in num.er of operators+ %ummar) o$ the comple2it) results/ The .M$ comple(ity parameters appear to be well within the initial constraints of the project. The .M$ speech codec comple(ity is slihtly hiher than the 23$ comple(ity (in wMOP! and $.M)- but the complete set of eiht codecs reBuires 3 times more $OM than the 23$. The channel codec comple(ity matches the initial project objectives (twice the 7$ channel codec comple(ity in 3ull 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3) $ate and once the 7$ channel codec comple(ity in 7alf $ate). The $OM reBuired for the full set of codecs represents around 8 times the $OM reBuired by the 7$ channel codec. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3" 1nne9 16 1@R 40a&a!,e&.Ja,.on <0ase ?ve&v.e% The .M$ 0haracteri)ation Tests were performed on version F4.@G of the .M$ speech codec source code + . Two host laboratories (.rcon and 0OM!.T- L!.) shared the responsibility of processin the speech samples initially provided by the different listenin laboratories. The host laboratories crossCchec"ed the processin performed by the other laboratory and provided the results of this cross chec"in to the 2T!& secretariat. 2iht listenin laboratories performed the correspondin subjective listenin tests in : different lanuaes (0hinese- 2nlish- 3rench- German- &talian S !panish). .ll listenin laboratories were reBuested to provide the results of the listenin tests they performed on an 2(cel 1or"boo" provided by the orani)ation responsible for the Global analysis of the results. The host laboratories and listenin laboratories also provided their own report and analysis to fulfil their contractual commitment. !even different e(periments and +; subCe(periments were specified in the .M$ 0haracteri)ation Test Plan. The primary objectives of the different e(periments are listed belowD 2(periment +a S +bD Performances in 0lean !peech in a 3ull $ate (+a) and 7alf $ate (+b) 2(periment 4D &nteroperability Performances in 0lean !peech (adaptation off) 2(periment 3a- 3b S 3cD Performances under bac"round noise conditions in a 3ull $ate 2(periment 3d- 3e S 3fD Performances under bac"round noise conditions in a 7alf $ate 2(periment 6a S 6bD Performances in dynamic error conditions in a 3ull $ate (6a) and 3ull $ate (6b) (with adaptation on) 2(periment 8D Performances in combined error conditions in 3ull $ate and 7alf $ate (with adaptation on) 2(periment :D &nfluence of the input speech level and Tandemin performances in 3ull $ate and 7alf $ate (adaptation off) 2(periment ;a S ;bD Performance of the 2'! =.5>5T? in 3ull $ate (;a) and 7alf $ate (;b) 2(periment ;a S ;bD Performance of the Motorola =.5>5T? in 3ull $ate (;c) and 7alf $ate (;e) The followin table provides a summary of the impairment conditions included in each e(periment. E2p6 ull Rate #al$ Rate Clean %peech Bc5grd .oise %tatic Errors D)namic Errors Adaptation &n 'andem +a ? ? ? +b ? ? ? 4 ? ? ? ? 3a ? ? ? 3b ? ? ? 3c ? ? ? 3d ? ? ? 3e ? ? ? 3f ? ? ? 6a ? ? ? ? 6b ? ? ? ? 8 ? ? ? ? ? ? : ? ? ? ? ;a ? ? ? ? ? ? ;b ? ? ? ? ? ? ;c ? ? ? ? ? ? ;d ? ? ? ? ? ? Ta.le A$): Summar6 of the AM Characterization Test conditions 2ach e(periment was performed by two different laboratories in two different lanuaes as shown in the followin table. , This version also includes version F(.(G of the =.5 Option 4. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 33 3a"orator)/ Arcon A'D' rance 'elecom Ber5om .ortel Cone2ant EB C&M%A' Aanuaes LsedD 2nlish 2nlish !panish 3rench German 2nlish 2nlish &talian 2nlish !panish 0hinese 'umber of 0onditions Tested 0 +a 3$ X .En(/ X .!er/ 0x$ +b 7$ X .En(/ X .!er/ 1x0 4 X .!er/ X .En(/ 1x$ 3a 3$ X .2ren/ X .En(/ 0x$ 3b 3$ X .2ren/ X .En(/ 0x$ 3c 3$ X .2ren/ X .En(/ 0x$ 3d 7$ X .2ren/ X .En(/ 1x0 3e 7$ X .2ren/ X .En(/ 1x0 3f 7$ X .2ren/ X .En(/ 1x0 6a 3$ X .Span/ X .En(/ %x3 6b 3$ X .Span/ X .En(/ %x3 8 X .En(/ X .4ta/ 1x2 : X .En(/ X .5hin/ 1x3 ;a (2'!) X .En(/ X .Span/ 1x3 ;b (2'!) X .En(/ X .Span/ 1x3 ;a (Motorola) X .En(/ X .Span/ 1x3 ;b (Motorola) X .En(/ X .Span/ 1x3 6ost lab: )rcon )rcon 57S)T 57S)T )rcon 57S)T )*57N 57S)T 57S)T Ta.le A$": Allocation of the 47periments to the Bistenin5 Ba.oratories The 0haracteri)ation tests were performed in .prilCMay +,,,. The results were distributed over the .M$ and !MG++ reflectors before May 4+- +,,,. The lobal analysis was under the responsibility of the G!M 'orth .merica .lliance. The full set of results and report provided by the different laboratories were reviewed and approved in !MG++M++ (Wune 6C;- +,,,) in Tampere- 3inland. The final report was approved by !MGM4, (Wune 4+C48- +,,,) in MiamiC3A- L!. 0 &n this table- the first number represents the number of impairment conditions (propaation errors- tandemin- input level- dynamic profileR). The second number represents the number of codec modes or number of confiurations under test. 3or 2(periments ;- both numbers represent impairment types 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 34 1nne9 A6 1@R Ve&.;.!a,.on <0ase ?ve&v.e% The selected .M$ speech codec was jointly proposed by 2ricsson- 'o"ia and !iemens. &t was identified durin the selection phase by the acronym 2'!+. The proponents had the responsibility to complete the codec optimi)ation after the approval by !MG of the selection phase results. The optimi)ation phase essentially consisted in bu fi(in and optimi)ation of the channel codin. To complete the standardi)ation- a number of Third Parties volunteered to participate to the verification phase by submittin contributions which served as the basis for this Technical $eport. They are listed below with reference to the previous sections of this report. %ections Description Contri"uting &rganiCations 8C, 0haracteri)ation Tests The 0haracteri)ation Tests (.nne( .) were funded by the G!M .ssociation- with additional contributions from 2ricsson- Motorola- 'o"ia- and !iemens +@ Performances with 5TM3 Tones /T (Tdoc. !MG++ +@8>,,) ++ Transparency to .nnouncement Tones 3rance Telecom S TCMobil (Tdoc. !MG++ +3>,,) +4 Performances with !pecial &nput !inals 3rance Telecom S 0one(ant (Tdocs !MG++ +4>,, S +@8>,,) Overload Performances /T (Tdoc !MG++ +@>,,) &dle 0hannel /ehavior /er"om S Aucent Technoloy (Tdocs !MG++ 86>,, S 88>,,) 0hannel 0odin Performances durin 5T? 'ortel 'etwor"s (Tdoc !MG++ :<>,,) Mutin /ehavior 'ortel 'etwor"s S Philips (Tdocs !MG++ :4>,, S :;>,,) +3 Aanuae 5ependency 'o direct contribution +6 Transmission 5elay 'ortel 'etwor"s (Tdoc !MG++ +8<>,,) +8 3reBuency $esponse 0!2AT (Tdoc !MG++ <>,,) +: 0omple(ity .lcatel- Philips- !T Microelectronics S Te(as &nstruments (Tdocs !MG++ ;8>,,- ++;>,,- +,6>,, and 3,<>,,) 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3' 1nne9 46 1''.,.ona/ 40a&a!,e&.Ja,.on Tes, Res+/,s This anne( contains few additional results from the 0haracteri)ation Tests. !pecifically- the followin sections provide a summary of the speech Buality measured for each codec mode under the different error conditions tested in 2(periments + and 3. . number of actual test results are also provided to show the dispersion between tests performed by different laboratories. A16, Per$ormances in Clean %peech in ull Rate mode The followin table shows the typical test results dispersion (in 2Buivalent 9) by comparin the results obtained for the 4 tests performed for 2(periment +aD C-I Best Codec per$ormance (re+uirement) 'est ,/ A'D' (English) 'est 0/ Ber5om (!erman) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 3@.4, "8&28 3+.<8 +.8: 4;.<4 "2&28 4<.<+ @.,, +: d/ 23$ 'o 2rrors 3@.4, "2&28 3@.:6 @.38 4;.<4 "2&28 4,.+4 +.3@ +3 d/ 23$ 'o 2rrors 3@.4, "2&28 3+.64 +.+3 4;.<4 "2&28 3+.+< 3.3: +@ d/ G.;4< 'o 2rrors '>. "8&28 3@.:6 '>. '>. 1&%# 3@.88 8.;, ; d/ G.;4< 'o 2rrors '>. 0&18 4<.4< '>. '>. 1&%# 4<.@, 3.36 6 d/ 23$ at +@ d/ 43.@3 #&%8 43.4: @.43 46.;8 #&%8 43.:; C+.@< Ta.le A3$)-): 47ample of test result dispersion for 47periment )a !1ull rate in clean speech+ The followin tables summari)e the performances of the different .M$ codec modes under the tested error conditions with respect to wellC"nown references in full rate mode and clean speechD Re$erence , Re$erence 0 ER .o Errors ER A ,=dB C-I Codec Mode E+uivalent to Re$erence , E+uivalent to Re$erence 0 *orse than Re$60 ,060 'o 2rrors down to +@ d/ 0>& ; d/ 0>& S below ,=60 'o 2rrors down to +@ d/ 0>& ; / 0>& 6 d/ 0>& :6<8 'o 2rrors down to ; d/ 0>& 6 d/ 0>& :67 'o 2rrors down to ; d/ 0>& 6 d/ 0>& 96: 'o 2rrors down to ; d/ 0>& 6 d/ 0>& 86< 'o 2rrors down to 6 d/ 0>& 86,8 'o 2rrors down to 6 d/ 0>& 76:8 'o 2rrors down to 6 d/ 0>& Ta.le A3$)-": AM speech codec mode performances in clean speech in full rate Re$erence , Re$erence 0 ER .o Errors ER at ,= dB C-I Error Condition E+uivalent to Re$erence , E+uivalent to Re$erence 0 *orse than Re$60 .o Errors +4.4- +@.4- ;.,8- ;.6 :.;- 8.,- 8.+8- 6.;8 ,1 dB C-I +4.4- +@.4- ;.,8- ;.6 :.;- 8.,- 8.+8- 6.;8 ,= dB C-I +4.4- +@.4- ;.,8- ;.6 :.;- 8.,- 8.+8- 6.;8 : dB C-I ;.,8- ;.6 +@.4- :.;- 8.,- 8.+8-6.;8 +4.4 7 dB C-I 8.,- 8.+8- 6.;8 :.;- ;.,8- ;.6 and hiher modes Ta.le A3$)-3: Performances of the AM speech codecs for different error conditions in clean speech in full rate 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3# A1606 Per$ormances in Clean %peech in #al$ Rate mode The followin table shows the typical test results dispersion (in 2Buivalent 9) by comparin the results obtained for the 4 tests performed for 2(periment +bD C-I Best Codec per$ormance (re+uirement) 'est ,/ A'D' (English) 'est 0/ Ber5om 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors G.;4< no errors 4:.+3 1&%# 43.,6 C4.+, 48.44 1&%# 4<.;3 3.84 +, d/ G.;4< no errors 4:.+3 1&%# 44.,6 C3.+, 48.44 1&%# 4;.4, 4.@< +: d/ G.;4< no errors 4:.+3 1&98 43.6+ C4.;4 48.44 1&%# 4;.@6 +.<3 +3 d/ 3$ at +3 d/ '>. #&%8 +,.:3 '>. '>. #&%8 43.8+ '>. +@ d/ 3$ at +@ d/ +:.3: #&%8 +:.3@ C@.@: +<.,4 #&"# 44.4+ 3.4, ; d/ 3$ at ; d/ +6.4+ 9&1# +s8.+6 @.,6 +:.;6 9&1# +,.;8 3.@@ 6 d/ 3$ at 6 d/ ;.;< 9&1# +@.8: 4.;< 8.;4 9&1# +4.@, :.3; Ta.le A3$"-): 47ample of test result dispersion for 47periment ). !/alf rate in clean speech+ The followin tables summari)e the performances of the different .M$ codec modes under the tested error conditions with respect to wellC"nown references in half rate mode and clean speechD Re$erence , Re$erence 0 !6:0; .o Errors R .o Errors Codec Mode E+uivalent to Re$erence , *orse than Re$6, Better than Re$60 E+uivalent to Re$erence 0 *orse than Re$60 :6<8 'o 2rrors down to +: d/ 0>& +3 d/ 0>& +@ d/ 0>& S below :67 'o 2rrors down to +: d/ 0>& +3 d/ 0>& +@ d/ 0>& S below 96: 'o 2rrors down to +: d/ 0>& +3 d/ 0>& +@ d/ 0>& S below 86< 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& S below 86,8 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& S below 76:8 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& S below Ta.le A3$"-": AM speech codec mode performances in clean speech in half rate Re$erence , Re$erence 0 Re$erence 1 !6:0; .o Errors ER at ,= dB C-I R at ,= dB C-I Error Condition E+uivalent to Re$erence , E+uivalent to Re$erence 0 *orse than Re$60 Better than Re$61 E+uivalent to Re$erence 0 *orse than Re$61 .o Errors ;.,8- ;.6- :.; 8.,- 8.+8- 6.;8 ,< dB C-I ;.,8- ;.6- :.; 8.,- 8.+8- 6.;8 ,9 dB C-I ;.,8- ;.6- :.; 8.,- 8.+8- 6.;8 ,1 dB C-I 8.,- 8.+8 :.;- 6.;8- ;.6- ;.,8 ,= dB C-I 8.+8- 6.;8 8.,- :.; ;.6- ;.,8 : dB C-I 6.;8 8.+8- 8.,- :.;- ;.6- ;.,8 7 dB C-I all Ta.le A3$"-3: Performances of the AM speech codecs for different error conditions in clean speech in half rate A1616 Per$ormances in Bac5ground .oise in ull Rate mode The followin tables show the typical test results dispersion (in 2Buivalent 9) by comparin the results obtained for the 4 tests performed for 2(periment 3a- 3b S 3cD C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 4<.@8 "8&28 21� C@.6, 48.<: "2&28 21&21 +.6@ +: d/ 23$ 'o 2rrors 4<.@8 "2&28 21� C@.6, 48.<: "2&28 20&"0 @.3@ +3 d/ 23$ 'o 2rrors 4<.@8 "2&28 20&$3 C+.44 48.<: "8&28 20&%" +.@8 +@ d/ G.;4,>3$ 'o 2rrors 43.;8 "8&28 2$&23 6.6< 48.<: "8&28 2$&32 4.6: ; d/ G.;4,>3$ 'o 2rrors 43.;8 "8&28 29&10 +.@+ 48.<: 0&18 29&1" C+.+: 6 d/ 3$ at +@ d/ 4@.,@ 0&18 23&00 4.;; 46.+8 #&%8 22 C+.8, Ta.le A3$3-): 47ample of test result dispersion for 47periment 3a !1ull rate in Street :oise+ 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3& C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 48.+, "8&28 20&$0 +.:; 4;.:; "8&28 2$&91 @.<@ +: d/ 23$ 'o 2rrors 48.+, "2&28 2#&98 @.4+ 4;.:; "2&28 20&$# C@.<4 +3 d/ 23$ 'o 2rrors 48.+, "2&28 2#&02 @.63 4;.:; "2&28 21&1% @.+3 +@ d/ G.;4,>3$ 'o 2rrors 43.6@ "8&28 21&10 6.3: 4:.44 "2&28 2%&98 3.+< ; d/ G.;4,>3$ 'o 2rrors 43.6@ "8&28 29&32 @.,4 4:.44 "8&28 2#&89 C+.+< 6 d/ 3$ at +@ d/ 4@.,6 #&%8 2"&%2 @.,; 43.4: #&%8 22&99 C@.<3 Ta.le A3$3-": 47ample of test result dispersion for 47periment 3. !1ull rate in Car :oise+ C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 3+.46 "8&28 33&8% +.<8 4,.3; "2&28 38&%8 +.83 +: d/ 23$ 'o 2rrors 3+.46 "2&28 38&"2 C+.+4 4,.3; "2&28 38&%8 +.83 +3 d/ 23$ 'o 2rrors 3+.46 "8&28 3"� @.34 4,.3; "2&28 38&%8 +.83 +@ d/ G.;4,>3$ 'o 2rrors 4:.:; "8&28 3"� 6.<, 4<.:4 "8&28 38&%8 4.4< ; d/ G.;4,>3$ 'o 2rrors 4:.:; 1&98 21&12 +.@6 4<.:4 0&18 2%&29 @.:4 6 d/ 3$ at +@ d/ 4+.34 #&%8 29&2" 4.<< 46.:< #&%8 2#&%3 +.4: Ta.le A3$3-3: 47ample of test result dispersion for 47periment 3. !1ull rate in ;ffice :oise+ The followin tables summari)e the performances of the different .M$ codec modes under the tested error conditions with respect to wellC"nown references in full rate mode under bac"round noise conditionsD Re$erence , Re$erence 0 ER .o Errors R .o Errors Codec Mode E+uivalent to Re$erence , *orse than Re$6, Better than Re$6 0 E+uivalent to Re$erence 0 *orse than Re$60 ,060 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& ; d/ 0>& S below ,=60 'o 2rrors down to +@ d/ 0>& ; / 0>& 6 d/ 0>& :6<8 'o 2rrors down to +: d/ 0>& +3 d/ 0>& down to +@ d/ 0>& ; / 0>& 6 d/ 0>& :67 'o 2rrors down to +: d/ 0>& +3 d/ 0>& down to ; d/ 0>& 6 d/ 0>& 96: 'o 2rrors down to +: d/ 0>& +3 d/ 0>& down to ; d/ 0>& 6 d/ 0>& 86< 'o 2rrors down to ; d/ 0>& 2Buivalent to 3$ at +@ d/ at 6 d/ 0>& 86,8 Lsually found below 3$ at +@ 0>& 76:8 Lsually found below 3$ at +@ 0>& Ta.le A3$3-4: AM speech codec mode performances under .ac85round noise conditions in full rate 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3- Re$erence , Re$erence 0 ER .o Errors R .o Errors Error Condition E+uivalent to Re$erence , *orse than Re$6, Better than Re$60 E+uivalent to Re$erence 0 *orse than Re$60 .o Errors +4.4- +@.4- ;.,8 ;.6- :.; 8., 8.+8- 6.;8 ,1 dB C-I +4.4- +@.4 ;.,8 ;.6- :.;- 8.,- 8.+8 8.+8- 6.;8 ,= dB C-I +@.4 ;.,8- +4.4 ;.6- :.;- 8., 8.+8- 6.;8 : dB C-I +@.4- ;.,8- ;.6- :.;- 8., 8.+8- +4.4- 6.;8 7 dB C-I .ll Ta.le A3$3-': Performances of the AM speech codecs for different error conditions under .ac85round noise conditions in full rate A1676 Per$ormances in Bac5ground .oise in #al$ Rate mode The followin tables show the typical test results dispersion (in 2Buivalent 9) by comparin the results obtained for the 4 tests performed for 2(periment 3d- 3e S 3fD C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 4+.3@ 1&98 2"&38 @.@@ 48.;3 1&%# 2# C@.44 +, d/ G.;4,>3$ 'o 2rrors +,.,, 1&%# 28	 @.88 43.<: 1&98 29&"% @.33 +: d/ G.;4,>3$ 'o 2rrors +,.,, 1&%# 28&28 @.4+ 43.<: 1&%# 2#&$# +.,, +3 d/ 3$ at +3 d/ +<.4+ #&%8 "1� C@.:8 48.4+ #&%8 22&%9 C4.4: +@ d/ 3$ at +@ d/ +;.8: #&"# "#&0% C+.<; 43.@, 9&1# 28 C4.8, ; d/ 3$ at ; d/ +6.,4 9&1# "#&"1 @.48 +,.,4 9&1# "$&09 C+.4< 6 d/ 3$ at 6 d/ 6.+< 9&1# 1&38 3.+4 ;.43 9&1# ""&98 6.+; Ta.le A3$4-): 47ample of test result dispersion for 47periment 3a !/alf rate in Street :oise+ C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 44.;+ 1&%# 22&38 C@.6+ 4;.4: 1&%# 2#&00 C+.8, +, d/ G.;4,>3$ 'o 2rrors 4@.4< 1&98 2"&## +.4< 43.+; 1&%# 29&12 +.88 +: d/ G.;4,>3$ 'o 2rrors 4@.4< 1&98 28&2$ @.@@ 43.+; 1&%# 29&12 +.88 +3 d/ 3$ at +3 d/ +;.:@ 0&18 "%	 +.,6 46.3@ #&%8 23&"1 C+.+3 +@ d/ 3$ at +@ d/ +;.:@ #&"# "0&0" C@.,, 43.@, 9&1# 28&30 C4.;3 ; d/ 3$ at ; d/ +6.8+ 9&1# "#&8" @.8@ 4+.4: 9&1# "% C+.;6 6 d/ 3$ at 6 d/ 4.3, 9&1# 1&2# 6.<: :.;: 9&1# ""&93 6.:; Ta.le A3$4-": 47ample of test result dispersion for 47periment 3. !/alf rate in Car :oise+ C-I Best Codec per$ormance (re+uirement) 'est ,/ Cone2ant (English) 'est 0/ rance 'elecom (rench) 9eB ($eB.) /est Mode .M$ 9eB. 5elta 9eB ($eB.) /est Mode .M$ 9eB. 5elta 'o 2rrors 23$ 'o 2rrors 3;.3: 0&18 31 @.+; 3+.,@ 1&%# 38&8$ C+.<4 +, d/ G.;4,>3$ 'o 2rrors 4;.;8 1&%# 21&39 C@.6+ 4<.4, 1&%# 2%&2% +.@@ +: d/ G.;4,>3$ 'o 2rrors 4;.;8 1&%# 20&03 C+.+4 4<.4, 1&%# 2%&$8 +.8+ +3 d/ 3$ at +3 d/ +,.4@ #&%8 22&$" 3.:+ 4;.,, #&%8 21&%8 C@.+@ +@ d/ 3$ at +@ d/ +,.4< 9&1# "%&8# C@.43 4;.@, #&%8 2#&29 C+.<6 ; d/ 3$ at ; d/ +;.@; 9&1# "1&$1 @.<@ 44.6, 9&1# 29&"9 +.:8 6 d/ 3$ at 6 d/ :.;+ 9&1# "8&"3 3.64 +4.43 9&1# "0&03 C+.<4 Ta.le A3$4-3: 47ample of test result dispersion for 47periment 3. !/alf rate in ;ffice :oise+ The followin tables summari)e the performances of the different .M$ codec modes under the tested error conditions with respect to wellC"nown references in half rate mode under bac"round noise conditionsD 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 3% Re$erence , Re$erence 0 Re$erence 1 ER .o Errors R .o Errors (see 'ote below) #R .o Errors Codec Mode E+uivalent to Re$erence , E+uivalent to Re$erence 0 *orse than Re$60 Better than Re$61 E+uivalent to Re$erence 1 *orse than Re$61 :6<8 'o 2rrors down to +: d/ 0>& +3 d/ 0>& +@ d/ 0>& S below :67 'o 2rrors +: d/ 0>& +3 d/ 0>& +@ d/ 0>& S below 96: 'o 2rrors +: to +3 d/ 0>& +@ d/ 0>& S below 86< 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& ; d/ 0>& S below 86,8 'o 2rrors down to +3 d/ 0>& +@ d/ 0>& ; d/ 0>& S below 76:8 'o 2rrors down to +@ d/ 0>& ; d/ 0>& S below Ta.le A3$4-4: AM speech codec mode performances under .ac85round noise conditions in half rate Re$erence , Re$erence 0 Re$erence 1 ER .o Errors R .o Errors (see 'ote above) #R .o Errors Codec Mode E+uivalent to Re$erence , E+uivalent to Re$erence 0 *orse than Re$60 Better than Re$61 E+uivalent to Re$erence 1 *orse than Re$61 .o Errors ;.,8- ;.6- :.; 8., 8.+8 6.;8 ,9 dB C-I ;.,8 ;.6- :.;- 8., 8.+8 6.;8 ,1 dB C-I 8.,- :.; 8.+8 6.;8- ;.,8- ;.6 ,= dB C-I 8.,- 8.+8- 6.;8 ;.,8- ;.6- :.; : dB C-I 6.;8 2Buivalent to 3$ at ; d/ 0>& 7 dB C-I 8.+8 S 6.;8 better than 3$ at 6 d/ 0>& Ta.le A3$4-': Performances of the AM speech codecs for different error conditions under .ac85round noise conditions in half rate 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4* 1nne9 D6 1@R <e&;o&man!es as a ;+n!,.on o; 25R an' RA5R &n this anne(- the characteri)ation test results are charted as a function of the 3rame 2rasure $ate (32$) or $esidual /it 2rror $ate ($/2$) as measured for each 2rror Pattern used for the subjective listenin tests. They are provided as an indication of the Buality deradation to be e(pected for the implementation of the .M$ speech codec in 3G networ"s. &n the followin diarams- the Buality deradation is e(pressed in MO! (or 5MO!) obtained by comparin the MO! (or 5MO!) obtained by the different codecs for each impairment condition with the MO! (or 5MO!) obtained by the 23$ in 2rror 3ree in the same e(periment. The results were compiled as e(plained belowD C &n all cases- the results represent the averae scores obtained over all tests performed for each e(periment as compiled in F5+G C The reference is always 23$ in 2rror 3ree as measured in the same e(periment. C The charts in clean speech (3iures 5+aC5+d) were obtained from the 0haracteri)ation test results for 2(periments +a and +b (Test performed by .TST and /er"om) C The charts in 0ar 'oise (3iures 5.aC54d) were obtained from the 0haracteri)ation test results for 2(periments 3b and 3e (Test performed by 3rance Telecom and 0one(ant) C The charts in !treet 'oise (3iures 53aC53d) were obtained from the 0haracteri)ation test results for 2(periments 3a and 3d (Test performed by 3rance Telecom and 0one(ant) C The charts in Office 'oise (3iures 56aC56e) were obtained from the 0haracteri)ation test results for 2(periments 3c and 3f (Test performed by 3rance Telecom and 0one(ant) C &n all cases- the actual results were manually altered to smoothen the shape of the curves. C The reference 32$ and $/2$ were e(tracted from F4G (document prepared in +4>,< for the selection of the .M$ 0hannel 0odin scheme). &t should also be noted that the diarams function of the 32$ are affected by the $esidual /it 2rror $ate for each test condition- while the diarams function of the $/2$ are also function of the 32$ present for each test condition. The two sets of diarams cannot be considered totally independent. 3inally- it should be pointed out that the 32$ and $/2$ estimates used to derive these diarams are based on the limited number of error patterns used for the .M$ characteri)ation phase. These could be affected by some inaccuracies that could e(plain the difference in shapes between the different speech codec modes. These results can also be compared to previous indications provided by !6 to $+ and !4 reardin the robustness of the .M$ !peech 0odec ($ef F3G and F6G). The followin section is e(tracted from a Aiaison !tatement sent to $+ F3G- the same reference is also used in F6G (Aiaison to !4)D The frame error rate re:uired for producin( hi(h speech :ualit- 'ith onl- small :ualit- de(radation compared to error free speech is t-picall- 2E* ; 8&#<& This re:uirement (uarantees retainin( the maximum :ualit- of= e&(&= the !S E2* codec& The :ualit- then de(rades (racefull- 'ith increasin( frame error rate& This 2E* limit should be considered as a conser,ati,e fi(ure& 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4) D1. Res+/,s .n 4/ean See!0 .n @?S6 Percei(ed 3ualit6 !M;S+ de5radation as a function of the 14 !1 Tests in Clean Speech+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !M;S+ de5radation as a function of the 14 !/ Tests in Clean Speech+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >)a: Cualit6 >e5radation function of 14 1i5ure >).: Cualit6 >e5radation function of 14 !1 Test esults+ !/ Test esults+ Percei(ed 3ualit6 !M;S+ de5radation as a function of the 24 !1 Tests in Clean Speech+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !M;S+ de5radation as a function of the 24 !/ Tests in Clean Speech+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >)c: Cualit6 >e5radation function of 24 1i5ure >)d: Cualit6 >e5radation function of 24 !1 Test esults+ !/ Test esults+ 0omments on the previous resultsD &n clean speech- it appears that all codec modes do not show any sinificant Buality deradation when the 3rame 2rasure $ate is lower than @.8N. &n some instances- the rane can even be e(tended to +N 32$ without any Buality deradation. &t is also interestin to note that at +N 32$ deradation- the hihest codec modes (+4.4 and +@.4) are still eBuivalent to the second tier of codec modes (;.,8 to 8.,) in error free. !imilarly- the middle rane codec modes (;.,8 to 8.,) present the same Buality at +N 32$ than the lower rate codec modes (8.+8 S 6.;8) in error free conditions. The e(periments in 7alf $ate have slihtly increased the differences between the codecs and with 23$ as could have been e(pected- but the same trends can be observed. The results as a function of the $/2$ are also very similar with a different rane of acceptable $/2$. The different codec modes do not present any sinificant Buality deradation when the $/2$ is below @.+N. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4" D2. Res+/,s .n 4a& 3o.se6 Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !1 Tests in Car :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !/ Tests in Car :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >"a: Cualit6 >e5radation function of 14 1i5ure >".: Cualit6 >e5radation function of 14 !1 Test esults+ !/ Test esults+ Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !1 Tests in Car :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !/ Tests in Car :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >"c: Cualit6 >e5radation function of 24 1i5ure >"d: Cualit6 >e5radation function of 24 !1 Test esults+ !/ Test esults+ 0omments on the previous resultsD &n car noise- no sinificant deradation is observed when the 32$ stays below +N and the difference in Buality between the different codecs is slihtly amplified compared to the results clean speech. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 43 D3. Res+/,s .n S,&ee, 3o.se6 Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !1 Tests in Street :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !/ Tests in Street :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >3a: Cualit6 >e5radation function of 14 1i5ure >3.: Cualit6 >e5radation function of 14 !1 Test esults+ !/ Test esults+ Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !1 Tests in Street :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !/ Tests in Street :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 7.95 >R 7.8 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >3c: Cualit6 >e5radation function of 24 1i5ure >3d: Cualit6 >e5radation function of 24 !1 Test esults+ !/ Test esults+ 0omments on the previous resultsD The results in street noise are in line with the previous results. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 44 D8. Res+/,s .n ?;;.!e 3o.se6 Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !1 Tests in ;ffice :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 14 !/ Tests in ;ffice :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K 100.000K (ER #M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >4a: Cualit6 >e5radation function of 14 1i5ure >4.: Cualit6 >e5radation function of 14 !1 Test esults+ !/ Test esults+ Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !1 Tests in ;ffice :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 12.2 10.2 7.95 2R 7.8 2R 6.7 2R 5.9 2R 5.15 2R 8.75 2R Percei(ed 3ualit6 !>M;S+ de5radation as a function of the 24 !/ Tests in ;ffice :oise+ -3.00 -2.50 -2.00 -1.50 -1.00 -0.50 0.00 0.50 0.001K 0.010K 0.100K 1.000K 10.000K R.ER #M! 7.95 >R 7.8 >R 6.7 >R 5.9 >R 5.15 >R 8.75 >R 1i5ure >4c: Cualit6 >e5radation function of 24 1i5ure >4d: Cualit6 >e5radation function of 24 !1 Test esults+ !/ Test esults+ 0omments on the previous resultsD !ame comment for the results in Office 'oise Re;e&en!es ,o 1nne9 D6 F5+GD .M$ 0haracteri)ation 0ombined Test $esults (spreadsheet)D !MG++ Tdoc 463>,,- !MG++M+@- Wune 6C++- +,,,- Tampere- 3inland F54GD .nne( 3 to the A! to !MG4 1P/ on alternative .M$ channel codin schemesD JObjective test results for alternative .M$ channel codin schemesK from 2ricsson>'o"ia>!iemens !MG++ Tdoc 34,>,<- !MG++M</is- 5ecember +;- +,,<- Aondon 7eathrow- L% F54GD !6 A! to T!GC$+ J$esponse to the T!GC$+ A! on !peech !ervicesK Tdoc +<8$>,,- T!GC!6M3- March 46C4:- +,,,- Uo"osu"a- Wapan F56GD !6 A! to T!GC!4- !4 9o! and $3 J2rror resilience in realCtime pac"et multimedia payloadsK Tdoc +;,$>,,- T!GC!6M8- Wune +6C+:- +,,,- Miami- 3ACL!. 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4' >.s,o&$ Document histor) v+.@.@ 5ecember +,,, /ased on G!M @:.;8 v;.@.@- Presented to T!GC!.M: for information v+.+.@ Wanuary 4@@@ Modification of document number to 4:.,;8 &ntroduction on Thresholds and 7ysteresis for 2(periment 6a S 6b &ntroduction of .nne( 5 3GPP 3G T "#$%&' ,)$)$* !"***-*)+ 4#
3rd Generation Partnership Project Technical Specification Group Radio Access Network Manifestations of Handover and SRNS Relocation (3G TR 25.832 Version 3.0.0)