1. Improving Voice Quality in International Mobile-to-Mobile Calls Aram Falsafi, Seattle, WA aram@aramfalsafi.com PIMRC 2008 18 September 2008

2. PIMRC 2008Aram Falsafi2 Presentation Outline Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

3. PIMRC 2008Aram Falsafi3 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

4. PIMRC 2008Aram Falsafi4 Mobile-to-Mobile Call E1-E3: Voice encoders D1-D3: Voice decoders Note: In the interest of simplicity, nodes that don’t affect the voice format (MSC and intermediate switches) are not shown.

5. PIMRC 2008Aram Falsafi5 Mobile-to-Mobile call with extra vocoders DCME: Digital Circuit Multiplication Equipment VoIP: Voice over IP Note: In the interest of simplicity, nodes that don’t affect the voice format (MSC and intermediate switches) are not shown.

6. PIMRC 2008Aram Falsafi6 International Mobile-Mobile Call: Up to five tandem vocoder pairs Voice quality degrades quickly with increased number of tandem vocoders Increase not necessarily linear Similar to effect of retransmission of a FAXed document (quantization error) Solution: Minimize number of tandem vocoders Tandem Free Operation (TFO) is already part of the 3GPP and 3GPP2 standards, for 2G and 3G equipment.

7. PIMRC 2008Aram Falsafi7 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

8. PIMRC 2008Aram Falsafi8 Mobile-to-Mobile Call

9. PIMRC 2008Aram Falsafi9 Mobile-to-Mobile Call (TFO setup) In-band TFO Negotiation

10. PIMRC 2008Aram Falsafi10 Mobile-to-Mobile Call with TFO Reduction from 2 vocoder stages down to 1.

11. PIMRC 2008Aram Falsafi11 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

12. PIMRC 2008Aram Falsafi12 International Mobile-Mobile Call

13. PIMRC 2008Aram Falsafi13 International Mobile-Mobile Call

14. PIMRC 2008Aram Falsafi14 International Mobile-Mobile Call Reduction from 3 vocoder stages down to 1! Significantly reduced delay Much lower-cost DCME/VoIP equipment

15. PIMRC 2008Aram Falsafi15 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

16. PIMRC 2008Aram Falsafi16 Requirements A “clean” 64 KBPS PCM file Serves as voice quality reference file Long enough to give a good estimate Encoders and decoders for standard GSM and UMTS networks For full rate (FR), enhanced full rate (EFR), half rate (HR), and different rates of adaptive multirate (AMR) Must be host-based Available in open source format Objective voice quality measurement software

17. PIMRC 2008Aram Falsafi17 Perceptual Evaluation of Speech Quality (PESQ) Standardized as ITU-T recommendation P.862 (*) Uses psycho-acoustic metrics – attempts to measure distortions only in the perceptually significant aspects of the voice signal Calculates an objective voice quality score, and maps it to “a MOS like scale, a single number in the range of –0.5 to 4.5, although for most cases the output range will be between 1.0 and 4.5, the normal range of MOS values …”(*) (*) ITU-T Rec. P.862, “Perceptual Evaluation of Speech Quality (PESQ), an Objective Method for End-to-End Speech Quality Assessment of Narrowband Telephone Networks and Speech Codecs,” Feb 2001

18. PIMRC 2008Aram Falsafi18 Instrumentation 3-minute sample file used PESQ software takes 2 input files & generates a single output value

19. PIMRC 2008Aram Falsafi19 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

20. PIMRC 2008Aram Falsafi20 Effect of Tandem Vocoders Applying the same vocoder multiple times Shows the effect of tandem vocoding Simple to simulate, but not common in the real world

21. PIMRC 2008Aram Falsafi21 Two Scenarios Case 1: High Voice Quality Requirement E2/D2 uses AMR12.2 vocoder For example Tier-1 operator in mature market Case 2: High Bandwidth Savings Requirement E2/D2 uses AMR4.75 vocoder For example severely congested backhaul, or satellite links

22. PIMRC 2008Aram Falsafi22 Case 1 – High Voice Quality With TFO: PESQ score of 3.93 (single AMR12.2 vocoder end to end) Without TFO:

23. PIMRC 2008Aram Falsafi23 Case 2 – High Bandwidth Savings With TFO: PESQ score of 3.41 (single AMR4.75 vocoder end to end) Without TFO:

24. PIMRC 2008Aram Falsafi24 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A

25. PIMRC 2008Aram Falsafi25 Conclusions Tandem vocoders can significantly impact voice quality And add to the total delay – an effect that has not been quantified in this study TFO has been created to address this problem in mobile to mobile calls TFO can be applied to other scenarios where tandem vocoders (including standard VoIP and proprietary DCME) are in the transmission path This has many advantages Measureable improvements in voice quality Lower delay Cheaper transmission equipment – fewer vocoders

26. PIMRC 2008Aram Falsafi26 Introduction & Problem Statement Current State of the Art Description of Proposed Architecture Instrumentation Methodology Results Conclusion Q & A