1. 3GPP2 Evolution Workshop - 2005 Multimedia Codecs and Protocols 3GPP2 TSG-C SWG1.2

2. Multimedia in PS Domain Transitioning from CS to PS –Shared channels as opposed to dedicated channels –User expectation: Similar to that with dedicated channels –Application Level Framing (ALF) in IP –Requires special considerations for wireless hops Codec and transport related challenges –IP fragmentation –Jitter in packet arrival –Codecs’ view of QoS: –Greater need for error resiliency and concealment –…

3. TSG-C SWG1.2 “Multimedia Services” ToR Evaluation and adoption of –Visual Codecs (Video, Image, Graphics), –Audio Codecs (synthetic and natural audio) –Audio/visual synchronization methods Evaluation and adoption of complementary protocols –RTSP/SDP, RTP/RTCP, to support real-time, near real-time and interactive multimedia services Development of fileformats to support –over the air transmission for streaming, pseudo-streaming, downloading –associated methods for content protection

4. IP Terminology ADU – Application Data Unit –Independently decodable data; minimum unit of error recovery –e.g. a vocoder frame, a video slice, … SDU – Service Data Unit –One or more ADUs, including required headers for transmission MTU – Maximum (size of a) Transmission Unit, in octets –e.g. 1500 octets on an Ethernet hop, 576 octets on PPP, … Fragmentation and Re-assembly –1 SDU  n PDU mapping, when SDU exceeds MTU on a given hop PDU – Protocol Data Unit –In this context, we consider physical layer frames –payload size in one physical layer frame –Not including MAC, CRC and tail bits overhead Observation: Significant fragmentation over cdma2000® hops

5. IP Fragmentation

6. “Goodput” above RTP From RFC2736, Best Current Practice for RTP payload design, –… relying on IP fragmentation is a bad design strategy as it significantly increases the effective loss rate of a network and decreases goodput –… if one fragment is lost, the remaining fragments (which have used up bottleneck bandwidth) will then need to be discarded by the receiver Pick ADU/SDU sizes not to exceed PDU sizes Do not concatenate SDUs (use stuffing bits)

7. What is optimal ADU size? One size does not fit all –Except for Constant bitrate (CBR) Channels and CBR codecs Most 3G networks provide multiple PDU sizes to support “Bandwidth on Demand” Entropy in multimedia content (video, speech, Audio, …) is not constant over a TTI –Variable bitrate (VBR) encoding: improved compression efficiency Examples –cdma2000 vocoders –Channel: RS-1 and RS-2 support 4 PDU sizes in addition to dtx –Source: cdma2000 vocoders generate 4 ADU sizes, based on source characteristics –SO60: ADU = PDU; CS like efficiency in PS! –ITU and MPEG Video codecs –VBR rate control constrained to generate ADUs of known size –Improved error resiliency when ADUs are matched to PDUs Significant statistical multiplexing gains when codecs are optimized to lower layers Challenges: Optimize 3G services, without cross-layer violations and maintain interoperability with other networks

8. Jitter in Packet Arrival

9. Jitter in PS networks Variance in the ADU inter-arrival time –SDUs queued behind cross-traffic in routers –Routing changes in end-to-end path De-jitter buffer –a buffer to restore constant inter-packet timing prior to decoding –Increases playout delay 20ms De-jitter buffer Decoder Even delivery of voice frames 20ms of voice per frame per frame 20ms Variable Delay, in-order delivery not guaranteed

10. Example: Jitter in packet arrival times

11. Example: Packet Loss due to Delay bound

12. Smart Decoders Reduce playout delay with variable rate rendering –Speech: Signal processing tricks such as “Time Warping” –Video: Audio visual synchronization (lip-synch) Improved concealment techniques to mitigate packet loss –Interpolation in time and/or frequency domains Challenges –Develop improved objective metrics and subjective evaluation methodologies –Study and characterize the effects of such enhancements on user experience GoodErasureGood Playback Time line CircuitVoIP

13. RTP Packet Loss Simulator SWG1.2 Simulation methodology to characterize multimedia content in cdma2000 wireless IP

14. Proposed Video Objective Metrics Foreman example Avg PSNR [dB] STD PSNR [dB]pDVD [%] Bitrate [Kbps ] Clean 32.350.80NA64.69 1.5% FER : Scheme A 26.927.0945.6764.69 1.5% FER : Scheme B 28.986.1135.6766.60 1.5% FER : Scheme C 32.032.573.6767.11 1.5% FER : Scheme D 31.573.518.3367.86 Percentage Degraded Video Duration (pDVD) Standard deviation of PSNR (STD_PSNR)

15. PSNR Traces: Foreman

16. Summary Optimize codecs and bearer protocols to lower layers for efficient IP services –e.g. VBR codecs and VBR channels in 3GPP2 CS voice –Rate control and RTP packetization to maximize “goodput” –Interoperability with other IP networks (cross-layer issues) Develop smart codecs –Improved error resiliency –Better concealment techniques –Variable rate rendering Develop/adopt suitable objective metrics and subjective evaluation methodologies –Characterize user experience in realistic 3GPP2 environments