1. Page 1 The department of Information & Communications Engineering Dong-uk, kim ghost@hufs.ac.kr A Survey of Packet Loss Recovery Techniques for Streaming Audio - Colin Perkins, Orion hodson, and Vicky hardman - Univ. college London Univ. college London IEEE Networks, Sep/Oct, 1998 IEEE Networks, Sep/Oct, 1998

2. Page 2 Table of Contents (1/1)  Overview  Multicast Channel Characteristics  Loss Repair techniques  Conclusion (Recommends)

3. Page 3 Table of Contents (1/2)

4. Page 4 The quality / complexity trade off 1. Splicing 2. Silence substitution 3. Noise substitution 4. Repetition 5. Repetition with Fading  The use of packet repetition with fading is recommended as offering a good compromise between achieved quality and excessive complexity 6.Waveform Substitution 7.Pitch waveform substitution 8.Time-scale modification 9.Interpolation of Transmitted state 10.Model Based Recovery

5. Page 5 Multicast Channel characteristics (1/2)  Indirection –Sender does not know the set of hosts which will receive a packet

6. Page 6 Multicast Channel characteristics (2/2)  Relatively high latency  Univ. Origon  Univ. London

7. Page 7 Forward Error Correction (Media-Independent)  Parity Code  Reed Solomon Code –original message is split into fixed length blocks –each block is split into m bit symbols RS(n,k) encoder takes k data symbols of m bits each, appends n - k parity symbols, and produces a code word of n symbols ( each of m bits).

8. Page 8 Forward Error Correction (Media-Specific)  Primary encoding –First transmitted copy of the audio data  Secondary encoding –Subsequent transmissions

9. Page 9 Congestion Control  The FEC repair data to a media stream is an effective means by which that stream maybe protected against packet loss –Should aware that Addition of large amounts of repair data will increase network congestion and hence packet loss  Network heterogeneity cause –Different sets of receivers Low-capa regions : Congestions High-capa regions: underutilized  Need layered encoding

10. Page 10 Interleaving

11. Page 11 Retransmission  SRM (Scalable reliable multicast) –A member of an session detects loss, will wait a random time (determined by distance from sender) and then request repair packet  Are not generally suitable for streaming media such as audio –They do not bound the transmission dealy –TCP has similar reasons

12. Page 12 Error Concealment  Producing a replacement for a lost packet which is similar to the original –These technique work for Relatively small loss rates ( < 15 percent) Small packets (4~40ms) Loss length approaches the length of phoneme(5~100ms) –Insertion based –Interpolation based –Regeneration based

13. Page 13 Insertion based Repair  Splicing –No gap is left due to a missing packet –Low loss rates and short clipping lengths (4~16ms) faired best –But Performs poorly  Silence Substitution –Fill the gap with silence –Short packet length (4ms) and low loss rate (2 percent) making suitable for interleaved audio over low-loss path  Noise Substitution –Filling in the gap left by a lost packet with silence, background noise is inserted instead.  Repetition –Replaces lost units with copies of the unit that arrived immediately before the lost –Subjective quality can be improved by gradually fading repeated units

14. Page 14 Interpolation Based Repair

15. Page 15 Recommendations (1/3)  Non-Interactive Applications –One –to- many transmissions (Radio broadcast) Latency is of considerably less importance then quality Receiver set is likely to be diverse and the group may include members behind low-speed links –Interleaving is compatible Some form of error concealment will still be need to compensate –Repetition with fading is acceptable –Retransmission-based repair is not appropriate A media independent FEC scheme will perform better than a retransmission based repair scheme

16. Page 16 Recommendations (2/3)  Interactive Applications –IP Telephony The principal concern is minimizing end – end delay Such delay imposed by use the –Interleaving –Retransmission –Media-independent FEC  not acceptable for interactive applications –Recommends media-specific FEC Tunable bandwidth overhead Repair is approximate due to the use of low-rate secondary encodings

17. Page 17 Recommendations (3/3)  Error concealment –Receivers must be prepared to accept some loss in an audio stream Current(1998)  use silence substitution to fill the gaps left by packet loss –Does not provide acceptable quality Recommends “Packet repetition “ –Simple to implement and having low computational overhead Other schemes discussed provide incremental improvements