Lost VOIP Packet Recovery in Active Networks

Transcription

1 Lost VOIP Packet Recovery in Active Networks Yousef Darmani M. Eng. Sc. Sharif University of Technology, Tehran, Iran Thesis submitted for the degree of Doctor of Philosophy m Department of Electrical and Electronic Engineering The University of Adelaide Supervisors: Professor Langford White Mr. Michael Liebelt June 2004

2 Contents Abstract xv Statement of Originality xvii Acknowledgments xix Publications xxi 1 Introduction Internet telephony. I 1.2 Problem statement Thesis outline and methodology 4 2 Packet loss models and metrics 2.1 Packet loss Causes of packet loss Loss rate and packet length Statistical information about packet loss Loss models n-order Markov chain Bernoulli model Gilbert model Extended Gilbert model IS Poisson Model Model evaluation Summary Voice Over Internet Protocol (VOIP) Properties of speech Voice quality measurement Voice Over Internet Protocol (VOIP) 22

3 3.4 Methods to cope with packet loss problem Sender and receiver based scheme Sender based scheme Receiver based recovery scheme Network and receiver based recovery scheme Packing methods used in the simulations Quality of Service (QoS) QoS model Protocol Internet Protocol (IP) User Datagram Protocol (UDP) Real-time Transport Protocol (RTP) RTP/UDPIIP Other protocols for VOIP Summary Active node Reasons to use active networks Active and passive networks Active node and lost packets Active path Virtual Private. letwork (VPN) Virtual Active Path (VAP) Active Packet Active node problems Active node in VOIP Node security Node structure Node resource reservation Node software specifications Node software structure for voice packet recovery Role of active node First analysis Two consecutive losses n consecutive losses Second analysis Two consecutive losses n consecutive losses 60

4 4.10 Third analysis 4.11 Active node location 4.12 Simulation and test environment Active packet structure in the simulations Active node implementation Simulation topology 4.13 Summary and conclusion Detached Samples Packing Method (DSPM) DSPM sender routine Packing Routine Main Routine Receiver or active node routine Protocol specification and verification Finite State Machine (FSM) model Specification and Description Language (SDL) model Results of DSPM Summary and conclusion 91 6 Detached Adaptive Differential Pulse Code Modulation (DADPCM) DADPCM sender routine First step: Separate samples Second step: Encoding and decoding samples Third step: packing the encoded samples Receiver routine Active node routine Results ofdadpcm Voice quality Data rate Processing time Important samples Fixed-sized packets Fixed distortion Next correlation Modified fixed distortion Summary and conclusion III

5 7 Detached Adaptive Differential Pulse Code Modulation 5 (DADPCMS) Sender routine First step: Separate samples Second step: Encoding and decoding samples Third step: Packing the encoded samples Receiver routine No loss Individual loss Heavy hurst loss Active node routine Results ofdadpcm Quality versus data rate Quality versus loss rate Processing time Summary and conclusion Detached Fourier Coeffcients Packing Method (DFCPM) Discrete Fourier Transform Decimation in time Motivation Compression: Lost packet effect compensation: Experiments First experiment Second experiment Fourth experiment Fifth experiment Detached Fourier Coeffcient Packing Method (DFCPM) Sender routine Receiver routine Active node routine Threshold Result of DFCPM Enhanced Detached Fourier Coeffcient Packing Method (EDFCPM) Sender routine Receiver routi n e Result ofedfcpm. 146

6 8.7 Summary and conclusion Enhanced Code Excited Linear Prediction (ECELP) Conjugate Structure Algebraic Code Excited Linear Prediction (CS-ACELP) Encoder Decoder ECELP ECELPI ECELP Redundancy and loss rate ECELP in active node Results of ECELP Individual loss Burst loss Summary and conclusion Complimentary Adaptive Differential Pulse Code Modulation (CADPCM) CADPCM Packing scheme Unvoiced block Voiced and Unvoiced2voiced blocks Voiced2unvoiced block Redundancy and loss rate CADPCM in active node CADPCM at the receiver Result ofcadpcm Voice quality Data rate Processing time Summary and conclusion Conclusion Contributions 11.2 Methodology 11.3 Major results 11.4 Future directions A SDLs ofdadpcm5 177

7 B Source codes 179 C Voice samples and test environment C.I Voice samples C.2 Test environment 184 Bibliography 185

8 Abstract Current best-effort packet-switched Internet is not a perfect environment for real-time applications such as transmitting voice-over the network (Voice Over Internet Protocol or VOJP). Due to the unlimited concurrent access to the Internet by users, the packet loss problem cannot be avoided. Therefore, the VOIP based applications encompass problems such as "voice quality degradation caused by lost packets". The effects of lost packets are fundamental issues in real-time voice transmission over the current unreliable Internet. The dropped packets have a negative impact on voice quality and concealing their effects at the receiver does not deal with all of the drop consequences. It has been observed that in a very lossy network, the receiver cannot cope with all the effects of lost packets and thereby the voice will have poor quality. At this point the Active Networks, a relatively new concept in networking, which allows users to execute a program on the packets in active nodes, can help VOIP regenerate the lost packets, and improve the quality of the received voice. Therefore, VOIP needs special voice-packing methods. Based on the measured packet loss rates, many new methods are introduced that can pack voice packets in such a way that the lost packets can be regenerated both within the network and at the receiver. The proposed voice-packing methods could help regenerate lost packets in the active nodes within the network to improve the perceptual quality of the received sound. The packing methods include schemes for packing samples from low and medium compressed sample-based codecs (PCM, ADPCM) and also include schemes for packing samples from high compressed frame-based codecs (G.729). Using these packing schemes, the received voice has good quality even under very high loss rates. Simulating a very lossy network using NS-2 and testing the regenerated voice quality by an audience showed that significant voice quality improvement is achievable by employing these packing schemes.