Joint source-channel coding for packet Network Transmission of low bit rate wide Band Speech

Federico M. Ang

(MS Graduated: 2nd Sem 2008-2099)

Abstract

The value of a speech codec in voice communication systems lies on its robustness against degradations in the channel. In the case of packet network communication such as the Internet, resilience from error is compounded by the additional problem of packet erasures, aside from the usual noise. In this research, the performance of two (2) low bit-rate wideband speech coders were investigated when transmitted over a channel subjected to noise, fading, and packet losses. The two codecs are the Wideband Linear Predictive (WLP) and Wideband Sinusoidal (WS) codecs. The design of the transmission model used source-controlled channel decoding. The joint source-channel coding scheme for error correction and concealment thus consists of the two codecs and the source-controlled channel decoder. It can be shown that as much as 5 dB of coding gain can be achieved from objective measurements such as bit error rate, root-mean-square errors and spectral distortion. We conclude that the joint source-channel scheme is better than the conventional packet error concealment of the GSM standard.

Subject Index : Computer networks, E-speak (Computer network technology)