Research On The Application Of G.729B Codec And Echo Cancellation On VoIP Gateway

With the development of IP (Internet Protocol) technology, the VoIP (Voice over Internet protocol) technology which can reduce the cost of communication has been widely applied all over the world. One of the key problems of VoIP is how to transmit the high quality speech signal in the limited bandwidth. For solving the problem, this paper studies and does research on two core technologies of VoIP: speech codec and echo cancellation.ITU-T G.729 is one of the most efficiently and broadly used standards that can process speech signal. Based on G.729, G.729B presents voice active detection (VAD) technology. Firstly, this paper introduces the algorithm structure of G.729B, focusing on the LP analysis, transformation of LP to LSP coefficients, pitch detection, searching of adaptive and fixed codebooks, VAD technology, etc.Secondly, the design of Echo Cancellation system based on double talk detection is presented. The coefficients of NLMS adaptive filter are adjusted according to the double talk status. This paper presents an improved Geigel algorithm by dividing the input signal X N( n ). The improved algorithm is more stable than normal Geigel algorithm. Combined with G.729B coder, the improved algorithm uses the results of VAD of past frames to improve the effect of echo cancellation.Finally, the design and realization of the high capacity Voice Trans Codec Board are described. This board uses TMS320C64X DSP as the major apparatus to apply G.729B codec and echo cancellation. In the aspect of hardware, the device of Power, Clock, outer-expanded SDRAM, EMIFB, McBSP and HPI of DSP are designed. In the aspect of software, echo cancellation and G.729B codes are programmed and optimized with multiple methods.The experimental results show that the optimized programs have reached the expectation in data storage and operational complexity. The distortion of the decoded speech is rather small. With the improved Geigel algorithm, the echo canceller can do better than classical Geigel algorithm and improve the quality of the speech.