Implementation Of Middle-Bitrate Audio Codecs On Fixed Point DSP Chip In VoIP System

VoIP is one of the hottest topics of computer network and communication technologies and one of the fastest growing Internet businesses at present. It is to transport speech signals in the form of packets through the IP packet switched network. Different from traditional circuit switched network, there are several problems exist in the packet switched network, such as limited bandwidth, packet loss and delay jittering, so we need to analyze and realize audio coding/decoding algorithms which are more suitable for the packet switched network, add some extra modules to deal with different network situations and implement these algorithms on IP phones in VoIP system. By analyzing different kinds of audio algorithms, we found that audio codecs such as ILBC and Speex had a low bit-rate and several modes to be selected according to different network situations; they also add some extra modules such as packet loss concealment and de-jittering. So these codecs are very suitable to transmit speech signals through internet. In addition, ILBC and Speex algorithms are open source and free, so they have great business value.According to the classifying standard of audio codecs, codec with bit-rate between 4.6kb/s～24kb/s is called middle-bitrate codec. So ILBC, G729 and most modes of Speex are middle-bitrate codecs. In this paper we aimed at analyzing and realizing middle-rate audio codecs such as ILBC, G729 and Speex which were better for the packet switched network with the simulating and developing tools such as PalmADSP, and Visual C++. Through the process of converting float point C codes to fixed point C codes, translating C codes to DSP codes, embedding the codes into DSP chips and executing system testing and optimization, we finally implemented these algorithms on DSP chips. During practical work, we solved two problems: how to choose a proper fixed point transforming scheme to assure data range and precision and how to enhance the efficiency of code execution while compressing the data and code space as much as possible for the limited data and program memory of DSP chip. Finally in this paper, we summarized some general float-point codes to fixed-point codes transforming methods through the fixed-point code transforming work of ILBC algorithm and put forward some useful C codes to DSP codes translating and optimizing methods for the development of DSP applications. Results of implementation on AR168G phones indicate that audio algorithms realized in the VoIP system could adapt to different network situations and gain good communication quality.