DSP subsystem for multiparty conferencing in VoIP

This paper deals with different types of configurations and network-architecture models used for multi-party conferencing in VoIP. A comparison of multi-party conferencing models is presented based on the bandwidth used by the Real Time Protocol (RTP) streams, quality degradation due to trans-coding if any, etc. Also, an efficient and simple architectural design of a media engine for decentralized conferencing with an end-point manager is presented. The investigation on the right placement of voice-enhancement modules in the media-engine is presented by analyzing the real time captured data. The proposed architectural design is extended for inter-operating with different VoIP-based terminals, which operate at different sampling rates such as 8 kHz, 16 kHz, 24 kHz etc. The proposed conferencing model contains ITU-T G.722, ITU-T G.722.1 Wideband codecs, Echo Cancellation (EC), Noise Cancellation (NC), Automatic Gain Controller (AGC) and other VoIP components implemented for IP-Phones running on desktop or laptop computers. The proposed system is able to connect up to ten conferees on a computer that has 1.6 GHz speed, 1GB RAM memory and Linux/Window 2000 or later operating system with wideband (16 kHz) voice quality.