The Reaserch On Algorithms Of Packet Loss Compensation In VoIP

With the rise and wide application of internet, VoIP (Voice over Internet Protocol) is newbusiness, as an IP telephony-based, and using the Internet technology for voicecommunication. In recent years, VoIP acquired rapid development and application, which iscontinually tapping into the market of traditional telephone service, and would replace it.However, IP networks provide a “best effet”service, which is not quality of service (QoS)guarantee and can’t guarantee the reliability of the network transmission. Therefore, the standor fall of network would inevitably cause jitter, delay and packet loss of speech signal and soon, and the packet loss is the main factors to influence the quality of Internet phone, and slovethis problem is the key to improve the quality of the voice.In order to solve the problem of packet loss in VoIP, now, domestic and internationalresearchers have developed a lot of packet loss process technology, which has played arelatively good results to make up for the lost packet. This article mainly does research in thepacket loss compensation algorithm based on the receiving end, from the voice signalgenerated models of linear and nonlinear to analyse, first of all, it research wavefrominsteading packet loss compensation algorithm based on linear theory, which is according toshort-term correlation of speech signals, to repair the missing wavefrom in the lost packetusing the speech wavefrom of the correctly received packet before the packet loss. To makethe speech wavefrom smooth and continuous in the wavefrom cross, it use the OLAalgorithm to realize it. And then, take the linear prediction analysis of speech signal,researching a packet loss compensation algorithm based on improved linear prediction model.The algorithm based on linear prediction calculates the correlation coefficient, LP coefficientsand estimate the pitch period and pitch period prediction signal according to the samples inthe buffer. Using the improved linear prediction model to estimate the samples of the lostpacket, and thus it achieve the speech packet loss compensation. The above studies are basedon linear theory analysis, but the speech signal generation is nonlinear, the speech signal is achaotic, so just from the point of view of linear analysis, it can’t predict the lost speech signalaccurately. Therefore, this article analyses the speech signals from the perspective of thenonlinear theory, taking phase space reconstruction for speech signals, combined with the theory of neural networks, it propose a packet loss compensation algorithm based on RBFneural network. From the theoretical analysis, the algorithm can get better effects forprocessing the packet loss in VoIP.Finally, the article build network simulation environment based on NIST Net software, inthe network simulation environment which has been builded, based on the UDP protocol toestablish Socket Communications and transfer a standard audio file, and take packet losscompensation for the received speech signal at the receving end, and then test out the PESQof the speech files after compensation,thus take analysis and evaluation of performance to allsorts of packet loss compensation algorithm. Experimental results show that the packet losscompensation algorithm based on RBF neural network has better compensation effect toslove the problem of continuous speech packet loss than the packet loss compensationalgorithm of the linear prediction.