An approach for improving performance of aggregate voice-over-IP traffic

The emerging popularity and interest in Voice-over-IP (VoIP) has been accompanied by customer concerns about voice quality over these networks. The lack of an appropriate real-time capable infrastructure in packet networks along with the threats of denial-of service (DoS) attacks can deteriorate the service that these voice calls receive. And these conditions contribute to the decline in call quality in VoIP applications; therefore, error-correcting/concealing techniques remain the only alternative to provide a reasonable protection for VoIP calls against packet losses. Traditionally, each voice call employs its own end-to-end forward-error-correction (FEC) mechanisms. In this paper, we show that when VoIP calls are aggregated over a provider's link, with a suitable linear-time encoding for the aggregated voice traffic, considerable quality improvement can be achieved with little redundancy. We show that it is possible to achieve rates closer to channel capacity as more calls are combined with very small output loss rates even in the presence of significant packet loss rates in the network. The advantages of the proposed scheme far exceed similar or other coding techniques applied to individual voice calls.