A scalable video transmission system using bandwidth inference in congestion control

Multimedia applications widely use UDP as the transport protocol because UDP can provide a more timely response to packet loss. Unlike TCP, which has a mature built-in window-based congestion control, UDP applications have to construct their own congestion control at the application level. Various protocols have been proposed for such multimedia applications. Packet loss, round-trip delay, and one-way delay are used in these congestion control protocols to adapt to the network dynamics. Most protocols try to achieve TCP-friendliness by using addictive-increase multiplicative decrease (AIMD) algorithm. In this paper, we present a new bandwidth inference congestion control (BIC) protocol, which adapts its rate by non-intrusive bandwidth inference. It relies on one-way delay trend detection to estimate the end-to-end available bandwidth. We also show an effective BIC protocol implementation based on an MPEG-4 FGS video coding system.