Adaptive Multicast Transmission of Live Video in ATM Networks

Introduction ATM variable bit rate (VBR) service provides a statistical multiplexing gain and supports guaranteed quality of service through preventive congestion control techniques. However, to do this it requires the source to accurately predict its traffic parameters at call admission time. For stored video, this a priori estimation of traffic characteristics is straightforward; the traffic characteristics are already fully known prior to transmission. For compressed, live video, however, this prediction of traffic characteristics is often difficult to perform accurately. Because of this lack of predictability, constant bit rate (CBR) service is often proposed as a solution for the transport of compressed, live video. By adaptively encoding the video signal to achieve a fixed output rate, the CBR source renders its video output process constant and thereby predictable. However, in order to obtain this traffic predictability, CBR compromises multiplexing gain and causes video quality to fluctuate [1]. We require a video service that does not rely heavily upon accurate a priori traffic parameter estimation (as VBR does) and does not unnecessarily sacrifice multiplexing gain and video quality (as CBR does). Furthermore, since a large number of live video applications will consist of multicast or broadcast communications, such a service must be capable of supporting point-to-multipointand multipoint-to-multipoint connections. We therefore propose a service, hereinafter called quasi-VBR service, with support for both unicast and multicast. In the unicast case, the service requires feedback-based rate control and adaptive encoder quantization adjustment (as in [2]). For the multicast case, we augment the service with probabilistic or scheduled feedback and adaptive hierarchical encoding. Feedback-Based Unicast Video Service Unicast quasi-VBR service requires feedback-based rate control and adaptive encoder quantization adjustment. We propose to use the closed-loop feedback algorithm already adopted by the ATM Forum to support available bit rate service for quasi-VBR service as well. This algorithm, known as the Enhanced Proportional Rate Control Algorithm (EPRCA), requires the source to specify its minimum cell rate at call admission time [3]. The service then guarantees this minimum cell rate while allowing the source to exceed it. To avoid congestion, EPRCA utilizes “positive” feedback. In other words, sources increase their allowable output rates only when explicitly allowed by network feedback. To adjust to changing output rates, the video encoder deploys adaptive quantization adjustment. As the allowable output rate and video signal fluctuate, so too does the occupancy of the encoder’s output buffer. The quantization (or coarseness) of the encoded video signal is gracefully increased as the buffer fills and decreased as it empties. Thus, a feedback loop from the network to the encoder is established; as conditions in the network change, so does the encoding process and the resulting video quality. Feedback-Based Multicast Video Service In a multicast environment, providing quasi-VBR service is more complicated. The amount of feedback received by the source increases with the number of destinations, leading to scalability problems. Furthermore, various paths in a multicast connection may experience different degrees of congestion, resulting in varying feedback signals. It is therefore not feasible for a single source to simply adjust its output rate (or its quantization step size) to respond to varying amounts of congestion in the network. Our proposed solution to these problems is to utilize a probabilistic feedback algorithm and adaptive hierarchical encoding in conjunction with adaptive quantization adjustment. To prevent quasi-VBR sources from being flooded with feedback, probabilistic feedback is utilized. Multicast destinations return positive (EPRCA) feedback as in the unicast case. Unlike the unicast case, however, multicast quasi-VBR destinations transmit their feedback with a probability less than one (as in [4]). This feedback response probability is a function of the total number of destinations in the multicast tree. Thus, the EPRCA feedback mechanism can scale to multicast