Transmission of real time media such as video requires that the end-to-end delay in the system (between encoder and decoder) be kept constant. Typical video compression algorithms generate a variable number of bits per video frame and thus require buffering of the information prior to transmission, with the buffer size being related to the end-to-end delay. Rate control is needed to prevent exceeding the maximum end-to-end delay or, equivalently, to avoid buffer overflow. In this talk we formalize the rate control problem for the cases of constant bit rate (CBR) and variable bit rate (VBR) transmission. After the constraints have been formalized we concentrate on approaches to rate control that are based on pre-analysis of the data to be coded. We show some experimental results based on MPEG-2 encoders where the complexity of the optimization is significantly reduced by using of the rate-distortion characteristics. We then discuss two different instances of VBR transmission. First we consider the situation where the encoder can transmit at a variable rate, with some restrictions on the rate variations (this would be the case for example of ATM transmission under Leaky Bucket constraints). This analysis allows us to provide a fair comparison between VBR and CBR video in terms of video quality, for given end-to-end delay constraints. Second, we show how transmission of video over a lossy channel, can also be seen as a variable rate scenario and we propose an algorithm where the video encoder optimizes its quantization selection for a given a priori channel model and observed channel state. This will have applications to video transmission over lossy channels, such as a wireless link or the Internet, where a feedback channel is available to inform the encoder of the state of the channel. Recent publications relevant to these topics can be found in:

http://sipi.usc.edu/~ortega/Research.html