Buffering of data generated by the coding of moving images

Data are generated at a fairly uneven rate when video telephone signals are coded by transmitting only the parts of the picture that change from frame to frame. Complete smoothing of these data is impracticable because of the size of the required buffer. Obviously, even a small buffer provides some advantage however. The object of this paper is to explore the relation between buffer requirements and channel rate under varying experimental conditions. The study was made by recording three minutes of data (covering a range of action) on a digital computer and simulating buffer behavior for various channel rates and operating conditions. With little or no buffering, a large rate is necessary. As the size of the buffer is increased, the required channel rate typically decreases quite rapidly until the buffer is large enough to smooth the data over an entire field. Beyond this point there is relatively little improvement until the buffer is large enough to smooth the data generated by a moving user from one movement to the next. At times data are generated at a faster rate than can be handled by the buffer-channel combination. Reduction of the rate of data generation during these periods can be controlled either by using the amount of activity in the picture as a control or by using the state of the buffer as a control. Both methods have distinctly different effects on the buffer-size versus channel-rate curve. The two modes of control can be effectively combined in developing a successful control strategy. The buffer size required to achieve within-field smoothing can be reduced dramatically if the data within the field are not taken in the order in which these data are generated but instead are interleaved in a systematic manner; this is because of the nonuniform rate of data generation within a field.