Scheduling methods for a low-bit-rate speech coding algorithm (ATC-VQ) on a multicomputer-type DSP system

For parallel implementation of a digital signal processing algorithm, the solution of scheduling problems that consist of task partitioning and task allocation are important. This paper considers an ATC-VQ (Adaptive Transform Coding using Vector Quantization) algorithm because it is a typically used complex algorithm, and describes the basic study of the scheduling techniques used for implementation of this algorithm on a multicomputer-type DSP system. With the ATC-VQ algorithm the scheduling methods for the LPC residual signal coding has a large effect on the overall performance, because this coding occupies more than 85 percent of the processing time. The LPC residual signals are divided first into multiple vectors and then vector quantized. Three scheduling methods are shown whose partition units quantize the LPC residual signal vectors, and the performance of each method is evaluated. Appropriate scheduling methods suitable to the ATC-VQ algorithm are discussed.