On unlimited parallelism of DSP arithmetic computations

The authors present a conceptually simple and practical technique which identifies necessary and sufficient conditions for achieving unlimited improvement in the throughput of a general DSP arithmetic computation (DAC). For the cases when necessary conditions are fulfilled, this technique is a basis for an approach which transforms an arbitrary DAC so that unlimited parallelism is realized. An important special case, linear computations, is addressed using a specially tuned version of the general algorithm so that the introduced latency and hardware overhead are minimized. The effectiveness of the new approach is demonstrated on several real life examples.<>