FPGA-based transformable computers for fast digital signal processing

FPGA-based computing systems provide a feasible and cost-effective platform for implementing fast parallel arithmetic circuits for digital signal and image processing. This paper reports the results obtained from embedding a highly parallel convolution algorithm on an FPGA-based "transformable" computer. Such a computer is intended to serve as a transformable co-processor for a standard microprocessor system. However, the transformable co-processor is reconfigurable and is capable of exploiting the concurrency of computations more than the "sequential" microprocessor. Our experiments show that a significant gain in speed can be achieved by using the transformable coprocessor. We present an example of performing a sequence of (independent) 16-point convolutions on 8-bit data, and show that the speed factor improves significantly as the number of convolutions to be performed increases.