Computational aspects of quadratic signal processing
William A. Porter · IEEE Transactions on Acoustics Speech and Signal Processing · 1990
Quadratic signal processing is often computationally more intense than the linear case, and hence attention to concurrent computation is an important concern, particularly in real-time applications. The author addresses the concurrent computation issue in the context of systolic array processors. As a prototype the data processing requirements inherent in the calculation of discrete Wigner distributions (DWD) are considered, extending their focus to bilinear (bipower) functions in general. The author summarizes a class of systolic arrays (cylindrical arrays) that prove to be well matched to the computational needs of the bilinear functions in question. He discusses the synthesis of the DWD and bilinear functions and summarizes the computational speeds of the DWD and the general bilinear case. The study makes use of recent results in fast form algorithms and their implementation by dynamically switchable systolic arrays.>