The regularized fast Hartley transform. Optimal formulation of real-data fast Fourier transform for silicon-based implementation in resource-constrained environments. (Q847617)

As is known, the discrete Hartley transform (DHT) is closely intertwined with the discrete Fourier transform (DFT). The aim of the author is to present a design for a generic double-sized butterfly for use by the fast Hartley transform (FHT) of radix-4 length, which lends itself to parallelization and to mapping onto a regular computational structure for implementation with parallel computing technology. Chapters \(1-3\) involve a short historical outline of the problem and an introduction of DFT and DHT. The known properties of DFT and DHT are presented without proofs. Unfortunately, some of these properties are not correctly described. All indices of vectors of length \(N\) must be computed modulo \(N\). Thus the reversal \(\{x[N-n]\}_{n=0}^{N-1}\) is correctly introduced, when \(N-n\) is computed modulo \(N\), i.e., \(x(N) =x(0)\). The shift property (3.30) of DHT contains a wrong sign. The cyclic convolution and auto-correlation are not defined on pp.~\(34-35\). All derivative formulas \((3.37) - (3.40)\) are wrong, since (3.37) is valid for Fourier coefficients, but not for DFT. For correct properties of DFT and DHT [see e.g. \textit{W.~L.~Briggs} and \textit{Van Emden Henson}, The DFT. An owner's manual for the discrete Fourier transform. Philadelphia, PA: SIAM, Society for Industrial and Applied Mathematics (1995; Zbl 0827.65147)]. The Chapters \(4 -7\) form the core of this book, which deal with the design of a highly-parallel formulation of FHT of radix-4 length. The efficient computation of radix-4 FHT is based upon generic double-sized butterfly, called regularized FHT. The author discusses a partitioned-memory single processing element computing architecture for the parallel computation of the regularized FHT. Further multiplications used by regularized FHT can be replaced by a hardware-based parallel arithmetic unit that yields a flexible-precision solution without need of trigonometric coefficient memory. Chapters \(8-9\) deal with some applications, namely efficient parallel computation of a real-data DFT of radix-2 length and computations of convolutions, auto-correlations and cross-correlations. Note that subsections 9.3.2 and 9.3.3 are not correct. This book closes with 2 appendices, where the source code in C required for regularized FHT is presented. The textbook is mainly written for students and researchers in engineering and computer science, who are interested in the design and implementation of parallel algorithms for real-data DFT and DHT.