Nonparametric spectral estimation with a computation engine based on VLSI vector signal processors

A three-level embodiment hierarchy, encapsulated within a design methodology and tools set, has been developed to provide a unified solution to the problem of implementing low-latency, high-resolution, and high-throughput nonparametric spectral estimation of a selected frequency band. The three-level embodiment hierarchy is composed of: (i) a VLSI 32-bit vector signal processor capable of direct execution of digital signal processing transform primitives such as fast Fourier transform, window, modulation, etc. on complex-data vectors; (ii) a computation engine (V8) that consists of an array of eight vector signal processors that perform the frequency domain algorithms and a front end that implements a general weighted overlap and add time-domain structure; and (iii) a multi-V8 composite that supports parallel, pipelined, and computational-wave type processing with controlled tradeoff between latency, throughput, and hardware complexity. The design methodology and the set of hardware and software tools enabled complete modeling of spectral-estimator execution time on the V8 and the derivation of the computation noise parameters, dynamic range, and signal-to-noise enhancement figures.<
>