Microcells for ICA-SOC for remote sensing of high energy radiation

In `radiation remote sensing\´ multiple unknown high energy sources are generally involved, and upon sensing the detectors must separate their contributions blindly for further analysis. That is, the objective is to recover the unobserved source signals from the observed mixtures without the knowledge of the mixing coefficients. An effective way to perform such separation is through Independent Component Analysis algorithm. This paper sets the background for this important application, shows successful application of the algorithm on actually measured multi-source data, develops an architecture for the ICA-SOC and, importantly, presents two major VLSI cells for it. A parallel architecture is presented utilizing two reconfigurable coarse-gain microcells with high functionality, performance, and reconfigurability. Fabricated in 0.5 micron CMOS process and tested, these comprise a two-cycle Universal Nonlinear (UNL) cell, and a two-cycle extended multiply accumulate (MA_PLUS) cell. The UNL includes two specialized functions for ICA, namely (1) the first derivative of an approximation to the negentropy function g(u) = G\´(u) = u exp(-u2/2), and (2) the second derivative of the approximation to the negentropy function g\´(u) = G"(u) = [1-u2]exp(-u2/2); The coarse-grain approach has the distinct advantages of reduced external interconnect, reduced design time, and manageable testability. The overall application goal is an intelligent sensor system for high energy radiation remote sensing, with integrated processing-and-decision-making capability, which can reliably separate threat radioactivity from the ambient one.