Experimental Evaluation of Independent Component Analysis Design with Enhanced Performance Metrics using ALJM

The goal of Independent Component Analysis (ICA) is to use statistical and computational methods to determine the underlying structure of a data collection consisting of independent variables, metrics or signals. The analyzed data with multiple variables is usually provided as a massive collection of instances and Independent Component Analysis creates a generative framework for this information. Several interesting hardware architectures for implementing Fast Independent Component Analysis are available, such as: Application Specific Integrated-Circuit (ASIC), Field-Programmable Gate Array (FPGA) and so on. In order to carry out the Eigen-Value-Deconstruction (EVD) necessary for the execution of Fast-ICA, a novel approach called the ALgebraic Jacobi Method (ALJM), is proposed. This ALJM simplify the calculation of the Jacobi rotations by making use of an approximate polynomial estimate and the Newton Raphson technique. The complexity of computation and energy usage of Fast-ICA are reduced without sacrificing its calculation speed by reusing hardware throughout both the Eigen-Value-Deconstruction phase and the Weight-Vector Estimation (WVE) phase. By applying Blind-Source-Separation (BSS) to linear combinations of predictable and unpredictable signals, the performance of the Fast-ICA assisted ALgebraic Jacobi Method is assessed and compare the findings to those of existing approaches. A Scalable-Systolic-Processing-Array (SPA) is proposed for the Fast ICA assisted ALJM and build it on Spartan6 FPGA after ensuring its performance and statistical consistency. The ALJM-based Fast ICA is concluded as one of the best and most effective methodologies for creating and distributing an ongoing deployable solution incorporating Fast-ICA by comparing it to current Fast-ICA implementations in terms of runtime, memory footprint, and power consumption. The envisioned symmetric computational array is designed in verilog-HDL, modeled with ModelSim-6.4C and synthesized with Xilinx Software, in which it implements the Fast-ICA assisted ALJM.