Optimized differential evolution algorithm for linear frequency modulation radar signal denoising

In radar systems, the problem of recovering the targets reflections has been a major concern for system designers for decades. One of the first steps for better signal recovery was done by initializing a stable radar signal with high repetition sequence of generated pulses. Stabilizing the radar signal and achieving a better recovery for the received signal, over the years, took a big part of extensive studies on pulse generators and led to the era of analog systems replacement with digital ones. Using the microelectronic circuitries have shown reliability prove in terms of signal generation stability. Chirp pulses are one of the most popular radar signals that can be easily generated using digital technology. In this thesis, Memory Based, and Direct Digital Synthesizer (DDS) architectures as the two most popular chirp signal generation techniques have been designed, by using Altera StratixIII FPGA by the use of Altera QuartusII software. The received signal recording was performed by using MATLAB Software code, connected to the FPGA for getting the received reflections from the HSMC FPGA daughter board that worked as an Analog-to-Digital and Digital-to-Analog converter. Both architectures gave precise results for different selections of chirp rate that could fit with system specifications. The main contention of this thesis is to investigate the development of new optimization technique based on Differential Evolution algorithm (DE), applied for radar signal denoising application. The choose of the Differential Evolution was mainly made because, of its simplicity, and reliability scheme that can provide especially, in the applications that require continuous spaces measurements, which was fit to our problem. An improvement to the conventional DE algorithm has been made to change it from its classical form to be possibly applied for ambiguous targets range detection for radar system. The standard DE algorithm is known as a fixed length optimizer, while our problem demands the need for methods that aren’t tolerated to a fixed individual size, and that was made by altering the mutation and crossover strategies as well as the selection operation. We propose an optimized crossover scheme that changes the crossover operation from being fixed-length to random-length, which has been designed to fit for the proposed variable length DE. We refer to the new DE algorithm as random variable length crossover DE (rvlx- DE) algorithm. The measurement results show high capability for target recognition in terms of frequency response and peak forming that has been clearly recognized from noise and clutter distortion, and that was shown more clearly when it was compared with Wavelet Transform and Hilbert-Huang Transform denoising techniques