Soft-core Processor by Neural Network Pulse Compressor accomplishment

Abstract

In pulse radar detection system, pulse compression was achieved by traditional matched filtering techniques. These approaches resulted in low signal-to-sidelobe ratio and were also fault intolerant. Hence, in this paper we suggest a neural network based pulse compression application. The traditional approach of digital pulse compression is first realized by designing a Finite Impulse Response (FIR) based matched filter in matlab simulink. Then, a multilayer feed forward neural network (MFNN), trained using back propagation algorithm with 13-element barker code as the input for pulse compression is realized by coding in matlab. The performance of this neural network is then tested by taking various performance measures into consideration, such as, resolution ability of two mixed pulse trains with delays and different input magnitude ratios, misalignment performance of the network when a particular bit in the input code sequence is discarded or duplicated, the effect of received signal magnitude and signal-to-sidelobe ratio. The performance of both FIR filter and neural network approaches for matched filtering are then compared to decide which of the two approaches has greater advantages and better efficiency by means of matlab simulations. The neural network pulse compressor is then practically implemented on a Vertex-4 Sx-35 Field Programmable Gate Array (FPGA) using Xilinx ISE Design Suite, and results are observed.