FPGA Implementation of a 1D-CNN Modulation Classifier for Radar Signals

In this paper, the VHDL implementation of a 1D convolutional neural network (CNN) intrapulse modulation classifier is discussed. The solution is designed for electronic warfare (EW) applications, whose main purpose is to detect and identify radar signals from potential threats in noisy environments, meeting stringent requirements related to classification accuracy and low latency. The presented classifier targets the Xilinx ZCU111 Development Board and is capable of identifying 14 modulation classes, as well as noise, under various signal-to-noise ratio (SNR) conditions. Results demonstrate that the field-programmable gate array (FPGA) implementation of the proposed classifier achieves an overall accuracy of 93% for signals with an SNR = -2 dB and at least 99% for SNR ≥ +10 dB, besides showing approximately 1.74× speed-up when compared to a processor-based approach of the same classifier.