Abstract:
Radar systems generate large amounts of data, which can create bottlenecks in the system data and signal processors if not handled properly. In recent years, Radio Frequency (RF) sampling has received popularity due to its potential benefits to the systems. However, RF sampling can generate a large amount of data, presenting processing, communication, and storage challenges. In the past, addressing these challenges required a combination of efficient signal processing algorithms, appropriate computational resources, robust data transmission, and fast access optimized storage. Despite advances in high-speed communication systems, data processing techniques like parallel processing and GPU, and storage technology, the large amount of data streaming for applications like radar systems and wireless communications is still a problem. Deep learning (DL) has emerged as a promising approach for processing radar data and extracting relevant features. In this paper, we investigate the use of Convolutional Neural Networks (CNNs) as a fast signal processor for classifying the direction of targets from radar returns. We evaluate the performance of CNNs in Direction of Arrival (DOA) recognition, particularly in extremely low Signal-to-Noise Ratio (SNR) conditions. Our proposed CNN-based DOA recognition system leverages a special configuration that minimizes delay in the signal processor, where Inphase and Quadrature (IQ) components of the data are fed directly to the trained network. Experimental results show that the proposed CNN-based DOA recognition system achieves high accuracy in low SNR conditions, demonstrating the effectiveness of CNNs for radar signal processing. This study provides insights into the design and optimization of CNNs for DOA estimation in radar systems.
Keywords:
Artificial intelligence, Deep learning, Convolutional neural networks, Radar signal processing, Direction of arrival, Low signal-to-noise ratio.
_______________________________________________________________________
