Interference Mitigation in Automotive Radar using ResNet Deep Neural Network Models

Current autonomous vehicles and Advanced Driver Assistance Systems (ADAS) integrate an array of sophisticated sensing techniques, including LiDAR, cameras, ultrasonic sensors, and radar. Among these, radar, owing to its distinctive attributes, finds application across various ADAS domains. However, the ubiquity of radar adoption amplifies vulnerability to radarto-radar interference, thereby compromising radar functionality. A plethora of research endeavors have diligently explored strategies for interference mitigation, encompassing conventional signal processing techniques and contemporary deep learning methodologies. This paper introduces a new technique by leveraging the capabilities of the ResNet deep neural network model to address interference-related challenges in ADAS applications through guided segmentation. It presents performance comparison between the best performing techniques previously published and our proposed approach. Benchmarking is done using a standard dataset that is well-known to the research community. Our results have shown the significant outcomes of using the residuals-based architecture for improving the performance of deep neural network models to mitigate interference between automotive radar modules.