Applicability and Performance of Standard Compression Methods for Efficient Data Transmission and Storage in Radar Networks

Modern sensor networks—those used for autonomous driving, security systems, human motion tracking, or smart city/smart factory applications—are shifting to a more centralized data processing approach to enable efficient multimodal sensor fusion for optimal environment perception in complex dynamic situations. Among lidars and cameras, radars are typical for these applications, but they generate huge amounts of data, which cannot be transmitted or stored effectively in current setups. Consequently, manufacturers usually have to process the data “on sensor.” This results in transmitting only a few extracted features as point clouds or object lists to a central processing unit, which usually causes a significant loss of information. With this approach, advanced processing—such as enhancement of resolution by coherent combination of sensors or ghost target removal with advanced algorithms—is hardly possible. To overcome this, we suggest an alternative method by using signal-based compression with defined losses. The following topology will be proposed: the sensors encode raw data without prior radar-specific processing and after transmission, a central unit decodes and processes the radar data, thus benefiting from its more powerful heterogeneous processing system. We will analyze lossless compression algorithms with rate savings of about 30% to 65%, but the focus is on lossy compression algorithms that incorporate higher compression ratios by allowing negligible errors. It is shown that state-of-the-art multimedia compression algorithms can obtain rate savings of 99%, and radar specific algorithms can add a 50-fold gain on top, reaching 99.98%. To assess the distortions of compressed data, we then present different radar-specific evaluation metrics.