Artificial Intelligence / Machine Learning (AI/ML)-Based Radar Data Compression

The contract seeks to develop artificial intelligence and machine learning-based compression techniques for next-generation synthetic aperture radar data, addressing the limitations of traditional image compression methods on raw radar returns. Phase I focuses on designing and evaluating complex-valued neural network autoencoders to compress radar pulses or range-Doppler maps into compact latent representations while preserving critical detection and analysis features, benchmarking performance against classical approaches in terms of compression ratio and distortion. Phase II transitions the most promising model to embedded hardware, employing quantization-aware training to reduce bit depth and enabling real-time compression on live radar feeds with validation of image and track fidelity after decompression. Phase III integrates the finalized system into fielded radar platforms or service-equivalent hardware emulators to significantly reduce bandwidth demands during data transmission. The technology is designed for dual-use applications beyond military radars, including high-rate sensors like unmanned aerial vehicle radar streams and high-resolution weather radar systems, with broad applicability to remote lidar and sensor networks where data volume and transmission efficiency are critical.