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

TECHNICAL POINT OF CONTACT (TPOC)
Thomas Provencher
PROJECTED CMMC LEVEL REQUIREMENT
Level 2 (Self)
TECHNOLOGY AREAS
Electronics
|
Sensors
MODERNIZATION PRIORITIES
Integrated Sensing and Cyber
|
Trusted AI and Autonomy
KEYWORDS
deep autoencoder, data compression, neural coding, rate-distortion, bandwidth reduction, synthetic aperture radar
OBJECTIVE
Use deep learning (e.g. autoencoders or transformer models) to compress raw radar data into low-bit representations for efficient storage/transmission, with reconstruction that preserves radar utility.
ITAR
The technology within this topic is restricted under the International Traffic in Arms Regulation (ITAR), 22 CFR Parts 120-130, which controls the export and import of defense-related material and services, including export of sensitive technical data, or the Export Administration Regulation (EAR), 15 CFR Parts 730-774, which controls dual use items. Offerors must disclose any proposed use of foreign nationals (FNs), their country(ies) of origin, the type of visa or work permit possessed, and the statement of work (SOW) tasks intended for accomplishment by the FN(s) in accordance with section 3.5 of the Announcement. Offerors are advised foreign nationals proposed to perform on this topic may be restricted due to the technical data under US Export Control Laws.
DESCRIPTION
Next-generation radars (especially synthetic aperture radar (SAR)) collect data with massive data rates. Traditional image compression is not optimized for raw radar returns. Recent work extends neural compression to the complex SAR domain. Phase I is to explore autoencoder architectures for radar data using, for example, a complex-valued neural network encoder/decoder that learns to compress raw pulses or range-Doppler maps into a latent code with minimal distortion. The system could be trained on representative radar scenes to prioritize features important for detection and analysis. Phase I will benchmark compression ratios vs. classical methods. Phase II will implement the best model on embedded hardware for real-time compression (this could involve quantization aware training to reduce bit depth) and test it on live radar feeds, evaluating image and track fidelity after decompression. Phase III will integrate into fielded radars or data links, or onto hardware emulating Service radar systems, to dramatically lower bandwidth needs. Dual Use: This is valuable for any high-rate sensor (remote lidar, sensor webs); for example, compressing unmanned aerial vehicle (UAS) radar streams or compressing high resolution weather radar data for comms.
PHASE I
Design and train autoencoder networks for radar data (range profiles and raw SAR data). Evaluate rate-distortion performance on recorded datasets (contractor generated and/or government provided) and its impact on target detection performance. Compare with baseline compression techniques. Deliverables include a feasibility study, performance evaluation results for the data sets utilized, algorithm implementation/software, and any synthetic data generated.
PHASE II
Optimize and implement the code in radar hardware. Demonstrate real-time compression of a SAR or moving target indication (MTI) data stream. Measure and quantify computational requirement/load and target detection performance. Deliverables include the hardware design, software (to include compression algorithm implementations), and performance analysis of the compression technique on the hardware including computation time and impact to target imagery and/or target detections/tracks after decompression.
PHASE III DUAL USE APPLICATIONS
Field it in Service radar systems, or in prototype systems emulating Service radars and associated concepts of operation, to reduce link bandwidth. Civil uses include weather radar data networks, and even medical imaging streams.
REFERENCES
R. M. Asiyabi et al., "Complex-Valued Autoencoder-Based Neural Data Compression for SAR Raw Data," in IEEE Journal of Selected Topics in Signal Processing, vol. 19, no. 3, pp. 572-582, April 2025, doi: 10.1109/JSTSP.2025.3558651.
QUESTIONS & ANSWERS
NOTE:
To ask a question, you must
log in or create an account
for the DSIP.