Thrust I – Culdar: On-array GPU Accelerated Target Tracking with Reinforcement Learning

Thrust II – Scalable & Distributed On-array Processing for Massive Array Radar System

Modern array systems have an increasing number of elements. However, the number of operations needed for executing conventional array processing algorithms is not scalable. As array size increases, these operations result in large processing overheads. On the other hand, conventional processing algorithms typically require centralized collection of array data, which makes the data transfer a major bottleneck of the system design.

This project is aimed at mitigating the two obstacles mentioned above by designing low complexity array processing method & by exploring the possibility of decentralized computational resources. In the first thrust, the major design objective is to reduce the complexity of processing algorithms and to utilize the massive parallelization capability of GPU. The project delivered a radar tracking pipeline through on-array GPU acceleration [1]. In the second thrust, the major design objective is to enable decentralized processing without centralized collection of array data while still making the array functioning coherently. Though currently under development, the project aims to provide spatial processing functionality including adaptive beamforming [2] and direction-of-arrival (DoA) estimation [3].

Publications:

[1] E. Krijestorac et al., “Machine Learning-Assisted Computationally Efficient Target Detection and Tracking in Massive Fully Digital Phased Arrays,” in IEEE Transactions on Radar Systems, vol. 1, pp. 353-367, 2023, doi: 10.1109/TRS.2023.3298340.

[2] R. Li and D. Cabric, “Covariance Denoising with Applications to Adaptive Beamforming,” 2024 IEEE International Symposium on Phased Array Systems and Technology (ARRAY), Boston, MA, USA, 2024, pp. 1-8, doi: 10.1109/ARRAY58370.2024.10880401.

[3] R. Li and D. Cabric, “A Coordinate Descent Approach to Atomic Norm Denoising,” in IEEE Transactions on Signal Processing, vol. 72, pp. 5077-5090, 2024, doi: 10.1109/TSP.2024.3486533.