Acceleration on Heterogeneous Architectures for Synthesis of Coherent Sparse Arrays

Radio frequency aperture synthesis from multiple free-flying collectors is traditionally dependent on highly accurate intra-constellation metrology and shared clocks. We demonstrate that coherent alignment of independent collectors with poor knowledge of relative positioning and clocking can be achieved through computational means in post processing. This allows the synthesis of a coherent sparse array of RF collectors with the time and position knowledge available from a cheap and commercial GPS receiver. This paper extends our previous publication in the 2018 IEEE Aerospace conference with several key advances, including on-GPU execution of the CAF algorithm, on-GPU multi-emitter tracking, and interpolation-based correction adaptation. The alignment algorithm has significantly increased in performance and capability since 2018. The Complex Ambiguity Function (CAF) is used for simultaneous estimation of time difference of arrival (TDOA) and frequency difference of arrival (FDOA) for multiple pairs of received signals. These time, frequency, and phase corrections are applied to the data, bringing them into alignment with each other. Maintaining this alignment over a longer time span is the challenge, addressed through overlapped sequential CAF estimates to estimate the evolving TDOA/FDOA of multiple emitters. Achieving good alignment over millions of samples requires an evolving model of the collection geometry. By breaking a long signal into successive, overlapping frames, we can estimate the signal parameters; however, the computation time is untenable on traditional CPUs. In this paper, we explore the use of heterogeneous architectures to speed up the CAF algorithm and to make it suitable for the alignment of larger signal lengths. We show that the optimized GPU implementation provides a performance improvement of 150x compared to the sequential implementation. We also demonstrate a dynamic load balancing scheme for MPI that can distribute the work across multiple GPU nodes. Using the accelerated code, we demonstrate the signal alignment using CAF with the NOAA satellite full-pass collects using multiple collectors on the ground around the Los Alamos townsite.