A Modular, Distributed and Scalable DOA Estimator for MIMO Systems

This paper proposes a hybrid model-based and data-driven approach for scalable DOA processors. A proxy spectrum computed from the sampled covariance matrix of the antenna array is fed to a Convolutional Neural Network (CNN) for DOA estimation. This input proxy being of fixed size makes the neural network input invariant of the array size, enabling it to handle multiple array sizes without requiring any modification of the neural network structure or model parameters. To reduce the computation of the covariance matrix and proxy spectrum, we employ a system of subarrays with Nearest-Neighbor communication. The solution was implemented on a Xilinx ZCU102 FPGA targeting 100 MHz frequency for 8 to 64-element arrays. We achieve a low latency of 650µs for an array of 64 antennas.