Recurrent NEAT Assisted 2D-DOA Estimation With Reduced Complexity for Satellite Communication Systems

Direction-of-arrival (DOA) estimation plays a vital role in the field of array signal processing. However, the need for heavy computing tasks in most traditional DOA algorithms, e.g., multiple signal classification (MUSIC), makes their engineering practicality significantly compromised in satellite communication systems. The neuroevolution of augmenting topologies (NEAT) can quickly search for appropriate topologies and weights of neural network functions, but its computational complexity is still too high for satellite systems. This paper proposes a modified NEAT architecture featuring a recurrent structure (RNEAT) that only needs a small number of phase components of the received signal covariance matrix as inputs to reduce the complexity and simplify the neural network architecture. The proposed RNEAT incorporated with multiple signal classification (RNEAT-MUSIC) features low complexity to achieve high resolution and low complexity simultaneously. Validation has been done by applying the proposed method in a two-dimensional direction of arrival estimation (2D-DOA) problem. Results show that the proposed RNEAT-MUSIC efficiently restricts the scanning region before forwarding the covariance matrix to the MUSIC stage. Consequently, the computational workload is reduced by 3/4 compared with the traditional 2D-MUSIC algorithm while maintaining satisfactory DOA resolution.