A Low-Complexity Blind Iterative Projections Approach for Beamforming and Channel Estimation

We propose a novel data-driven approach for beam-forming and channel estimation based on a computationally efficient alternative projections method. Unlike traditional blind beamforming techniques, our method does not require training symbols or specific signal properties such as constant modulus or finite-alphabet symbols. This makes our proposed approach more versatile and robust. Our approach is provably convergent, unlike algorithms such as the Least Means Squares that can diverge if the step size parameter is not properly chosen. Additionally, our method converges to the optimal eigenvector solution as the batch size increases. We show that the iterations can be drastically reduced, and the performance can be enhanced in a static channel by initializing the algorithm with the estimate formed at the prior batch. Through rigorous analysis and simulations, we demonstrate the effectiveness of our proposed method.