Harnessing Imperfection: Deep Learning-Based Receiver for Non-Coherent Distributed Transmission

This paper investigates the feasibility of distributed multi-transmission in non-coherent systems using deep learning (DL), a concept we term <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">non-coherent distributed transmission</i> (NCDT). In this approach, multiple transmitters non-coherently send the same content to a receiver. We leverage DL to learn and capture the complex patterns arising from signal combinations with frequency and phase offsets. By engineering relevant features and designing a deep neural network with an adaptable auto-tuning mechanism, we enable precise mapping of received signals to transmitted symbols within the NCDT framework. Additionally, we develop a likelihood ratio test (LRT)-based detector as a baseline for comparison. Our simulations and experiments demonstrate that the proposed DL-based receiver achieves lower bit error rates (BER) than the other state-of-the-art DL-based and LRT-based detectors while eliminating the need for synchronization, channel state information (CSI) estimation, or precoding. Experimental validation using software-defined radios corroborates our simulation results, confirming that NCDT can be practically integrated into wireless systems without requiring hardware modifications. Notably, our results show that the proposed method achieves comparable BER performance at nearly half the transmission power, corresponding to an approximate 3 dB gain.