Deep Learning Based Universal Beamforming Design With Finite Blocklength Transmission

In low-latency and high-reliability communication systems, a channel coding operates within finite blocklengths, making the data rate a function of both decoding error probability and blocklength. This paper proposes a deep learning based universal beamforming design for a multi-user downlink systems, considering finite blocklength transmission. By leveraging blocklength as a side information in deep neural network (DNN), we develop an efficient beamforming design that is universally applicable to the varying blocklengths, aiming to maximize the sum rate under finite blocklength and decoding error conditions. Through numerical results, we demonstrate the superiority of the proposed scheme over existing beamforming design methods.