End-to-End Deep Learning Architectures for Joint Modulation and Signal Detection in Next-Generation 6G Communication Systems

Abstract

Development of the customary wireless communication solutions is based on discrete, specialist-designed modules of modulation and signal recognition. Such systems are effective in an ideal situation but cannot satisfy the increasing demands of the next-generation 6G networks that need ultra-low latency, high reliability, and real-time responsiveness to a variety of different and dynamically changing channel conditions. The proposed paper presents a new end-to-end deep learning framework that is trained to learn modulation strategy, and signal detecting strategy simultaneously through the use of data. The communications pipeline can be expressed as a trainable autoencoder: the transmitter, the receiver, and the channel are carried by neural networks, and the channel is implemented in a differentiable layer that simulates additive noise, fading, and non-linear distortions. The model suggested is trained based on supervised learning to reduce bit reconstruction error at different signal-to-noise levels (SNRs) and channel situations. Experimental test results prove that the end-to-end solution is much better than the traditional modulation techniques (e.g. QAM, BPSK), superior bit error rate (BER) performance in non-ideal channel transmissions is observed over Rayleigh fading and composite channels. It also has robustness and generalization over types of channels without any manual tuning within the model. The findings support the viability of end-to-end learning as a possible method to the problem of the physical layer of 6G communication systems so that data-driven, adaptive, and intelligent transceiver design becomes feasible. Future extensions will consider the extension to MIMO, real-time hardware deployment, and combining with reinforcement making the system online adjustable.