Data-Driven Advanced Symbol Modulation in the PHY Layer

Next generation Terabit per second (Tbps) wireless communication systems aim for sub-THz and THz bands where noise and hardware impairments are known to be dominantly non-linear and lack accurate closed form analytical models. The physical layer (PHY) blocks designed under linearity and Gaussian noise assumptions will fall short of satisfying the requirements of these next generation systems. Symbol modulation, which is traditionally QAM based and also designed according to the linearity and Gaussian noise assumptions, is one of the key blocks in PHY layer, including 5G NR systems. However, AI/ML based advanced symbol modulation can bring substantial gains in terms of bit error rate and throughput. In this study we focus on methods and procedures that enable trainable symbol modulation in the PHY layer. We propose an end-to-end architecture that includes trainable symbol modulation, a scalable DNN to cover any M-ary modulation and new signalling procedures required for practical training and inference procedures.