DNN-Assisted Constellation Design for SIC-Free Hierarchical QAM-Based Multiple Access

QAMA, a recently proposed multiple access technology, utilizes hierarchical quadrature amplitude modulation (QAM) to achieve the same rate region as conventional multi-user superposition transmission (MUST) without requiring successive interference cancellation (SIC) at receivers. This is particularly advantageous for low-cost Internet of Things (IoT) devices. The key process involves adjusting the constellation to generate transmission modes that enable SIC-free operation and span the desired rate region. However, determining the optimal constellation currently requires either an exhaustive search over all Euclidean distances, which is computationally intensive, or storing a lookup codebook, which demands significant memory. To address this, a deep neural network (DNN)-based approach is proposed for QAMA constellation design. The approach employs a binary classification network to identify transmission modes under given user signal-to-noise ratios (SNRs), followed by a regression network to predict optimal Euclidean distances. Numerical results demonstrate that the DNN-based method significantly reduces computational complexity while maintaining performance comparable to exhaustive search.