Abstract
Efficient sparse codebook design is a fundamental research problem in sparse code multiple access (SCMA) systems. This paper proposes an advanced nonlinear SCMA (NL-SCMA) framework for downlink channels, which subsumes conventional (linear) SCMA as a special case. Specifically, the proposed NL-SCMA enables a direct mapping of user messages to a superimposed codeword through a nonlinear mapping mechanism, eliminating the need of per-user based codebooks. The design problem therefore shifts from the conventional user codebook design to the nonlinear mapping optimization and superimposed constellation design. We first propose a Lattice constellation-based superimposed constellation by leveraging its advantages in terms of large minimum Euclidean distance (MED), compact constellation volume, design flexibility, low peak-to-average power ratio (PAPR) and favorable shape gain. By analyzing the error patterns of the lattice-based superimposed constellation using pairwise error probability, we prove that the MED of the proposed nonlinear codebook is lower bounded by the so-called "single error pattern". Motivated by this, we propose an error pattern-inspired nonlinear mapping strategy to maximize the MED. Simulation results demonstrate that the proposed nonlinear codebooks significantly outperform state-of-the-art linear codebooks in terms of PAPR, MED, and uncoded and coded error rate performance over Rician fading channels.