From Biology to Communication: A Practice for the Non-uniform Constellations Design

Non-uniform constellations (NUCs) present a promising technique to improve the spectral efficiency whereas the existing NUCs do not fully exploit the degree of freedom for the constellation design. This paper performs an interesting practice of the biology-motivated communication design to geometrically arrange the constellations. Specifically, we reformulate the traditional constellation design problem as the seed arrangement problem within a circular region. The centroidal Voronoi tessellation is employed to model the seed growth for enlarging the Euclidean distances among the constellations. Besides, we propose a particle swarm optimization based constellation design to minimize the averaged bit error probability. Finally, a lightweight machine learning enhanced demodulation is developed to reduce the complexity introduced by the asymmetric feature of NUCs. Simulation results demonstrate that the proposed scheme outperforms quadrature amplitude modulation and gold angle modulation, highlighting the potential of biology-inspired design techniques for communication systems.