A Memory-Efficient Learning Framework for Symbol Level Precoding with Quantized NN Weights

This paper proposes a memory-efficient deep neural network (DNN) framework-based symbol level precoding (SLP). We focus on DNN with realistic finite precision weights and adopt an unsupervised learning (DL) based SLP model (SLP-DNet). apply stochastic quantization (SQ) technique to obtain its corresponding quantized version called SLP-SQDNet. The proposed scheme offers scalable performance vs memory trade-off, by quantizing percentage of the weights, we explore binary ternary quantizations. Our results show that while SLP-DNet provides near-optimal performance, versions through SQ yield ~3.46× ~2.64× compression for binary-based ternary-based SLP-SQDNets, respectively. also find our proposals offer ~20× ~10× computational complexity reductions compared optimization-based SLP-DNet,