Deep Unfolded Parameter Quantization for Multiplier-free MIMO Receivers

The demand for high-performance, low-complexity receivers in future wireless communication systems has spurred research into model-driven design combined with Deep Unfolding (DU) techniques. While advanced receivers based on these principles, such as Deep Unfolded Interleaved Detection and Decoding (DUIDD), demonstrate superior performance, the hardware implementation of their parameters remains challenging. This paper focuses on the learnable parameters within such DU-based receiver architectures, investigating their performance under low bitwidth quantization. We propose and evaluate a post training quantization (PTQ) simulation framework employing Fixed-Clipping Additive Power-of-Two Parameter Quantization (FCAP-PQ), which targets hardware-friendly multiplier-free operations, benchmarked against Fixed-Clipping Uniform Parameter Quantization (FCUPQ). Simulations were conducted for a MIMO-OFDM system under both ideal (Perfect CSI Rayleigh) and realistic ray tracing (CEst REMCOM) channel conditions, assessing the impact of 4-bit FCAP-PQ and FCUPQ on Block Error Rate (BLER). Results indicate that the 4-bit FCAP-PQ scheme enables the DU receiver to maintain BLER performance close to its counterpart under both channel conditions, with particularly minimal degradation in scenarios involving channel estimation errors. Furthermore, FCAP-PQ shows potential to outperform or match the performance of FCUPQ at the same bitwidth. This study validates the feasibility of applying low-complexity, multiplier-free oriented Power of Tow(PoT) quantization to critical parameters of model-driven DU receivers, offering valuable strategy insights for their efficient hardware deployment on resource-constrained platforms.