Approximate dynamic precision computation in large-scale MIMO systems: analysis and applications

Abstract

As 6G technology advances, it promises to deliver ultra-high transmission speeds and enhanced network capabilities, supported by developments in Massive MIMO and Extremely Large MIMO technologies. However, the expansion of transmission antenna arrays significantly increases signal processing complexity. In this paper, we analyze approximate dynamic precision computation in large-scale MIMO systems and explore dynamic precision-aware approximate computation as a means to counteract the rising complexity. We develop a theoretical framework that clarifies the relationships between quantization bit width, spectral efficiency, and Bit Error Rate (BER). First, we determine a critical point (a stationary point) beyond which further increases in Signal-to-Noise Ratio (SNR) yield negligible gains in spectral efficiency. We show that this SNR stationary point shifts by approximately 6 dB for each additional bit of quantization precision. Moreover, we establish a relationship between quantization bit width and BER, identifying a threshold bit width beyond which further increases no longer yield meaningful BER reductions for a given modulation order and SNR. Using these findings, we propose several dynamic bit-width adaptation strategies to optimize system performance. Specifically, we introduce techniques to adjust the quantization bit width based on real-time SNR, modulation scheme, and target BER. These strategies move away from traditional worst-case fixed-precision designs toward more resource-efficient and performance-tailored approaches. Our methods offer a new framework to manage quantization complexity, paving the way for innovative designs in future 6G communication systems.