不同基站计算架构Massive MIMO基带能效建模和趋势研究

大规模多输入多输出(massive MIMO)是5G系统的标志性技术，可利用大规模天线有效地提高频谱利用率。未来的5G-Advanced和6G演进massive MIMO将支持更多的天线和更复杂的算法，基带能效(EE)会成为持续提升网络能效的关键挑战之一。在MIMO系统中，基站计算架构可分为以ASIC为主的专用计算和以CPU为主的通用计算。由于缺乏对基带计算需求和EE的定量建模，选择最佳计算架构(专用或通用计算架构)是非常困难的。因此，有必要研究与组合逻辑单元及处理周期有关的不同计算架构的功耗模型。基于提出的功耗模型，得到每瓦特每秒单位浮点运算的EE方程式的封闭形式。数值结果表明，目前专用计算的EE分别比通用计算(带硬件加速)和CPU通用计算架构高30倍和200倍。

Energy Efficiency Modeling of Massive MIMO Baseband Processing with Different Base Station Computing Architectures

Massive multiple-input multiple-output (MIMO) is a key enabling technology for future 5G-Advanced/5G mobile networks to effectively increase spectrum utilization by using large-scale antennas. It is expected that with the evolution to 6G massive MIMO will support more antennas and more complex algorithms, and thus baseband energy efficiency (EE) will be one of the crucial challenges to improve network energy efficiency. In such a system, base station (BS) computing architectures consist of dedicated (ASIC) and general-purpose (CPU) computing architectures. It is very difficult to choose the optimal computing architecture due to the lack of quantitative modeling of the computational requirements and EE of the baseband. Hence, it is necessary to study the power consumption model of different computing architectures related to combined logic units and processing cycles. Based on the proposed power consumption model, the closed forms of EE equations are derived with unit floating point operations per-second per-Watt. Numerical results show that the current EE of dedicated computing is 30 times and 200 times higher than that of the general-purpose computing (with hardware acceleration) and CPU general-purpose computing architecture respectively.