Lin Chunxiao, Azmine Muhammad Farhan, Liang Yibin, Yi Yang
Bradley Department of Electrical and Computing Engineering, Virginia Tech, Blacksburg, VA, United States.
Front Comput Neurosci. 2024 Feb 21;18:1345644. doi: 10.3389/fncom.2024.1345644. eCollection 2024.
The field of wireless communication is currently being pushed to new boundaries with the emergence of 6G technology. This advanced technology requires substantially increased data rates and processing speeds while simultaneously requiring energy-efficient solutions for real-world practicality. In this work, we apply a neuroscience-inspired machine learning model called echo state network (ESN) to the critical task of symbol detection in massive MIMO-OFDM systems, a key technology for 6G networks. Our work encompasses the design of a hardware-accelerated reservoir neuron architecture to speed up the ESN-based symbol detector. The design is then validated through a proof of concept on the Xilinx Virtex-7 FPGA board in real-world scenarios. The experiment results show the great performance and scalability of our symbol detector design across a range of MIMO configurations, compared with traditional MIMO symbol detection methods like linear minimum mean square error. Our findings also confirm the performance and feasibility of our entire system, reflected in low bit error rates, low resource utilization, and high throughput.
随着6G技术的出现，无线通信领域目前正被推向新的边界。这种先进技术需要大幅提高数据速率和处理速度，同时需要节能解决方案以实现实际应用。在这项工作中，我们将一种受神经科学启发的机器学习模型——回声状态网络（ESN）应用于大规模MIMO-OFDM系统中的符号检测这一关键任务，大规模MIMO-OFDM系统是6G网络的一项关键技术。我们的工作包括设计一种硬件加速的储层神经元架构，以加速基于ESN的符号检测器。然后通过在实际场景中的Xilinx Virtex-7 FPGA板上进行概念验证来验证该设计。实验结果表明，与线性最小均方误差等传统MIMO符号检测方法相比，我们的符号检测器设计在一系列MIMO配置中具有出色的性能和可扩展性。我们的研究结果还证实了整个系统的性能和可行性，具体表现为低误码率、低资源利用率和高吞吐量。
