Massive multiple-input multiple-output (MIMO) can serve huge amounts of users simultaneously with high spectral and energy efficiency, which makes it one of the key technologies for 3GPP fifth generation (5G) wireless networks to achieve high throughput. With the increasing number of antennas, the overhead and complexity of the processing of channel estimation also scale up. More and more machine-learning-based approaches have been proposed to handle this complex challenge. However, as 5G standards continue to evolve, the need for greater bandwidth, lower latency network access, and more energy efficient implementations emerge.
The continuous development of edge computing technology has attracted more and more attention. As the complement of the traditional centralized cloud computing, Edge computing performs data processing tasks at the edge of the network, which enhances data processing capability through distributing the computations to the edge devices. It not only greatly reduces the network traffic and network delay due to this reduced distance between the data processing unit and data generating sources, but also increased the data processing speed with less but more relevant data at the network edges.
In this project we evaluate the feasibility of applying Distributed Machine Learning (DML) combining edge computing technology to conduct channel estimation in 5G massive MIMO through comparing the performance improvement on throughput, latency, as well as the accuracy of the results. Through the evaluation on the performance gains or losses of the proposed solutions over a centralized machine learning solution, this project shall build a solid foundation leading to the next stage of a bigger research program that is the implementation of a suitable architecture as well as a software platform that maximizes the potential of machine learning-based solutions to best elevate massive MIMO technology with the highest throughput, lowest latency and maximized data processing speed and accuracy while preserving data privacy.