Research On Reinforcement Learning-based Cell-Free Massive MIMO MEC Offloading Algorithms

One of the technical indicators pursued by mobile communication technology is a faster data transmission rate.However,the current cellular network may impose restrictions on this.So a new network concept called Cell-Free Massive MIMO or CF was proposed.At the same time,with the greatly enhanced computing power of mobile terminal devices,some computing-intensive tasks such as VR gradually need to be completed by mobile devices.But users often hope that these tasks can be completed with low latency,so as to bring a good experience for themselves.The portability of the device will limit its computing power,making it difficult to achieve the above goals.Therefore,a concept called Mobile Edge Computing(MEC)was proposed.It enhances the computing power of the user equipment by equipping the edge server on the network access side to perform these tasks instead of the user equipment.On the one hand,the realization of MEC requires a higher information transmission rate to reduce the impact of transmission delay on the total delay.On the other hand,it requires users to obtain almost the same transmission rate at any location.This is a challenge for traditional networks.This thesis first conducts a theoretical analysis of Cell-Free Massive MIMO,including the different signal transmission stages and uplink reachable spectrum efficiency.Through the numerical simulation of the obtained uplink spectrum efficiency expression,it can be known that it can achieve higher spectrum efficiency and better user fairness.This allows CF to become a strong competitor for MEC infrastructure.The CFMEC model is obtained by combining CF and MEC.Then,because the CF-MEC network has a large number of APs available for server deployment on the access side,the problem of computing offloading for determining the destination of computing tasks becomes very important.Based on the discussion of related issues in the existing literature,this paper analyzes the limitations of existing methods and proposes a CF-MEC computational offloading algorithm based on deep reinforcement learning.This thesis first considers the centralized computing offloading algorithm.It transfers the offloading decision of the computing task to the network center,that is,the CPU.This method optimizes the offloading object of the current computing task by using global information.After setting up scenarios where computing tasks frequently arrive,this article simulates and compares and analyzes the centralized computing offloading method based on deep reinforcement learning.Finally,this thesis considers a distributed computing offloading algorithm based on deep reinforcement learning.The additional transmission delay introduced by centralized computing offloading can reduce user experience in some cases,while distributed computing offloading can overcome this shortcoming.It reduces the transmission delay by allowing users to choose local calculation decisions at the right time.By using the same scenario as the centralized offloading decision,this paper simulates and analyzes the performance,advantages and disadvantages of the distributed computing offloading algorithm..