Handoff Algorithms Using Machine Learning

With continuous construction of rail transit in our country,the demand for people to access the mobile Internet in the process of take rail transit is also growing.WLAN has become one of the most popular wireless communications systems,because it is cheap and easy to deploy.Efficient handoff algorithms are both accurate and timely,and result in good link quality,high system capacity and low probability of dropped calls.The need for better handoff algorithms is increasing as Access Point density increase in order to serve more users per unit area.Traditional handoff algorithms use comparisons of instantaneous indicators of signal strength to decide on handoff.They do not make use of memory to achieve both accuracy and timeliness.Instead,there is a tradeoff between the number of unnecessary handoffs and probability of dropped calls or delay in the handoff decision time.We show how memory can be used in two ways.Firstly,the handoff problem can be treated as a supervised learning problem,applying supervised learning algorithm to learn the function relationship between handoff position and packet loss rate.Secondly,it can be treated as a Markov Decision Problem,applying reinforcement learning algorithm to learn the perfect policy,which action to take in certain position state.My research investigates how using memory in these two ways can improve the performance of handoff algorithm.The machine learning handoff algorithm mainly involves a decision scheme that we have invented.This scheme uses training set to learn the relationship between handoff position and the finally result.Then,the system will get the best position for handoff and initiates a handoff when a user is at certain preselected locations.Extensive computer simulations have been run to compare the performance of traditional handoff algorithms and the machine learning handoff algorithm in a handoff scenario called the canonical handoff problem.Both traditional handoff algorithms and the machine learning handoff algorithm have been simulated and the machine learning based handoff algorithm has been found to perform better than others,having the capability to make decisions more accurate.