MTC Network Access Strategy And Resource Allocation Based On Machine Learning

The demand of today's world is high-speed ubiquitous connectivity around the clock to realize a fully mobile and connected society.This demand for the Internet gave birth to the concept of the Internet of Things(IoT).Machine-Type Communication(MTC)is the enabler of the Internet of Things,allowing smart objects to communicate with each other without human intervention.With the advent of the fifth generation of mobile communication era and the further development of the Internet of Things,high-speed growth devices have greatly tested the access capabilities of wireless access networks.Therefore,for the current emerging machine type communication network,how to effectively access the large-scale machine type communication device(MTCD)is a hot issue that has been urgently needed to be solved.Recently,combined with the development of Machine Learning(ML)technology,many mobile Internet scenarios introduce intelligent solutions.Machine learning is a statistical model used by computer systems that does not require the use of special instructions to effectively perform specific tasks,but rely on patterns and inference instead.Machine learning acquires new knowledge by observing the environment and self-exploration to continuously improve its performance.Combined with the cogitation of machine learning,this paper studies the insufficient access capability and resource allocation in MTC networks,and proposes effective access strategies and resource allocation schemes in different MTC scenarios.The main results are summarized as follows:According to the characteristics of MTCD,which possess dominant uplink and is mainly small data,an MTC access strategy based on self-organizing map neural network clustering algorithm is proposed.By using the clustering algorithm,the MTCD selforganizing spontaneously trains to obtain suitable data aggregation points,thereby reducing the number of devices directly interacting with the base station,and effectively avoiding the congestion problem caused by large-scale MTCD access to the base station.In addition,the clustering algorithm preserves the mapping relationship,making the MTC network environment highly inclusive for the newly added MTCD.Considering that the preambles are orthogonal to each other during the random access procedure,the access problem is simplified to competitive access problem that at least one MTCD competes for the same preamble.Using the backoff mechanism,the double queue model of MTCD is set,and the queuing model of the access request is abstracted into a Markov decision process.Therefore,the dynamic programming algorithm of Markov decision process is used to solve the competitive access problem.It can be obtained from the simulation results.The algorithm effectively improves the access capability of the MTC network,reduces the network collision probability,and reduces the access delay.For the case of MTCD with multiple service requirements in the MTC network,the deep reinforcement learning algorithm is used to obtain an efficient resource allocation scheme.For the problem of large-scale MTCD resource allocation,using traditional Q learning to solve this problem will cause the state action to be too large,and the Q value table occupies the memory explosion problem.Therefore,the deep neural network and Q learning are combined to obtain relatively optimal resource allocation.Program.Compared with the traditional resource allocation scheme,the resource allocation scheme based on deep reinforcement learning maximizes the network throughput while ensuring different service quality requirements of the MTCD.By analyzing the characteristics of MTC communication,this paper uses the machine learning method,targeted design access strategy and resource allocation scheme to improve the access capability of the network,and provides a new idea for solving the MTC network access problem.