Research On Parameter Estimation And Application Of Wireless Edge Cache Based On Machine Learnin

With the growth of data,the traditional communication architecture mode has been difficult to meet people’s requirements for communication service quality.Aiming at solving this problem,there have been many effective solutions,among which wireless edge caching is a good research direction.The wireless edge caching technology refers to cache popular content in the entire communication system in the edge device of wireless communication system,so as to reduce the load and delay of the network.Due to the limited storage space,we need to cache the content frequently accessed by users as much as possible.And because the content popularity will change with the external environment,the content popularity prediction is proposed.Among them,the prediction accuracy of content popularity is a very critical evaluation index,and whether the prediction is accurate or not can directly affect the overall performance of the system.In addition to considering the popularity of content,it is of great help to improve the overall performance of the communication system if content frequently accessed by users can be transmitted through channels with good channel conditions.Against this background,we did the following research.Firstly,previous studies on content popularity prediction are usually carried out in the case of multi-user demand for a single content set,while there are few studies on the situation of multi-user demand for multiple content set.Moreover,the prediction accuracy of previous research algorithms still need to be improved.Under such a premise,this paper studies the content popularity prediction under multiple content sets.Specifically,an algorithm based on self-supervised learning is proposed to predict the popularity of multiple content sets,and compared with the commonly used LSTM algorithm.Experimental results show that the proposed method outperforms LSTM-based prediction algorithm in terms of prediction accuracy.The distribution of content popularity is usually a Zipf distribution,and users are more concentrated on the front content when Zipf factor is larger.In the past,the prediction of content popularity mostly focused on the prediction of the distribution,but few studies focused on the estimation of the Zipf factor.The representation of the Zipf distribution is usually divided into a two-parameter representation and a single-parameter representation.Based on this,this paper proposes the gradient descent method and the least squares method to estimate the Zipf distribution factor in the two-parameter form and the single-parameter form.Secondly,for the problem of cache deployment strategy,previous research usually only consider the popularity of content,but rarely consider the communication overhead of content.Therefore,in the research of this paper,the consideration of content transmission channel selection is also considered in the case of predicting the popularity of multiple content sets.Specifically,a DDQN-based algorithm is proposed to learn a strategy when content popularity is unknown,so that content with high content popularity can be transmitted through channels with good channel conditions as much as possible.Simulation results show that the proposed algorithm can effectively improve the overall spectrum efficiency of the system and reduce the bit error rate of the system.