Wlan Performance Estimation And Radio Frequency Tuning Based On Deep Learning

With the rapid development of wireless communication technology,wireless local area networks(WLAN)are taking on more and more important responsibilities in enterprise networks.In the centralized WLAN of most enterprises,the wireless access point(AP)adopts a dense deployment method to achieve the goal of complete wire-less signal coverage.However,due to the limitation of wireless spectrum resources,the co-frequency interference between APs caused by the high-density deployment of APs reduces the wireless office and life experience.Aim-ing at the enterprise centralized WLAN scenario where APs are densely deployed,this paper studies the maximum throughput estimation,timing throughput prediction,and RF tuning of WLAN.The main work of the thesis is as follows:1.Aiming at the problem of obtaining the maximum throughput tag of WLAN,the maximum throughput tag collection APP was developed.The verification showed that the APP can accurately obtain the maximum uplink and downlink rate of the test terminal in the actual network to make the maximum throughput tag.Aiming at the problem of estimating the maximum throughput of WLAN,a maximum throughput estimation algorithm based on Deep Neu-ral Network(DNN),a maximum throughput estimation algorithm based on densely connected network(Dense Net),and domain-based adaptive model accuracy improvement algorithm.DNN extracts features better by deepening the network structure,and Dense Net allows each layer to obtain the gradient information of all previous layers through a feedforward method.The domain adaptive algorithm improves the performance of DNN and Dens Net models on the test set by resampling the training set and giving large weights to points whose distribution is close to the test set.The experimental results prove that the proposed algorithms can accurately estimate the maximum throughput and effectively evaluate the network performance when only the network parameters are known.2.Aiming at the prediction problem of wireless local area network time series throughput,the prediction al-gorithm based on gaussian process regression(GPR)model and the prediction algorithm based on recurrent neural network are studied.The GPR model can give reliable confidence intervals,but the prediction error is large,and the Long short-term memory(LSTM)prediction error is small,but it consumes too much computing resources and time,and does not meet the needs of the enterprise.In response to the needs of enterprises for the rapid prediction of WLAN time series throughput and the robustness of the prediction results,a time series throughput prediction al-gorithm based on a shared weighted long short-term memory network and gaussian process regression hybrid model(SWLSTM-GPR)is proposed.SWLSTM-GPR accelerates network convergence by sharing network weights.Ex-periments have proved that the algorithm meets the needs of enterprises in terms of time complexity and hardware overhead,obtains a reliable confidence interval,and improves prediction accuracy,which has enterprise application value.3.For WLAN radio frequency optimization,in order to reduce the co-frequency interference between network APs and ncrease the throughput of the entire network,the minimum coverage boundary of AP wireless signals is defined,and a AP power control algorithm based on the minimum coverage boundary is proposed.The algorithm uses the downlink received signal strength of a large number of terminal users to search for the minimum coverage boundary,and gives power control decisions through the minimum coverage boundary path loss.At the same time,in response to the scalability requirements of WLAN,a multi-layer perceptron-based AP power estimation method is proposed to quickly provide power configuration for re-deployed APs.Practical tests show that the power strategies given by the two algorithms in their respective scenarios are both effective and reliable,effectively reducing co-frequency and improving terminal user's experience while ensuring that the throughput of the entire network does not decrease.