Performance Measurement And Immediacy Improvement Of The Aggregate Interference Prediction Algorithm

In today's information age,wireless communication technology is developing rapidly,which makes communication more and more convenient and communication quality better.At the same time,more and more wireless communication-based systems have been applied to life(such as automatic driving system).But the rapid development of wireless technology is accompanied by some problems.With the increasing scarcity of spectrum resources,spectrum sharing between secondary users and primary users has become a solution to the problem of spectrum scarcity.However,when a large number of secondary users dynamically access the spectrum of primary users,a large number of aggregated interference will be generated on the receiver of primary users.Once aggregated interference exceeds the tolerance limit of the receiver of primary users,it will have a harmful impact on the communication of primary users.In this context,the main work of this paper will focus on the following two issues.The first problem is how to predict the cumulative interference in spectrum sharing.In our laboratory,a cumulative interference prediction algorithm based on neural network is proposed to evaluate the aggregate interference.This paper will use software radio platform(Sora)to test the performance of cumulative interference prediction algorithm.Firstly,the performance difference of the algorithm is compared when the number of input parameters of the neural network is different.Secondly,the performance differences between the polar coordinate system and the rectangular coordinate system are compared.Preliminary results show that(1)The more accurate the prediction results when the input parameters are more.When the number of iterations is sufficient,the predicted mean square error is 3.13 when the number of input parameters is 4,the mean square error is 8.32 when the number of input parameters is 3,and the mean square error is 12.08 when the number of input parameters is 2.(2)When using the polar coordinate system,the algorithm predicts the result more accurately than when using the Cartesian coordinate system.The second problem is the real-time problem of the algorithm.With the introduction of new technologies such as automated driving systems into life,the lack of real-time algorithms has led to many accidents.Some of the current traffic accidents occur because the vehicle is in an automatic driving mode,and the poor real-time performance of the algorithm results in an unresponsive response to unexpected situations.In this paper,the cumulative interference prediction algorithm based on neural network is taken as an example to measure the real-time performance of the cumulative interference prediction algorithm.The operation of the algorithm can be divided into two parts,one is neural network training,and the other is neural network to predict cumulative interference.The measured results show that the total time of the algorithm running once is about 0.12s.The neural network prediction time is only about 5 percent of the total running time of the algorithm in about 5 ms,and the training time of the neural network accounts for 95 percent of the total running time of the algorithm.In other words,the speed of neural network training will determine the real-time performance of the algorithm.This paper will compare the neural network training time under the three kinds of optimizers:stochastic gradient descent method(SGD),momentum method(Momentum)and variable learning rate method(AdaGrad).The measured results show that the real-time results are best when using the variable learning rate method(AdaGrad),followed by the momentum method(Momentum)and the stochastic gradient descent method(SGD).