Research On Wideband Spectrum Fast Sensing Technology Based On Clustering And Multi-step Prediction

Cognitive radio technology,on one hand,aims to improve spectrum efficiency and on the other hand to enhance communication quality,and it is mainly used in spectrum management,adaptive communication and so on.While spectrum sensing is the prerequisite of spectrum analysis,decision making and reliable transmission in cognitive radio system.In order to make more efficient use of spectrum idle opportunities,the demand for wideband spectrum sensing technology is increasing.Because the wideband spectrum sensing technology based on prediction can reduce the detection channel,it has attracted much attention from researchers.In order to achieve wideband spectrum fast sensing,a method is designed to improve the sensing speed,which combines time and frequency domain to prediction and integrates coarse sensing and fine sensing,therefore it decreases the sensing bandwidth in frequency domain and reduces the prediction time in time domain.The wideband spectrum fast sensing method is applied to the adaptive communication system and the sensing performance is tested.The main research contents are as follows:1.For fast spectrum sensing of 100MHz-3 GHz frequency band,an off-line coarse sensing method is proposed based on correlated channels clustering,estimation and multi-step prediction.Wideband spectrum which will be sensed is divided into many sub-channels equally,and the strongly correlated sub-channels are clustered according to the density based clustering algorithm.The occupancy states of the most continuous channels in the cluster are detected and estimated,and then the continuous channels with the largest number of idle states are selected for multi-step prediction.Finally,according to the result of coarse sensing and the demand of service qualit,the channels that need further fine sensing are selected,which reduces the sensing bandwidth and prediction time.2.As the accuracy of multi-dimensional multi-step spectrum prediction by single neural network needs to be improved,a combined neural network model is set up for wideband spectrum multi-step prediction based on Seq2Seq of Long Short-Term Memory(LSTM)and Convolutional Neural Networks(CNN).The multi-step prediction of each channel is realized by parallel Seq2Seq model,and the correlation characteristics between channels are fully mined through the CNN.and the spectrum prediction accuracy is greater than 0.9 in particular steps.3.For the performance of spectrum prediction still cannot meet the needs of secondary users,the simulation of wideband spectrum fine sensing is realized based on reconfigurable filter bank and energy detection.A method is proposed to reconfigure the filter bank subband number according to the subbands idle rate of spectrum sensing results.Meet the prerequisites of detection probability greater than 0.9,and the computational complexity is reduced by selecting the appropriate number of subbands.4.In order to engineer the wideband spectrum sensing method,it is applied to the adaptive communication system,and the communication architecture of adaptive transmitter and receiver is designed.Meanwhile,the wideband spectrum sensing performance is tested on the software radio defined platform,and through the experimental analysis,the test results all meet the requirements of users.Based on spectrum fine sensing results and Quality of Service requirements,the communication parameters such as transmission power and modulation mode are optimized by chaotic quantum dolphin swarm decision algorithm.