Research And Implementation Of Spectrum Sensing Algorithm

Cognitive radio can effectively improve the utilization of spectrum,and spectrum sensing is one of the key technologies of cognitive radio.Therefore,this paper will study spectrum sensing algorithms and its implementation.Firstly,three spectrum sensing algorithms were studied and proposed.In the light of noise influence in narrowband spectrum sensing algorithm based on power spectrum,a spectrum sensing algorithm(PSEENVR)based on power spectrum extremum and noise variance estimation is proposed.The algorithm uses the ranking coefficient to estimate the noise variance relatively accurately,thus reducing the influence of noise variance and improving the performance of the algorithm against frequency offset and noise power uncertainty.In accordance with the problem that the spectrum sensing algorithm based on multi-node signal covariance matrix is difficult to obtain the accurate threshold,and the original signal information is not fully utilized,a cooperative spectrum sensing algorithm based on convolutional neural network and covariance matrix(abb.CNN-based algorithm)is proposed.The algorithm uses the normalized covariance matrix of the received I and Q orthogonal signals as an input signal matrix,and then directly extracts the feature of the covariance matrix using the convolutional neural network,and obtains the classifier by training.The trained classifier is used to spectrum sensing.This algorithm has a high detection probability at low SNR,and the spectrum sensing performance is improved with the increase of the cooperative nodes.In accordance with the problem that the fusion rules for cooperative spectrum sensing are difficult to determine,and the CNN-based algorithm does not fully utilize the original information of the signal,a cooperative spectrum sensing algorithm based on long-short term memory network(LSTM)is studied.Since the signal sequence features is different when the primary user signal is present or not,LSTM is used to extract the temporal feature of each secondary user's received signal sequence.And then,the feature is fused at the fusion center using the fully-connected layer and the Softmax layer to get a classification decision.The detection performance of the algorithm is better than that of the CNN-based algorithm under low SNR.Then,an implement of cooperative spectrum sensing algorithms based on the USRP X310 was studied.A spectrum sensing software system based on CNN and LSTM on the host using GNU Radio and Keras deep learning framework was built,and offline training model and online spectrum sensing module were designed.The real-time spectrum sensing results for actual signals show that the two cooperative sensing algorithms have certain anti-frequency offset performance.Due to the demand of practical engineering applications of spectrum sensing,an implement system of PSEENVR spectrum sensing algorithm based on TMS320C6678 was finally studied.The AD9361 realizes the transmission and reception of RF signals.The FPGA and DSP chips realize the processing of zero-IF signal data.The DDR3 memory realizes the data sharing between FPGA and DSP.And the host computer provides the human-computer interaction interface.The software system which includes the data and instruction interaction modules between DSP and FPGA,LCD touch screen,and between FPGA and DDR3 was designed.The performance test results for actual signal sensing show that the system can accurately receive signals and sense 42 channels in the broadband,and has certain anti-frequency offset ability.PSEENVR algorithm has higher actual signal detection performance than the PSEGAR algorithm.