Research On Intelligent Algorithm For Sepctrum Sensing In Cognitive Radios

With the advent of the information times, wireless communication system has penetrated into all areas of society, and is playing an increasingly important role in the socio-economic development. However, the widespread application and development of wireless communication services have been increasingly limited to fixed spectrum allocation. Cognitive radio is the key technology to solve the current uneven distribution of spectrum resources. As the primary task of cognitive radio, but also the most important task of establishing cognitive radio systems, spectrum sensing is mainly to understand the degree of interference and the current channel occupancy of the main users, then determine the unoccupied bands as fast and accurate as possible, finally improve spectrum utilization.Faced with the complex realities of communication environment, the awareness of the intelligence of cognitive systems have become increasingly demanding. Traditional spectrum sensing algorithms have been unable to meet the requirements, so the spectrum sensing algorithm based on machine learning emerges. The most representative is the spectrum sensing algorithm based on support vector machine (SVM). To some extent, it improved the test accuracy and immunity of the system, however, due to the difficulty of determining the parameters, it consumes a lot of time and has a poor real-time performance.To address this problem, this thesis presents a spectrum sensing algorithm based on Extreme Learning Machine (ELM). It first introduces the ELM algorithm, and analyzes the feasibility of ELM for spectrum sensing. Then, this thesis proposes a spectrum sensing model based on ELM, to identify the primary user signal. First, calculate the energy value and the cyclic spectrum value of the sample signal, and train the model, so that it has the best performance to judge the test signal quickly. Simulation results show that, the proposed algorithm is superior to the traditional spectrum sensing algorithm and SVM algorithm on the detection performance, and has a great performance boost in training speed compared with SVM algorithm. But it uses the trial-and-error method to confirm the parameters, which has a poor controllability.In order to solve the above problems, improve the generalization ability, and shorten the training time of the spectrum sensing algorithm, in this thesis, an improved model is proposed, which uses a spectrum sensing algorithm based on nuclear ELM. Simulation results show that, compared to the above algorithm, the detection performance of the algorithm improved much more, the training time is shorter, and the parameter settings controllable.