Research On Spectrum Sensing In Cognitive Radio System

With the growing demand for wireless communication services, the spectrum resources have become increasingly scarce, which seriously limits the development of communication technologies. To solve this problem, cognitive radio(CR) is introduced. It allows secondary users(SU) to opportunistically use the spectrum when the primary users(PU) are idle, thus to achieve the goal of improving spectrum utilization.As the most important part of cognitive radio, spectrum sensing has received many attentions in recent years. However, to avoid the interference to PU caused by the SU, meanwhile, to provide more opportunities for SUs to access the network, it is necessary to propose spectrum sensing methods with high accuracy and efficiency. This dissertation focuses on the spectrum sensing in cognitive radio system, and the main work and contributions are outlined as follows:1. In the basic model of spectrum sensing, three traditional sensing methods, including matched filter method, energy detector and the cyclostationary feature detector, are systematically analyzed under the framework of likelihood ratio test. The test statistics of three methods are also re-derived, respectively. Finally, the advantages and disadvantages of these methods are illustrated, which laid the theoretical foundation of this dissertation.2. For spectrum sensing under ultra-low SNR environment, a spectrum sensing model based on Duffing oscillator is proposed by combining with the theory of chaos in nonlinear dynamics. Since Duffing oscillator has the characteristics of initial sensitivity and noise immunity, we can make a decision on whether there is a primary user by setting the received signal as the driving force of the Duffing system. In the practical applications, the frequency to detect is random. However, the original Duffing system can only detect a signal at a fixed frequency. To solve this problem, an equivalent Duffing model is proposed through the introduction of a dimensionless transform coefficient. In the equivalent model, we only need to modify the transform coefficient to detect signals at different frequencies. Besides, it is usually necessary to detect a wide frequency band in practice, while the Duffing spectrum sensing model can only detect signals within a narrow band. To overcome this, an array Duffing oscillator spectrum sensing system is further designed. In addition, since the Duffing method is sensitive to the phase offset, we introduced a pre-processing module in the system to make it more suitable for the practical application. Finally, numerical simulations proved the effectiveness of the proposed method, and the detection performance under AWGN channel is also analyzed. The results show that the proposed Duffing spectrum sensing method can achieve high detection accuracy at very low SNR(-25 dB), thus, it is more applicable in practice.3. Local variance detectors are developed for spectrum sensing in multi-antenna system. Based on the fact that the correlation structure of the received signals differs between the cases where the PU is present and absent, a new concept of local variance is presented. And then three sensing methods are proposed by the local variance. The asymptotic theoretical thresholds of three methods are also derived according to the asymptotic distribution theorem. Since the detail information of the sample covariance matrix is used and the test statistics are constructed by the second-order statistical characteristic of the sampling covariance matrix(SCM), the proposed methods have better performance. To verify this, the detection performance comparisons with other related methods are simulated. The results show that the proposed methods outperform other algorithms and only need small sample numbers, thus the local variance methods have higher sensing accuracy and efficiency.4. A novel spectrum sensing method for noncircular signals is proposed in this dissertation. In the practical communication system, the noncircular signals are frequently encountered. For noncircular signal, its second-order statistical characteristic relies on not only the usual covariance matrix but also the complementary covariance matrix. However, the latter is usually ignored in the existing methods. Based on this, a NC-LMPIT(noncircular-locally most powerful invariant test) sensing method is proposed by using the sample covariance matrix and the sample complementary covariance matrix to construct the test statistic. The corresponding asymptotic theoretical threshold is also derived based on the asymptotic distribution theorem. Numerical results are included to demonstrate the superiority of the proposed method. Since the NC-LMPIT method is able to exploit full statistical property of the noncircular signals, the detection performance has been greatly improved. In addition, there is no need to have any prior information about the channel, the noise and the PU signals. Furthermore, the proposed method can achieve better performance with small sample size and low SNR. Hence, it can be widely used in the spectrum sensing applications.