The Study Of The Spectrum Sensingalgorithm Based On Compressed Sensing

With the rapid development of communication technology and the growing increasement in wireless communication services, people pay much attention to the increasingly scarce spectrum resources. The technology of cognitive radio provides a strong technical support in solving the scarce spectrum problem, management for the dynamic spectrum and improvement in spectrum efficiency. Spectrum sensing of cognitive radio (CR) requires the ability of sensing spectrum quickly and accurately in up to GHz-bandwidth-wide spectrum, asking for high-rate analog-to-digital (AD) converters to match for. It challenges people to search for a new method sampling under traditional rate. The framework of Compressed Sensing (CS) is such a way to break the Nyquist sampling limit, making it possible to reconstruct and estimate signals via fewer measurements than that is required traditionally. However, reconstruction algorithms based on CS are non-deterministic polynomial hard (NP-hard) in mathematics, which is difficult to obtain the real-time analysis result using the general computing methods because of the computing complexity of such a problem .This paper proposes some new compressed sensing scheme to deal with such problems.First of all, taking both the characteristics in the spectrum distribution and the request for spectrum sensing into consideration, this thesis proposes a new spectrum sensing algorithm for the wideband cognitive radio based on Differential Signal Compressed Sensing (DSCS) by the use of the sparsity of spectrum. Based on energy detection technology, the theory of compressed sensing (CS) is used to acquire signals at reduced rates, rather than the classical Shannon-Nyquist rate. To reduce the complexity, as well as to improve the sensing performance, differential signals, instead of normal ones, are taken to acquire spectrum information. What's more, a new tolerance limitation is proposed to halt the iterations for a further reduction in complexity. This new limitation makes the algorithm much easier to match the practical requests-both in time and in computation. Related experiments demonstrate that with a proper tolerance, the wide spectrum sensing methods for Wideband Cognitive Radio based on DSCS leads to a great reduce in both iteration and computation, in addition to a better sensing performance.Secondly, performing badly in co-using and comprehensive treatment for multi-source, multi-dimensional information, single-node detection is weak in reliability, accuracy and practicality, thus multi-node detection is needed. Multi-node centralized data fusion has the disadvantages such as fully data collection, no information lost, and high confidence in final decision, etc. However, it needs large volumes of data transmission, a long time cost in information processing and heavy load in fusion center. Based on compressed sensing theory, this thesis proposes a differential signal based multi-node centralized detection, which inherits advantages of centralized data fusion while with greatly reducement in sampling rate, data transmission and storage time.Finally, a new compressed wide spectrum sensing scheme based on BP neural network is proposed. In this scheme, the BP neural network technology is added into the normal CS-based detection scheme for wideband signals, replacing the reconstruction process. In this way, the computational complexity transfers from reconstruction and estimation to network training process, which can be done before spectrum sensing. As with blocky sparsity character, signals can be detected without destructive reconstruction, leading to input signals without completely retained. So 1-bit quantification is carried out by which the network load can be mitigated. Simulation results show that with 1-bit quantization, the system responds in a short period of time. Compared to normal CS-based detection scheme, our new scheme presents a much shorter consumption time as well as a better robustness performance to noise.