Research On Wideband Specrum Detection Method Based On Compressed Sensing

In recent decades, wireless telecommunication technology is undergoing a rapid development, of which the most representative are cellular mobile communication systems, wireless local area networks, satellite communications, short distance wireless communications and group communications, etc. In current fixed mode of spectrum allocation, spectrum resources become extremely scarce for redistribution or direct use, while the utilization efficiency of authorized users is very low. Hence hot topic "cognitive radio" attracts much attention, which is a smart system improving the spectral efficiency. CR not only grants primary users priority to occupy frequency bands, but also provides secondary users with access, as long as they don’t cause performance degradation. Based on such goal, secondary users should communicate in idle frequency bands through self-adjustment. In fact a variety of literatures written by domestic and foreign scholars have carried out in-depth theories and achievements, including spectrum detection, cognitive engine dynamic spectrum management, end-to-end reconfiguration etc. Among them, spectrum sensing is one of the most significant issues in CR system.Spectrum sensing involves two purposes. First, once collided with primary users, secondary users must conduct frequency switching or limit transmission power, in an effort to keep performances of licensed users over the required lower bound. Second, secondary users should utilize spectrum holes as much as possible to meet the requirement of cognitive radio system. Apparently, the accuracy of spectrum sensing is the crucial point here. Scholars launch the spectrum sensing researches from the aspects of spectral energy, spectrum characteristics, collaborative detection etc. Several detection algorithms are proposed to improve the detection accuracy and reduce the burden of individual cognitive nodes, e.g. energy detection algorithm, matched filter detection algorithm, circular feature detection algorithm, cooperative spectrum detection algorithm.Nonetheless, an overwhelming trend of higher mobility, wider bandwidth and faster data speed makes future communication systems more challenging. For instance, broadband wireless communication should support up to Gbps, leading to much bigger bandwidth, increasing difficulty of resource allocation and spectrum sensing. The spectrum sensing should not only be performed in a wide band range, but also keep fast, efficient and dynamic characteristics. Therefore, researches on wideband spectrum sensing in cognitive radio systems are necessary to adapt to the future high data-rate wireless communication system. Generally, methods of the wideband spectrum detection are classified into two categories:local single node of narrow band parallel detection; and single node broadband detection. For the former method, each node is equipped with one RF front end, and samples the entire wide band spectrum through a single ADC. According to Shannon sampling theorem, it is hard to sample very large spectrum range without the ultra high-speed ADC chip, which is very expensive and difficult to obtain although allowing GHz sampling rate. This problem has become a bottleneck of broadband spectrum sensing development.Compressed sensing theory has brought revolutionary breakthrough for data acquisition technology. By combining the compressed sensing theory and broadband spectrum detection, one can solve the problem of insufficient sampling frequency of ADC hardware. Wideband spectrum detection technique based on compressed sensing is an emerging signal processing technique, which performs the wideband spectrum detection using the sparsity of signals. However, this technique is still in the initial stage. Most researchers directly apply the CS theory to the wideband spectrum detection in cognitive radio, which makes the back-end recovery algorithm very complicated and results in the problem of serious delay in reality.In fact, wideband cognitive radio system is environmentally dynamic, spectrally heterogeneous, and extreme wide in frequency band. In this thesis, the internal relationship between AIC and broadband heterogeneous spectrum is first analyzed based on the compressed sensing theory. Then, according to the demand characteristics of the wideband spectrum detection in cognitive radio, a dynamic spectrum detection model is built for wideband cognitive radio system, through analyzing the inherent coupling mechanism between the sparse signal model and prior information of signal source. Finally, under the primary users’network interference tolerance constraints, a suitable framework of current cognitive radio system is studied and designed, together with a wideband spectrum detection strategy in accordance with the specific spectrum characteristics of the signal. Our work is able to make cognitive radio technology more suitable for future broadband communication system, maximize the utilization of spectrum resources, and provide alternative solutions for the further application of cognitive radio technology.The main contributions of this paper are as follows:1. A sounding signal detection method is proposed for non-sparse signals. The spectrum estimation algorithm is based on compressed sensing, which simply exploits data obtained from observation of compressed samples, so as to reduce the data amount, algorithm complexity and detection delay. The aim is to solve the contradiction between limited computational ability at cognitive nodes and high complexity of compressed sensing recovery algorithm.Under the framework of compressed sensing, the method uses priori information of primary users’spectrum allocation to design matched pattern of sounding signals. Without recovery of the sampling signal, it performs linear operation on sampling data in the compressed domain to retain sounding signals only in spectrum holes, using the linear arithmetic properties of DFT. Then, through back-end signal processing model, parameter estimates of non-sparse signals are directly obtained from the observed compressive sampling value. Last, a simulation experiment is designed to verify the proposed sounding signal detection method. Simulation results indicate that the method can improve the detection accuracy, reduce the complexity of reconstruction, and enhance the robustness against received signal types.2. According to the compressed sensing framework, a frequency sampling structure is proposed based on frequency-domain sampling theory. The structure adds a transforming matrix in the sampling process, and makes full use of the sparse characteristic of the primary user signals in the transformed domain. A crucial problem that transform bases don’t exist in current sampling structure is successfully solved. Moreover, it improves the detection accuracy of frequency domain sparse signals.Through the study of the existing sampling structure, a frequency domain sampler is also designed, which is called the frequency-domain random demodulator (FRD). The sampler is comprised of multiple parallel channels, which preprocess the received signal with separate random sampled signal in frequency domain. Then the processed signal streams sequentially enter an integrator, a low speed ADC and the decision device, obtaining the compressed measurement value of analog signals. Verified by the simulation experiment, the proposed FRD sampling structure can correctly perform frequency-domain sampling under the current hardware condition, which can improve detection accuracy of frequency-domain sparse signals and reduce the complexity of reconstruction compared with the existing sampler. Therefore, the application scope of front-end hardware is also greatly expanded.3. An OFDM signal detection method based on block compressed sensing is proposed, which utilizes the structured feature of OFDM signal in frequency-domain. Via mining the block-sparse characteristic, the structural measurement matrix takes the place of random measurement matrix. In terms of signals with high-standard sparsity or low block sparsity, the method efficiently reduces high reconstruction complexity and error probabilities.At the same time, by analyzing the signal with known special structure, a block sparse model is constructed, which relies on prior information of signal and structural characteristics of OFDM signal. Given the measurement matrix matching the signal structure, extremely little measurement data is required to perform combination subspace vector transformation. Still, a simulation platform is designed to verify the detection performance. Compared with traditional method, the proposed method improves the detection accuracy of the block sparse signal, reduces the reconstruction complexity of the OFDM signal, and expands the standard sparsity condition to a wider range.