Cyclostationary Feature Detection Based On Compressed Sensing Research In Cognitive Radio

With the fast development of wireless communication, the radio spectrum is a requisite natural resource, but the usage efficiency in some band is very low. Therefore, cognitive radio is proposed to solve the contradiction. Cognitive radio means that the secondary user can dynamic access to the spectrum holes to share the wireless resource without causing the interference to the primary user. As one of the most important tasks for dynamic spectrum access and spectrum sensing in CR, the fast and accurate spectrum sensing over an ultra-wide bandwidth is a primary task. Acquiring the information of the interest is essential for spectrum sensing, Shannon Theorem that sampling rate is at least twice of the highest frequency of the bandlimited width lays the foundation of modern signal processing. In terms of the explosive data service requirements nowadays, wideband analog signals push contemporary analog-to-digital conversion (ADC) systems to their performance limits, the application of compressed sampling (CS) in communication is for this. CS theory asserts that one can recover certain signals and images from far fewer samples or measurements than traditional methods use. Thus, compressed sensing is a hot pot in research.This thesis focuses on the cyclostationary feature detection based on spectrum sensing of cognitive radio and compressed sampling, describe the principle and realization of compressed sampling in detail and analyze the feasibility of combining with the cyclostationary feature detection. CS can deduce the sampling rate in RF terminal, while the cyclic spectrum of the signal is sparse and is effective in low SNRs for detecting spectrum holes, thus, ultra-wide spectrum sensing in low SNR can be realized by combining the both technology.To sum up, the thesis is constituted of the following aspects.First of all, this paper makes a deep investigation on compressed sampling and cyclostationary feature detection and analyzes the feasibility of combining them basedon their own characteristics. Communication signals are all modulated signals withperiodicity, showing distinct feature in cyclic specrum, so they can be detected bycyclostationary feature detection effectively. However, much more measurements arerequired for cyclostationary feature detection. Cyclic spectrum is highly sparsesatisfying the condition of compressed sampling, so it can be recovered bycompressed samples. Cyclostaionary feature detection based on compressed samplingis the research emphasis, dyanmic compressed sampling mechanism based oncyclostationary feature detection and a feature detector based on compressed sensingand wavelet transform are proposed innovatively. The traditional compressed sampling aims at reconstructing the signal perfectly,however, we just use the samples to detect the spectrum holes, regardless of thesymbol rate, modulation type, carrier frequency, only to unveil the fact that whetherprimary user occupy the channel, lowering the requirement of recovery accuracy. Afeedback mechanism of using the feature statistic to guide the sampling rate isproposed to deduce the sampling rate required to detect spectrum holes. The reductionin measurements means lower sensing costs and lower energy consumption. The CS process will inevitably bring in noise because of the compression oforiginal data into smaller number of sample points. The noise introduced by CS willcause degradation to sensing performance as it will cut down SNR, making noisereduction after CS necessary. Otherwise, missing detection probability is increased,causing interference with the primary users. Noise deduction of wavelet transformtheory is investigated in the thesis, besides, SCF is viewed as a grey image withfrequency in the horizontal direction and cyclic frequency in the vertical directionwith the cyclic feature points correspondence to the edges of the image. A shrinkagefactor is derived to modify the wavelet coefficients then inverse wavelet transform isconducted to recover the SCF to make detection. The proposed feature detector canget a better probability of detection in low SNR cases, especially for QPSK. Finally, the conclusion is given in the end of this thesis, there are still some limitsin the proposed schemes and the author also points out the research filed in the nearfuture.