Research On Modulation Recognition Of Communication Signals Based On Compressive Measurements

This paper mainly researches the algorithms of modulation recognition in compressive domian without reconstruction orignal communication signal. As modulation recognition is an important step between signal detection and demodulation.,correct recognition of modulation significantly influences the following signal processing. As the bandwidth of communication signals become larger, the sample rate becomes higher and the amount of data needed to be processed is larger. The most prominent advantages of compressive sensing is that it can require the wideband sparse signal at the sampling rate far below Nyquist, so compressive sensing is a new choose to solve the sample and process problem of wideband signals. In order to solve the problem of compressive modulation recognition, compressive sensing as the core theory has been studied deeply firstly. The four most important parts of compressive sensing:the representation of sparse signal, the process of compressive measurement, the restricted isometry property and reconstruction algorithm have been introduced. Besides, the analog to information converter for sampling the analog signal based on compressive sensing theory and its mathematical description are analyzed. Second, three compressive modulation recognition algorithms have been presented. In the end, based on the system of developing wideband signal compressive sensing receiver, the initialization configuration of AD9739A、W5300、ADF4350 is fulfilled by simulate SPI of ARM. The communication protocol between ARM and FPGA based on dual-port RAM has been designed and realized. Based on UDP, the communication between PC and W5300 is realized. Then the wideband waveform for compressive sampling has been generated by high-speed digital-to-analog converter AD9739A. The key innovations are summarized as follows:1. To solve problem of the recognition between different modulations, this paper presents an algorithm called compressive high-order cumulants modulation classification (CS-HCMC). Using the property of random subsampling matrix, the algorithm infers the compressive high-order cumulants of different modulations, and constructs some characteristics to class {2ASK(2PSK),4ASK,MFSK,4PSK,16QAM} the five modulations. Compared with the traditional cumulants modulation classification algorithm, CS-HCMC needs fewer samples to obtain the same correct recognition ratio.2. To solve the problem of recognizing the different order of MPSK, this paper presents an algorithm called joint compressive maximum-likelihood MPSK classification and demodulation (CS-MLCD). This algorithm conducts multivariate hypothesis of MPSK, and builds different mathematical models. The probability density function and the likelihood function is inferred in the compressive domain. To realize MPSK modulation recognition, the compressive maximum-likelihood classifier chooses the hypotheses whose value of maximum likelihood function is the max. When the modulation is recognized, the intermediate variables is used to realize demodulation. Compared with the traditional Nyquist maximum-likelihood modulation recognition algorithm, CS-MLCD can reduce the amount of data and promote the real-time property of communication system.3. To solve the problem of recognizing the different order of MFSK, this paper presents an algorithm called compressive Correlation-Index blind classification (CS-CIBC). As MFSK has prominent peaks in frequency domain, this algorithm use the inner product between the compressive measurements and sensing matrix to build the compressive Correlation-Index which is the classification characteristics to realize blind MFSK recognition wthout any priori information. Compared with the spectrum peak modulation recognition algorithm, CS-CIBC can reduce the computation complexity and promote the real-time property of communication system.Besides, based on the developed compressive sampling hardware platform, practical signals are compressive sampled. Using the compressive measurements verifies a modulation recognition algorithm presented in this paper successfully. The results show that the algorithm is effective.