A Study Of Algorithms For Single Carrier Modulation Classification In Non-cooperative Communication

Research of modulation recognition techniques is one of the key technologies ofreceiver in the non-cooperative communication systems. It is of significance in bothcivilian and militarily applications. However, along with the rapid development ofcommunication technologies, it is far away to say maturity of this field due to thecomplex non-cooperative transmission environment, the lasting emergence of newmodulation formats. It is partially reflected in the following points: correct classificationrate is very low for high order QAM signals in low signal to noise ratio (SNR) inGaussian channel, the intra-class recognition of QAM signals is difficult in the presenceof frequency offset and phase offset in fading channel, the recognition results ofdifferent single-carrier modulation signals are not satisfactory in multipath fadingchannel, and how to optimize the classifier is still a problem in the case of smallsamples. This dissertation focuses on these new problems. Our works can besummarized mainly as follows:1. In order to solve the recognition rate of QAM modulation signals is low inGaussian channel when carrier frequency is not known, firstly put forward a modifiedalgorithm for estimating carrier frequency which can improve the accuracy of carrierfrequency estimation. Furthermore, the novel clustering method is proposed which usesthe combination of subtractive clustering (SC) and particle swarm optimization (PSO)to extract the discriminating features, and applied to the large sample and the smallsample. Compared with existing methods, the new algorithm can improve theidentification rate.2. The performance analysis of large sample estimation of higher-order cumulant ofQAM is addressed. Taking the large sample estimation of fourth order cumulant forexample, we derives the Cramer-Rao lower bound（CRLB）from the theory, andcompare the obtained CRLB with the sampling estimation variance of the fourth ordercumulant. Simulation results show that as the SNR increases, estimation variancegradually approaches the CRLB and average normalized absolute deviation of theestimation is almost close to zero when SNR is2dB. It verifies that this estimation isprogressive unbiased consistent estimation. These results also can be extended toperformance analysis of other large sample estimation of higher-order cumulant andprovide a theoretical basis for later chapters which use higher-order cumulant to extractclassification feature.3. The modulation classification for QAM signals in Rayleigh fading channels is studied in the presence of frequency offset and phase offset, and two modified methodsare proposed. The first method can classify QAM signals using the feature parameterwhich is extracted from two-order, four-order and six-order cumulants without the priorinformation. The performance of the classifier to phase jitter and frequency offset isvery well. The second modified method is addressed based on the equal gain diversityreception, which uses non-linear least-squares estimator to estimate the frequencyoffsets and phase offsets, and uses cumulants of the received signals to extract thefeatures. Simulations demonstrate that the proposed scheme is efficient in Rayleighfading channels.4. The classification algorithm of different single carrier signals in fading channelsis studied. A new classification method using wavelet transform (WT) and higher-ordercyclic-stationary cumulants (HOCC) is presented which extracts the features using thelinearity property of WT and cumulative property of HOCC. Through theoreticalanalysis, it is certified that the extracted feature parameter can eliminate the influence ofmulti-path channel parameter. Monte Carlo simulation results show that in identificationof2ASK、BPSK and QPSK signal the proposed method yields the correct rate of100%in multi-path fading channel when the signal-to-noise ratio (SNR) is0dB. Comparedwith the higher-order cumulants based method, this method can achieve betteridentification performance.5. The dissertation studies the support vector machine classifier of modulationrecognition algorithm in the case of small samples. A single-carrier modulated signalsclassification algorithm using the PSO and SVM is proposed, which utilizes globalsearching property of particle swarm optimization algorithm to optimize thehyper-parameter of SVM, and implements the automatic classification of the modulatedsignals as2PSK,4PSK,8PSK,16QAM and64QAM. Simulation results show themethod can effectively improve the classification performance.