Modulation Recognition For Communication Signals In The Alpha-stable Distribution Noise

The aim of modulation recognition for communication signals is to determinemodulation types and some parameters of the received signals in the case of certain noise, andprovide the foundation for subsequent signal analysis and processing. Modulation recognitionfor communication signals has many important applications in both military and civilianfields. Most of previous researches on modulation recognition employed Gaussiandistribution as the model of background noise, but in fact it often exists some significantshort-time impulse noise with large amplitude that can not be simply described by Gaussiandistribution in wireless communications channels. More and more researchers confirmed thatthe Alpha-stable distribution is a more effective noise model. Therefore, researches onmodulation recognition for communication signals in Alpha stable distribution noise havemore practical significance.Alpha-stable distribution noise does not have second-order and higher order statistics, somany of the signal recognition algorithms in Gaussian noise have no longer applies. Thisdissertation aims to study the modulation recognition algorithms for communication signals inAlpha-stable distribution noise, and the main work including:Firstly, the modulation recognition algorithms based on fractal theory are studied. Fractalbox dimension and multifractal spectrum are important concepts in fractal theory, and the twoparameters are not easily affected by the Alpha-stable distribution noise whose characteristicindex ranges from1to2. This dissertation studies the fractal box dimension and multifractalspectrum features of Alpha-stable distribution noise, on this basis, modulation recognitionalgorithms based on the fractal box dimension and multifractal spectrum are separatelyproposed. Recognition algorithm based on the fractal box dimension extracts the fractal boxdimension of signal phase as the recognition feature, and uses BP neural network as classifierto achieve the effective modulation recognition for BPSK, QPSK, OQPSK, MSK and GMSKsignal. The singularity exponent corresponding to the maximum spectrum value and thedifference between maximum spectrum value and minimum spectrum value are effectiverecognition features, and recognition algorithm based on the multifractal spectrum employs athreshold decision method based on these two features to recognize2FSK,4FSK and8FSK signal. Advantage of algorithms based on fractal theory is the small amount of calculation,and easier to project implementation.Secondly, the modulation recognition algorithm based on fractional low-order cyclicstatistics is studied. Fractional low-order cyclic statistics is an effective tool for signalprocessing in Alpha-stable distribution noise, and it has been widely used. Signal parameterestimation is one of the key technologies for modulation recognition preprocessing. MPSKsignals parameter estimation algorithm based on fractional low-order cyclic spectrum isproposed. On the basis of analysing the relationship between signal parameters that to beestimated and corresponding parameters of fractional low-order cyclic spectrum, algorithmuses fractional low-order cyclic spectrum to achieve the effective estimation of carrierfrequency and symbol rate. In addition, this dissertation extends the traditional second-ordercyclic spectrum coherent coefficient by using fractional low-order transform, and putsforward the concept of fractional low-order cyclic spectrum coherent coefficient. Thefractional low-order cyclic spectrum coherent coefficient characteristics of signals to berecognized are analyzed, and its cyclic frequency domain characteristics are extracted asrecognition feature to achieve the effective recognition for BPSK, QPSK,2FSK, MSK andAM signal. Fractional low-order cyclic statistics can effectively inhibit the Alpha-stabledistribution noise, so above two algorithms have better anti-noise performance. But itsfractional exponent needs to be set according to the characteristics index of noise, that resultscertain limitations of algorithm.Finally, modulation recognition algorithms based on the generalized second-order cyclicspectrum and the generalized quartic spectrum are studied. Referencing the idea of fractionallow-order nonlinear transform in the fractional low-order statistics. By using generalizedtransform, this dissertation extends the traditional second-order cyclic spectrum and thetraditional quartic spectrum, and proposes new concepts of generalized second-order cyclicspectrum and the generalized quartic spectrum, that are applied to the communication signalsrecognition in Alpha-stable distribution noise. Using the characteristic that different signalshave different cyclic frequencies, algorithm based on the generalized second-order cyclicspectrum recognizes BPSK, QPSK and OQPSK signal according to the cyclic frequency.Using the characteristic that different signals have different spectrum line positions, algorithmbased on the generalized quartic spectrum achieves effectively recognition for MPSK signals. These two algorithms do not need to consider the characteristics index of noise, so they havea large range of application, and the former is superior to the latter from anti-noiseperformance.