Automatic Modulation Recognition Of Digital Signal

Automatic modulation recognition is one of the key technologies in the areassuch as collaborative communication,signal monitoring and so on, is also animportant part of the software radio technology. Both in the field of military electronicconfrontation and in the civilian spectrum regulatory, modulation recognitiontechnology plays an important role. At present, to improve recognition accuracy andreduce the the algorithm complexity has become the goal of this area. In this paper,based on the cyclic autocorrelation theory of communication signals, some research isdone in this paper in aspects of improvement of estimation algorithm, analysis of thecyclostationary features of modulating signals, the method of modulation recognition,BP neural network classifier and the improvement of its transfer function throughmathematical deduction and computer simulation. In the process of research, someideas are proposed and validated.First of all, in this paper, the estimation algorithm on the basis of frequencysmoothening method is researched in depth. Furthermore, aiming at the randomfluctuating phenomenon of cyclic spectrum caused by the interference of noise andthe restricted number of samples for computation, the wavelet de-noising andsuperposition averaging are introduced to the estimation algorithm. Theoreticalanalysis and simulation results show that the improved algorithm could greatlyeliminate the random fluctuation of the cyclic spectrum, make the signal spectrumcharacteristics clearer.The spectral correlation characteristics of several digital modulating signals areresearched through the improvement of estimation algorithm. These signals includeAmplitude Shift Keying(ASK) signals, Frequency Shift Keying(FSK) signals, PhaseShift Keying(PSK) signals, Minimum Shift Keying(MSK) signal andPhase-Incoherent Frequency Shift Keying(FSK*) signals. On this basis, fourclassification features of the modulating signals with great distinction degree andstable antinoise performance are extracted. These features include normalization ofspectral correlation function f-profile area, and fprofiles absolute maximumratio, signal square spectrum correlation function and zero center normalizedinstantaneous amplitude.Finally, the BP neural network classification is constructed for recognition ofabove modulating signals. Through the analysis of the transfer function of the BP network, aiming at the existing problem of the transfer function, put forward theimproved algorithm, and the simulation verify the effectiveness of the improvedalgorithm.