PSTN Voice Band Data Classification And Modulation Recognition

Along with the development of communication transmission technology to high speed digitized and integrated direction, non-voice service proportion that occupies in the public switch telephone network is getting more important, especially the VBD service. The research process is a process of using almost all kinds of signal processing theory to analysis the PSTN signal, including modern signal processing, speech recognition, modulation analysis, pattern recognition, optimized computation, etc. Therefore, PSTN data's research and application have the important theory significance and application value.In this paper, the signaling and telephone timeslots relationship analysis, VBD signal identification and the modulation recognition is presented.1. Based on the dynamic programming thought, an optimal path retrieval algorithm is firstly used in corresponding relationship analysis on the signaling and the telephone timeslots. First, in the process of communication establishment and release, the time and frequency characteristic are extracted, which are used to judge the telephone process's start and the end. Then, based on the parse of No.7 signaling, the signaling information and the telephone timeslot's corresponding relationship is found out. Last, through establishing the distance matrix and calculating the minimum edit distance, the process of optimal path retrieval is completed. The obtain ability of voice and VBD signals is improved.2. Based on ERRB algorithm, a selective ensemble SVM classification algorithm is used in the voice and VBD classification. First, a definition of classifier ensemble's difference is given. Then, based on two level structure's dynamic stacking algorithm, the classifier output is given. In the training stage, this method can accurately choose the classifiers that have the high recognition precision and the high difference. In the test stage, a dynamic ensemble algorithm is used to guarantee that the achievement of VBD signal's effective separation. Experiment result indicates that the algorithm is proposed in this paper have good classification effect and low operation complexity.3. In this paper, a differential power spectrum sliding compensating(DSC) algorithm is presented for frequency estimation of single sinusoids in colored noise at low SNR. The power spectrum is smoothed by comparing every frequency bin and a set of nearby bins. Then, the frequency estimation will be given by peak-searching algorithm. By comparing with other correlative algorithm, the result of simulation indicates that the new algorithm can work well.The DSC-CZT algorithm is proposed, which is used in the Baud rate estimation. This method can effectively eliminate the colored noise's influence and improve the frequency resolution by the CZT. At the low SNR condition, using short length data can obtain the high estimate precision.The average spectrum conversion Rife(ASCR) algorithm is proposed, which is used in the carrier frequency estimation. The influence of noise is reduced by using segmental averaging spectrum. By converting the frequency, the Rife algorithm's estimation performance is improved.4. To the modulation classification, based on the DSC spectrum, two new feature parameters are proposed, and a time domain parameter is improved. These parameters have low complexity and good anti-noise ability. To RBF neutral network, the hidden layer training algorithm is optimized by simulated annealing algorithm. Then, an optimal stopping algorithm is proposed to avoid the overfitting issue. Based on optimal neutral network, a most optimal weights'ensemble algorithm is proposed. Experiment indicates that the identification rate as well as generalization ability is improved.The application of neutral network ensemble algorithm to the VBD recognition is described. According to the VBD signal characteristics, two new feature parameters and the special characteristic vectors construction method are proposed.Finally, we summarize our research work for this thesis and discuss the further research topics and directions.