Study On Some Novel Issues Of Digital Communication Signal Modulation Classification

In many communication applications, it is required for the receiver to monitor the activities of spatial signals, identify their characteristics, even to intercept the signal information content with little prior knowledge. For instance, surveillance of electromagnetic spectra activities is used in Electronic Warfare (EW), and management of civilian communication signals is duty of the governmental authorities.Modulation format is one of the most important characteristics used to distinguish communication signals. The automatic recognition of modulation type is a key mission of analyzing communication signals intercepted. Given a received communication signal, the objective of modulation recognition is to decide the modulation format without knowledge of the transformed content, and to provide necessary information for further signal processing tasks.The modulation recognition may be treated as a hypothetical test problem in the sense of signal detection, or a pattern recognition problem with multi-unknown-parameters in the sense of statistical pattern recognition. During the last several decades, considerable progress has been made in the modulation recognition filed, both in features extracting, classifier constructing and classifying algorithm designment and optimalization. However, it is far away to say maturity of this field due to the complex non-cooperate transmission environment, the lasting emergence of new modulation formats and the intrinsical attribute of modulation recognition. A uniform system is absent as the existing methods lack extending ability and a standard for classifying performance is wanted. Further works in this field may include extracting features for recognition which are useable in low signal to noise ratio (SNR) and are robust to variation of SNR, investigating algorithms which are less dependence of unknown parameters or more robust to error of parameters estimation, constructing more efficient classifiers, treating multi-signals in co-channel, classifying novel complex modulation formats, and so on. Based on previous works, this dissertation focus on some of the new problems in modulation recognition of digital communication signals.Our works can be summarized mainly as follows:(1) The background and significance of the study is discussed in detail. We have summed up the lately research status in this field extensively and analyzed some of the existing techniques to bring up new issues of modulation recognition.(2) Based on the advantage of hierarchical decision tree and RBF neural networks, and by adopting some existed excellent modulation recognition algorithms, we have proposed an integrated modulation classification system. This system can recognize the 2ASK,4ASK,2FSK,4FSK,MSK,BPSK,QPSK,8PSK,4QAM,16QAM and 64QAM modulated signals under additional white Gaussian noise (AWGN) environment with quite accurate.(3) We have studied the modulation recognition problem under fading environments. The fading characters cause the response of the transmitting channel to be time-variabling and the produce of intersymbol interference (ISI). The hybrid likelihood function with channel parameters estimation for classifying is discussed over flat slow fading channels. We have also analyzed the performance of modulation recognition algorithm that adopts spatial diversity obtained by multi-antennas.(4) We have researched the modulation recognition problem in non-Gaussian noise. The noise of transmitting channel, though is generally supposed to obey Gaussian distribution in traditional modulation recognition techniques, can be modeled as alpha-stable distribution more suitably. Since there does not exist finite second and higher order statistics of the alpha-stable distribution, we treat the modulated signal as an Auto Regression (AR) process of a stable random variable passing through a linear filter. By using Fractional Lower Order Moments (FLOM), the instantaneous frequency and bandwidth are extracted from the received signals as distinguishing features for modulation recognition.(5) The fractal features for modulated recognition are studied. We have extended the box dimension, which is used as a distinguishing fractal feature in existing works, to multi-fractal spectra features and provided further theoretical demonstration that the new features are robust to SNR variation. These features are extracted based on phase-space reconstruction and correlation integration theory. With the multi-fractal spectra features and RBF classifier, the modulation recognition is completed with more accurate result.(6) We have illustrated that the prediction of hopping-frequency-modulating code is an application of modulation recognition in Spread Spectrum Communication applications. Based on the chaotic character of the code and chaotic series are short-time predictable, we have proposed an algorithm for predicting chaotic time series from the fractal structure of strange attractor and self-affine property of chaotic series. The algorithm exploits the iterative function system to track current chaotic trajectory and constructs prediction model according to attractor and coverage theorem. When use this model to predict hopping frequency code, satisfying results are reported.To sum up, based on the two elementary method for modulation, hypothesis tests and pattern recognition, this dissertation discusses some novel issues of modulation recognition such as the extracting of distinguish features, the design of classifier, modulation classification in fading and non-Gaussian noise environment, the prediction of frequency hopping code. The results of our research reflect the further study direction of modulation recognition and are of use to its theory development as well as to engineering practice.