| Communication reconnaissance is one of the most important methods to obtain military information and is also the basis in implementing electronic interference and damage. With the increasing information level in modern wars, lots of new technologies such as high speed frequency-hopping, burst communication, complicated modulation and code are widely used in military communication field and make military communication present tendency of software, intelligence, broadband and network. However, the application of these new techs would bring big challenges to communication reconnaissance, such as: (1) the intercepted signal could be probably multi-signal with time-frequency overlapping seriously; (2) only small size data samples can be used due to the low interception probability communication technology; (3) the received signal's qulity would be much more worse because of non-line-of-sigh reconnaissance, sidelobe reconnaissance and multipath effects.Therefore, signal processing in communication reconnaissance is not a simple task but a problem more diversity and complex. Base on the consideration, a new complete signal process should have the capability of multi-signal separation, channel equalization, parameters estimation adapted to low SNR and small size data, and modulation classification in complicated envirement. The main contributions of this dissertation are studying the solutions to such problems, they are summarized as following:1. The non-data aided signal to noise ratio (SNR) estimation problem is first studied. After comparing various information criteria performance in estimating SNR, the suitable criterion is given in practical application. The effects when small size data is used in calculating eigenvalues of sample covariance matrix are also analized with the help of random matrix theory. By using the properties that the likelihood function of AIC criteria depends only on function of noise space eigenvalues, a new robust SNR estimation method is proposed with the application of modified-AIC criterion. 2. Based on the characteristics of two parallel uniform linear arrays (ULAs), a novel 2-D ESPRIT algorithm has been proposed. The proposed method employs the shift invariance property of the array geometry and it can estimate precisely for avoiding decreasing the array aperture. This proposed method not only has an advantage of automatically parameter alignment but also handle sources with common 1-D angles.3. The limitation of nc-FastICA method based on Newton iteration is analized, which can not separate the modulated non-circular signal, for instance, BPSK signal. The reason is that the kurtsis of modulated BPSK signal changed and exceeded the stable range of nc-FastICA method. Therefore a general non-linear function is proposed and applied to the separation algorithm, and the robust separation of modulated non-circular signal is completed.4. With the aim of solving problem that channel blind identification methods based on second-order characteristics perform badly when DSSS signals are used to be sources, a criterion is derived to determine appropriate starting point of receiving data as a preprocessing algorithm. The new method avoids channel order estimation and shows more robust in the situation when small leading and/or trailing terms are existed.5. Finally two time-frequency methods are proposed which are based on wavelet and Hilbert-huang transformings and complete the intra and inner modulation classification of higher order PSK-type and QAM-type signals. By using the uncorreration property of Gaussian noise at different moments, a new frequency domain classification algorithm is presented which can adapt to work under lower SNR scenario. An idea of joining array receiving mode and dual mode blind equalization is also given as to the modulation classification problem under multipath channel. |
PDF Full Text Request