R. Hippenstiel

Wavelet-based detection of frequency hopping signals

We investigate the application of wavelet transforms in the detection and estimation of spread spectrum frequency hopping signals. The technique developed makes only two basic assumptions of a minimum hopping time and a minimum frequency hopping differential. The approach is based on the phase information of the temporal correlation function and the resulting discrete wavelet transform of the phase information is used to estimate the hopping time of frequency hopping signals. Results show the proposed scheme is robust to additive white Gaussian noise degradations for SNR levels of 3 dB and above.

Instantaneous Power Spectrum

This paper examines the spectral content of time varying signals using a modified version of the Instantaneous Power Spectrum (IPS). The Pseudo Wigner-Ville Distribution (PWD) has been used extensively to display the time varying spectral information. Problems of PWD are spectral cross terms and endpoint resolution. The modified IPS is shown to be much less sensitive to these types of problems. Furthermore, it can operate with real signals sampled at the Nyquist rate. Representative cases are provided.

Wavelet transforms of correlation functions of frequency hopped signals

This work uses the so called instantaneous correlation function to extract features and parameters of frequency hopped signals. Wavelet transforms allow a faster processing throughput than conventional Fourier transforms do. Wavelet analysis of the correlation surface is performed to identify the modulation and to obtain estimates of signal parameters.

Modulation identification using neural networks and wavelet domain based approaches

Digital communication signals are processed to identify the type of modulation. This paper presents several techniques for the classification of digital modulations. One of the techniques is based on training and testing feedforward neural networks using time or frequency-domain information, or the Shannon entropy. The second technique is based on estimation of higher order moments obtained in the wavelet domain. The classification techniques are tested and compared using a set of simulated signals at different SNR. In particular, ASK, FSK, BPSK, QPSK, and 16PSK modulation schemes are investigated.

The use of wavelets to identify frequency hopped signals

Digital signal processing is used in the interception of digital communication signals. Due to the inherent security features, frequency hopped signals (FH) are widely used in both military and commercial communication applications. The interception of FH signals can be addressed in several ways. In this work we use the instantaneous correlation function (ICF) to represent non-stationary signals. We relate the wavelet transform of the ICF surface of a signal to its Wigner-Ville distribution (WVD). The signals can be observed in the scale surfaces of the wavelet domain. An energy analysis is applied to the surfaces (in the wavelet domain) to identify the scale of the FH signal and to estimate the hop frequency. The FH signals can also be identified by inspecting the pattern of the scales of a multiple-hop-observation interval. If all hop frequencies are within one wavelet scale, then the FH signal can be identified by a set of distinct hop frequencies.

On time-series analysis and signal classification - part I: fractal dimensions

Time series analysis is becoming an increasingly reliable tool for the study of complicated dynamics in measurements across many fields of science and engineering. This paper explores the applications of nonlinear time series analysis for digital communication signal classification. In particular, the fractal dimension was investigated as a tool for signal classification. Primarily, the fractal dimension was calculated for a set of white Gaussian noise as well as for a pure sinusoid. The effect of added DC component as well as the noise variance on the fractal dimension was also investigated. Moreover, the fractal dimension of a set of simulated signals is calculated and investigated for possible use as tool for modulation classification. Furthermore, a time-domain feed-forward neural network was trained and tested for digital signal classification. The success rate of the neural network was used as benchmark for assessment. The method is applied to several examples of synthetic signals, of digital modulations such as ASK, FSK, BPSK, QPSK, and 16PSK.

Wavelet-based denoising: comparisons between orthogonal and non-orthogonal decompositions

In this study we present a denoising scheme applied to a non-orthogonal wavelet transform and compare the performances obtained with those based on orthogonal decompositions. We investigate the effect due to small sample sizes, specific thresholding levels, and initial thresholding scales, and compare results obtained for orthogonal and non-orthogonal transforms.

Feature extraction from digital communication signals using wavelet transforms

Proportional bandwidth processing and wavelet transforms are applied to extract transient features from digital communication signals. Switching times of noisy BPSK, QPSK, FSK, and ASK signals are detected. The scalogram based on a variety of wavelet functions is used to detect the switching times above a threshold signal to noise ratio. Classical wavelets, proportional bandwidth processing schemes and the Morlet wavelet transform are applied.

Difference of arrival estimation using wavelet transforms

Accurate localization of a mobile wireless communication unit via a Time Difference of Arrival (TDOA) method is addressed. Wavelet denoising is used to enhance the TDOA estimation. A 79 to 81 percent improvement relative to raw data processing is obtained using a modified approximate maximum likelihood method or its time varying enhancement, respectively.

Classification of wideband transient signals using spectral-based techniques

Spectral-based classification schemes designed to separate various wideband transient signals are considered and their performances compared to those obtained using a back-propagation neural network implementation. Spectral-based measures considered include the Bhattacharyya distance, the divergence, the normalized cross-correlation coefficient, and the modified normalized cross-correlation coefficient. Results show that accurate classification may be obtained using spectral-based measures and that the performances compare, or are sometimes better, to those obtained using neural networks when the training data used to train the neural network is small. In addition, the spectral-based measures are simple and computationally inexpensive.<<ETX>>