André Quinquis

Some radar imagery results using superresolution techniques

The key problem in any decision-making system is to gather as much information as possible about the object or the phenomenon under study. In the case of the radar targets, frequency and angular information is integrated to form a radar image, which has very high information content. Salient features can then be extracted in order to characterize or to classify radar targets. The quality of the reconstructed image is mainly related to the resolution performed by the radar system both in slant range and in cross range. The Fourier-based reconstruction methods are fast and robust, but they are limited in resolution and dynamic range. Subspace eigenanalysis based methods, such as multiple signal classification (MUSIC) or estimation of signal parameters by rotational invariance techniques (ESPRIT), are able to provide superresolution and to accurately recover the scattering center locations even for a small number of correlated samples. The aim of the paper is to present some results of our ongoing research on the application of these techniques for radar imagery.

Time-frequency analysis using warped-based high-order phase modeling

The high-order ambiguity function (HAF) was introduced for the estimation of polynomial-phase signals (PPS) embedded in noise. Since the HAF is a nonlinear operator, it suffers from noise-masking effects and from the appearance of undesired cross-terms when multicomponents PPS are analyzed. In order to improve the performances of the HAF, the multi-lag HAF concept was proposed. Based on this approach, several advanced methods (e.g., product high-order ambiguity function (PHAF)) have been recently proposed. Nevertheless, performances of these new methods are affected by the error propagation effect which drastically limits the order of the polynomial approximation. This phenomenon acts especially when a high-order polynomial modeling is needed: representation of the digital modulation signals or the acoustic transient signals. This effect is caused by the technique used for polynomial order reduction, common for existing approaches: signal multiplication with the complex conjugated exponentials formed with the estimated coefficients. In this paper, we introduce an alternative method to reduce the polynomial order, based on the successive unitary signal transformation, according to each polynomial order. We will prove that this method reduces considerably the effect of error propagation. Namely, with this order reduction method, the estimation error at a given order will depend only on the performances of the estimation method.

Time-frequency detection using Gabor filter bank and Viterbi based grouping algorithm

The problem of signal detection, followed by a characterization stage, is considered in this paper. The main difficulties arising in the detection stage are caused by noise, which acts in a real environment, and by multiple time-frequency (TF) structures of the signal. In this paper a detection method based on the adaptive grouping of the TF information provided by a Gabor filter bank is proposed. A Viterbi-type algorithm is used as a tool for grouping of TF components. The results obtained for real data prove the capability of the proposed approach for accurately detect and characterize signals with a complex TF behavior.

Supervised self-organizing classification of superresolution ISAR images: an anechoic chamber experiment

The problem of the automatic classification of superresolution ISAR images is addressed in the paper. We describe an anechoic chamber experiment involving ten-scale-reduced aircraft models. The radar images of these targets are reconstructed using MUSIC-2D (multiple signal classification) method coupled with two additional processing steps: phase unwrapping and symmetry enhancement. A feature vector is then proposed including Fourier descriptors and moment invariants, which are calculated from the target shape and the scattering center distribution extracted from each reconstructed image. The classification is finally performed by a new self-organizing neural network called SART (supervised ART), which is compared to two standard classifiers, MLP (multilayer perceptron) and fuzzy KNN ( nearest neighbors). While the classification accuracy is similar, SART is shown to outperform the two other classifiers in terms of training speed and classification speed, especially for large databases. It is also easier to use since it does not require any input parameter related to its structure.

Localization in Underwater Dispersive Channels Using the Time-Frequency-Phase Continuity of Signals

Time-frequency representations constitute the main tool for analysis of nonstationary signals arising in real-life systems. One of the most challenging applications of time-frequency representations deal with the analysis of the underwater acoustic signals. Recently, the interest for dispersive channels increased mainly due to the presence of the wide band nonlinear effect at very low frequencies. That is, if we intend to establish an underwater communication link at low frequencies, the dispersion phenomenon has to be taken into account. In such conditions, the application of the conventional time-frequency tools could be a difficult task, mainly because of the nonlinearity and the closeness of the time-frequency components of the impulse response. Moreover, the channel being unknown, any assumption about the instantaneous frequency laws characterizing the channel could not be approximate. In this paper, we introduce a new time-frequency analysis tool that aims to extract the time-frequency components of the channel impulse response. The main feature of this technique is the joint use of time-amplitude, time-frequency, and time-phase information. Tests provided for realistic scenarios and real data illustrate the potential and the benefits of the proposed approach.

A time-frequency approach to blind deconvolution in multipath underwater channels

Blind deconvolution is presented in the underwater acoustic channel context, by time-frequency processing. The acoustic propagation environment was modelled as a multipath propagation channel. For noiseless simulated data, source signature estimation was performed by a model-based method. The channel estimate was obtained via a time-frequency formulation of the conventional matched-filter. Simulations used a ray-tracing physical model, initiated with at-sea recorded environmental data, in order to produce realistic underwater channel conditions. The quality of the estimates was 0.793 for the source signal, and close to 1 for the resolved amplitudes and time-delays of the impulse response. Time-frequency processing has proved to overcome the typical ill-conditioning of single sensor deterministic deconvolution techniques.

Denoising underwater signals propagating through multi-path channels

In multipath configurations, an estimation of the channel impulse response is useful for who wants to reduce or cancel undesirable multipath effects. However, the underwater noise does not fulfill the classical white-noise assumption for which matched filer-based algorithms performs optimally. In this context, three denoising methods are studied. The first two are classical methods, based on wavelet packet (WPD) decomposition and uniform filter bank (UFB) decomposition. As an alternative, we propose a novel time-frequency denoising approach, founded on the joint use of unitary warping operators and finite impulse response filters. As shown through experimental results, the WPD-based and UFB-based denoising tools are not well-suited in a multipath context. By contrast, our proposed warping-based method gives good results and leads to an improved estimation of the channel impulse response.

Toward The Use Of The Time-Warping Principle With Discrete-Time Sequences

This paper establishes a new coherent framework to extend the class of unitary warping operators to the case of discrete–time sequences. Providing some a priori considerations on signals, we show that the class of discrete–time warping operators finds a natural description in linear shift– invariant spaces. On such spaces, any discrete–time warping operator can be seen as a non – uniform weighted resampling of the original signal. Then, gathering different results from the non– uniform sampling theory, we propose an efficient iterative algorithm to compute the inverse discrete –time warping operator and we give the conditions under which the warped sequence can be inverted. Numerical examples show that the inversion error is of the order of the numerical round– off limitations after few iterations.

A Time-Frequency Characterization Framework for Signals Issued from Underwater Dispersive Environments

Time-frequency representations constitute the main tool for analysis of non-stationary signals arising from environmental systems. Recently, the interest for underwater dispersive channels appears since dispersivity phenomena act at very low frequencies which are well suited for long range underwater communication. In such a case, a main interest is to perform estimation of the impulse response of such channel for processing purposes. In this paper we introduce a time-frequency analysis tool that aims to extract the time-frequency components of the channel impulse response. This technique is based on the adaptive time-frequency filtering whose parameters are defined by a local chirp matching procedure. Tests provided for realistic scenarios illustrate the potential and the benefits of the proposed approach.

Horizon Picking on Subbottom Profiles Using Multiresolution Analysis

Abstract Maroni, C.-S., Quinquis, A., and Vinson, S., Horizon Picking on Subbottom Profiles Using Multiresolution Analysis, Digital Signal Processing11 (2001) 269–287 A fully automatic algorithm is proposed for the mapping of sediment layers on subbottom profiles. This mapping should significantly speed up data analysis and sedimentary data base building. The proposed method combines two techniques: edge following algorithms and multiresolution analysis using the wavelet transform. Image analysis at different scales allows us to follow sediment layer contours without interruption, and with adequate accuracy. In order to reduce detection errors a thresholding with hysteresis is first performed. The thresholds are automatically set according to data distribution. Performances of the algorithm are discussed, from testing on a set of real subbottom profile data.