The subject of this thesis is to study Bayesian estimation/detection algorithms suitable for P and T wave analysis in ECG signals. In this work, different statistical models and associated Bayesian methods are proposed to solve simultaneously the P and T wave delineation task (determination of the positions of the peaks and boundaries of the individual waves) and the waveform-estimation problem. These models take into account appropriate prior distributions for the unknown parameters (wave locations and amplitudes, and waveform coefficients). These prior distributions are combined with the likelihood of the observed data to provide the posterior distribution of the unknown parameters. Due to the complexity of the resulting posterior distributions, we propose to use Markov chain Monte Carlo algorithms for (sample-based) detection/estimation.

Nonlinear power amplifiers (transistors and tubes) are used in telecom transmission because of their power efficiency. These amplifiers produce intermodulation noise (in same bandwidth and in adjacent bandwidth) that impairs the spectrum efficiency. When using a linearizer, a telecom engineer will get a better optimum between spectrum and energy efficiency. Nobody has been able to write a test specification for a good non linearity except that it should be easy to linearize and give good intermodulation results when linearized. In the last 2 decades, cellular telephony base station manufacturers have specified that RF non linear components should be measured under best linearization conditions. Test benches now include an adaptive linearizer for these measurements. After AM/AM and AM/PM measurements of the device, the test bench computes the best possible pre-distortion for the device and then measures efficiency and distortion (such as ACPR)using RF signals that have been pre-distorted by a digital linearizer. The base station manufacturer will then build its power amplifier together with a practical linearizer (real time analogue or digital) targeting this best linearized performance. In the next years, this will be applied also to RF and microwave amplifiers and systems such as telecom satellite links and microwave LANs. Future benches will be more demanding: - Higher centre frequency (up to 60 GHz) - Higher bandwidth (up to some GHz) and much higher sample frequency (keeping the same memory length and depth or more) - Higher spectrum efficiency signals (time and frequency packing, low roll-off, single carrier, multicarrier or OFDM) - Best compromise between energy and spectrum efficiency taking into account nominal RF power and operating point of the device, thermal noise, intermodulation noise and interference (SNIR or SNIIR). The first 2 points will need new hardware, particularly high frequency 12 bits ADC and DAC and fast memory storage. The last ones will need improvements in the simulation of new test signals, the computation of best linearization curves (particularly in the case of memory non-linearity) and the measurement of ACPR, EVM or NPR.

For blind deconvolution of an unknown sparse sequence convolved with an unknown pulse, a powerful Bayesian method employs the Gibbs sampler in combination with a Bernoulli–Gaussian prior modeling sparsity. In this paper, we extend this method by introducing a minimum distance constraint for the pulses in the sequence. This is physically relevant in applications including layer detection, medical imaging, seismology, and multipath parameter estimation. We propose a Bayesian method for blind deconvolution that is based on a modified Bernoulli–Gaussian prior including a minimum distance constraint factor. The core of our method is a partially collapsed Gibbs sampler (PCGS) that tolerates and even exploits the strong local dependencies introduced by the minimum distance constraint. Simulation results demonstrate significant performance gains compared to a recently proposed PCGS. The main advantages of the minimum distance constraint are a substantial reduction of computational complexity and of the number of spurious components in the deconvolution result.

This paper presents a nonlinear mixing model for hyperspectral image unmixing. The proposed model assumes that the pixel reflectances are nonlinear functions of pure spectral components contaminated by an additive white Gaussian noise. These nonlinear functions are approximated using polynomial functions leading to a polynomial postnonlinear mixing model. A Bayesian algorithm and optimization methods are proposed to estimate the parameters involved in the model. The performance of the unmixing strategies is evaluated by simulations conducted on synthetic and real data.

Monitoring and detection of ships and oil spills using synthetic aperture radar (SAR) have received a considerable attention over the past few years, notably due to the wide area coverage and day and night all-weather capabilities of SAR systems. Among different polarimetric SAR modes, dual-pol SAR data are widely used for monitoring large ocean and coastal areas. The degree of polarization (DoP) is a fundamental quantity characterizing a partially polarized electromagnetic field, with significantly less computational complexity, readily adaptable for on-board implementation, compared with other well-known polarimetric discriminators. The performance of the DoP is studied for joint ship and oil-spill detection under different polarizations in hybrid/compact and linear dual-pol SAR imagery. Experiments are performed on RADARSAT-2 -band polarimetric data sets, over San Francisco Bay, and -band NASA/JPL UAVSAR data, covering the Deepwater Horizon oil spill in the Gulf of Mexico.

Wake vortices are associated to the generation of lift when an aircraft is flying. During the take-off and landing phases, wake vortices are hazardous if encountered by other flying aircrafts. In order to ensure flight safety and increase airports capacity as a constraining minimum distance between successive aircrafts has been defined to avoid them, wake vortex monitoring in real time has emerged as one of the key challenge in air traffic management. In this paper, the potential use of an X band Doppler radar for detecting and monitoring wake vortices in rainy weather is assessed by simulation. The Doppler signature measured by an X band radar in presence of a wake vortex in rainy weather is simulated accounting for the backscattering of each individual raindrop in the volume surrounding wake vortices.Starting from a given DSD and a homogeneous repartition of the raindrops in still air, their trajectory is computed assuming a generic air flow induced vortex and a simple model of drag. The descent velocity of the vortex due to the local reduction of buoyancy in the vortices is also taken into consideration in the computation of the trajectory. A scheme for computing the radar signals from the raindrops within wake vortices is described. The X band radar signatures in scanning mode are computed for raindrops in each concerned radar cell. According to the simulation results, the Doppler spectrum width of the raindrops disturbed by wake vortices is extended, thus providing a mean to identify the potential location of wake vortices in the scanned area and therefore to localize the hazards. The potentiality of this tool for the design of inversion algorithms from wake vortices signatures will also be addressed.

This article proposes a stochastic model to obtain the end-to-end delay law between two nodes of a Delay Tolerant Network (DTN). We focus on the commonly used Binary Spray and Wait (BSW) routing protocol and propose a model that can be applied to homogeneous or heterogeneous networks (i.e. when the inter-contact law parameter takes one or several values). To the best of our knowledge, this is the first model allowing to estimate the delay distribution of Binary Spray and Wait DTN protocol in heterogeneous networks. We first detail the model and propose a set of simulations to validate the theoretical results.

Electrical Flight Control Systems (EFCS) now constitute an industrial standard for commercial applications, ensuring a more sophisticated control of the aircraft and flight envelope protection functions. In the scope of a global optimization towards extended system availability and more easy-to-handle aircraft, sophisticated tools are necessary to diagnose sensor behaviour and to prevent from faulty measurements. In this context, a signal processing approach using Partial Least Squares (PLS) has been proposed to increase the EFCS autonomy. This algorithm establishes a linear relation between observed flight parameters - issued from possibly faulty sensors - and independent sensor measurements thanks to an iterative process. A monitoring strategy based on the regression coefficient dynamics, provided by the PLS, has been implemented and tested on real flight data and through an Airbus high-fidelity simulator, including realistic failure scenarios. The simulation results demonstrate the method performance robustness, in compliance with current real-time constraints. This paper focuses on the application of the method to the industrial context. A special attention is dedicated to the simulation process and to the performance analysis.