Efficient sampling from high-dimensional distributions is a challenging issue which is encountered in many large data recovery problems involving Markov chain Monte Carlo schemes. In this context, sampling using Hamiltonian dynamics is one of the recent techniques that have been proposed to exploit the target distribution geometry. Such schemes have clearly been shown to be efficient for multi-dimensional sampling, but are rather adapted to the exponential families of distributions with smooth energy function. In this paper, we address the problem of using Hamiltonian dynamics to sample from probability distributions having non-differentiable energy functions such as ℓ1. Such distributions are being more and more used in sparse signal and image recovery applications. The proposed technique uses a modified leapfrog transform involving a proximal step. The resulting non-smooth Hamiltonian Monte Carlo (ns-HMC) method is tested and validated on a number of experiments. Results show its ability to accurately sample according to various multivariate target distributions. The proposed technique is illustrated on synthetic examples and is applied to an image denoising problem.

With the need to provide the Internet access to deprived areas and to cope with constantly enlarging satellite networks, enhancing satellite communications becomes a crucial challenge. In this context, the use of Random Access (RA) techniques combined with dedicated access on the satellite return link, can improve the system performance. However conventional RA techniques like Aloha and Slotted Aloha suffer from a high packet loss rate caused by destructive packet collisions. For this reason, those techniques are not well-suited for data transmission in satellite communications. Therefore, researchers have been studying and proposing new RA techniques that can cope with packet collisions and decrease the packet loss ratio. In particular, recent RA techniques involving information redundancy and successive interference cancellation, have shown some promising performance gains. With such methods that can function in high load regimes and resolve packets with high collisions, channel estimation is not an evident task. As a first contribution in this dissertation, we describe an improved channel estimation scheme for packets in collision in new RAmethods in satellite communications. And we analyse the impact of residual channel estimation errors on the performance of interference cancellation. The results obtained show a performance degradation compared to the perfect channel knowledge case, but provide a performance enhancement compared to existing channel estimation algorithms. Another contribution of this thesis is presenting a method called Multi-Replica Decoding using Correlation based Localisation (MARSALA). MARSALA is a new decoding technique for a recent synchronous RAmethod called Contention Resolution Diversity Slotted Aloha (CRDSA). Based on packets replication and successive interference cancellation, CRDSA enables to significantly enhance the performance of legacy RA techniques. However, if CRDSA is unable to resolve additional packets due to high levels of collision, MARSALA is applied. At the receiver side, MARSALA takes advantage of correlation procedures to localise the replicas of a given packet, then combines the replicas in order to obtain a better Signal to Noise plus Interference Ratio. Nevertheless, the performance of MARSALA is highly dependent on replicas synchronisation in timing and phase, otherwise replicas combination would not be constructive. In this dissertation, we describe an overall framework ofMARSALA including replicas timing and phase estimation and compensation, then channel estimation for the resulting signal. This dissertation also provides an analytical model for the performance degradation of MARSALA due to imperfect replicas combination and channel estimation. In addition, several enhancement schemes forMARSALA are proposed likeMaximum Ratio Combining, packets power unbalance, and various coding schemes. Finally, we show that by choosing the optimal design configuration for MARSALA, the performance gain can be significantly enhanced.

With the need to provide the Internet access to deprived areas and to cope with constantly enlarging satellite networks, enhancing satellite communications becomes a crucial challenge. In this context, the use of Random Access (RA) techniques combined with dedicated access on the satellite return link, can improve the system performance. However conventional RA techniques like Aloha and Slotted Aloha suffer from a high packet loss rate caused by destructive packet collisions. For this reason, those techniques are not well-suited for data transmission in satellite communications. Therefore, researchers have been studying and proposing new RA techniques that can cope with packet collisions and decrease the packet loss ratio. In particular, recent RA techniques involving information redundancy and successive interference cancellation, have shown some promising performance gains. With such methods that can function in high load regimes and resolve packets with high collisions, channel estimation is not an evident task. As a first contribution in this dissertation, we describe an improved channel estimation scheme for packets in collision in new RA methods in satellite communications. And we analyse the impact of residual channel estimation errors on the performance of interference cancellation. The results obtained show a performance degradation compared to the perfect channel knowledge case, but provide a performance enhancement compared to existing channel estimation algorithms. Another contribution of this thesis is presenting a method called Multi-Replica Decoding using Correlation based Localisation (MARSALA). MARSALA is a new decoding technique for a recent synchronous RAmethod called Contention Resolution Diversity Slotted Aloha (CRDSA). Based on packets replication and successive interference cancellation, CRDSA enables to significantly enhance the performance of legacy RA techniques. However, if CRDSA is unable to resolve additional packets due to high levels of collision, MARSALA is applied. At the receiver side, MARSALA takes advantage of correlation procedures to localise the replicas of a given packet, then combines the replicas in order to obtain a better Signal to Noise plus Interference Ratio. Nevertheless, the performance of MARSALA is highly dependent on replicas synchronisation in timing and phase, otherwise replicas combination would not be constructive. In this dissertation, we describe an overall framework of MARSALA including replicas timing and phase estimation and compensation, then channel estimation for the resulting signal. This dissertation also provides an analytical model for the performance degradation of MARSALA due to imperfect replicas combination and channel estimation. In addition, several enhancement schemes forMARSALA are proposed likeMaximum Ratio Combining, packets power unbalance, and various coding schemes. Finally, we show that by choosing the optimal design configuration for MARSALA, the performance gain can be significantly enhanced.

Crowdsensing is, for a few years, a hot topic. Until now, research on crowdsensing mainly focused on scenarios with devices such as smartphones with huge memory and high computive skills. With the development of the Internet of Things (IoT), crowdsensing can be envisaged with other constraints. Indeed, some IoT nodes are mobile but with limitations about storage and processing capabilities, then connectivity disruptions might occur between the nodes. These issues are tackled by a Disruption Tolerant Networking architecture. In this article, we focus on a subset of IoT, Mobile Sensing Networks (MSN). We propose then, a mechanism which respects the constraints of the nodes and maintains high performance. This mechanism, the multi-level FREAK, uses the mean frequency of contacts with the destination. The metrics drives the transmission. Since some nodes might not meet the destination nor nodes meeting the destination, we had the idea of a multi-level metrics to allow these “disconnected” nodes to transmit data to the destination. We evaluate our proposal through simulations based on several real mobility traces. Our solution outperforms reference replication and quota-based DTN solutions.

Spectral analysis of sea level anomalies (SLA) is widely used in the altimetry community to understand the geophysical content of the measured signal, to assess and compare the missions’ performances. Spectral content of SLA is used to characterize the ocean at different scales as well as instrumental noise. Based on the SLA spectrum, one can estimate the spectral slope at medium to large scales (relied to the Surface Quasi-Geostrophic (SQG) ocean dynamics theory) and the measurement noise (observed as a noise plateau at smallest scales). It has already been shown that the spectral slope strongly depends on ocean variability, both in time and space domains [1]. However, spectral analysis based on Fourier transform requires stationary signals and is well-known to suffer from a convolutive bias and a high variance of estimation [2]. Thus, using Fourier transforms for SLA spectral analysis requires mathematical caution and needs to be fully managed. This study aims at reviewing applicability of Fourier transform-based methods to SLA analysis and comparing it to other spectral methods. Such comparison has been performed on both simulated SLA signals obtained from theoretical spectra and real signals from a high-resolution altimeter (Orbit – Range – Mean Sea Surface). Finally, a parametric spectral analysis method is proposed and suggested for use by the wider Cal/Val and altimetry science community. [1] C. Dufau et al., Mesoscale capability of along-track altimeter data in LRM & SARM, OSTST Meeting, 2014. [2] P. Stoica, R. Moses, Introduction to spectral analysis, Prentice Hall, 1997.

Monitoring the drive train of a wind turbine is still a challenge for reducing operationand maintenance costs and therefore decreasing cost of energy. In this paper, astandalone, data-driven and automatic tracking analyzer, entitled AStrion and alreadypresented in this conference, is applied on vibration data acquired during one full yearon a set of sensors located in the nacelle of two wind turbines in a wind farm in thePyrénées (France). These experimentations were realized thanks to KAStrion projectfunded by KIC InnoEnergy program.In the context of a particular case study, the main bearing failure of one of the two windturbines, this paper will highlight three main assets of AStrion strategy. A first asset willbe the application of the data validation module. According to the value of anonstationary index, the data measured on the sensor located on the main bearing closeto the failure have been discarded. This was justified afterwards by a dysfunction of thesensor. Then from the validated data acquired with a more remote sensor, a second assetwill be the trends of global features computed by AStrion which proved a strong linkwith maintenance operations on the mechanical components such as the greasing. Thethird asset will be the reading of other AStrion features associated to one specificcomponent. Indeed the trends of the features of the main bearing show evolutionsthroughout the year. A real time reading would have led to the conclusion of a severeevolution of the condition of this main bearing eight months before the failure and thestop of the machine. This study was carried out thanks to a narrow collaboration withthe operator of the wind farm.

Offering a secured shared hosting environment for web applications is not a trivial task. In addition to a well secured system configuration, up-to-date shared hosting are still exposed to security threats by compromised web applications that serve as spam relay, distributed denial of service actors, phishing page hosting and drive-by download page hosting to name a few. As a result, the availability of the server could suffer from a bad IP address reputation and thus, blocked access to all accounts in the server, not only the compromised account. The emergence of web application firewalls (WAF) manages to close the gap by thoroughly analysing HTTP requests in search of known vulnerabilities. However, as any misuse type mechanism, it falls short at discovering zero-day attacks or already compromised environment. In this paper, an anomaly detection model is proposed as a very helpful tool to start building an efficient intrusion detection system adapted to a specific web application or to assist a forensic analysis. The learning phase does not need past activities nor prior knowledge of the web application and its underlying architecture, making it a very simple yet powerful tool for reducing the access log entries for further analysis.

A detection device comprising a first data processing unit configured to determine a first trend of the control surface control order as a function of time, the first trend being defined according to a reference parameter, a second data processing unit configured to determine at least one second trend of the control surface position value as a function of time, the second trend being also defined according to the reference parameter, and a monitoring unit configured to compare the first and second trends in order to verify the existence of a correlation between these first and second trends so as to detect an oscillatory failure as soon as a loss of correlation between the first and second trends appears, if necessary.

Hyperspectral unmixing is aimed at identifying the reference spectral signatures composing an hyperspectral image and their relative abundance fractions in each pixel. In practice, the identified signatures may vary spectrally from an image to another due to varying acquisition conditions, thus inducing possibly significant estimation errors. Against this background, hyperspectral unmixing of several images acquired over the same area is of considerable interest. Indeed, such an analysis enables the endmembers of the scene to be tracked and the corresponding endmember variability to be characterized. Sequential endmember estimation from a set of hyperspectral images is expected to provide improved performance when compared to methods analyzing the images independently. However, the significant size of hyperspectral data precludes the use of batch procedures to jointly estimate the mixture parameters of a sequence of hyperspectral images. Provided that each elementary component is present in at least one image of the sequence, we propose to perform an online hyperspectral unmixing accounting for temporal endmember variability. The online hyperspectral unmixing is formulated as a two-stage stochastic program, which can be solved using a stochastic approximation. The performance of the proposed method is evaluated on synthetic and real data. A comparison with independent unmixing algorithms finally illustrates the interest of the proposed strategy.

This paper proposes a robust fast multi-band image fusion method to merge a high-spatial lowspectral resolution image and a low-spatial high-spectral resolution image. Following the method recently developed in [1], the generalized Sylvester matrix equation associated with the multi-band image fusion problem is solved in a more robust and efficient way by exploiting the Woodbury formula, avoiding any permutation operation in the frequency domain as well as the blurring kernel invertibility assumption required in [1]. Thanks to this improvement, the proposed algorithm requires fewer computational operations and is also more robust with respect to the blurring kernel compared with the one in [1]. The proposed new algorithm is tested with different priors considered in [1]. Our conclusion is that the proposed fusion algorithm is more robust than the one in [1] with a reduced computational cost.