### Article de revue avec comité de lecture (14)

AN Lin, LI Ming, BOUDAREN Mohamed El Yazid, PIECZYNSKI Wojciech**Unsupervised segmentation of hidden Markov fields corrupted by correlated non-Gaussian noise**. International journal of approximate reasoning, november 2018, vol. 102, pp. 41-59

abstract

*Pixel labeling problem stands among the most commonly considered topics in image processing. Many statistical approaches have been developed for this purpose, particularly in the frame of hidden Markov random fields. Such models have been extended in many directions to better fit image data. Our contribution falls under such extensions and consists of introducing two new models allowing one to deal with non- Gaussian correlated noise. The first one is purely probabilistic, whereas the second one calls on Dempster-Shafer theory of evidence, both being particular triplet Markov fields. The interest of the proposed models is assessed in unsupervised segmentation of sampled and real images. While both models exhibit significant improvement with respect to classic models, the evidential model turns out to be of particular interest when the hidden label field presents fine details*

CHEVALIER Pascal, CHAUVAT Rémi, DELMAS Jean-Pierre**Enhanced widely linear filtering to make quasi-rectilinear signals almost equivalent to rectilinear ones for SAIC/MAIC**. IEEE transactions on signal processing, march 2018, vol. 66, n° 6, pp. 1438-1453

URL: http://www-public.tem-tsp.eu/~delmas/articlesPDF.JPD/IEEE/49_2017.pdf

[PDF]

abstract

*Widely linear (WL) receivers have the capability to perform single antenna interference cancellation (SAIC) of one rectilinear (R) or quasi-rectilinear (QR) co-channel interference (CCI), a function which is operational in global system for mobile communications (GSM) handsets in particular. Moreover, SAIC technology for QR signals is still required for voice services over adaptive multi-user channels on one slot (VAMOS) standard, a recent evolution of GSM/EDGE standard, to mitigate legacy GSM CCI in particular. It is also required for filter bank multi-carrier offset quadrature amplitude modulation (FBMCOQAM) networks, which are candidate for 5G mobile networks, to mitigate inter-carrier interference (ICI) at reception for frequency selective propagation channels in particular. In this context, the purpose of this paper is twofold. The first one is to get more insights into the existing SAIC technology, and its extension to multiple antenna called MAIC, by showing analytically that, contrary to what is accepted as true in the literature, SAIC/MAIC implemented from standard WL filtering may be less efficient for QR signals than for R ones. From this result, the second purpose of the paper is to propose and to analyze, for QR signals and frequency selective fading channels, a SAIC/MAIC enhancement based on a three-input WL frequency shift (FRESH) receiver, making QR signals always almost equivalent to R ones for WL filtering in the presence of CCI. The results of the paper, completely new, may contribute to develop elsewhere new powerful WL receivers for QR signals and for both VAMOS and FBMC-OQAM networks in particular.*

CHEVALIER Pascal, DELMAS Jean-Pierre, SADOK Mustapha**Third-order Volterra MVDR beamforming for non-Gaussian and potentially non-circular interference cancellation**. IEEE transactions on signal processing, september 2018, vol. 66, n° 18, pp. 4766-4781

URL: http://www-public.imtbs-tsp.eu/~delmas/

[PDF]

abstract

*Linear beamformers are optimal, in a mean square (MS) sense, when the signal of interest (SOI) and observations are jointly Gaussian and circular. Otherwise, linear beamformers become sub-optimal. When the SOI and observations are zero-mean, jointly Gaussian and non-circular, optimal beamformers become widely linear (WL). They become non-linear with a structure depending on the unknown joint probability distribution of the SOI and observations when the latter are jointly non-Gaussian, assumption which is very common in radiocommunications. In this context, the paper aims at introducing, for small-scale systems, third-order Volterra minimum variance distortionless response (MVDR) beamformers, for the reception of a SOI, whose waveform is unknown but whose steering vector is known, corrupted by non-Gaussian and potentially non-circular interference, omnipresent in practical situations. Properties, performance, complexity and adaptive implementation of these beamformers in the presence of non-Gaussian and potentially non-circular interference are analyzed in this paper. These new beamformers are shown to always improve, in the steady state, the performance of Capon beamformer for non-gaussian/circular interference, whereas some of them improve the performance of the WL MVDR beamformer for non-Gaussian/non-circular interference. These new beamformers open new perspectives for spectrum monitoring of non-Gaussian signals and for radiocommunication networks using such signals.*

COURBOT Jean-Baptiste, MONFRINI Emmanuel, MAZET Vincent, COLLET Christophe**Oriented triplet Markov fields**. Pattern recognition letters, février 2018, vol. 103, pp. 16-22

abstract

*Hidden Markov Field modeling is widely used for image segmentation. However, it sometimes lacks power to handle complex situations, e.g. correlated noise, textures or non-stationarities. This is why Pairwise, and then Triplet Markov Fields were introduced to handle in a generic fashion more complex observations. In this paper, we tackle the problem of anisotropic image modeling by introducing an Oriented Triplet Markov Field model, able to explicitly deal with oriented structures. Using oriented features in the framework of Triplet Markov Field modeling, we compare the behavior of this model towards other Markovian modeling on images containing such oriented pattern. We present experiments on synthetic data for segmentation, and application to real data from remote sensing images*

DELANDE Emmanuel, FRUEH Carolin, FRANCO Jose, HOUSSINEAU Jérémie, CLARK Daniel**Novel multi-object filtering approach for space situational awareness**. Journal of guidance, control, and dynamics, january 2018, vol. 41, n° 1, pp. 59-73

[PDF]

abstract

*Surveillance activities with ground-based assets in the context of space situational awareness are particularly challenging. The observation process is indeed hindered by short observation arcs, limited observability, missed detections, measurement noise, and contamination by clutter. This paper exploits a recent estimation framework for stochastic populations for space situational awareness surveillance scenarios. This framework shares the flexibility of the finite set statistics framework in the modeling of a dynamic population of objects and the representation of all the sources of uncertainty in a single coherent probabilistic framework and the intuitive approach of traditional track-based techniques to describe individual objects and maintain track continuity. We present a recent multi-object filtering solution derived from this framework, the filter for distinguishable and independent stochastic populations, and propose a bespoke implementation of the multitarget tracking algorithm for a space situational awareness surveillance activity. The distinguishable and independent stochastic populations filter is tested on a surveillance scenario involving two ground-based Doppler radars in a challenging environment with significant measurement noise, limited observability, missed detections, false alarms, and no a priori knowledge about the number and the initial states of the objects in the scene. The tracking algorithm shows good performance in initiating tracks from object-generated observations and in maintaining track custody throughout the scenario, even when the objects are outside of the sensors' fields of view, despite the challenging conditions of the surveillance scenario*

GORYNIN Ivan, GANGLOFF Hugo, MONFRINI Emmanuel, PIECZYNSKI Wojciech**Assessing the segmentation performance of pairwise and triplet Markov models**. Signal processing, april 2018, vol. 145, pp. 183-192

abstract

*The hidden Markov models (HMMs) are state-space models widely applied in time series analysis. Well-known Bayesian state estimation methods designed for HMMs, such as the Baum-Welch algorithm and the Viterbi algorithm, allow state estimation with a complexity linear in the sample size. We consider recent extensions of HMMs, specifically the pairwise Markov models (PMMs) and the triplet Markov models (TMMs), in which the Baum-Welch algorithm also has a complexity linear in the sample size. However, the state process is not necessarily Markovian in PMMs and TMMs, which offers a considerable flexibility of modeling. This study explores potential performance gains achievable if PMMs and TMMs are used to describe the state-space system rather than HMMs. This is done through extensive comparative Monte-Carlo experiments among HMMs, PMMs and TMMs in the case of discrete state space models. A simple comparative example of the use of PMMs and HMMs to predict market direction is also given. These experiments confirm the interest of PMMs and TMMs in the time series modeling: specifically, the classification rate can be improved by nearly fifty percent. These findings mean that PMMs and TMMs may be more suitable than classic HMMs for real-world applications*

HOUSSINEAU Jérémie, CLARK Daniel**Multitarget filtering with linearised complexity**. IEEE transactions on signal processing, september 2018, vol. 66, n° 18, pp. 4957-4970

abstract

*An algorithm for the estimation of multiple targets from partial and corrupted observations is introduced based on the concept of a partially distinguishable multitarget system. It combines the advantages of engineering solutions like multiple hypothesis tracking with the rigor of point-process-based methods. It is demonstrated that under intuitive assumptions and approximations, the complexity of the proposed multitarget estimation algorithm can be made linear in terms of the number of tracks and the number of observations, while naturally preserving distinct tracks for detected targets, unlike point-process-based methods*

LAMBERTI Roland, PETETIN Yohan, DESBOUVRIES François, SEPTIER François**Semi-independent resampling for particle filtering**. IEEE signal processing letters, january 2018, vol. 25, n° 1, pp. 130-134

URL: https://arxiv.org/pdf/1710.05407.pdf

abstract

*Among sequential Monte Carlo methods, sampling importance resampling (SIR) algorithms are based on importance sampling and on some (resampling-based) rejuvenation algorithm that aims at fighting against weight degeneracy. However, this mechanism tends to be insufficient when applied to informative or high-dimensional models. In this letter, we revisit the rejuvenation mechanism and propose a class of parameterized SIR-based solutions that enable us to adjust the tradeoff between computational cost and statistical performances*

LAMBERTI Roland, SEPTIER François, SALMAN Naveed, MIHAYLOVA Lyudmila**Gradient based Sequential Markov Chain Monte Carlo for multi-target tracking with correlated measurements**. IEEE transactions on signal and information processing over networks, september 2018, vol. 4, n° 3, pp. 510-518

URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8049338

abstract

*Measurements in Wireless Sensor Networks (WSNs) are often correlated both in space and in time. This paper focuses on tracking multiple targets in WSNs by taking into consideration these measurement correlations. A Sequential Markov Chain Monte Carlo (SMCMC) approach is proposed in which a Metropolis within Gibbs refinement step and a likelihood gradient proposal are introduced. This SMCMC filter is applied to case studies with cellular network Received Signal Strength (RSS) data in which the shadowing component correlations in space and time are estimated. The efficiency of the SMCMC approach compared to particle filtering, as well as the gradient proposal compared to a basic prior proposal, are demonstrated through numerical simulations. The accuracy improvement with the gradient-based SMCMC is above 90% when using a low number of particles. Thanks to its sequential nature, the proposed approach can be applied to various WSN applications, including traffic mobility monitoring and prediction*

LEHMANN Frederic, FRIGNAC Yann**Performance analysis of a new calibration method for fiber nonlinearity compensation**. IEEE communications letters, june 2018, vol. 22, n° 6, pp. 1176-1179

abstract

*Digital signal processing for fiber nonlinearity com- pensation is a key enabler for the ever-increasing demand for higher data rates in coherent optical transmissions. A major challenge of existing techniques is that the fiber nonlinear coefficient needs to be scaled properly during compensation in order to reach the achievable signal quality increase. We solve this problem using a low-complexity algorithm adaptively optimizing a metric based on the soft-decision bitwise demodulator used for modern FEC decoders. An analytical model shows that the proposed scheme converges to the optimal scaling factor with a predictable precision, that is validated by numerical results*

SCHLANGEN Isabel, DELANDE Emmanuel, HOUSSINEAU Jérémie, CLARK Daniel**A second-order PHD filter with mean and variance in target number**. IEEE transactions on signal processing, january 2018, vol. 66, n° 1, pp. 48-63

[PDF]

abstract

*The Probability Hypothesis Density (PHD) and Cardinalized PHD (CPHD) filters are popular solutions to the multi-target tracking problem due to their low complexity and ability to estimate the number and states of targets in cluttered environments. The PHD filter propagates the first-order moment (i.e. mean) of the number of targets while the CPHD propagates the cardinality distribution in the number of targets, albeit for a greater computational cost. Introducing the Panjer point process, this paper proposes a second-order PHD filter, propagating the second-order moment (i.e. variance) of the number of targets alongside its mean. The resulting algorithm is more versatile in the modelling choices than the PHD filter, and its computational cost is significantly lower compared to the CPHD filter. The paper compares the three filters in statistical simulations which demonstrate that the proposed filter reacts more quickly to changes in the number of targets, i.e., target births and target deaths, than the CPHD filter. In addition, a new statistic for multi-object filters is introduced in order to study the correlation between the estimated number of targets in different regions of the state space, and propose a quantitative analysis of the spooky effect for the three filters*

SHMALIY Yuriy, LEHMANN Frederic, ZHAO Shunyi, AHN Choon K.**Comparing robustness of the Kalman, H_inf, and UFIR filters**. IEEE transactions on signal processing, july 2018, vol. 66, n° 13, pp. 3447-3458

abstract

*This paper provides a comparative analysis for robustness of the Kalman filter (KF), H_infinity filter derived using the game theory, and unbiased finite impulse response (UFIR) filter, which ignores the noise statistics and initial values. A comparison is provided for Gaussian models by studying the effects of errors and disturbing factors on the bias correction gain. It is shown that the rule of thumb of optimal filtering in terms of accuracy, UFIR*

*
*

UNEY Murat, MULGREW Bernard, CLARK Daniel**Latent parameter estimation in fusion networks using separable likelihoods**. IEEE transactions on signal and information processing over networks, december 2018, vol. 4, n° 4, pp. 752-768

[PDF]

abstract

*Multi-sensor state space models underpin fusion applications in networks of sensors. Estimation of latent parameters in these models has the potential to provide highly desirable capabilities such as network self-calibration. Conventional solutions to the problem pose difficulties in scaling with the number of sensors due to the joint multi-sensor filtering involved when evaluating the parameter likelihood. In this article, we propose a separable pseudo-likelihood which is a more accurate approximation compared to a previously proposed alternative under typical operating conditions. In addition, we consider using separable likelihoods in the presence of many objects and ambiguity in associating measurements with objects that originated them. To this end, we use a state space model with a hypothesis based parameterisation, and, develop an empirical Bayesian perspective in order to evaluate separable likelihoods on this model using local filtering. Bayesian inference with this likelihood is carried out using belief propagation on the associated pairwise Markov random field. We specify a particle algorithm for latent parameter estimation in a linear Gaussian state space model and demonstrate its efficacy for network self-calibration using measurements from non-cooperative targets in comparison with alternatives*

ZHENG Fei, DERRODE Stéphane, PIECZYNSKI Wojciech**Parameter estimation in switching Markov systems and unsupervised smoothing**. IEEE transactions on automatic control, 2018, pp. 1-6 (document in press - published online 06 Aug 2018)

abstract

*Stationary Jump Markov Linear Systems (JMLSs) model linear systems whose parameters evolve with time according to a hidden finite state Markov chain. We propose an algorithm for parameter estimation of a recent class of JMLSs called Conditionally Gaussian Pairwise Markov Switching Models (CGPMSMs). Our algorithm, named Double-EM (DEM), is based on the Expectation-Maximization (EM) principle applied twice sequentially. The first EM is applied to the couple (switches, observations) temporarily assumed to be a Pairwise Markov Chain (PMC). The second one is used to estimate the remaining conditional transitions and conditional noise matrices of the CGPMSM. The efficiency of the proposed algorithm is studied via unsupervised smoothing on simulated data. In particular, smoothing results, produced with CGPMSM in an unsupervised manner using DEM, can be more efficient than the ones obtained with the nearest classic "Conditionally Gaussian Linear State-Space Model" (CGLSSM) based on true parameters and true switches*

### Communication dans une conférence à comité de lecture (5)

AUSSEL Nicolas, PETETIN Yohan, CHABRIDON Sophie**Improving performances of log mining for anomaly prediction through NLP-based log parsing**. MASCOTS 2018: 26th International Symposium on the Modeling, Analysis, and Simulation of Computer and Telecommunication Systems, Los Alamitos : IEEE Computer Society, 25-28 september 2018, Milwaukee, United States, 2018, pp. 237-243, ISBN 978-1-5386-6886-3

abstract

*Failure prediction of industrial systems is a promising application domain for data mining approaches and should naturally rely on log messages which are a prime source of data as they are generated by many systems. However, before extracting relevant information of such log messages, another critical step is to parse the logs, that is to say to transform a raw unstructured text from the log messages into a suitable input for data mining. These two problems (log parsing then log mining) are often studied separately while they are directly related in the context of failure prediction ; moreover, few performance benchmarks are publicly available. In this paper, we focus on the impact of log parsing techniques via natural language processing on the performances of log mining on two datasets. The first one is a log of an industrial aeronautical system comprising over 4, 500, 000 messages collected over one year of operation ; the second one is a public benchmark set from an HDFS cluster. On the latter, we show that it is possible to raise the F-score from 96% to 99.2% while using simpler and more robust log parsing techniques that require less parameter tuning provided that they are correctly combined with log mining techniques*

CAMPBELL Mark, CLARK Daniel**Time-lapse estimation for optical telescope sequences**. URSI-France 2018 : Geolocation and navigation in space and time, URSI, 28-29 march 2018, Meudon, France, 2018, pp. 1-8

URL: https://ursi-france.sciencesconf.org/187227/document

[PDF]

abstract

*Typical tracking scenarios rely on the assumption that there is a constant time lapse between observations. In real life applications, this assumption is often untrue. In Space Situational Awareness (SSA) applications accurate target estimation is of importance to obtain orbital information. This paper presents recent developments in multi target detection and tracking techniques, exploiting the Single Cluster Probability Hypothesis Density (SC-PHD) filter, in order to jointly estimate the dynamic objects and the time lapse between images*

Les scénarios de suivi typiques reposent sur l'hypothèse qu'il y a un laps de temps constant entre les observations. Dans les applications réelles, cette hypothèse est souvent fausse. Dans les applications SSA, l'estimation précise de la cible est importante pour obtenir des informations orbitales. Cet article présente les développements récents dans les techniques de détection et de suivi de cibles multiples, en exploitant le filtre SC-PHD, an d'estimer conjointement les objets dynamiques et le laps de temps entre les images

Les scénarios de suivi typiques reposent sur l'hypothèse qu'il y a un laps de temps constant entre les observations. Dans les applications réelles, cette hypothèse est souvent fausse. Dans les applications SSA, l'estimation précise de la cible est importante pour obtenir des informations orbitales. Cet article présente les développements récents dans les techniques de détection et de suivi de cibles multiples, en exploitant le filtre SC-PHD, an d'estimer conjointement les objets dynamiques et le laps de temps entre les images

CHEVALIER Pascal, DELMAS Jean-Pierre, SADOK Mustapha**Performance of a third-order Volterra MVDR beamformer in the presence of non-Gaussian and/or non-circular interference**. EUSIPCO 2018: 26th European Signal Processing Conference , IEEE Computer Society, 03-07 september 2018, Rome, Italy, 2018 (document in press)

URL: http://www-public.imtbs-tsp.eu/~delmas/

[PDF]

abstract

*Linear beamformers are optimal, in a mean square (MS) sense, when the signal of interest (SOI) and observations are jointly Gaussian and circular. When the SOI and observations are zero-mean, jointly Gaussian and non-circular, optimal beamformers become widely linear (WL). They become non-linear with a structure depending on the unknown joint probability distribution of the SOI and observations when the latter are jointly non-Gaussian, assumption which is very common in radiocommunications. In this context, a third-order Volterra minimum variance distortionless response (MVDR) beamformer has been introduced recently for the reception of a SOI, whose waveform is unknown, but whose steering vector is known, corrupted by non-Gaussian and potentially non-circular interference, omnipresent in practical situations. However its statistical performance has not yet been analyzed. The aim of this paper is twofold. We first introduce an equivalent generalized sidelobe canceller (GSC) structure of this beamformer and then, we present an analytical performance analysis of the latter in the presence of one interference. This allows us to quantify the improvement of the performance with respect to the linear and WL MVDR beamformers*

CLARK Daniel, DE MELO Flavio**A linear-complexity second-order multi-object filter via factorial cumulants**. FUSION 2018: 21st international conference on Information Fusion, Los Alamitos : IEEE Computer Society, 10-13 july 2018, Cambridge, United Kingdom, 2018, pp. 1250-1259, ISBN 978-0-9964527-6-2

[PDF]

abstract

*Multi-target tracking solutions with low computational complexity are required in order to address large-scale tracking problems. Solutions based on statistics determined from point processes, such as the PHD filter, CPHD filter, and newer second-order PHD filter are some examples of these algorithms. There are few solutions of linear complexity in the number of targets and number of measurements, with the PHD filter being one exception. However, the trade-off is that it is unable to propagate beyond first-order moment statistics. In this paper, a new filter is proposed with the same complexity as the PHD filter that also propagates second-order information via the second-order factorial cumulant. The results show that the algorithm is more robust than the PHD filter in challenging clutter environments*

LAMBERTI Roland, PETETIN Yohan, SEPTIER François, DESBOUVRIES François**A double proposal normalized importance sampling estimator**. SSP 2018: IEEE Statistical Signal Processing Workshop, Los Alamitos : IEEE Computer Society, 10-13 june 2018, Freiburg, Germany, 2018, pp. 238-242, ISBN 978-1-5386-1571-3

abstract

*Monte Carlo methods rely on random sampling to compute and approximate expectations of interest in signal processing. Among Monte Carlo methods for integration, Importance Sampling is a variance reduction technique which consists in sampling from an importance distribution which is not necessary the original target distribution. The performance of the resulting estimate is strongly related to the critical choice of such an important distribution. In this paper we revisit the rationale of the normalized importance sampling technique and show that it is possible to improve the classical importance sampling estimate by approximating the expectation of interest via two importance distributions. The choice of these two importance distributions is optimized w.r.t. the variance of the final estimate. Our results are validated via numerical simulations*

Modifié le 30 juin 2010