The reason to present this proposal is the necessity of efficient distributed communication and information extraction methods to exploit the potential of current and future distributed sensing and communication systems. The combination of di erent and complementary theoretical disciplines is mandatory to fulfill the objectives of this project, and this methodology is applied at different and again complementary levels of the distributed system.

At the lowest level of information extraction, machine learning, information, and optimization theories will be used to obtain compressed sensing methods to reduce the information to be transmitted for each node. Then, the combination of local classifiers and estimators based on learning, and probabilistic parametric methods for global classification and estimation, is an efficient methodology to work with uncertain and correlated measurements and we will combine disciplines such as machine learning, graph theory, multitask learning, information theory, large deviation theory, bootstrap or random matrices to develop distributed detection and estimation methods, suitable to work under practical communication constraints, and to analyze their performance. State-space based methodologies are appropriate to deal with dynamic systems, and Bayesian and secuential Monte Carlo theoretical frameworks will be applied to global dynamic optimization and distributed inference in complex dynamic systems at the highest level.

With respect to data transmission, machine learning methods (support vector machines and Gaussian processes), approximate inference algorithms (extensions of the belief propagation), coding theory, and alternative loss functions ( -divergence), will be combined to obtain: i) nonlinear channel equalizers and improved decoding algorithms; ii) joint source and channel coding with the goal of achieving maximum capacity; iii) optimal and suboptimal detection and coding in cooperative mesh netwoks. Moreover, we propose to bridge the gap between the optimal precoder in MIMO systems (the current result is non-constructive and its complexity is exponential in the number of bits) and an implementable approach that can also be used in the blind estimation of the channel at the receiver, a critical issue in mobile scenarios.

Background and previous results

The initial hypothesis of this proposal is supported by numerous previous results. The research topics to be addressed are of current interest and present a high research activity in the scientific community. A bibliographic review of the recent technical literature reveals this interest and the appropriateness of the proposed theoretical tools and methodology. Section 2 summarizes some of the most relevant previous scientific results supporting the hypothesis of the proposal.

The background of the two applicant groups in the framework of the proposal is evidenced by a large number of previous results at di erent levels. In the following, we will cite the most relevant ones:

• Recent publication of results in international journals and conference proceedings

  Members of the research team included in this proposal have authored numerous papers published in international journals or in the proceedings of international conferences where the corresponding works were presented. These publications are included in the authors’ CVs (more than 170 publications in the last 5 years in the field considered in this proposal).

• Participation in competitive calls for related research projects

  The two groups composing the research team have received public funding in competitive national and regional research programs for several research projects with topics related with the ones considered in this proposal. The most recent examples are projects TEC2006-13514-C02-01/TCM MONIN, S-0505/TIC/0223 PRO-MULTIDIS, and CSD-2008-00010 COMONSENS. The
  rst one can be considered the main predecent of this proposal, which continues the work carried jointly by the two groups in that project. The second one, granted by the Comunidad Autónoma de Madrid, is currently in process and supports the realization of activities among groups in the field of application of the present proposal The last one, funded by the program \CONSOLIDER-INGENIO 2010″, call 2008, integrates the two groups applicants of this proposal into a greater consortium with interdisciplinary expertise and the goal of analyzing the fundamental limits of distributed wireless communication systems. The Signal Processing Group of Universidad Carlos III de Madrid participated in the network of excellence CRUISE (CReating Ubiquitous Intelligent Sensing Environments, IST4027738), funded by the European Comission within the 6th FP, and aimed to focus and coordinate the research on communications and applications aspects of wireless sensor networks in Europe. The group also participates in projects oriented to the transfer of technology to the industry, like project CEN-2002-2007 \TIMI”, a project of the \Consorcios Estratégicos Nacionales en Investigación Técnica” (CENIT) (\Strategic National Consortiums in Technical Research”) program of the Spanish \Centro para el Desarrollo Tecnológico e Industrial” (CDTI, Center for the Technological and Industrial Development) coordinated by Atos Origin, where the group is responsible of technical activities in the field of distributed localization and tracking for indoor navigation.

Objetives

1. To develop distributed compressed sensing methods for e ciently gathering the sensed information, specially in the case of correlated sources.
2. To obtain distributed methods for detection, estimation, and classi
   cation, suitable to work under realistic communication constraints in distributed scenarios where measures are uncertain and correlated, by combining classifiers using parametric and non parametric techniques.
3. To obtain global dynamic optimization methods, and distributed inference methods in complex dynamic systems, in the context of Bayesian and sequential Monte Carlo theoretical frameworks.
4. To develop new improved precoding, non linear equalization, and decoding methods to improve the capacity of communication systems making use of approximate inference algorithms, graph theory, and alternative loss functions to minimize the e ect of loops in the graphs.
5. To obtain novel coding algorithms for application in joint source channel coding and coding in cooperative mesh networks based on the application of LDPC and IRA codes.
6. To evaluate the performance and applicability of the methods developed in the project, by implementing and developing a multipurpose platform for demonstration, within four well defined and real practical scenarios: indoor navigation, domestic monitoring, security in large-scale indoor facilities and communications.