WP2

Applied methodologies

This WP will investigate the design of efficient signal processing algorithms and coding techniques for multipoint-to-multipoint communication and sensor networks, based on the theoretical guidelines obtained in WP1. Special emphasis will be placed on distributed algorithms, optimizing processes such as channel estimation, synchronization, MIMO coding, cognitive sensing and radio transmission, routing, source coding, etc. Schemes will be evaluated from the viewpoints of robustness, scalability and adaptability, with practical complexity constraints imposed. In the context of ad-hoc self-organized networks, the output of this WP will substantially increase the understanding of their performance in terms of robustness, throughput, adaptability, power efficiency, latency, and interactions between source and channel statistics.

There has been active research in distributed wireless sensor networks since the early 1990's, guided by the trend to move from centralized platforms to dense networks of possibly unreliable and heterogeneous low-cost nodes (usually termed motes), equipped with sensing, processing and wireless communication capabilities, and with the potential to perform complex tasks co-operatively. However, most of the algorithms that have been developed to date are neither fully decentralized nor fully scalable and, moreover, they rely on a number of unrealistic assumptions. We will investigate the design, under realistic complexity and communication constraints, of novel decentralized algorithms that co-operatively perform tasks (detection, statistical inference with consensus, estimation, distributed storage, control, localization, etc) that are building blocks in the provision of services in areas such as environmental monitoring, object localization and tracking, etc. We will seek to understand the interaction between the different distributed algorithms that are necessary to extract useful information from the environment, and the various issues that hamper this task: a) existing restrictions associated to nodes in terms of power, network connectivity, processing capabilities and mote/link failures, b) influence and potential exploitation of mobility, c) heterogeneity in both processing and communication capabilities, as well as in sensing, and d) influence of packet losses in applications with severe delay constraints.

This workpackage is led by UVEG and UC3M, and involes the following partners: UPC, UPM, UC, CEIT, UPF, US, UVIGO and UDC. The activities contained in this workpackage are described in more detail below.

A2.1 - Efficient Communication Schemes for the Network of the Future

In their ongoing evolution, wireless networks are incorporating features such as cooperation, opportunism and cognitive sensing in order to accommodate the ever-growing levels of mobility, security, capacity and monitoring requirements that emerging applications demand. These new functionalities will affect the design of communication schemes at multiple levels and will provide means to close in on theoretical performance bounds. Practical considerations, however, impose constraints on system resources such as computational complexity, power, data rate, latency and bandwidth. Thus, it becomes necessary to develop robust and efficient schemes by following a multidisciplinary approach in which networking, communications and signal processing closely interact. To this end, several key enabling tools have been identified, namely the use of side information, advanced signal processing and channel coding techniques for cooperative communication, joint optimization of multiple layers of the protocol stack, the use of multiple-antenna transceivers, and dynamic spectrum access. The objectives of this activity are:

 Exploiting side information for efficient transmission. Side information may refer to interference due to other users/systems, CSI, or even data correlation with the primary information source. Whenever available, information-theoretic results suggest that the use of side information substantially increases capacity. We will address the design of practical schemes exploiting side information, either at the transmitter (based on the frameworks of 'Dirty Paper Coding' for multi-terminal communication, watermarking for secure information exchange, and cooperative communication) or at the receiver (based on the framework of Wyner-Ziv source coding as well as on joint source-channel coding, which is of particular relevance to sensor networks due to the correlations within the observed data).

 Signal Processing for cooperative communication. Reaping the benefits in system capacity and diversity promised by information theory requires sophisticated signal processing algorithms for many transceiver tasks such as channel estimation, digital compensation of transmission distortions, interference cancellation, and adaptive modulation schemes. One of our research lines will focus on the application of concepts from such diverse disciplines as machine learning, pattern recognition, and multivariate statistics, collectively known as Cognitive Information Processing (CIP), to the design of these signal processing algorithms. Particular interest will be devoted to synchronization, a critical issue in cooperative communication.

 Channel coding for cooperative communication. In the last decade, considerable effort has been devoted to devising practical channel encoders and decoders based on the promising concepts of Turbo and Low Density Parity Check (LDPC) codes. We propose to investigate the design of channel codes for multi-terminal scenarios, with special emphasis on serially-concatenated low-density generator matrix codes with irregular degree profile, low-rate codes, rate-compatible codes, and multilevel codes. The interplay between popular space-time signal encoding strategies and practical channel coding will also be analyzed in order to attain performance close to capacity in realistic scenarios.

 Routing, network coding and cross-layer schemes. Traditionally, routing has been the approach of choice to manage network information flows. The highly dynamic nature of future wireless networks will make decentralized routing a necessity, especially with mobile nodes. Mobility, however, may be exploited in order to improve energy efficiency and throughput by applying spatial diffusion ideas to the design of routing algorithms. Even in static networks, the idea of projecting communication constraints and parameters across different layers of the protocol stack (cross-layer designs) is generally seen as a useful means to ensure energy efficiency, suitable medium access procedures and, in the case of sensor networks, robust detection capabilities. As an alternative to routing, network coding has revealed itself as a promising technique to improve network throughput. In this sense, and capitalizing on the theoretical results developed in Activity 1.2, it is necessary in practice to investigate scenarios in which network coding is applied and a utility function (e.g., network energy consumption) is to be optimized. This activity will also explore the possibility of combining network coding with other layers in a cross-layer design.

 Multiple Antenna Transceivers. Performance of both infrastructure-based and ad-hoc wireless networks can be noticeably improved if terminal nodes are provided with the MIMO ability of transmitting and receiving employing multiple antennas. It is therefore important to investigate novel signal processing and coding techniques to push the state of the art closer to the theoretical capacity limits of wireless networks with MIMO terminals. In addition, the important role that cooperation is expected to play in future wireless networks motivates the study of virtual (or distributed) MIMO systems, in which the antennas are not co-located.

 Cognitive radios. Many spectrum measurements have evidenced great temporal and geographical variability in the usage of licensed spectral bands. In conjunction with fixed spectrum assignment policies and the limited amount of available radio spectrum, this results in low efficiency in the use of this scarce resource. Cognitive radios provided with Dynamic Spectrum Access (DSA) capabilities will improve this situation by granting unlicensed 'secondary' networks access to licensed bands in an opportunistic manner, as long as interference with licensed 'primary' networks is kept at tolerable levels. Two different scenarios will be considered: ad-hoc scenarios, where the cognitive use and the spectrum sensing take place at the same terminal; and wireless-sensor-aided cognitive radio, where there is a separate wireless sensor network deployed to perform the sensing. In both cases, there is a need to develop novel sensing algorithms capable of detecting primary users in the very low-power regime. Decentralized and co-operative detection will be also studied as a means to deal with shadowing and varying path-loss conditions. Finally, we will also consider the design of actuation strategies that avoid the interference caused by the cognitive user to the primary links.

go top

A2.2 - Distributed Source Coding for Ambient Sensing

Wireless sensor networks are becoming a new scientific instrument in numerous applications because they offer the possibility of sampling physical processes at a level of detail that was unaffordable to date. This will allow verifying new theories, developing better models, posing and addressing new scientific questions, and fostering multidisciplinary research of great interest.

The deployment of wireless sensor networks gives rise to several intriguing issues:

a) A possibly time-varying spatio-temporal sampling of a physical source process, which in turn generates a certain spatial correlation across different sensors. This offers opportunities that should be exploited.

b) The interplay between the physics of the source and the physics governing the inter-node communication channels, which should be analyzed in the various code designs.

c) Because of the strong constraints in terms of resources such as power, memory and computation capabilities, the motes in a real wireless sensor network are subject to failures. This motivates the use of appropriate distributed storage mechanisms that allow for self-recovery.

 Collaborative data acquisition and distributed sampling in wireless sensor networks. Within this subtask, we plan to investigate three topics: a) new sampling, quantization and reconstruction methods to deal with larger classes of non-bandlimited physical fields using the underlying physics-based knowledge governing the spatio-temporal behavior of these fields, b) the interplay between sampling theory and the various network mechanisms that might cause irregular and time-varying sampling in wireless sensor networks, such as loss of connectivity and node mobility, and c) novel compressive sensing techniques, where the sampling and data compression are performed jointly via a sparse spatial representation of the process being sampled thereby reducing drastically the number of samples in time and space (i.e., sensors) that are necessary for a given approximation quality.

 Decentralized joint source-channel coding for wireless sensor networks: the role of measure-matching. Since in a wireless sensor network the communication takes place in a multi-point to multi-point manner and the source model plays an important role, Shannon's source-channel coding separation theorem does not apply in general. It is hence imperative to understand the best way of designing novel joint source-channel coding algorithms. We will consider the implication of uncoded transmission and other joint source-channel (low-complexity) coding systems on the sensing of quantities governed by physical laws. First, we will investigate how certain physical signal models fit a coded or uncoded paradigm. Second, trade-offs between sensing and communication will be explored, using both a traditional coded approach with separation and a measure-matching approach.

 Distributed and robust storage for wireless sensor networks based on dynamic erasure channel codes. Recent work has explored the benefits of using network coding in distributed storage applications of sensor networks. Combinations with the so-called distributed erasure and fountain codes have been discussed. In general though, the issue of how to combine network coding with source coding, and in particular sparse coding, has not been addressed. This activity will focus on the problem of distributed storage in sensor networks based on fountain-like codes, with k data nodes generating spatial correlated data by sensing the environment and with n (n > k) storage nodes having limited memory. We propose theoretical approaches, where the parity equations underlying the code are generated randomly in order to opportunistically match the instantaneous network state. Another crucial problem is how to perform joint detection in a dynamic environment, where the number of devices transmitting simultaneously is random and unknown to each receiver. Work on dynamic multiuser joint detection has merely started and we will aim at obtaining results also in this area.

go top

A2.3 - Distributed inference under communication constraints

The convergence of communication and sensor networks will pose major challenges that must be addressed in order to create new detection, estimation, prediction, tracking and control methods that work efficiently under practical communication constraints. Most of the technical problems that have to be tackled in the medium term will be related to one or more of the following key issues.

a) The heterogeneity of future networks, which will integrate a very broad range of functions, devices and algorithms. Many different types of sensor nodes will have to cooperate in the processing of data subject to practical communication impairments.

b) Dynamic network topologies will become increasingly common. The combination of mobile and fixed nodes generates difficulties because of the (possibly frequent) setup and suppression of network connections.

c) Decentralized methods are urgently needed. While current inference techniques available for sensor networks are essentially centralized, the technological focus is progressively shifting to highly distributed setups where the combination of local processing and cooperation will be preeminent.

d) The complexity of the inference problems to be addressed will grow rapidly with new applications of sensor networks. Highly complex probabilistic models will emerge for to adequately represent the physical phenomena of interest along with the features (connectivity, reliability, etc) of the network nodes. Such issues set up a rich scenario where existing algorithmic tools are clearly insufficient. Our research will involve the construction of formal models that capture the key features of wireless sensor networks, such as time-varying network topologies, link capacities, transmission reliability, data losses, etc. In turn, efficient distributed algorithms matched to these complex models will have to be devised and analyzed. We currently identify the following topics to be investigated.

 Optimal distributed detection. The construction and analysis of distributed detection tests when statistical models for the performance of local detection (at the network nodes) are unavailable is an important yet still open problem. A thorough understanding of how network communication parameters (such as network size, density of nodes, latencies, reliability of links, availability of CSI at the sensors, etc) is needed for the design of efficient schemes.

 Cooperative statistical methods. Cooperative methods for estimation, tracking and prediction will be pursued, both because of the trend to network decentralization and because they are much more robust to practical communication impairments. Bayesian learning provides flexible and powerful tools that can be intertwined with network and communication constraints and implemented using advanced methods from the field of computational statistics. Geo-location is a particularly relevant application of such techniques. Since global navigation satellite systems fail to work in indoor and multipath radio propagation environments, accurate and robust positioning, navigation and tracking methods that exploit heterogeneous measurements from sensor networks are needed.

 Gossip and consensus algorithms. In wireless sensor networks, local decisions usually have to be made by nodes having access only to partial and imprecise measurements. Moreover, due to various realistic communication impairments, the resulting topology will be highly dynamic. In this scenario, gossip algorithms are executed across the different nodes, following an order that depends on the time-varying connectivity graph of the network and on propagating local decisions. We will study both performance limits and algorithmic design for these methods, analyzing their dependence on the channel physics and network connectivity. The goal of gossip algorithms is to achieve consensus with respect to some function of the sensor data. In this case, we will investigate conditions that ensure convergence to the consensus, as well as the rate of this convergence.

go top