Gernot Fabeck

Chernoff information-based optimization of sensor networks for distributed detection

This paper addresses the scalable optimization of sensor networks for distributed detection applications. In the general case, the jointly optimum solution for the local sensor decision rules and the fusion rule is extremely difficult to obtain and does not scale with the number of sensors. In this paper, we consider optimization of distributed detection systems based on a local metric for sensor detection performance. Derived from the asymptotic error exponents in binary hypothesis testing, the Chernoff information emerges as an appropriate metric for sensor detection quality. By locally maximizing the Chernoff information at each sensor and thus decoupling the optimization problem, scalable solutions are obtained which are also robust with respect to the underlying prior probabilities. By considering the problem of detecting a deterministic signal in the presence of Gaussian noise, a detailed numerical study illustrates the feasibilty of the proposed approach.

Optimization of cooperative spectrum sensing and implementation on software defined radios

Reliable detection of primary user activity by spectrum sensing is a crucial issue of cognitive radio systems. The objective of cooperative spectrum sensing is to combine the detection results of multiple cognitive radios in order to maximize the probability of detecting unused spectrum while meeting a required reliability of detecting primary user activity. In this paper, a Kullback-Leibler distance-based optimization approach for the local decision thresholds of cooperative spectrum sensing is proposed. It is both computationally efficient and scalable with the number of cognitive radios. To validate the concept, real spectrum sensing results are used. The employed practical setup is based on software defined radio and detects a WiMAX-like OFDM signal. The presented numerical results illustrate the feasibility and effectiveness of the approach.

Power-aware distributed detection in IR-UWB sensor networks

The interplay between signal processing and wireless networking plays a crucial role in sensor networks deployed for detection and estimation applications. In this paper, an opportunistic power assignment strategy for IR-UWB sensor networks is presented which is designed to optimize detection performance in terms of the global probability of error. The opportunistic power assignment strategy utilizes both the detection error probabilities of individual sensors as well as network topology information, leading to significant performance gains compared to uniform power assignment.

Tight Performance Bounds for Distributed Detection

Evaluating the performance measures of distributed detection in sensor networks is important for design procedures aiming at optimal configurations. Direct computation of the global error probabilities is a difficult problem and feasible only in some special cases. In this paper, we present an approach that provides closed-form upper bounds on the detection error probabilities in the parallel fusion network which are both computationally simple and numerically tight. The bounds are derived by combining a probability inequality formulated by Hoeffding with a multiplicative form factor which is due to Talagrand. We demonstrate that the bounds apply to sensor networks of varying size, an arbitrary number of local quantization levels, and non-identical sensors.

Performance evaluation of sensor fusion with side information

The efficient evaluation of fusion algorithms becomes particularly important when different fusion schemes have to be compared with respect to an underlying performance metric. In this paper, we present explicit expressions for the global error probabilities of sensor fusion with side information for distributed detection applications. In the considered distributed detection problem, the sensors compress their observations independently and transmit local decisions to a fusion center that combines the received decisions with respect to available side information and computes the final detection result. In the special case of identical sensors, computationally efficient expressions are obtained by using the multinomial distribution. Numerical results reveal the influence of different qualities of side information on the overall detection performance.

Optimization of Linear Wireless Sensor Networks for Serial Distributed Detection Applications

Typical applications of wireless sensor networks include infrastructure monitoring and surveillance, where in many cases the geometry of the monitored object determines the topology of the deployed network. For example, important applications like pipeline monitoring and border surveillance feature a linear arrangement of wireless sensors. In this paper, we address the scalable optimization of linear sensor networks for serial distributed detection applications. In serial distributed detection, signal detection is performed collaboratively by multiple sensors arranged in serial until a final detection result is reached. By locally maximizing the Chernoff information at each sensor in the serial network, scalable solutions are obtained which only rely on local information. By considering the problem of detecting a deterministic signal in the presence of Gaussian noise, a detailed numerical study reveals interesting trade-offs and dependencies between communication constraints and detection performance.

Power-Aware Sensor Selection for Distributed Detection in Wireless Sensor Networks

The limited energy of sensor nodes in wireless sensor networks strongly recommends power-aware design methodolo- gies. In this paper, a power-aware sensor selection strategy for wireless sensor networks is presented that is especially designed for distributed detection with soft decision fusion. The objective is to minimize the global probability of error at the fusion center under a total network power constraint. The cross-layer approach for the selection of a proper subset of sensors is based on a measure of individual sensor detection quality as well as location information. It corresponds to a low-complexity power allocation algorithm and enables significant performance gains in terms of reduction of the global probability of error compared to the inclusion of all sensors.

Power Allocation and Node Clustering for Distributed Detection in IR-UWB Sensor Networks

In wireless sensor networks, the limited energy of the nodes should be utilized in such a way that the performance measure of the sensing application is optimized. In this paper, power-aware design of IR-UWB sensor networks for distributed signal detection is discussed. The design approach consists of two parts which aim at minimizing the global probability of detection error. First, an application-specific node clustering algorithm is performed. Based on the generated topology a resource allocation scheme adapted to distributed detection is carried out. It is based on both, information from the network topology and individual sensor detection quality. Numerical results indicate significant performance gains for sensor networks with realistic resource constraints. I. INTRODUCTION The initial task of many applications of wireless sensor networks is the detection of targets in a region of interest. In distributed detection, the sensor nodes process their observa- tions locally and make preliminary decisions about the state of the monitored environment. The local decisions are transmitted to a fusion center where they are combined to obtain a final detection result with high reliability. In practice, the detection performance of wireless sensor networks is influenced by resource constraints like limited available energy or a restricted maximum transmission range. Hence, resource allocation and networking algorithms should be adapted to the detection application (1) in order to optimally design the sensor system. In the parallel fusion network, where all sensor nodes transmit their local decisions directly to a fusion center, the maximum area that can be covered is limited by the maximum transmission range of a node. The covered area can be extended by a clustering of the network into a tree structure resulting in hierarchical transmission of node decisions. This requires algorithms that perform the clustering. In the litera- ture, several algorithms with different optimization objectives and different complexity have been suggested. The authors of (2) consider a TDMA scheme. In (3) two algorithms are presented, which are combined with an impulse radio ultra- wideband (IR-UWB) specific multiple access scheme with non-orthogonal channels. Usually, clustering algorithms are not adapted to specific applications. The asymptotic detection behavior of a tree network for distributed detection has been analyzed in (4). However, for realistic numbers of sensor nodes, these asymptotic results provide only limited informa- tion. In this paper, we present an algorithm which considers the interdependency between energy consumption and the over- all detection performance by including the individual sensor detection performance in the process of cluster head elec- tion and cluster formation. Based on the generated topology, we furthermore suggest an application-specific assignment of transmission power levels that depends both on individual sensor qualities as well as the generated topology. It aims at minimizing the global probability of detection error given a budget of transmission power. As enabling technology for wireless sensor networks, we consider IR-UWB transceivers. Due to the possibility of power control, IR-UWB transceivers are well suited to adapt networking algorithms to specific applications. Compared to our preceding work (5), where the reduction of transmission energy for a given detection performance is analyzed, we ask the complementary ques- tion of how much the probability of detection error can be decreased given a fixed total power budget. Furthermore, we conduct a direct comparison of the detection performance of the parallel and the tree network with and without limitations of the transmission range, which reveals which topology is advantageous in which parameter range. Moreover, a trade- off between the power budget and the number of nodes is discussed.

Cross-layer design of cluster formation and power allocation in IR-UWB sensor networks

Wireless sensor networks are usually deployed for specific purposes. A cross-layer design of networking algorithms adapted to the considered application can significantly improve overall performance metrics. In this paper, distributed signal detection in IR-UWB sensor networks with severe resource constraints is considered. The presented algorithms for power assignment aim at minimizing the energy necessary to meet an overall detection performance by exploiting dependencies between application and physical layer. To account for a limited transmission range of the sensor nodes, the approach is furthermore combined with an application-specific formation of node clusters which allows for a hierarchical transmission of the sensor decisions to a fusion center. Numerical results validate that this cross-layer approach leads to substantial energy savings compared to uniform power assignment.

Cross-Layer design of IR-UWB sensor networks for distributed detection applications

Cross-layer design for wireless networks aims at optimizing system-wide performance measures by exploiting dependencies between different network layers. In this paper, an opportunistic power assignment algorithms for IR-UWB sensor networks is presented that is especially designed for distributed signal detection under resource constraints. Specifically, the objective is to minimize the global probability of error of distributed detection systems, given a fixed level of total transmission power. The cross-layer approach for the allocation of transmission power is based on individual sensor detection quality as well as location information. It leads to significant performance gains compared to uniform power assignment for both the parallel and the serial sensor network topology.