Benedito J. Fonseca

A Distributed Procedure for Carrier Sensing Threshold Adaptation in CSMA-Based Mobile Ad Hoc Networks

This paper discusses the problem of determining the carrier sense threshold (CSth) of nodes in a CSMA-based Mobile Ad Hoc Network (MANET) and proposes a practical, fully distributed, CSth adaptation procedure. The procedure enables a node to select its CSth in order to improve the data throughput of the system by maximizing the number of successful transmissions in its neighborhood. Simulation results show that the procedure is able to improve the data throughput in more than 85% of random topologies. The average throughput improvement over all topologies were 11% and 18% in two different scenarios, with some topologies showing up to 75% data throughput gains.

Least Favorable Distributions for the Design of Randomly Deployed Sensor Detection Systems

The design of a detection system in which sensors are randomly located to detect a low-power signal emitter in a random location tends to be difficult because the measurements are conditionally dependent in general and the alternative hypothesis is composite. It is shown that there are conditions that allow a system designer to deal with these problems by assuming a least favorable distribution for the emitter location that not only makes the alternative hypothesis simple and ensures a detection performance, but also causes the measurements to become conditionally i.i.d., making models more amenable for analysis. Since a design based on a least favorable distribution may be considered too conservative, this paper proposes the use of a most favorable distribution for the emitter location and uses the theory of asymptotic relative efficiency (ARE) to evaluate how conservative the design based on a least favorable distribution is. It is further shown that, if the system designer can place sensors in an enlarged deployment region, then there are situations in which the design based on a least favorable distribution becomes less and less conservative as the dimensions of the region of interest increase.

Analysis of randomly deployed sensor detection systems under least favorable distributions

Consider the design of a system in which multiple sensors are randomly deployed in a circular region to detect the presence of a signal emitter in a random location. If the distribution of the emitter location is unknown, the design is difficult because the detection problem involves a composite hypothesis and the sensor measurements may be conditionally dependent. In this paper, we show that using the least favorable distribution for the emitter location not only is a robust design approach that solves the composite hypothesis issue, but also helps in dealing with the conditional dependency issue. We show that there are conditions under which the least favorable distribution for the emitter location causes the sensor measurements to become conditionally i.i.d. when using either the maximin or other practical fusion functions. Motivated by the form of the least favorable distribution, we also explore an alternative random sensor deployment strategy.

Improving the scan statistic to design sensor detection systems

When designing a sensor system to detect a target emitter in a large region of interest, it is desirable to use a fusion rule that avoids fusing strong measurements from sensors close to the emitter with weak measurements from sensors far from the emitter. To satisfy this criterion, the system designer may use a fusion rule based on the scan statistic. While this rule offers improved detection performance over other rules, this paper shows that its performance suffers when the scanning process encounters a varying number of sensors. In order to mitigate this degradation, it is proposed here that the fusion rule be based on a normalized scan statistic.

Operation Cost as a Performance Metric of Wireless Sensor Networks

This paper proposes a new performance metric for the evaluation of wireless sensor networks that reflects the cost of operation per unit time. This metric includes traditional metrics and considers the distribution of sensor energy consumption, which was not considered before. Focusing on a single-hop wireless sensor network, it is shown that it is not enough to consider just the expected energy consumed per measurement, unless sensors are equipped with unlimited energy resources. The paper illustrates the use of the metric in large wireless sensor networks for detection applications, relating it to a previously proposed metric and providing further insight in the selection of modulation techniques.

How conservative is a sensor detection system with the scan statistic designed using the union bound

In a sensor system to detect an emitter in a large region of interest, only a subset of sensors collects strong measurements, motivating fusion rules based on the scan statistic. Determining the false alarm probability for the scan statistic is however difficult; and using simple bounds such as the union bound would presumably result in too conservative designs. This paper argues that, at typical configurations, the design obtained with the union bound is not as conservative as expected. In fact, for a large class of distributions, the resulting design becomes less and less conservative as the system size grows.

Designing Conservative Sensor Detection Systems With Emitter Location Uncertainty

The design of sensor systems to detect an emitter at a random location with an unknown distribution is difficult because measurements are conditionally dependent and the hypothesis test is composite. This paper shows that, when sensors are at deterministic locations, these problems can be circumvented and a conservative design is achieved by adopting a least favorable distribution for the emitter location. An algorithm to achieve the design and the application to generalized likelihood ratio test detectors are presented.

Enlarged Deployment Regions to Circumvent the Conditional Dependence and Composite Hypothesis Problems in Sensor Detection Systems

It is usually difficult to design randomly deployed sensor systems to detect a signal emitter in a region of interest because measurements are conditionally dependent in general and the alternative hypothesis is composite. To circumvent these problems, this paper presents two system design approaches: in Approach 1, a modified decay function is considered; in Approach 2, a modified region of interest and a suitable distribution for the emitter location are considered; and both approaches use enlarged sensor deployment regions. It is shown that both approaches cause the measurements to become conditionally independent and identically distributed, cause the alternative hypothesis to become simple, and generate designs that ensure a detection performance. This paper further evaluates how conservative each approach is and compares them, helping a designer choose the most suitable approach for a situation.

Vanishing decay functions and enlarged deployment regions to facilitate the design of randomly deployed sensor detection systems

The design of randomly deployed sensor detection systems to detect a low-power signal emitter tends to be difficult because the measurements are conditionally dependent in general and the designer usually does not have all the information to characterize the distribution of measurements, making the alternative hypothesis composite. In this paper, it is shown that these two issues can be solved if the system designer considers a vanishing decay function and an enlarged sensor deployment region. It is shown that there are sufficient conditions under which such a design approach causes the measurements to become independent of the emitter location, causes the measurements to become conditionally independent and identically distributed, and generates a design that ensures a detection performance.

Least favorable distributions for the design of sensor detection systems in non-circular regions of interest

When designing a sensor system to detect the presence of an emitter whose random location has an unknown distribution, least favorable distributions can be used to avoid complications associated with composite hypotheses. Furthermore, sufficient conditions exist under which the least favorable distribution for the emitter location also makes the measurements conditionally independent and identically distributed, avoiding complications associated with conditionally dependent measurements. One such sufficient condition is having the region of interest be circular. We show that it is not trivial to extend this result to non-circular regions, and we identify a new set of sufficient conditions for a large set of non-circular regions. We also provide results to help choose among regions of interest.