Method for adaptively modifying the observed collective behavior of individual sensor nodes based on broadcasting of parameters

Abstract

A method for continually controlling the collective behavior of a set of computing devices in a distributed data processing system. A gateway node disseminates a specification request comprising a set of parameters to a set of computing devices. The gateway node may be unaware of the number and identity of individual computing devices. Each computing node receiving the request determines whether its attributes satisfy the predicates expressed in the specification request. If so, the node processes the parameters in the specification request and modifies its own behavior based on the values in the parameters of the specification request. Subsequently, the gateway node may also observe the quality of information (QoI) values communicated from the set of computing devices. The gateway node iteratively modifies the parameters disseminated in subsequent specification requests based on a divergence between a computed quality of information value and a desired quality of information value.

Description

BACKGROUND OF THE INVENTION [0001] 1. Field of the Invention [0002] The present invention relates generally to computer and communications networks, and more specifically to the emerging field of “sensor networks”, typically consisting of a large number of individual “sensor” devices, which send back samples of some environmental state to certain designated nodes. [0003] 2. Description of the Related Art [0004] Sensor networks are an emerging and very promising new category of computer networks, characterized by the development of very-low cost sensor devices with combined sensing and communication (often wireless) capabilities. Most applications of sensor networks rely on combining information from multiple sensor devices to establish or infer some composite state or event of the sensed environment (often called the “sensing field”). Examples include the desire to “compute an average of readings taken by 10 different temperature sensors,” or to “compute the minimum out of 30 distin...

AI Technical Summary

Benefits of technology

[0019] Another aspect of this invention relates to the continual monitoring of the QoI obtained from the activated set of sensors, and the subsequent adjustment of the parameters in subsequent specification requests. Clearly, such continual monitoring allows the gateway node to rapidly re-tune the behavior of the sensor network and maintain conformance to the target QoI objective, even though the underlying sensor network may change dynamically. To make the convergence process faster, the gateway node may maintain some “state” that it infers about the current network resources, and make the adjustment of parameters a function of this estimated “state”. Many well-known techniques from control theory, such as Proportional-Integral-Derivative (PID) control, Kalman filtering, or fuzzy control, may be used for this purpose. The mechanism of the present invention may employ any closed loop control technique for rapid adaptation of individual sensor node behavior. As a continuing example, after the activation probability p has converged to approximately N / S, let us assume that some catastrophic event causes the number of available sensors to suddenly drop to N / 2. At this point, the gateway node will observe a rapid drop in the number of distinct sensors reporting data, and may then rapidly adjust the probability to a higher value, so that a higher proportion of the available sensor node population is activated to meet the desired QoI of the application.

Problems solved by technology

Examples include the desire to “compute an average of readings taken by 10 different temperature sensors,” or to “compute the minimum out of 30 distinct humidity sensor readings.” Scalability is a major technical challenge in such future networks, which are projected to contain thousands, and possibly hundreds of thousands, of tiny sensor nodes deployed fairly densely over the sensing field.

A key characteristic of many such operating environments is that such networks often exhibit significant redundancy, in that most applications do not normally require the use of sensor data from all of the available sensor nodes.

Indeed, the physical density of sensor networks may often vary greatly over the sensing field due to choice (e.g., network designers may deploy more nodes in an area where finer location accuracy is required), lack of precise control (e.g., when nodes over a remote terrain are deployed by being dropped from an airplane), or failures (e.g., when sets of sensor nodes turn out to be defective or die due to exhaustion of battery resources).

The main challenge in supporting such energy-efficient operation is that, while applications may be able to express the amount of resources (such as the number or resolution of the sensor nodes), they typically have no idea of the actual number or layout of the sensors deployed.

Accordingly, an application may not be able to decide on the appropriate settings (e.g., on or off, high or low zoom) for each individual sensor.

However, such a mechanism is not useful in future sensor network environments for two distinct reasons.

For example, the sensor network substrate can be modified either due to the occasional addition of new nodes, the death or removal of existing nodes and due to other unforeseen reasons (such as catastrophic node failures).

This imposes a substantial reporting overhead and cannot scale to large sensor network environments, since every such change in the network topology must be propagated to the database.

If a static configuration scheme is employed instead, it may quickly become inappropriate for the given operating environment.

However, if subsequently 50 nodes suffer a catastrophic failure (e.g., due to a natural disaster), the application will be left only with data from the remaining 50 nodes.

Although all of these techniques define “how” some queries or requests are propagated over the sensing field, these techniques do not define “what” is propagated or how individual nodes respond to the “what” they receive.

Moreover, they do not define or provide an adaptive method by which the “what” (i.e., the content) may be used to continually meet the QoI requirements of the application, even though the topology and other physical properties of the sensor network changes.

However, DD does not provide a method by which the application's QoI requirements can be met by iteratively issuing directed broadcasts, with modifications to the parameters contained in the broadcast.

However, like DD, the AR technique does not propose the use of an iterative mechanism to match the amount and quality of sensor data to an application's QoI requirements, thereby avoiding redundant operation of sensor nodes.

However, all of the proposals and prior art in this domain deal with mechanisms by which nodes get to know of this probability, and not with methods by which such probabilities can be used to iteratively control the collective behavior of a set of sensor nodes.

However, these approaches do not utilize a control loop for adjusting the probability to the “right level” desired by the controller, and do not talk about using the probabilities or other parametric values to adjust other behaviorial parameters (such as data reporting frequency or camera zoom level) of a subset of the sensor nodes.

However, in LEACH, there is no notion of a control loop being used by a gateway node to dynamically tune the value p to ensure that it meets a target value N. Moreover, LEACH does not utilize broadcasts from a gateway node that are targeted to a specific sub-set of the sensor nodes (identified for example by the type or location of the sensor nodes) to activate varying numbers of sensors with different attributes (e.g., type or location).

They do not provide a method of dynamically controlling the activation probabilities without knowing the node density (and for non-uniform node densities), or discuss a broadcast-based technique to activate the right number of sensors without addressing them individually.

Images