In this paper, the problem of determining faulty readings in a wireless sensor network without compromising detection of important events is studied. By exploring correlations between readings of sensors, a correlation network is built based on similarity between readings of two sensors. By exploring Markov Chain in the network, a mechanism for rating sensors in terms of the correlation, called SensorRank, is developed. In light of SensorRank, an efficient in-network voting algorithm, called TrustVoting, is proposed to determine faulty sensor readings. Performance studies are con- ducted via simulation. Experimental results show that the proposed algorithm outperforms majority voting and dis- tance weighted voting, two state-of-the-art approaches for in-network faulty reading detection.
Sensors are prone to failure in harsh and unreliable envi- ronments. Faulty sensors are likely to report arbitrary read- ings which do not reflect the true state of environmental phe- nomenon or events under monitoring. Meanwhile, sensors may sometimes report noisy readings resulted from interfer- ences [3]. Both arbitrary and noisy readings are viewed as faulty readings in this paper. The presence of faulty readings may cause inaccurate query results and hinder their useful- ness. Thus, it is critical to identify and filter out faulty readings so as to improve the query accuracy.
As mentioned earlier, prior works only take the distance between sensor nodes into consideration when modeling the correlation of sensor readings. However, it is also possible that the readings of two geographically close sensor nodes to have dramatically different readings. Thus, it’s critical to truly capture the correlation of sensor readings rather than their distance.
SensorRank is to represent the trustworthiness of sensor nodes. By our design, two requirements need to be met in deriving SensorRank for each sensor.
Requirement 1: If a sensor has a large number of neigh- bors with correlated readings, the opinion of this sensor is trustworthy and thus its vote deserves more weight.
Requirement 2: A sensor node with a lot of trustworthy neighbors is also trustworthy.
These two requirements ensure that 1) a sensor node which has a large number of similar neighbors to have a high rank; and 2) a sensor node which has a large number of ’good ref- erences’ to have a high rank. Given a correlation network G = (V, E) derived previously, we determine SensorRank for each sensor to meet the above two requirements.
Here we describe our design of the TrustVoting algorithm, which consists of two phases: a) self-diagnosis; and b) neigh- bors diagnosis phase. In the self-diagnosis phase, each sensor verifies whether the current reading of a sensor is unusual or not. Once the reading of a sensor goes through the self- diagnosis phase, this sensor can directly report the reading. Otherwise, the sensor node consults with its neighbors to further validate whether the current reading is faulty or not. If a reading is determined as faulty, it will be filtered out. The sensor nodes generating faulty readings will not partic- ipate in voting since these sensors are likely to contaminate the voting result. Note that TrustVoting is an in-network algorithm which is executed in a distributed manner. The execution order of algorithm TrustVoting has an impact on faulty reading detection. We will discuss this issue later.
When a set of sensor nodes is queried, each sensor in the queried set performs a self-diagnosis procedure to verify whether its current reading vector is faulty or not. Once the reading vector of a sensor node is determined as normal, the sensor node does not need to enter the neighbor-diagnosis phase. To execute a self-diagnosis, each sensor si only main- tains two reading vectors: i) the current reading vector at the current time t (denoted as bi (t)); and ii) the last correct reading vector at a previous time tp (expressed by bi (tp)). bi (tp) records a series of readings occurred in the previous time and is used for checking whether the current reading behavior is faulty or not. If these two reading vectors are not similar, bi (t) is viewed as an unusual reading vector. Once a sensor node is detected an unusual reading vector, this sensor node will enter the neighbor-diagnosis phase to fur- ther decide whether the unusual reading behavior is faulty or not. Note that when bi (t) is identified as a normal vector through the neighbor-diagnosis, bi (tp) is updated so as to reflect the current monitoring state.
We simulate a synthetic environment, where sensors are deployed in a 500 by 500 to monitor temperatures. The tem- perature reading range is [ 25, 275]. Moreover, events with unusual readings are randomly generated in the monitored field. The model of generating events are the same as in [5, 6]. The faulty sensor rate (abbreviated as faulty rate) is the ratio of the number of faulty sensors and the total number of sensors deployed. Each sensor will report noisy readings according to the parameter noise prob. A faulty sensor always report faulty readings and thus noise prob is set to 1 for faulty sensors. On the other hand, a normal sensor is still likely to report noise or faulty readings. Thus, for normal sensors, we set the noise prob to 0.1. A noise reading (referred to as a faulty reading) is randomly biased from the normal reading generated and the amount of bias is within the range of [ 50, 50]. A query is submitted to wireless sensor networks with its query region as a rectangle and query region size varied from 80 by 80 up to 160 by 160. To evaluate the simulation result, two performance metrics are employed: faulty detection rate and false positive rate. Specifically, a query is issued to a query region B to obtain the current readings sensed by the sensors, where the set of these current readings is denoted as XB .
With the presence of faulty readings, the accuracy of query results in wireless sensor networks may be greatly affected. In this paper, we first formulated the correlation among readings of sensors nodes. Given correlations among sensor nodes, a correlation network is built to facilitate derivation of SensorRank for sensor nodes in the network. In light of SensorRank, an in-network algorithm TrustVoting is devel- oped to determine faulty readings. Performance evaluation shows that by exploiting SensorRank, algorithm TrustVot- ing is able to efficiently identify faulty readings and out- performs majority voting and distance weighted voting, two state-of-the-art approaches for in-network faulty reading de- tection.
Wen-Chih Peng was supported in part by the National Sci- ence Council, Project No. NSC 95-2211-E-009-61-MY3 and NSC 95-2221-E-009-026, Taiwan, Republic of China. Wang- Chien Lee was supported in part by the National Science Foundation under Grant no. IIS-0328881, IIS-0534343 and CNS-0626709.
CS 340 Milestone One Guidelines and Rubric Overview: For this assignment, you will implement the fundamental operations of create, read, update,
Retail Transaction Programming Project Project Requirements: Develop a program to emulate a purchase transaction at a retail store. This
7COM1028 Secure Systems Programming Referral Coursework: Secure
Create a GUI program that:Accepts the following from a user:Item NameItem QuantityItem PriceAllows the user to create a file to store the sales receip
CS 340 Final Project Guidelines and Rubric Overview The final project will encompass developing a web service using a software stack and impleme
