Wireless Sensor Networks (WSNs) finds application in a variety of fields, including agriculture, healthcare, and renewable energy. These sensors are used to monitor and record physical conditions such as temperature, humidity, etc.
The increase in the applications of WSNs has made them more vulnerable to attacks such as hacking.
These attacks aim to manipulate the normal functioning of the sensor nodes and thus disrupt the overall signal-processing capabilities of the WSNs.
Hence, there is a need to operate these WSNs in such a way that, in spite of attacks, they can accurately estimate a parameter such as temperature or humidity.
The solution to counteract this problem is to develop consensus algorithms resilient to such attacks. Recently two algorithms based on the consensus+innovations approach have been developed, which ensure both the normal sensors, and the sensors under attack reach consensus over the true value of the parameter to be estimated. However, these algorithms assume bidirectional communication links between the sensor nodes. In many practical scenarios, the power levels at which sensor nodes broadcast information or their interference and noise patterns, differ from node to node. Therefore, the communication between nodes in such cases is unidirectional. For consensus-based algorithms to work properly, it is required to ensure averaging of the information among the agents over time. While for an undirected network, this gets taken care of by network connectivity, it is not that trivial for a network with unidirectional communication. This necessitates the need to scale the information appropriately by introducing proper weights, termed weight balancing, so that, with time, the averaging of the information in the network is facilitated.
In this study conducted by Mr. Shamik Bhattacharyya and Prof. Rachel Kalpana Kalaimani from the Department of Electrical Engineering, Indian Institute of Technology (IIT) Madras, Chennai, India, and Mr. Kiran Rokade from the Electrical and Computer Engineering Department, Cornell University, Ithaca, New York, USA, a resilient distribution estimation algorithm was devised that ensures both the normal sensors and the sensors under attack reach consensus over the true value of the parameter to be estimated even when unidirectional communication exists across the sensors. This algorithm was named the Resilient Estimation through Weight Balancing (REWB) algorithm which achieves averaging on the information in the network, required for the unidirectional case by introducing appropriate weights to scale the information of the nodes, termed weight balancing.
The REWB algorithm is a distributed estimation algorithm over directed graphs resilient to sensor spoofing type attacks. To the best of the authors’ knowledge, this is the first time that such an algorithm has been devised. The only condition for the sensors to asymptotically converge to the parameter value to be estimated is that the number of sensors affected is less than half of the total number of sensors.
Future works are to consider other models of adversarial attacks.
Dr. Swanand Khare, from the Department of Mathematics, Indian Institute of Technology (IIT) Kharagpur, West Bengal, India, gave his analysis of this study and appreciated the authors’ efforts with the following comments: “In a multi-agent system, distributed estimation of an unknown parameter in the presence of sensor spoofing-type adversarial attacks was previously addressed using bidirectional communication links between agents.
In this paper, Dr. Rachel and her students consider the communication links between agents to be unidirectional, modelled as a directed graph.
To tackle the challenge that a directed graph is not inherently weight-balanced, the authors assign time-varying weights to the nodes and suitably update them to achieve weight-balancing.
Resilience to the manipulation of sensor data by adversaries is achieved by designing the update-law based on the consensus+innovations approach and using a suitable damping term in the innovations part.
Both ideas of weight-balancing and resilience are nicely integrated to develop the novel Resilient Estimation through Weight Balancing (REWB) algorithm, the essential contribution of this paper.
Interestingly, the REWB algorithm achieves the same resilience guarantee for directed graphs as the best-case guarantee for undirected graphs.”
Article by Akshay Anantharaman
Click here for the original link to the paper
