Approximate replication of data using adaptive filters in Wireless Sensor Networks

Approximate replication of data is a technique commonly used in distributed systems where the exact value from the source (client) is not required at the destination (server), rather a fairly accurate estimate of the actual data is sufficient for the end application to infer a result. A Wireless Sensor Network (WSN) is a special case of distributed system where energy is the prime concern governing its life span. Approximate replication when deployed inWSN causes heavy energy savings if the application at the sink is intelligent enough to deduce the result from an estimate of actual sensed data. The crux of approximate replication is the adaptation and prediction mechanisms that cause the sensing node to transmit only a subset of actual sensed data. In this paper, we have modified the traditional approach of approximate replication through coordinating filters, by deploying different adaptive algorithms at the sink and the node. We have also modified the existing model by properly initializing the filter parameters to facilitate faster convergence, whenever there is a change in the working mode of the model. We also bring about some changes in the LMS algorithm to make sure that it suits to the prediction conditions of the approximate replication platform. In the end we have demonstrated by means of simulation that the new modifications achieve better energy savings than the existing peer models by reducing the size of the subset of the information transmitted.