Torsha Banerjee

Fault tolerant multiple event detection in a wireless sensor network

With an increasing acceptance of Wireless Sensor Networks (WSNs), the health of individual sensor is becoming critical in identifying important events in the region of interest. One of the key challenges in detecting event in a WSN is how to detect it accurately transmitting minimum information providing sufficient details about the event. At the same time, it is also important to devise a strategy to handle multiple events occurring simultaneously. In this paper, we propose a Polynomial-based scheme that addresses these problems of Event Region Detection (PERD) by having a aggregation tree of sensor nodes. We employ a data aggregation scheme, TREG (proposed in our earlier work) to perform function approximation of the event using a multivariate polynomial regression. Only coefficients of the polynomial (P) are passed instead of aggregated data. PERD includes two components: event recognition and event report with boundary detection. This can be performed for multiple simultaneously occurring events. We also identify faulty sensor(s) using the aggregation tree. Performing further mathematical operations on the calculated P can identify the maximum (max) and minimum (min) values of the sensed attribute and their locations. Therefore, if any sensor reports a data value outside the [min, max] range, it can be identified as a faulty sensor. Since PERD is implemented over a polynomial tree on a WSN in a distributed manner, it is easily scalable and computation overhead is marginal. Results reveal that event(s) can be detected by PERD with error in detection remaining almost constant achieving a percentage error within a threshold of 10% with increase in communication range. Results also show that a faulty sensor can be detected with an average accuracy of 94% and it increases with increase in node density.

Tree based data aggregation in sensor networks using polynomial regression

In this paper, we propose a tree based regression algorithm, (TREG) that addresses the problem of data compression in wireless sensor networks. By function approximation based on multivariable polynomial regression and passing only the coefficients returned by the regression function instead of aggregated data, TREG achieves the following goals: (1) the sink can get attribute values in regions devoid of sensor nodes for attribute values that show smooth spatial gradation (2) readings over any portion of the region can be obtained at one time by querying the root instead of flooding those regions, thus incurring significant energy savings. As size of the data packet transmitted, from one tree node to another remains constant, the proposed scheme scales well with growing network density. Extensive simulations are performed on real world data to demonstrate the effectiveness of our aggregation algorithm. Results reveal that for a network density of 0.0025, the optimal tree-depth should be 4 in order to restrict the absolute error to less than a threshold of 6%. A data compression ratio of about 0.02 is achieved using our proposed algorithm, which is almost independent of tree depth.

Distributed data aggregation in sensor networks by regression based compression

In this paper we propose a method for data compression and its subsequent regeneration using a polynomial regression technique. We approximate data received over the considered area by fitting it to a function and communicate this by passing only the coefficients that describe the function. In this paper, we extend our previous algorithm TREG to consider non-complete aggregation trees. The proposed algorithm DUMMYREG is run at each parent node and uses information present in the existing child to construct a complete binary tree. In addition to obtaining values in regions devoid of sensor nodes and reducing communication overhead, this new approach further reduces the error when the readings are regenerated at the sink. Results reveal that for a network density of 0.0025 and a complete binary tree of depth 4, the absolute error is 6%. For a non-complete binary tree, TREG returns an error of 18% while this is reduced to 12% when DUMMYREG is used

Using polynomial regression for data representation in wireless sensor networks

Unlike conventional sensor networks, wireless sensors are limited in power, have much smaller memory buffers, and possess relatively slower processing speeds. These characteristics necessitate minimum transfer and storage of information in order to prolong the network lifetime. In this paper, we exploit the spatio-temporal nature of sensor data to approximate the current values of the sensors based on readings obtained from neighbouring sensors and itself. We propose a tree based polynomial regression algorithm (TREG), that addresses the problem of data compression in wireless sensor networks. Instead of aggregated data, only the coefficients computed by the regression function, TREG are passed to achieve the following goals: (i) the sink can get attribute values in the regions devoid of sensor nodes, and (ii) readings over any portion of the region can be obtained at one time by querying the root of the tree. As the size of the data packet from each tree node to its parent remains constant, the proposed scheme scales very well with growing network density or increased coverage area. Since physical attributes exhibit a gradual change over time, we propose an iterative scheme, UPDATE_COEFF, which obviates the need to perform the regression function repeatedly and uses approximations based on previous readings. Extensive simulations are performed on real world data to demonstrate the effectiveness of the aggregation algorithm, TREG. Results reveal that for a network density of 0.0025, a complete binary tree of depth 4 could provide the absolute error to be less than 6%. A data compression ratio of about 0.02 is achieved using our proposed algorithm, which is almost independent of the tree depth. In addition, our proposed updating scheme makes the aggregation process faster while maintaining the desired error bounds. Copyright

LIMOC: Enhancing the Lifetime of a Sensor Network with Mobile Clusterheads

This paper proposes a scheme for enhancement of network Lifetime using MObile Clusterheads (LIMOC) in a wireless sensor network (WSN). The low energy, static sensor nodes sense physical parameters and route the data to the highly energy-rich nodes called ClusterHeads (CHs) which are mobile and can transmit data directly to the base station (BS). A collaborative strategy among the CHs increases the lifetime (hence residual energy) of the network further. Simulation has shown that residual energy of the network can be improved by 45% by making the CHs always move towards the event in an event-driven network. For general cases, increased energy savings is obtained by making the CH move towards the center of equilibrium w.r.t. to both the total residual energy and data flow of the network.

Increasing Lifetime of Wireless Sensor Networks Using Controllable Mobile Cluster Heads

This paper proposes a scheme for improvement of network lifetime and delay by employing a connected group of mobile cluster heads in a wireless sensor network (WSN). As data sensing is triggered by an event, the sensors relaying the aggregated data to the base station run out of energy at a much faster rate than sensors in other parts of the network. This gives rise to an unequal distribution of residual energy in the network, making the nodes with lower remaining energy level to die much faster than others. To distribute the remaining energy more evenly in the network, some energy-rich nodes are designated as cluster heads which move in a controlled manner toward locations rich in energy and data. This reduces the transmission energy required by the static sensors to send data and thus increases the overall lifetime of the network. Along with transmission energy, time taken for transmitting data to the BS is also reduced as the CHs follow a connectivity strategy to always maintain a connected path to the BS. Simulation shows that lifetime of the network can be increased by 42% over existing scheme by making the CHs always move towards a stable equilibrium point, i.e., a point where the total residual energy of the network and data are concentrated. Our connectivity algorithm also provides 40% improvement in the transmission delay as compared to existing schemes.

Wireless Sensor based Dynamic Channel Selection in Cellular Communication by Cognitive Radio Approach

In a cellular communication scenario, wireless sensors can be deployed to sense the interference power of the frequency band. In an ideal channel, interference temperature (IT) which is directly proportional to the interference power can be assumed to vary spatially with the frequency of the sub channel. We propose a scheme for approximating ITs over an extended C-band (licensed and unused television band) by fitting sub channel frequencies and corresponding ITs to a regression model. Using this model, IT of a random sub channel can be calculated by the base station (BS) for further analysis of the channel interference. Our proposed model based on readings reported by sensors helps in dynamic channel selection (S-DCS) in extended C-band for assignment to unlicensed secondary users. S-DCS maximizes channel utilization and proves to be economic from energy consumption point of view. It also exhibits substantial amount of accuracy with error bound within 6.8%. Again, users are assigned empty sub channels without actually probing them, incurring minimum delay in the process. Overall channel allocation efficiency is also maximized along with fairness to individual users

PERD: Polynomial-based Event Region Detection in Wireless Sensor Networks

We propose a polynomial-based scheme that addresses the problem of event region detection (PERD) for wireless sensor networks (WSNs). Nodes of an aggregation tree perform function approximation of the event using multivariate polynomial regression. Only the coefficients of this polynomial are then transmitted up the tree instead of aggregated data. PERD includes two components: event recognition and event report with boundary detection. In addition, PERD is capable of detecting single event or multiple events simultaneously occurring in the WSN. Since PERD is implemented over a polynomial tree on a WSN in a distributed manner, it is easily scalable and computation overhead is very light. Results reveal that event(s) can be detected by PERD with error in detection remaining almost constant achieving a percentage error within a threshold of 10% with increase in communication range.

Achieving Fault Tolerance in Data Aggregation in Wireless Sensor Networks

This paper identifies faulty sensor(s) in a polynomial-based data aggregation scenario, TREG proposed in our earlier work. In TREG, function approximation is performed over the entire range of data and only coefficients of a polynomial (P) are passed instead of aggregated data. Performing further mathematical operations on the calculated P can identify the maximum (max) and minimum (min) values of the sensed attribute and their locations. Therefore, if any sensor reports a data value outside the [min, max] range, it can be identified as a faulty sensor. We achieve the following goals: (1) uncorrelated readings from a specific sensor helps in detecting a faulty sensor, (2) faulty sensors are detected near the source and isolated preventing them from affecting the accuracy of the overall aggregated data and reducing the overall delay. Results show that a faulty sensor can be detected with an average accuracy of 94 %. With increase in node density, accuracy in faulty sensor detection improves as more nodes are able to report the information to their nearest tree node.

Exploiting Spatial Correlation in a three dimensional Wireless Sensor Network

Physical attributes such as temperature exhibit gradual change in value with space. Observations from the sensor nodes are thus highly correlated. In this paper, we propose a scheme to exploit the spatial correlation of data in a three dimensional wireless sensor network (WSN), in a way so as to reduce the number of transmissions in the network, thereby saving energy. The presented work is an extension of the earlier proposed TREG scheme which is only applicable to two dimensional WSN. Our scheme involves formation of binary tree and transmission of data along the tree nodes. Simulation has been carried out for different tree heights. Simulation results show that a tree height of four provides results with optimal values.