Chandra Shekhar Rai

Analysis of Some Feedforward Artificial Neural Network Training Algorithms for Developing Localization Framework in Wireless Sensor Networks

Wireless sensor networks (WSNs) have gained global attention of both, the research community and various application users. Localisation in WSN plays a crucial role in implementing myriad of applications such as healthcare management, disaster management, environment management, and agriculture management. Localization algorithms have become an essential requirement to enhance the effectiveness of WSNs demonstrating relative estimation of sensor node position of anchor nodes with their absolute coordinates. We have done a comprehensive performance evaluation of some feedforward artificial neural networks (FFANNs) training algorithms for developing efficient localization framework in WSNs. The proposed work utilizes the received signal strength observed by anchor nodes by means of some multi-path propagation effects. This paper aims for best training algorithm output while comparing results of different training algorithms. The FFANNs is designed with 3-dimensional inputs and one hidden layer with variable neurons and two outputs. For hidden layer tansigmoid transfer function while for output layer linear transfer function is used. The best training algorithm of FFANNs based model can provide better position accuracy and precision for the future applications. We have analysed and proposed the usage of training algorithms that improves the accuracy and precision of localization algorithms. The simulation results demonstrate the effectiveness and huge potential in optimizing hardware for localization module in sensor nodes.

Near---Far Resistant ICA based Detector for DS-CDMA Systems in the Downlink

The performance of a conventional single user DS-CDMA receiver is severely limited by multiple access interference (MAI) and near–far effects. This severity has motivated research into adaptive filtering, near–far resistant detectors and power control strategies for DS-CDMA systems. In this paper, we propose a near–far resistant detector based on independent component analysis (ICA) of the received signal. Since ICA is a blind technique, the proposed ICA based detector has the potential to combat the near–far problem. The ICA is a higher order statistical technique based on the assumption of independence of source signals. The assumptions in ICA algorithm are the realistic conditions in a DS-CDMA system and therefore ICA algorithm can be applied successfully to detect the signal of the desired user. The focus of this paper is to illustrate the near–far resistance capability of the ICA based detector. Simulation studies performed on the proposed detector show that it is resistant to the near-far problem and has low bit error rate.

TiD: Standalone software for mining putative drug targets from bacterial proteome.

TiD is a standalone application, which relies on basic assumption that a protein must be essential for pathogens survival and non-homologous with host to qualify as putative target. With an input bacterial proteome, TiD removes paralogous proteins, picks essential ones, and excludes proteins homologous with host organisms. The targets illustrate non-homology with at least 40 out of 84 gut microbes, considered safe for human. TiD classifies proposed targets as known, novel and virulent. Users can perform pathway analysis, choke point analysis, interactome analysis, subcellular localization and functional annotations through web servers cross-referenced with the application. Drug targets identified by TiD for Listeria monocytogenes, Bacillus anthracis and Pseudomonas aeruginosa have revealed significant overlaps with previous studies. TiD takes <2h to scan putative targets from a bacterial proteome with ~5000 proteins; hence, we propose it as a useful tool for rational drug design. TiD is available at http://bmicnip.in/TiD/.

Analysis of Radial Basis Function network for localization framework in Wireless Sensor Networks

Wireless Sensor Networks (WSNs) are nowadays extensively preferred for collecting data in the field of disaster management, military operations, habitat monitoring, medical monitoring, and environment monitoring. The location of the sensor which is sending this data is very important for developing efficient routing algorithms, energy efficient communication protocols, and other Quality of services (QoS). Localization is the process by which the sensor motes in the network can identify their own location in the overall network. In this paper, we analyse Radial Basis Function (RBF) Network for developing localization framework in WSNs. RBF based localization framework is to be analysed for faster speed of convergence and low cost of computation. We present analysis of RBF through probabilistic neural network and generalized regression neural network in this paper. The proposed method can be used for designing cost-effective localization framework.

Fair adaptive resource allocation algorithm for heterogeneous users in OFDMA system

In the future, high capacity downlink would be required in cellular system with quality of service provisioning. The quality of services (QoS) means acceptable data transfer rate, signal to noise ratio (SNR), and bit error rate (BER). It can be achieved by management of resources in orthogonal frequency division multiple access (OFDMA) system. The main issue in OFDMA systems is allocation of sub-carriers and power among different types of users. In this paper, an algorithm is proposed for heterogeneous users in OFDMA system. Two different types of users are considered on the basis of required quality of services that is sensitive user (SU) and general user (GU). While the priority is given to the sensitive users, some resources are also allocated to general users to keep the system fair. The complexity of algorithm is also a major issue and it can be reduced by allocation of subcarrier and power sequentially. The water-filling method applies for power allocation among users in SU category and equal power distribution among the users in GU category to avoid the calculation burden of CPU.

Measuring the effectiveness of Multi Path Routing as a function of increased path availability

Mobile Ad-hoc networks (MANETs) are self organizing, infrastructure-less and multi-hop packet forwarding networks. Apart from the normal responsibility, each node has the additional responsibility that it also acts as a router. The dynamic behavior, distributed control and scarcity of resources makes routing in ad hoc networks, a tricky task. Several studies proclaim the benefits accrued from multi path routing, when compared with single path routing. These studies have compared multi path routing vis-à-vis single path routing and demonstrated the advantages associated. But not much work has been done to demonstrate that the benefits of Multi Path Routing become less pronounced, as more and more alternate paths are put to disposal. For this we define a term Multi Path Routing Index (MRI), as the maximum number of paths selected for routing data packets, out of the total available data paths. In this paper, a new multi path routing protocol is implemented, which allows us to vary and control actual number of alternate paths. We try to investigate the effects of increasing alternate paths at our disposal, on the overall network performance.

Performance Evaluation of IEEE 802.11 DCF in Single Hop Ad Hoc Networks

The mathematical modeling of IEEE 802.11 CSMA/CA has received considerable attention in the past few years. However there is lack of significant work considering the presence of buffer in wireless networks. Also most of the work has not followed IEEE specifications regarding inclusion of retry limits. This paper presents a new analytical model for performance evaluation of IEEE 802.11 single hop networks. The work presented here takes into consideration the presence of first order memory buffer along with the effect of traffic arrival which obeys Poisson distribution. In addition, we have considered the short retry limit to accommodate IEEE 802.11 specifications. On one hand, lack of buffering leads to high delays and under utilization of the channel. Also the absence of short retry limit tends to overestimate the throughput. By accommodating both the shortcomings, it has been proved that our model predicts the behaviour more accurately. We have been carrying out extensive simulations to validate results of our model. We have presented the performance evaluation of both the access mechanisms present in 802.11 MAC protocol.

Metrics for Weight Stuck-at-Zero Fault in Sigmoidal FFANNs

In this paper, a class of weight fault model known as single weight stuck-at-zero for a single hidden layer (with sigmoidal nodes) feed forward artificial neural networks is analyzed. Fault measures/metrics are derived for weight stuck-at-zero fault. Experiments are conducted for four function approximation tasks wherein a set of 30 networks are trained for each task. A network which has a least validation error is selected for further analysis of weight stuck-at fault for the four function approximation tasks. The average change in the prediction error on single fault seeding is measured and compared with the predicted fault measure. Correlation between the derived fault measures and empirical measure of single fault seeding demonstrate that the correlation is significant at 0.10 level for both the derived measures, for one o f the derived measures, the correlation is significant at 0.05 level. Thus, these two derived measures are shown to be good metrics for the measurement of the fault tolerance of the network to single weight fault, at least for the two function approximation tasks. Further experimentation is required to empirically assess the validity of these measures.

Sensitivity Measurement of Neural Hardware: A Simulation Based Study

Artificial Neural Networks are inherently fault tolerant. Fault tolerance properties of artificial neural networks have been investigated with reference to the hardware model of artificial neural networks. Limited precision of neural hardware lead to study the sensitivity of feedforward layered networks for weight and input errors. In this paper, we analyze the sensitivity of feedforward layered network. We propose a framework for the investigation of fault tolerance properties of a hardware model of artificial neural networks. The result obtained indicates that networks obtained by training them with the resilient back propagation algorithm are not fault tolerant.

ICA based Blind Multi User Detection for DS-CDMA Systems

In this paper, we present an Independent Component Analysis (ICA) based detection approach for Direct Sequence Code Division Multiple Access (DS-CDMA) systems. Similar to conventional matched filter receivers, the proposed receiver is based on only the spreading code and timing information of the desired user. However, its performance is much better than the conventional receiver and it is comparable to other multiuser detection techniques based on training sequences or knowledge of spreading codes, delays and amplitudes of all the interfering users. In a practical DS-CDMA system, the interfering users keep on changing. This makes the wireless environment highly dynamic. Therefore, the multiuser detection techniques requiring complete information about interfering users are not spectrum efficient. The proposed technique does not require information about the interfering users and thus can handle changing wireless conditions. We illustrate the comparison of Bit Error Rate (BER) of the proposed approach with conventional matched filter receiver and other multiuser detection approaches. Simulation results show the effectiveness of the proposed approach.