Priyadip Ray

Bayesian joint analysis of heterogeneous genomics data

SUMMARY A non-parametric Bayesian factor model is proposed for joint analysis of multi-platform genomics data. The approach is based on factorizing the latent space (feature space) into a shared component and a data-specific component with the dimensionality of these components (spaces) inferred via a beta-Bernoulli process. The proposed approach is demonstrated by jointly analyzing gene expression/copy number variations and gene expression/methylation data for ovarian cancer patients, showing that the proposed model can potentially uncover key drivers related to cancer. AVAILABILITY AND IMPLEMENTATION The source code for this model is written in MATLAB and has been made publicly available at https://sites.google.com/site/jointgenomics/. CONTACT catherine.ll.zheng@gmail.com SUPPLEMENTARY INFORMATION Supplementary data are available at Bioinformatics online.

Subspace-based cooperative spectrum sensing for cognitive radio

Cognitive radio with its capability to sense the radio environment and dynamically access the vacant spectrum opportunistically, is an important technology to improve the efficiency of current spectrum utilization. In this paper, a cooperative wideband spectrum sensing approach based on subspace methods is proposed to estimate the number of primary user signals present in the band of interest and their carrier locations. Specifically, each secondary user implements eigen-decomposition of its sample covariance matrix to provide a local estimate. The fusion center generates a global estimate through a weighted sum of these local estimates. Experimental results demonstrate the efficiency of the proposed algorithm in detecting the correct number of primary users and estimating their carrier frequencies.

Distributed Detection in Wireless Sensor Networks Using Dynamic Sensor Thresholds

This paper presents a new approach for distributed target detection in wireless sensor networks (WSNs). Contrary to the conventional practice where every sensor uses an identical threshold for decision-making, an unequal and dynamic local sensor threshold selection scheme is proposed. This threshold selection scheme is based on a recently proposed statistical metric for multiple testing problems called the False Discovery Rate (FDR). Assuming a signal attenuation model, where the received signal power decays as the distance from the target increases, various performance indices like the system level probability of detection and the probability of false alarm are studied. Simulation results are provided to demonstrate the effectiveness of this approach.

Estimation of spatially distributed processes in wireless sensor networks with random packet loss

This paper studies the effect of wireless channel imperfections on the transport and estimation of spatially distributed events using wireless sensor networks (WSNs). It is observed that the quality of event estimation at the sink (fusion center) degrades considerably with correlated packet losses during transmission from the sensors. A novel diversity technique based on field estimation is proposed to mitigate the effects of packet losses on the quality of estimation at the sink. Dense deployment of sensor nodes and the spatial nature of the observed physical phenomenon result in the sensor observations being noisy spatial samples of an unknown underlying function. The proposed algorithm exploits this feature, using supervised learning to achieve diversity. A new information fusion methodology based on approximate likelihood is proposed to integrate the information obtained from the learning algorithm into the classical estimation framework. Simulation results are provided to demonstrate the performance of the proposed approach.

Factor Graph Based Cooperative Spectrum Sensing in Cognitive Radio over Time-Varying Channels

A normal factor graph (NFG) based approach for cooperative spectrum sensing in cognitive radio over time varying and frequency non-selective fading channels is presented in this paper. An NFG based representation of a distributed cognitive radio system is first presented and then a Sum-Product- Algorithm (SPA) based analysis is developed for inference. The spectrum sensing problem is modelled as a distributed binary hypothesis testing problem. A Neyman-Pearson (NP) based likelihood ratio test statistic is derived for optimal sensing. As exact theoretical analysis of the system level probability of detection and probability of false alarm is very difficult, we provide an approximation which performs satisfactorily in the moderate to high signal-to-noise ratio (SNR) regime. The proposed NFG based spectrum sensing approach is computationally scalable to large networks and performs well under time varying channel conditions. Extensive simulation results are provided to validate our proposed approximation.

Radar Target Detection Framework Based on False Discovery Rate

A new framework is presented for the detection of targets in clutter and noise. Conventional radar detection techniques involve testing each test cell separately at a predefined probability of false alarm. In this paper, we formulate a novel surveillance paradigm that controls the false discovery rate (FDR) for a specified surveillance area (SA), which consists of a number of test cells. FDR is defined as the expectation of the ratio of the number of false alarms to the total number of cells classified as targets over the entire SA. It is observed that control of FDR results in an increase in the probabilities of detection and false alarm with an increase in the number of targets in the SA. This is achieved without the actual knowledge of the number of targets in the SA. To control the increase of the probability of false alarm and to confine it within a prespecified range, we propose a hybrid detection strategy involving both SA-based testing and per-cell testing. Our proposed hybrid approach, based on the concept of algorithm fusion, provides substantial improvement in detection performance in target-rich environments at the cost of a controlled increase in the false alarm rate. Analytical and simulation results are provided to demonstrate the performance of the proposed approach.

Coupled Detection and Estimation Based Censored Spectrum Sharing in Cognitive Radio Networks

A novel spectrum sharing strategy based on coupled detection and estimation is proposed for cognitive radio networks. The proposed approach is able to tradeoff throughput for reduced interference at the primary user (PU) via censored transmissions. We derive the optimum censoring strategy that maximizes the throughput of the cognitive radio system under an average interference power constraint at the PU. We then extend the proposed framework to jointly optimize the censoring and the power allocation strategies of the secondary user (SU) that maximize the throughput of the secondary network under average transmit power and average interference power constraints. Finally, we provide extensive simulation results to demonstrate the enhanced performance of the proposed censoring based spectrum sharing approach.

Efficient wideband spectrum sensing using random projection

Subspace based spectrum estimation is a powerful technique for wideband spectrum sensing. Unlike narrowband sensing, wideband sensing requires no prior information about the band-structure or bandwidth of the primary users of the spectrum, thus making it an attractive alternative to narrowband spectrum sensing. Typically, subspace based techniques require eigen-decomposition of the sample covariance matrix, which is computationally very expensive. As the expected number of primary users in the system increases, the size of the covariance matrix increases, thus increasing the spectrum sensing time, resulting in the reduction of overall throughput. In this paper, an efficient approach to perform subspace based spectrum sensing via random projection is proposed. In the proposed approach, spectral decomposition of a significantly lower order matrix is required for wideband spectrum sensing. The time complexity of the proposed approach is shown to be much better than conventional subspace based techniques. In addition to improved time complexity, the regularization imposed via the low rank approximation, improves the spectrum sensing performance of the proposed approach over the conventional subspace based approach, especially with limited observations. Simulations results are provided to demonstrate the effectiveness of the proposed approach.

Heart rate estimation from photoplethysmogram during intensive physical exercise using non-parametric Bayesian factor analysis

Estimating heart rate from a photoplethysmogram (PPG) signal in the presence of severe motion artifacts is a challenging problem. Previous approaches have primarily relied on accelerometer data to remove motion artifacts. In this paper, we propose a non-parametric Bayesian factor analysis based approach to estimate heart rate from a PPG signal, in the presence of severe motion artifacts. The novelty of our approach lies in the fact that it relies only on a single channel PPG signal to estimate heart rate, and does not require accelerometer data. The accuracy and robustness of the proposed approach has been demonstrated on a range of datasets corresponding to different intensive physical exercise scenarios.

A false discovery rate based detector for detection of targets in clutter and noise

This paper presents a new framework for detection of targets in clutter and noise. Conventional radar detection techniques involve testing each range cell separately at a predefined probability of false alarm. In this paper, we propose a novel surveillance paradigm that controls the false discovery rate (FDR) for a specified surveillance area (SA), which consists of a number of range cells. This approach is adaptive to the number of targets in the surveillance area, and improved detection performance may be achieved at the cost of slightly higher false alarms over the surveillance area.