Current research involves one or more of the topics from my areas of interest. However, the application may be GM confidential.

Areas of Interest
  • Digital Signal Processing
  • Detection and estimation
  • Adaptive data modeling
  • Pattern recognition and classification
  • Non-parametric Bayesian estimation
  • Machine learning
  • Data mining
  • Sensor/data fusion
  • Sensor optimization
  • Time-frequency signal analysis
  • Stochastic filtering (Kalman filtering, Particle filtering, Bayesian filtering)
  • Multi-core implementation


Sponsor: Air Force Office of Scientific Research (AFOSR) MURI Grant.

Asst. Universities: Arizona State University, Johns Hopkins University, University of Southern California, Virginia Tech

Advisor: Prof. Antonia Papandreou Suppappola

Co-researcher: Dr. Narayan Kovvali

Category: Research for doctoral thesis (2006 - 2010)

This project entailed designing of novel detection and estimation algorithm methods, using joint time-frequency features. Matching pursuit decomposition (MPD) is the primary feature extraction algorithm. Preliminary classification algorithms based on MPD correlation was discussed. Further, a more stochastic approach, hidden Markov model (HMM) was used for data modeling and classification. When data from multiple sensors were available, Bayesian sensor fusion was incorporated to enhance performance. Recently, a more adaptive approach was used to overcome the limitations of training overheads and incorporate variations due to operating conditions. A MPD based probability density function (MPD-PDF) was established to estimate statistical measure of similarity between data. This similarity feature was then used for Dirichlet process based clustering for unsubervised learning. To allow learning actual labels for the data, Bayesian filter was used in a state-space formulation which incorporated using a theoretical state model. In this approach, active data selection is incorporated maximize information diversity. Currently, I am looking at transfer learning approaches.

Project Objective:
In this MURI project, established in 2006, we are aiming at producing a major advance in the ability to provide reliable life cycle estimates for current and future aircraft systems. The current states of structural health monitoring (SHM), damage diagnosis and prognosis will be transformed by the introduction of a hierarchical framework of sensor data, information, models and algorithms that span and integrate scales from microstructure to structural level. The proposed effort addresses the RCAs of this MURI topic through integration of material characterization (including high temperature), computational modeling, sensor instrumentation, information management, damage detection and benchmark laboratory experiments. The proposed effort will be undertaken by an interdisciplinary research team with specific expertise in material, structural, mechanical, electrical and systems engineering and with extensive experience in interdisciplinary collaborative research projects under DoD sponsorship. To ensure Air Force relevance, the team has already started engaging AFRL personnel to help identify crucial issues; the team plans to aggressively pursue collaborations with DoD laboratory personnel during the project to optimize our understanding of crucial challenges, the use of available test data, and the likelihood of transitioning results from the project into DoD practice. We are supported by the Department of Defense MURI program through the Air Force Office of Scientific Research (AFOSR).


Support from: DARPA, Raytheon.

Advisor: Prof. Antonia Papandreou Suppappola

Co-researcher: Prof. Darryl Morrell, Dr. Bhavana Chakraborty, Camron Vossberg, Abhisek Swain

Category: Former Research (2006)

Acoustic signals created by footsteps can be detected due to their unique time-frequency signature. Motes, placed at random form a latent sensor array that can transpond acoustic signals produced by the footsteps of people walking. The footstep detection algorithm used filtering of data, and a heuristic thresholding approach based on prior training and validation. This was implemented on board, using the fixed point DSP processor. Detection result and energy estimates were communicated wirelessly to a central computing facility where particle filter was used for target tracking.

Project objective:
Motes are wireless sensor platform gives the flexibility to create powerful, tether-less, and automated data collection and monitoring systems. Algorithms are in the the progress of development of accurate target tracking. The areas of application include power plants to medical devices, automobile engines to environmental monitoring systems, navigation to safety, sensors are increasingly important in many aspects of our daily lives. This project is aimed at perimeter security. Presense of target is sensoned using acoustic sensors and the energy information is further used to perform tracking.