Session S8

Session S8: Adaptive Sensing

Time:                 Tuesday, May 13, 17:00-19:00
Chair:                 Robin L. Murray, Naval Undersea Warfare Center
Co-Chair:          Gang Qian, Arizona State University

S8-1: Fast-Converging Adaptive STAP for Non-Rayleigh Clutter Suppression
Tiange Shao, Y. Rosa Zheng University of Missouri-Rolla, Rolla MO, Genshe Chen DCM Research Resources LLC, Germantown MD, Erik Blasch, Air Force Research Laboratory, Wright-Patterson Air Force Base OH and Robert Lynch Naval Undersea Warfare Center, Newport RI
A newly proposed space-time adaptive processing (STAP) algorithm, namely the normalized fractionally lower-order moment (N-FLOM) algorithm has the advantage in suppressing heavy-tailed clutters for phased array radar systems. In contrast to the commonly used normalized least mean square (NLMS) algorithm which uses the second order moments of the data for adaptation, the N-FLOM algorithm uses the lower order moments to adapt the weight coefficients. The normalization is also performed based on the data sample dispersion rather than the variance. In this paper, the N-FLOM algorithm is evaluated under correlated compound-K clutters. Processing results using simulated data show that the N-FLOM algorithm converges faster than the NLMS algorithms in both Rayleigh and Compound K clutter environments. It also provides better Signal-to-Interference–and-Noise-Ratio (SINR) at the STAP output than the NLMS algorithm.

S8-2: Computationally Efficient Particle Filtering Using Adaptive Techniques
Bhavana B. Manjunath, Antonia Papandreou-Suppappola, Chaitali Chakrabarti, Arizona State University, Tempe AZ and Darryl Morrell, Arizona State University Polytechnic Campus, Mesa AZ
Most particle filtering algorithms used currently are based on sequential importance sampling that can represent the posterior density function by a set of random particles sampled from an importance sampling density and by their associated weights. Based on the number of particles used and the choice of the importance sampling density, these algorithms can be computationally intensive; they can differ in performance, efficiency, and computational cost. We adaptively use two such particle filter variations, the sequential importance resampling (SIR) particle filter and the unscented particle filter (UPF). The UPF performs more accurately than the SIR for highly nonlinear dynamic systems, but it is more computationally intensive. We propose a computationally efficient particle filtering algorithm that adaptively chooses between the SIR and UPF algorithms based on the use of the Kullback-Leibler distance (KLD) sampling. We apply this technique to a scalar estimation problem. We provide simulation results to demonstrate that the new algorithm is more accurate than the SIR and is faster than the UPF for systems characterized by highly nonlinear measurement models.

S8-3: Sample-Efficiency-Based Adaptive Particle Filter
Gang Qian and Feng Guo Arizona State University, Tempe AZ
This paper presents an adaptive particle filter framework according to sample efficiency. Sample efficiency measured by effective sample size can serve as an indicator of the tracking performance in particle filtering. To overcome degeneracy, sampling with importance resampling (SIR) is often used in particle filter to improve sample efficiency. However insufficient sample size or sudden changes in system dynamics can still lead to decreased sample efficiency and result in tracking failures even when SIR is applied. Auxiliary particle filters and its variants provide improved sample efficiency, nevertheless usually at largely increased computational cost. To obtain a balance between sample efficiency and computational cost, the proposed adaptive particle filter dynamically switches between an improved sample-efficiency-optimized auxiliary particle filter (SEO-APF) and a baseline SIR particle filter according to the sample efficiency. In general, the adaptive particle filter runs in the SIR mode and it will only switch to the SEO-APF mode when the sample efficiency is below certain prechosen threshold. Experimental results show the efficacy of the proposed algorithm.

S8-4: Steady-State Analysis of the Discrete-Time Kalman Filter with One-Step Optimal Measurement Selection
Daniel Fuhrmann Washington University in St. Louis, St. Louis MO
In this paper, we examine the steady-state behavior of the discrete-time Kalman filter incorporating optimal measurement selection. We concentrate on the case where the state behaves like a random walk with different diffusion coefficients in different directions, as described by the eigenstructure of the noise covariance. If the process noise covariance is full-rank, then in steady state, the posterior state covariance will be an identity at each time step. This follows directly from the waterfilling solution for the optimal measurement matrix. Given this, we can relate the posterior variance of each state component to the eigenvalues of the process noise covariance. Using convex optimization arguments, it is shown how the performance improvement that can be obtained by adaptive sensing is related to the spread of these eigenvalues. Furthermore, it is shown that, in situations where there is the most to be gained by adaptive sensing, there is also the most to be lost in performance through the incorrect choice of the measurement. If the process noise covariance is not full-rank, then in steady state, the posterior state covariance will have r eigenvalues equal to a, and the remaining eigenvalues are 0, where r is the rank of the process noise covariance. This is the situation that leads to the greatest performance improvement from adaptive sensing, since it implies that there exist “modes” of the state space which do not need to be interrogated by the measurement, and hence the available energy is better used elsewhere. The gains to be found in the posterior variances are mathematically equivalent to those resulting from the SNR gain in adaptive sensor array processing for passive sensing.

S8-5: Cost Efficient Target Tracking in a Distributed Sensor Network Using Non-myopic Sensor Scheduling
Himanshu Shah Arizona State University, Tempe AZ and Darryl Morrell, Arizona State University Polytechnic Campus, Mesa AZ
Non-myopic sensor scheduling can be beneficial when tracking a target in a sensor network with constrained resources. The use of integer programming has been beneficial for myopic sensor scheduling. We extend its use to non-myopic sensor scheduling to perform cost efficient target tracking. In particular, we consider a scenario where a target is tracked in a distributed sensor network and formulate an integer non-linear program to schedule sensors over an M step planning horizon. We then solve this integer non-linear program using Outer Approximation to obtain the long term sensor schedule and show using Monte Carlo simulations how this scenario benefits from non-myopic sensor scheduling.

S8-6: The Predictive Enterprise: Transforming Business via Proactive Computing Technologies - Changing the Transportation Industry through Wireless Sensor Networks
Mary Murphy-Hoye Intel Corporation
After an extended focus on efficiency and cost-containment, businesses are beginning to investigate and invest in new ways to create competitive advantage using emerging technologies for a more Predictive Enterprise. This paper will explore wireless sensor networks experiments where companies are moving technology from the Data Center to the “edge” of the Enterprise. The purpose is to change the beat rate and flow of data about their business processes and key assets. This will create a more responsive infrastructure, a new form of real-time information assessment, and actuation capabilities at the edge based on changes captured in the environment. In order to learn about these technologies and the needs of large scale business environments, we have deployed wireless sensor network applications targeted at the transportation industry. The primary framework for this discussion is the Intelligent Container project, which focused on “networked” security and commercial benefits for the 20 million world-wide ship-based cargo containers. An overview is provided as well as results of real-life trials and the scale requirements identified. Insight into end-to-end architecture and mesh network protocol standards’ needs as well as deployment considerations (including power harvesting, rugged-ization, and self-configuring network management) will also be included.

S8-7: Adaptive Path Design of a Moving Radar
Martin Hurtado and Arye Nehorai Washington University in St. Louis, St. Louis MO
We consider the problem of designing the trajectory of a radar mounted on a moving platform. We develop an adaptive algorithm that, at each time step, optimally selects the radar path in response to the estimated and predicted target parameters to improve the tracking accuracy. We derive our approach under a framework of sequential Bayesian filtering. We apply a sequential Monte Carlo method (particle filter) to implement the filter for the case of nonlinear measurement models. We design the criterion for the path optimization based on the posterior Cramer-Rao bound.

S8-8: Real Time Human Activity Recognition Using Tri-Axial Accelerometers
Narayanan C. Krishnan, Dirk Colbry, Colin Juillard and Sethuraman Panchanathan Arizona State University, Tempe AZ
In this paper, we describe a real time system for detecting and recognizing lower body activities (walking, sitting, standing, running and lying down) using streaming data from tri-axial accelerometers. While there have been various attempts to solve this problem, what makes our system unique is that it uses a minimal set of sensors and works in real time. We have divided the system into three components: preprocessing, feature extraction and classification. This paper describes each component and addresses the issue of locating the sensors on a human body. We also discuss different elementary signal processing techniques that we experimented with to extract salient features from the sensory stream, bearing in mind the computation costs of each method. We used the AdaBoost algorithm built on decision stumps for classification, and our system is able to recognize each activity (walking, sitting, standing, running, lying down) with 95% accuracy.

S8-9: Progressive Estimation and Detection
Yingbo Hua and Yi Huang, University of California, Riverside CA
For large-scale multi-hop wireless sensor networks, progressive estimation and detection algorithms are developed. For progressive estimation, the classical principles of best linear unbiased estimation (BLUE) and linear minimum mean square error (LMMSE) estimation are applied. The difference and similarity between the two are revealed. For progressive detection, a tradeoff between detection performance and limited decision propagation is studied. The progressive estimation and detection algorithms are distributed and performed at the application layer of wireless sensor networks, which are fully adaptive to the operations at the networking and other lower layers that can be designed for high spectral efficiency.

S8-10: Subcomplemnatry Pulse Compression Sequences-India MST Radar Observations
K. Raja Rajeswari Andhra University, Visakhapatnam, India, Pathipati Srihari Dadi Institute of Engineering and Technology, Anakapalle, India, P. Rajesh Kumar Andhra University, Visakhapatnam, India and B. Visvasvara Rao Chaithanya Engineering College, Kakinada, India
Phase coding and linear frequency modulation are commonly used in radar systems for pulse compression to achieve high range resolution. In this paper, some experimental observations of Indian MST radar are incorporated. Range versus Signal to Noise Ratio (SNR) plots were drawn for different length complementary pairs, sets and subcomplemntary pairs. These sequences usually find applications in weather and mesosphere-stratosphere-troposphere (MST) radar systems to enhance range resolution performances.