JSJason E. Summers

Jason E. Summers

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Jason E. Summers has a total of 32 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Deep reinforcement learning for cognitive sonar

    by Jason E. Summers, Jason M. Trader, Charles F. Gaumond, Johnny L. Chen

    Current developments in cognitive sonar have leveraged explicit semantic representations such as ontologies to develop cognitive agents that enable adaptation of sonar settings to optimize behaviors such as waveform selection during active tracking. However, such cognitive systems based on explicit knowledge representations can be time-and-labor intensive to update in operation and limited in ultimate performance. In other applications, such as computer Go, breakthrough performance was achieved by going beyond learning from all prior games and actions of experts to allowing the algorithm to compete with itself to develop and learn from the outcome of new approaches. This hybrid approach of learning from experts and then learning via self-competition is vital for sonar applications such as active antisubmarine warfare (ASW) because ground-truthed performance data are sparse and may not display optimal solutions. Here, we discuss our application of reinforcement learning to active ASW in a simulated environment based on computational-acoustics models, including an assessment of the goal states and performance metrics we have used. The fidelity of the simulated environment is discussed in terms of the interaction with reinforcement learning and the impact on generalization from learning in simulated environments to application in real environments. [Work supported by ARiA IR&D.]


  • Dimension-reduced adaptive pulse compression with model error compensation

    by Jonathan Botts, Jason E. Summers, Frank W. Bentrem, Charles F. Gaumond

    In active acoustic imaging, range sidelobes from competing clutter and interferers in acoustically complex environments may mask signals of interest. In order to unmask these signals, increase range resolution, and decorrelate the time-series we introduce dimension-reduced adaptive pulse compression (APC) with model-error compensation that operates after a conventional matched filter. Structured covariance estimates minimize the required sample support while dimension reduction occurs naturally and a priori after the matched filter. We compensate for predictable signal distortions such as Doppler through both covariance matrix tapers and truncation of the covariance matrix. Model-based APC is computationally intensive, scaling with the third or fourth power with replica length, so algorithmic modifications are introduced to reduce computation by orders of magnitude. We evaluate the algorithm's robustness and computational feasibility on simulated and archived data. [Portions of this material are based upon work supported by the Naval Sea Systems Command.]

  • Design of a signal normalizer for high-clutter active-sonar detection

    by Frank W. Bentrem, Jonathan Botts, Jason E. Summers

    We present normalizer design considerations for detection of targets in a high-clutter environment. Several classes of constant-false-alarm-rate (CFAR) normalizers have conventionally been used to bring out a target signal from the background noise and reverberation in active sonar signal processing. The special challenges of detection in high-clutter environments are outlined, and the normalizer classes are compared for their suitability of use in these environments. We consider statistical and multiple-look normalizers and give emphasis to the class of split-window mean-power normalizers and implementations that preserve target signal-to-noise ratio (SNR) and highlight structure while mitigating artifacts that are sometimes produced near clutter edges. Statistical comparisons and specific examples are presented, and we propose an efficient CFAR normalizer for high performance in regions of high clutter.


  • Combining interactive simulation with scaffolded visualization in ESAIL to teach fundamental concepts of acoustical oceanography

    by Jason E. Summers, Mark A. Ross, Zachary Walker, Paul Longerbeam, Daniel Redmond

    To understand and interpret active-sonar returns and effectively employ active sonar systems for purposes such as antisubmarine warfare (ASW), operators must understand fundamental concepts of acoustical oceanography that govern how sonar interacts with the environment. Experts achieve high levels of performance in analysis and employment not by solving acoustic-propagation equations, but rather by employing simplified mental models that characterize the functional relationships governing how the environment affects acoustic propagation. Thus, a key goal in educating operators is helping them to develop and use such mental models. Here, we describe the development and employment of ESAIL, the Environment for Surface ASW Interactive Learning, a learning platform that combines interactive simulation with scaffolded visualization to facilitate instructors and students in conducting and assessing the outcome of “what if…?” experiments. In ESAIL, users are scaffolded in performing “what if…?” experimentation by tools that automatically classify user-modified acoustical environments as well as predictive acoustical models. This capability for direct exploration of cause-and-effect relationships with scaffolding to help develop expectations facilitates operators actively learning the mental models and mental-simulation capabilities experts use to interpret displays and reason about the underlying tactical scenario to achieve more optimal employment. [Work supported by the Naval Sea Systems Command.]

  • In-situ estimation of transmission loss based on learned dictionaries and sparse reconstruction

    by Jonathan Botts, Mark A. Ross, Jason E. Summers, Charles F. Gaumond

    Transmission loss is an important and notoriously difficult quantity to estimate in real underwater environments. Used to evaluate detections and threshold exceedances, estimates of transmission loss may be affected by approximate knowledge of weather conditions, bathymetry, bottom properties, and sound-speed profile—among other sources of environmental uncertainty. For applications in multistatic antisubmarine warfare, a relatively small number of in-situ measurements are available. In this work, in-situ measurements of transmission loss are used to reconstruct transmission loss fields using matching pursuit over a learned dictionary of transmission-loss fields for a given operational area. The dictionary is trained on simulated transmission-loss fields with inputs derived from historical data and plausible variations of environmental parameters. A principal challenge is constraining the problem in such a way that fields may be meaningfully reconstructed given a small set of measured data. The feasibility of reconstruction is evaluated for both full-field and depth-independent transmission-loss fields. Plausibility of practical application is evaluated with respect to realistic sampling conditions. [Portions of this material are based upon work supported by the Naval Air Systems Command.]

  • Evaluating simulations of acoustic environments by learning time-frequency kernels that optimally separate classes of simulated signals

    by Jason E. Summers, Jonathan Botts, Charles F. Gaumond

    Previously, the authors have presented techniques for classifying the aural accuracy of computed impulse responses for room-acoustic simulation and for sonar-operator training using two-dimensional distance metrics operating on time-frequency representations (TFRs) calculated using computational auditory models and by a generalized TFR formed by an affine radial-Gaussian correlation kernel operating on the Wigner-Ville distribution (i.e., Cohen’s class), with kernel parameters optimized for binary classification [J. E. Summers et al., J. Acoust. Soc. Am. 115, 2514 (A) (2004); J. E. Summers et al., Proc. 19th Int. Cong. Acoust., 8, 4591-4596, RBA-05-015-IP (2007)]. Here, we revisit the use of classification-optimized TFR for distinguishing between measured and simulated impulse responses and between impulse responses simulated by different computational approaches in the domains of room-acoustics and submerged target scattering. The learned parameters of the optimized TFR kernels are investigated with respect to the information they provide about the nature of the differences between impulse-response classes, particularly as they indicate temporal and spectral scales of those phenomena that particular computational models fail to represent. [Work supported by ARiA IR&D and the Office of Naval Research.]

  • Model-error-compensated reiterative adaptive beamforming on a physically shaded cylindrical array

    by Jonathan Botts, Jason E. Summers, Charles F. Gaumond

    The authors formulate and compare beamspace and element-space formulations of Reiterative Superresolution (RISR) [S. D. Blunt et al., IEEE Trans. Aero. & Elec. Sys. 47, 332-346 (2011)] on a physically shaded cylindrical array. The beamspace adaptive approach minimizes computational load while use of a model-based structured covariance estimates enables adaptive beamforming with Doppler sensitive waveforms having little sample support. While structured covariance estimates reduce sample-support requirements and facilitate stable inversion, they also introduce degradation due to model-mismatch errors. In space-time adaptive processing (STAP), covariance matrix tapers (CMT) have been used to counter the effects of off-axis arrivals and other forms of model mismatch [J. R. Guerci, Space-Time Adaptive Processing for Radar (Artech, 2014)]. In adaptive pulse compression, CMT have been used to increase range sidelobe suppression in the presence of Doppler [T. Cuprak, M.S. Thesis (2013)]. In this work, we have formulated CMT to address model mismatch by accounting for interbeam arrivals. We also consider the degree to which CMT can be used to address model-mismatch in RISR resulting from the physical shading effects on the directionality of elements in a large cylindrical array. [Portions of this material are based upon work supported by the Naval Sea Systems Command.]

  • Deep neural networks for learning classification features and generative models from synthetic aperture sonar big data

    by Johnny L. Chen, Jason E. Summers

    Autonomous synthetic aperture sonar (SAS) imaging by unmanned underwater vehicles (UUVs) provides an abundance of high-resolution acoustic imagery useful for studying the seafloor and identifying targets of interest (e.g., unexploded ordnance or mines). Unaided manual processing is cumbersome as the amount of data gathered by UUVs can be enormous. Computer-vision and machine-learning techniques have helped to automate classification and object-recognition tasks, but often rely on hand-built features that fail to generalize. Deep-learning algorithms facilitated by emergence of graphics-processing unit (GPU) hardware and highly optimized neural-network implementations have recently enabled great improvements in computer vision. Autoencoders allow for deep unsupervised learning of features based on a reconstruction objective. Here, we present unsupervised feature learning applied to seafloor classification of SAS images. Deep architectures are also capable generative models. We illustrate this with generative networks that are capable of generating realistic SAS images of different seafloor bottom types. Deep models allow us to construct algorithms that learn hierarchical and higher-order SAS features, which promise to improve automatic target recognition (ATR) and aid operators in processing the large data volumes generated by UUV based SAS imaging. [Work supported by the Office of Naval Research.]


  • Compressive beamforming on a cylindrical array

    by Charles F. Gaumond, Jonathan Botts, Jason E. Summers

    Previously, we presented subarray compressive beamforming for Doppler sensitive midfrequency active sonar on a cylindrical array, focusing on practical application for mitigating Doppler spreading caused by sidelobes and the width of the main lobe [J. Acoust. Soc. Am. 139, 2050 (2016)]. Here, we describe ongoing work focusing on theoretical challenges of adapting compressive beamforming to a cylindrical array. Unlike linear arrays, for example, the optimal sampling for the sound field incident on a cylinder array is not evident. Formal design guidelines for compressive beamforming on a cylindrical array are discussed, guided by the statistical restricted isometry property (StRIP). Prior work has demonstrated subarray compressive beamforming by combining sparse estimates from arbitrarily chosen subarrays. Here, a formal approach is outlined in which StRIP is used to specify minimal subarrays that can individually reconstruct a sound field of a given sparsity, using a new strategy for combining individual estimates. Finally, differences between the fundamental assumptions and resulting limitations of compressive, conventional, and adaptive beamforming techniques are discussed for the cylindrical array geometry and continuous-wave (CW) waveforms of interest, as demonstrated with high-fidelity simulated data. [Work supported by a NAVSEA Phase I SBIR award.]

  • Adaptive pulse compression for clutter mitigation in midfrequency active sonar

    by Jason E. Summers, Jonthan Botts, Charles F. Gaumond, Ian Cummings

    Midfrequency active sonar can achieve a combination of fine range resolution and good signal-to-noise ratio (SNR) by transmitting low-crest-factor frequency-modulated (FM) waveforms that are pulse compressed by match filtering. While this linear filter optimizes SNR for signals in additive Gaussian noise, it has large-amplitude range sidelobes that can allow strong sources of clutter, such as fish schools, to mask weaker nearby targets. To address range sidelobes we consider the reiterative minimum mean-square error (RMMSE) adaptive-pulse-compression algorithm [Blunt and Gerlach, Proc. IEEE Intl. Radar Conf., Sept. 2003]. RMMSE performance is known to degrade for small Doppler shifts in the received waveform (e.g., due to clutter internal motion), which, depending on the form of the FM sweep, can be partially mitigated by covariance-matrix tapers [Cuprak, M.S. Thesis, George Mason University, 2013]. We note that RMMSE is analogous to the minimum-variance distortionless-response (MVDR) beamformer: for each range cell it steers nulls the location of strong returns at nearby range samples. Motivated by this similarity and expectation of target sparsity in range, we extend the compressive beamforming approach developed in prior work to learning a compressive-sensing match filter. Preliminary results from this work are discussed. [Work supported by a NAVSEA Phase I SBIR award.]