Publications

This paper presents a description of the MIT Lincoln Laboratory language recognition system submitted to the NIST 2009 Language Recognition Evaluation (LRE). This system consists of a fusion of three core recognizers, two based on spectral similarity and one based on tokenization. The 2009 LRE differed from previous ones in that test data included narrowband segments from worldwide Voice of America broadcasts as well as conventional recorded conversational telephone speech. Results are presented for the 23-language closed-set and open-set detection tasks at the 30, 10, and 3 second durations along with a discussion of the language-pair task. On the 30 second 23-language closed set detection task, the system achieved a 1.64 average error rate.

Graphs are canonical examples of high-dimensional non-Euclidean data sets, and are emerging as a common data structure in many fields. While there are many algorithms to analyze such data, a signal processing theory for evaluating these techniques akin to detection and estimation in the classical Euclidean setting remains to be developed. In this paper we show the conceptual advantages gained by formulating graph analysis problems in a signal processing framework by way of a practical example: detection of a subgraph embedded in a background graph. We describe an approach based on detection theory and provide empirical results indicating that the test statistic proposed has reasonable power to detect dense subgraphs in large random graphs.

We propose supervised and unsupervised learning algorithms to extract dialect discriminating phonetic rules and use these rules to adapt biphones to identify dialects. Despite many challenges (e.g., sub-dialect issues and no word transcriptions), we discovered dialect discriminating biphones compatible with the linguistic literature, while outperforming a baseline monophone system by 7.5% (relative). Our proposed dialect discriminating biphone system achieves similar performance to a baseline all-biphone system despite using 25% fewer biphone models. In addition, our system complements PRLM (Phone Recognition followed by Language Modeling), verified by obtaining relative gains of 15-29% when fused with PRLM. Our work is an encouraging first step towards a linguistically-informative dialect recognition system, with potential applications in forensic phonetics, accent training, and language learning.

The long-term objectives for the NextGen Weather Processor (NWP) include consolidation of today's multiple weather systems, incorporation of recent and emerging Federal Aviation Administration (FAA) infrastructure (Federal Telecommunications Infrastructure (FTI), System Wide Information Management (SWIM), NextGen Network-Enabled Weather (NNEW)), leveraging National Oceanic and Atmospheric Administraiton (NOAA) and/or commercial weather resources, and providing a solid development and runn-time platform for advanced aviation weather capabilities. These objectives are to be achieved in a staged fashion, ideally with new components coming on-line in time to replace existing capabilities prior to their end-of-life dates. As part of NWP Segment 1, a number of alternative implementations for the NWP as it might exist in the 2013 time frame have been proposed. This report examines the alternatives form a top-down technical perspective, assessing how well each maps to a high-level NWP architecture consistent with the long-term NextGen information sharing vision. Tehcnical challenges and opportunities for weather product improvements associated with each alternative are discussed. Additional alternatives consistent with the high-level NWP architecture, as well as a number of suggested follow-on analysis efforts are also presented.

Airspace encounter models, providing a statistical representation of geometries and aircraft behavior during a close encounter, are required to estimate the safety and robustness of collision avoidance systems. Prior encounter models, developed to certify the Traffic Alert and Collision Avoidance System, have been limited in their ability to capture important characteristics of encounters as revealed by recorded surveillance data, do not capture the current mix of aircraft types or noncooperative aircraft, and do not represent more recent airspace procedures. This paper describes a methodology for encounter model construction based on a Bayesian statistical framework connected to an extensive set of national radar data. In addition, this paper provides examples of using several such high-fidelity models to evaluate the safety of collision avoidance systems for manned and unmanned aircraft.

Ultralow-power electronics will expand the technological capability of handheld and wireless devices by dramatically improving battery life and portability. In addition to innovative low-power design techniques, a complementary process technology is required to enable the highest performance devices possible while maintaining extremely low power consumption. Transistors optimized for subthreshold operation at 0.3 V may achieve a 97% reduction in switching energy compared to conventional transistors. The process technology described in this article takes advantage of the capacitance and performance benefits of thin-body silicon-oninsulator devices, combined with a workfunction engineered mid-gap metal gate.

The Traffic Alert and Collision Avoidance System (TCAS) is designed to reduce the risk of mid-air collisions by providing resolution advisories to pilots. The current version of the collision avoidance logic was hand-crafted over the course of many years and contains many parameters that have been tuned to varying extents and heuristic rules whose justification has been lost. Further development of the TCAS system is required to make the system compatible with next generation air traffic control procedures and surveillance systems that will reduce separation between aircraft. This report presents a decision-theoretic approach to optimizing the TCAS logic using probabilistic models of aircraft behavior and a cost metric that balances the cost of alerting with the cost of collision. Such an approach ahs the potential for meeting or exceeding the current safety level while lowering the false alert rate and simplifing the process of re-optimizing the logic in response to changes in the airspace and sensor capabilities.

When operationally significant weather affects the National Airspace System (NAS) a Severe Weather Avoidance Program (SWAP) is initiated. Each SWAP event is a unique mix of demand, weather conditions, traffic flow management (TFM) initiatives and traffic movement. Following a SWAP, the day's events are reviewed and the TFM initiatives used are evaluated to understand their impact on the traffic flows, benefits, and disadvantages. These analyses require an accurate representation of the conditions during SWAP and objective, data-driven metrics to determine the effectiveness of the implemented TFM initiatives, and to identify opportunities for improved decision making in future events. As part of the ongoing development and evaluation of the Route Availability Planning Tool (RAPT), a departure management decision support prototype currently deployed in New York, several detailed metrics were developed to streamline these analyses. This paper focuses on metrics that address the most significant concern regarding departure flows from New York airports: the timely reopening of departure routes that have been closed due to convective weather impacts. These metrics are derived from two datasets: flight tracks from the Enhanced Traffic Management System (ETMS) to monitor the flight traffic, and route blockage from the Route Availability Planning Tool (RAPT) to determine the impact of weather on routes. RAPT automatically identifies Post-Impact-GREENs (PIGs) - the period of time when routes are clear ('GREEN') after being blocked by convective weather. Identifying PIGs early is a key element of the RAPT concept of operations, which enables traffic managers to restart traffic flow sooner along these routes, alleviating backed up ground conditions and reducing delay times for waiting flights. An automated system, that correlates PIGs identified by RAPT with departure traffic flows, calculates both the time from the appearance of each PIG until the first departure along the PIG route, and the departure rate on the route during the PIG period. Short times to first departure and high departure rates during PIGs indicate efficient departure management during SWAP. Arrival aircraft deviating into departure airspace is also managed by closing the departure route until the danger from incurring flights has passed. Arrival incursions are sometimes recorded in the National Traffic Management Log (NTML), but the extent to which the deviations occur is unmeasured. Lack of details regarding deviations limits evaluation of implemented responses and alternative actions. New algorithms comparing clear weather vs. SWAP traffic flows enables the locations and durations of incursions to be identified. Exact figures detailing incursions allows for thorough review as well as recognition of areas of frequent incursions and the potential for developing a targeted response for like situations. Full flight tracks of arriving and departing flights provide significant insight into the status of the NAS. During SWAP when the airspace capacity is decreased and airport operation rates are limited, airborne aircraft by protocol receive priority. Arrival numbers can completely dominate operations at these times both in the air and on the ground, draining the resources available for departures in particular flows or for an entire region. To convey cases where departure infrequency results from these conditions, arrival and departure counts grouped according to direction of travel are calculated on an hourly basis. Results from the automated analysis are made available on the RAPT Evaluation and Post Event Analysis Tool (REPEAT) website by 7AM ET for the FAA Northeast tactical review teleconferences, and are being tracked over the convective season for further analysis of operational performance. This paper will present the techniques used in the automated system and initial results from the analysis of operational data.

Convective storms are responsible for causing a predominant number of delays in the summer when air traffic peaks. Several studies have shown that there is a critical need for timely, reliable, and high-quality forecasts of precipitation and echo tops with forecast time horizons of up to 12 hours in order to predict airspace capacity (Robinson et al. 2008; Evans et al. 2006; FAA 2007). While a variety of convective weather forecast systems are available to strategic planners of the National Airspace System (NAS), these products don't meet Air Traffic Management (ATM) needs fully. In addition, a multitude of forecast products increases the potential of having conflicting information available in the planning process, which can cause situational awareness problems between the operational facilities, ultimately leading to more potential delays and perhaps safety problems.

The effective management of convective weather in congested air space requires decision support tools that can translate weather information available to air traffic managers into anticipated impact on air traffic operations. The Convective Weather Avoidance Model (CWAM) has been under development at Lincoln Lab under sponsorship of NASA to develop a correlation between pilot behavior and observable weather parameters. To date, the observable weather parameters have been the Corridor Integrated Weather System (CIWS) high resolution Vertically Integrated Liquid (VIL) precipitation map and the CIWS Echo Top product. The CWAM was dependent upon a crude model to define pilot deviations based upon finding weather encounters and then comparing the distance between the planned and actual flight trajectories. Due to a large number of false deviations from this crude model, a significant amount of hand editing was required to use the database. This paper will focus on two areas of work to improve the performance of the enroute convective weather avoidance models. First, an improved automated algorithm to detect weather-related deviations that significantly reduces the percentage of false deviation detections will be presented. This new model includes additional information on each deviation, including the location the decision was made to deviate. The additional information extracted from this algorithm can be used to evaluate the conditions at the decision time which may impact the severity of weather pilots are willing to penetrate. The new deviation detection algorithm has also reduced the amount of hand editing required by removing short cuts taken to reduce the flight time, deviations that occur well past the decision time, and non-weather related reroutes. The second focus of this paper will be the comparison of three different convective weather avoidance models that have been proposed, based upon the analysis of an expanded database of flight deviations. Six weather impact days from 2007 and 2008 have been added to the existing case set from 2006, tripling the number of flight trajectories that can be used in validating the models. In addition to validating the existing CWAM, we will look at additional parameters that may improve the performance of the CWAM.