All projects
Completed 2015 – 2017 Flight Dynamics & Control / Hybrid Systems Lab, Purdue University

State Estimation under Communication Uncertainties

Estimators that keep tracking accurate when sensor data is lost in transit, including losses that cluster in predictable regions, and that let a network of sensors agree on one estimate. Demonstrated on aircraft tracking.

Collaborators: Suraj Jayaprakash Nandiganahalli, Inseok Hwang, Raj Deshmukh, Cheolhyeon Kwon

Paper M.S. thesis slides
Left: an aircraft trajectory passing through three jammer regions. Top right: a two-mode measurement model that switches off inside those regions. Bottom right: the estimator predicting through missed measurements and updating when they arrive.
An aircraft flies past radio jammers (left). Losing measurements inside the jammed regions is modeled as a switching measurement model (top right), and the estimator predicts through the gaps and updates whenever a measurement gets through (bottom right). Adapted from Kalman filtering with state-dependent packet losses, IET Control Theory & Applications, 2019.

Problem

Air traffic surveillance, like many tracking systems, relies on measurements sent over imperfect communication links. Packets get dropped, sometimes at random and sometimes in predictable places, such as regions with poor coverage or radio jamming. A standard Kalman filter assumes every measurement arrives, so its estimates degrade exactly where losses cluster. When many sensors track the same aircraft, each sees only part of the picture and has to agree with its neighbours over a network whose connections keep changing.

Approach

The first line of work models packet arrival explicitly. For random losses typical of real communication channels, and for losses that depend on where the tracked object is, the estimator keeps the familiar Kalman filter structure but updates its uncertainty to reflect the chance that each measurement arrives, using prior knowledge of where losses are likely.

The second line of work spreads estimation across a sensor network. Each sensor runs a local estimator for a system that switches between operating modes and shares its estimates with neighbouring sensors, so the whole network reaches agreement even as its connections change over time.

Three stacked sensor nodes. Each runs two mode-matched filters, mixes them, and updates and combines its estimate, while exchanging initial estimates and edge-error covariances with the nodes above and below.

How one sensor (center) updates its estimate: it runs one filter per operating mode, mixes and combines them, and trades estimates and error covariances with neighbouring sensors (above and below). Adapted from Distributed state estimation for a stochastic linear hybrid system over a sensor network, IET Control Theory & Applications, 2018.

Outcome

  • More accurate tracking under patchy communication. On an aircraft tracking example with location-dependent packet losses, the new filter outperforms the baseline packet-loss filter at similar computational cost.
  • Distributed estimation on changing networks. The consensus-based estimator outperforms existing hybrid estimators at comparable per-sensor complexity, while allowing the network’s connections to change.
  • Published. Two IET Control Theory & Applications papers (2018, 2019) and an IEEE CDC 2017 paper. The packet-loss work formed the basis of an M.S. thesis at Purdue, and the distributed estimation work was supported by the NSF (CMMI 1335084).