State Estimation Adaptable to Cyberattack Using a Hardware Programmable Bank of Kalman Filters

Abstract

Sensor-estimator systems provide critical information on the state of cyber-physical plants. Often, these units operate in an environment of constrained computational resources. This condition makes them vulnerable to cyberattacks that aim especially to degrade their processing capability and effectively incapacitate them. In the event that computational nodes are lost, an approach to adapt the estimator’s algorithm and reprogram the adapted form on the surviving hardware is presented. To prepare the sensor-estimator system for degradation, the following co-design steps are developed: 1) the estimation algorithm, a bank of Kalman filters (KFs), is distributed so that multiple elemental filters are implemented on a collection of field-programmable gate arrays (FPGAs) and 2) the matrix operations of the conventional KF are programmed on the FPGAs using Faddeeva’s elimination. After the attack, adaptation of the filter bank is realized by leveraging dynamic partial reconfiguration (DPR) of the surviving FPGAs. A high-authority agent monitors the likelihood of all elemental filters, a measure of which filters currently provide the best estimates, and replaces the least likely elements of the bank with the most likely ones. The latter are loaded onto the freed-up fabric of the remaining FPGAs, while these units are running other elemental filters in order to process sensor data without interruption. We have demonstrated their method on a prototype system that uses a radar sensor to estimate the kinematics of a maneuvering unmanned surface vehicle (USV).