Writing · Sole-authored, Cal Poly Pomona · 2024
Analyzing the Resilience of UAV Communication Systems
How much noise a flight control system tolerates before operational integrity goes, and why machine learning is a better way to find that threshold than flying it and seeing.
UAVs have moved from military use into agriculture, surveillance, logistics and environmental monitoring, and every one of those roles puts a flight controller in a position where losing signal integrity is a safety event rather than an inconvenience.
A flight controller is an embedded real-time system. It fuses gyroscopes, accelerometers and GPS to estimate orientation, altitude and velocity, and drives motors and servos from that estimate. Everything downstream of a corrupted input is a corrupted control output, which is what makes signal integrity a flight-safety property rather than a communications nicety.
The interference does not only come from outside. Competing signals degrade the command link into control latency or outright loss, but power supply fluctuation and electromagnetic interference generated by the vehicle's own electronics corrupt sensor accuracy and actuator response from inside the airframe. Both paths end in the same place.
The question the paper sets is quantitative: how much noise can these systems absorb before integrity is compromised? That threshold is what determines whether a design is safe, and it is normally established anecdotally — fly it and see what happens.
The proposed alternative uses machine learning as an exploration tool. By generating scenarios with varied noise and interference and letting ML algorithms characterize how each disturbance moves system performance, the noise-tolerance boundary can be mapped rather than sampled at a handful of points. That both tests resilience under adversarial conditions and produces the knowledge needed to design against them.