Geolocation Analytics
F-35 Lightning II
Foreword
As an intern at Northrop Grumman, I am required to retain a Secret clearance. I verified with my supervisors that I may not share the details of the work I performed during my internship to prevent a compromise of security and proprietary information. More specifically, I can talk broadly about what I did, but I must refrain from referring to names, displaying sensitive images, sharing code, and quantifying results.
Overview
Geolocation technology enables aircraft to precisely identify the location of potential threats. This capability relies on radar technology, which analyzes radio frequency signals to determine the spatial coordinates of objects within the airspace. Geolocation plays a crucial role in enhancing a pilot’s situational awareness and response capabilities. As an intern, my responsibility was to further develop a piece of MATLAB software that analyzes the performance of various geolocation techniques.
The analytics software simulates the flight of an aircraft and the emittance of RF signals from a target at a distance to the aircraft. The flight path of the aircraft, the location of the target, and the signal being emitted are parameters that can be modified to test different scenarios. The first step in analyzing the performance of a geolocation technique during a specific flight scenario is taking measurements of the RF signal. Different geolocation techniques require different types of measurements. For instance, one technique may measure the gain differences between different modal patterns detected on a spiral antenna, while another may measure the phase differences between an array of antennas spaced-out on the wing of a plane. Based on the scenario parameters, the theoretical, true values of these measurements can be determined.
The second step was to add noise to these measurements to simulate the collection of real world data. The source of noise is another parameter that is varied and be as simple as specifying a signal-to-noise (SNR) ratio and as complex as importing recorded noise characterized on specific antennas. Since the sources and magnitudes of noise introduce uncertainty, a Monte Carlo simulation is performed with the number of trials being another user input. The outcome of this step is a set of noisy measurements, each representing a possible measurement that could have been taken in real life.
The last step involves calculating the emitter’s or target’s position. This calculation varies depending on which geolocation technique is being performed. A non-linear least squares fitting determines the signal’s origin from the set of measurements collected during the Monte Carlo trials. Then, a statistical analysis can be conducted on the set of predicted locations, yielding an uncertainty ellipse. The ellipse provides insight into the likelihood that the signal was emitted from the center of the predicted locations.
One of the challenges involved in developing this software was integrating the distinct needs of the various geolocation techniques. As previously mentioned, each technique used a different configuration of antennas and types of antennas. Each, also, measured different characteristics of the emitted signal, such as amplitude and phase, and subsequently, had a distinct method of deriving the emitter’s location, some involving look-up tables and others, an analytical solution to a matrix equation.
I approached this problem by modularizing the program so that steps in the process that differed depending on the technique used were hot-swappable. Sets of function specific to the techniques were defined. Moreover, each technique had their own measurement, corruption (introduction of noise), and solving (which calculates the emitter’s location from the measurements) function. In addition, the program was sectioned into two levels, top and low. The top-level script contained all the scenario parameters, hyperparameters (i.e. number of Monte Carlo trials), and various toggles to switch between options (including which geolocation technique and least squares fitting algorithm to use). The low-level portion handled the simulation and analytics, remaining separate from the top-level script.
At the end of my internship, I was able to cleanly integrate 4 geolocation techniques into the analytics software. The portion of code that ran the Monte Carlo simulations was parallelized, allowing for multiple trials to be run at once. Several test scenarios were run to ensure the proper functioning of the geolocation techniques. Plots of the predicted emitter locations, uncertainty ellipses, and convergence time series were generated and cross-validated against academic theories. The program was developed with adherence to proper version control practices (using Git), as well as documented with descriptive function headers and comments.
