Artillery Detection & Localization using a Two-Microphone Array on a Moving Platform
Authors
Schultz, Kasper Holmbo ; Løvendal, Jacob
Term
4. semester
Education
Publication year
2025
Submitted on
2025-06-04
Pages
126
Abstract
This project examines whether artillery can be detected and located using a mobile two-microphone array. Assuming a known muzzle blast, a Neyman–Pearson (NP) detector is implemented, a statistical decision rule that aims to maximize detection for a fixed false-alarm rate. The localization method combines Time Difference of Arrival (TDoA, the difference in arrival time between the microphones) and Frequency Difference of Arrival (FDoA, differences in frequency content at the microphones). A simulated acoustic environment was developed for testing, including propagation of both the muzzle blast and the shockwave. The NP detector performed reliably across scenarios, but performance dropped above 95 dB SPL noise and was notably affected by the shockwave. TDoA estimated with a matched filter showed a negative bias (systematic underestimation) but had narrower uncertainty intervals than GCC-PHAT. FDoA estimation, based on matching frequency-response amplitudes, was highly error-sensitive and did not yield reliable localization. Although the subsequent intersection solver using the Levenberg–Marquardt (LM) algorithm worked as intended, overall accuracy was constrained by the limitations of the FDoA approach. Final tests showed large errors at long ranges and wide angles, with mean errors around 6 km at a 15 km range. At 1 km, accuracy improved, but overall performance remains insufficient for real-world use. The study concludes that a two-microphone TDoA/FDoA system is promising in principle, but the current implementation is not ready for practical deployment.
Dette projekt undersøger, om artilleri kan opdages og lokaliseres med et mobilt array med to mikrofoner. Under antagelse af en kendt mundingsknald implementeres en Neyman–Pearson (NP) detektor, en statistisk metode der søger at maksimere detektionsraten for en fast falsk-alarmrate. Lokaliseringsmetoden kombinerer Time Difference of Arrival (TDoA, forskellen i ankomsttid mellem de to mikrofoner) og Frequency Difference of Arrival (FDoA, forskelle i frekvensindhold ved mikrofonerne). Et simuleret akustisk miljø blev udviklet til test, inklusive udbredelse af både mundingsknald og chokbølge. NP-detektoren var pålidelig på tværs af scenarier, men ydelsen faldt ved støjniveauer over 95 dB SPL og blev mærkbart påvirket af chokbølgen. TDoA estimeret med et matched filter viste en negativ bias (systematisk undervurdering), men gav snævrere usikkerhedsintervaller end GCC-PHAT. FDoA-estimatet, baseret på matching af amplituden i frekvensrespons, var meget følsomt over for fejl og gav derfor ikke pålidelig lokalisering. Selvom den efterfølgende skæringsløser med Levenberg–Marquardt (LM) algoritmen fungerede som forventet, blev den samlede nøjagtighed begrænset af FDoA-tilgangen. Sluttelige tests viste store fejl på lange afstande og brede vinkler, med gennemsnitlige fejl omkring 6 km ved 15 km afstand. Ved 1 km blev nøjagtigheden bedre, men samlet set er systemet utilstrækkeligt til praktisk brug. Konklusionen er, at et to-mikrofon TDoA/FDoA-system rummer konceptuelt potentiale, men den aktuelle implementering er ikke egnet til brug i virkeligheden.
[This apstract has been rewritten with the help of AI based on the project's original abstract]