Session: 06-11-01: Applications of Artificial Intelligence/Machine Learning in Aerospace Engineering

Paper Number: 165409

Acoustic-Based Navigation Using Trilateration and Neural Network-Enhanced Tdoa Adjustment 

Accurate positioning using acoustic signals is a fundamental challenge in various aircraft and general navigation applications, including autonomous systems, underwater localization, military applications and surveillance. This work presents a novel approach to source localization based on trilateration using multiple microphones, enhanced by a neural network for Time Difference of Arrival (TDOA) correction.

The proposed system consists of an array of five spatially distributed microphones that capture acoustic signals emitted by a sound source. By measuring the time differences at which the signal reaches each microphone, the system estimates the relative distances between the source and the receivers. Using these TDOA values, a trilateration algorithm determines the source’s position in a two-dimensional or three-dimensional space. However, TDOA measurements are susceptible to errors due to environmental noise, signal reflections, and propagation anomalies. These inaccuracies degrade localization performance, making traditional TDOA-based methods unreliable in complex acoustic environments.

To mitigate these issues, we incorporate a neural network to refine the TDOA estimates. The neural model is trained on a dataset composed of both simulated and real-world acoustic signals, where ground-truth positions and corresponding TDOA values are known. The network learns to identify and compensate for systematic errors introduced by multipath effects, microphone synchronization discrepancies, and background noise. By correcting these errors before the trilateration step, the system significantly improves localization accuracy compared to conventional approaches.

The neural network architecture consists of multiple fully connected layers with nonlinear activation functions, allowing it to learn complex mappings between noisy TDOA inputs and corrected values. The model is trained using supervised learning, where the loss function minimizes the discrepancy between predicted and actual TDOA values. Additionally, data augmentation techniques are employed to improve the model’s generalization to unseen environments, ensuring robustness across diverse operational conditions.

Experimental validation is conducted in controlled and real-world scenarios, demonstrating the effectiveness of the proposed approach. Results indicate that the neural-enhanced TDOA correction reduces localization errors by a significant margin, outperforming traditional statistical filtering techniques. Moreover, the system exhibits stable performance even in environments with high levels of reverberation and noise, where classical TDOA methods typically struggle.

Potential applications of this work include indoor navigation for autonomous robots, underwater vehicle localization where GPS is unavailable, and aircraft surveillance systems that rely on passive acoustic monitoring, including some military applications. The integration of machine learning with classical trilateration techniques represents a promising direction for improving the reliability of acoustic-based positioning systems.

In conclusion, this research presents a hybrid approach that combines physics-based trilateration with data-driven neural network correction for TDOA-based source localization. By leveraging deep learning to refine time delay estimates, the proposed method achieves superior accuracy and robustness in challenging acoustic environments. Future work will explore alternative neural architectures, real-time processing capabilities, and further extensions to dynamic source tracking.

Presenting Author: Raul De Celis Rey Juan Carlos University

Presenting Author Biography: Raúl de Celis gained experience in GN&C for aerial platforms (2010-2016) in aerospace companies, including INTA, Airbus (Altran), and Escribano. Since 2016, he has been a full-time professor at Universidad Rey Juan Carlos, focusing on GN&C, air transport, and AI applications. He has published over 20 papers in top-tier journals, participated in competitive research projects, and collaborated with the aerospace industry.

Authors:

Raul De Celis Rey Juan Carlos University