Two-Stage Approach to Community Noise Prediction in Advanced Air Mobility Operations

Public acceptance of Advanced Air Mobility (AAM) operations depends critically on minimizing community noise impact. Yet, accurate noise prediction for eVTOL aircraft remains challenging due to their unique multi-rotor configurations and operating conditions. This study introduces a modular two-stage framework for predicting noise levels of AAM, focusing on a coaxial multicopter X8 configuration. In the first stage, flight state vectors are extracted from measurement data, as these parameters directly influence noise output through changes in propulsion characteristics, aerodynamic loading, and noise directivity patterns. These flight states are given as input to a second-stage CNN-BiLSTM-Attention network for A-weighted sound pressure level prediction. Unlike physical approaches requiring explicit noise source modeling, this framework directly maps flight states to overall acoustic output. The modular architecture allows the first stage to be implemented either as a measurement-based system (as demonstrated in this work) or as a physics-based dynamic model when flight data are unavailable. This framework enables the evaluation of noise impacts for different flight trajectories. Results demonstrate predictive performance with an R^2 value of 92.11% and MAE of 1.902 dB(A).