Determine whether overfitting occurred in the high‑order spherical harmonics radiation-pattern model

Determine whether overfitting occurs when learning the decoupled antenna radiation patterns of two quadrotor UAVs using a spherical harmonics series expansion of order 28 (784 parameters) fitted via ridge regression within the Friis transmission equation framework to received-signal-strength data collected during the proposed synchronized calibration trajectory.

Background

The paper proposes learning and decoupling the antenna radiation patterns of two heterogeneous quadrotor UAVs directly from in‑flight received-signal-strength data using a Friis-based residual minimization. Among several models, a spherical harmonics representation is evaluated, with increasing polynomial order enabling finer detail.

In the high-complexity setting (Legendre order 28, 784 parameters), the authors observe strong performance by Akaike information criterion and stable cross‑validation topology but raise concern about potential overfitting and possible sampling bias introduced by the calibration trajectory. They explicitly note uncertainty about whether overfitting occurred and defer a full analysis to future work.