CNN-Transformer improves angle detection accuracy below one degree
CNN-Transformer with Sigmoid Label for High-Resolution Sub-Degree DOA Estimation
Artificial Intelligence
Summary
Estimating the direction from which sounds or signals come is often done by dividing the space into fixed steps, but this limits how precisely you can tell the direction. The authors created a new label style using sigmoid functions that better captures slight shifts in angle. By training a neural network combining CNN and Transformer components with this label, they achieve more precise direction estimates than before, going below one degree of error. Their method works better than previous labeling styles on this task.
What this means in practice
- •For acoustic engineers: Use improved angle estimation for sound source localization with sub-degree accuracy in audio systems and environments.
- •For wireless communication teams: Enhance antenna array direction estimation for better beamforming and signal reception precision using the proposed CNN-Transformer method.
Authors
Lifeng Wang, Xintong Li, Aifei Liu
Abstract
This paper addresses the grid quantization error of deep-learning (DL)-based direction-of-arrival (DOA) estimation. Classification- based DOA methods discretize the continuous angular space into a grid, so that with a peak-based read-out, the accuracy and resolution are limited by the grid step and by the width of the label peak. To address this issue, we propose a sigmoid label that encodes the sources' DOAs as the sum of angle-shifted monotonic sigmoid functions. Differentiating the output of a CNN-Transformer network trained with the sigmoid label produces sharp spectral peaks at the source DOAs. Combined with a moderately refined grid step of 0.25 degrees, the CNN-Transformer with the proposed sigmoid label attains sub-degree accuracy and a 1-degree resolution, breaking the limit of the grid quantization error. Under the unified CNN-Transformer backbone, we compare the proposed Sigmoid label with the Gaussian label, the 0-1 hard label, and the binary cross-entropy (BCE) baseline. Numerical results demonstrate that the Sigmoid label attains the highest accuracy and superior resolution compared to other labels.