PFANC uses CRNN to predict noise, beating reactive methods
A new proactive noise control method predicts the next noise frame using deep learning.
Active noise control (ANC) traditionally reacts to noise as it happens, causing a lag that worsens with rapidly changing sounds. A new paper from Nanyang Technological University researchers introduces PFANC, a proactive system that uses a convolutional recurrent neural network (CRNN) to analyze multiple past noise frames and predict the control filter for the very next frame. This shifts ANC from reactive to predictive, reducing lag significantly. The team provides a theoretical justification using high-order Markov chains to show why multiple frames improve prediction accuracy.
In numerical simulations with linear/logarithmic chirps and real-world dynamic noises, PFANC consistently outperforms the existing GFANC method and its variations. It also shows strong transferability—the model works well across different acoustic environments without retraining. While the paper is purely academic and hasn't been deployed in hardware yet, it offers a clear path toward faster, more responsive noise cancellation for headphones, cars, and industrial settings. The code and data are not yet released, but the arXiv paper provides full technical details.
- PFANC uses a CNN-RNN hybrid (CRNN) to predict the next-frame noise filter from multiple consecutive frames.
- It eliminates the tracking lag inherent in reactive methods by proactively anticipating noise changes.
- Outperforms GFANC in simulations with chirp signals and real dynamic noises, and generalizes across acoustic paths.
Why It Matters
Predictive ANC could drastically improve noise cancellation for rapidly changing environments like traffic or conversations.