A multi-channel active noise control system using deep learning-based method to estimate secondary path and normalized-clustered control strategy for vehicle interior engine noise
•A normalized clustered control strategy is adopted to a new normalize the step size of the two-channel FxLMS algorithm for each local controller, which balances the system convergence rate and the steady state error, thus improving the noise reduction performance of the system, especially the track...
Uloženo v:
| Vydáno v: | Applied acoustics Ročník 228; s. 110263 |
|---|---|
| Hlavní autoři: | , , , , , |
| Médium: | Journal Article |
| Jazyk: | angličtina |
| Vydáno: |
Elsevier Ltd
15.01.2025
|
| Témata: | |
| ISSN: | 0003-682X |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
| Shrnutí: | •A normalized clustered control strategy is adopted to a new normalize the step size of the two-channel FxLMS algorithm for each local controller, which balances the system convergence rate and the steady state error, thus improving the noise reduction performance of the system, especially the tracking ability of the noise reduction under non-stationary acceleration conditions.•A deep learning method is proposed to estimate the secondary paths, which avoids the frequent re-estimation of secondary paths using the traditional offline estimation method under the disturbance of the dynamic environment. Then, a genetic algorithm is used to estimate a neural network model with the optimal number of hidden layer nodes to ensure the accuracy of secondary path estimation. Finally, to deal with the real-time problem of estimating secondary paths by the neural network, this study adopts the interpolation method to substitute the secondary paths estimated based on the deep neural networks (DNN) method into the ANC system for filtering convolution calculation.•A series of real vehicle experiments are conducted based on the proposed multi-channel ANC system. It has practical engineering guiding significance.
Although the multi-channel active noise control (ANC) system based on the traditional clustered control strategy solves the problems of high algorithm complexity and fragile stability, the step size setting of each local controller relies on the trial-and-error method, which makes it difficult to trade-off the convergence speed and the steady-state error of the algorithm, and the secondary path estimation is susceptible to the interference of the dynamic environment. To solve the above problems, this study proposes an efficient multi-channel ANC system, which accurately estimates the secondary paths based on the deep learning prediction model and adopts a normalized-clustered control strategy to normalize the step size of the two-channel FxLMS algorithm of each local controller, which balances the system convergence speed and the steady-state error. A series of real-vehicle ANC experiments are conducted. The results show that under stationary conditions, the proposed control strategy control converges quickly and has better noise reduction performance than the traditional clustered control strategy. Under non-stationary conditions, after normalizing the step size of the local controller, the proposed control strategy can better balance the steady-state error and the convergence speed of the control strategy and improve the noise reduction tracking ability. Finally, it is verified that the proposed deep learning method can accurately estimate the secondary path after changes and ensure the noise reduction performance of the proposed control strategy. |
|---|---|
| ISSN: | 0003-682X |
| DOI: | 10.1016/j.apacoust.2024.110263 |