Research On Active Noise Control Method Based On Deep Learning Optimization

With the continuous progress of social progress and the search for a better living condition,the noise issue is receiving more and more extensive concern.The core idea of Active Noise Control(ANC)is to analyze the incoming noise through a specific control algorithm and generate an acoustic wave with the same amplitude and opposite phase of the noise signal to cancel the noise.Traditional active noise control methods generally use a fixed-step strategy,which makes it difficult to achieve a trade-off between the convergence speed and stability of the algorithm.In order to improve the noise attenuation effect and balance the relationship between the convergence speed and stability of the algorithm,this article researches and explores the key technologies of ANC,mainly including adaptive filters,adaptive algorithms and deep learning networks in the speech field,and on this basis,from the perspective of adaptive algorithm improvement and deep learning network fusion,explores the new scheme of step size update and deep learning technology based The new techniques of ANC based on deep learning techniques are explored and the effectiveness of the algorithms is verified by building a test system.The main work accomplished in this article is as follows:Firstly,the ANC key technology is analyzed,and the Energy Adaptive Fx LMS(En A-Fx LMS)algorithm is proposed for the problem of unbalanced convergence speed and stability of the adaptive algorithm due to the step size update strategy in the LMS algorithm.The step update function is improved on the basis of the traditional Fx LMS algorithm,and the iterative step size is adaptively updated based on this.Experiments show that the Normalized Mean Square Error(NMSE)of the En AFx LMS algorithm is lower than that of the Fx LMS algorithm by about 2.5 and lower than that of the normalized Fx LMS algorithm by about 1.1 under different noise conditions,showing good noise reduction and generalization;Next,for the problem of slow convergence speed of traditional adaptive algorithm,Deep SF-Fx LMS algorithm is proposed.The algorithm takes advantage of the features of Fx LMS algorithm with strong adaptivity and low steady-state error but slow convergence speed,and combines the features of SFANC with fast convergence speed but weak adaptivity and large steady-state error to design a new filtering method based on Fx LMS algorithm and SFANC,which realizes a filtering method for ANC fast adaptive filtering technique.Experimentally,the approach can achieve faster response rates compared to the Fx LMS algorithm;has lower steady-state error and adaptability compared with SFANC,and thus can be adapted to different noise environments;Then,to address the problem of inaccurate filter selection in SFANC in(2),an enhanced residual convolutional neural network(Improved Res CNN)is proposed to achieve the selection of optimal filter parameters for different frequency ranges.By integrating multiple convolutional layers and residual blocks into the same residual convolutional neural network(Res CNN),deeper and more complex feature extraction requirements are obtained for active noise reduction needs;furthermore,a noisy signal database with high frequency distinction is constructed to compensate for the problems of low classification labeling accuracy and inconspicuous frequency distinction of existing noisy data sets,and to enhance the training and validation of the neural network.The experiments show that the classification accuracy of Improved Res CNN network structure reaches 99.2%,and its classification accuracy is significantly better than the commonly used multiple residual network methods.Finally,on the basis of the above research,the speech collection module and the noise control module were built separately using Simulink to complete the overall construction and testing of the active noise control system and realize the functional simulation at the software level,which laid the theoretical foundation for the application of the active noise reduction system.