Research On Active Noise Control Algorithm Combined With Machine Hearing

An active noise control(ANC)system usually contains a control filter to generate an anti-noise sound that equal in magnitude and opposite in phase to the primary noise for noise reduction purposes.The filter-x least mean square(FxLMS)algorithm is a very clas-sic and widely used adaptive ANC algorithm,which updates the control filter coefficients in real time through iteration.However,the FxLMS algorithm is primarily designed to minimize the power of the residual sound,without consideration to the auditory sensation.When the FxLMS algorithm has fully converged,the residual sound of the ANC system is perceived to be different from the primary noise.For example,when an ANC system run-ning the FxLMS algorithm was piloted in a substation’s control room,workers expressed discomfort with the residual sound,even though the noise level was significantly reduced.This was mainly because workers had been used to the process of identifying the operating state from the sound that they hear.After hearing the residual sound of the ANC system,workers would mistakenly identified the residual sound as an anomaly operating state,which annoyed them.It is preferable to keep the residual sound to represent the original operating state,in order for the sound environment to be comfortable and familiar.At the same time,due to the development of machine learning in recent years,it has helped or even replaced the human auditory system to perform tasks in some fields.In order to solve the above problems,the active noise control algorithm that combines the machine hearing to keep the original operating state to consider the human sense of hearing is studies in this thesis.This thesis introduced an anomaly sound detection(ASD)model based on sauxiliary machine type classification for production scenarios,and attempted to use the model per-formance as an objective indicator to replace the subjective indicator of human hearing.The HC-FxLMS algorithm was proposed based on the homothety constrained between the residual sound of the system and the original machine noise.The PE-FxLMS algo-rithm was proposed based on the perceptual equalization path which describes the impulse response relationship between the reference signal and the target primary noise signal at the working position.Simulations were carried out in this thesis,and the results verified the effectiveness of the two algorithms.Based on further research and expansion,the SPE-FxLMS algorithm was proposed and the selection criteria was derived in this thesis.Theoretical analysis showed that the improvement of the algorithm has no effect on the convergence,but the complex-ity increases to a certain extent.By comprehensively comparing the effects of the three algorithms which are controllable noise reduction FxLMS,improved brodband ANE and SPE-FxLMS algorithms through simulations under multiple machine types,experiments in the laboratory environment,and subjective tests,the superiority of the SPE-FxLMS algorithm in reducing noise level and shaping the residual sound to create a sense of audi-tory familiarity for the workers was verified.It is worth noting that it is also an important innovation of this thesis to improve the FxLMS and brodband ANE algorithms and unify each algorithm to accurately control the sound pressure level of the residual sound to meet the predetermined target in a same framework.