Study On The Reconstruction Of Near-field Sound Pressure Based On Sparse Sampling Cylindrical Array

For columnar sound source structures,cylindrical near-field acoustic holography is an effective tool for near-field acoustic quantity reconstruction and sound source identification.The reconstruction accuracy of the image from traditional cylindrical near-field acoustic holographic technology is closely related to the measured aperture and the number of measurement points.To achieve high reconstruction accuracy and resolution,large and dense microphone array must be arranged in actual operation,which increases the cost of measurement.How to use a small amount of test data to obtain a large amount of accurate prediction data has become an urgent problem to be solved.First,this dissertation studies the mechanism of action and the solution of inherent error of cylindrical near-field acoustical holography based on spatial Fourier transform.Then the estimation method of signal to noise ratio is designed.The algorithm principle of the statistically optimal cylindrical near-field acoustic holography is stated.To treat the inverse ill-posed problem,the Tikhonov regularization method is used.And the standard and regular solution is gotten.In the end,this chapter studies the reconstruction accuracy of the two techniques by sampling interval,sampling array size and the location of the sound source relating to sampling array.The complex sound pressure from the holographic surface has the band-limit characteristics of wave number domain.According to this condition,the related chapter puts forward a patch cylindrical near-field acoustic holography algorithm based on PGa.It extends the Patch theory which widely used in plane near-field acoustic holography to cylinder.By combining the extrapolation of the data outside the sampling array with the interpolation of data among the known data points,a large and dense sampling of the complex sound pressure on the holographic surface can be completed with few measurement points.This algorithm can effectively reduce the number of measurement points.The experimental results can also prove that this algorithm has good feasibility.Combining the stacked auto encoder with 3D-convolutional neural network in the field of deep learning together,the combined CS3 C model is proposed.The normalized complex sound pressure is transformed through a cylindrical translation window to expand the receptive field of the boundary unit,then it is used to train the neural network.The result shows that this model can greatly reduce the sampling rate with small reconstruction accuracy loss and nearly constant spatial resolution.Thinking of the ability to restore the details of sound field,it performs better than the patch cylindrical near-field acoustic holography algorithm based on PGa.