双面声场反向神经网络声压外推分离

针对有限测量孔径造成的窗效应和离散传声器阵列带来的卷绕误差对双面声场分离技术的影响,提出了一种利用反向传播神经网络的全息面声压外推方法。该方法首先利用孔径内部测点平面坐标和相应声压值组成的学习样本训练神经网络,拟合出两者的函数关系。接着代入孔径外部测点坐标得到对应声压值,实现有限孔径外推。最后将已有测量值和外推测量值组成的虚拟大孔径导入双面声场分离处理。与传统外推方法相比,该方法不需要先验知识,操作简单,计算抗干扰能力强,结果准确性高。数值仿真和实验进一步验证了该方法在改善声场分离结果方面的可行性和有效性。 

Double-plane sound field separation after sound pressure extrapolation using back propagation neural networks

For the influence of the window effect caused by the finite aperture and the wraparound error due to the discrete microphone array on the double-plane sound field separation, a hologram pressure extrapolation method using back propagation neural network is proposed. Firstly, the learning samples consist of the planar coordinates of measurement points inside the aperture and the corresponding sound pressures. They are used to train neural networks, leading to the function fitting between both parameters. Next, the planar coordinates outside the aperture are introduced into the networks to generate the corresponding pressures, achieving the finite aperture extrapolation. Finally, the virtual large aperture including the measurement data and the extrapolated result is introduced into the double-plane sound field separation processing. Theoretical analysis shows that compared to the conventional extrapolation methods, the proposed method does not require a priori knowledge and is easy to use, yet of strong anti-interference capability and high accuracy. Numerical simulations and experiments further verify the feasibility and effectiveness of this method.