Construction of a pockels effect-based free-space voltage sensor with neural network for signal processing

Abstract

A system which is capable of measuring strong (∼kV range) and fast (nanosecond range) electric fields in free space is proposed and developed. The sensor utilizes an electro-optic crystal, which reacts to electric fields and changes its refractive index to create birefringence. When a polarized laser beam is passed through the crystal, the birefringence changes the polarization of the laser beam, depending on the electric field strength felt by the crystal. The polarization change is measured by converting it into intensity modulation by the use of two orthogonal polarizers. Advantages of the sensor include fully dielectric components, small size for excellent resolution, and large bandwidth. However, one problem with using electro-optic crystals is the occurrence of piezoelectric ringing which create fluctuations when measuring fast and strong pulses. The use of neural network is investigated and its effectiveness is shown to reliably remove fluctuations from the measurement. The neural network is found to be highly effective in removing the piezoelectric fluctuations, while maintaining the original pulse shape. Pulses not included in the training data are also handled well, indicating good generalization of the network for real field applications. Comparison with the conventional method of handling piezoelectric ringing is also done to emphasize the superiority of the neural network method.