Magnetic Substrate Coupled Broadband and Miniaturized Electromagnetic Interference Shielding Structure Using Deep Neural Network

Abstract

Owing to the augmentation of wireless communication, the development of electromagnetic interference (EMI) shielding structures is an open area for research. The devices working at microwave frequencies are hazardous to human health and can also result in the malfunctioning of various sensitive equipments. Various techniques are reported by the researchers for the development of an efficient EMI shielding structure, some of them are composite materials, nanocomposites, metamaterials, and frequency selective surfaces (FSSs) [1]-[3]. FSSs are an array of periodic slots or apertures having excellent electromagnetic (EM) wave filtering properties. Major challenges across the research community are to create a miniaturized, angular stable, wideband, polarization-insensitive, and flexible EMI shielding structure. Multiband structures are often used by researchers to get a wide bandwidth. Whereas, miniaturization is attained either by techniques like 3-D, convoluted geometries, and lumped capacitances. All these methods of bandwidth improvement and miniaturization result in complex design and therefore fabrication becomes tedious. Almost all reported EMI shielding structures uses pure dielectric substrates like FR4 and Rogers. However, bandwidth and miniaturization can be improved by increasing the overall inductance of the shielding structure. The introduction of lumped inductors can be a solution but it not only increases the fabrication complexities but also may result in unwanted parasitic capacitance. Also introducing active or reactive components limits the conformal applications of shielding structures. One possible solution is to use magnetic substrates, which will increase the overall inductance of the shielding structure. Therefore, in this work electromagnetic properties of two magnetic substrates (Ferrite and magnetic silicon rubber) are measured and investigated to improve the bandwidth and miniaturization of an EMI shielding structure. Further, a polarization-insensitive, angular stable FSS is designed and the equivalent circuit model (ECM) is developed to have a physical insight into the structure. Moreover, a competent deep neural network (DNN) approach is employed for fast and accurate optimization of the shielding structure. At last, a prototype is developed and measured for validation of the proposed technique.The schematic of the proposed shielding structure is depicted in Figure 1(a). It comprises an array of periodic unit cells, a magnetic ferrite substrate, and a silicon rubber sheet. Figure 1(b) demonstrates an FSS unit cell with corresponding geometrical design variables. Here, ro is the radius of the outer circle and ri corresponds to the radius of the inner circle. d is the length of the patch, w is the width, and periodicity of the unit cell is represented by p. The electromagnetic properties of magnetic substrates are measured using a waveguide-based microwave measurement technique. Figure 1(c) depicts the measured relative permittivity (εr) and permeability (μr) of both the substrates. The real part of μr varies between 2.10 and 1.49 and the imaginary part varies between 0.34 and 0.0025 for magnetic silicon rubber when the frequency changes from 8.2 to 12.4 GHz. The real part of εr is 5.50 and the imaginary part varies between 0.20 and 0.005. On the other hand, the ferrite substrate real part of μr changes varies between 1.17 and 1.13 and the imaginary part varies between 0.04 and 0.001.Moving a step further, ECM of the FSS backed substrates is developed and the overall impedance is calculated. A huge database is created using the ECM and a deep neural network is trained using the Levenberg-Marquardt algorithm. The entire dataset is divided into three sections randomly for training, validation, and testing in the proportion of 70%, 15%, and 15%. Once the training, testing, and validation of the network is done, the shielding structure is optimized for resonant frequency and bandwidth. A prototype is then fabricated from the optimized parameters using the printed circuit board (PCB) technique. Figure 2(a) shows the image of the fabricated prototype. Meanwhile, the effect of the magnetic properties of the substrates on the transmission performance of the shielding structure is investigated. When the permittivity of the substrates dominates, a bandwidth of 2.75 GHz is obtained with resonant frequency at 10.42 GHz as shown in Figure 2(b). The bandwidth of surfaces with quite high impedance is directly proportional to the square root of overall inductance. To increase the bandwidth and to achieve miniaturization substrates with dominant magnetic properties are used. The overall inductance of the structure is increased by using substrates with dominant μr. Figure 2(b) depicts when substrates with dominant permeability are used in place of permittivity dominant substrates resonance shifts to lower value as the overall inductance of the structure increased. Also, the bandwidth obtained from such magnetic substrate increased to 3.37 GHz from 2.75 GHz. Therefore, it can be concluded that using magnetic substrates bandwidth enhancement and miniaturization of the shielding structure can be achieved. **