In the last few years, convolutional neural networks (CNNs) have demonstrated good performance while solving various computer vision problems. However, since CNNs exhibit high computational complexity, signal processing is performed on the server side. To reduce the computational complexity of CNNs for edge computing, a lightweight algorithm, such as a MobileNet, is proposed. Although MobileNet is lighter than other CNN models, it commonly achieves lower classification accuracy. Hence, to find a balance between complexity and accuracy, additional hyperparameters for adjusting the size of the model have recently been proposed. However, significantly increasing the number of parameters makes models dense and unsuitable for devices with limited computational resources. In this study, we propose a novel MobileNet architecture, in which the number of parameters is adaptively increased according to the importance of feature maps. We show that our proposed network achieves better classification accuracy with fewer parameters than the conventional MobileNet.

[1], [2] 를 제안 한다. 제안된 안테나는 뮤-네거티브(MNG: Mu-negative) 전송선로의 직렬 인덕턴스와 커패시턴스를 사용하는 뮤-제로 영 차공진 안테나이며, 간단한 구조의 기생패치를 추가함으로써 직렬 커패시턴스를 증가시켜 소형화 하였다. 추가된 기생 패치는 기존의 직렬 커패시턴스에 추가적인 병렬로 커패시턴스를 만들어 공진 주파수를 결정하는 직렬 커패시턴스를 등가적으로 증가시킨다. 기생패치는 HFSS [3] 를 이용한 모의실험을 통하여 최적화되었다. 제안된 안테나는 0.59의 값을 가지며, 기존 뮤-제로 영차공진 안테나 크기 대비 24 % 소형화 되었으며, 92 %의 효율과 6.57 dBi의 이득을 보였다. 최종 설계된 안테나는 제작 및 측정되었으며, 측정 결과는 모의실험 결과와 잘 일치함을 확인하였다. AbstractIn this paper, a small mu-zero zeroth order resonance(ZOR) antenna [1], [2] based on meta structure is proposed using parasitic patch at 5.8 GHz. The mu-zero ZOR antenna is designed by utilizing the resonance of series inductance and capacitance of mu-negative transmission line and its size can be further reduced by a simple parasitic patch. The parasitic patch can increase series capacitance of mu-negative transmission line related to a resonant frequency. We have simulated and optimized dimension of the parasitic patch using Ansys commercial simulator(HFSS [3] ). As a result, the antenna has the following characteristics: kr of 0.59, efficiency of 92 %, and gain of 6.57 dBi. Also, its size is reduced by 24 % compared to a conventional mu-zero ZOR antenna [1], [2] . The measured results are in good agreement with the simulated results.

This paper suggests an optimisation method to design multi-band antenna using artificial neural networks. The proposed network, surrogate-based model using auto-encoder (SBM-AE), is composed of two parts, ordinary neural network and auto-encoder. First, the front neural network predicts the encoded antenna characteristics and then decodes the predicted data to obtain the antenna characteristics. After training the encoder to obtain a characteristic signature vector, the front neural network regresses only the characteristic signature vectors, reducing the complexity and number of parameters of the neural network. This not only reduces the training time but also significantly reduces the number of training data required. We confirm the effectiveness of the design method by designing a multi-band antenna where the proposed SBM-AE required 270 training data while the conventional neural network without auto-encoder needed 1350 training data to achieve comparatively the same error rate. K E Y W O R D S auto-encoder, broadband antennas, neural nets, optimisation, surrogate based modelThis is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

A dual band omnidirectional circularly polarized (CP) antenna is presented based upon new concept of combining two orthogonal modes of epsilon‐zero resonance (EZR) and mu‐zero resonance (MZR) in this article. The antenna has the CP characteristic by two circularly ring‐structured double negative (DNG) transmission lines. One DNG transmission line has higher frequencies of EZR and lower frequency of MZR while the other DNG transmission line has higher frequencies of MZR and lower frequency of EZR. The EZR and MZR modes at the same frequency have a vertical and a horizontal polarization with 90° phase difference, respectively. In the EZR mode, a vertical polarization is generated like an electric monopole so that the EZR antenna radiates omnidirectionally. The MZR antenna makes a horizontal polarization by the circular ring current like a magnetic dipole loop which also radiates omnidirectionally. The analysis and design of the antenna are performed according to theory and simulation based on a dispersion diagram and E‐field distribution. Finally, the reasonable performances of the antenna such as a reflection coefficient, a CP gain, an axial ratio, and a far‐field radiation pattern are obtained in dual band.

No abstract