主催: 電気・情報関係学会九州支部連合大会委員会
会議名: 平成30年度電気・情報関係学会九州支部連合大会
回次: 71
開催地: 大分大学
開催日: 2018/09/27 - 2018/09/28
In this paper, we propose a fast synthesis of training dataset used for a CNN learning. This dataset is a combination of an image of the planar filter circuit and its filter response. The frequency response is a cascaded production of F-parameters representing partitioned planer circuits. It is obtained from a look-up-table which is calculated by an EM simulator in advance. In an experiment, filter responses are calculated from randomly generated images of planar filter circuits. As a result, 10,000 datasets can be generated in 80 minutes using an Intel Core i3 @ 2.53 GHz CPU.