http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
동적 저궤도 위성 네트워크에서 실시간 라우팅을 위한 FPGA 기반 컨볼루션층 추론 병렬화 기술
김대연,이헌철,원동식,한명훈 한국정보기술학회 2023 한국정보기술학회논문지 Vol.21 No.8
This paper addresses the real-time routing problem in Low Earth Orbit (LEO) satellite networks. Existing routing algorithms have been found to struggle to adapt effectively to dynamic satellite network environments. As such, this study proposes a reinforcement learning-based routing approach and implements it using a dueling deep Q-network model. However, the inference process on satellites faces challenges in meeting real-time requirements due to limited computational capabilities. To resolve this, we propose an approach that accelerates inference speed by parallelizing the convolutional layer's inference process. Experimental results show that our proposed method has reduced the computation time of the convolutional layer by 90.2% and the total algorithm execution time by 29.0% compared to the existing methods.
동적 저궤도 위성 네트워크를 위한 Dueling DQN 기반 라우팅 기법
김도형,이상현,이헌철,원동식,Dohyung Kim,Sanghyeon Lee,Heoncheol Lee,Dongshik Won 대한임베디드공학회 2023 대한임베디드공학회논문지 Vol.18 No.4
This paper deals with a routing algorithm which can find the best communication route to a desired point considering disconnected links in the LEO (low earth orbit) satellite networks. If the LEO satellite networks are dynamic, the number and distribution of the disconnected links are varying, which makes the routing problem challenging. To solve the problem, in this paper, we propose a routing method based on Dueling DQN which is one of the reinforcement learning algorithms. The proposed method was successfully conducted and verified by showing improved performance by reducing convergence times and converging more stably compared to other existing reinforcement learning-based routing algorithms.
동적 저궤도 위성 네트워크에서 온보드 강화학습 기반 라우팅을 위한 이종 프로세서 기반 추론 병렬화 기술
김도형(Dohyung Kim),이민준(Minjoon Lee),이헌철(Heoncheol Lee),원동식(Dongshik Won),한명훈(Myoung-Hun Han) 한국정보기술학회 2023 한국정보기술학회논문지 Vol.21 No.7
This paper addresses the routing problem in dynamic low-orbit(LEO) satellite networks for on-board computer (OBC). Deep reinforcement learning can be applied for routing in networks with dynamic connectivity between LEO satellites. However, it is difficult to apply the inference process with deep reinforcement learning models to real-time OBCs because it causes excessive execution time due to the calculation of multiple convolution layers. To solve the problem, we propose a practical method based on heterogeneous processors which can reduce the execution time by parallelizing the inference process, which is performed sequentially in a Central Processing Unit. The performance of the proposed method was evaluated using an actual OBC based on heterogeneous processors, and the routing result was the same as that of the existing method, but the overall execution time was significantly reduced.