A Neural Network Approach for Controller Area Network Message Scheduling Control
Chuan Ku Lin, Hao-Wei Yen, Mu-Song Chen, Chi-Pan Hwang, Nguyễn Minh Thạnh · 2006
Abstract—In this paper, a novel approach for Controller Area Network (CAN) message scheduling control is proposed. The proposed method utilizes a single hidden layer neural network, i.e. Radial Basis Function Network (RBFN), such that the CAN’s bandwidth can be shared properly among different messages. CAN was initially developed for the automotive industry to solve the cabling problems found in some kinds of vehicles. However, CAN is not able to satisfy all the communication needs of the industrial automation and process control scenarios. In particular, the basic protocol in CAN can neither ensure deterministically bounded transmission times for RT (real-time) data exchanges nor a fair share out of the network bandwidth among non-RT application processes. Therefore, CAN must carry both RT messages as well as non-RT messages in order to meet the requirements of a dynamic distributed system. All these messages must be properly scheduled on the network so that RT messages meet their deadlines while co-existing with non-RT messages. To solve these problems, we present a neural network approach to dynamically schedule and distribute the system bandwidth. At the same time, the transmission time can meet the basic requirements. The experimental results clearly show the effectiveness of the proposed technique in solving the message scheduling problems in CAN.