Self-organization and Optimization of Priority-based Communication Buses

With the increasing complexity in distributed embedded systems, the communication between the nodes becomes an essential part. A popular approach is to connect components by using priority-based communication buses. Due to cost and physical constraints, however, the amount of communication is limited. This thesis pushes the limitations of priority-based communication buses by two methods: Access control and data rate increase. The access control targets two weaknesses of priority-based communication buses. First, it does not allow a fair distribution of bandwidth for equally important message streams. Second, the response time for messages with low priority increases with high bus utilization. Based on a game-theoretic model, a reinforcement learning algorithm is proposed that uses simple local rules to establish fair bandwidth sharing. Furthermore, a dynamic o� set adaptation algorithm for scheduling messages is introduced that may reduce the average response times. This in turn allows to increase the utilization while maintaining the same response times. Both algorithms are executed fully distributed during run-time, require little computational e� ort and no additional communication. An increase of data rate on protocol level is proposed by extending the Controller Area Network (CAN). By sending additional data in those time slots where CAN-conform nodes do not listen, devices not dedicated to these boosted data rates can still be used. Prototype implementations of all introduced methods demonstrate the feasibility and gains in � exibility, response times and data rate on a real physical setup