Ethernet Train Network Real-time Modification Research

Introducing Time-Division Multiplexing Access (TDMA) industrial Ethernet into train communication network can avoid the conflicts of data sending and overcome the disadvantages of traditional field bus based train network, such as low speed, simple network topology structure, high design and using cost. This paper analyzes the influencing factors of the real-time performance in Ethernet. Under this background, slave clock compensation and train gateway queue scheduling problems are discussed.Precise time synchronization is needed in the TDMA network. Due to the sudden change of IEEE1588direct clock compensation, Kalman filter based frequency indirect compensation method is proposed. Firstly, Kalman filter is used to pretreat the IEEE1588time offset data to enhance the precision of time offset, and then time-counter frequency is compensated accordingly. In this way, time offset between master clock and slave clock is compensated indirectly and system stability is improved.Concerning about the heavy traffic in the train gateway, a reinforcement learning based adaptive weights change queue scheduling algorithm is proposed in order to reduce the queuing delay. The basic elements of reinforcement learning are defined. According to the measured queue length and the predefined length, the state set is designed. The revenue function is designed according to the change of queue length after each queue scheduling action. The scheduling weigths of each queue are changed through reinforcement learning. This method can schedule each queue dynamically and adapt to the traffic variation well. Using this strategy can guarantee the delay requirement of each queue as well as enhance the bandwidth utilization.Finally, we achieve train Ethernet communication nodes based on ARM microprocessor and embedded operating system. Two types of communication nodes are designed, one is bus-Ethernet network interface card which has clock synchronization and protocol conversion functions, and the other is train gateway node which provides simple routing and queue scheduling functions. The experiments based on above nodes demonstrate the reasonability of proposed clock compensation and queue scheduling method.