Synchronization between a SystemC based offline Restbus Simulator and a HIL FlexRay network

Transcription

1 Synchronization between a SystemC based offline Restbus Simulator and a HIL FlexRay network Gilles Bertrand Defo, Wolfgang Mueller University of Paderborn / C-LAB Fürstenallee Paderborn Germany bertrand@upb.de 1 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

2 Outline Introduction Hardware In the Loop (HIL) and Restbus Simulation (RBS) Restbus simulation framework Architecture Synchronization Experimental results Summary 2 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

3 Typical automotive HIL test system Test automation software (e.g.: dspace ControlDesk) Electronic Control Unit (ECU) Hardware with embedded software Real-time simulator (e.g: dspace Simulator) Plant model (vehicle or engine) Electrical emulation of sensors and actuators I/Os Road characteristics (e.g.: ABS) Eventually Signal adaptation fault simulation hardware 3 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

4 Restbus Simulation (RBS) Application: Development, Testing or Debugging of ECU (Electronic Control Unit) software. RBS when: Complete hardware is not available Only parts of the hardware are under test are available State of the art: Typically RBS runs on dedicated hardware Expensive, too application specific Our approach PC-Based SystemC Focuses on FlexRay TM (Time Division Multiple Access-TDMA) FlexRay Bus Based on IEEE standards Application of EDA testbenches Restbus simulation example [1] 4 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

5 Architecture of SystemC Simulator Simulator: SystemC TLM 2.0 (LT, AT) Testbenches: UVM, PSL, PC SystemC simulator Process 1 Process n Implements synchronization and communication with hardware Interconnect Hardware Simulator and Transactor run as separate OS processes Communicate via IPC Synchronization Buffer Transactor Hardware Driver Interface 5 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

6 Synchronization Focus is simulation of: Monitoring systems Active control systems Testbenches Slow system Challenge: SystemC simulator and real (physical) infrastructure do not have the same time base. Simulation can be slower than Hardware for complex SystemC models Undersampling and oversampling effect Downsampling Transactor Upsampling Undersampling: Missing data are filled by interpolation (Upsampling) fast system Oversampling: Data reduction (Downsampling), however causes loss of data Loss of peaks E.g.: shock of ramp (steer-by-wire), small ice spots (ABS) Need a mechanism for data reduction with peak detection 6 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

7 RBS Hardware: Upsampling Transactor interpolates by repeating the last valid value (simplest variant) PDUs FlexRay Cycle Generated data Interpolated data Fig.: Upsampling example Other complex interpolation is possible as well. 7 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

8 Simulator Hardware: Downsampling Synchronizationbuffer Receive queue 1 Sender Process queue Receiver 2 Send queue Fill send queue 8 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

9 Simulator Hardware: Downsampling Initialization phase: 1.b Send queue full, fill Process queue Synchronizationbuffer Receive queue 1 Sender Process queue Receiver 2 Send queue... Fill process queue and empty send queue 9 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

10 Simulator Hardware: Downsampling Main phase: 1.a Process queue full, fill receive queue Synchronizationbuffer Receive queue 1 Sender Process queue Receiver 2 Send queue... Process queue full, fill receive queue 10 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

11 Simulator Hardware: Downsampling Main phase: 1.b Send queue empty, start downsampling process. Synchronizationbuffer Receive queue 4. downsampling 1 Sender Receiver 2 Process queue 3. move Send queue... Send queue empty, downsampling 11 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

12 Downsampling Interval 12 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

13 Peak detection 1. Data smoothing using Robust-LOWESS (LOcally WEighted Scatterplot Smoothing) introduced by Cleveland in 1979 [2] Uses a linear polynomial regression method Local regression: smoothed value of each data is only influenced by neighboring data points defined with the span. Lowess smoothing example Robust: in addition to the regression weight function a robust weight function is used which makes the process resistant to outliers Robust smoothing example 13 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

14 Peak detection 14 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

15 Synchronizationbuffer Before downsampling Sender Send Queue.size() Receive queue Process Queue.size() Data rate of sender Data rate of receiver Receiver Process queue After downsampling Send queue 15 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

16 Impact of Send Queue size on peak detection 16 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

17 Implementation: delay Hardware PC SystemC Simulator FlexRay Bus Transactor Interconnect SWC_1 Boost Interprocess Boost Interprocess Synchronization Buffer FlexRay PCI Card SWC_2 FlexCard Driver Interface 17 Steering Wheel Wheel Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

18 Summary Approach used for synchronization between an offline SystemC based Restbus simulator and a FlexRay network PC-Based SystemC Focuses on FlexRay TM (Time Division Multiple Access-TDMA) Downsampling with focus on peak detection Based on IEEE standards Application of EDA testbenches References: [1] Cleveland, W.S. (1979) Robust locally weighted regression and smoothing scatter-plots, journal of the American Statistical Association, V. 74, pp [2] Analysis and residual bus simulation for FlexRay networks (Roman Hofmann) [3] 18 Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA

19 Thank you for your attention! C-LAB Fürstenallee Paderborn Germany Fon Fax Gilles Bertrand Defo, 17th NASCUG San Jose, CA USA