CSE 522 Simulink Computer Science & Engineering Department Arizona State University Tempe, AZ 85287 Dr. Yann-Hang Lee yhlee@asu.edu (480) 727-7507 Some of the slides were based on lectures by Lee & Seshia (UC Berkeley) and Fainekos (ASU) Actor-Oriented Design Object orientation: What flows through an object is sequential control class name data methods call return Actor orientation: What flows through an object is data streams actor name data (state) Input data parameters ports Output data (http://ptolemy.eecs.berkeley.edu/presentations/04/Parc_Lee.ppt) Actor Orientation vs. Object Orientation Actor oriented Object oriented TextToSpeech initialize(): void notify(): void isReady(): boolean getSpeech(): double[] OO interface definition gives procedures that have to be invoked in an order not specified as part of the interface definition. actor-oriented interface definition says “Give me text and I’ll give you speech” Identified limitations of object orientation: Says little or nothing about concurrency and time Concurrency typically expressed with threads, monitors, semaphores Components tend to implement low-level communication protocols Re-use potential is disappointing (http://ptolemy.eecs.berkeley.edu/presentations/04/Parc_Lee.ppt) 3 Example of an Actor-Oriented Framework Signal flow graph with linear, time-invariant components Synchronous concurrent composition of components basic abstraction mechanism is hierarchy. (http://ptolemy.eecs.berkeley.edu/pres entations/04/Parc_Lee.ppt) 4 Matlab/Simulink Developed by Mathworks (http://www.mathworks.com/) Matlab An high-level programming language and interactive environment for scientific computing good quality numerical algorithms easy-to-use graphics and visualization capabilities real and complex vectors and matrices (including sparse matrices) The syntax of the language closely resembles the way we write mathematical equations Easy extensibility, by the user or via packages of M-files (which contains a computer code) and GUIs known as toolboxes It has a large number of toolboxes as add-ons The academic and scientific communities also create toolboxes 5 Example Mathlab Code Trapezoidal Rule function y = corrtrap(fname, fpname, a, b) % % % % Corrected trapezoidal rule y. fname - the m-file used to evaluate the integrand, fpname - the m-file used to evaluate the first derivative a,b - endpoinds of the interval of integration. h = b - a; y = (h/2)*(feval(fname,a) + feval(fname,b))+ (h^2)/12*(feval(fpname,a) feval(fpname,b)); 6 Simulink A graphical environment for multi-domain simulation and Model-Based Design for dynamic and embedded systems. Based on block diagrams and data flow modeling Not a programming language Hierarchical, component-based modeling Extensive library of predefined blocks MATLAB integration Application-specific libraries available Open Application Program Interface (API) It has a large number of toolboxes as add-ons Code generation Verification, Validation, and Test 7 Verification Methodology (1) Model in the loop Define mathematical models of the plant and the controller. sensor controller (model) test vectors plant (model) actuator Test process outputs Simulation platform Can the controller and algorithm fulfill the specification? The set of tests will be used as “an oracle” in the next steps. 8 Verification Methodology (2) Software (code) in controller (software) plant (model) the loop Test process Simulation platform Hardware in the loop Simulation platform Processor in the processor (software) plant (model) loop Test process 9 Simulink Blocks (1) Block: an actor Consists of some functionality and an arbitrary number of ports can be pre-defined blocks from Simulink library, S-function blocks (writing your own function in C, Fortran, etc.), or subsystem blocks S-functions are dynamically linked subroutines that the MATLAB interpreter can automatically load and execute Signals connect block‘s ports to pass data between blocks To calculate the values of the output ports based on the values of the input ports and the internal states. Sample time: how often and when the functionality of a block is evaluated. 10 Simulink Blocks (2) Continuous blocks have an infinitesimal sample time, e.g., integrator and derivative blocks Discrete block configured by a sample time parameter can be inherited either from the block connected to its input (inheritance) of its output (back inheritance). Block update (for all blocks within a system): to compute the block‘s outputs, the block‘s states, and the time for the next time step Direct feed-through ports The calculation of the output values depends on the input values of the current sample time, e.g., sum block An algebraic loop occurs when a signal loop exists with only direct feed-through blocks within the loop 11 Simulink Signal and Subsystem Each Signal Object is associated to: Data Type, dimension, sample rate, complexity (real, complex or auto), minimum/maximum values, initial value, unit of measure (only for doc), description Subsystems -- decompose the model in components and reuse subsystems virtual non-virtual atomic documentation library conditional enabled triggered control flow 12 Conditionally Executed Subsystems A subsystem whose execution depends on the value of an input signal. Enabled Subsystem Executed if the control signal has a positive value Control Flow Subsystem Executed if the control flow condition (e.g., if, while) evaluated to true Triggered Subsystem: execute each time a trigger event occurs positive or negative edges function call signals (from function-call generator blocks or S-function blocks) 13 Mathlab/Simulink Code Generation (formerly Real-Time Workshop) Embedded MATLAB: (http://blogs.mathworks.com/seth/2011/04/ 08/welcome-to-the-coders/) a subset of the MATLAB language that supports efficient code generation for deployment in embedded systems can use Real-Time Workshop to convert MATLAB programs to C programs Generated code can run in real time (physical clock) or simulated time (steps) 14 Simulink Model Execution (1) Model compilation Identification of the signal types, sizes Propagation of the types Optimization of the structure Flattening of the virtual subsystems Link phase Allocation of memory structures Connection between elements Loop phase 15 Simulink Model Execution (2) Block defines the time invariant relation between its input and its output. Virtual time – computation takes no time and when it is done, advance to next time step. Major step to produce output Minor step to improve the accuracy of continuous solvers (http://synchron2011.di.ens.fr/slides/chapoutot_ simulink_semantics.pdf ) input: x0, d0, t0, h0; n = 0; loop until tn tend evaluate g(tn, xn, dn) ; update d = fd (tn, xn, dn) ; solve ẋ(t) = fx (t, x(t), dn) over interval [tn, tn + hn] to get x(tn + hn) ; find zero crossing ; compute hn+1 ; compute tn+1 ; dn+1 = d; xn+1 = x(tn + hn) ; n=n+1; end loop 16 Simulink Model Execution (3) Order of block updates Updated before any of the blocks if it drives via direct-feed-through ports Blocks that do not have direct-feed-through inputs can be updated in any order. Atomic (compute at once) to Virtual Subsystem (flattened to the level of the parent system) Block priority 17 Execution Models of Generated Code (1) single-rate multi-rate multi-rate singletasking • model_main() spawns a base rate task, tBaseRate, which is task is activated by a clock semaphore and calls model_step singletasking • model_main() spawns only a base rate task, tBaseRate. All rates run under this task. • The base rate task is activated by a clock semaphore provided by VxWorks. • On each activation, tBaseRate calls model_step. • model_step in turn calls the rate_monotonic_scheduler utility, which maintains the scheduling counters that determine which rates should execute. multitasking • model_main() spawns a base rate task and sub-rate tasks which call model_step with their associated tids. • The base rate task and model_step are responsible for maintaining event flags and scheduling counter. • Task priorities are assigned by rate. 18 Execution Scheduling In simulation: computation results occur instantaneously In real execution non-zero non-deterministic block execution time task overrun detection communication through data variables in memory Single-tasking multiple-rate execution rates are harmonic a timer ISR to drive at base rate assume 3 tasks of sampling times of 1, 2 and 4 T1 T2 T4 T1 T1 T2 T1 T1 T2 T4 (from Simulink Coder User Guide) 19 RTW Main program (no RTOS) main() { Initialization (including installation of rt_OneStep as an interrupt service routine for a real-time clock) Initialize and start timer hardware Enable interrupts While(not Error) and (time < final time) Background task EndWhile Disable interrupts (Disable rt_OneStep from executing) Complete any background tasks Shutdown } (from RTW Embedded Coder User Guide) Periodically interrupted by a timer. rt_OneStep is either installed as a timer interrupt service routine (ISR), or called from a timer ISR at each clock step 20 RTW Main program (no RTOS) rt_OneStep() { Check for base-rate interrupt overrun Enable "rt_OneStep" interrupt Determine which rates need to run this time step Model_Step0() -- run base-rate time step code For N=1:NumTasks-1 -- iterate over sub-rate tasks If (sub-rate task N is scheduled) Check for sub-rate interrupt overrun Model_StepN() -- run sub-rate time step code EndIf EndFor (from RTW Embedded Coder User Guide) } sequences calls to the model_step functions for multi-rate model reinterruption of rt_OneStep by the timer is an error condition and rt_OneStep signals an error and returns immediately. 21 Rate Transition (1) Multi-rate and multi-tasking model multiple threads (tasks) with preemption one thread for each rate and, for periodic tasks, the faster rate task has higher priority Faster to slower transition The input of slower block may be changed during its execution Slower to faster transition (from Simulink Coder User Guide) 22 Rate Transition (2) Rate transition block to copy data runs in high priority, but low rate output value is held constant while the slower block executes rate transition update and output (from Simulink Coder User Guide) 23 Semantics-preserving (1) Synchronous: atomic reactions indexed by a global logical clock, each reaction computes new events for its outputs based on its internal state and on the input values the communication of all events between components occur synchronously during each reaction. Cycles of reading inputs, computing reaction and producing outputs Synchronous = 0-delay = within the same cycle No interference between I/O and computation Why? deterministic semantics in the presence of concurrency. 24 Semantics-preserving (2) A_2 A_1 a high priority task B arrives and receives inputs from A (from A_1 or A_2?) A_1 A_1 C B C A_2 A_2 B C B if Pri(A) > Pri(C)>Pri(B), depending upon the execution time of C, B may receive inputs from A_1 or A_2 25 Semantics-preserving (3) If execution time = 0, then the computation is determined by the order of arrivals, not the arrival instances, nor execution time Can we memorize the arrival order and then fetch data from buffer Unit delay is necessary when a higher priority task reads from a lower priority task A_1 C B A_2 B A_2 C B y_1 A_1 C y_1 y_2 26 Asynchronous Support in RTW/vxWorks Asyns interrupt -- generate interrupt-level code Task sync -- spawn a vxWorks task that calls a function call subsystem Protected RT -- enable data integrity when transferring data between blocks running as different tasks Unprotected RT -- use an unprotected/nondeterministic mode 27