A MATLAB TOOLBOX FOR REAL-TIME CONTROL USING C167 MICROCONTROLLERS

Transcription

1 A MATLAB TOOLBOX FOR REAL-TIME CONTROL USING C167 MICROCONTROLLERS F. WÖrnle, R. Murillo Garcia Glasgow Caledonian University School of Engineering, Science and Design Cowcaddens Road, Glasgow, G4 0BA, Scotland, UK Tel.: +44 (0) , ( ), ( ) Abstract "RTC167-Target" is a MATLAB/Simulink block-set for real-time control applications on Infineon C167 microcontrollers. Representing the key hardware units of the microcontroller by intuitive Simulink blocks, the time consuming process of low-level coding in C/C++ or assembler, followed by the inevitable testing and debugging, is not required anymore. At present, RTC167-Target provides access to the 16 channels of the A/D converter and to all available digital I/O lines. A separate pair of pulse width modulation channels is also available. Additional hardware units can easily be included in form of user-supplied s-functions. The microcontroller can communicate with a host machine (PC) through an RS-232 serial connection. This opens the door to real-time monitoring of process data as well as the on-the-fly downloading of control parameters. Host-to-target communication is based on the external mode interface of Simulink, giving RTC167-Target the look and feel of a typical Simulink real-time target. Published under the GNU General Public License (GPL), the presented block-set is available free of charge and can be downloaded from the Internet. facilitate the study of dynamic systems in general and their control in particular, a simulation is always limited to the undertaking of virtual case studies. The gap between the theoretic analysis of a control engineering problem and the practical implementation in the context of a real-world application remains unresolved. Much of the hands-on experience is thus left to the imagination of the students, thereby reducing their chances of building a solid base of applicable knowledge. Amongst the reasons for this problem are cost and time factors: Modern laboratory equipment is often very cost intensive and delicate, requiring more careful handling than a simulation software package. Another aspect may be the limited time which can be spent on building and testing prototypes. Many experiments would simply require too much time to be covered in a single laboratory session. An attempt to remedy this situation has recently been undertaken at Glasgow Caledonian University s School of Engineering, Science and Design. A microcontroller based mobile robot has been designed to provide a flexible platform for student projects and control engineering applications (Figure 1). Keywords Real-time, embedded, MATLAB, Simulink, C167 microcontroller 1. Introduction At university level, control engineering is commonly considered to be a part of the sciences of applied mathematics. In consequence, lectures on this subject often turn out to be of a predominately theoretical nature: The term mathematics is emphasised at the expense of the prefix applied. It seems to be difficult to achieve a good balance between the textbook theory and its use in practical applications. Attempting to solve this problem, many universities have turned towards modern simulation tools such as MATLAB and Simulink. However, while these teaching aids greatly Figure 1: Mobile robot unit Central to this robot is a C167 microcontroller (Infineon [1]) which can be programmed and controlled through a wireless serial link (Bluetooth module [2]). A variety of sensors can be fitted to explore the environment surrounding the robot. Interesting applications can be studied using a camera based object detection system such as the Phytec pcgrabber [3] in conjunction with

2 the freely available image processing library CMVision [4]. The programming of a microcontroller and the subsequent debugging process are very time consuming tasks. From the point of view of a control engineer these tasks are of little or no interest at all. With this in mind, a MATLAB toolbox was developed giving direct access to the hardware units of the C167 from within the highlevel development environment Simulink. Making use of MathWorks Real-Time Workshop, a regular Simulink model can be built into a real-time module for the C167. Following the download to the target, the code execution can be controlled through the external mode communication interface of Simulink. Process data can be monitored live using Simulink Scope or Display blocks and parameters can be tuned while the controller is running (on-the-fly). Optionally, freely customisable user telegrams can be exchanged between the host and the target. The toolbox RTC167-Target is available free of charge under the terms of the GNU General Public License [5]. Source code files, documentation and sample models can be obtained from [6]. The remainder of this contribution highlights the key features of RTC167-Target and outlines typical applications. 2. Hardware and Software requirements RTC167-Target has been developed for the Phytec [3] single-board computer phycore-167 in conjunction with the rapid development board HD-200 (Figure 2). Centred around the ROM-less microcontroller C167CR- LM (Infineon), the phycore-167 includes 256 kbyte of RAM, 256 kbyte of on-board programmable FLASH ROM and a standard RS-232 transceiver for one serial interface (UART). Figure 2: HD-200 with phycore-167 (centre) Once programmed the small phycore-167 module can be unplugged from the development board and connected to the specific target hardware. Originally, RTC167-Target was written for and tested with MATLAB 6.1 (R12.1). The release of MATLAB 6.5 (R13) brought about a number of significant changes of the source code generated by Real-Time Workshop [7]. This made it necessary to adapt the way the toolbox controls the build process as well as a the implementation of the external mode interface. All of these modifications are fully transparent to the user. The toolbox can thus be used either with R12.1 or with R13. The toolbox is currently aimed at the KEIL crosscompiler C166 [8]. However, for portability reasons sparse use has been made of compiler specific symbols and keywords. Should the toolbox be extended by additional functionalities, i. e. through customised S- Function blocks (see section 3), a native C-compiler is also required. All host sided modules of the toolbox have been compiled with Microsoft Visual C [9]. 3. Working principle of the toolbox The build process from a Simulink block diagram to a C167 target module is governed by a template makefile c167.tmf. The format of this makefile is compatible to the widely used development tool GNU make. A Windows based port of this tool has been included in the distribution of the toolbox [10]. The toolbox is organised as a collection of functionally distinct modules, including both hardware dependent as well as platform independent code sections. Central element is the interrupt based timing engine. A single timer is used for Simulink models with control cycles of up to 3.6 seconds. Slower controllers require a cascade of two timers. Typical controller base rates range from 200 Hz to 500 Hz. User access to the hardware of the microcontroller is provided through a set of S-functions. The current distribution includes modules for the A/D-converter unit, two separate PWM channels and the digital I/O lines of the controller. This modular concept facilitates future inclusions of additional hardware units. RTC167-Target can be used to generate both standalone controllers as well as host-driven applications. All hosttarget communications make use of MathWorks external mode interface, implemented on a 3-wire serial connection (null modem). The principal services of this interface are the on-the-fly parameter tuning as well as the monitoring of signals in real-time. This calls for a robust communication protocol with flow control and mechanisms to avoid data congestion. For example, the temporary unavailability of the host should not affect the stable operation of the target. Every data upload is therefore initiated by the host. Should a host be busy with a more imminent task, this upload trigger remains unsent, causing the target to hold further data

3 monitoring. All Simulink Scopes and Display blocks are updated once the host returns to normal operation. The communication parameters of the serial interface can be set using the options page of Real-Time Workshop (Figure 3). Figure 3: Code generation options An extension of the external mode interface provides an optional messaging system with a maximum of 10 user channels. Freely structured messages of up to 250 bytes can be transmitted between the host and the target. The host sided end points of all user channels needs to be defined as a separate block diagram, the so-called host model. A target application can exploit user channels to access host-based server applications such as data bases, network resources or data acquisition cards. Figure 4 illustrates the use of the messaging system in a simple remote control of a mobile robot: the target model (foreground) receives velocity information and directional data from the associated host model (background). build process and the installation of the target module in the ROM space of the hardware, the controller is started by a target reset. Code execution will prevail until the power is cut or a further reset is issued. In its present form, the toolbox can be configured produce code for the compact, the large or the huge memory model. The compact memory model allows for code sizes of up to 64 kbyte and 64 kbyte RAM. This is commonly sufficient for most reasonably sized block diagrams. More complex control tasks, however, may require the use of the large or the huge memory model. While both memory models allow for unlimited code size and unlimited data size, the large memory model restricts the maximum extension of any one data object to a single data page (16 kbytes); the huge memory model does not impose this restriction. The required memory model can be specified via the options page of Real-Time Workshop (see Figure 3). Another important code generation option which can be selected from this page is the generation of timing signals. Currently supported signals are the display of the cycle time and the monitoring of the serial reception. The cycle time measures the actual duration of the base rate interrupt service routine (ISR). A positive rectangular pulse is produced, showing the duration of the active control cycle. The other signal which can be monitored is associated with the reception characters on the RxD line of the microcontroller. A short positive pulse is produced for every character received from the host. Figure 5 shows the screen shot of a typical oscilloscope trace; channel 1 (top) represents the cycle time, whereas the serial reception is monitored using channel 2 (bottom). Figure 5: Timing signals (cycle time, serial reception) Figure 4: Optional user communication Standalone solutions do not make use of the external mode interface. The runtime code is therefore much more compact and allows for slightly faster base sample rates. Typical reductions of the size of the target module range from 50 % to 70 %. Following the successful The timing signals allow a designer to ensure that the chosen base sample rate of a particular Simulink block diagram does not exceed the capabilities of the microcontroller or of the serial communication interface.

4 4. The blocks of RTC167-Target Upon the successful installation of the toolbox, a new entry can be found in the Simulink library browser: Real-time C167 toolbox. This new library gives access to 8 new blocks. The user communication interface is represented by the blocks Transmit User Telegram and Receive User Telegram. Block mask parameters include the channel number (0 10), the number of elements to be sent and the data type of these elements (single precision floating point number, integer, byte). The total length of each user telegram is currently limited to 250 bytes. The other blocks provide access to various hardware units of the microcontroller: The 16 channels of the A/D conversion unit can be read using the block A/D converter. Pulse width modulated (PWM) signals can be generated using the PWM block. The toolbox currently provides two individual PWM channels. Background monitoring of the generated PWM signals can be chosen through the block parameter mask. Control of the digital I/O lines of the microcontroller can be achieved using the blocks Digital input and Digital output. The latter allows definition of a hysteresis: A corresponding line is set to logical high when the block input exceeds the upper threshold V up ; it returns to logical low if the block input signal falls below the lower threshold V down. Finally, the blocks Toggle a pin of port 2 (enabled) and Toggle a pin of port 2 (triggered) allow inversion of the logical state of a chosen pin of port 2. On the Phytec rapid prototyping board HD-200 this can be used to display warning signals such as the flashing on-board LED, etc. Figure 6: Camera driven robot control These images are digitised with a Phytec framegrabber card pcgrabber-3 [3] and processed using the public domain vision recognition software library CMVision [4]. Selected colour coded objects can thus be detect and their co-ordinates within a reference frame can be determined. A special Simulink block has been derived to extract the centroid co-ordinates of objects which fit a predefined colour scheme. Using this set-up, typical machine vision applications can be put together within very short times. Figure 7 shows the images seen by the camera together with the detected object, here a red dot near the front of the robot. For illustrative purposes, the co-ordinate detection block allows this information to be displayed while the image processing is in progress. 5. Application examples A number of illustrative examples are distributed with RTC167-Target to explain the common use of each of the blocks of the toolbox. Where a model includes the optional user communication blocks, both host model and target model are provided. The host model is thereby given the postfix _pc. To discuss all of these models would certainly be beyond the scope of this contribution. A detailed description can be found in the manual of the toolbox which is part of the distribution. This document also includes a step-by-step tutorial of how to use the toolbox. At Glasgow Caledonian University, the RTC167-Target is used to experiment with a set of small mobile robots (see Figure 1). An interesting application is the camera driven control of a robot. A stream of live images is supplied by a small camera which has been mounted at the ceiling of the laboratory (Figure 6). Figure 7: Picture as seen by the camera The co-ordinates are sent to the robot via a wireless serial link (Bluetooth). By comparing the detected coordinates with those of a preset reference position, the robot can work out a positional error. This error signal can then be used to drive the motors. The convenient block diagram based programming of the microcontroller makes it very easy to compare different strategies for the approach of the destination position.

5 This also includes the variation of controller internal elements, e. g. the replacement of a PID block by a fuzzy controller, etc. To facilitate the training of the vision recognition software for the detection of objects of a particular colour, a special training tool has been developed. This MATLAB script file allows the graphical definition of the object of interest. A range of HSV-triplets is returned, providing important guidance for the final specification of the colour definition files as required by CMVision. Figure 8 illustrates the use of the teaching tool. RTC167-Target is distributed as Free Software under the terms of the GNU General Public License (GPL) [5]. The source code can be downloaded from the internet at [6]. References [1] Infineon technologies, C167CR Derivatives User s Manual, Web page accessed: March 2003, [2] ConnectBlue Inc., OEM Serial port adapter, Web page accessed: March 2003, [3] Phytec Technology Holding AG, Web page accessed: March 2003, [4] The CORAL Group, Carnegie Mellon University, The Color Machine Vision Project - CMVision, Web page accessed: March 2003, www-2.cs.cmu.edu/~jbruce/cmvision [5] The GNU Project web server, Web page accessed: March 2003, Figure 8: Training mode: detection of the red dot 6. Conclusions A MATLAB/Simulink toolbox for the rapid prototyping of C167 microcontroller based control applications has been presented. Eliminating the need for tedious handcoding and debugging of microcontroller programs, this toolbox allows enormous savings of design time to be made. In an educational environment, this time can now be spent on studying the actual control problem at hand. The entire stage of the familiarisation with a compiler package and/or a, possibly new, programming language has thus been cut out. This makes it possible to include realistic control problems, such as machine vision based control and pathfinder robots, into the regular curriculum. These can be treated in form of student projects or small case studies which can be accommodated within the duration of one or two laboratory sessions. [6] F. WÖrnle, R. Murillo-Garcia, RTC167-Target, Web page accessed: March 2003, Target.htm [7] The MathWorks Inc., Web page accessed: March 2003, [8] KEIL Software Inc., Web page accessed: March 2003, [9] Microsoft Visual Studio, Web page accessed: March 2003, [10] Mozilla.org, Building Mozilla on Microsoft Windows 32-bit platforms, Web page accessed: March 2003, ftp.mozilla.org/pub/mozilla/source/wintools.zip Experience shows that, the toolbox has helped to reduce the frequently significant gap between the theoretical investigation (including dynamic simulations) and its actual application on a real prototype. Due to the completeness of the undertaken case studies the participants reveal a greatly improved motivation, leading to a remarkably raised learning outcome.