The Beam and Ball Project - Part 2 Servo Control

 

Software (continued)


Servo Control

To control the servo the microcontroller need to produce this 50 Hz waveform shown below. The resolution is defined by the microcontroller because the servo is an analog device.


There are mainly three ways of doing it.

Delays

Set output, then do delay. This is a bad way of doing it. The program is halted for every delay. However if you want to mash up some code to test your servo, this is the way to go.
  • Consumes the whole CPU
  • Good and quick technique to test servos.


Hardware PWM

The most positive about hardware PWM is that the overhead is 0. The CPU usage is practically zero, and only one or two instruction cycles are used to set a new pulsewidth.
Running the PWM at 50Hz demands that the PIC16F887 is ran in max 1 MHz due to max prescaler settings. So at the beginning of the project I went with 1 MHz.

The 10bit capable Hardware PWM uses a CCPx module, and TMR2. TMR2 is 8 bit, however the remaining two least significant bits are made possible by the fact that the oscillator is four times faster than the system clock. These bits has the be written into another register, but that is a small price to pay. The PWM period is set by register PR2, which basically resets the timer when TMR2=PR2. The pulsewidth (dutycycle) is set by the CCPRxL register.

Only a small portion of the pulsewidth is used. Only 1 of 20 ms is used to vary the servoposition. In theory, the resolution possible is then 1024 (10bit) * 1ms/20ms = 51 steps. In practice I used 80 steps.

For a while i used this technique. 1 MHz seemed to be more than enough. The beam vibrated for every step which seemed to cause the ball to move unexpectedly. Filtering the distance measurements helped a lot, but still the quantization noise (if I can call it that) of only having 51 steps resolution was quite poor.
  • 0% CPU usage
  • Demands max Fosc=1 MHz to produce ~50Hz PWM.
  • Only 51 of the 1024 steps of the PWM is usable for servo control. Practically, I used 80 steps though.

Interrupts

For 50 Hz PWM, using interrupts is recommended. You'll need four interrupts in 20ms, which at Fosc=8Mhz isn't so bad. In 20 ms, the cpu can do (Fosc/4) * 0.02 s = 40 000 .. Say there is about 5-6 instructions in each interrupts.. Round it up to 10 to include pushing the stack and moving program counter (PC). That's 40 instructions in one PWM periods. The overhead will then be 40 / 40 000 = 0.1 % . Unless you have some time sensitive communication going in your code the overhead is negligiable.

You can almost fit a full period of TMR0 into 1 ms. At Fosc = 8 Mhz, one instruction cycle is 1 / Fosc/4 = 0.5 us. If you set the TMR0 prescaler to 1:8, then the timer runs at 4 us/tick. The nice thing about these settings is that 250 of the 256 TMR0 ticks fits into 1 ms. You get 5 times the resolution of the hardware PWM above (or 3 times the practical. see above).

At 1:8 prescaler, 8 instructions are exectued for each tick. How short time can you set until next interrupt? It is possible to set the TMR0=254 and get 1 tick until next interrupt if the interrupt service routine is short enough. You can also compensate by making the initil 1ms a couple of ticks shorter and then offset the pwm pulsewidth by a couple of ticks because there are 6 ticks of the 256 ticks not used. I didn't care about that in my code.


The red dots in the figure above signifies cpu interrupts. Four interrupts are needed:
  • t = 1 ms, output=1
  • t = dutycycle
  • t = 1 ms - dutycycle (the complement of dutycycle)
  • t = 18 ms
The first interrupt sets output to high and sets the timer to 1 ms. Prescaler is set to 1:8, TMR0=6 because then there are 250 ticks left for 1 ms.
The second interrupt defines the variable part of the dutycycle. It can be 0-1 ms long. TMR0=PWM.
The third interrupt sets output to low and fills the rest of the 1 ms period to ensure a fixed 20 ms period independent of the PWM value. TMR0=-PWM (Which isn't technically correct, maybe adding 6 ticks to that will make it perfect but I didn't bother with that.
The last interrupt sets the timer to 18 ms. The prescaler is set to 256 and TMR0=141.
The interrupts are sequenced with if sentences and a variable that increments between 0..3.

Kommentarer

Legg inn en kommentar

Populære innlegg fra denne bloggen

Reverse engineering PCBs with KiCAD 7

In this Article we will be looking at replicating a PCB using KiCAD 7, and it's new Background Bitmap image feature. With the bitmap image feature it is much easier to replicate PCB designs. Before I have used sprint layout and kicad in a complicated process. Capturing images A scanner is hands down the best tool you can use for getting a good result. You can use a camera for smaller boards, but for larger boards you quickly run into problems. Image scaling Because KiCAD 7 not yet have separate x/y scaling, and x and y scaling is not equal usually, you have to scale them separately in a photo editor, like GIMP. When the image is loaded, set the DPI of the image to the same as the settings of the scanner. It is usually 600, but sometimes 1200 (or even 300). When you have set the correct DPI, the ruler tool should yield distances close to the real thing. Find two points on the board in the X direction that you know the distance of.  Use the (GIMP) ruler tool to measure them. Take not
Read more »

The Videopac G7000 composite mod

Bilde
   Introduction This is a write-up of my experience with a custom composite mod board for the Videopac G7000 game console, and the mod installation. The write-up is based on my youtube video below (the second one). In this blog post we will demonstrate the installation and picture quality. The sites gamesx and coolretroprojects linked below have installation instructions and schematics. Coolretroprojects sell a board, and they have a nice installation manual in PDF format. Links: http://gamesx.com/wiki/doku.php?id=av%3Ag7000_av#picture_of_g7000_internals http://www.coolretroprojects.com/Mods-Kits.html Videos: Video: Building a circuit for a friend of mine. Video: Installing the composite mod. This is the video you are reading a write-up of right now. Background The Videopac G7000 also known as Odyssey 2, however different internally and different composite mod board for it. The G7000 was introduced to the market in 1978. It has a RF modulator only..
Read more »

Designing the Videopac SDCart. 2: The SD Cart demo

Bilde
Introduction I managed to get several ROM binaries on an SD card to interface and run on the Videopac G7000. Last weekend I made a video about the progress so far. It was not all easy. The damn roms wouldn't run from the SRAM. Some problems were ironed out, but remaining was the problem of which address lines to use. The buses were isolated with three 8-bit transcievers, but two of them might be replaced by resistors. Sören Gust's gold mine recide in the archive.org. It was complete with design files for the G7000 RAM cart where you can connect a PC and transfer a ROM via serial. He also shared his source code and documents on the bios library. The Program Memory Map Often address line A10 is not used in game carts. Lets have a look at the memory map. In the program memory map for the Intel 8048, the first 1kb is reserved for the internal program memory, and the remaining 3kb is external program memory. These are accessed with A11..A0. Port1 pins P0 and P1 can be
Read more »