Teensy Clock Fixed

Paul - the creator of the Teensy has provided a software update (Rev 1.04), that fixes the clock problem that I observed (see previous post) After a 24 hour test, my Teensy has only gained about 1 second (as compared to a wall clock) that is only about 11 ppm error, which is good considering ambient temp and generic 16mHz cpu crystal.

Another test with a known good reference clock is necessary for more resolution.

Thanks Paul, for your efforts to provide a fix.

--

Teensy Programming Shield

I had some time this morning to install the headers on the Teensy Programming Shield which I just received from the FAB Shop (see previous post). With plated through holes, soldering of parts (headers) is much easier than experienced with Homebrew double sided boards.

Headers Installed

Bottom

Top

--

DorkBotPDX in Portland

Monday night Jeff - KO7M and I attended the 7-11pm  DorkBotPDX meeting in Portland, there were many projects available to review and interesting people to meet.

An unexpected pleasurable encounter was with Paul - the creator of the "Teensy", which is the cpu board that I have mentioned in previous posts (see search label: Teensy). Paul also, video taped some of the high lights of the meeting. My GUI Development Board and 9 Volt Micro Transmitters can be seen.

http://youtu.be/seyTMkTGZ3E

It was a great meeting, even thought I did not get back to Seattle and home until 3:30am, I plan to attend often.

--

Teenys Clock Trouble

The only time I have had to work on other projects was late a night. My Teensy (an Arduino work-a-like, see previous posts) has been giving me fits. The clock is FAST by about 24 seconds per 6 hours. All previous efforts to correct the problem resulted in No Change.  Jeff - KO7M and I had some time to review my progress (or lack there of) and found that the Teensy does not use the same Clock file (wiring.c) as used by the Arduino (I was modifying the wrong file), in fact it not even named the same under the Teensy directory structure (pins_teensy.c).

void TIMER0_OVF_vect()
{
        asm volatile(
                "push   r24"                            "\n\t"
                "in     r24, __SREG__"                  "\n\t"
                "push   r24"                            "\n\t"

                "lds    r24, timer0_fract_count"        "\n\t"
                "subi   r24, 256 - %0"                  "\n\t"
                "cpi    r24, 125"                       "\n\t"
                "brsh   L_%=_fract_roll"                "\n\t"

        "L_%=_fract_noroll:"                            "\n\t"
                "sts    timer0_fract_count, r24"        "\n\t"
                "lds    r24, timer0_millis_count"       "\n\t"
                "subi   r24, 256 - %1"                  "\n\t"
                "sts    timer0_millis_count, r24"       "\n\t"
                "brcs   L_%=_ovcount"                   "\n\t"

        "L_%=_millis_inc_sext:"
                "lds    r24, timer0_millis_count+1"     "\n\t"
                "sbci   r24, 255"                       "\n\t"
                "sts    timer0_millis_count+1, r24"     "\n\t"
                "brcs   L_%=_ovcount"                   "\n\t"
                "lds    r24, timer0_millis_count+2"     "\n\t"
                "sbci   r24, 255"                       "\n\t"
                "sts    timer0_millis_count+2, r24"     "\n\t"
                "brcs   L_%=_ovcount"                   "\n\t"
                "lds    r24, timer0_millis_count+3"     "\n\t"
                "sbci   r24, 255"                       "\n\t"
                "sts    timer0_millis_count+3, r24"     "\n\t"
                "rjmp   L_%=_ovcount"                   "\n\t"

        "L_%=_fract_roll:"                              "\n\t"
                "subi   r24, 125 - %0"                  "\n\t"
                "sts    timer0_fract_count, r24"        "\n\t"
                "lds    r24, timer0_millis_count"       "\n\t"
                "subi   r24, 256 - %1 - 1"              "\n\t"
                "sts    timer0_millis_count, r24"       "\n\t"
                "brcc   L_%=_millis_inc_sext"           "\n\t"

        "L_%=_ovcount:"
                "lds    r24, timer0_micros_count"       "\n\t"
                "subi   r24, 256 - %2"                  "\n\t"
                "sts    timer0_micros_count, r24"       "\n\t"
                "brcs   L_%=_end"                       "\n\t"
                "lds    r24, timer0_micros_count+1"     "\n\t"
                "sbci   r24, 255"                       "\n\t"
                "sts    timer0_micros_count+1, r24"     "\n\t"
                "brcs   L_%=_end"                       "\n\t"
                "lds    r24, timer0_micros_count+2"     "\n\t"
                "sbci   r24, 255"                       "\n\t"
                "sts    timer0_micros_count+2, r24"     "\n\t"

        "L_%=_end:"
                "pop    r24"                            "\n\t"
                "out    __SREG__, r24"                  "\n\t"
                "pop    r24"                            "\n\t"
                "reti"
                : 
                : "M" (TIMER0_FRACT_INC), "M" (TIMER0_MILLIS_INC),
                  "M" (TIMER0_MICROS_INC)
        );
}

Teensy Clock Code from Original File

The Teensy Clock file is "pin_teensy.c", which is mostly written in Assembly Language. I unfortunately do not know Assembly Language well enough to make modification to attempt to correct the Clock. Jeff suggested we replace a section (the clock) with known working "C" code from Arduino.

The first attempt included a section to try to slow the Clock by an amount equal to the previously observed error. After 6 more hours, the clock was NOW slow by about the same amount as it was FAST before. Jeff suggested we remove the correction and obtain a new base line with just the "C" code installed.

SIGNAL(TIMER0_OVF_vect)
{
 // copy these to local variables so they can be stored in registers
 // (volatile variables must be read from memory on every access)
 unsigned long m = timer0_millis_count;
 unsigned char f = timer0_fract_count;

 m += MILLIS_INC;
 f += FRACT_INC;
 if (f >= FRACT_MAX) {
  f -= FRACT_MAX;
  m += 1;
 }
 
 timer0_fract_count = f;
 timer0_millis_count = m;
 timer0_overflow_count++;
}

Copied Arduino Clock Code

Due to the work in the Shop (listed above) I did not get back to the Clock project for 29 hours. After 29 hours, the base line results indicated that the Clock was spot-on, only less than a second difference could be observed when compared to a wall clock. Wow!

But, the "C" code replacement did something to the "Interrupts" for the Teensy, my push buttons no longer work as expected. We damaged something by removing the Assembly Language code. More investigation will be needed.

For now, I have reverted to the Assembly Language Clock routing, so that all of my GUI development environment works correctly, but with a Clock that runs a little fast.


UPDATE: The new release of Teensyduino 1.04 fixed the problem.

--

Shop Work, While E Projects on Hold

The last few weeks I have not had much time for Ham Radio, Computers or Electronic projects. My Son and I have been working on building/Installing/updating new infrastructure for my Shop.

• Low Bay door was insulated
• Water Heater and Pressure Tank were relocated
• New High Volume Compressed Air will be installed
• Power Lines to the Compressor were installed
• Air line will be install to all areas of the Shop
• Major Shop Tools will be wired  and placed
• High Volume Saw Dust Vacuum System will be installed
• Freeze Relocated
• A small Force Air Heater will be install in the High Bay (only to remove the chill while occupied)
• Install New Cabinets in Kitchen and Laundry
• Close Air Leaks at Window Step

--

Teensy Programming Shield - To FAB Shop

After a little rework on the previous circuit board, we got it working. The +5V trace was broken at the header pin. This type of problem is always a concern with Homebrew Double Sided Boards as they are not through hole plated, and therefore must be soldered on the pads of the two sides. The pad and connecting trace are very fragile.

As Designed and Sent to DorkBotPDX

To make the board more useful (and rugged) we decided to have it sent to my favorite fab shop; DorkBotPDX. The boards should be back (in less than 2 weeks) by Thanksgiving.

The Back Side

The board has been changed a little, as the fab shop can do useful things, that are avoided when doing HB PCB's.


The board is .7'' x 1.8" with 8 mil traces, 6 mil ground grid, and the text font is 3pt.

 I enjoy building very small projects!

--

Teensy Programming Shield

Jeff - KO7M and I have been working on several Arduino and Teensy projects, to control QRP and Beacon Transmitters. To expand and progress to the point where we will use raw chips in planned projects we decided to produce a Teensy Programming (under) Shield. For several projects we have used the Teensy as the development platform.

As Designed in DipTrace

The Teensy will plug into the top of the Shield. After several iterations and corrections of schematic I we have a layout.

Today, I used the Toner Transfer Method to create the board. High resolution two sided boards are tough but they can be done. The Teensy Programming Shield is .75" x 1.75" with 8 mil traces.

Etched and Ready for Inspection

There are not many traces so it was not hard to route or build. If I had need for more than a few boards I would have sent it out for FAB.

The Etch Looks Good
and both sides look like they are in good alignment
These are 8 mil Traces, and 6 mil Ground Grid

After Laser Toner Resist is remove, the board is cleaned and polished, and then placed into a small dish of Tinnit. In about 5 minutes the board is bright and shiny.

Tinnit was used for Tin Plating

Tinnit works well, with good coverage,
but it is very thin (.004").

The Results:

Parts Loaded

Double sided HB board are difficult as the socket pin are typically used to transfer the traces between sides and therefore both sides of the socket pins require solder. With care this can be done by raising the socket a small amount and soldering with a very sharp iron. Also, small errors of alignment between the image of each side make for difficult drilling and therefore pad are not always centered on the component pins.

Cut, Drilled and Parts are loaded, in this case Headers and one jumper.

Ready to Use !

As you can see the Header were raised about 1/10" to facilitate top side soldering the few pads which are connected via traces. Raising the parts would not be necessary if this board was from a FAB shop, plated through holes would make loading parts much easier.

In use, the Teensy will cover the first 24 pins (far end), short jumpers will be used if need for the none standard Teensy pin locations. For ISP Programming effort, only one jumper will be required - the RST pin.

With the Teensy plugged in hopefully there will be enough space to plug in the programmers? To use the Shield to program the Teensy, a short jumper is still needed between the RST pin (on end of the Teensy) and the third pin on the near open socket holes. That hole, is connected to the RST pin on the 6 pin Programming Header. We could have included dedicated RST pin and socket as part of the Shield, but that would require we alter the Teensy with a dedicated pin, disallowing it to be used directly with simple circuits on protoboards.

The Teensy is Installed, Ready for Programming,
It can be used
In or Out of a Protoboard

Now it is time to do some Direct AVR Programming - Jeff!

--

Tennis Ball Launch Trigger Program

Finally, Jeff - KO7M and myself got together to work on my Tennis Ball Launcher Trigger Program (see previous post). The launch trigger program will help avoid accidental firing of the launcher. The launcher, of course, will be use to string antenna lines into the trees. So far, the weighted tennis balls have been shot over 160' trees. Tess, my dog loves the launch activity!

Jeff came prepared with his subroutines to perform most of the required functions. Our time together was used to assemble the mainline program and test the firing sequence functions. My Teensy Test Jig as describe in previous posts, was used to test the program. A few things are yet to be resolved, but progress was good.

An unanticipated value; was the opportunity to watch over Jeff shoulder, as he programmed the task in the Arduino IDE. A lot can be learned by watching a programming Master - Jeff is good! (it is his day job) The interaction is worth much more than just the resulting program listing.

The lessons learned will be used in my future embedded micro projects. - Thanks, Jeff

--

Teensy Test Jig Schmatic

Here is the Teensy Test Jig Schematic that I am using for the programs shown in previous posts.

Teensy Test Jig

Teensy  Test Jig Schematic
Rev: F

Teensy Test Jig - In Fritzing Format

Note: The Rotary Ecoders are show as "Trimmer Pots" with two extra pins at the top for the push button switch, the encoder part is not available in the Fritzing Library, I may need to create a custom part. I have updated the Fritzing diagram above - I had to build a custom Fritzing Rotary Encoder part.

I had to re upload the schematic, the first and second were flawed.

Note: The Photo, Schematic, and the Fritzing view are NOT all "exactly" the same.

This is a "work-in-progress" project.

--

ISR Lessons Learned

If you follow my Blog, (see previous posts) you will know that I have been playing with an Arduino work-a-like known as the Teensy. The plan is use it to help make my Homebrew Projects smarter and maybe more interesting.

So far, I have been "playing" with it to build multi-tasking template for future projects. The first hurtle was to write a predicable Rotary Encoder Test routine. Which I have done and published on the previous post. Actually, the code that I published has been replaced three time as I learn more. Finally I think the code is solid and would not mind others to commit or post reviews.

It was a struggle getting to this point. I have several chats with my friend Jeff - KO7M, about several major issues that were . . . just kicking my butt. The program just did not operate the way I had in mind. But in the end, it now works better than expected.

There were several lessens learned along the way, many are simple and probable known by heavy software types, but for me they were large stumbling blocks.

The lessons learned:

• When writing software it is very important to have a friend to review your code. The process of expainning code bring new eyes on the problem.

• Interrupt Service Routines (ISR) should be as short as possible, I instinctive knew this, but initially did not follow my own advice. It is just too easy to add one more line to the ISR. Note: only the second (PinB) needs to be read here as the other (PinA) is already known to be HIGH, as the RISING edge of the pulse is what created the Interrupt.

void doEncoder0() {    // Rotary Encoder
    digitalRead(encoder0PinB) ? g_encoder0Pos++ : g_encoder0Pos--;
    return;
}

• The variable used by the ISR should be declared as global and type "volatile byte", as:

// Set Initial Encoder Values
volatile byte g_encoder0Pos = 0;

• Jeff suggested the "g_" conventions for global variables.

• The main program routine that takes advantage of the data from the ISR should read it with one machine instruction, so that the process can NOT be interrupted by yet another ISR event. For an 8 bit processor that means use an assign statement between two variables of type byte. In my case, the flag that indicates that the assignment was complete (ISR data extracted), was another byte assignment of value zero to the source variable. Once the data is assigned, it can be accumulated or used as necessary.

int doGetEncoderValue() {
    byte tmp;

   // Get Encoder input
   tmp = g_encoder0Pos;
   g_encoder0Pos = 0;

   . . . . 

}

• Each of these two assignments are NOT interrupt able and therefore does not pose a problem if another ISR event occurs. Yes, there is a very-very narrow window where one event may be lost, but for my Rotary Encoder application that will not be a problem. If I had used; "int", "long", or  "float" data type multiple machine instructions would have been necessary to complete an assignment. And then, if an event occurred during the the assignment, there would have been a good chance the data would have be corrupt.

The methods used within the test program, have already been incorporated into another more complex multi-tasking program that I am working on.

The following is an excerpt from that program:

        byte btmp; int itmp;

        lcd.setCursor(0,0);
        lcd.print("Adj Backlight");
        // Get and Decode Rotary Encoder Data
        btmp = g_encoder0Pos;
        g_encoder0Pos = 0;
                
        if(btmp < 128) itmp = btmp;
        if(btmp > 127) itmp = (btmp - 256);
          
        level += itmp * 16;
        
        if(level < 0) level = 0; // Constrain
        if(level > 256)level = 256;

        lcd.setCursor(0,1);
        if(level<10) lcd.print("0");
        if(level<100) lcd.print("0");
        lcd.print(int(level));
        
        // Some Sound Feedback, just for fun
        if(itmp) {
            tone(speaker, 2000 + 2 * level, 10);
            analogWrite(lcd_BL_pin,min(max(level,4),255));
            g_HoldOffReset = true;
            state=1;
        }

It will be just more fun stuff.


--