Write and debug your Arduino programs on your desktop part 3: Fading

In the first post I said we would be doing a Larson scanner, and all we’ve done so far is make a light that goes back and forth in different colors. That is cool and all, but what about the FADING!

In standard Larson scanner, this is pretty easily done; you just need a way to go backwards for <n> steps and then you just write the bright color at the current spot and then dim it as you go backwards. With the color wheel, however, you would need to keep track of what the colors of the previous spots were and dim that color.

Seems like a lot of work to me.

Instead, we’re going to be building something that I think is a little cooler – a fading LED strip. Set a point to a specific color, and over <N> steps, it will automatically fade to black.

Hmm. It sounds like what we need is some code that can blend from a color to black. We already have a class that can do that – the ColorBlend class. We can just leverage that to do what we want. Here’s a test:

static void TestSingleFade()
{
     FadingLedStrip fadingLedStrip(4);
     LedStrip ledStrip;


    LedColor ledColor;


    fadingLedStrip.setColor(1, 255, 0, 0);


    fadingLedStrip.show(ledStrip);
     ledColor = ledStrip.getColor(1);
     Assert::AreEqual(255, ledColor.Red);


    fadingLedStrip.show(ledStrip);
     ledColor = ledStrip.getColor(1);
     Assert::AreEqual(191, ledColor.Red);


    fadingLedStrip.show(ledStrip);
     ledColor = ledStrip.getColor(1);
     Assert::AreEqual(127, ledColor.Red);


    fadingLedStrip.show(ledStrip);
     ledColor = ledStrip.getColor(1);
     Assert::AreEqual(63, ledColor.Red);


    fadingLedStrip.show(ledStrip);
     ledColor = ledStrip.getColor(1);
     Assert::AreEqual(0, ledColor.Red);
}

We set a color and then each time we call show(), it dims the color down. In this case, the dim count is set to 4, so it will take 4 more steps to dim all the way down.

The code for FadingLedStrip is here:

class FadingLedStrip
{
     int _steps;


    ColorBlender _blenders[15];


public:
     FadingLedStrip(int steps)
     {
         _steps = steps;
     }


    void setColor(int ledNumber, int red, int green, int blue)
     {
         _blenders[ledNumber].blendToColor(LedColor(red, green, blue), 0);
         _blenders[ledNumber].blendToColor(LedColor(0, 0, 0), _steps);
     }


    void show(LedStrip& ledStrip)
     {
         for (int i = 0; i < 15; i++)
         {
             LedColor ledColor = _blenders[i].getCurrentColor();
             ledStrip.setColor(i, ledColor.Red, ledColor.Green, ledColor.Blue);
             _blenders[i].step();
         }


        ledStrip.show();
     }
};

It keeps an array of ColorBlenders – one per LED. When we set a color, we tell the blender to immediately switch to that color, and we also tell it to blend to black of the specified number of steps.

The show() method then walks through all of the blenders and copies the color of each blender to the real strip and tells the blender to step.

Here’s a video of the final result:


Larson Scanner from Eric Gunnerson on Vimeo.


Write and debug your Arduino programs on your desktop part 2: Automated Testing

Read the previous post before you read this one.

In the previous post, I showed how to use hand-verification – and perhaps a debugger – to get your code working. That works well in many cases, but sometimes you have code that you think is going to evolve over time or code where it is tedious to do the hand verification.

The alternate is to automate that verification, using what is commonly known as “Unit Tests”.

Blending colors

The current implementation only uses red, green, and blue. It would be much nicer if it could smoothly change between colors. I’m going to be building a way to blend from the current color to a new color in a specified number of steps.

I’m going to do this implementation in small steps, using a technique where I write the test before I write the code. To start, we need to switch to a new color immediately when the user chooses zero steps.

Here’s my test code:

static void TestZeroSteps()
{
     ColorBlender colorBlender;


    colorBlender.blendToColor(LedColor(255, 0, 255), 0);


    LedColor color = colorBlender.getCurrentColor();
     Assert::AreEqual(255, color.Red);
     Assert::AreEqual(  0, color.Green);
     Assert::AreEqual(255, color.Blue);
}

The test blends to (255, 0, 255) – purple – in zero steps, so the next time getCurrentColor() is called, it should return that color.

The Assert::AreEqual() statements are verifying that the values we get back are the ones we expect; if they are not, a message will be written out to the console.

This test code lives in the ColorBlenderTest.h file.

The code for ColorBlender lives in the arduino project, and looks like this:

class ColorBlender
{
     LedColor _targetColor;


    public:


    LedColor getCurrentColor()
     {
         return LedColor(_targetColor.Red, _targetColor.Green, _targetColor.Blue);
     }


    void blendToColor(LedColor targetColor, int steps)
     {
         _targetColor = targetColor;
     }
};

When run, that produces no errors. In the next test, we’ll do the blend in one step. Here’s a new test:

static void TestOneStep()
{
     ColorBlender colorBlender;


    colorBlender.blendToColor(LedColor(255, 0, 255), 1);


    LedColor color = colorBlender.getCurrentColor();
     Assert::AreEqual(0, color.Red);
     Assert::AreEqual(0, color.Green);
     Assert::AreEqual(0, color.Blue);


    colorBlender.step();


    color = colorBlender.getCurrentColor();
     Assert::AreEqual(255, color.Red);
     Assert::AreEqual(0, color.Green);
     Assert::AreEqual(255, color.Blue);
}

The initial color should be black, and then after calling step(), it should move to the new color. When this is run, we get the following:

Assert: expected 0 got 255
Assert: expected 0 got 255

We get those errors because there is no implementation to make the test work. This code will make it work:

class ColorBlender
{
     LedColor _currentColor;
     LedColor _targetColor;


    public:


    LedColor getCurrentColor()
     {
         return LedColor(_currentColor.Red, _currentColor.Green, _currentColor.Blue);
     }


    void blendToColor(LedColor targetColor, int steps)
     {
         _targetColor = targetColor;


        if (steps == 0)
         {
             _currentColor = _targetColor;
         }
     }


    void step()
     {
         _currentColor = _targetColor;
     }
};

and now, onto two steps. Here’s the test:

static void TestTwoSteps()
{
     ColorBlender colorBlender;


    colorBlender.blendToColor(LedColor(255, 0, 255), 2);


    LedColor color = colorBlender.getCurrentColor();
     Assert::AreEqual(0, color.Red);
     Assert::AreEqual(0, color.Green);
     Assert::AreEqual(0, color.Blue);


    colorBlender.step();


    color = colorBlender.getCurrentColor();
     Assert::AreEqual(127, color.Red);
     Assert::AreEqual(0, color.Green);
     Assert::AreEqual(127, color.Blue);


    colorBlender.step();


    color = colorBlender.getCurrentColor();
     Assert::AreEqual(255, color.Red);
     Assert::AreEqual(0, color.Green);
     Assert::AreEqual(255, color.Blue);
}

and the updated code:

class ColorBlender
{
     float _red = 0.0F;
     float _green = 0.0F;
     float _blue = 0.0F;
     float _redDelta = 0.0F;
     float _greenDelta = 0.0F;
     float _blueDelta = 0.0F;


    LedColor _targetColor;


    public:


    LedColor getCurrentColor()
     {
         return LedColor((int) _red, (int)_green, (int)_blue);
     }


    void blendToColor(LedColor targetColor, int steps)
     {
         _targetColor = targetColor;


        if (steps == 0)
         {
             _red = _targetColor.Red;
             _green = _targetColor.Green;
             _blue = _targetColor.Blue;
         }
         else
         {
             _redDelta = (_targetColor.Red – _red) / steps;
             _greenDelta = (_targetColor.Green – _green) / steps;
             _blueDelta = (_targetColor.Blue – _blue) / steps;
         }
     }


    void step()
     {
         _red = _red + _redDelta;
         _green = _green + _greenDelta;
         _blue = _blue + _blueDelta;
     }
};

That works. One more test to add; we should stop blending even if we go beyond the specified number of steps. Here’s a test for it:

static void TestThreeStepsAndHold()
{
     ColorBlender colorBlender;


    colorBlender.blendToColor(LedColor(10, 0, 0), 3);


    Assert::AreEqual(0, colorBlender.getCurrentColor().Red);
     colorBlender.step();


    Assert::AreEqual(3, colorBlender.getCurrentColor().Red);
     colorBlender.step();


    Assert::AreEqual(6, colorBlender.getCurrentColor().Red);
     colorBlender.step();


    Assert::AreEqual(10, colorBlender.getCurrentColor().Red);
     colorBlender.step();


    Assert::AreEqual(10, colorBlender.getCurrentColor().Red);
}

That fails on the last assert, as it keeps adding and gives us 13. I added some code and ended up with this:

class ColorBlender
{
     float _red = 0.0F;
     float _green = 0.0F;
     float _blue = 0.0F;
     float _redDelta = 0.0F;
     float _greenDelta = 0.0F;
     float _blueDelta = 0.0F;


    LedColor _targetColor;
     int _steps;


    public:


    LedColor getCurrentColor()
     {
         return LedColor((int) _red, (int)_green, (int)_blue);
     }


    void blendToColor(LedColor targetColor, int steps)
     {
         _steps = steps;
         _targetColor = targetColor;


        if (steps == 0)
         {
             _red = _targetColor.Red;
             _green = _targetColor.Green;
             _blue = _targetColor.Blue;
         }
         else
         {
             _redDelta = (_targetColor.Red – _red) / steps;
             _greenDelta = (_targetColor.Green – _green) / steps;
             _blueDelta = (_targetColor.Blue – _blue) / steps;
         }
     }


    void step()
     {
         if (_steps != 0)
         {
             _red = _red + _redDelta;
             _green = _green + _greenDelta;
             _blue = _blue + _blueDelta;
             _steps–;
         }
     }
};

All of that was written without and tested without any interaction with my microcontroller.

Doing something nice with the blender…

The blender by itself isn’t that useful; we need something to drive it through different colors. Here’s ColorWheel.h:

class ColorWheel
{
     int _stepCount;
     ColorBlender _colorBlender;
     LedColor _colors[6] = {
         LedColor(255, 0, 0),
         LedColor(255, 255, 0),
         LedColor(0, 255, 0),
         LedColor(0, 255, 255),
         LedColor(0, 0, 255),
         LedColor(255, 0, 255) };
     int _colorIndex = 0;


    public:
         ColorWheel(int stepCount)
         {
             _stepCount = stepCount;
             _colorBlender.blendToColor(_colors[_colorIndex], 1);
         }


        LedColor getNextColor()
         {
             _colorBlender.step();
             LedColor ledColor = _colorBlender.getCurrentColor();
             if (_colorBlender.isDone())
             {
                 _colorIndex = (_colorIndex + 1) % 6;
                 _colorBlender.blendToColor(_colors[_colorIndex], _stepCount);
             }


            return ledColor;
         }
};

It uses a ColorBlender, and whenever a color blender is done – which is checked through a new “isDone()” method – it will add a blend to the next color in the sequence. So it continuously cycles through the 6 main colors (Red, yellow, green, cyan, blue, purple).

It has tests:

#pragma once
#include “..\Arduino\Larson\src\ColorWheel.h”


class ColorWheelTest
{
     static void TestSingleStepWheel()
     {
         ColorWheel colorWheel(1);
        
         LedColor ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(  0, ledColor.Green);
         Assert::AreEqual(  0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(255, ledColor.Green);
         Assert::AreEqual(  0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(  0, ledColor.Red);
         Assert::AreEqual(255, ledColor.Green);
         Assert::AreEqual(  0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(  0, ledColor.Red);
         Assert::AreEqual(255, ledColor.Green);
         Assert::AreEqual(255, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(  0, ledColor.Red);
         Assert::AreEqual(  0, ledColor.Green);
         Assert::AreEqual(255, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(  0, ledColor.Green);
         Assert::AreEqual(255, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(  0, ledColor.Green);
         Assert::AreEqual(  0, ledColor.Blue);
     }


    static void TestFourStepWheel()
     {
         ColorWheel colorWheel(4);


        LedColor ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(0, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(63, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(127, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(191, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(255, ledColor.Red);
         Assert::AreEqual(255, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);


        ledColor = colorWheel.getNextColor();
         Assert::AreEqual(191, ledColor.Red);
         Assert::AreEqual(255, ledColor.Green);
         Assert::AreEqual(0, ledColor.Blue);
     }


public:
     static void RunTests()
     {
         TestSingleStepWheel();
         TestFourStepWheel();
     }
};

and that makes the Animater simpler. Here’s the running code:

#define NUM_LEDS 15


class Animater
{
     int _last = 0;
     int _current = 0;
     int _increment = 1;


    int _color = 0;
     ColorWheel _colorWheel;


    public:


    Animater() : _colorWheel(20) {}


    void doAnimationStep(LedStrip &ledStrip)
     {
         ledStrip.setColor(_last, 0, 0, 0);
         LedColor ledColor = _colorWheel.getNextColor();


        ledStrip.setColor(_current, ledColor.Red, ledColor.Green, ledColor.Blue);
         ledStrip.show();


        _last = _current;
         _current = _current + _increment;


        if (_current == 0 || _current == NUM_LEDS – 1)
         {
             _increment = -_increment;
         }
     }
};









Write and debug your Arduino programs on your desktop

Working on projects with an Arduino – or with other microcontrollers – can be a lot of fun. It can also be a frustrating experience; you write some code and then you need to wait for it to be compiled, packaged up, downloaded to your microcontroller, and then run. And when it doesn’t work, it can be difficult to figure out why it isn’t working; generally, the best you can do look at the output from Serial.println() or look at the signals on an oscilloscope, if you own one.

As a (now former) professional developer, I was sure that there was a better way, and after some experimentation I came up with the method described in these posts.

Basically, the approach is pretty simple. We are going to structure our software so that there are two discrete parts; a first part that is directly dependent on the microprocessor and libraries and a second part that is generic. The second part will contain the bulk of the code that we will be writing.

And then, we are going to use a generic C++ environment to run the code from the second part and verify that it works. Once we have it working in that environment, we can then move over to the arduino environment and test it on the real hardware.

In addition to making it easier to write and verify code, this will also break our code into parts and make it simpler to understand.

Tools

We will be using two different development environments. For our Arduino code, we need an Arduino environment – what I would call an “IDE”. I’m going to be using Visual Studio Code with the Platform IO package, but you can use the Arduino IDE if you’d rather. Both of those are free.

We will also need a desktop/laptop environment that can run C++ code. There are several good options here; I’m going to stay true to my roots and use Visual Studio Community with C++ support installed (also free) for that development, but you can use whatever environment you would like.

The project

A quick look at my blog will indicate that I am quite devoted to LEDs, so that’s what we’re going to build. Specifically a Larson Scanner:

Or at least something like that; I’m not sure we’re going to get to the dimming trail part. If you want to follow along, you’ll need a microcontroller that can run the FastLed library (I’m going to use an ESP8266 board), and a strip of WS2812/Neopixel LEDs. There’s a nice intro to using FastLED in their wiki here, and I suggest getting something working in your environment using their directions before trying to follow along.

I’ll add a note here that if you are using the ESP8266, Makuna’s NeoPixelBus is a better choice than FastLed as it has hardware support, but FastLed is more popular so that’s why I’m using it here.

First Version

All of the code lives in the LarsonScanner repository. I will try to keep my commits small and informative so you can see what the changes are.

I started by writing a quick and minimal version of the code. It looks like this:

#include <Arduino.h>
#define FASTLED_ESP8266_NODEMCU_PIN_ORDER
#include <fastled.h>


#define NUM_LEDS 15
#define DATA_PIN 3


CRGB leds[NUM_LEDS];


void setup() {
   FastLED.addLeds<NEOPIXEL, DATA_PIN>(leds, NUM_LEDS);
}


int last = 0;
int current = 0;
int increment = 1;


void loop() {
  leds[last] = CRGB::Black;


  leds[current] = CRGB::Red;
   FastLED.show();
   last = current;
   current = current + increment;


  if (current == 0 || current == NUM_LEDS – 1)
   {
     increment = -increment;
   }


  delay(30);
}

Most of this is boilerplate FastLED code. The code itself is simple; it has three variables:

  • current defines the next LED that we need to turn on
  • last defines the LED that is currently on that we will need to turn off
  • increment defines the direction we are moving with the animation

The loop code sets the previous LED off and turns the new one on, and then increments to move to the next LED. If that puts us at the ends of the strip – either LED 0 or LED NUM_LEDS-1 – then that tells us we need to start going the other direction and we negate increment to do that.

Running this code on the desktop

Our goal is to be able to run the code that we wrote – the code to do the animation – on the desktop. But we can’t do this because that code calls the FastLED library, and there’s no FastLED library on the desktop. What we will need to do is provide a way for our animation code to *use* the FastLED code indirectly rather than referring to it directly.

In software development terms, we’re doing what is called “encapsulation”; taking all of the code related to a specific operation and separating it from the rest of our code.

We will take all the FastLED code and move it into a separate class. It looks like this:

#define FASTLED_ESP8266_NODEMCU_PIN_ORDER
#include <fastled.h>


#define NUM_LEDS 15
#define DATA_PIN 3


class LedStrip
{
     CRGB _leds[NUM_LEDS];


    public:
     void setup()
    {
         FastLED.addLeds<NEOPIXEL, DATA_PIN>(_leds, NUM_LEDS);
     }


    void setColor(int ledIndex, int red, int green, int blue)
     {
         _leds[ledIndex] = CRGB(red, green, blue);
     }


    void show()
     {
         FastLED.show();
     }
};

This new class now keeps track of the details of dealing with FastLED. Note that the “leds” array has been renamed “_leds”; that is a naming convention to make it easier to know that it belongs to this class.

Our main code now looks like this:

#include <Arduino.h>
#include <LedStrip.h>


LedStrip ledStrip;


void setup() {
   ledStrip.setup();
}


int last = 0;
int current = 0;
int increment = 1;


void loop() {


  ledStrip.setColor(last, 0, 0, 0);
   ledStrip.setColor(current, 255, 0, 0);
   ledStrip.show();


  last = current;
   current = current + increment;


  if (current == 0 || current == NUM_LEDS – 1)
   {
     increment = -increment;
   }


  delay(30);
}

Better. There are no FastLED details in here, but there are still arduino details that would get in the way of using it from the desktop. Just as we took all the FastLED details and put them in a class, we will now put all of the animation details into a separate class. It looks like this:

#define NUM_LEDS 15


class Animater
{
     int _last = 0;
     int _current = 0;
     int _increment = 1;


    public:


    void doAnimationStep(LedStrip &ledStrip)
     {
         ledStrip.setColor(_last, 0, 0, 0);
         ledStrip.setColor(_current, 255, 0, 0);
         ledStrip.show();


        _last = _current;
         _current = _current + _increment;


        if (_current == 0 || _current == NUM_LEDS – 1)
         {
             _increment = -_increment;
         }
     }
};

That puts all the code in a method named doAnimationStep; we pass in an LedStrip, and it does whatever it needs to do.

Our main program code gets even simpler:

#include <Arduino.h>
#include <LedStrip.h>
#include <Animater.h>


LedStrip ledStrip;
Animater animater;


void setup() {
   ledStrip.setup();
}


void loop() {
   animater.doAnimationStep(ledStrip);
   delay(30);
}

This demonstrates quite well why encapsulation is a good thing; instead of having one main program with different things going on, we have three sections of code; the code that only deals with FastLED operations, the code that deals with the animation, and then a very simple bit of code that hooks them together. If your arduino code is getting complicated and hard to understand, using encapsulation will help immensely.

The desktop version

We are now ready to run our animation code. I’ve created a Visual Studio C++ project named “ConsoleTest” next to the arduino project. My goal is to run the code in Animater.h in this environment, and to do that, I’m going to need a different implementation of LedStrip.h. Here’s what I create in the ConsoleTest project:

class LedStrip
{
public:
     void setColor(int ledNumber, int red, int green, int blue)
     {
         printf(“LED %d: (%d, %d, %d) \n”, ledNumber, red, green, blue);
     }


    void show()
     {
         printf(“Show: \n”);
     }
};

Instead of talking to an LEDStrip, it just writes out the information it is called with to the console.

The ConsoleTest.cpp file in this project looks like this:

#include “stdafx.h”
#include “LedStrip.h”
#include “..\Arduino\Larson\src\Animater.h”


int main()
{
     LedStrip ledStrip;
     Animater animater;


    for (int i = 0; i < 30; i++)
     {
         animater.doAnimationStep(ledStrip);
     }


    return 0;
}

It includes the printing version of LedStrip in the test project, but it then includes Animater.h from the arduino project. The main() function then calls the animation code the same way the arduino code would call it. It generates the following output:

LED 0: (0, 0, 0)
LED 0: (255, 0, 0)
Show:
LED 0: (0, 0, 0)
LED 1: (255, 0, 0)
Show:
LED 1: (0, 0, 0)
LED 2: (255, 0, 0)
Show:
LED 2: (0, 0, 0)
LED 3: (255, 0, 0)
Show:
LED 3: (0, 0, 0)
LED 4: (255, 0, 0)
Show:
LED 4: (0, 0, 0)
LED 5: (255, 0, 0)
Show:
LED 5: (0, 0, 0)
LED 6: (255, 0, 0)
Show:
LED 6: (0, 0, 0)
LED 7: (255, 0, 0)
Show:
LED 7: (0, 0, 0)
LED 8: (255, 0, 0)
Show:
LED 8: (0, 0, 0)
LED 9: (255, 0, 0)
Show:

We are now able to see the calls the animation code would make to the FastLED library when it runs on the arduino and see if it is behaving as expected.

Digression for experienced developers

If you aren’t an experienced developer, you can safely ignore this section.

This technique – which I call “abstraction by include file” – likely looks a little weird. The “right” way to do this in C++ is to define a pure abstract class named ILedStrip with pure virtual functions that are then overwridden by LedStrip in the arduino code and by a LedStripTest class in the console project.

I’ve implemented this technique both my way and the “right” way, and I’ve found that the right way requires an extra interface definition and doesn’t really help the resulting code. And it requires understanding virtual methods. But that’s an aesthetic choice; feel free to make the opposite choice.

Modifying our animation…

Let’s say that we now want our animation to change colors each time it switches direction. Can we write that code and test it without downloading it to the Arduino?

Here’s my crappy implementation:

#define NUM_LEDS 15


class Animater
{
     int _last = 0;
     int _current = 0;
     int _increment = 1;


    int _color = 0;


    public:


    void doAnimationStep(LedStrip &ledStrip)
     {
         ledStrip.setColor(_last, 0, 0, 0);
         if (_color == 0)
         {
             ledStrip.setColor(_current, 255, 0, 0);
         }
         else if (_color == 1)
         {
             ledStrip.setColor(_current, 0, 255, 0);
         }
         else
         {
             ledStrip.setColor(_current, 0, 0, 255);
         }
         ledStrip.show();


        _last = _current;
         _current = _current + _increment;


        if (_current == 0 || _current == NUM_LEDS – 1)
         {
             _increment = -_increment;


            _color = _color + 1;
             if (_color == 3)
             {
                 _color = 0;
             }
         }
     }
};

Basically, it has a color variable that increments each time we switch directions, and we check that variable to decide what color to set.

By examining the output, we can see if the program is doing what we expect. Or we can use the debugger that is built into Visual Studio Community to have the program stop at any line in our code so that we can see what the values of variables are and what code is being executed. That is much much easier than trying to figure out what is going on in code running on the arduino. There’s a nice introduction to using the debugger here.

Tracking state

To verify an animation, we have to read a lot of output and keep track of which LEDs are which colors. We can make that a little easier by modifying our test LedStrip class so that it keeps track for us. First, we’ll need class that can hold the state of one LED:

class LedColor
{
public:
     int Red;
     int Green;
     int Blue;


    LedColor() : LedColor(0, 0, 0)
     {
     }


    LedColor(int red, int green, int blue)
     {
         Red = red;
         Green = green;
         Blue = blue;
     }
};

And then we can use that class in our LedStrip class:

class LedStrip
{
     LedColor _colors[15];


public:
     void setColor(int ledNumber, int red, int green, int blue)
     {
         _colors[ledNumber] = LedColor(red, green, blue);
     }


    void show()
     {
         printf(“Show: “);
         for (int i = 0; i < 15; i++)
         {
             printf(“(%d,%d,%d)”, _colors[i].Red, _colors[i].Green, _colors[i].Blue);
         }
         puts(“”);
     }
};

This generates the following output:

Show: (255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)
Show: (0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(0,0,0)(255,0,0)(0,0,0)(0,0,0)(0,0,0)

I find this approach to be a bit more visual and easier to understand.

That’s a good place to stop for this post. The next post will explore automated testing using this approach.

Part 2: Automated testing


Sequence Controller Part 2–Board design and MOSFET testing…

Having chosen MOSFETs, I went off to do some board design. I’m hoping this will be a very simple design; it needs to provide power to the ESP, connect ESP outputs to the driving MOSFETs, and provide connections for the loads to the MOSFETs.

Here’s the schematic:

image

On the right we have all of the LOAD outputs; we’re using N-channel MOSFETS to switch to ground, so there are 8 outputs plus a ground. Somewhat conveniently – assuming I’ve read the data sheets right – there are 8 PWM outputs on the right side and 8 on the left.

In my WS2811 extender I put both positive and negative terminals for the load on the board, but in this case I don’t have room so only the ground connections show up.

The other two 9-pin connectors – ExtOut1 and ExtIn1 – are for a feature that I’m hoping will be very cool, but it will be oh-so-easier to explain when I have boards in hand.

One question I already had was whether the ESP could put out enough current to switch the MOSFETs quickly enough. The time spent switching is time the MOSFETs spend in their linear region, and the Rds is much higher during that period. The SOT-32 package doesn’t give much opportunity for heat dissipation.

I didn’t have any protoboards to mount the MOSFET on, but I did have some WS2812 LED boards that I made. Two of the solder pads matched and I used a short wire to hook on the third one.

IMG_9584

That’s wired up to the ESP.

IMG_9582

The ESP running very simple code that ramps up to full brightness and then back down.

I then needed a test load. I don’t actually have a good 2-3 amp 5V test load, so this was my first test:

IMG_9581

That’s 5 of my ornament kits stacked on top of each other. At full brightness they are pulling just over an amp, which is my design point (more would be better). I let it ran on that for a few hours, and the MOSFET was maybe a little warmer than ambient, but barely. I threw on my 12V light bulb testing rig, and got 1.5 amps, and it was also fine with that. Two of those bulbs in parallel would unfortunately be 4 times the power which is more than the MOSFET is rated for, so I’ll need a different load to finish my testing.

I am a little concerned that the ESP may have issues driving more than 1 channel as there could easily be 8 (or 16) channels trying to change all at once. The ESP has 16 independent PWM channels and I’m thinking that if I desync the frequencies slightly (say, 500 Hz, 501 Hz, etc.), the transition points for the PWM will generally not be at the same time.

Anyway, I considered that enough of a test to do the board design. I had to do a custom component and footprint for the ESP because I couldn’t find one that matched my 30-pin DEVKIT board.

One of these times I’m going to remember to do a video of the layout process, but I usually enjoy it so much that I don’t remember.

image

The MOSFETS live at the bottom to minimize the length of the traces that carry the most current, and to put them all near the bottom. The high-current traces are 1mm wide; I could likely go to 1.5 or even 2mm but that seems like a bit of overkill for the currents I expect. The driving traces from the ESP are 0.5mm because I want to get charge into and out of the gates as quickly as possible.

There is a bit of creativity on the left side; the pins on the 9-pin connector are quite a bit offset from the ones on the ESP, so I took pin 12 and 13 and ran them up to the two top pins to make the rest easy to layout.

The board is meant to be an “undershield”; I plan on putting female headers in the 15 pin connectors of this board and those will mate with the male headers that are already on the board. The power and load connectors should probably use right-angle headers.

I spun a small order of these boards for testing.






DORMA

Preparation

A few years ago, I got the idea of doing a solo ride along the RAMROD (my 2013 writeup here) route but in the reverse direction, which would obviously be named “DORMAR” (RAMROD backwards). It looked like it was doable from a water perspective, and would have the advantage of putting all of the climbing in the first 75 miles of the ride, rather than putting a huge climb in at 90 miles into the ride. That would be better because it would be cooler. And you could do *all* the climbing, including the last 600’ up to the Paradise visitor’s center which the Park Service no longer lets RAMROD do.

I was excited enough to create a Facebook group and write a Rider’s Guide. Even scheduled a day for it.

And, for two years, I found excellent reasons not to do the ride. Which means that I’m thinking that I really can’t finish the thing comfortably.

This year, my lovely wife offered to drive SAG for me, which means I could ride it solo – which is really my preference – but without having to be wholly self-sufficient for the whole ride. My tentative plan was to meet her in Eatonville (115 miles in, after the majority of the climbing) to have a lunch, and then I would ride the last 40 miles.

That was the working plan, and this time I actually came up with a date – the Monday after RAMROD, July 29th this year.

Further planning commenced. And I realized that while I was really looking forward to the climbing part of the day – and the descents, I love the descents – I was unexcited about 40 miles of mostly flat riding in the heat of the afternoon after all the climbing. 3+ hours in the heat of the day after already doing all the hard work? I knew that I didn’t like the last 30 miles of RAMROD, and that at least was downhill (though with a headwind).

Hmm. What would you call such a ride? It’s a little bit shorter than the DORMAR, so let’s chop a letter off of the name and we’ll call it “DORMA”. It will be 116 miles and over 8000’ of climbing.

So, I spent a few hours creating the following in Inkscape:

image

I am *exceedingly* happy with how that turned out. It’s supposed to look something like this. For those of you who would prefer English, from left to right, Greenwater, Crystal Mountain Blvd, Cayuse Pass, Grove of the Patriarchs, Backbone Ridge, Inspiration Point, and Paradise (yes, it’s not really a Mont, but from the way climbs are named I think that makes the most sense).

So, there’s a 32 mile climb from Enumclaw to the park entrance at Crystal Mountain, then a climb up Cayuse (1700’ over 6.1 miles), a climb up Backbone ridge (1330’ over 5.6 miles), and a climb up to Paradise (2621’ over 12.5 miles). None of the climbs are particularly steep – last weekend I did 4000’ of climbing at 9% – but there’s certainly a lot of it.

For Kim to be SAG, she needs to have some idea of when I’m going to be at various points along the route. That is a bit of a challenge. For the main climbs, I actually have some data from last summer of the first climb, and it showed that I did that section in 50 minutes, which meant I was climbing at 34 feet per minute (at 195 watts, if you care). For the descents and easier climb at the beginning, I based it off other numbers I had or just a decent guess. That gave me the following timeline:

image

That’s 9:20 minutes total, 8:50 riding and :30 resting. That’s a little over 13 MPH average, and while my last run of RAMROD came in at an even 15 MPH, that included 50-60 paceline miles which bumps the average up a bit.

I put this all together in a guide for my wife, along with a map of the Eatonville end location (the Visitor’s center that RAMROD uses as the first stop).

T-1

A list of tasks done in preparation for the ride, in random order:

  • Charging of electronics
    • Headlight to deal with the early start
    • Rear blinkie (these are technically required when riding in the park but are spottily enforced)
    • Di2 shifters (very happy to remember this)
    • MP3 player
  • Downloading of podcasts
  • Inflation of tires (80 PSI for my 700×28 Conti GP4s.)
  • Lubing of chain
  • Replacement of backup rear blinkie
  • Testing of backup backup rear blinkie
  • Clothing Selection
    • My best Castelli bibs
    • A Nike Dryfit underlayer (it’s not going to be very hot)
    • My Rails to Trials Jersey (a bit too big, but bright yellow, so good for visibility and not too hot. And really big pockets)
  • Pack bike bag (I have a cloth shopping bag that holds all of my basic riding stuff; arm/leg warmers (hope to skip these), light vests and coats, shoes, helmet, gloves, etc. This is a “grab and go” bag, so all I really needed to do was make sure I had my gloves and headband, which get hung up to dry after rides.
  • Tightened the BOA retention wheel on my left shoe
  • Food and drink prep (I don’t need much food these days, so it’s a lot more than I need)
  • Three servings of hydration mix (Bio Steel), one in a bottle and two in a pack.
  • A bag each of:
    • Mixed nuts
    • Trail mix
    • Cheez-its (my traditional long-ride fuel)
  • A bag of homemade jerky thawing in the fridge that I hope not to forget
  • A serving of SuperStarch to drink before I start.
  • Put sunscreen, chamois butt’r, and MP3 player/headphones in my small bag.
  • There are only 5 or so turns on the entire route and I have them in my head, so there’s no need to bring any maps or use my GPS to navigate. If I was doing the whole ride back to Enumclaw, I’d probably have done nav for the last section.

    I got a decent but not great night of sleep two nights before, and that is the one that really matters; the night before I typically never sleep well and since I’m getting up at 4AM I’m going to be messing with my REM sleep anyway.

    We headed down to Enumclaw, checked in at the Rodeway Inn, and went out for dinner at the Rainier Bar & Grill. I had a decent burger.

    From wakefullness to ridefullness

    After a malfunctioning AC and a hot crappy night of sleep, I woke up at 4AM for my 5AM departure. I don’t need that much time to get ready, but my eyes are much happier with contacts if I’m up for a bit.

    As part of getting ready, I took a look at the current weather. I had planned my gear based on mid-50s and then warming up as the morning went on – which for me would mean a vest and maybe arm warmers. Even though I’m wearing my “Rails to Trails” jersey which has giant pockets, I have a lot of stuff and clothes take up a lot of space, so I don’t want to take too much.

    51 degrees.

    Yikes. The logical thing to do would be to add my leg warmers, but leg warmers are really bulky and not easy to carry when you take them off (you can stuff them inside your bibs but they will keep you warmer than you like. So, it’s vest & arm warmers and hope that it warms up quickly.

    I drink 3 servings of SuperStarch mixed in water. SuperStarch is modified cornstarch, so go to your kitchen, take about 1/3 of a cup of cornstarch and add it to a glass of water, and drink it down. I’ll wait.

    Nasty, wasn’t it? But I’ve had good luck with it as a time-release glucose source.

    I glance at my watch as I ride out of the hotel parking lot, and it says 5:01. Perfect

    Enumclaw => Park Entrance

    32 miles, 2146’ of up

    As I head out on highway 410 – with front and back lights as it’s still *dark* – it’s a nice and still night and 51 degrees doesn’t feel as cold as I expected. I’m cold, but not that bad. Bodes well.

    The moon is out as a waning crescent; just the smallest slice bright and the rest slightly illuminated from earthshine. Pretty.

    The first 3 miles climbs about 500’, and I warm up pretty well during this section. There’s pretty much nobody at all out on the road; a few trucks heading to the gravel plant for the first mile and then it’s almost empty. I’m listening to Radiolab podcasts on this ride to keep me occupied (yes, I can still hear cars & trucks approaching), and after the first hill section I’m spinning along at about 170 watts. I want to keep a decent pace but I don’t want to use too much energy or legginess (legity), so I’m trying to stay in the sweet spot in between.

    After about an hour, I zip my vest back up because my hands are getting a bit cold. I flip over to another screen on my GPS to see the temperature…

    41.7

    Damn. I can tolerate the cold pretty well and my core temp is okay so far, but my knees do not like being this cold. Nothing to do but press on.

    The rest of this section passes slowly and it does warm up *slightly* as I keep going; the base of Crystal Mountain Blvd – where the National Park starts – is all the way up to 44 degrees.

    Cayuse Pass

    6.1 miles, 1700’ of up

    This section is a big misleading; the first two miles after you enter the park are the same gradient so you think it’s going to be easy, and then the pass begins.

    I haven’t been able to find out who designed Cayuse Pass, but he was pretty bull-headed. Starting at the top point, the route wraps around the hill contours but barely wavers from a constant 6%. That means it’s easy to find a groove and stick to it, but there’s pretty much no variety to be had. I’ve been eating a bit to keep my reserves up; a bit of trail mix and some Cheez-Its. And drinking, despite me not sweating much, as they last thing you want in the mountains is to get behind on hydration. There’s a nice view of the southeast side of the mountain at one point, but I keep climbing. The sun is up but I’m on the west side of the hills and therefore still in shadow.

    Eventually, I finally top out at the top into the sun and 47 degrees. My timeline estimate was that this would take me 3:20 and it actually took me 3:40, which I’ll note is pretty much exactly 10% slow. I was a little lower on wattage than I had expected, but it’s a long day and this is not the part of the ride to try to push.

    Cayuse Pass Descent

    11 miles, 2584’ of down

    The Cayuse descent is a bit like the climb I just finished; 8 miles of 6% and then a flatter 3 mile section. I really like mountain descents, the road is good, and the constant 6% gradient means that I can cruise along at around 31-32 MPH at about 150 watts. I rarely coast on downhills as spinning keeps my legs warm. There are two sections at the top that head eastish and are therefore in the sun, and I warm up a bit, but most of the route is once again on the west-facing side and are pretty cool. My next climbs are going to be on hills facing east, so I’ll have plenty of chance to warm up soon. A glorious 14 minutes of fast descending takes me to the runout section, and I get to the park entrance at 4:04 into the ride, or 24 minutes behind my estimate.

    A young park ranger takes my $15 – she does not offer the “Just go ahead” discount that I got when I climbed Sunrise last year – and I head to the Grove of the Patriarchs stop.

    If you are in the area, this is a great stop; the trees are truly massive and the short loop hike is worth the effort. I’m only here to use the bathroom and to refill my water bottles. My hydration state seems okay so I mix a bottle of BioSteel to replace the one I had just finished and fill my second bottle as much as possible – which is only about 50% given the water fountain stream. I drink a bit extra, refill it as much as possible, and head out after a quick 10 minute stop. 5 minutes of time made up on the stop.

    Backbone Ridge

    5.6 miles, 1330’

    This is the baby climb of the ride, but at 1330’ of up, it’s still quite a bit of vertical. I’m still trying to climb at a reasonable pace, and it seems that my pace today is a bit slow; it takes me 50 minutes to do this climb and I’m only climbing at 441 meters per hour; my usual rate is closer to 600 so this doesn’t bode particularly well. The climb isn’t very steep – only about 4% – and the temp is in the 50s and there are sections of sun. I feel decent, I’m just not riding very fast. Sometimes it happens.

    And my butt is hurting. I’ve had a saddle sore for a while, and it’s flaring up. That means I need to stand up every few minutes, which I do fairly often anyway to stretch my legs but not this much. Both of those are having an effect on my speed.

    On the way up there is a cycling group with matching jerseys that pass me going down. There are few cars.

    The descent is a fun one, and there’s a bit of flat. A miscalculation means that I’m done with RadioLab, so I switch over to music.

    My data says that I’m 22 minutes behind my timeline at this point, but other than a general sense of where I was at the top of Cayuse, I don’t know it at the time. It’s only relevant for Kim driving SAG, and there’s little I could do about it even if I knew.

    Paradise

    12.5 miles, 2621’

    This climb is 300’ shorter than the climb up to Sunrise that I’ve done a few times, and it’s dwarfed by the 5000’ Hurricane Ridge climb on the Olympic peninsula, but that’s still quite a bit of climbing.

    Nothing to do but HTFU and climb it. I feel decent but not strong, so I settle into a constant pace, which later data shows is a disappointing 170 watts.

    I climb, climb, and then I climb some more as I work my way up the ridge. After 46 minutes I take a quick stop to eat some jerky (on this climb I think trying to eat it on the bike will end up using it to decorate the roadway, providing an unexpected protein windfall to the local fauna). This is *not* my jerky – which I did manage to leave in the fridge at home – but a decidedly inferior substitute purchased at a gas station.

    Then it’s back on the bike to ride the rest of the way to the top. During the climb I stand up 19 times to rest my butt.

    Eventually I hit the switchbacks and reach a point where I can actually see the mountain, and then a bit more climbing and I reach Reflection Lakes where it flattens out, and then after a bit of downhill it’s just the short 600’ climb to the Paradise Visitor’s Center. Lots more traffic on that section and I’m pretty toasted, but after what feels like another 10 hours on the bike, at 11:45 I hit the top, where I get off my bike to fill my bottles, take a rest, and look at The Mountain. Despite it being a weekday, the place is packed.

    List of things on Eric that hurt:

    • Back
    • Butt
    • Feet
    • Toes
    • Knees
    • Pride

    My feet are really tender, likely from all the standing, and my knees – which pretty much never bother me even on really hilly courses – are hurting a lot. If I was smart, I’d take a couple of ibuprofen, but apparently I’m not.

    My target time for the climb is 80 minutes, and it ends up taking me 100 minutes, so about 20 minutes slow, or about 45 minutes behind in total. A little of that can be attributed to altitude; the average altitude of the climb was about 3750’, and – looking at some references on Alveolar O2 and altitude – I can calculate that I’m down about 13% on oxygen, and that goes up to 18% at the top.

    Math Pop Quiz:

    Q: It is currently 9:45. You are going to perform an activity that you expect to take 80 minutes. What will the time be when you finish? Please show your work.

    A: Well, 80 minutes is 1:20, so at means 10:65, but that’s not a real time, so normalize it to 11:05.

    Did you pass? I didn’t, as a look at my timeline will show that I ended up with 12:05 as my expected endpoint for this climb. So, rather than being 40 minutes late, I’m suddenly 20 minutes early.

    I obviously didn’t realize this at the time;I just knew that I had beat Kim to the top. I expect this will work out well, as she’s going to come up here and hang out a bit before following me down.

    I know you are wondering what my music was, so here’s the playlist I used from a number of years ago (it’s generated by my Personal DJ program):

    • My World – Avril Lavigne
    • Any Way You Want It – Journey
    • Be My Girl – The Police
    • Holiday – Scorpions
    • Silicon World – Eiffel 65
    • Roll the Bones – Rush
    • Time  – Pink Floyd
    • Sister – Creed
    • When I Come Around – Green Day
    • You Give Me All I Need – Scorpions
    • Analog Kid – Rush
    • Bastille Day – Rush
    • Crystal Baller – Third Eye Blind
    • A Praise Chorus – Jimmy Eat World
    • Tie Your Mother Down – Queen
    • I’ll Be Over You – Toto
    • Doug’s First Job – Uncle Bonsai
    • Out of the Vein – Third Eye Blind
    • Stranger in Town – Toto
    • God of Whine – Third Eye Blind
    • Questioned Apocalypse – One Fell Swoop
    • Always Somewhere – Scorpions
    • Wake Me Up When This Climb Ends – Green Day
    • Warning – Green Day
    • Summer Song – Joe Satriani
    • Cult of Personality – Living Colour
    • Sing Child – Heart
    • Suite Madame Blue – Styx
    • Trees – Tripod
    • It’s Easy (taking it pitch by pitch) – Boston
    • Going To California – Led Zeppelin

      After a bit of sitting and resting, I pick all the cashews out of the mixed nuts I brought, eat three brazil nuts, and head out for the rest of the ride.

      Descent and runout…

      43.8 miles, 5331’ down, 725’ up.

      It starts with about 11 miles of 4% grade and then gets flatter as the section goes on.

      As previously noted, several body parts are painful, but the first part of the descent is what I expect it to be; fast parts with some tight technical turns where I show that I am not the fastest descender in the peleton. As I turn off the top part, Kim passes me going up.

      The descent is just what I thought it would be, and would be fun to ride.

      Except. For. The. Headwind.

      One of the truisms of RAMROD is that the ride back to Enumclaw is always windy. I don’t know if that’s a truism here, but it’s certainly true today. It ranges from a few MPH when I’m in the forest to gusty sections that remind me of the time I rode part of the Kona Ironman course on the Big Island. On an average, it’s cutting off 5-10 MPH from my speed, which is just pissing me off. All the effort to climb up and then I get ripped off going down. When I finally get back outside the park, I text Kim so she has a time check for when I hit a certain part of the course.

      I have Elbe on my mind. On the north side of the road at the west end of town, there’s a gas station with attached store, and there’s an ice cold Coke Zero in one of their coolers, calling my name.

      The wind is doing nothing to improve my mood or reduce the pain in my body, but I can still ride and my power levels aren’t horrible. As a ride leader, I spend a lot of time out in the wind, but this one is nasty and relentless. I pass Ashland and it gets better for a couple minutes, but then comes back with a vengeance. 104 miles is my target, and I slowly watch the miles count up. At 100 miles my GPS loses the decimal point and time slows down.

      I finally get to Elbe, where I get the aforementioned Coke Zero and read the advertisements on the community bulletin board. Good price for tree grinding. I text Kim again, and then head inside to get an ice cream bar. I want something simple, but I end up with a Heath bar crunch. I haven’t eating an ice cream bar in a *long* time. It is sickeningly sweet and not very appealing, but I was raised to eat the food I took, so I finish it and feel a bit queasy.

      I’m getting ready to head out for the the last 12 miles, but I receive word that the organizers have decided to neutralize the remainder of the route, so I instead wait for my team car to show up, and we head to Eatonville to have a nice lunch at the Cottage Bakery and Cafe.

      Thoughts and other stuff

      That was a really hard ride, though if my knees/butt/feet were better, the section to Eatonville would have been simple as it was only another 12 miles.

      I tend to do my long rides solo, but at this distance companionship would have been welcome, though my climbing pace might have been problematic. Omitting the ride from Eatonville back to Enumclaw was a good decision, given my current fitness level. It was fun, for “long ride in the mountains” levels of “fun”.

      Stats:

      Distance: 103.36 miles
      Riding Time: 8:02:11
      Elapsed Time: 8:30:50
      Speed 12.9 MPH
      Power 146W
      Calories 4217
      Whines 3845

      My food for that day was three servings of SuperStarch before the ride, a handful of trail mix, half a package of jerky, 15 cashews, and about 50 cheez-its.

      Stava here, RideWithGPS route here.




        Sequence Controller

        I’m working on a new display for the upcoming holiday season – actually a couple of them – and I need some new controller hardware to drive them.

        Here are the basic requirements that I jotted down:

        • 8 outputs (perhaps expandable to 16)
        • Each output can drive 1 amp at 12V
        • Designed to deal with sequential animation (do this on output 1, do something else on output 2, etc.)
        • Dimming support if practical
        • Easy setup and and configuration
        • Compact & cheap (within reason)…

        With those in mind, I started thinking about components…

        My go-to microcontroller has been the ESP8266 (in NodeMcu mini d1 form) for a while, because it’s so small and cheap. But it’s a bit weak on output pins; you can get 7 pretty easily, but to get more you may have to play tricks. Supposedly you can get to 11 with those tricks which would be okay for 8 but would make 16 possible without some sort of I/O expander.

        Which brings me obviously to the ESP32. Which is honestly a ridiculously capable device; 160 Mhz, 520K of SRAM, dual core (if you need it), Wifi, bluetooth, and pretty much all the I/O support you could want. It’s a little more pricey, about $4 from China in single quantities.

        For this project, it has loads of output pins, and 16 independent PWM channels, which fits pretty well into my requirements. And I’m hoping I can adapt my existing controller software – which is optimized to drive WS2812s – to work in this new scenario.

        MOSFETs

        The switching will of course be handled by n-channel MOSFETS. My WS2812 expander uses DPAK (TO-252/TO-263) packages, which work great but take up a lot of real estate. That was okay for a small number of channels, but for 8 channels I’d like something smaller and I don’t need to be driving 10 amps per channel, which was my design goal for the expander.

        So, my requirements are:

        • 1 amp @ 12V
        • Switchable from 3.3V outputs (I could add a transistor to drive, but I’d rather avoid the complexity)
        • Low Rds at 3.3V
        • Small package
        • Enough power dissipation

        I started doing some parametric searches in DigiKey and on Octopart, narrowed things down, and came across the BSR202N from Infineon. How does it stack up?

        • 3.8 amps @ 12V (25 degrees, 3.1 at 70 degrees)
        • Specified behavior down to 2.5V.
        • Rds of 33 milliohms at 2.5 V.
        • SOT23 compatible package
        • 500 mW power dissipation

        Those specs are honestly ridiculously good, especially the Rds. If I pull 3 amps through one of the channels, that gives me 0.033 ohms * 3 = 0.1 watts. Just a tenth of a watt to switch 3 amps. If I did that with a bipolar, it would be in the range of 1.8 watts (I’d definitely need a heatsink) and I’d lose 0.6 volts in the process.

        In reality, it will likely be a little better than that since the Rds is lower at 3.3V, but I don’t know how much 3 amps will be heating it up and that will make the Rds worse. It will take some testing to see.

        My only big concern whether the ESP32 has enough drive to deal with the gate capacitance while doing PWM, as with PWM it’s switching all the time, and slow transitions mean slower switching, more heating, and therefore worse Rds and more heating. I’ll need to do some testing, but my guess is that with a PWM rate of 250 Hz it probably won’t be a significant problem in typical usage patterns.

        If it does turn out to be an issue, I’ll add a small bipolar in front of the mosfet. That will give me lots of drive for very fast switching plus a higher Vgs for a lower Rds. It will invert the PWM so I’d have to flip things in software, but that’s simple enough. I’m hoping to avoid it because it will require two resistors per channel, so it’s a nice 8 MOSFETs by themselves or with an added 8 bipolars and 16 resistors, which makes building the boards more of a pain (the cost is of the bipolars + resistors is a few cents per channel).

        Expandability

        My current thought is to make the boards stackable like arduino shields, and I think I have a scheme that works.

        I have the ESP32 boards in my hot hands, but I need to get my hands on some of the MOSFETS to do testing. In parallel, I’m going to start the board design.



        Fixing Bally/Williams Opto troughs

        I had a problem with my WCS serving multiple balls, and I thought I’d share the approach I used to fix it.

        I had looked at the switch tests, but the problem was somewhat intermittent so that didn’t really help.

        I pulled the whole trough unit out; that took:

        • Two screws from the bottom
        • Removing the bottom playfield cover below the flippers
        • Removing 5 (?) screws from the top

        That loosens the trough. I then took out the 4 screws that hold the solenoid to the trough, unplugged the connectors, and it was out.

        Test the LEDs

        I started by testing the LEDs. Through trial and error, I found that a 1K ohm resistor and a 5V supply resulted in a current of about 4mA, and since that’s within the spec for most LEDs, I stuck with that. Hook one end of the supply to the common and the other to the individual LED pins, and verify that they all light up.

        They’re infrared LEDs, so you can’t see them, but pretty much any digital sensor can; a camera, your phone, etc. It’s simpler to remove the board from the trough before you do this.

        All the LEDs on my board checked out.

        Test the phototransistors

        Keep your setup to turn on the LEDs as we’ll need it for this step.

        Using a ohmmeter, connect one end to the common and then connect the other end to the pin for the LED that you currently have on. You should see about 4K ohm when the LED is on and something around 1M ohm when your hand is blocking the light. If you don’t see any difference, swap the leads from the ohmmeter around. You may have to turn the lights out to get 1M on some of the phototransistors as you can get room light reflecting into them.

        Work your way down through each LED and phototransistor and verify that you are getting the right settings. If you find one that isn’t reading correctly, or consistently, it is *most likely* a connection issue.

        I would start by verifying the connections; with one lead connected to the common pin, verify that you have continuity to all of the phototransistors; one of the pins on every one should be zero ohms (or close to it) and connected to the common.

        Then repeat that from each of the LED pins on the connector to the non-common phototransistor pin. You should see zero ohms on each of those as well.

        My issue turned out to be a rework issue; the #6 phototransistor was replaced by somebody and they either messed up the through-hole or didn’t resolder it correctly, so it was only making contact on the LED side of the board sporadically. Rather than pull the board off and try to resolder the phototransistor, I added a small jumper wire from the pin to the phototransistor.

        Everything tested fine, and no more double balls.


        Pogo pins + laser cutter = test fixture

        I sell a small WS2811 expander board on Tindie:

        WS2811 / WS2812 Extender 1

        It’s not a particularly complex board, but it still needs to be tested, and at minimum that test requires 9 connections to the board. For the prototypes I just soldered wires on, but for the ones I sell that would not be a good idea, and it would also be a fair bit of extra work.

        I decided to build a pogo-pin test jig, and since the approach I came up with was different than the other approaches I’ve seen I thought it would be worth sharing. I’m going to be targeting my laser cutter for fabrication, though I could have chosen to use my 3D printer instead.

        Pin design and layout

        I’m going to be using Kicad for my examples here; if you use something different, you’ll need to figure out how to do some operations ourselves.

        Here’s the starting design:

        image

        For testing, I need to provide connections to a subset of the all of the headers. I’m going to do the design for all of the headers and then just populate the ones that I need. For many boards, you would have test pads that are unpopulated as the targets for your testing.

        I need a way to get this into a format I can use with my laser cutter, and SVG is the one I’d like. Kicad can export to SVG just fine; you use the Gerber export and choose “SVG” as the format (no, it doesn’t really make a lot of sense). I’ll be using the pins to connect to the copper, so I’m going after the copper layer.

        Once it’s exported, I can open it up in Inkscape for editing:

        image

        I’m going to clean this up to get rid of all the parts that I don’t need. In some cases, the components are grouped and I need to ungroup them.

        image

        What I want to do is put a pogo pin at the middle of each of these. These are the pogo pins that I’m using:

        image

        They are spec’d to have 1mm shafts and they’re quite close to that, so we’ll plan our design using that.

        At this point, we need to account for one of the things that Inkscape does. The UI allows you to set the size of a circle, but the size that you set is the outer side of the circle, and that includes the stroke width. If your stroke width is 0.25mm and you set the circle to 1mm, the actual circle will only be 0.5mm.

        This confused me for a while when I did my first prototype. And then it confused me again when I did this version. The fix is to set the stroke width to 0 – or something very close to 0 – and use a filled circle instead.

        Here’s a picture of a pad and a 1mm red circle I want to center inside of it:

        image

        I need to get that red circle centered inside the pad. Because of the way Inkscape works, you need to start with the small circle on the lower left. I don’t know why. Then do the following operations:

        • Drag select both objects.
        • Choose “align top”
        • Choose “align right”
        • That should move only the red circle.
        • Align center on the horizontal axis
        • Align center on the vertical axis.

        To select the circles I need to drag a region; it doesn’t work trying to select the object. I don’t know why.

        That gives us the following:

        image

        Eventually, I will want one of those for every pin. But I need to do some test sizing first.

        The beam on a laser cutter is pretty thin, but it still has some width. That means if I cut out a 1.0mm circle, I’ll get one that is just a bit bigger than that, and the pins will be loose.

        I’m going to use the row of 6 pads at the top for test sizing.

        image

        You can’t tell from the picture, but these are sized 1.0 mm, 0.975 mm, 0.950 mm, 0.925 mm, 0.900 mm, and 0.875 mm.

        Off to the laser cutter!

        IMG_9553

        The test fixture cut out of 0.1” (2.6mm) maple plywood.

        The 0.875mm hole is the only one that is close; it’s a tiny bit snug in the middle (the laser beam is shaped like a very tall “X”, so the hole is thinnest at the focus point in the middle and a little bigger at the top and bottom).

        Based on the step in sizing between holes, I’m going to size them all to 0.85 mm.

        image

        That’s the completed design. In my laser cutter (a GlowForge), it groups the elements by color, so it’s easy for me to tell it to cut the red circles and not the black pads. If you want to simplify the cutting part, you might want to delete the pads.

        Back to the cutter.

        IMG_9554

        I cut 3 identical plates plus a spacer. The stack will have two plates with holes, then the spacer level where the wires will be soldered to the pins and finally a plate at back. The wires will all come out through the spacer holes on the left. It’s hard to tell from the picture, but the pins are inserted so that they stick out the back enough for the bottom of the pin to be flush when the spacer and bottom plate are added.

        Oh, and that lower-left piece is upside down…

        Next is hooking up the test wires. There are 22 pin holes, but two of them need only to be bridged and seven connected with wires for testing.

        IMG_9560

        That’s a rather poor picture of the wires attached to the pogo pins. After they are all soldered on, the back piece goes on and then I taped it together with blue tape so the pieces are apart.

        That would normally be the last step, but the high-current pins on this board are set up using 3.96mm headers instead of the 2.54mm that my pogo pins are designed for, so those pins just go right through the bulbs. With a little bit of play using a soldering iron, you can get a blob of solder on each of those pins and then it will work fine.

        Here’s the completed tester with a board just sitting on it.

        IMG_9562

        The yellow and green wires come from an ESP8266 that I use to drive it, the red and black are 12V power from a repurposed XBox 360 power supply, and then the white wires are the grounds for the loads (the board provides both positive and ground for each LED, but I only need the grounds to do the testing).

        I made a quick video showing the tester in action.


        MVI_0093 from Eric Gunnerson on Vimeo.


        EagleDecorations Ornament Creation Instructions

        Thank you for buying one of our ornament kits. These are the generic instructions that apply to all of our ornaments; please look at specific instructions for your kits for more details.

        Tools & Supplies

        You will need the following supplies:

        • A small soldering iron
        • Solder
        • Needle nose pliers
        • Diagonal cutters or other tool to trim leads and wire
        • A power supply for the kit your ordered – either a 5V USB charger or a 12V power supply.

        LEDs and Resistors

        To keep LEDs from burning up, we will be including resistors that will limit the flow of current through the LEDs and equalize the brightness between different ornaments.

        Depending on the color of the LED that we are using the the voltage we are using for the ornament – either 5 volts or 12 volts – we will be connecting chains of 1, 2, or 4 LEDs to a single resistor. The instructions for your kit will tell you how many LEDs to put in the chain for each resistor. If there are multiple colors in your ornament, each color may use a different number of LEDs in the chain.

        Creating chains

        Here is an example of creating chains of 2 or 4 LEDs, taken from the yellow star ornament:

        IMG_7062

        Note that the LEDs are placed with the longer lead towards the outside of the ornament. That is the basic pattern we use for all of the ornaments.

        IMG_7063IMG_7064

        In these pictures, we are making chains of 2 LEDs. In the left picture, the longer lead on the closest LED is bent towards the shorter LED of the next LED. In the second picture, the short LED on the second LED is bent back towards the long lead from the first LED. Connections between LEDs should always be done in this manner.

        Here is what it looks like after creating two chains of 2 LEDs:

        IMG_7065

        A 4 LED chain looks like this, with 4 LEDs connected in a chain.

        IMG_7066

        A full set of chains

        The outline of an ornament will be a series of chains; it will look like this:

        IMG_7068

        Adding resistors

        After the chains are created, we will need to add a resistor for each chain. The resistors are always connected to the inside (shorter) lead at one end of the LED chain:

        IMG_7069

        When all the resistors are connected, it will look like this:

        IMG_7070

        Hooking the chains together

        The next step is to hook all of the chains together. We will do the insides first. This is done with some of the bare copper wire included in the kit. Start by taking the wire and bending it into a rough approximation of the template, and then put that inside the wire.

        We will be connected the currently unconnected end of each resistor to the bare wire.

        IMG_7071

        As shown in this picture, you may need to reroute the resistor wire a bit to make it easier to connect to the bare wire. Here’s a close up of that:

        IMG_7072

        Once all the resistor wires are soldered on, trim the resistor wires. Next up are the outer wires. The outer wires run around the perimeter of the LEDs and are soldered to the remaining unconnected LED lead. Make sure the outer wire does not touch any other wires.

        It is very useful to clamp the outer wire down as you are routing it around. I use a little alligator clip:

        IMG_7074

        Here’s what it looks like when finished:

        IMG_7076

        At this point we would test by applying the appropriate voltage to the inner and outer bare wires.

        Adding the power cord

        Locate the power cord – either the USB one with the 5V kit or another one if you are building the 12 volt version.

        At the bottom of the ornament, you will find two tiny laser-cut holes. The are for the zip-tie that will hold the power cable in place. Pass the zip-tie from back to front and then to the back again, place the power cable in approximately the location you want and lightly secure it with the zip tie. Solder the power cord wires to the two bare wires, verify that it works, and then tighten the zip tie. Cut off the extra.

        Success! You have completed the ornament:

        IMG_7080

        Protecting the wires

        The wires only carry low voltage, so there is little shock hazard.

        If you want to waterproof the ornament, I have had good luck with 100% clear silicone sealant. Make sure to cover the base all the LEDs and over and under the resistors and all wires. This approach has survived multiple holiday seasons outside in wet and cold weather, but there is no warranty for outside use.


        Observational Studies and causation

        There’s a problem with observational studies.

        Let’s say you tell people to do something – eat less red meat, for example – you are hoping to change their behavior. You end up with some people who totally avoid red meat, some people who reduce the amount of red meat they eat, and some people who just ignore you.

        Then you come back 10 or 20 years later and do an observational study and look at how much red meat people eat and how healthy they are, and – lo and behold – you find out that those people who eat less red meat are healthier.

        So, you publish your study and a bunch of other people publish their studies.

        Unfortunately, there’s a problem; the act of telling people what to do is messing with your results. The people who listened to your advice to give up red meat are fundamentally more interested in their health than those that didn’t listen in a myriad of ways. Those differences are known as “confounders”, and studies use statistical techniques to reduce the impact of confounders on the results, but they can never get rid of all the confounders. Which leaves us with a problem: we don’t know big the residual confounders are in comparison with any real effect we might be seeing.

        Residual confounding is why those studies can never show causality; if you look at the studies themselves, they will say there is an association with red meat consumption and increased mortality.

        But in the press releases from the research groups or universities, causality is often assumed.



        Pages:1234567...34