The problem with fading is that requires ramping PWM up and down, which requires the MCU to be awake. Awake, the 416 will probably use 1mA (@3.3MHz). Asleep, it should be around 1uA. (datasheet pages 439 and 445, not my measured values). If battery drain isn’t too big of a concern, that definitely is a possibility!
Ok… I tested it in-flashlight. It’s a Convoy M1 with a Nanjg105c & 750 Ohm bleeder. Results seem mostly positive.
Firstly, it works. But a few observations:
- I’ll definitely need to adjust the voltage detection levels. A 4.09V battery gave 6/8 LEDs. A 3.54V battery gave 3/8 LEDs.
- The LEDs are dim during the animation, I’m guessing because the MCU is taking up most of the current. Not a big deal. I might even remove the animation. Not sure yet.
- The first voltage check reads low. This is despite the fact that I take 16 ADC readings and trash them, then grab 8 readings and average them. I guess it’s just because it’s taking the reading on the coat tails of the animation? Dunno, I can probably code around it.
- I can probably reduce the MCU current draw in active mode (ie, animation) by reducing the CPU frequency and making sure I have some other things disabled (like BOD). Freq is currently 3.3MHz, I think. I just rolled with the default, which I think is 20MHz with a divider of 6. Any idea whats a safe, low frequency to use?
gchart,
This is sort of off topic but what do you have your buck driver attached to for power? I’m looking for a similar setup.
An update…
I managed to decrease the MCU power draw considerably by changing the main clock to be driven from the 32.767 KHz internal ULP (ultra low power) RC oscillator instead of the prescaled 20MHz one.
I also managed to get a good grasp on the ADC readings by using the 8 LEDs to display the significant digits of the ADC readings in binary (each LED representing one bit of the ADC reading). It worked really slick. I took 7 measurements across the typical voltage range and got a great trend line with an R^2 value of 0.99.
So now that I have a calibrated formula, I just need to plug that into the source code, re-uploaded, and we should be in business!
Thanks! What are you using as a power source?
Mr gchart I like your project very much! I am looking forward to more update and improvements!!
Give that man a Bells! 
Watching close with interest
Updates:
- I made a few code tweaks yesterday and re-flashed it. I think it’s all done now! I put in several batteries at various voltages and the LEDs are lighting up exactly as they should.
- Switching to using the 32.768KHz oscillator for the clock really helped with power consumption during active state (and a bit in standby as well). The LEDs are no longer dim during the animation, as there is plenty of current available now
- I used LED series resistors with less resistance than what I desired (used 30K, probably should have used 90K or so). With all the LEDs lit during standby, current consumption is 0.35mA (I typically aim for 0.12-0.18mA depending on the LED colors used). I think almost all of that is going to the LEDs. Only an immeasurable amount to the MCU (well, immeasurable with my DMM that only reads to 0.00mA).
- During the active state with animation (one LED lit), the consumption is 0.05mA. That’s actually about how much current I figure the LED itself is consuming… battery voltage at 4.1V with measured LED vF of 2.5V using a 30K Ohm resistor: (4.1V - 2.5V)/30000Ω = 0.05mA. So even during active state (not sleep), the MCU is using almost no current.
Well, now that I know that power consumption during active state can be highly reduced using the ULP oscillator (~12uA), this actually could happen! :+1:
Hey, I think maybe you might have missed post #97 ? Is this what you’re looking for? If it’s not, just let me know.
I love where you’re going with your illuminated tailcap development gchart. I’ve just built one of your LVP tailcaps as my first non-emitter reflow and I’d love to build your voltage indicator design one day.
Even without a code change I think it would look good with four green, two yellow, and two red LEDs.
This is so sexy!
Nice implementation with the MCU! Now how do you take your ADC voltage readings exactly, with the first one being wrong? I might have a simple solution, if the first value affects the average too much.
It seems to be working fine at the moment, but I’d be curious to hear your ideas! Here’s an overview of the code:
Main()
- Initialize the clock and pins
- Run the animation
- Shut the LEDs off and pause for 20ms to allow things to settle down
- Run Set_Battery_Status
- Initialize the RTC to wake us up every 60 minutes
- Go to sleep (standby)
The RTC interrupt just clears the overflow flag and then calls Set_Battery_Status
Set_Battery_Status()
- Turn off all the LEDs
- Wait 10ms for things to settle down
- Run Get_Voltage
- Use the returned voltage to turn on desired LEDs
Get_Voltage()
- Set Vbandgap ref to internal 2.5V
- Set ADC ref to VDD
- Enable the ADC, set it to 10-bit resolution
- Get 15 ADC readings and throw them away
- Get 8 ADC readings and add them together
- Get the average of the 8 readings (bitshift: sum >> 3)
- Use a calibrated formula to get the voltage. Standard formula is (1023*2.5)/reading. My formula is (2643.3/reading)+0.0267.
- Turn off the ADC
- Return the reading
It’s probably a bit wasteful, but I have 4K of program space to work with and I’m just turning on and off a few indicator LEDs so I wasn’t very concerned about peak efficiency.
Good, I think I can contribute a bit there:
So you get 8 readings with your ADC, and then simply average them. I had these ADC reading problems with Arduinos and ATTinys as well, and I found an alternative to averaging to get better readings. I don't know if it requires less or more bytes though. Never tested it.
So the solution: get the median.
The median is the value that separates the higher half , from the lower half. To get that: Do your readings, and put them in an array. Then order them (ascending or descending, whatever you like), and take the value in the "middle" of the array. That way, a bad reading doesn't affect the value you get.
Small tip: use an array with an uneven length. If it would be an even number, you would have 2 middle values and need to take an average of those two.
I don't know how familiar you are with programming, so I'll just write it. Please let me know if you don't need code bits, it takes so long with the advanced post editor.
in C++:
//sorting array in ascending order
int i, j, temp, median;
int length = 9; //uneven is better
int array[length];
for(i=0, i<length, ++i){ //get value in array
for(j=i+1, j<length, ++j){ //get value next to it in array
if(array[j] < array[i]){ //if value next to it is higher, swap them
temp = array[i];
array[i] = array[j];
array[j] = temp;
}
}
}
median = array[(length-1)/2] //for an even length it would be (array[(length-1)/2]+array[(length)/2])/2
There's even a way without using a temporary int, here's just the principle:
int x = 5;
int y = 7;
x = x + y; //x now 12
y = x - y; //y now 5
x = x - y; //x now 7, swap done
Median - nice idea! I’m actually using that same methodology for my DIY lumen tube-logger. Its a Wemos D1 Mini running Arduino hooked up to a good light sensor via I2C. Anytime there is a change, the first reading is typically off. So anytime I do a reading, I actually grab three and return the median. That’s a pretty small sample size, but it works for what I’m using it for. But yeah, median can be helpful for dealing with outliers. I could probably get away with far fewer ADC readings by using it.
Ahh, good ol’ Bubble Sort! My undergrad degree was in Computer Science, and my go-to back then was C (I use mostly database and web languages now).
I assembled another smart tailcap and put it in a S41S (stainless, not colored).
New firmware this time: when starting up, it “dials up” the battery readout and displays it for 3 seconds. Then it goes into a slow rotate animation.
Hmm… I wonder how a pseudo-random “twinkling” animation would look?
I always thought it might look neat to do sort of a roulette wheel animation while waiting for voltage to settle, then take a reading and slow down the animation until it stops at a point representing the current charge. After that, I figured I would have it go into a low-power beacon mode which blinks the result periodically.
That was for a RGB tailcap though. I guess on a 8-of-the-same-color design it would display based on how many LEDs are illuminated.
As designed, I could use multiple colors of LEDs (perhaps 4 each of 2 different colors). I’m not sure how that would work for a roulette animation, but it could be a start. And I considered having this rotating animation pause occasionally to display the battery level… not sure yet. Or perhaps just stop the animation and switch to a beacon once it hits a low-batt threshold? Dunno yet.
PS - thoughts on how to efficiently do an 8-bit pseudo-random pattern? Bitshifting a predefined pattern? Using int rand() from <stdlib.h>?
Pull the code bytes out of flash with an incrementing pointer and output them. When your through, bend it back to the beginning.