Flashlight Firmware Repository

Hi, just wanted to mention that the code repo now has some attiny25 tools and firmwares. It’s only a small amount so far, but it’s a start.

TK, I was still puzzled by the much lower than advertised PWM frequencies we see (and hear) on the BLF A6 and re-flashed Nanjg drivers.
I get 13 kHz - 14 kHz, as opposed to the expected 18.75 kHz.
The delay_loop_2 function that is supposed to take 4 CPU cycles is also way off, as you probably noticed.

I stumbled across this document:

There is also this in 17.3 of the data sheet:

It seems the specified 10% clock accuracy is only valid if using the default 9.6 MHz internal oscillator (and at 25 C and at 3 Vcc).
Maybe I am just late to the party, otherwise I hope this bit of info is helpful.

A fix, without getting into the user calibration procedure, would be to manually write the 4.8 MHz calibration data (it is there, just not used) to the calibration register.

Personally I decided to run at 9.6/8, phase-correct PWM only (2.352 KHz). Any higher and the PWM signal turns ugly on an oscilloscope. Probably not a good option if you want to avoid noise with a high-current driver. Also slightly higher MCU power consumption.

PS. Your s7.c firmware got me started in this. Thank you :slight_smile: !

Hi TK,

Thanks for this thread and the work. I was wondering if you might include Tido’s work in the repository (or maybe it’s there and I missed it), the BLF-VLD:

Jim

I didn’t have that in the repo, but it’s included now. :slight_smile:

I’m not sure how I missed that one. It looks like it’s pretty decent, has some interesting tricks, and is free under the GPL.

Thanks!

This probably explains why the attiny25 appears to run so much faster than the attiny13. And why selfbuilt measured “15 kHz circuit noise” instead of 19 kHz. I expected the tiny25 to be 166% as fast, but I measured about 210% instead, perhaps because the 13 is running slower than expected. I suppose I should probably figure out how to activate the attiny13’s 4.8 MHz calibration data…

As for 9.6 MHz / 8 (2.3 kHz PWM), it’s not used because faster clock speeds use more power… and because I find 2.3 kHz PWM both loud and obnoxiously slow. Anything under 10 kHz is hard for me to not see, so I try to keep the pulses fast enough that they’re a little beyond the human perceptual range.

BTW, how do you measure the PWM speed? Above ~3 kHz I can’t usually get a loud enough signal to measure it with a microphone. I’ve been tempted to try an optical sensor instead, maybe wiring up an LED in reverse, shining a light at it, and sending its output to a nice DMM, the DMM might be able to measure voltage pulse speeds… but I haven’t tried it.

I've used a scope in the past to measure high PWM's, just needed help from guys here @work to set it up. Great though seeing the exact waveforms, sliding the markers to get a precise measurement - pretty cool.

For the 25/45/85's, I'm really liking the 8 Mhz operation - the timing for every driver/MCU I tried has been dead-on accurate with no calibration -- 90 secs turbo timeout is accurate to within a second, and even that slop is probably me using my stopwatch.

I know it must be using more power than the 4.8 Mhz 13A, but thinking it's pretty insignificant in terms of LED ON and standy mode power - dunno the detials though.

I’ve been wondering if I should use 6.4 MHz instead.

At 8 MHz, the PWM should be 31.25 kHz, which is a bit fast. Or 15.625 kHz with phase-correct. At 6.4 MHz, it would be about 25 kHz or 12.5 kHz.

However, 8 MHz seems to work well enough in my initial testing. BLF-A6 runs fine on it, and moon mode even seems reasonably stable.

Yikes, that PWM is pretty darn high. Didn't realize that - good or bad? Hhmm...

Hi,

Thanks. BTW, it appears you looked at the code? Might you have any idea why the “fixed mode” version (not the “simple” one and not the “extended” one) might be giving me only two modes (kind of “high” and “medium”) and it seems to be skipping clicks, i.e., I’ll click, click, click, and then light will light then click, click…… click and then it goes off, then click, click…. and then different mode. Its like it’s missing the clicks?

Jim

I am still debating the 2.3 kHz PWM. I can see it when flicking a credit card, otherwise not. But I much prefer the resulting signal on the oscilloscope. Power consumption also seems lower actually. I get ~1 mA in between the flashes of a 5 s beacon. A 4.8 MHz clock gives ~2 mA. (I am running the MCU with the 9.6 MHz and 8 pre-scale fuses set, so 1.2 MHz.)

With an oscilloscope it is of course easy to measure the frequency. I also have a few FFT apps on the phone that seem to be reasonable accurate (E.g. FrequenSee HD and Speedy Spectrum). FFT with logarithmic amplitude scaling is the way to detect frequencies. The base and harmonics really pop out. They have no problem picking up the BLF A6, especially using your black cloth trick.

Using a regular 5 mm diode as a photo detector does work with the oscilloscope (I use a green or red one). Mode 2/7 on the A6 gives a ~0.5 Vpp signal with ~1.5 V DC offset. Just the bare LED between the scope probe and ground at 2” from the light. Neat to check strobe frequencies etc. without opening the light.
I just tried it on 3 different multimeters, it only worked on an older one.

I posted on my other thread that I just tried nlite.hex and that seems to work well (3 modes) on one of my NANJGs.

Does anyone know which ones are designed to work with NANJGs/7135 drivers? I’ve tried a few and many behave similar to the BLF-VLD with NANJG, i.e., the modes are somewhat inconsistent.

Thanks,
Jim

I didn’t look at the code very closely, so I’m not sure how the UI works. However, I think it uses on-time memory… so the only “clicks” which really count are the ones which happen immediately after it gets power. I’m not sure what the on-time timeout is though, or how exactly it handles memory.

Mostly, I noticed that Tido used some interesting code techniques which are totally different than STAR. It looks like he has done a lot more C programming than I have; mostly I use Python and shell so my C is rusty.

Something else I noticed is that it only uses 3 slots for wear-levelling… so it’ll wear out the eeprom about 10X as fast as STAR or 20X as fast as blf-a6. Most of the eeprom is used to store the mode table, instead of putting that in regular ROM or RAM.

Anything tagged with attiny13a and ‘ontime’ memory should work fine on a nanjg.

Yeah, 2.3 kHz seems too slow. Even 4.5 kHz stands out to me during regular use, and I can see up to about 20 kHz with the credit card trick. (if I look really closely, I can just barely see the 31.25 kHz PWM on attiny25)

I wonder if the clock divider puts the MCU to sleep for 7 of every 8 cycles. That would explain the lower power use. Given Atmel’s design goals, I bet it does something like that. Or, it could actually be running at a slower speed, underclocked.

Now that you’ve confirmed it works, I really want to try the LED photosensor trick. Until today it was just a random idea I had, but hadn’t actually tried… I’m pretty sure my DMM has some sort of relevant mode for that, though I haven’t explored that yet. It’s a Fluke 8846A which is mostly used for automated testing at work, but I can do other stuff with it between tests.

The more I look at this, the more I realize full power PWM is not the way to go.
LEDs really crave current regulation. If only we had an LD-2 with an ATMEL MCU :slight_smile:
In the meantime I will hack an 8x7135 into two or three controllable groups of 7135s, as others have done. Only first group with PWM to get the very low modes.

For the photosensor, just connect the LED between the ‘LO’ and ‘HI’ inputs on your meter and select frequency. If the signal is large enough for the frequency counter to trigger, you will have a reading. If not, the LED signal will need to amplified and the DC offset removed.

Well, that was ridiculously easy. I should have tried this sooner!

Step 1: Connect an LED to the DMM. I used an old XM-L T6.
Step 2: Press the DMM’s “FREQ” button.
Step 3: Shine a light at the LED.

Instant and precise PWM speed measurement! :slight_smile:

I measured a few things…

  • Two production BLF-A6 units: 15.49 kHz, 13.35 kHz
  • S10-Ti w/ MELD: 488 Hz (same speed my phone mic measured)
  • Blackshadow Terminator: 187.9 Hz (same speed my phone mic measured)
  • CNQG Brass Beauty w/ brass-edc.c: 16.84 kHz
  • Stock BLF EE X6v1: 9.38 kHz
  • Stock original Convoy S3: 4.43 kHz
  • iTP A3 EOS (final model): 2.45 kHz
  • CNQG brass AA (stock): 304.2 Hz low, 1000.0 Hz med
  • SK-68: 128 Hz… then 127… then 126… decreasing with battery charge
  • And an attiny25 test host: 31.74 kHz

I was also able to measure the party strobe modes on my EDC:

  • “12Hz mode”: 12.4 Hz
  • “24Hz mode”: 23.9 Hz
  • “60Hz mode”: 64.6 Hz

It’s neat to see that I got them pretty close through guesswork.

Hi,

Could someone build the .hex for luxdrv and for STAR?

Thanks,
Jim

I’ve been able to build a .hex - just using the minidrv for now.

I’ve been messing around with the modes, and I was wondering what is the minimum value for that in the array? From the comments, it looks like minimum is 5, but is that the lowest mode I can get? If so, it still seems kind of bright, so how do they get those moonlight modes that are lower than that?

Thanks,
Jim

TK,

Apparently nickelflipper had an alpha of his NANJG ramping driver:

Should that be in your repository (you were on that thread too :laughing:?

Jim

Ah, good. It looks like you figured out the build process.

The PWM values can go from 0 to 255, but the minimum depends on your exact hardware. For moon on a 7135 chip, I normally use a value of 3/255 or sometimes even 2/255, both running with phase-correct PWM. For fast PWM, the minimum will be a bit higher. For a FET driver, the minimum will be lower.

Maybe. However, it’s only an alpha debug version and there is no source code available.

Hi TK,

For the Dr. Jones minidrv, he has:

Is that 9kHz PWM considered “fast PWM”?

If it’s not, then would “1” or “2” work? [I guess what I’m asking is if there is some minimum value at which PWM stops working?]

Also, does the number of 7135s on the NANJG (in my case, there are 12 of them) determine how low the moonlight would be?