Attiny25/45/85 FW Development Thread

Mike - ahh, always forget bout you… Of course! Are your designs (boards and firmware) open/available? The board design may be so different, there’s not much in common. I don’t use PWM’s on the FET much - for typical 5 modes, only mode #4 uses PWM on the FET. I can cause hickups going from full FET mode, to OFF, to moon mode on the 7135 for example, so PWM’s on the FET is never used in this sequence.
The bank of 7135’s typically means the 2nd highest mode is off the 7135’s, while ours is PWM on the FET — that’s about the only difference, but still, a big difference. Going to 7 modes or more would be more PWM FET modes.

I’d like to see this design from you. We need all the help we can get. :wink:

Wouldn’t it overheat at some point? I think that somewhere in the middle range it would need to get rid of close to 1W. Hardware is not my field but I expect that would require some reasonably good heatsink or a different package (or both). And if I’m not completely lost, this could be avoided by switching to PWM above perhaps 300mA or so.

No, not all - negates advantages e-switch firmware has, plus I've seen these glitch's happening with clicky lights/firmware as well. The BLF X5/X6 v2 problems were apparently solved by changing the cap from 10 uF to 12 uF. On clicky switch lights w/25 I've built, I usually go with 2 stacked 10 uF caps to fix the issues.

You have to be very careful operating FETs in their linear region. Most FETs are only specified for pulse/saturated operation. When running them in the “linear” region you can run into all sorts of issues. Hot spots can develop and kill the device, etc. Check the safe-operating-area specs/curves carefully… I know a lot of people were unpleasantly surprised when they tried to use the wrong FETs in adjustable power supply load testers.

I’ll have to take your word for it. My worry was that you’ll take the power being dissipated by eight 7135 and now have it all in a single FET. It seems like a lot to me but I’ll admit it’s just a very rough guess, I haven’t looked up any specs. I might try it and come back with pictures if I burn something :slight_smile:

Granted, but it’s convenient because you already have a heat sink for the LED anyway. Also, it’s not obvious to me that running the LED at very high current is much worse for efficiency than wasting power in the FET or 7135s. This probably depends a lot on the LED involved.

Anyway, how would you go about it? All I can think of for good control over the whole range is to build a crude 1bit DAC for the FET gate. I think it could be really poor if feedback from one of the ADCs is used to hold the output steady instead of having to rely on precalibration.

Hardly… the FETs designed 25+ years ago tend to be quite different from more modern FETs (e.g. smaller process geometry, cell sizes) You have to look closely at the spec sheets to see how they will work in DC/linear operation. Assuming your shiny new FET will work the same as your 25 year old beastie is not a good idea. Now a flashlight running at 5 amps may not be an issue, but if you are using a tiny SOT23 fet, it may be. The only way to be sure is to check the datasheets… and often the relevant info can is very well hidden / non-obvious.

I have designed power controllers for up to 7.5 megawatts. My CD spot welder does 500+ kW pulses. At those power levels the devil is in the details. Magic smoke / silicon plasma wants nothing more than to be free.

Led4power made some drivers which work this way. He has a whole family of drivers available for purchase.

Efficiency was better and there was no pulsing in the output, but the driver itself has overheating issues in the medium-to-high modes. It’s fine on low and on maximum output, but in-between it tends to cook itself. Moon (~4 lumens, not 0.2) operates separately due to problems doing moon from the FET. I recall some issues with “low” mode too, like it was a little unstable and it’s ~60 lumens instead of a single digit and it can’t really go any lower (probably for the same reason the moon mode operates separately).

Unfortunately, led4power hasn’t released any designs or code and the MCU isn’t a type which can be reflashed on-board. It’s a nice product, unsurpassed in several ways, but it hasn’t been very widely used.

Based on led4power’s drivers and the test results for it, I think a linear FET would be a nice way to increase efficiency on modes between ~80 lumens and maximum output, if the driver has good heat sinking. A separate power channel would still probably be required for modes between 0.1 lm and 80 lm though.

Some common code has already been exported into shared headers to avoid duplication. Other parts tend to be different per project, and very different for different types of switches. There is still more de-duplication to be done, but this is very limited due to the “tiny” nature of the attiny.

Nothing available yet. I know a lot of people advocate releasing often but I have had enough talks with people that prefer to wait for a version that is tried, tested and approved. I have a whole pile of old obsolete versions of driver boards that are of no use to anyone. I would hate to have had people waste time and money on those versions, only to hear me say that I don’t support them anymore because I have a new design. Also, I have the FET on the dreaded reset pin (all other pins occupied). It took me a lot of swearing and hair pulling before I managed to build a device for high voltage fuse resetting and get it working. Updating the firmware is by no means an easy task if you do not have a working HVSP fuse resetter, because then you have to make/get one and figure all that crap out.

There are two major differences. First, I have E-switch, voltage monitoring and off time cap all on the same pin. This requires a whole different approach to the firmware. For example I have to have delays on the voltage monitoring because of the off time cap charging and E-switch presses. I have to run the watchdog interrupt at full speed in order to detect E-switch presses instantly, and disable all voltage monitoring until a moment after releasing it. The voltage divider resistors are much higher than “regular” ones as R2 on the divider becomes a bleed/pull-down resistor for the off time cap.
The second difference is that I run any number between one and eight AMC7135s fully on (constant current) and use a single PWM dedicated AMC7135 for all brightness levels between. So all of my mode values consist of two bytes, a PWM value for the PWM decicated AMC7135, and then the amount of AMC7135s I want fully on. It’s a constant current/PWM hybrid.

I have had no such issues. I don’t have any gate resistors or capacitors on the FET, and can happily switch between moonlight low PWM values on the PWM AMC7135 and full blast on the FET without hickups, using both off time mode switching and E-switch mode switching. I’ve been following these discussions with interest (where to put the input cap and so on) but as I have not had any of these issues myself I have nothing to contribute towards solving them. The FET I use is the NXP PSMN3R0-30YLD if that matters.

Except awesomeness, and possibly some of the answers. :slight_smile:

Seeing the differences in board design might make it clear why one design has issues while another does not.

Plus, the things you built by yourself are some of the most fascinating and exciting projects on BLF. You may think of your old boards as obsolete junk, but the ideas within, and the details, could potentially be the basis or inspiration for all sorts of other projects. It’s not just the final revision which matters; the things you learned during development would also be beneficial for other people to know. And you’ve downplayed the value of your code by saying it requires different hardware and different approaches, but variety is exactly what BLF needs to grow and move forward.

It’s a good idea. However the FETs don’t perform well at moon to low levels in either PWM or linear modes, so a separate power channel is still likely required to get those levels to run well.

I’d have to check again, but I think led4power’s drivers use a separate thermal sensor to regulate themselves since heat is so much more important on a linear FET design. As for learning algorithms or PID, I looked into PID while doing the bistro thermal code. A fairly simple PID algorithm doesn’t fit onto the MCU, much less a nice-quality PID algorithm. DrJones recently tried to do it too, reduced the size until it would fit, and found that it behaved very badly. He ended up taking the same approach I used, which is basically just an incremental correction factor with a really strong lowpass filter to avoid bouncing.

So, it can work, and it improves efficiency quite a bit on medium modes. However, it also adds some significant complications. Eventually someone will probably decide the complexity is worth it though, and dive in. We already know it works; it just needs someone to do it.

There are other interesting ideas to explore too, like boost and/or buck drivers. It’s possible those might be developed first, since at least one is already in progress. There’s also the many-7135s approach already in use, which provides some of the linear FET’s benefits but without the heat issues. All sorts of interesting stuff happening lately.

comment for subscription purposes. Enjoying the conversation and brainstorming going on in here.

I’ve seen PIDs implemented in a TINY13’s…

Check out Dale Wheat’s LED dimmer code. He uses a PID (actually a PD) for doing smooth fades in his LED dimmer. Link to the source code is in the description. I’ve used his dimmer for lots of stuff. Modified the code and for a TINY85 to control a 5000 watt composite curing oven using a PID. I modified his circuit board with a better FET/heat sink/regulator. Also modified one to do PID based control of a fan prevision temperature control of ovenized crystal oscillators. By changing the regulator to an LM2936 and a couple of tweaks to the firmware the off-state current draw is under 10 microamps.

http://www.dalewheat.com/news/12v-dimmer-kit-v2-now-available/

Cool. I’ll have to look at that in detail next time I’m working on the code.

I didn’t go into much depth when I checked out PID… just deep enough to see that the reference code was orders of magnitude bigger than I had room for, so I went with an extremely simple approach instead.

Edit: I was too curious, so I looked at the code. He’s doing pretty much the same algorithm I did for the “soft start” thing in bistro, which makes smooth transitions between different output levels. That’s not really sufficient to make thermal regulation behave, but it works well for traversing arbitrary adjustment distances in constant time with a smooth-looking ramp.

Edit 2: The “soft start” and dimmer adjustment algorithm is:

  1. Are we there yet? If so, stop. Otherwise…
  2. Go about a quarter(ish) of the remaining distance, minimum 1 unit.
  3. Go back to step 1.

Looking around a little more, I think I might have found a potentially-useful approach. Thermal regulation is rather unpleasantly like trying to steer a really heavy ship. The high degree of inertia makes it hard not to over-steer. Bistro’s solution is to steer very very slowly.

This looks promising:

I know it doesn’t do much good either with or without source, but INFERION claims to have implemented PID in the Indigo 5.x firmware. Indigo 5.0, или "не совсем форумный" драйвер? (Часть 2) | Мастерская: INFERION

Hopefully I won’t put my foot in my mouth, but it seems to me that PID shouldn’t require a lot of space. Here is an Atmel document on the subject (with code provided in a zip), as linked from the ATtiny85 docs page. Atmel claims that their PID function compiles to 534 bytes.

http://www.atmel.com/images/AVR221.zip

Well, I do have some true PID code for my reflow oven. It runs on a MEGA2561 and uses floating point code. It is pretty much the same PID/fan control code I used my GPSDO (GPS Disciplined Oscillator) temperature stabilizer. I have seen it control temperature with a 24 hour RMS error of 1 microdegree, with peak errors of less than 5 millidegrees… Might be a bit much for a flashlight.

A simple temperature control PID is not much more complex than Dale’s code. Implementing the I (intergral) term can be a bit tricky… you have to handle “integral windup” where the integral (long term compensation) value keeps accumulating in the same direction and “runs away”

Tuning a PIDs control parameters is always a pain. Most simple method is to start with P/I/D values at 0. Increase the P value until the output starts oscillating and then increase the D value to stop the oscillation. Then adjust the I value to minimize long term drift. My oven / GPSDO code has some auto-tuning code to that does a lot of the work.

Hmm, 534 bytes. Not bad at all. Certainly better than the ~5k bytes one I looked at. :slight_smile:

I’ve got a copy of the Indigo 5.2 firmware, but it’s a little … opaque. Assembly code with sparse Russian comments. I think Atmel’s version will be a more useful reference, and another Kalman link I found which deals specifically with compensating for situations where one or more values lag behind.

It wouldn’t be a big deal to devote an extra ~500 bytes to PID on a tiny45/85. I wonder how much the response curve can be improved with PID and/or Kalman instead of the “steer really slowly” approach, and if it can anticipate over-heating conditions and start to correct for it in advance. I also wonder if it’ll need to use the full 10 bits of precision instead of 8, since temperature measurements are rather coarse so far.

Too many questions/unknowns. I need to just do it and find out. … … after I get more parts and some time.

Belive me, you don’t want to go down the Kalman rabbit hole. Insanity awaits those foolish enough to enter that dark realm. I’ve been there and barely escaped with my life. Bits of brain still ooze out of my nose whenever I think about it.