Well, it depends. I used to take the polite way of trying to settle things without resorting to threats. But given the last run-in I had with DX's customer support...
Give them a chance to rectify the situation, but file a dispute as soon as they start haggling. If you get the "We can give you $2 store credit, but that's all we can do" reply, you know what to do ;)
Tido, I programmed the same program into the BLF AA-Y4E as I did to the AK-47 board, but in the AA-Y4E it only starts up in the first level and do not react to tapping. I tried to remove brown-out detection in high fuse but it did'nt help. With brown-out detection the light flashes a few times (fast) at each start-up and is then steady.
Can you think why it does'nt react on taps as the AK-47 board? (I have re-fitted the two components that were removed to make space). I suspect it has something to do with slow rising and falling of the voltage to the ATtiny due to the PAM 2803 (but nkildal's BLF AA-Y4E worked).
It sounds like the buffer cap keeps the µC running for too long when the light is switched off. Are you sure you re-soldered the components correctly? Looking at the last picture nkildal posted, the black parts on the left of the µC look like diodes. The lower one seems to be part of a voltage monitoring circuit, which would help draining the cap. If you soldered it back the wrong way around, the cap would be isolated from the circuit and retain its charge much longer.
The diode was unsoldered in one end only. It comes (with kathode) from pin 2 on tiny13 in parallel with a 100kOhm. Where they end I have not found out yet. Perhaps the tiny13 is active in draining the capacitor?. The upper diode feeds the tiny13 pin 8 from VOut on the PAM2803. The only thing I changed from the original SW was the timing (PWM appr. 1 kHz, ON appr. 15 us in lowest mode). The PWM seems to work fine all the way to the LED's negative end, only the tapping, slow or fast, does not work.
The voltage to tiny13 collapses in about 10msec so it is unlikely that the watchdog should count for 2 sec
I'm afraid that I have destroyed the solder pad under one end of the capacitor near pin 4 on tiny13. If that is the reason for malfunctioning then I apologize, Tido.
Now I have no idea of where that capacitor goes to, I can't see the trace disappearing under the tiny13. nkildal, do you by any chance have an BLF AA-Y4E open so you can help me out?
Edit: I think I have found it - it goes to the end of the diode and 100 kOhm from pin 2 on tiny13 where I found no connection before!
I'm at work right now (probably shouldn't be following the forum then :-) ) but I would be happy to investigate it, as soon as I get home.
I'm happily retired, so I can do what I want, but I know your problem 
. As I added in my previous post I think I found it but thank's anyway, now let's see what happens:
Nothing!. Re-connecting the condenser did'nt help.
Perhaps I should try the hex-files that nkildal got to work. Can you mail me them, nkildal?. Then I could see if it's a HW or a SW problem.
Yep - I will send them to you, as soon as I get home...
Thank You for the files and the very fascinating soft-beacon function that You invented. One can almost feel the light beam sweeping by...
Your files worked right away, and mine also, after playing around with some editing - I never found out what the problem was.
About the mode shift:
I got an UltraFire WF501B with MC-E today and loved the mode shift, the mode is stored about half a second after turn-off. You don't have to think about starting a 2 sec period before You can switch to next level.
I think this is ideal in a light with only level shifts. If there are other functions, seldom used (Beacon etc.), perhaps it would be better to have no memory and start up in medium, followed by high - low1 - low2 - Beacon or whatever.
I wonder if there is time to write the eeprom during a brownout?. How do we address that?, Tido, agenthex, anyone?
Release v0.3 is ready for download, grab it here. Here is the changelog:
2010-12-09 Tido Klaassen <tido@4gh.eu>
* BLF-VLD: release v0.3
* driver.c: mode functions can now be en/disabled without rearranging
modeline offsets
* driver.c: EEPROM image configurations containing disabled mode
functions will result in compile time error
* driver.c: added support for battery monitoring and handling of low
battery voltage
* driver.c: added wear levelling to mode switching to protect EEPROM
* driver.c: added timer prescaling config option
* driver.c: added mode functions for SOS, alpine distress signal and
fading beacon
* README: added info on how to set up battery monitoring
It would take some modifications to the stock PCBs for this to work.
You could build a "simple" driver without the extended mode group and disable mode memory with the NOMEMORY #define.
This cannot be done with the brown-out detection. The BOD works by stopping the µC once Vcc drops below the configured threshold, and resetting it when the power comes back. The only difference to a normal start-up is a flag set in a status register. Besides, to use the BOD for switching modes, the buffer cap would have to be chosen with the exactly right value, which depends on things like power consumption of the µC and resistance of the battery monitoring circuit. (BTW, I did a calculation some time ago on how long a 100nF buffer cap would keep the µC running. The voltage divider alone would drain it below the µC's minimal operation voltage in about 2ms. I don't think it's possible to use BOD for mode switching on a stock 101-AK.)
The only way to get this working would be to add a really big cap (10-100µF), remove the voltage divider and add a wire from the battery to one input pin, bypassing the protective diode. This way an interrupt could be triggered when the light is switched off (battery voltage is zero, but µC can still run from the buffer cap). If the power comes back before the cap is drained, it was just a short tap. But this would take some fundamental changes to the driver software to get it working.
Hi Tido,
I think You just disencouraged me to dig further into this. But as I know myself I can't let it be.
I thought about using the Brown-out interrupt vector rather than polling the flag BORF. If I could find out how to catch Brown-out interrupt (I don't know much about this special processor) I would turn off the load and count 4 watch dog interrupts (1 sec) and then store the tap_none flag to the EEPROM. I thought that it could be that simple.
I went through my collection of lights with memory and found 7(6) that saves the mode after power-down and only 3(2) that does it after power-up:
Store mode after power-down:
TrustFires: EF23: 1 sec, F20: 0.5 sec, Z1: 1 sec
UltraFires: C3 SS: 10 sec, another C3 SS: 5 sec, 501B MC-E: 2sec
Skyray S-A1: 1 sec
Store mode after power-up:
Tank007 TK-566: 2 sec, TrustFire R5-A3: 2 sec, Romisen RC-C6: <1 sec but always go to next.
Now I should disassemble the first group and spy on the configuration, but the only thought of it......
I am not the only one that like store after power-down (=only one tap for next mode), this is from brted's review of UltraFire 501B-MC:
I'd be glad if you proved me wrong, as I also prefer this kind of mode switching. But the PCB needs to be designed for this and the available driver boards using an ATtiny just aren't.
Using an interrupt routine to trigger on the BOD is impossible, as there is no such interrupt. When Vcc drops below the threshold, the µC's internal reset line is pulled low and kept this way until Vcc reaches operating level again. BORF is there to inform the program that it is not starting up from a complete loss of power. This is useful, if there are peripheral components that need to be initialized or explicitly reset in such a situation.
That's too bad, but thanks anyway for clarifying thing to me 
Could'nt we just use a diode from plus on ATtiny (pin8) to a capacitor to ground, then a few hundred kiloOhms from the capacitor to an unused input pin, fx. PB3 (pin2) . Then at power-up, if the pin reads as '1' there was a short tap and if '0' there was a longer pause.
Sure, that would work, if it takes longer to charge the empty capacitor to logic high level than the program needs for starting up and reading the pin. Or we could charge the capacitor through the pin while the light is on and read the state on start up. I just measured the resistance of the I/O pins when the chip is powered down and they are tri-stated, so there would be no discharging through the ATtiny.
Problem is, this would require altering the PCB and a significant rewrite of the driver software.
Yeah, something strikes me!. Do you remember my problems with the Manafont BLF? I had unsoldered two components. A diode was going from pin 2 in parallel with 100 kOhm, they were connected to the capacitor at the other end of the ATtiny and nowhere else.
I think that those components are just what we are talking about!!!!
Yes, the capacitor could be rapidly charged across the diode and slowly drained via the resistor. Although I don't see how the capacitor would lose its charge in an exactly defined time if the ATtiny was switched off.
It is a small capacitor, perhaps 100 nF, and there will always be a (very) small leak current into the powered-off ATtiny port. About accuracy, look at the different timings on my lights a few post's backwards, there are two 'identical' C3-SS, one with 5 sec and one with 10 sec untill storing the mode. They properly have equal circuits.
I can see no other use for this circuit than providing this storing-timing. I really think we spotted 'how they do it'. 
According to the datasheet, there is a maximum leak current of 1µA into a tri-stated I/O pin at 5.5V, which equals a minimal resistance of 5.5MΩ. I measured ~12MΩ on one of my chips. A high level is >= 0.6 * Vcc, so if we take these two values as lower and upper bounds, and assume the capacitor is 100nF, this circuit would register taps from 0.3s to 0.6s in length. That sounds plausible.