[Technical discussions] Sofirn SP35 - SST40, 21700 size, buck driver

My updated SP35 with “working” ATR showed up yesterday. I still think the SP35 is an amazing light & value. That said… my first runtime test is underway and (as expected) I’m seeing the same swings that Lux-Perpetua observed. I wish a little more focus was put to improving the ATR algorithm to help smooth this out.

The peaks are, on average, 1 minute and 45 seconds apart. The ramp up and down is very smooth; swinging between ~500 and ~1000 lumens. So in real life, it’s really not as noticeable as this graph might make you believe.

Once the tests are done, I’ll overlay graphs of the ATR and non-ATR versions and post those up.

I wonder how it would behave with fan cooling? Just thinking about which PID parameters should be adjusted. My guess is the fan wouldn’t change the amplitude much, just the mean output. Could be wrong…

On second thought. It might ‘raise the valleys’ by cooling the light faster once current is pulled back. This is of course based on an assumption that light to air conduction + radiation is much slower (ie limiting) than internal conduction.

I’m with you on the second part. It would likely “raise the valleys”. With regards to PID, I doubt that is relevant here. I imagine it’s more of a “ohh, too hot! now too cold!” algorithm without a bunch of extra math/prediction.

I bet you are right. In that case. There could be a simple two stage method. Let’s assume its 50c target. Set a ~2 degree window with a low gain value, then beyond that a high gain. If these two gain values are tuned, oscillations would reduce like a pid initially and then reach a continuous low amplitude ring.

IMHO, they (also) need to put more focus on the hysteresis effect with regard to thermodynamics and dissipation of excess heat.

https://en.wikipedia.org/wiki/Hysteresis

Compared to a real world scenario SP35 acts like someone driving a car on the highway, frequently pressing and releasing the foot on the gas pedal to keep an average speed.

I also noticed that other flashlights allow a wider tolerance for their ATR threshold, especially after turbo is triggered. For instance, Acebeam L17 allows up to 70°C excess heat during the leveling phase before the ATR threshold settles at 55°C.

Sofirn‘s SP35 with ATR or even their new TF84 model have a very conservative, strict throttling. Consequently, the rising temperature is being reduced in an early stage to prevent significant overheating above the set threshold of 50°C (SP35) or 45°C (TF84). On a positive note, customers would probably complain less about flashlights getting very hot.

Got mine yesterday. Overall really like it. The size/power/price combination is just perfect. The ATR is very acceptable. Running flat top VTC6 with no problems and no rattle. For $24 with battery its just about the best deal in town.

I just had a look at that listing and maybe they changed it but it’s got ATR in title listing & description. :frowning:

Ya I didn’t quite your post but this Amazon listing appears notw to have ATR. :frowning:

I’m happy having no ATR version. Just awesome!

Little beam comparison between new and old reflector.

Visually for me the new reflector has a larger hotspot and more smooth transition from spot to spill. Less tint shift and artifacts. I absolutely prefer the new one!

Collected for the archives:

SF20L082 - no ATR, old reflector.

SF20L151 - no ATR, new reflector.

SF21C052 - yes ATR, new reflector.

I’ve run full tests on both Turbo and High using the new ATR version (as well as the non-ATR version).

  • Turbo without ATR: gets too hot to handle in around 5 minutes
  • High without ATR: runs great, temperatures are fine (at least in a 21°C room)
  • Turbo with ATR: bright, but regulates down before temperatures get out of control; it ramps up and down to maintain a safe temperature (avg 54°C)
  • High with ATR: for some reason, output seemed a bit higher than on the No-ATR version and it gets hotter faster, then it begins oscillate much like it did on Turbo.

While the ramping up and down looks crazy; in reality it doesn’t actually look as bad. I just shot a video and will have it uploaded soon.

Here’s a nice, long, boring video of the ATR version running on Turbo. It’s exposure-locked so that you should be able to see fluctuations. The thing is, they’re so gradual that you can’t really see them. Each up and down cycle takes 1.5+ minutes to complete and is very smooth (not stepped).

Thanks a lot for the interesting and detailed analysis, Gabe. Glad to see that my observations are being confirmed in most points. I only wonder why my sample only ran 2h 55min on High using a fully charged 5000mAh Samsung 50E battery while yours ran outstanding 3h 11min with only 4000mAh. Can you please reconfirm that you used the Sofirn 4000mAh battery?

I also noticed a completely different „phase out“ behavior between your sample and mine. While yours shuts down very quickly at the end mine slowly reduces its brightness over a course of 30minutes. Any idea what can cause such a huge difference? The 50E I used is about 2 years old and had only a couple of cycles yet.

Regarding these brightness fluctuations I agree with you that these are barely noticeable outside „in the wilderness“ ;-) but looking onto the white wall I could easily track them with my bare eyes. While most customers not knowing about these fluctuations may not complain I still hope that Sofirn‘s engineering can tweak their ATR for a more stabilized output.

Yes, I did use the 4000 mAh battery that came with the original SP35 (the exact battery). I did this to retain an “apples to apples” comparison. But the new one did come with a 5000 mAh, which is nice!

That difference in the “phase out” is very odd. And the fact that we got virtually the same runtime with 4000 mAh (myself) and 5000 mAh (you) batteries. Bizarre! The only thing I can see is that it looks like the peaks of each wave is a bit higher on yours.

Thanks for doublechecking. I agree the new Lishen LS2170SD (5000mAh) is pretty good, especially for lights with moderate power consumption like SP35. Provided that our samples are absolutely similar in terms of driver specs and firmware, I could only imagine two options:

a) My Samsung 50E battery is worn out and lost probably 20% of it‘s capacity (which would worry me quite a bit).

b) Sofirn put a 4000mAh wrap around your „4000mAh“ battery but in fact they sent you a 5000mAh battery in disguise. :-))

Maybe the prolonged phase out on mine could be a matter of voltage dropout between battery and buckdriver? Maybe it depends on how fast voltage drops in the final stage of the battery depletion? I have no idea but I could try testing it once again using a high drain 30T to see some difference if any. OTOH, I do not expect any big impact on the result since it is not an FET but a buckdriver.

Right. This could also explain why I am a bit more displeased about the fluctuations, even noticeable by my eyes if I look very closely.

Has anyone measured the parasitic drain or know if it will fit a protected 21700?

Just checked. Parasitic drain is only 23 uA. The only protected 21700 is an Acebeam cell with built-in USB-C charging; it’s too long to fit.

Dang, I thought there would more protected 21700’s on the market by now. This looks like a great light to gift someone, but I always give them a protected cell (and usually a clicky switch). I guess with parasitic drain that low it’s not that big an issue. Thanks

Sorry, I meant to say that the Acebeam cell is the only protected 21700 that I have. There’s got to be a few others out there.

Btw here’s my runtime measurement with the Sofirn 5000 mAh battery.