Decay of luminescence on IR LEDs?

They vary a lot. The quickest ones are about 70 MHz BW, and they should slow down like an RC--the minority carrier lifetime in GaAs and its relatives is very short (ballpark 1 ns).

JL and I had occasion to measure a bunch of them recently for the PH200.

Cheers

Phil Hobbs

But 70MHz at 3dB is miles away from 1nsec.

Reason I ask is something weird that happens at a long distance client: They modified a TIA per my instructions so it would recover from brutal overdrive (factor 100 and more) very fast. SPICE confirmed it. Then they measured and it's still as slow as before. In our case the TIA would come out of "pegging" at about 1% luminescence and my suspicion is that either our LED or the PD takes their sweet time to get there. We are seeing many microseconds. Can't imagine any PD being this slow but not sure about the IR-LED. It's listed around 10MHz BW.

So far I have mostly dealt with laser diodes and those were always super fast.

Yup, diode lasers are very much faster. I think it's basically an RC effect in LEDs.

Cheers

Phil Hobbs

Well, one reason for gigabit fiber communications devices being more expensive than 100Mbit is supposedly the need to use lasers rather than LEDs, which argues in favor of lasers being significantly faster (and I think "faster to shut off" was specifically mentioned.) At an in-between price point there are Vertical Cavity Surface Emitting Lasers - I don't know if they are also intermediate in speed, or as fast as "normal" lasers.

At a low price point, if you are willing to rip open SFPs, those are often available used, with IR lasers (and PDs), absurdly cheap. I got a pile for my project (not ripping them open, just using them as intended) at about $5 each last fall (4GB capable).

Do you think the spontaneous recombination lifetime can be modeled as RC? That would be cool. The datasheets say that the optical output rise/fall times are about 50nsec and this is for 10% and 90%. If that is RC then I could model it in SPICE.

Unfortunately there is precious little literature on the topic unless one is a member of the respective engineer's society or has a subscription to one of those $2k/year journals.

Yeah, I know, but unfortunately the diode is beyond our influence. The transmit stuff comes with it :-(

I really like VCSELs, except for their polarization mode hop noise.

There are some VCSELs with really good polarization stability. Check out ULM Photonics.

They work by adding a deep subwavelength grating to one of the mirrors. While this doesn't cause diffraction, the modulated evanescent field causes a change in the reflection phase in the polarizations parallel and perpendicular to the grooves. (The evanescent field is different, and there's no diffraction, so changing the reflection coefficient is all it can do.)

Because the cavity is so short, the reflection phase shift splits the P and S polarized cavity resonances by a lot, enough that one of them is shifted off the gain peak and so doesn't oscillate.

I don't know how stable the polarization is over long periods, but the short term stability is very very good--better than Fabry-Perot diode lasers.

Cheers

Phil Hobbs

The minority carrier population probably decays exponentially, because the rate is proportional to n_e * n_h, and one species will be in large excess.

LEDs have huge capacitances, especially in forward bias--70 to 100 pF is not unusual. You might be able to speed it up by transient reverse bias, the way you do with BJT stored charge.

LED bandwidth is (iiuc) measured with a small modulation on top of a large DC bias, so that the differential resistance of the diode is only an ohm or two. That'll speed up the RC time constant pretty dramatically, so I wouldn't be at all surprised if the on-off behaviour is quite a bit slower.

Cheers

Phil Hobbs

Thanks, Phil, very good hint. If anyone else wants to look:

They are in my old home country and they named it after their city (Ulm, at the Danube river). For us it won't work right now, this project has been completed a few years ago and they currently don't offer long enough wavelengths, only 948nm max. But that could change some day, got to keep them on my watch list.

Ok, so an RC decay would be a good approximation then. Thanks, Phil.

Aha! Sounds like a good old marketing trick to make it look better. Nobody in their right mind would use an IR LED as a transmitter that way. Reminds me of the range claims of some wireless companies. I put one of them on the spot and then an engineer admitted that, yeah, the data pretty much applies when both stations are in outer space.

Phil, why are their VCSELs 100MHz in spectral width? That sounds ghastly compared to the ones we had a couple years ago.

Or did they measure unfavorably? Can't imagine why anyone would do that. We had the VCSELs in a loop which cuts down on some of the noise mechanisms, just like you can clean up a fairly noisy oscillator by tying a PLL around it.

VCSELs have very wide Schawlow-Townes line widths. The S-T limit is basically caused by the counting statistics of both the emission and loss in the cavity. The AM and PM this causes depends on how many photons there are in the cavity. VCSELs have very short cavities, which makes the linewidth wider.

You can make them better behaved with an external cavity--by locking to an etalon or a big long fibre Michelson. (You need a Faraday mirror for the fibre approach.)

Cheers

Phil Hobbs

But this wide? About two years ago we had some that were already natively better.

We had to have flexibility in wavelength so a hard optical lock wasn't possible. But I ran them in a PID loop which really made a difference.

I'm not certain about these exact devices, but a SWAG is that they probably make the cavity very short even for a VCSEL in order that the polarization splitting make more of a tuning shift.

Besides, they probably current-tune really fast, so just wrap a loop around it. ;)

Cheers

Phil Hobbs

That would not be so cool.

No joke, I had a loop around them most of the time. But it cannot be a fast loop so if the wavelength meanders around somewhat slowly, ok. If it's a super wideband hash it's all toast.