Maker Pro
Maker Pro

Re: Spice models of laser diodes?

J

Joerg

Lostgallifreyan said:
If anyone is following this and wants to try modelling their own stuff, I
found what might be a way.
Intusoft make a tool called SpiceMod which is part of a package they call
ICAP4 though it seems the demo setup doesn't have that tool, just some very
good noted on it, and modelling in general:
The file WkwModels.pdf from the demo install answers a lot fo the questions I
had about modelling diodes, which parameters to tweak, and extraction from
data sheets.

I don't know if the lack of response to me is because of a thousand experts
silently screming RTFM at me, or because it's actually asomething they DON'T
KNOW. Given that Intusoft explain that this is a serious challenge for
experts too, I'm assuming maybe they really don't know, so they might benefit
from that file as much as I will.


Most of us just use a regular diode to simulate because all one
(usually) wants to know is that there definitely won't be any ever so
slight spike in diode current because LDs can go poof in microseconds.
Simulating the optics part would be a major challenge, I think.
 
J

Joerg

Lostgallifreyan said:
Thought that might provoke an expert. :) I agree, same here, that's all I
usually need too, though like that EDN manual says, a more detailed model
that allows models to give warning of imminent demise is useful. Modelling
for ESD is likely daft, better that we just take care and put in TVS's and
such, but when it comes to fast modulation, a model definitely helps. Surely
you'd have a use for that, no?

I have designed LD circuitry but I am not an expert. Senor Hobbs would
be one. TVS don't work well. Their cutoff isn't terribly well defined
and they have too much capacitance for this. Just handle the things with
care. I've never killed one with ESD, knock on wood ;-)

While I found that a string of four 1N4148's produced a modelled overshoot
almost exactly like what I saw on an oscilloscope months before I considered
looking at spice, I got a very different result when trying four 1N4001's so
it really does need something better than reaching for a standard diode, in a
model OR as a dummy diode in a real circuit (where a optically dead laser
diode is best anyway). So it really comes down to trying to get something
usefully close! EDN's model seems ideal, aimed at solving this problem for
general use, as opposed to the elaborate models in private university
publications. EDN's is probably tested too, proofread and verified before
publishing.

http://www.e-insite.net/ednmag/archives/1998/071698/pdf/15di.pdf
(Halfway through file).

"File not found" :-(

Some of the 4000 series behave more like PIN diodes.

If anyone can help by making that into asy and sub files for LTspice it will
help me a lot, and probably a lot of other people too. Four days of searching
have found lots of other people askign questions, and very few answers, and
NONE complete and verified. Someone could get well known for solving this so
other people can have an easier time of it.


It's going to be lots of work. You'll need to get into behavioral models
and while people have modeled large chunks of jet engines with that it
was a ton of work.
 
J

Joerg

Lostgallifreyan said:
Lostgallifreyan wrote:

[...]
"File not found" :-(

Sorry, right file, wrong location. Try this:
http://www.edn.com/archives/1998/071698/pdf/15di.pdf

Well, laser diodes are a lot more complicated that this simple model.
Onset of lasing which happens sort of digital at xx percent of max
current, non-linear relationship between output and input power,
overload behavior, wavelength drift with heat and so on.

I've been finding (and been sent) spice models that are in various stages of
completion and complexity. Ignoring the rest for now, I'm focussing on that
EDN one because it looks like it's meant to be complete. But it's NOT in a
single subcircuit file that can be adapted to LTspice, or anything else. I
don't think it needs modelling, just translation.

For modelling, I'll look into that too, as equations are given in that PDF
file I mentioned earlier, from Intusoft. In their descriptions of how to use
SpiceMod (which I learned costs 600 bucks(!) and is therefore the most
expensive few tens of kilobytes I've ever heard of) they show a screen that
neatly prioritises the data sheet specs for pluging into the equations. While
the details for transisotrs (bar a MOSFET) are not similarly revealed, there
ought to be enough there in their diode explanations to model a laser diode,
electrically. But again, all I'm asking for here right now is for someone who
knows how, to translate that EDN article to a subcircuit ready to plug into
LTspice. (I can make my own symbol file).


Seriously, it's a lot more intricate than that. Just to give you an
example from the non-LD world: One of the transistor models of an
amplifier I recently ran is about two dozen lines of SPICE entries.
Those are the sims where the fans come on hard and the office temp
creeps up another 3F or so.

I don't think you'll get around behavioral models here.
 
J

Joerg

Lostgallifreyan said:
Interesting. Robin Bowden didn't make strong recommendations for them either.
Just a resistor/capacitor network, and a zener. I concur, though at the time
I was all for varistors too as I'd found a LOT fo them cheaply on eBay at the
time.. But they apparently change characteristics rapidly over time fi they
see anything like the conditions they have to limit, so I'm not too keen now.

How do you want to protect a laser diode with any of these?

Trying not to get away from the focus of the thread because I really do want
direct help with model translation as it's unlikely I can do it alone, but
this IS interesting to me.. PIN are those fast photodiodes, right? Often used
for detecting very short laser pulses and such? When I modelled my driver
using the 1N4005 model I found, I got a lot of ringing. I also saw a
correlation with the changing of Vf with hard changes between zero and full
drive. I analogised it with the hitting of a hard object as opposed to a
softer one that inherently damps the impact. If a PIN diode is to respond
fast it seems reasonable to think that it is capable of hard brittle
responses, otherwise it could not hope to react in time to register a very
short laser pulse. Is this a reasonable way to view it? And if so, why so
with rectifiers and not signal diodes? Or LED's which I imagine are similar
to laser diodes, being usually GaAs based.


Photodiodes are reverse biased if you need speed while PIN diodes used
as controlled RF resistors are forward biased and the current sets the
resistance. But that works only if the carrier lifetime is sufficient
for the frequencies it has to work at.

For more data regarding LD models I'd get in contact with a few
university research labs. It is not a trivial task at all.
 
J

Joerg

Lostgallifreyan said:
EDN (who I trust quite deeply on matters of useful working ideas) seem to
think that a moderately simple diode model can be enough when making basic
driver models. All I'm asking for is translation of that model to a
subcircuit file I can use in LTspice. I'm new to spice but it's already clear
to me that it is UNWISE to model more detail than is strictly needed. So I
think EDN have the right idea. I'm after something that other people can
also easily use and benefit from, I'm not after a perfect detailed model I
can win prizes with.


You'll have to do that piece by piece. LTSpice's preferred method is
graphics entry. For example, instead of writing out the netlist line for
G you place a voltage dependent current source via point and click and
connect it up in the schematic editor.

If you are new to SPICE I strongly suggest not to start with a project
like this but first use some of the supplied "jigs" or example circuits
and play around with those. Unless you really understand the program it
is very easy to reach wrong sim results.
 
J

Joerg

Lostgallifreyan said:
I read somewhere that it was a better version of an earlier one in a Pspice
library. How good does it have to be? ...


That depends on you. If you want the sims to model all the way to
failure modes I'd really be surprised if PSPICE had something in a
library that would even come close. My PSPICE license is too old so I
wouldn't know. But that would be one tough job.

Again, a laser diode is not a linear device when you look at lasing
output. Not at all.

... No point in modelling more detail than
needed as a starting point. We don't even HAVE that much, any of us hobbyists
and small scale designers. If experts raise their hands in horror saying
there's no point in simple modelling rather than too much detail (with which
Intusoft, who really know this stuff, would solidly disagree), ...


Let them disagree. Will they pay for your dead laser diodes?

... then we'll all
continue to have nothing to go on except dead laser diodes! Surely someone
who knows how might change this?


All I am saying is that I believe it is impossible to correctly model a
laser diode by trying to find electrical equivalents for all its
behaviors. You need a behavioral model in addition. All it takes for a
LD to die is optical overload inside the cavity. A brief wiggle of a
fiber connector, a spike of a few usec ... poof. The EDN model isn't
helping you with that.
 
J

Joerg

Lostgallifreyan said:
I suspected as much. :) Actually my guess is that their equivalent circuit
needs to be constructed then turned into a subcircuit. There's no doubt it
CAN ultimately be written as a single subcircuit file, I don't know why EDN's
contributor didn't go all the way.

Yes, it should be feasible to place it in a subcircuit. However, since
you said you just started out with LTSpice I suggest not to do that (yet).

I always reality-check things. I only intend to use LTspice the way I do my
case designs and other physical models in SketchUp. Sure, a posh CAD tool can
design me a proper screw thread while SketchUp hasn't a chance of this (at
least, not the version 4 I choose to stay with), but the idea is to quickly
aid visualisation, not to substitute for reality.

I use DesignCAD 3D for that. Ten bucks at a liquidator. Mostly because
it can read in AutoCAD files.
 
J

Joerg

Lostgallifreyan said:
No, but I'm looking at electrical INput..

That's the easy part :)

BTW, I do not understand why EDN placed the SPICE list as a graphic.
Somehow that doesn't leave a very professional taste.

No, but they DO take my point when I tell them that SpiceMod shouldn't cost a
personal user more than a very few of them. :)

The ones I dealt with last were about $1200 a pop. Lots of them. The
client would have been very p....d if I had blown some.

Well, Tim G at Intusoft kindly said if I sent him some model numbers, he'd
see what he can do. I sent him 5 data sheets. :) Got to try my luck, no?
Besides, they might give a workable average for the kinds of high power red
single mode diodes hobbyists are using.

From what I've seen of my dead diodes, they are still electrically very
similar to live ones so I'm not concerned with advanced optical modelling.


Often you can still use them as LEDs :)

What killed them?
 
J

Joerg

Lostgallifreyan said:
Ah, but Ghost lets me retain a known working OS config, recalled at will, and
an INTENSELY useful side effect of this is infinitely renewable demo periods.
:)


:)

I don't do that. IMHO it's quite borderline from an ethics POV to fool
demo SW into a fresh time period. Then I either buy it or move on.
 
J

Joerg

Lostgallifreyan said:
During the transitions, perhaps, but in steady state operation there seems to
be a remarkable electrical similarity between a dead diode and a live one.
What matters is the way the diode responds to hard electrical changes on the
input, as that's what makes the ringing and damaging overshoots. This is true
with NO consideration of optical nature, and just modelling that alone,
realistically for real laser diodes, is a lot more than we currently have.
And likely not that big an ask, it's just not been done much, it seems.


Ringing? I never had that. Don't drive them through a built-in inductor :)

LDs are usually current driven. The prudent way is to impose a constant
DC current, very well stabilized and equipped with belts, suspenders,
cushions, airbags. Then the fast signals are fed in via a current
"robbing" shunt circuit to ground. That pretty much makes sure you can't
fry it.
 
T

Tim Williams

Lostgallifreyan said:

Wow, I didn't know people were still using that term. Last time I heard it
was in talking about tube televisions from the 1960s!

Snivets is where the sweep output tube undergoes oscillations (Barkhausen
oscillation) between the plate and screen, when plate voltage is lower than
screen (Vp(sat) typ. 30V for these types, at Vg2 = 125V). It's supposed to
make an electron-beam-driven resonant cavity, producing UHF oscillations
something like a klystron I suppose. The solution is to raise the voltage
on the beam former grid to about 30V.

Can I get a modern definition for "snivets"?

Tim
 
J

Joerg

Phil said:
Ohhhhhh..... You're using a voltage driver, and trying to trick it into
being a current driver. Bad, bad, bad, bad news.

Due to the rolloff in the loop gain, the output of a voltage regulator
appears inductive, which will reliably give you a big noise peak if the
output cap is too big, and some ringing if it's too small. As Joerg
said, you're way better off using current drive. It isn't difficult,
just an op amp and a Darlington. Use the Darlington's collector as the
output, and sense the current in its emitter. Adding an outboard
current limit is easy then too.

You can build a current drive out of a LM317, I've even used it as an AM
modulator. But 500kHz, nope, it ain't going to go quite that far.
 
J

Joerg

Lostgallifreyan said:
Here's something to consider: I think many posts here are assuming a
switching modulator wheras I am after a continuously proportional and DC
coupled type. I can hybridise with a PWM control, and eventually will go to
that, probably, but for now, what do you do if you want linear control? Never
mind that diode light output isn't at all linear, we might as well start with
what we CAN control... So do we really use a Darlington? I understand that
those are made to switch fast, they're not chosen for a long linear operating
region... So that leaves some fast single stage transistor, and a sense
resistor, and a controller, likely an op-amp.....


Is this getting through? :) The LM317 in data sheet approved constant
current mode IS just that, though not a shunt mod. But it does the opamp, the
sense resistor, and the hefty transistor to drive up to 1.5A. It's not as
daft as it looks, and it performs better than anyone ever told me it could.

Risque if you want to do 500kHz as you mentioned in another post.

Anyway, assuming I do shunt part of a constant current from the diode, what
might be the simplest way to do it? As far as I know, something like Robin
Bowden's 'Die4drive' circuit might have a basis for this with its MOSFET,
opamp amd sense resistor, but that's not a shunt mod either...

In a DC application (well, sort of, low kHz stuff) I did it the same way
as fast applications: A top current source that makes sure the DC
current never ever goes above xxx milliamps. Can be built around a LM317
if you wish. Then an opamp-controlled current sink from the LD anode to
ground (the LD cathode is also on ground). This current sink then
"wastes" x amount of the current from the upper source by sinking it
into ground, depending on the control signal on the IN+ of the opamp.
What you control is in essence how much of the top source current goes
through the LD and how much is wasted.

The super-prudent approach is a high voltage and resistors. The sum of
the resistors makes sure the LD can never get too much current. Then
current-rob like above. Resistors can't go unstable on you :)
Getting back to the topic, whatever we do, it would REALLY help if we all had
access to some reasonable semblance (electrically) of a single mode laser
diode in spice. It's long overdue. It would save students and schools and
hobbyists a lot of money if it was there. Linear Technology have given us
LTspice to use for free, now we need things like this to put it to use.

Don't hold your breath. For many people (like me) a LD model would only
make sense if the optical side is in it as well and that makes one heck
of a complicated model. Otherwise it isn't terribly useful.

In other news, I seem to have dropped alt.lasers from the cross-post list.
Annoying. :)


I think that's ok. Those guys know all this stuff I suppose.
 
J

Joerg

Lostgallifreyan said:
Want to bet? >:) I grant that it won't go further, I tried the model at 1 MHz
and it degraded to crude sine waves, but 500 KHz is ok. As far as I can tell,
the main weakness seems to be a small range of acceptable capacitance on my
controlling op-amp's feedback loop so parasitic capacitances alone may, or
may NOT, make it work well. And it's very dependent on the diode too which is
why I want an LD model. But when I built my first real circuit for this
design a few months back the LD showed characteristics similar to the 1N4148
model which are much more favourable to damping high frequency ringing than
if it had been a 'hard' diode like a 1N4005 that barely responds with Vf
changes to current changes and forces. I think LD's do vary Vf a lot with
current so they damp the ringing well enough.

It can work but as Phil had mentioned the LM317's stability becomes iffy
beyond a certain frequency when operated as a current source. It's like
taking a small car past 120mph.

But I'll know more when I've ate the pudding.


Remember, laser diodes die quietly. There is no tchk ... *phut* but
there is that painful hit in the bank account.
 
J

Joerg

Lostgallifreyan said:
Good stuff. I thoughyt they were fairly linear too, above threshold (and my
design sets the low peak to just sub-threshold too. They can mode hop though,
which doesn't help, but if they're going to do that they'll do it anyway and
impair the effect of a clean drawn line in graphics.

The PWM idea isn't great for graphics either, but it depends how fast the
main cycle is, if it's possible to modulate on a 10 MHz carrier without
awkward exotic precautions, it isn't going to show up as a dashed line using
any scanner currently available. That makes it an attractive idea for other
types of laser too, and also AOM's too, which is cool because proportional
PCAOM's cost more. (Not that I've even had the money and availaibility of
such to try those).


10MHz is sluggishly slow for a decent laser diode.
 
J

Joerg

John said:
Lostgallifreyan wrote:
[...]
"File not found" :-(

Some of the 4000 series behave more like PIN diodes.

The higher-voltage ones ARE pin diodes. They can make excellent drift
step-recovery diodes and impact avalanche diodes. Need 1000 volts in
100 picoseconds?

Do you have some examples and pics? That would be interesting, I have
use the 1N4007 as a RF PIN diode but never as SRD.

One of my next projects needs the opposite, briefly and automatically
disconnecting an RF amp from a several kV pulse path so it doesn't go up
in smoke. The kind of stuff where people often say "you can't do that".
I love it when a project like this comes along.
 
J

Joerg

Lostgallifreyan said:
I'm all for pushing the envelope, there's no hide at stake here. :)
It sure looks like it will work. It's 'competing' with systems that can do
100 or 200 KHz at best generally, and none of these devices are needed to go
further. They work alongside laser systems incapable of more that 10 K, often
as not.

The main point is to see what's possible, and to establish the safe operating
area so when I really want to use this thing I can have a very good idea how
to make it right, at more reasonable figures. And I'll know it will draw a
very clean line, which is what the people most likely to use it would want it
to do.

You can only try that on a breadboard, not SPICE. And then you find out
that it can do 250kHz but after switching to a different manufacturer's
LM317 ... *POOF*

But I still want a good laser diode electrical model.. :)


I'd rather prefer a nice bottle of Chateau d'Yquem :)
 
J

Joerg

Lostgallifreyan said:
Sure, but again, that's part of the appeal of making widgets that all use a
single device or small number of parts. Means such adjustments don't have to
be made that often. And again, this is as much as exercise in establishing
safe operating areas, as much as anything else...

For a one-off prototype you could be ok.

Does it talk? (Python joke :)


No, but it can make you talk :)
 
J

Joerg

John said:
John said:
Lostgallifreyan wrote: [...]

While I found that a string of four 1N4148's produced a modelled overshoot
almost exactly like what I saw on an oscilloscope months before I considered
looking at spice, I got a very different result when trying four 1N4001's so
it really does need something better than reaching for a standard diode, in a
model OR as a dummy diode in a real circuit (where a optically dead laser
diode is best anyway). So it really comes down to trying to get something
usefully close! EDN's model seems ideal, aimed at solving this problem for
general use, as opposed to the elaborate models in private university
publications. EDN's is probably tested too, proofread and verified before
publishing.

http://www.e-insite.net/ednmag/archives/1998/071698/pdf/15di.pdf
(Halfway through file).

"File not found" :-(

Some of the 4000 series behave more like PIN diodes.
The higher-voltage ones ARE pin diodes. They can make excellent drift
step-recovery diodes and impact avalanche diodes. Need 1000 volts in
100 picoseconds?
Do you have some examples and pics? That would be interesting, I have
use the 1N4007 as a RF PIN diode but never as SRD.

Google "Grehkov diode". He's a Russian guy who discovered the DSRD and
impact avalanche effects in cheap power diodes.

Also look for papers and patents by Thomas E. McEwan.

Thanks, I will check that out.

We did one water-cooled DSRD pulser that makes -2KV pulses, about 2 ns
wide, at 500 KHz. We forward-bias a secret diode at +48 volts for
about 80 ns, to let the current build up to 50 amps or so, then
reverse-bias it from a 400 volt supply and wait for it to snap.

Here's the pulser head...

http://www.highlandtechnology.com/DSS/T220DS.html

with the serious parts bolted to a gold-plated copper block. A
water-flow cold plate gets bolted to the bottom.

I can show you innards privately. The HV, high power PIN diode turned
out to be unusual.

That would be nice. My "Reply to" address is valid as is and lands in my
biz inbox. But only if you feel comfortable doing that.

Was this the project from several years ago that started with the FR804?
Take a look at this puppy, fresh out of the gate:

http://www.nxp.com/acrobat_download/datasheets/BYV29FX-600_1.pdf

In the project coming up I need a lot more volts though (10kV range) and
it has to be small.
 
Top