Laser, DFB module: TEC reaction speed

[Joerg]

Advanced control algorithms presuppose that the actual "plant" state
can be accurately measured as the thing to feed back against. It's not
obvious to me that you have an accurate realtime wavelength sensor,
either in the production system or even for initial laser/tec
characterization.

There is a really nice optical spectrum analyzer. Sweet. Works like a
champ. Long story short I got the PID thing going today but it's too
slow for my taste, around 3secs per 100pm for TEC control. The
difficulty in this case is that we must scan until we hit a resonance
and then lock. The location of that resonance isn't known a priori.
Pretty much like an etalon where it's all flat and then there is a
valley of just a few pm into which we have to lock. Ok, we also have the
diode current as a lasso but it still feels like barreling down Lombard
Street in a tanker truck and suddenly having to stop on a dime.

Hell of a storm here last night. There's debris all over, lots of palm
tree bits on Dolores Street, and some traffic lights are out. Jim and
the boys should get it next.

Same here east of Sacramento. It even tore the pool sweep
quick-disconnect off which never happened before. Now a cold front rolls
in. But when I came home from the client my wife had already stoked the
wood stove really good. Nice! Time for an Irish coffee.

 

[Joerg]

My understanding of TEC coolers is that they cool at a rate proportional
to KP*I - R*I^2 -- and the 'R' term really is a resistance. If this is
truly the case (and if you can characterize it) then it should be easy
to calculate a desired rate of heating, then calculate the necessary
current.

Of course, that doesn't mean that your laser's response to temperature
changes is at all linear.

That response is what I'll still have to measure but from a coarse test
today it looks quite linear. The TEC doesn't seem to follow any easy
formula though and all the delays in there are a bit unorthodox. But at
least the PID is basically running. Not fast enough though, about
3sec/degreeC. I'd like to spiff that up a bit.

The crux is that I have to drag the TEC until a rather sharp valley is
hit, then bring it to a screeching halt. We typically won't know where
that lone resonance is going to be and it's rather abrupt, without the
usual LC type slope. Just flat and then off the cliff. And we have to
cling to the face of that cliff.

The finish of the last PID tuning round was greeted by one of the DIP
switches breaking off. Oh man, those things are so cheesy these days.

Eww. I don't like lookup tables, unless there's some linear
interpolation going on -- it always seems there's something odd
happening at the transitions.

That's where wave digital filters come in handy if there aren't enough
MIPS. They behave quite gracefully on transitions and drop-outs.

Phil Hobbs's suggestion of using a heater with the TEC going full blast
sounds good to me -- I suppose it's too late to slip a heater in there,
isn't it?

Maybe. But its a DFB module where you can't get to all that. Got to live
with what's in that can :-(

 

[[email protected]]

My understanding of TEC coolers is that they cool at a rate proportional
to KP*I - R*I^2 -- and the 'R' term really is a resistance. If this is
truly the case (and if you can characterize it) then it should be easy
to calculate a desired rate of heating, then calculate the necessary
current.

Check out my paper in Measurement Science and Technology, volume 7,
pages 1653-1664, (1996). Appendix A gives an expression for the heat
transferred by a Peltier junction as a function of the temperature
difference across the junction and the current through it in terms of
the maximum heat transferred at zeo temperaturedifference and the
maximum temperature difference at zero heat transfer - both available
from the manufacturer's data.

If you can estimate thermal resistances accurately enough, this can be
reworked to give heat transferred per unit current from the controlled
region to ambient - also discussed in Appendix A.

The expression isn't particularly precise - it looks as if some of the
coefficients are temperature dependent - but it is good enough to keep
a PID loop pretty close to critically damped.

I've seen the expression published elsewhere - in EDN a few years later
- and I got it into a comment published in Rev.Sci. Instrum. in 2004
-volume 75 pages 788-89.

It ought to be in Pelter junction application notes, but I've yet to
see it anywhere useful.

 

[Joerg]

Hello Folks,

We need to lock onto one side of a sharp but moving resonance similar to
a Fabry-Perot dip. Sharp meaning around 10pm. Long story short, when
scooting the TEC via a ADN8831 chip it's tough to achieve regulation
without overshoot. IOW tuning the PID loop is a bear and the locking
mechanism overshoots the target which shall not happen in this case.
It's a DFB module (JDS and one from Sumitomo).

What is the fastest TEC tuning speed that the laser expert here have
achieved with those modules without "ringing"?

Of course it's possible to move 10C in a couple of seconds but when
checking how professional equipment such as the ILX Lightwave controller
is the lab does that it overshoots by a lot more than 1C and takes
several seconds to stabilize. They seem to really be stepping on it for
the derivative part which is ok in telco apps where you usually set the
wavelength once and then leave it there. In our case we must remain
agile. The total range we need is around 10C (hopefully...). A wee
overshoot of 0.1-0.2C would be ok but when tuning the PID loop towards
that goal the whole thing becomes like molasses.

Ok, folks, just in case anyone else runs into this issue: I got it down
to a little under 3sec/degreeC for modest overshoot (under 0.1sec). I
can push it down to below 2sec but then the overshoot will stray out too
much in some regions. We need a range of up to 20C and the TEC behavior
is very non-constant over that wide range. My feeling is that the only
way to push it faster would be PID with adaptive coefficients, IOW where
the PID coefficients can be changed depending on the temperature the TEC
is sitting right now. Some brute force digital algorithm might work as
well but currently we don't have a micro controller on there.