Forums

TEC for laser diode

Started by mt September 28, 2013
On Tue, 01 Oct 2013 14:02:23 -0400, Phil Hobbs wrote:

> On 10/01/2013 12:37 PM, George Herold wrote: >> On Tuesday, October 1, 2013 11:19:24 AM UTC-4, Phil Hobbs wrote: >>> On 10/01/2013 11:04 AM, George Herold wrote: >>> >>>> On Saturday, September 28, 2013 9:23:39 AM UTC-4, Glenn B wrote: >>> >>>>> On 28/09/2013 1:33 PM, mt wrote: >> <snip> >> >>>> Hmm I bought one of their TEC controllers years ago. I had to hack >>>> it to get a longer integration time... and then just ended up making >>>> my own. >>> >>>> The longer integration time gets to the debate of where to put the >>>> temperature sensor. Near the TEC, or near the laser diode. I opted >>>> for near the diode, and so a longer time constant. But others think >>>> otherwise, and I very well could be wrong. >>> >>> >>>> >>>> George H. >>> >>> >>>> >>> There's no reason to have just one sensor. You can wrap the actuator >>> in a local first-order loop with some fairly crude but fast sensor >>> right at the actuator, and use that as an improved actuator in an >>> integrating loop with a sensor where you actually want to control the >>> temperature. >>> >>> That gives you high accuracy plus decent rejection of thermal forcing. >>> >> OK, having your cake and eating it too. >> Hmm, so you roll off the fast loop output at some frequency where the >> slow loop takes over, (high pass filter on the fast one) and then sum >> them. >> The details of that hand-off are -interesting-. >> Is a single pole roll off the way to go? >> (Drawing some PI bode plots I seem to be confusing myself. Do you roll >> the slow loop with a single pole low pass too?) >> >> >> >> ^ LOG(gain) >> | >> |\ \ >> | \ \ >> | \ \ <--fast loop | \ \ >> | \ \ >> | \ \ >> | \<--slow | >> | >> |---------------------->LOG(freq) >> >> >> ^gain | >> |\ >> | \ >> | \ /\ >> | \ / \ >> | \ \ >> | \ \ >> | >> | ^-----rolled off here. >> | >> |---------------------->freq >> >> But how does one pick the frequency? >> >> > It's usually pretty simple IME--you make the inner loop as fast as you > can with decent stability, and then do the same with the inner one. > Normally there's enough bandwidth difference that any small phase > whoopdedoos in the closed-loop response of the inner loop don't affect > the outer loop much.
After you tune the inner loop twice, then be sure to tune the outer loop. (or fix your typos). ((snicker)) -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
On Tue, 01 Oct 2013 09:37:49 -0700, George Herold wrote:

> On Tuesday, October 1, 2013 11:19:24 AM UTC-4, Phil Hobbs wrote: >> On 10/01/2013 11:04 AM, George Herold wrote: >> >> > On Saturday, September 28, 2013 9:23:39 AM UTC-4, Glenn B wrote: >> >> >> On 28/09/2013 1:33 PM, mt wrote: > <snip> > >> > Hmm I bought one of their TEC controllers years ago. I had to hack >> > it to get a longer integration time... and then just ended up making >> > my own. >> >> > The longer integration time gets to the debate of where to put the >> > temperature sensor. Near the TEC, or near the laser diode. I opted >> > for near the diode, and so a longer time constant. But others think >> > otherwise, and I very well could be wrong. >> >> >> > >> > George H. >> >> >> > >> There's no reason to have just one sensor. You can wrap the actuator >> in a local first-order loop with some fairly crude but fast sensor >> right at the actuator, and use that as an improved actuator in an >> integrating loop with a sensor where you actually want to control the >> temperature. >> >> That gives you high accuracy plus decent rejection of thermal forcing. >> > OK, having your cake and eating it too. > Hmm, so you roll off the fast loop output at some frequency where the > slow loop takes over, (high pass filter on the fast one) and then sum > them. > The details of that hand-off are -interesting-. > Is a single pole roll off the way to go? > (Drawing some PI bode plots I seem to be confusing myself. Do you roll > the slow loop with a single pole low pass too?) >
In general when you do a two-loop control system you don't try to roll off the control of the outer loop or any of that razzmatazz -- you tune the inner loop for good behavior in a conventional way, then you say "gosh, look at this plant that I have" and you tune the outer loop to the inner loop's characteristics. You may do some iteration between inner loop and outer loop tuning (usually to relax the inner loop's gains -- see below), but in general you want to tune the inner loop in isolation, then tune the outer loop as if the inner loop's characteristics are fixed. You will, in general, find that the outer loop's bandwidth is significantly less than the inner loop's bandwidth -- but you will also often find out that the control system as a whole is much better behaved, and often faster, with two loops. You often don't want to tune the inner loop as tightly as you might if it were the only loop -- having an inner loop that is Really Well Damped is usually better than having one that has the World's Highest Bandwidth. The best compromise is between these two extremes, but leaning toward well damped. If you want better bandwidth from the inner loop it's usually better to use some judicious feed-forward; that usually gives you better bandwidth and phase roll-off characteristics at the cost of accuracy, but since you're supplying accuracy with the outer loop that's usually OK. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com
On 10/1/2013 2:46 PM, Tim Wescott wrote:
> On Tue, 01 Oct 2013 14:02:23 -0400, Phil Hobbs wrote: > >> On 10/01/2013 12:37 PM, George Herold wrote: >>> On Tuesday, October 1, 2013 11:19:24 AM UTC-4, Phil Hobbs wrote: >>>> On 10/01/2013 11:04 AM, George Herold wrote: >>>> >>>>> On Saturday, September 28, 2013 9:23:39 AM UTC-4, Glenn B wrote: >>>> >>>>>> On 28/09/2013 1:33 PM, mt wrote: >>> <snip> >>> >>>>> Hmm I bought one of their TEC controllers years ago. I had to hack >>>>> it to get a longer integration time... and then just ended up making >>>>> my own. >>>> >>>>> The longer integration time gets to the debate of where to put the >>>>> temperature sensor. Near the TEC, or near the laser diode. I opted >>>>> for near the diode, and so a longer time constant. But others think >>>>> otherwise, and I very well could be wrong. >>>> >>>> >>>>> >>>>> George H. >>>> >>>> >>>>> >>>> There's no reason to have just one sensor. You can wrap the actuator >>>> in a local first-order loop with some fairly crude but fast sensor >>>> right at the actuator, and use that as an improved actuator in an >>>> integrating loop with a sensor where you actually want to control the >>>> temperature. >>>> >>>> That gives you high accuracy plus decent rejection of thermal forcing. >>>> >>> OK, having your cake and eating it too. >>> Hmm, so you roll off the fast loop output at some frequency where the >>> slow loop takes over, (high pass filter on the fast one) and then sum >>> them. >>> The details of that hand-off are -interesting-. >>> Is a single pole roll off the way to go? >>> (Drawing some PI bode plots I seem to be confusing myself. Do you roll >>> the slow loop with a single pole low pass too?) >>> >>> >>> >>> ^ LOG(gain) >>> | >>> |\ \ >>> | \ \ >>> | \ \ <--fast loop | \ \ >>> | \ \ >>> | \ \ >>> | \<--slow | >>> | >>> |---------------------->LOG(freq) >>> >>> >>> ^gain | >>> |\ >>> | \ >>> | \ /\ >>> | \ / \ >>> | \ \ >>> | \ \ >>> | >>> | ^-----rolled off here. >>> | >>> |---------------------->freq >>> >>> But how does one pick the frequency? >>> >>> >> It's usually pretty simple IME--you make the inner loop as fast as you >> can with decent stability, and then do the same with the inner one. >> Normally there's enough bandwidth difference that any small phase >> whoopdedoos in the closed-loop response of the inner loop don't affect >> the outer loop much. > > After you tune the inner loop twice, then be sure to tune the outer loop. > > (or fix your typos). > > ((snicker)) >
Your mother was a hamster and your father smelled of....Elderberries! (Apologies to Monty Python) Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 USA +1 845 480 2058 hobbs at electrooptical dot net http://electrooptical.net
On Tuesday, October 1, 2013 2:42:36 PM UTC-4, Tim Wescott wrote:
> On Tue, 01 Oct 2013 07:56:45 -0700, George Herold wrote: >
<snip>
> > The most important thing (IMHO) for a TEC is sizing the heat sink. If > > it's free air cooled then you need a pretty big heat sink compared to > > the size of the TEC. It's just a matter of energy conservation, all the > > heat that flows through the TEC has to also leave the heatsink. > > Don't you mean the heat that flows through as well as the heat produced > by? I thought that TEC's generated heat on their own, which accounted > both for much of the nonlinearity and for their renowned inefficiencies. >
Yeah that too! (thanks) ~1/2 the time I'm using TEC's as heaters, where the internal heat generated is a 'bonus'. The great thing about TEC's is they go both ways. George H.
> > -- > > > > Tim Wescott > > Wescott Design Services > > http://www.wescottdesign.com
On Tuesday, October 1, 2013 2:51:53 PM UTC-4, Tim Wescott wrote:
> On Tue, 01 Oct 2013 09:37:49 -0700, George Herold wrote: >=20
<snip.. go upstream to read previous stuff>
> In general when you do a two-loop control system you don't try to roll=20 > off the control of the outer loop or any of that razzmatazz -- you tune=
=20
> the inner loop for good behavior in a conventional way, then you say=20 > "gosh, look at this plant that I have" and you tune the outer loop to the=
=20
> inner loop's characteristics. You may do some iteration between inner=20 > loop and outer loop tuning (usually to relax the inner loop's gains --=20 > see below), but in general you want to tune the inner loop in isolation,=
=20
> then tune the outer loop as if the inner loop's characteristics are fixed=
.
>=20 > You will, in general, find that the outer loop's bandwidth is=20 > significantly less than the inner loop's bandwidth -- but you will also=
=20
> often find out that the control system as a whole is much better behaved,=
=20
> and often faster, with two loops. >=20 > You often don't want to tune the inner loop as tightly as you might if it=
=20
> were the only loop -- having an inner loop that is Really Well Damped is=
=20
> usually better than having one that has the World's Highest Bandwidth. =
=20
> The best compromise is between these two extremes, but leaning toward=20 > well damped. If you want better bandwidth from the inner loop it's=20 > usually better to use some judicious feed-forward; that usually gives you=
=20
> better bandwidth and phase roll-off characteristics at the cost of=20 > accuracy, but since you're supplying accuracy with the outer loop that's=
=20
> usually OK. >=20 > --=20 >=20 > Tim Wescott >=20 > Wescott Design Services >=20 > http://www.wescottdesign.com
Gee that was nice, thanks Tim and Phil, =20 I think I'd have to build/(or tweak) one to really understand. I guess I'm trying to see this in the frequency domain. (let's stick with = thermal control with a PI loop, and no long thermal paths) I picture the ga= in as like the open loop response of an opamp. (Maybe that's wrong?) So is= the DC gain of the fast/inner loop always less than the slower/outer? And= I can then add them together with 'no worries'? =20 George H.
On Tue, 01 Oct 2013 19:45:43 -0700, George Herold wrote:

> On Tuesday, October 1, 2013 2:51:53 PM UTC-4, Tim Wescott wrote: >> On Tue, 01 Oct 2013 09:37:49 -0700, George Herold wrote: >> > <snip.. go upstream to read previous stuff> >> In general when you do a two-loop control system you don't try to roll >> off the control of the outer loop or any of that razzmatazz -- you tune >> the inner loop for good behavior in a conventional way, then you say >> "gosh, look at this plant that I have" and you tune the outer loop to >> the inner loop's characteristics. You may do some iteration between >> inner loop and outer loop tuning (usually to relax the inner loop's >> gains -- see below), but in general you want to tune the inner loop in >> isolation, then tune the outer loop as if the inner loop's >> characteristics are fixed. >> >> You will, in general, find that the outer loop's bandwidth is >> significantly less than the inner loop's bandwidth -- but you will also >> often find out that the control system as a whole is much better >> behaved, and often faster, with two loops. >> >> You often don't want to tune the inner loop as tightly as you might if >> it were the only loop -- having an inner loop that is Really Well >> Damped is usually better than having one that has the World's Highest >> Bandwidth. The best compromise is between these two extremes, but >> leaning toward well damped. If you want better bandwidth from the >> inner loop it's usually better to use some judicious feed-forward; that >> usually gives you better bandwidth and phase roll-off characteristics >> at the cost of accuracy, but since you're supplying accuracy with the >> outer loop that's usually OK. >> >> -- >> >> Tim Wescott >> >> Wescott Design Services >> >> http://www.wescottdesign.com > > Gee that was nice, thanks Tim and Phil, > I think I'd have to build/(or tweak) one to really understand. > > I guess I'm trying to see this in the frequency domain. (let's stick > with thermal control with a PI loop, and no long thermal paths) I > picture the gain as like the open loop response of an opamp. (Maybe > that's wrong?) So is the DC gain of the fast/inner loop always less > than the slower/outer? And I can then add them together with 'no > worries'?
I think that using an op-amp as an example is problematical -- there's subtleties there that could trip you up if you think of them and I don't, or visa-versa. There's not a whole lot of universal statements you can make on this, but often you don't use integral control on the inner loop, and you do on the outer. In those cases then the inner loop's DC gain is lower, because it's finite compared to the outer loop's infinite DC gain due to the integrator. If there _is_ a universal statement that can be made, it's that you do the multiple loop thing when you have to give up too much to do things with a single loop. Often you're implementing an inner loop with an extra sensor, and that extra sensor is letting your control system see some state that's not easily accessible to the "main" sensor. In a motion control system that inner sensor might be velocity, or motor current (or both, with more than one "inner" loop). In your heating system, having that inner sensor be a monitor on the temperature of your heating element itself lets you know how much heat is going to be flowing into your system after you turn things off. -- Tim Wescott Control system and signal processing consulting www.wescottdesign.com
"George Herold" <gherold@teachspin.com> wrote in message 
news:597b4f0e-097f-4a5b-984f-57e50337d3f1@googlegroups.com...
> OK, having your cake and eating it too. > Hmm, so you roll off the fast loop output at some frequency where the > slow loop takes over, (high pass filter on the fast one) and then sum > them. > The details of that hand-off are -interesting-. > Is a single pole roll off the way to go? > (Drawing some PI bode plots I seem to be confusing myself. Do you roll > the slow loop with a single pole low pass too?)
Don't think loops side-by-side, or split in parallel: think composition, putting one loop inside another. f(g(x)), not f(x) + g(x). As an example, suppose you have a dumb buck converter: a PWM modulator driving a transistor. So, whatever control voltage sets PWM%, that's what it's chopping at. The buck feeds an inductor, which (ignoring what's after it) integrates PWM into switching current. If you just set PWM willy-nilly, you'll get all sorts of crazy currents, because for a fairly small increase in PWM, current just keeps climbing and climbing and climbing! So, we put a current sensor in there (a shunt resistor, or a DC current probe, whatever), and put a loop around the PWM so as to control current. Since PWM is first order to current, this loop can be pretty fast: generally, the error amp's pole will be near maybe half the switching frequency, so as to filter out switching ripple, without being needlessly slow, or so fast as to be unstable. But a buck converter is usually for constant-voltage output, so we need *another* loop. This one sets the current setpoint for the inner loop, and monitors the output voltage on the filter capacitor. The cap (in and of itself) first-order integrates current, so again, we can put a fairly simple error amp around it. Typically this pole will be 1/3 to 1/10th the inner loop, and you want to have extra on there in case the load has extra bypass caps or whatever. One might ask, if voltage is directly proportional to PWM, why not control PWM by voltage in the first place? And the problem is that pesky inductor inbetween. You always want to control inductor current first, since it's also the switch current, and limiting that keeps the magic smoke in! Tim -- Seven Transistor Labs Electrical Engineering Consultation Website: http://seventransistorlabs.com
On Wednesday, 2 October 2013 00:56:45 UTC+10, George Herold  wrote:
> On Saturday, September 28, 2013 9:20:34 AM UTC-4, Bill Sloman wrote:=20 > > On Saturday, 28 September 2013 15:33:32 UTC+10, mt wrote:=20 > > > I have a laser diode that is cooled by a fan and a heat sink. I have =
a small TEC lying around and I would like to integrate that to the laser di= ode head. How do I go about controlling the voltage supply to maintain a st= able temperature.=20
> > =20 > > You need to think in terms of current control rather than voltage - TEC=
's transfer watt per amp,=20
> > =20 > > Sadly, the watts transferred per amp varies fairly rapidly with tempera=
ture difference you are trying to sustain across the TEC, and a proper cont= roller needs to measure the temperature on both sides of the junction. =20
> > =20 > > Sloman A.W., Buggs P., Molloy J., and Stewart D. =93A microcontroller-b=
ased driver to stabilise the temperature of an optical stage to 1mK in the = range 4C to 38C, using a Peltier heat pump and a thermistor sensor=94 Measu= rement Science and Technology, 7 1653-64 (1996)
> > > > spells it all out in fair detail, including a formula for heat transfer=
per amp, which none of the application notes seem to include. E-mail me - = at bill.sloman@ieee.org - if you can't get hold a of a copy.
> =20 > Well as Bill says current control is a bit better. But if you are wrappi=
ng a control loop around it (PI) then you can do it with just voltage too. = =20 Keeping the control loop critically damped becomes an issue. As the voltage= difference across the TEC changes the gain around a simple loop change als= o changes. =20
> The most important thing (IMHO) for a TEC is sizing the heat sink. If it=
's free air cooled then you need a pretty big heat sink compared to the si= ze of the TEC. It's just a matter of energy conservation, all the heat tha= t flows through the TEC has to also leave the heatsink. =20 Worse, the current you are driving through the heat sink to move the heat g= enerates more heat in the resistance of the heat sink. Big heat sinks can be vital. --=20 Bill Sloman, Sydney
On 10/1/2013 10:45 PM, George Herold wrote:
> On Tuesday, October 1, 2013 2:51:53 PM UTC-4, Tim Wescott wrote: >> On Tue, 01 Oct 2013 09:37:49 -0700, George Herold wrote: >> > <snip.. go upstream to read previous stuff> >> In general when you do a two-loop control system you don't try to roll >> off the control of the outer loop or any of that razzmatazz -- you tune >> the inner loop for good behavior in a conventional way, then you say >> "gosh, look at this plant that I have" and you tune the outer loop to the >> inner loop's characteristics. You may do some iteration between inner >> loop and outer loop tuning (usually to relax the inner loop's gains -- >> see below), but in general you want to tune the inner loop in isolation, >> then tune the outer loop as if the inner loop's characteristics are fixed. >> >> You will, in general, find that the outer loop's bandwidth is >> significantly less than the inner loop's bandwidth -- but you will also >> often find out that the control system as a whole is much better behaved, >> and often faster, with two loops. >> >> You often don't want to tune the inner loop as tightly as you might if it >> were the only loop -- having an inner loop that is Really Well Damped is >> usually better than having one that has the World's Highest Bandwidth. >> The best compromise is between these two extremes, but leaning toward >> well damped. If you want better bandwidth from the inner loop it's >> usually better to use some judicious feed-forward; that usually gives you >> better bandwidth and phase roll-off characteristics at the cost of >> accuracy, but since you're supplying accuracy with the outer loop that's >> usually OK. >> >> -- >> >> Tim Wescott >> >> Wescott Design Services >> >> http://www.wescottdesign.com > > Gee that was nice, thanks Tim and Phil, > I think I'd have to build/(or tweak) one to really understand. > > I guess I'm trying to see this in the frequency domain. (let's stick with thermal control with a PI loop, and no long thermal paths) I picture the gain as like the open loop response of an opamp. (Maybe that's wrong?) So is the DC gain of the fast/inner loop always less than the slower/outer? And I can then add them together with 'no worries'? > > George H. >
Tim and I are talking about wrapping the outer loop around the inner loop, like using a closed-loop buffer inside an op amp feedback loop. Adding them is a more difficult problem, but not impossible as long as the slow loop can be made to have lower gain at all frequencies. Otherwise you get whoopdedoos where the gain curves cross. For instance, I made a laser frequency locker a couple of years ago that used current and temperature in a single loop. It locked up using temperature, and when the (much narrower range) current-tuning loop came into range, its higher gain immediately dominated. (The temperature loop kept the current loop centred.) Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 USA +1 845 480 2058 hobbs at electrooptical dot net http://electrooptical.net
On 10/01/2013 11:23 PM, Tim Wescott wrote:
> On Tue, 01 Oct 2013 19:45:43 -0700, George Herold wrote: > >> On Tuesday, October 1, 2013 2:51:53 PM UTC-4, Tim Wescott wrote: >>> On Tue, 01 Oct 2013 09:37:49 -0700, George Herold wrote: >>> >> <snip.. go upstream to read previous stuff> >>> In general when you do a two-loop control system you don't try to roll >>> off the control of the outer loop or any of that razzmatazz -- you tune >>> the inner loop for good behavior in a conventional way, then you say >>> "gosh, look at this plant that I have" and you tune the outer loop to >>> the inner loop's characteristics. You may do some iteration between >>> inner loop and outer loop tuning (usually to relax the inner loop's >>> gains -- see below), but in general you want to tune the inner loop in >>> isolation, then tune the outer loop as if the inner loop's >>> characteristics are fixed. >>> >>> You will, in general, find that the outer loop's bandwidth is >>> significantly less than the inner loop's bandwidth -- but you will also >>> often find out that the control system as a whole is much better >>> behaved, and often faster, with two loops. >>> >>> You often don't want to tune the inner loop as tightly as you might if >>> it were the only loop -- having an inner loop that is Really Well >>> Damped is usually better than having one that has the World's Highest >>> Bandwidth. The best compromise is between these two extremes, but >>> leaning toward well damped. If you want better bandwidth from the >>> inner loop it's usually better to use some judicious feed-forward; that >>> usually gives you better bandwidth and phase roll-off characteristics >>> at the cost of accuracy, but since you're supplying accuracy with the >>> outer loop that's usually OK. >>> >>> -- >>> >>> Tim Wescott >>> >>> Wescott Design Services >>> >>> http://www.wescottdesign.com >> >> Gee that was nice, thanks Tim and Phil, >> I think I'd have to build/(or tweak) one to really understand. >> >> I guess I'm trying to see this in the frequency domain. (let's stick >> with thermal control with a PI loop, and no long thermal paths) I >> picture the gain as like the open loop response of an opamp. (Maybe >> that's wrong?) So is the DC gain of the fast/inner loop always less >> than the slower/outer? And I can then add them together with 'no >> worries'? > > I think that using an op-amp as an example is problematical -- there's > subtleties there that could trip you up if you think of them and I don't, > or visa-versa. > > There's not a whole lot of universal statements you can make on this, but > often you don't use integral control on the inner loop, and you do on the > outer. In those cases then the inner loop's DC gain is lower, because > it's finite compared to the outer loop's infinite DC gain due to the > integrator. > > If there _is_ a universal statement that can be made, it's that you do > the multiple loop thing when you have to give up too much to do things > with a single loop. Often you're implementing an inner loop with an > extra sensor, and that extra sensor is letting your control system see > some state that's not easily accessible to the "main" sensor. In a > motion control system that inner sensor might be velocity, or motor > current (or both, with more than one "inner" loop). In your heating > system, having that inner sensor be a monitor on the temperature of your > heating element itself lets you know how much heat is going to be flowing > into your system after you turn things off. >
With a TEC and decent insulation, the main external thermal forcing comes by conduction through the TEC itself. Thus having an inner loop measuring the cold plate temperature right at the TEC improves your forcing rejection by a lot. Cheers Phil Hobbs -- Dr Philip C D Hobbs Principal Consultant ElectroOptical Innovations LLC Optics, Electro-optics, Photonics, Analog Electronics 160 North State Road #203 Briarcliff Manor NY 10510 hobbs at electrooptical dot net http://electrooptical.net