Closed-loop power control of lasers

[heroic]

So, in the TEC thread, I mentioned that I'd been playing around with an idea to make lasers more linear in modulation response by using a photosensor in the spill light near the output and varying the modulation input using a microcontroller.

That's only useful if you're building your own DPSS laser (or modifying an existing). I wouldn't want to use it with a direct injection laser.

I don't see why not- being able to time-delay the DI's mod input so that it lines up with the response of your DPSSes seems quite valuable. It's probably possible to do this with some software, but the come-on time of my cheap Chinese DPSS lasers seems to vary substantially with temperature. I also get quite bad not-quite-blanking on my green (mutter, mutter cheaper than the blue mutter) when it's hot; I suspect the threshold moves. All of this can be corrected with a little feedback.

If you do, you must consider that the wavelength changes with temperature.

That only matters if I am concerned about absolute intensity or stability on a minute-by-minute basis rather than on a second-by-second basis. The fast changes are more visually intrusive than the gradual ones (high spatial frequencies are more visible than low ones, see any book on the discrete cosine transform) so it's those I want to get rid of first.

Also that longer wavelengths have a higher photon flux at a given radiant power, but at the same time Rayleigh scattering will be more efficient at shorter wavelengths. Looking at the relevant equations I get that the wavelength dependence of the photodiode current is (λ[1]/λ[2])^3.

I'm actually using light-to-voltage convertor sensors (a snip at a buck a pop, and including temperature compensation, dark current compensation, bias current generator and preamp in one package!) which ameliorates some of these concerns, but again, it's not a major issue.

The path length between the opening of the laser housing and the sensor is less than a centimetre, and I am working not off the scattered light, but off the off-axis light produced by imperfections in the cheap optics inside the device. The sensor is very highly directional but the degree of crosstalk remains to be seen- it may or may not be a problem. Finally, we recalibrate the scale every time the laser is commanded to go to the maximum or minimum voltage. The control scheme I have implemented turns out to be rather clever if I do say so myself.

That is of course assuming there are no particles in the air. If there is, Mie scattering will dominate and the equation becomes nearly impossible to solve.

Again, you're assuming I care about quantitative results, when a relative result is good enough for this application.

For a DPSS laser it is of course completely different, but what you are talking about here is close to building a whole driver circuit.

Not really. I don't have to worry about protecting the diodes, driving heavy currents, or doing any of the heavy lifting of actually driving a pump diode array.

It goes like this:

Control voltage from the controller card into one of the ADCs on the Atmel.

Output of photodiode device into a second ADC on the Atmel.

SPI on Atmel to a 12-bit DAC, and the output of the DAC to the modulation input of the laser.

One digital pin of the Atmel to the shutter interlock circuit

When power is applied to the Atmel, it pulls the shutter interlock low to prevent the shutter from being opened. It drives the laser to zero and takes a reading from the photosensor. This is zero. It then ramps up the laser modulation voltage until it gets a non-zero (with a little bit of noise allowed for) reading from the photosensor. This is control point one. Then it continues to ramp up until the reading from the photosensor no longer climbs. This is control point two. The laser is turned off, the shutter is released (tri-stated, so the normal safety system can take over) and things start happening. :-)

The reason this isn't just plain PID is that PID is really sensitive to error, and we have error creeping in in two places- in our measurements, and in the response of our controlled device.

A lookup table is constructed mapping 256 evenly spaced points on the continuum of photosensor voltages to the appropriate modulation voltages. Since the Atmel has 10-bit ADCs and the DAC is a 12-bit part, there is plenty of accuracy to lose in this way without succumbing to confusion.

The controller then enters a loop, reading the command voltage, estimating the appropriate modulation voltage using the table, reading the sensor output and keeping a count of the error percentage. When the error percentage exceeds a certain cumulative limit, it's time to recalibrate. Recalibration is triggered by the command voltage either exceeding the previous record high (read low) (a running "max_command_intensity" variable is kept) or the error percentage cumulative total exceeding limits.

Running recalibration is done not by shuttering the laser and recalibrating the way the base calibration is done, but by waiting for the command input to include a suitable range of inputs. Since many patterns include 100% brightness and 100% blanking, this happens quickly.

I am not completely sure that this running recalibration is the best way to do things, either, but it seems to work.

The output routine feeds a circular buffer which is emptied under interrupt to implement the variable delay. The three processors- one for each colour- measure their delay constantly (since it is naturally measured by the feedback sensor after every change in intensity) and communicate this to each other via I2C. (Currently they just communicate it to me, personally, via USB, but still.)

Even so, I don't think you should use 'spill' light; the error will simply be too big unless that photodiode is inside of the laser module, which must be hermetically sealed and dust free. I am of the opinion that the sensor must be in the beam.

The fun of this is that I'm not sure you're not right- I guess we're going to find out! This is a fun project with lots of things to learn.

[heroic]

[not using spill light]
I've been thinking that too. Though as a photodiode can be sensitive, and shrouded easily if a neat beam is to enter, it might be best to use a bit of microscope slide glass at a Brewster angle to part off just enough light for the sensor while leaving most intact.

I've also been thinking about putting two of my sensors behind the dichros, so they get the beam that goes straight through; my cheap dichros leak quite a lot and what goes straight through seems to be single wavelength from my not-so-careful inspection with a diffraction grating.

[mixedgas]

If I can get some other stuff done this weekend, I will finish hooking up a lexel light feedback card to the DPSS.


The thing I did in the past for linear fades when I only had a ON/OFF ttl input AOM was PWM, 8 bit PWM at 6 PWM cycles per point, gives really nice linear fades while keeping the light source at constant brightness. This gets rid of that sinking DPSS feeling. The AOM could modulate on/off at 7 mhz.

Thats something to keep in mind. I ran a flash A/D into a simple TTL counter/comparator PWM circuit, and everybody who ever saw it was amazed at the linearity. Since the clock for the PWM and flash A/D was not tied to the graphics generator (I hate the term dac) , it did a sliding average between points.
cost was about 16$ for the flash AD and 3$ worth of TTL. I gave the boards to a friend who needed RGB, so I'd have to find a new A/D chip, as harris intersil doesn't make the RCA part I used.

here is a online redraw of a lexel light loop:

http://repairfaq.ece.drexel.edu/sam/88csch.gif

Heroic, I can mail you a proper pickoff glass if you need one, its a simple not quite AR coating. Usually .5% for small lasers. I also have some commercial ALo2 attenuator disks as well. When repairing a lexel argon photo pickoff , you build up stacks of the sintered Alo2 disks and faint blue theatrical gel to correct the photocell response VS wavelength against a thermal power meter. So they are stock parts around here.

Steve

[heroic]

If I can get some other stuff done this weekend, I will finish hooking up a lexel light feedback card to the DPSS.


The thing I did in the past for linear fades when I only had a ON/OFF ttl input AOM was PWM, 8 bit PWM at 6 PWM cycles per point, gives really nice linear fades while keeping the light source at constant brightness. This gets rid of that sinking DPSS feeling. The AOM could modulate on/off at 7 mhz.

Obviously I can't modulate the DPSS that fast, even my TTL only goes up to 30 kHz because of the time constant in the feedback loop it uses for the pump diode power control. Even if I did, that 800 usec high state lifetime for Nd is going to bite. Fortunately, I actually do have analogue control

Heroic, I can mail you a proper pickoff glass if you need one, its a simple not quite AR coating. Usually .5% for small lasers.

That's very kind of you, but I want to see if I can get this working without one (or three, for that matter). For one thing I don't want to consume any more of the light that's currently hitting the galvos if I can avoid it, for another thing I want to make this system as cheap as humanly possible so that when I release the design someone else might have a chance of using it, and for the other other thing I don't think I actually have room in my optics enclosure for more than the very smallest extra stuff- it's less than a foot long and about seven inches wide, and it contains three laser heads and three mirrors already.

[weartronics]

I hope it won't diminish your enthusiasm to build such a system yourself, but you might be interested to know that Eye Magic also sell a system to accomplish this. I believe it uses a PID loop, similar to galvo feedback.

http://www.eyemagic.gr/laser-iris-colorsafe.html

Best regards,

weartronics

[Pangolin]

Hi Heroic,

This whole discussion is actually what Francesco has done at Pangolin over the summer. Remember when I said ESD protection wasn't really our main intention when the project started, and we felt that was "step 1"? Well, we got past "step 1" and investigated a bit of what you are talking about.

One problem we found is that the thermal time constant of TEC inside the lasers is on the order of 10 seconds!! If the TEC was much smaller, and the people building lasers was much smarter, the time constant could have been made on the order of less than 1 second, and then what you are talking about would be more practical.

There is another problem too. The lasers are highly nonlinear and, at times, even non-monotonic. This is especially the case for DPSS blue lasers, but we also see it with DPSS green as well.

However, wrapping a servo loop around the light output of a laser (that isn't DPSS) or simply disregarding the TEC does in fact work, and there is another company doing just this with OK results.

In any event, your idea is a good idea. We had the same idea too, and investigated it. If only a) the thermal time constant was much much much faster; and b) the laser output was not noisy and non-monotonic, it might have worked...

Bill

[heroic]

There is another problem too. The lasers are highly nonlinear and, at times, even non-monotonic. This is especially the case for DPSS blue lasers, but we also see it with DPSS green as well.

However, wrapping a servo loop around the light output of a laser (that isn't DPSS) or simply disregarding the TEC does in fact work, and there is another company doing just this with OK results.

Hi Bill,

I'm only trying to control the temperature as a side effect of my real intent, which is to control the light output directly. Non-linearity is, after all, the hallmark of any system with gain, and the A in laser does stand for amplifier.

Monotonicity does not bother me. I have a background in control systems and having put the laser on the bench and scoped it, I think that though the short term modehops are difficult to handle, I can make a substantial improvement in the longer-term drift and modulation linearity.

I have a test bench in front of me that shows this to be the case- now all I have to do is speed the code up by a factor of ten... shouldn't be too hard :-)

[Pangolin]

Hi Heroic,

Well, sounds like a good project for you. Be sure to let us know how it goes...

Bill

[heroic]

Hi Heroic,

Well, sounds like a good project for you. Be sure to let us know how it goes...

Bill

I certainly will- I'm going to release the code and designs as open source!

So far I've managed to achieve 30 kHz bandwidth, closed-loop controlling an LED, and a little less than this driving a 5 mW laser pointer diode. Still, is promising, no?

[drlava]

Like Bill said, there are many instabilities in the blue and green due to non-ideal thermal matching and control. Even with a 1s thermal time constant there would be noticeable changes in brightness as the average power dissipation in the laser changed. You might be able to feed back fast enough to combat some of the noise but not all of it. Also as the modulation level increases the feedback control will not be perfect as you have less and less correction headroom. In the end, it's still a good project and a useful tool, especially for getting accurate greys.