Brainstorm please! High speed (>20krpm) IR reflective optical trigger

[max_torque]

Hi all!

Anyone fancy a brain storm??   


I'd like to generate a TTL level edge coherent to the edges of the teeth (probably 6 teeth) on an optical trigger disc, that could be spinning at between 20 and 30krpm!  That TTL signal needs to be consistent in it's position verses the true position of the teeth, but can have a latency / phase offset, as long as that offset remains reasonably constant (or is small enough to not worry about) across the speed range.

Previously (different project) i used an IR LED / LDD, and fired 200ns pulses from the LED at a 1.5MHz rate, and with a transimpedance amp generated a return signal from the LDD.  A  suitable high pass filter made the system robust against ambient IR levels.

This time, i'd like to get less latency and jitter, which due to the way the previous system worked could be as much as 1us.

So, ideas please?? 

(some guess system design values could be, subject to change:  50ns light pulses, 5MHz rate, max 500ns latency and "Low" jitter)


NB currently i am only considering an optical system, and not any alternatives (ie VR / hall sensor etc)  to date i can't really find too much info about how those electrical systems cope with high speed (it strikes me the inductance of those sensors is going to result in a phase shift vs speed)

[sokoloff]

That's actually not all that fast electrically. It's 500Hz at the outside at one pulse per rev or 3 kHz at 6 pulses per.

Have you considered/tried a comparator based system where you shine a constant light at the target, sense the return, and compare the instantaneous return to the average return via a comparator? You might also be able to look into a PLL because the frequency of this pulse train will not change instantaneously due to the mass of the rotating system.

Alternatively, are you able to modify the system under test to include a slit and use a slotted sensor?

[max_torque]

The average frequency of the return signal is not fast

15krpm with 6 teeth is 'just' 1.5khz ie an edge every 666.7us.  However, i need to be able to (ideally) resolve 1rpm in 15,000 at each of those teeth, which is about 11ppm.  That needs a 22Mhz clock for the tooth interval counter, for a clock resolution of 45ns. So whilst absolute phase offset from each physical tooth is irrelevant, the jitter in the tooth edge trigger signal needs to be very low indeed.......

[Siwastaja]

An IR LED, and a fast, reverse biased photodiode, and a fast comparator, closely coupled? Use enough illumination power and use a tight optical bandpass filter on the photodiode to reject ambient light in a simple way, without the need of correlated double sampling tricks.

What is LDD?

[sokoloff]

The rotational inertia of the system is your friend here (assuming the available torque to accelerate it is reasonably low).
You may need 11ppm, but you (almost surely) don't need that level of precision from a single measurement.

[Conrad Hoffman]

Don't forget that the accuracy of the target and especially the centering, will be critical. If mechanically possible, I'd be wanting a chrome on glass target, used as an interrupter, not reflective. Then do what Siwastaja says.

[max_torque]

LDD was my short hand for (any) form of Light Detecting Diode!

That system you suggest is pretty much how the last system worked. It's limitations were in terms of driving the LED with a sharp enough pulse and in getting enough gain and enough bandwidth into the detector without getting swamped by noise (mostly noise that comes from the LED driver actually!)   All easy things to say, but more complex and time consuming to actually do!

[max_torque]

The system will use an "inertial driven coastdown" calibration to work out the intrinsic geometrical errors in the target wheel.

[Siwastaja]

I don't understand the need for blinking the LED? It would just quantize the detection into discrete intervals, which only adds quantization noise / jitter?

Why not just keep the LED on and measure the reflective pulse?

Anyway, even if you need to, a small LED should blink just fine with about 20ns rise and fall times (90%-10%), without fancy drive circuits (not even push-pull required). Driving it at larger current makes the LED actually turn off faster (or so I have read; I tried to measure the effect and didn't see much difference, so take this as a grain of salt.) If you really need shorter rise/fall, then a laser diode is the answer.

[max_torque]

The blinking is only to provide a known AC component of the reflected signal, to which the detector can be tuned in a fairly narrow band, and hence attempt to reject spurious IR sources (the LDD i used previously included a reasonably narrow band optical filter) because i cannot ensure the sensor is operated in total darkness

[jbb]

You mention that noise from the LED driver itself is an issue. Have you carefully reviewed the electrical design of the system? Used power supply filtering and shielding on the photodiode and TransImpedance Amplifier (TIA) front end? We did this on an IR comms project and shielding helped a lot.

I too think you should look at continuous LED operation. In order to deal with noise, I have two thoughts:
1) could you add a 2nd photodiode which is close to the first one, but does not see reflected pulses? Maybe it could be used to detect abient noise in some differential manner.
2) How about feeding the noisy photodiode output into some kind of filter, such as a PLL? At the expense of some response time delay, a PLL effectively operates as a very narrow-band tunable filter and might be able to reject abient noise (note: not good at very low speeds)
3) Using a constant-on laser might be effective by simply providing massive signal to drown out the noise. Make sure the laser is eye safe (for the love of god don’t use a green eBay one, they’re super dodgy), apply warning labels (laser not eye safe if someone goes looking with a magnifying glass) and maybe apply an attenuator right on top of the photodiode to prevent saturation.

[max_torque]

A few interesting ideas there!  Perhaps the massive-intensity-drown-out solution has some merit!      One small advantage of the pulsed solution is that i can fairly easily work out the latency / response time of the detector circuit, which is a bit harder with a constantly on light source, because then you have to somehow measure the physical "chopper" to work out what the delay is??

Another thing i need to look at is the physical form factor of the optics, because a thin window produces a sharper edge, as compared to a wide one.  (because it's area reduces faster per angular degree as the interrupter disc rotates.  Some form of thin transparent film (ie selotape!!) could be used i think, sandwiched into some sort of opaque material to provide a narrow slit type arrangement. Obviously the alignment of the two halves becomes more significant, but i can easily machine a solid metal type support......

[jbb]

I guess that you’d have to measure the photodiode response with a pulsed circuit during design verification. Multi MHz bandwidths are possible using a fairly standard photodiode and TIA.

I think you could form a slit with a laser printer and an OverHead Projector (OHP) sheet (perhaps glued to something stronger) (also make sure the OHP sheet is rate to go through a prenter; you don’t want it melting!!).

If you want to get the highest bandwidth, here are some thoughts:
- photodiodes have stray capacitance which limits bandwidth
- if you bias them with a moderately high reverse voltage (could be 10 - 30V), the capacitance drops and bandwidth increases
- make sure the bias voltage is clean and well filtered because voltage noise will here will pass through the photodiode
- placing the TIA right next to the photodiode minimises cable capacitance and the area most sensitive to electrical noise
- don’t use excessive gain on the TIA because it will reduce bandwidth. You can add a second gain stage if you need it
- consider using an AC coupled gain stage to reject changes in DC levels. Then use a comparator to for the edges.

I think Analog Devices have app notes about this kind of thing.

[StillTrying]

"A few interesting ideas there!  Perhaps the massive-intensity-drown-out solution has some merit!"

Definitely! 10-20mA would be more than enough through a super bright visible LED, even with almost no sheilding of extenal light and some EMI, edge jitter should be only ~5ns.

"Another thing i need to look at is the physical form factor of the optics, because a thin window produces a sharper edge, as compared to a wide one."

Yes, the resolutions all depend on the mechanics.
Assuming the 6 toothed optical disk was 30mm diameter, and the light beam was only 0.1mm diameter, then it still takes 2us for a tooth edge to cross the beam width at 30krpm. You'd have to always trigger at the analogue center of the 2us transition. 0.01mm differences in the width of the teeth would cause 200ns of jitter.

But with such a strong signal you don't need a too fancy TIA.

[TurboTom]

A typical application of a laser-based frequency measurement arrangement is an RPM pickup for micro gas turbine engines. I've worked on such systems up to 250,000 RPM (single through-hole) so frequency was up to 8.333kHz. LED-based systems also work but the laser has got the huge advantage of a much more narrow beam and higher intensity over a very small space angle. This makes the pulses it generates much better defined. What's been told before regarding reverse-biasing the PD is correct, if possible also use narrow-bandwidth optical filtering to eliminate as much stray light as possible. A high-speed current / voltage converter (inverting fast OPAMP) as the first amplifier for the photo diode will result in the voltage across the diode to stay for all practical means constant, thus almost eliminating the effect of the PD's capacitance. After that, some schmitt trigger can be used to provide a proper digital signal.

All that said, in the mentioned field (micro gas turbine engines), common sense is to do without optical measurements if possible at all and rather use magnetic / inductive systems. This is simplified by the fact that usually it's not necessary to measure RPM of the engine rotor right down to zero. In applications where the engine is driving an alternator, a dedicated RPM pickup obviously isn't necessary.

[hans]

I guess that you’d have to measure the photodiode response with a pulsed circuit during design verification. Multi MHz bandwidths are possible using a fairly standard photodiode and TIA.

I think you could form a slit with a laser printer and an OverHead Projector (OHP) sheet (perhaps glued to something stronger) (also make sure the OHP sheet is rate to go through a prenter; you don’t want it melting!!).

If you want to get the highest bandwidth, here are some thoughts:
- photodiodes have stray capacitance which limits bandwidth
- if you bias them with a moderately high reverse voltage (could be 10 - 30V), the capacitance drops and bandwidth increases
- make sure the bias voltage is clean and well filtered because voltage noise will here will pass through the photodiode
- placing the TIA right next to the photodiode minimises cable capacitance and the area most sensitive to electrical noise
- don’t use excessive gain on the TIA because it will reduce bandwidth. You can add a second gain stage if you need it
- consider using an AC coupled gain stage to reject changes in DC levels. Then use a comparator to for the edges.

I think Analog Devices have app notes about this kind of thing.

And add a spot for a feedback capacitor parallel to the feedback resistor of the TIA.

The TIA will most likely cause severe peaking in bode plot (thus horrible ringing if you drive a square/pulse test signal), since the input impedance of a TIA looks inductive and the photodiode (and input of your opamp) is capacitive. The feedback cap adds a zero to the amplifier which you can tune to be just in front of the resonance frequency you may observe. That way you can get a flatband response.

You may still want to try several feedback resistors even if you're able to get -3dB at your desirable frequency. Sometimes you can get more gain (in absolute terms) but with a little bit more roll off (relatively speaking) at the highest frequency you need.

Be careful with the layout, because typically at several kohm's feedback resistance and several MHz means you only need a few pF of feedback capacitance to make it stable again.

[max_torque]

All good stuff, thanks all   

My current photdiode system is reverse biased at 20V (max allowed) and uses a TIA with an LT1226 amplifier.  The issue is that it's hacked together on stripboard, and i think that is the limiting factor (stray C's and L's)!!

Because the current system is reflective, and has about 3m of Plastic Optical fibre between the trigger disc and the diodes it needs to have a fair amount of gain, which is slowing thing down, and with the LED drive on the same bit of strip board, those high currents from the pulsing are causing havoc over on the sensitive photodiode amplifier) side..

With a continuous light source (led on all the time) there should be a lot less noise around, and with a interrupter type arrangement, i can get the LED and PD adjacent to each other (perhaps 3mm apart maybe), so get a much larger optical signal, meaning less gain required.

As mentioned, the waveform coming back from the PD is going to have a variable dc offset, and have short pulses with relatively fast edges when the trigger wheel teeth pass the sensor.  Ideally i'd want to have a circuit that works out the "no tooth" and "tooth" voltage levels, and perhaps sets a comparitor to one shot trigger at say 50% of that voltage, to try to get a repeatable, aligned edge?  I'd need to look into analogue peak hold circuits perhaps?  Depending on how the disc is arranged (ie is a tooth a "hole" or is a tooth a "not a hole") and depending on the average duty cycle of the disc (50:50 mark space ratio) perhaps i can recover an "average" voltage by just using a low pass filter, and then setting some hysterisis on either side of that value for the comparitor??

[voltsandjolts]

Ideally i'd want to have a circuit that works out the "no tooth" and "tooth" voltage levels, and perhaps sets a comparitor to one shot trigger at say 50% of that voltage, to try to get a repeatable, aligned edge?

Sounds like a data slicer.

[max_torque]

ok, looks like the first job is to knock up a decent pulsed IR light source that can be used to determine the response of a detector circuit.  Be nice to get 50ns to 500ns pulses, adjustable, at perhaps a 1kHz rate, with the sharpest photon increase rate ?

Then, once i have that, i can develop a "constant" high intensity IR source, and a low latency / low jitter detector for use on the actual sensor. 

If i measure current into the driving led, and correct for the capacitance of that part, does current = photons?  (datasheet gives a quantum yield of 93%) ie i know the current and the voltage, so i can calc how many electrons go into the devices capacitance and how many get converted into photons and so i can estimate the IR intensity vs time during the pulse??

[StillTrying]

ok, looks like the first job is to knock up a decent pulsed IR light source that can be used to determine the response of a detector circuit.  Be nice to get 50ns to 500ns pulses, adjustable, at perhaps a 1kHz rate, with the sharpest photon increase rate ?

I'm glad someone else is going to do some LED fast flashing experiments.

There's some info here.
https://www.eevblog.com/forum/beginners/floating-probe!-for-$2-50/msg1834925/#msg1834925
edit. I've given up tryng to fix the link.

For very narrow pulses I use something like this, 10R will probably be too low for an IR LED's 1.25V.
I don't think you don't need to worry about the LED's physical capacitance, it only adds up to a couple of ns difference on the edges.

[snarkysparky]

BF862 jfet amp with bipolar cascode load is pretty fast for photodiode detector.


http://www.ti.com/lit/an/snoa620/snoa620.pdf

perhaps a phase lock loop to smooth out jitter ?

[max_torque]

Ok, that's one 10 to 50 nanosecond high intensity pulsed IR source knocked up:



Looks like a right dogs dinner, but works a charm! 

It should allow me to build and test the detector stage and get an idea of the responsive and delays inherent in that side of the device.

One thing i've been wondering, is how to drive the signal data up the cable and into the counter unit.  It will probably end up a couple of feet away from the IR head.  I guess a fast buffer driving into a coax is the answer, but i wonder if a balanced (differential) twisted pair might be better?  Or perhaps a low impedance current loop?  The environment will be electrically noisy, and i need to get the nice sharp electrical signal out of the sensing head and up to the measurement unit without falling foul of noise or reflections etc.  Any suggestions anyone?

[StillTrying]

"10 to 50 nanosecond high intensity"

You'll be lucky! Does a mobile phone camera see anything of the IR pulses.

[max_torque]

nope, not even close, i can't find a CCD device that sees the output at all.

a fast photodiode with 30V backwards across it with a low gain transimpedance amp  repeatably shows 10ns pulses, which match the pulse time (with about 3ns of latency) of the trigger pulse.  There's a bit of ringing for around 5 to 10 ns on the off egde, because i am using a inductor to provide a quick reverse bias to the LED when the fet switches off

[StillTrying]

10ns pulses and 3ns latency will just be noise, real light pulses will be 10X slower than that.