Optical position sensing, minimizing jitter

[CC]

Hi:

I have to design an improved version of a rotating wheel optical
position sensor. There is a slit of about 1mm width at 44mm radius on
the wheel with tangential velocity of about 110m/s rotating at 400Hz.

The past version used a IRLED shining through a similar slit onto a
PIN3CD photodiode through a 350Hz wheel. The rest of the circuitry is
shown here:

http://web.newsguy.com/crcarl/images/shutter-pd.png

Is hasn't much bandwidth, but has a lot of gain (4Meg effective
transimpedance gain) which was needed to deal with the relatively low
light reaching the detector from the LED.

In the new detector, I plan to use a 665nm VCSEL to create a well
focussed beam. The main goal is to get a much faster optical edge off
the slit. So the optical risetime will be roughly equal to the beam
diameter / the velocity. For instance, with a 100um beam:

(100*10^-6 m) / (110 m/s) = 910ns

Question: If you wanted some 't' precision of position sensing accuracy
(jitter/noise less than 't' seconds) what would the risetime of the
optical edge have to be? Equal? Smaller? Larger Ok? Let's say we
want to shoot for 85ns sensing jitter. What risetime of the optical
signal is needed?

The exact requirement is presently still poorly specified in my
application with something like 85ns being a possible future extreme.
So for academic purposes let's shoot for now at a detector bandwidth of
at least 3x what is implied by the 910ns risetime, ie. about 1MHz bandwidth.

First of all, I plan to use only a transimpedance stage and no second
gain stage since I should have a reasonable fraction of 1.5mW of light
hitting the photodiode (say, 0.5mW). I will use a UDT BPW-34B 12pF (at
10V bias, sensitivity 0.45A/W @ 650nm) photodiode. Thus, I will get a
photocurrent of 225uA. A 22k transimpedance will give a 5V pulse. To
get a BW of 1MHz this is easily done with something like an OPA356.

The next question is, how best to deal with this photodiode signal to
get the minimum jitter from the comparator output?

There would appear to be two possible strategies:

1. Since the beam power and alignment will change very little once it's
all set up, tune the gain of the transimpedance stage or the optics to
give a pulse of peak amplitude 'A' and set a fixed comparator threshold
of A/2. Then the threshold should be roughly where the maximum rate of
change is occuring in the optical signal, leading to minimum jitter.

Is it detrimental to use too much gain and saturate the opamp, for the
purpose of getting a faster slew rate just before the opamp saturates,
then putting the comparator threshold nearer the saturation voltage?

2. Use some sort of adaptive trigger threshold circuit.

I'd like to avoid the complexity of the adaptive trigger. Since there
is little fluctuation in optical signal strength, option 1 seems acceptable.

3. Other?

Comments appreciated.


Thanks for input.


Good day!

 

[John Popelish]

Hi:

I have to design an improved version of a rotating wheel optical
position sensor. There is a slit of about 1mm width at 44mm radius on
the wheel with tangential velocity of about 110m/s rotating at 400Hz.

The past version used a IRLED shining through a similar slit onto a
PIN3CD photodiode through a 350Hz wheel. The rest of the circuitry is
shown here:

http://web.newsguy.com/crcarl/images/shutter-pd.png (snip)
3. Other?

Since you do not need anything like a linear measure of the light
level, but only a time that it passes through some fixed amplitude,
you might think about how you can use a gracefully saturating
amplifier in place of the opamp to magnify that part of the optical
signal. Opamps are inherently slow, to allow closed loop operation,
and you have no need of that. Think about the choices based on common
base amplifiers or emitter coupled logic or limiter amplifiers. These
all run open loop, are very fast, and handle overload without slowing
down, much.

 

[Phil Hobbs]

Hi:

I have to design an improved version of a rotating wheel optical
position sensor. There is a slit of about 1mm width at 44mm radius on
the wheel with tangential velocity of about 110m/s rotating at 400Hz.

The past version used a IRLED shining through a similar slit onto a
PIN3CD photodiode through a 350Hz wheel. The rest of the circuitry is
shown here:

http://web.newsguy.com/crcarl/images/shutter-pd.png

Is hasn't much bandwidth, but has a lot of gain (4Meg effective
transimpedance gain) which was needed to deal with the relatively low
light reaching the detector from the LED.

In the new detector, I plan to use a 665nm VCSEL to create a well
focussed beam. The main goal is to get a much faster optical edge off
the slit. So the optical risetime will be roughly equal to the beam
diameter / the velocity. For instance, with a 100um beam:

<snip>

To estimate your position jitter, you really have to take account of
laser speckle and mode partition noise. VCSELs are typically highly
multimode devices, with as many as 12 transverse modes running
simultaneously. (They're usually single longitudinal mode due to their
very short cavities.) Multimode devices produce focused spots that
dance around at megahertz rates, which will be a significant source of
jitter and 1/f noise.

To do this really right, you need a single transverse mode device such
as an index-guided, cleaved-cavity laser diode. Single longitudinal
mode will help too, but is less important than a single spatial mode,
since the total intensity noise is typically *dramatically smaller* than
the noise of any individual mode. This is because the bias current sets
the total pumping rate, so the sum of the mode intensities tends to stay
pretty constant--the power just wanders around from mode to mode.

Cheers,

Phil Hobbs

 

[CC]

<snip>

To estimate your position jitter, you really have to take account of
laser speckle and mode partition noise. VCSELs are typically highly
multimode devices, with as many as 12 transverse modes running
simultaneously. (They're usually single longitudinal mode due to their
very short cavities.) Multimode devices produce focused spots that
dance around at megahertz rates, which will be a significant source of
jitter and 1/f noise.

To do this really right, you need a single transverse mode device such
as an index-guided, cleaved-cavity laser diode. Single longitudinal
mode will help too, but is less important than a single spatial mode,
since the total intensity noise is typically *dramatically smaller* than
the noise of any individual mode. This is because the bias current sets
the total pumping rate, so the sum of the mode intensities tends to stay
pretty constant--the power just wanders around from mode to mode.

Cheers,

Phil Hobbs

Right. But if the beam is spatially averaged over the detector, then
what matters is the total of power distributed throughout the modes,
which will be constant to within the noise specification of the emitter.

Where the noise would begin to appear in a signal is if you sample a
spatial fraction of the beam approaching the spatial resolution of the
transverse mode interference pattern. Then you can have meaningful
portions of the total beam power sometimes present in your sampling
area, and sometimes not present, leading to substantial noise.

Now the question is whether the slit interrupting the beam qualifies as
a sampling of a spatial fraction that will lead to such observed noise.
It might, when only a small fraction of the beam is passing. In fact
then, we might surmise that the noise due to sampling a fraction of the
total beam area will be in inverse proportion to the area sampled.

But for the amplitude midpoint and above regions of the signal pulse, I
expect that since this is sampling => 50% of the beam area, that the
noise should be very minor.

It would be interesting to take a look for this noise though. The
actual VCSELs I am interested in are Firecomms 665nm visible devices.
They will be quoting me for some samples (should have gotten the quote
yesterday...) And they seemed to suggest that single TEM mode devices
are available. We will see. I will likely obtain some OPTEK 850nm
devices as well, but don't hope to deploy them since aligning an
invisible beam is obviously more difficult.

Thanks for input.


Good day!

 

[Joerg]

Hello John,

Since you do not need anything like a linear measure of the light level,
but only a time that it passes through some fixed amplitude, you might
think about how you can use a gracefully saturating amplifier in place
of the opamp to magnify that part of the optical signal. Opamps are
inherently slow, to allow closed loop operation, and you have no need of
that. Think about the choices based on common base amplifiers or
emitter coupled logic or limiter amplifiers. These all run open loop,
are very fast, and handle overload without slowing down, much.

Or use a regular common source architecture and employ a few of those
hot TV tuner transistors. Those can do edges in the sub-nsec range.

 

[Phil Hobbs]

Where the noise would begin to appear in a signal is if you sample a
spatial fraction of the beam approaching the spatial resolution of the
transverse mode interference pattern. Then you can have meaningful
portions of the total beam power sometimes present in your sampling
area, and sometimes not present, leading to substantial noise.

Now the question is whether the slit interrupting the beam qualifies as
a sampling of a spatial fraction that will lead to such observed noise.
It might, when only a small fraction of the beam is passing. In fact
then, we might surmise that the noise due to sampling a fraction of the
total beam area will be in inverse proportion to the area sampled.

But for the amplitude midpoint and above regions of the signal pulse, I
expect that since this is sampling => 50% of the beam area, that the
noise should be very minor.

It isn't important only for large obscurations, believe me.

As a data point, I once had a *single longitudinal mode* argon laser
(also single transverse mode because of the much larger frequency
spacing) whose noise went up by *20 dB* when half the beam was obscured
by a knife edge. The side modes were still noisy even though they
weren't quite oscillating.

Cheers,

Phil Hobbs