Compensating tempcos of a photodiode

[Jim Thompson]

Greetings,

I'm running a photodiode with an op amp in a photovoltaic mode. From the
datasheets I can find that the responsivity tempco is 0.35%/C and dark
current tempco is 2X/10C. I wonder how to best compensate for these
effects. Since I have an extra and so far unused ~identical photodiode,
would it be possible to use it (blinfolded) to cancel the temperature
effects of the other one by connecting it to inverting terminal to
cancel the effects of the other one?

Jack Middleton

Responsivity is **gain**, so you'd need to compensate for that using
(perhaps) a thermistor. You could probably rig the extra diode to
compensate for the dark current, but I'd have to see your setup before
I could say if the inverting terminal is the place for it.

What I usually do with photo-link situations is use a PWM signal
source and a DC loop to keep the output centered.

...Jim Thompson

 

[Jack Middleton]

Greetings,

I'm running a photodiode with an op amp in a photovoltaic mode. From the
datasheets I can find that the responsivity tempco is 0.35%/C and dark
current tempco is 2X/10C. I wonder how to best compensate for these
effects. Since I have an extra and so far unused ~identical photodiode,
would it be possible to use it (blinfolded) to cancel the temperature
effects of the other one by connecting it to inverting terminal to
cancel the effects of the other one?

Jack Middleton

 

[Phil Hobbs]

Responsivity is **gain**, so you'd need to compensate for that using
(perhaps) a thermistor. You could probably rig the extra diode to
compensate for the dark current, but I'd have to see your setup before
I could say if the inverting terminal is the place for it.

What I usually do with photo-link situations is use a PWM signal
source and a DC loop to keep the output centered.

The other thing is that the actual tempco of responsivity is strongly
dependent on wavelength and bias voltage, because of at least three things:

1. (for narrow line illumination, e.g. lasers) etalon fringes in the
photodiode window, which cause sensitivity variations of a few percent,
varying sinusoidally with temperature. The rapid drift of these fringes can
cause responsivity slopes of ~1% per kelvin--10,000 ppm/K--which makes
absolute nonsense of a precision laser measurement if it isn't cured.

2. Absorption depth of the light in the silicon. Carriers generated below
the depletion region give rise to much less actual photocurrent, because they
mostly recombine instead. This is strongly dependent on reverse bias,
because the depletion depth can change by a factor of ~10 with bias.

3. (for high light levels and large diodes) Resistivity variation in the
silicon changing the lateral voltage drops in the epi layer, causing
different parts of the diode to run at different bias levels.

How much these things matter depends a lot on what you're trying to do with
the photodiode. What is it?

Cheers,

Phil Hobbs

 

[Joerg]

Hi Jack,

If things get too complicated there is still the "not so elegant but
efficient" method: Measure temp with whatever device is handy, then feed
that info into a uC. Dual slope conversion is often adequate and any uC
can do that.

Now the uC can hold several transfer functions that can be as
complicated as they need to be. It can adjust offset, gain and other
parameters according to temperature and a function for each. With an
MSP430 this scheme doesn't even require much in terms of power.

Regards, Joerg

 

[Jack Middleton]

The other thing is that the actual tempco of responsivity is strongly
dependent on wavelength and bias voltage, because of at least three things:

1. (for narrow line illumination, e.g. lasers) etalon fringes in the
photodiode window, which cause sensitivity variations of a few percent,
varying sinusoidally with temperature. The rapid drift of these fringes
can cause responsivity slopes of ~1% per kelvin--10,000 ppm/K--which
makes absolute nonsense of a precision laser measurement if it isn't cured.

2. Absorption depth of the light in the silicon. Carriers generated
below the depletion region give rise to much less actual photocurrent,
because they mostly recombine instead. This is strongly dependent on
reverse bias, because the depletion depth can change by a factor of ~10
with bias.

3. (for high light levels and large diodes) Resistivity variation in
the silicon changing the lateral voltage drops in the epi layer, causing
different parts of the diode to run at different bias levels.

How much these things matter depends a lot on what you're trying to do
with the photodiode. What is it?

I'm sorry for the delay in responding, I've been kept away from my desk
these couple of days, now it is looking brighter again.

It is just a regular general purpose light meter. I have an old folder
camera that needs some work. The first phase is to measure/fix shutter
speeds (mostly done) and aperture values. The next goal (after fixing
the camera) is to use it in photography to determine exposures. It is
connected to a microcontroller via 10-bit adc (0 - 5v). Once I get the
measurement data in, the rest is easily handled with software and
displayed with lcd.

Due to the limitations of adc I have about 9 bits to play with to
measure from darkness to sun. That should be enough for ordinary
photographic purposes (assuming logarithmic measurement).

Since measurement is mostly done in natural environment I assume that
the effects of tempcos in voltaic mode are significant.
They would be visible through adc within 2-3 degrees change of
temperature even without amplification.

I can either measure the temperature separately and compensate in
software (a lot more hardware) of try to eliminate the effects before
adc (hopefylly less hardware).

As you already may have surmised, I'm more at home with the digital part
of the operation than with the analog and need some help there.

Basically the operation looks like this (taken from earlier post, thanks
John!)

.----------------------------.
| |
.-------------. |
| | | |
.-|\ | |
| \-->|-----+--/\/\--+--/\/\--+
| / | |
.-|/| | ___
| | | -
| | \\ | |
.------>|--------------+--/\/\--+
| |
.---------------------------.

Reading what you both said about the natures of tempcos, it doesn't look
like its too easy to compensate for them, even with an almost identical
part. I did a search on some application notes and couldn't find any
standard solution. I wonder if that is the reason why the
photoconductive mode is advertized so much. But then I suffer with the
adc. One can never win (the second law of thermodynamics .

Jack

 

[Phil Hobbs]

I'm sorry for the delay in responding, I've been kept away from my desk
these couple of days, now it is looking brighter again.

It is just a regular general purpose light meter. I have an old folder
camera that needs some work. The first phase is to measure/fix shutter
speeds (mostly done) and aperture values. The next goal (after fixing
the camera) is to use it in photography to determine exposures. It is
connected to a microcontroller via 10-bit adc (0 - 5v). Once I get the
measurement data in, the rest is easily handled with software and
displayed with lcd.

Due to the limitations of adc I have about 9 bits to play with to
measure from darkness to sun. That should be enough for ordinary
photographic purposes (assuming logarithmic measurement).

Since measurement is mostly done in natural environment I assume that
the effects of tempcos in voltaic mode are significant.
They would be visible through adc within 2-3 degrees change of
temperature even without amplification.

I can either measure the temperature separately and compensate in
software (a lot more hardware) of try to eliminate the effects before
adc (hopefylly less hardware).

As you already may have surmised, I'm more at home with the digital part
of the operation than with the analog and need some help there.

Basically the operation looks like this (taken from earlier post, thanks
John!)

.----------------------------.
| |
.-------------. |
| | | |
.-|\ | |
| \-->|-----+--/\/\--+--/\/\--+
| / | |
.-|/| | ___
| | | -
| | \\ | |
.------>|--------------+--/\/\--+
| |
.---------------------------.

Reading what you both said about the natures of tempcos, it doesn't look
like its too easy to compensate for them, even with an almost identical
part. I did a search on some application notes and couldn't find any
standard solution. I wonder if that is the reason why the
photoconductive mode is advertized so much. But then I suffer with the
adc. One can never win (the second law of thermodynamics .

Jack

OHHHH. That's much less complicated. For a wide-range photometer, I'd
certainly do the photovoltaic thing, because you'll need the logarithmic
response. Film latitude is very wide--even a stickler doesn't worry about
quarter-stops, which is a 19% change in photocurrent. Even at 0.35%/K, you'd
need a 50-kelvin temperature swing to have anything to worry about. In
photovoltaic mode, you don't need to worry about dark current, so it's just
the shift in the forward voltage of the silicon that's going to matter. Save
the photodiode, digitize the Vf of a 1N914 run at constant current (say 10
uA), and use that for first-order temperature compensation. Just subtract
the two voltages (with the same gain, of course). If you do the subtraction
in analogue, it will also do your level shifting for you. The simple method
is like this:




Your PD will be reasonably accurately logarithmic over 4 or 5 decades of
light intensity, which is better performance than your film. Nine bits of
resolution over that range will give you 2% resolution, which is about 1/40
stop, much more accurate than the ISO rating of your film.

A more serious problem is that the PD's spectral sensitivity is not a good
match for film--bad enough in daylight and *atrocious* under tungsten light.
Cad sulphide is much better for this, but of course it has its own
problems, so a blue-filtered photodiode is the usual medicine.

Cheers,

Phil Hobbs

 

[Jack Middleton]

OHHHH. That's much less complicated. For a wide-range photometer, I'd
certainly do the photovoltaic thing, because you'll need the logarithmic
response. Film latitude is very wide--even a stickler doesn't worry
about quarter-stops, which is a 19% change in photocurrent. Even at
0.35%/K, you'd need a 50-kelvin temperature swing to have anything to
worry about. In photovoltaic mode, you don't need to worry about dark
current, so it's just the shift in the forward voltage of the silicon
that's going to matter. Save the photodiode, digitize the Vf of a 1N914
run at constant current (say 10 uA), and use that for first-order
temperature compensation. Just subtract the two voltages (with the same
gain, of course). If you do the subtraction in analogue, it will also
do your level shifting for you. The simple method is like this:




Your PD will be reasonably accurately logarithmic over 4 or 5 decades of
light intensity, which is better performance than your film. Nine bits
of resolution over that range will give you 2% resolution, which is
about 1/40 stop, much more accurate than the ISO rating of your film.

That would be a good thing. The idea is to create a meter that can be
used reasonably over the range of natural light in different contexts.
The first being photography, but the rest being mainly dependent on the
software in the microcontroller. Maintaining accuracy over changes in
temperature is my main concern. I will expect that 50-kelvin temperature
swing to happen quite easily. That 2% resolution might come handy in
some other context.

A more serious problem is that the PD's spectral sensitivity is not a
good match for film--bad enough in daylight and *atrocious* under
tungsten light. Cad sulphide is much better for this, but of course it
has its own problems, so a blue-filtered photodiode is the usual medicine.

I have an access to good source of filter material. I was thinking of
trying to optimize the spectral response to daylight since that camera
will be used mainly in natural light photography.

I've seen different diagrams for photovoltaic mode and wonder which is
the most appropriate in this situation. It is curious that none of them
has included temperature compensation - assuming that 25 degrees Celcius
is maintained universally. Anyone seen one in an application note?

Jack