Photodiode Amplifier Noise Effects

[WhiteDog]

I have been playing with my shiny new photodiode amplifier (BW=75MHz
A=35k) and my trusty spectrum analyzer set to 'noise' mode. I'm
dissecting the thing to try and find areas where some simple tweaks
might improve the SNR.

The photodiode is a 500um PIN device with a NEP of 0.008pW/RtHz and a
dark current under -5V bias (on the anode) of about 1.25nA.

The amplifier circuit under examination takes it's input from the
photodiode cathode through a common-base amplifier (BFG25A/X) and from
there into a low noise op-amp transimpedance amp. The common-base amp
is biased with 12k resistors to +/-5V to get the required bandwidth
from the circuit. Basically the circuit from Phil Hobbs' book in the
'Photodiode Amplifiers' section.

Could someone please explain the following effect?:

1. When the photodiode is removed (i.e. just a biased cascode into the
TIA) , the output noise is 80nV/RtHz at 50MHz. This matches back-of-
the envelope calculations of what this should be.
2. When the photodiode is connected, I get 160nV/RtHz!! Note that I
have covered the photodiode window with copper tape to remove the
possibility of DC bias, cut down the device leads, etc -- I think all
I should be getting at this point is dark current, correct?

This additional 80nV/RtHz seems excessive -- though the system still
meets spec, I would like some 'margin' in there and wouldn't mind
reducing the noise if possible.

The only source of 'noise' current that I could think of would be from
the bias supply, however, I am using a low-noise LT1964 LDO for this,
and am supplying the bias through two 10k resistors with a 1uF and a
1000pF ceramic cap arranged in a 'pi' configuration. This -ve supply
IS shared with the -ve rail of the TIA.

Before I tear into the bias of the PIN (building a second -ve bias
supply from an LDO), is there any other source I have missed or could
this really be all from the bias voltage?

Thanks for any help you can offer

 

[[email protected]]

I have been playing with my shiny new photodiode amplifier (BW=75MHz
A=35k) and my trusty spectrum analyzer set to 'noise' mode. I'm
dissecting the thing to try and find areas where some simple tweaks
might improve the SNR.

The photodiode is a 500um PIN device with a NEP of 0.008pW/RtHz and a
dark current under -5V bias (on the anode) of about 1.25nA.

The amplifier circuit under examination takes it's input from the
photodiode cathode through a common-base amplifier (BFG25A/X) and from
there into a low noise op-amp transimpedance amp. The common-base amp
is biased with 12k resistors to +/-5V to get the required bandwidth
from the circuit. Basically the circuit from Phil Hobbs' book in the
'Photodiode Amplifiers' section.

Could someone please explain the following effect?:

1. When the photodiode is removed (i.e. just a biased cascode into the
TIA) , the output noise is 80nV/RtHz at 50MHz. This matches back-of-
the envelope calculations of what this should be.
2. When the photodiode is connected, I get 160nV/RtHz!! Note that I
have covered the photodiode window with copper tape to remove the
possibility of DC bias, cut down the device leads, etc -- I think all
I should be getting at this point is dark current, correct?

This additional 80nV/RtHz seems excessive -- though the system still
meets spec, I would like some 'margin' in there and wouldn't mind
reducing the noise if possible.

The only source of 'noise' current that I could think of would be from
the bias supply, however, I am using a low-noise LT1964 LDO for this,
and am supplying the bias through two 10k resistors with a 1uF and a
1000pF ceramic cap arranged in a 'pi' configuration. This -ve supply
IS shared with the -ve rail of the TIA.

Before I tear into the bias of the PIN (building a second -ve bias
supply from an LDO), is there any other source I have missed or could
this really be all from the bias voltage?

Thanks for any help you can offer

I don't think a bit of noise on the photodiode bais voltage is much of
a problem. When you remove the photodiode try putting in a
capacitor whose value is equal to the capacitance of the biased
photodiode and meause the noise again.

George Herold

 

[WhiteDog]

Thanks George,

When you remove the photodiode try putting in a
capacitor whose value is equal to the capacitance of the biased
photodiode and meause the noise again.

Interesting.

Subbing in a 10pF NPO cap (the max capacitance of the PD at -5V as per
the datasheet) yields a somewhat reduced level of about 140nV/RtHz.

That seems to indicate to me that some of the objectionable signal is,
in fact, coming from the bias supply, and some is from some other
unanticipated effect of connecting the InGaAs photodiode assembly to
the cascode amplifier, correct?

 

[GregS]

Thanks George,


Interesting.

Subbing in a 10pF NPO cap (the max capacitance of the PD at -5V as per
the datasheet) yields a somewhat reduced level of about 140nV/RtHz.

That seems to indicate to me that some of the objectionable signal is,
in fact, coming from the bias supply, and some is from some other
unanticipated effect of connecting the InGaAs photodiode assembly to
the cascode amplifier, correct?

Why don't you remove supply and add ground, and see what happens.

greg

 

[GregS]

Thanks George,


Interesting.

Subbing in a 10pF NPO cap (the max capacitance of the PD at -5V as per
the datasheet) yields a somewhat reduced level of about 140nV/RtHz.

That seems to indicate to me that some of the objectionable signal is,
in fact, coming from the bias supply, and some is from some other
unanticipated effect of connecting the InGaAs photodiode assembly to
the cascode amplifier, correct?

This is very interesting to me. Please post results.

greg

 

[[email protected]]

Thanks George,


Interesting.

Subbing in a 10pF NPO cap (the max capacitance of the PD at -5V as per
the datasheet) yields a somewhat reduced level of about 140nV/RtHz.

That seems to indicate to me that some of the objectionable signal is,
in fact, coming from the bias supply, and some is from some other
unanticipated effect of connecting the InGaAs photodiode assembly to
the cascode amplifier, correct?

I suggest you read and go through the noise analysis of the photodiode
TIA circuit. There is 'noise gain' in the circuit that is due to the
photodiode capacitance.
The idea behind Phil H's cascode input is to reduce this effect. I've
not done any noise analysis of the TIA circuit with a cascode input.
(I've dreamed of building this circuit, but as of yet have not had the
free time or application that needed it.) You might estimate the
impedance of the 5pF at 75 MHz and compare this to the impedance of
the cascode. (Are you running any DC bias current through the
cascode?) I think you can drive the cascode harder and reduce it's
impedance.

I'm not sure why you are focused on the bias supply. The purpose of
the bias supply is to reduce the capacitance of the photodiode. A
little noise on the bias voltage will mean a bit of fluctuation in the
photodioe capacitance, but this should hardly effect the noise at
all. (Try using a noisier voltage bias and see if this increases your
output noise.)

Did you build this photodiode preamp yourself?


George Herold

 

[WhiteDog]

George,

Grounding the bias end of the 'fake' photodiode (the 10pF cap) yields
a similar noise figure (138nV/rtHz). Essentially this is AC noise

Yes, Phil's circuit is meant tor educe the noise gain effect of Cd by
isolating it from the TIA via the cascode amplifier. The Cd is then
replaced by the (hopefully) smaller Cs of the cascode amp's
collector.

This was the point of removing the photodiode from it (there's not
much I can do about the Cs of the cascode, outside of replacing the
transistor). By simply removing the photodiode what I have left is the
cascode (biased on) plus the TIA, amplifying it's own noise plus the
input from the cascode.

Without the photodiode gives me:
TIA noise + cascode noise

Installing the photodiode should give me:
TIA noise + cascode noise + (photodiode NEP * responsivity) + Bias
supply noise

The NEP on the photodiode yields sub nV/RtHz after the TIA, so all I
have left is the bias supply
I wanted to see if there was anything I could do on the very front
end to quiet things down a little in the 30-60MHz region.

impedance of the 5pF at 75 MHz and compare this to the impedance of
the cascode. (Are you running any DC bias current through the
cascode?) I think you can drive the cascode harder and reduce it's
impedance.

You can bias the cascode and reduce the re of the emitter (which I
have done). The amp is biased to 417uA, which seems like a lot, but I
needed that to get the response flatness over the BW I required. The
effects of this should be included as described above.

I'm not sure why you are focused on the bias supply. The purpose of
the bias supply is to reduce the capacitance of the photodiode. A
little noise on the bias voltage will mean a bit of fluctuation in the
photodioe capacitance, but this should hardly effect the noise at
all.

I am focused on the bias supply because that seems to be all that is
left (as far as I know) for a noise source. Noise on the photodiode
bias will be divided across the impedance of the bias voltage filter,
the photodiode (362 ohms), and the cascode emitter (.

My astrophysicist buddy suggested thermocouple effects as well (I'll
try that next with a hunk of wire and some cold spray)

Did you build this photodiode preamp yourself

Schematic plus layout, yes. Fab + assembly done by board house.

 

[[email protected]]

George,

Grounding the bias end of the 'fake' photodiode (the 10pF cap) yields
a similar noise figure (138nV/rtHz). Essentially this is AC noise

Yes, Phil's circuit is meant tor educe the noise gain effect of Cd by
isolating it from the TIA via the cascode amplifier. The Cd is then
replaced by the (hopefully) smaller Cs of the cascode amp's
collector.

This was the point of removing the photodiode from it (there's not
much I can do about the Cs of the cascode, outside of replacing the
transistor). By simply removing the photodiode what I have left is the
cascode (biased on) plus the TIA, amplifying it's own noise plus the
input from the cascode.

Without the photodiode gives me:
    TIA noise + cascode noise

Installing the photodiode should give me:
    TIA noise + cascode noise + (photodiode NEP * responsivity) + Bias
supply noise

The NEP on the photodiode yields sub nV/RtHz after the TIA, so all I
have left is the bias supply
 I wanted to see if there was anything I could do on the very front
end to quiet things down a little in the 30-60MHz region.


You can bias the cascode and reduce the re of the emitter (which I
have done). The amp is biased to 417uA, which seems like a lot,  but I
needed that to get the response flatness over the BW I required. The
effects of this should be included as described above.


I am focused on the bias supply because that seems to be all that is
left (as far as I know) for a noise source. Noise on the photodiode
bias will be divided across the impedance of the bias voltage filter,
the photodiode (362 ohms), and the cascode emitter (.

My astrophysicist buddy suggested thermocouple effects as well (I'll
try that next with a hunk of wire and some cold spray)


Schematic plus layout, yes. Fab + assembly done by board house.

Whitedog,
My copy of Phil's book is at home, but I'll look later and see if he
does any noise analysis in it. (Better yet he may log on and and you
can hear it right from the horses mouth. My guess (and this is really
a guess mind you) is that you are seeing the TIA op-amp noise coupled
to the output when you put either the photodiode (PD) or equivalent
capacitor in the circuit. You would certainly see the same type of
thing if you removed the PD from a 'traditional' TIA without the
cascode front end. (Less noise without the PD and more with PD or
equivalent capacitor.)

You should be able to 'filter the piss' out of the bias supply. A
simple RC should be more than adequate. If you are really worried
stick in the 'capacitor multiplier' circuit also in Phil's book. I've
used this circuit to reduce 15 Volt power supply noise to the 1 nV/
rtHz level.

What opamp are you using in the TIA? What level of photocurrent are
you trying to sense? Can you increase the bias voltage on the PD?


"Schematic plus layout, yes. Fab + assembly done by board house."

Ahh I'm jealous... Well someday I hope to build a fast photodiode
front end of my own.

George Herold

(I'd forget the thermocouple effects. This can give you some DC
offsets, but I've never heard of it affecting the noise at 30-60 MHz.)

 

[[email protected]]

Whitedog,
My copy of Phil's book is at home, but I'll look later and see if he
does any noise analysis in it.  (Better yet he may log on and and you
can hear it right from the horses mouth.  My guess (and this is really
a guess mind you) is that you are seeing the TIA op-amp noise coupled
to the output when you put either the photodiode (PD) or equivalent
capacitor in the circuit.  You would certainly see the same type of
thing if you removed the PD from a 'traditional' TIA without the
cascode front end.  (Less noise without the PD and more with PD or
equivalent capacitor.)

You should be able to 'filter the piss' out of the bias supply.  A
simple RC should be more than adequate.  If you are really worried
stick in the 'capacitor multiplier' circuit also in Phil's book.  I've
used this circuit to reduce 15 Volt power supply noise to the 1 nV/
rtHz level.

What opamp are you using in the TIA?  What level of photocurrent are
you trying to sense?  Can you increase the bias voltage on the PD?

"Schematic plus layout, yes. Fab + assembly done by board house."

Ahh I'm jealous... Well someday I hope to build a fast photodiode
front end of my own.

George Herold

(I'd forget the thermocouple effects.  This can give you some DC
offsets, but I've never heard of it affecting the noise at 30-60 MHz.)- Hide quoted text -

- Show quoted text -

Whitedog,
I was thinking of this on the ride home. Your numbers suggest that
you're using an opamp with a BJT front end. (The 1-2 pA/rtHz of
current noise through the 35 k ohm resistor is your 80 nV/rtHz. of
'preamp' noise.) You really need to stick an FET opamp your
circuit.

If your supplies are only +/- 5 Volts there is a "fast FET" from
analog you might try. (If you bias every thing right you might be
able to run these from 0 to +10 V)

George

 

[JosephKK]

I have been playing with my shiny new photodiode amplifier (BW=75MHz
A=35k) and my trusty spectrum analyzer set to 'noise' mode. I'm
dissecting the thing to try and find areas where some simple tweaks
might improve the SNR.

The photodiode is a 500um PIN device with a NEP of 0.008pW/RtHz and a
dark current under -5V bias (on the anode) of about 1.25nA.

The amplifier circuit under examination takes it's input from the
photodiode cathode through a common-base amplifier (BFG25A/X) and from
there into a low noise op-amp transimpedance amp. The common-base amp
is biased with 12k resistors to +/-5V to get the required bandwidth
from the circuit. Basically the circuit from Phil Hobbs' book in the
'Photodiode Amplifiers' section.

Could someone please explain the following effect?:

1. When the photodiode is removed (i.e. just a biased cascode into the
TIA) , the output noise is 80nV/RtHz at 50MHz. This matches back-of-
the envelope calculations of what this should be.
2. When the photodiode is connected, I get 160nV/RtHz!! Note that I
have covered the photodiode window with copper tape to remove the
possibility of DC bias, cut down the device leads, etc -- I think all
I should be getting at this point is dark current, correct?

This additional 80nV/RtHz seems excessive -- though the system still
meets spec, I would like some 'margin' in there and wouldn't mind
reducing the noise if possible.

The only source of 'noise' current that I could think of would be from
the bias supply, however, I am using a low-noise LT1964 LDO for this,
and am supplying the bias through two 10k resistors with a 1uF and a
1000pF ceramic cap arranged in a 'pi' configuration. This -ve supply
IS shared with the -ve rail of the TIA.

Oops! Do not share these.

Before I tear into the bias of the PIN (building a second -ve bias
supply from an LDO), is there any other source I have missed or could
this really be all from the bias voltage?

Thanks for any help you can offer

I read the thread. Separate the photodiode bias supply from the TIA
supply. "Yellow wire" it if you have to.