Ideas on causes of excess noise in transimpedance pre-amp?

[Bret Cannon]

I'm trying to understand the noise I measure in a couple of versions of a
trans-impedance pre-amp that I'm using with a quartz tuning fork, such as is
used in digital watches, but removed from its can and used to measure
acoustic signals from a laser beam focused between the tines of the tuning
fork. The output noise spectra do not match what I calculate using the
Johnson noise of the feedback resistor, the specified voltage and current
noise of the op-amps, the measured feedback capacitance and my estimate of
the capacitance at the input of the op-amp.

The initial version of the trans-impedance amp used a Burr-Brown OPA656 with
a 10 Mohm feedback resistor on a 2-sided board. I measure an output noise
of 410 nV/rt-Hz at frequencies below 1 kHz, which matches the Johnson noise
of the 10 Mohm feedback resistor, but the noise floor starts increasing
above 1 kHz. This increasing output noise floor is qualitatively similar to
that due to the voltage noise of the op-amp amplified by the noise gain.
However the numbers don't workout. The specified input capacitance of the
OPA656 is 4.5 pF common mode, 0.7 pF differential. There are the
capacitances of the 1/2" long traces from the op-amp inputs to an SMA
connector soldered to the board, the SMA connector, an SMA to BNC adapter,
and a bulkhead BNC connector to which the fork is soldered. I estimate
these connector capacitances total about 3 pF. Adding in the shunt
capacitance of the quartz tuning fork, about 1 pF, the I estimate the total
input capacitance to be 9.2 pF. I measure a 3 dB bandwidth of 106 kHz from
which I conclude the stray capacitance of surface mount feedback resistor is
0.15 pF. With these capacitances and the 10M feedback resistor, I calculate
that the output noise should not rise above the Johnson noise of the
feedback resistor by 5% until about 100 kHz.

However I measure the noise reaching 500 nV/rt-Hz at 30 kHz. (With a quartz
tuning fork connected that is still in its can, there is a noise peak at the
fork resonance of 32764 Hz, but this noise peak has a width of 0.4 Hz and
doesn't not contribute significantly to the noise spectrum except within
about 20 Hz of the resonance frequency.)

This pre-amp has another problem, the trans-impedance gain varies with the
equivalent series resistance of the tuning fork's equivalent series RLC
circuit. The fork series resistance changes with pressure and temperature,
so the system gain changes with gas pressure. The change in trans-impedance
gain with source resistance is well fit by a model using a finite open-loop
gain of 62 dB for the OPA656 which is in fair agreement with the "typical"
open loop gain of 65 dB up to about 100 kHz.

So to reduce the variation in trans-impedance gain with source resistance, I
decided to replace the OPA656 with the OPA657. The OPA657 has a higher
typical open loop gain (75 dB) and a larger gain bandwidth product, but the
same specified input capacitance and resistance as the OPA656, the same
current noise (1.3 fA/rt-Hz), and slightly lower voltage noise (5 nV/rt-Hz
for the OPA657 and 7 nV/rt-Hz for the OPA 656). To my surprise and
disappointment, with the OPA 657, the noise increased at 30 kHz from 500
nV/rt-Hz to 600 nV/rt-Hz and peaks at 50 kHz. With the OPA656, the noise
kept increasing to 100 kHz, the upper limit of my spectrum analyzer.

To further confuse me, another version of this pre-amp was made with a
4-layer board that has provision for doing active bandpass filtering. We
found that the trans-impedance input stage with an OPA657 has twice the
noise at 30 kHz as the previous board with an OPA657. This higher noise
level is present when the components for filtering are removed. Even when a
quartz tuning fork is soldered to the 4-layer circuit board, the noise level
near 30 kHz is still about 3 times the Johnson noise level of the feedback
resistor.

I have measured the noise spectrum at the power pins for the OPA657 and it
is flat at 200 nV/rt-Hz from about 128 Hz until it starts rolling off at
about 40 kHz. I have also looked at the output of these pre-amps with a 400
MHz bandwidth scope and can not find evidence of oscillation. I found a
problem with current leakage on one of these boards that was fixed by
scrubbing with isopropyl alcohol and cotton swabs, so we've periodically
cleaned these boards without any reduction in noise since that first time.

The one other obvious difference in the two circuit boards is that on the
4-layer board, the ground plane extends under the op-amp and perhaps under
the traces from the op-amp inverting input to the fork connections, while
the original 2-layer board does not have a ground plane under the op-amp and
the inverting input.

My basic questions are:
1) What could be the cause of the excess noise that became significantly
worse when switching to a faster and higher gain op-amp that has less
voltage noise and that depends on the layout of the circuit board?

2) What things should be done in the layout of the next circuit board to
minimize the noise near 32 kHz for a trans-impedance pre-amp with gains of 1
to 10 Mohms?

Thanks for any suggestions,
Bret Cannon

 

[[email protected]]

I'm trying to understand the noise I measure in a couple of versions of a
trans-impedance pre-amp that I'm using with a quartz tuning fork, such as is
used in digital watches, but removed from its can and used to measure
acoustic signals from a laser beam focused between the tines of the tuning
fork. The output noise spectra do not match what I calculate using the
Johnson noise of the feedback resistor, the specified voltage and current
noise of the op-amps, the measured feedback capacitance and my estimate of
the capacitance at the input of the op-amp.

The initial version of the trans-impedance amp used a Burr-Brown OPA656 with
a 10 Mohm feedback resistor on a 2-sided board. I measure an output noise
of 410 nV/rt-Hz at frequencies below 1 kHz, which matches the Johnson noise
of the 10 Mohm feedback resistor, but the noise floor starts increasing
above 1 kHz.

The corner at 1kHz implies a capacitance to ground of 16pF - a bit more
than your estimated 9.2pF.Printed circuit tracks are microstrip
transmission lines, and the capacitance per unit length depends on the
width. 50R tracks are fairly wide and run at 3.8pF/inch (150pF/metre).
75R tracks look like regular trances and run at 1.7pF/inch.

Philips use 2.2pF/inch in one of their application notes on using high
speed logic.

The extra 6pF suggests that you have neglected to allow for some three
inches of track over ground somewhere in your system.

This increasing output noise floor is qualitatively similar to
that due to the voltage noise of the op-amp amplified by the noise gain.
However the numbers don't workout. The specified input capacitance of the
OPA656 is 4.5 pF common mode, 0.7 pF differential. There are the
capacitances of the 1/2" long traces from the op-amp inputs to an SMA
connector soldered to the board, the SMA connector, an SMA to BNC adapter,
and a bulkhead BNC connector to which the fork is soldered. I estimate
these connector capacitances total about 3 pF. Adding in the shunt
capacitance of the quartz tuning fork, about 1 pF, the I estimate the total
input capacitance to be 9.2 pF. I measure a 3 dB bandwidth of 106 kHz from
which I conclude the stray capacitance of surface mount feedback resistor is
0.15 pF. With these capacitances and the 10M feedback resistor, I calculate
that the output noise should not rise above the Johnson noise of the
feedback resistor by 5% until about 100 kHz.

That sounds wrong. If the gain corner is at 1kHz, the noise gain at
30kHz is 30, so the 7nV/root Hz input noise of the op amp will look
like 210nV/root Hz at the output at 30kHz, 700nV/Hz at 100kHz

However I measure the noise reaching 500 nV/rt-Hz at 30 kHz. (With a quartz
tuning fork connected that is still in its can, there is a noise peak at the
fork resonance of 32764 Hz, but this noise peak has a width of 0.4 Hz and
doesn't not contribute significantly to the noise spectrum except within
about 20 Hz of the resonance frequency.)

The root means square sum of 210nV/root Hz and 400nV/root Hz is
452nV/root Hz.

The noise spectrum of the amplified inut noise won't be white, because
the high frequency components are amplified more than the low, so rms
summation may be over-optimistic.

This pre-amp has another problem, the trans-impedance gain varies with the
equivalent series resistance of the tuning fork's equivalent series RLC
circuit. The fork series resistance changes with pressure and temperature,
so the system gain changes with gas pressure. The change in trans-impedance
gain with source resistance is well fit by a model using a finite open-loop
gain of 62 dB for the OPA656 which is in fair agreement with the "typical"
open loop gain of 65 dB up to about 100 kHz.

So to reduce the variation in trans-impedance gain with source resistance, I
decided to replace the OPA656 with the OPA657. The OPA657 has a higher
typical open loop gain (75 dB) and a larger gain bandwidth product, but the
same specified input capacitance and resistance as the OPA656, the same
current noise (1.3 fA/rt-Hz), and slightly lower voltage noise (5 nV/rt-Hz
for the OPA657 and 7 nV/rt-Hz for the OPA 656). To my surprise and
disappointment, with the OPA 657, the noise increased at 30 kHz from 500
nV/rt-Hz to 600 nV/rt-Hz and peaks at 50 kHz. With the OPA656, the noise
kept increasing to 100 kHz, the upper limit of my spectrum analyzer.

To further confuse me, another version of this pre-amp was made with a
4-layer board that has provision for doing active bandpass filtering. We
found that the trans-impedance input stage with an OPA657 has twice the
noise at 30 kHz as the previous board with an OPA657. This higher noise
level is present when the components for filtering are removed. Even when a
quartz tuning fork is soldered to the 4-layer circuit board, the noise level
near 30 kHz is still about 3 times the Johnson noise level of the feedback
resistor.

I have measured the noise spectrum at the power pins for the OPA657 and it
is flat at 200 nV/rt-Hz from about 128 Hz until it starts rolling off at
about 40 kHz. I have also looked at the output of these pre-amps with a 400
MHz bandwidth scope and can not find evidence of oscillation. I found a
problem with current leakage on one of these boards that was fixed by
scrubbing with isopropyl alcohol and cotton swabs, so we've periodically
cleaned these boards without any reduction in noise since that first time.

The one other obvious difference in the two circuit boards is that on the
4-layer board, the ground plane extends under the op-amp and perhaps under
the traces from the op-amp inverting input to the fork connections, while
the original 2-layer board does not have a ground plane under the op-amp and
the inverting input.

My basic questions are:
1) What could be the cause of the excess noise that became significantly
worse when switching to a faster and higher gain op-amp that has less
voltage noise and that depends on the layout of the circuit board?

The stray capacitances on the printed circuit board seem to be higher
than you have estimated. Have you tried measuring the stray capacitance
on an unpopulated printed circuit board?

2) What things should be done in the layout of the next circuit board to
minimize the noise near 32 kHz for a trans-impedance pre-amp with gains of 1
to 10 Mohms?

Minimise the stray capacitance - shorter tracks, narrower tracks,
remove ground plane under the crucial tracks, as recommended in the
Burr-Brown application notes. Using a two-layer layout on the outer
layers of a four layer board reduces the distance to the ground plane
by a factor of three, and roughly triples the capacitance of any
microstrip tracks, even before you add in any extra ground plane.

Make sure that you have both the recommended 100nf decoupling capacitor
to ground close to the op amp's power pins, and a couple of uF of
tantalum capacitorless than an inch (25mm) away, as recommended by the
application notes. I tend to throw in a ferrite chip/bead between the
capacitors and the power rail - they are fairly cheap, don't introduce
any DC drop, and - being lossy non-wound devices - look resistive
rather than capacitative up to very high frequencies. The of the order
of 1uH of inductance can resonate with the decoupling capacitor, but as
long as you have a few uF of tantalum capacitor present, the resonance
is around 100kHz and the impedance at resonance low enough - at around
0R6 - that the ESR of the tantalum bead will kill any peaking.

 

[Bret Cannon]

The corner at 1kHz implies a capacitance to ground of 16pF - a bit more
than your estimated 9.2pF.Printed circuit tracks are microstrip
transmission lines, and the capacitance per unit length depends on the
width. 50R tracks are fairly wide and run at 3.8pF/inch (150pF/metre).
75R tracks look like regular trances and run at 1.7pF/inch.

Philips use 2.2pF/inch in one of their application notes on using high
speed logic.

The extra 6pF suggests that you have neglected to allow for some three
inches of track over ground somewhere in your system.


That sounds wrong. If the gain corner is at 1kHz, the noise gain at
30kHz is 30, so the 7nV/root Hz input noise of the op amp will look
like 210nV/root Hz at the output at 30kHz, 700nV/Hz at 100kHz


The root means square sum of 210nV/root Hz and 400nV/root Hz is
452nV/root Hz.

The noise spectrum of the amplified inut noise won't be white, because
the high frequency components are amplified more than the low, so rms
summation may be over-optimistic.


The stray capacitances on the printed circuit board seem to be higher
than you have estimated. Have you tried measuring the stray capacitance
on an unpopulated printed circuit board?


Minimise the stray capacitance - shorter tracks, narrower tracks,
remove ground plane under the crucial tracks, as recommended in the
Burr-Brown application notes. Using a two-layer layout on the outer
layers of a four layer board reduces the distance to the ground plane
by a factor of three, and roughly triples the capacitance of any
microstrip tracks, even before you add in any extra ground plane.

Make sure that you have both the recommended 100nf decoupling capacitor
to ground close to the op amp's power pins, and a couple of uF of
tantalum capacitorless than an inch (25mm) away, as recommended by the
application notes. I tend to throw in a ferrite chip/bead between the
capacitors and the power rail - they are fairly cheap, don't introduce
any DC drop, and - being lossy non-wound devices - look resistive
rather than capacitative up to very high frequencies. The of the order
of 1uH of inductance can resonate with the decoupling capacitor, but as
long as you have a few uF of tantalum capacitor present, the resonance
is around 100kHz and the impedance at resonance low enough - at around
0R6 - that the ESR of the tantalum bead will kill any peaking.

Thanks for your response.

I measured the stray capacitance on both boards and the connectors. Both
are larger than I had estimated, but not enough to explain my measured
noise. The first board with connectors and the op-amp input capacitance is
16-19 pF while for the second board the capacitance is 18-21 pF. The
measured capacitance on the first board now gives good agreement when the
op-amp is the OPA656 (at 33 kHz the model gives about 475 nV/rt-Hz while I
measure 500 nV/rt-Hz.

These larger capacitance values still leave unexplained why when I changed
to the OPA657, which has the same current noise and input capacitance specs
as the OPA656 and slightly lower voltage noise (5 nV/rt-Hz versus 7
nV/rt-Hz), the noise at 33 kHz increased to 600 nV/rt-Hz.

Also unexplained is why with the second board and an OPA657, the noise is
1000 nV/rt-Hz, which is more than twice what I calculate.

I did confirm that the power pins have 100 nf decoupling capacitors, 10 uF
capacitors within 1/4" of the power pins and a ferrite beam between the
capacitors and the power rail.

Bret Cannon

 

[Winfield Hill]

Bret Cannon wrote...

These larger capacitance values still leave unexplained why when I changed
to the OPA657, which has the same current noise and input capacitance specs
as the OPA656 and slightly lower voltage noise (5 nV/rt-Hz versus 7
nV/rt-Hz), the noise at 33 kHz increased to 600 nV/rt-Hz.

In this situation, you should measure your various opamp's
actual voltage noise, rather than rely on a spec.

BTW, a possible noise source that hasn't been mentioned is a
falling opamp power-supply rejection at higher frequencies,
and high-frequency power-supply noise. Usually your bypass
caps help reduce the latter, etc., but it's worth examining.

 

[[email protected]]

Thanks for your response.

I measured the stray capacitance on both boards and the connectors. Both
are larger than I had estimated, but not enough to explain my measured
noise. The first board with connectors and the op-amp input capacitance is
16-19 pF while for the second board the capacitance is 18-21 pF.

"Connectors" implies lengths of coaxial cable, with a capacitance of
about 100pF/metre. I presume that this is included in the measured
capacitance.
What are you using to measure your capacitances?

The measured capacitance on the first board now gives good agreement when the
op-amp is the OPA656 (at 33 kHz the model gives about 475 nV/rt-Hz while I
measure 500 nV/rt-Hz.

These larger capacitance values still leave unexplained why when I changed
to the OPA657, which has the same current noise and input capacitance specs
as the OPA656 and slightly lower voltage noise (5 nV/rt-Hz versus
7 nV/rt-Hz), the noise at 33 kHz increased to 600 nV/rt-Hz.

The OPA656 is going to run out of gain at about a MHz, at a gain of
about 100, set by the 0.15pF parallel capacitance of the 10M resistor,
so the output will have some
800nV/root Hz of random noise. The closed loop bandwidth is about a
MHz, so the rms noise will about 800uV, about 5mV ptp.

The OPA657 has six times the bandwidth, so the broadband noise on its
output will be higher by root six, say to 12.5mV peak-to-peak.

This makes the output relatively noisy. Do you have a damping resistor
on the op amp output, to isolate it from any capacitative loads? The
data sheet recommends 100R for a 10pF load, progressively reducing as
the capacitative load increases (see the curve at the bottom of page 7
of the OPA657 data sheet). The op amp may not oscillate without it, but
the closed loop gain can develop a signficant peak before it starts
rolling off, which might boost the noise to the point where the output
stage starts getting non-linear and producing difference products
between the noise signal and harmonics in the tuning fork output.

Also unexplained is why with the second board and an OPA657, the noise is
1000 nV/rt-Hz, which is more than twice what I calculate.

The extra bandwidth of the OPA657 does make it better at picking up
externally generated noise.- once, early on in my career, I found that
a significant part of the "noise" on a circuit I was looking at was the
local AM radio transmission. Somehow the circuit had managed to
configure itself as a moderately effective AM receiver. More careful
shielding dealt with that.

I did confirm that the power pins have 100 nf decoupling capacitors, 10 uF
capacitors within 1/4" of the power pins and a ferrite beam between the
capacitors and the power rail.

Sounds good.

Bill Sloman, Nijmegen

 

[Bret Cannon]

"Connectors" implies lengths of coaxial cable, with a capacitance of
about 100pF/metre. I presume that this is included in the measured
capacitance.
What are you using to measure your capacitances?




The OPA656 is going to run out of gain at about a MHz, at a gain of
about 100, set by the 0.15pF parallel capacitance of the 10M resistor,
so the output will have some
800nV/root Hz of random noise. The closed loop bandwidth is about a
MHz, so the rms noise will about 800uV, about 5mV ptp.

The OPA657 has six times the bandwidth, so the broadband noise on its
output will be higher by root six, say to 12.5mV peak-to-peak.

This makes the output relatively noisy. Do you have a damping resistor
on the op amp output, to isolate it from any capacitative loads? The
data sheet recommends 100R for a 10pF load, progressively reducing as
the capacitative load increases (see the curve at the bottom of page 7
of the OPA657 data sheet). The op amp may not oscillate without it, but
the closed loop gain can develop a signficant peak before it starts
rolling off, which might boost the noise to the point where the output
stage starts getting non-linear and producing difference products
between the noise signal and harmonics in the tuning fork output.


The extra bandwidth of the OPA657 does make it better at picking up
externally generated noise.- once, early on in my career, I found that
a significant part of the "noise" on a circuit I was looking at was the
local AM radio transmission. Somehow the circuit had managed to
configure itself as a moderately effective AM receiver. More careful
shielding dealt with that.


Sounds good.

Bill Sloman, Nijmegen

For the first board, the connectors are an SMA bulkhead connector soldered
to the board, an SMA to BNC adapter, and a BNC male union. To measure the
capacitance, I connect a Fluke 4 1/2 digit multi-meter (I don't remember the
model) via a twin banana-plug to BNC adapter. The capacitance reading of
the Fluke with the banana-plug adapter in place is a quite stable 0.067 nF
and when I connect the first circuit board with its connectors, the
capacitance reading increases to 0.074 nF if the red terminal of the
multimeter goes to the center pin of the BNC. If I reverse the banana to
BNC, then this last reading is 0.077 nF, while the "zero" is still 0.067 nF.
I measure the capacitance of a tuning fork soldered to an isolated BNC
bulkhead as 4 pF independent of polarity.

I see noise spikes on the spectrum analyzer at 19 kHz and its multiples and
at 43 kHz unless I shield the input. (The 19 kHz is from switching supplies
in the lab and I think that the 43 kHz is from the fluorescent light
ballasts.) However, with several layers of aluminum foil connected to the
case of the pre-amp or putting a metal box around the input, these noise
spikes disappear into the noise floor. I haven't used an rf spectrum
analyzer to look for any pickup above 100 kHz, but I don't see anything on a
400 MHz bandwidth scope.

The output of the OPA656 and OPA657 on both boards has a 50 ohm resistor in
series to the output connection and a 100 ohm resistor to ground, which I'm
told reduces oscillation problems.

By the way, I'm measuring the noise with a Stanford Research FFT spectrum
analyzer, so I assume that the filtering on the spectrum analyzer takes care
of the noise at frequencies above 100 kHz. Is this wrong? If not, then why
would the wider gain bandwidth of the OPA657 increase the noise I see with
the spectrum analyzer?

thanks,
Bret Cannon

 

[Mark]

It was my understanding (perhaps wrong) that to really characterize the
noise of an op-amp correctly , you need 4 values....

the current noise of the + input,

the voltage noise of the + input,

the current noise of the - input

and the voltage noise of the - input.

Depending on what you have connected to the op-amp, these 4 numbers
will have varying significance. If you are connecting Hi Z sources,
the current noise becomes very important and the voltage noise less so.

see:

http://focus.ti.com/lit/an/slva043a/slva043a.pdf


Mark

 

[Terry Given]

Bret Cannon wrote...



In this situation, you should measure your various opamp's
actual voltage noise, rather than rely on a spec.

BTW, a possible noise source that hasn't been mentioned is a
falling opamp power-supply rejection at higher frequencies,
and high-frequency power-supply noise. Usually your bypass
caps help reduce the latter, etc., but it's worth examining.

that bit me a while back. I had 10dB of *GAIN* thru the supply at 100kHz.

Cheers
Terry

 

[[email protected]]

The output of the OPA656 and OPA657 on both boards has a 50 ohm resistor in
series to the output connection and a 100 ohm resistor to ground, which I'm
told reduces oscillation problems.

The 50R resistor in series with the output connection os good - not
only does it isolate the amplifier from any capacitative load, but it
also acts as a series termination for driving 50R transmission lines
(like regular coaxial cable).

The 100R resistor to ground is a mistake - the amplifier is specified
for driving a 100R load, but that is your 50R source (series)
termination resistor, plus the 50R parallel terminating resistor at the
other end of your 50R coaxial cable. Adding a 100R resistor to ground
in parallel with the cable load doesn't do anything useful, and
overlaods your output stage.

Whoever used it to kill and oscillation didn't know what they were
doing - the extra load may kill the amplifier's gain enough to
stabilise the loop, but at the expene of crippling the amplifier.

Get rid of it.It may be indirectly contributing to your noise
problems.

By the way, I'm measuring the noise with a Stanford Research FFT spectrum
analyzer, so I assume that the filtering on the spectrum analyzer takes care
of the noise at frequencies above 100 kHz. Is this wrong? If not, then why
would the wider gain bandwidth of the OPA657 increase the noise I see with
the spectrum analyzer?

The wider gain bandwidth of the OPA657 is going to mean that there is
more wide-band noise on its output than there is on the output of the
OPA656. This means more wide-band current flowing into and out of the
power supply pins, and more opportunity for high -frequency noise
components to mix and generate low frequency difference products that
may add to your Johnson noise sources.

If you are over-loading the output stage, you may be forcing that to
operate in a voltage current region where it isn't all that lnear, and
the non-linearity can allow high-frequency components to mix and
generate extra low frequency noise..

 

[Winfield Hill]

Bret Cannon wrote...

I see noise spikes on the spectrum analyzer at 19 kHz and its
multiples and at 43 kHz unless I shield the input. (The 19 kHz
is from switching supplies in the lab and I think ...

I'm pretty sure the 19kHz you see is coming from the radiated
magnetic field from the SRS analyzer's own CRT sweep circuits.
But there's very little energy in these narrow spikes and they
can be safely ignored.

Did you see my point about measuring the opamp's actual voltage
noise density, rather than relying on a datasheet typical spec?
That's an important step, as is checking the opamp's power-supply
high frequency rejection, and your power-supply's noise spectrum.

 

[Bret Cannon]

Bret Cannon wrote...

In this situation, you should measure your various opamp's
actual voltage noise, rather than rely on a spec.

BTW, a possible noise source that hasn't been mentioned is a
falling opamp power-supply rejection at higher frequencies,
and high-frequency power-supply noise. Usually your bypass
caps help reduce the latter, etc., but it's worth examining.

Thanks for the response.

I measured the input voltage noise density of the OPA657 on the first board
today and it matches the datasheet value of 4.8 nV/rt-Hz at 100 kHz and was
only slightly above 5 nV/rt-Hz at 10 kHz.

I've also measured the power supply noise densities at both the positive and
negative and the are well below the output noise density from at least 100
Hz to 100 kHz. There is a slight noise peak at about 60 kHz on in negative
input power to the metal case, but the -5 V regulator and the filtering on
the circuit board seem to eliminate that noise bump.

Bret Cannon

 

[Bret Cannon]

It was my understanding (perhaps wrong) that to really characterize the
noise of an op-amp correctly , you need 4 values....

the current noise of the + input,

the voltage noise of the + input,

the current noise of the - input

and the voltage noise of the - input.

Depending on what you have connected to the op-amp, these 4 numbers
will have varying significance. If you are connecting Hi Z sources,
the current noise becomes very important and the voltage noise less so.

see:

http://focus.ti.com/lit/an/slva043a/slva043a.pdf


Mark

The current noise spec is 1.3 fA/rt-Hz and this is much less than the
Johnson current noise of the 10 M feedback resistor of about 40 fA/rt-Hz.
With the non-inverting input shorted to ground, the current noise at this
input would not contribute anyway.

I have seen some op-amps with different noise specs for the inverting and
non-inverting input, but the datasheet for the OPA656 and OPA657 don't
mention any difference for these parts.

Bret Cannon

 

[Bret Cannon]

The 50R resistor in series with the output connection os good - not
only does it isolate the amplifier from any capacitative load, but it
also acts as a series termination for driving 50R transmission lines
(like regular coaxial cable).

The 100R resistor to ground is a mistake - the amplifier is specified
for driving a 100R load, but that is your 50R source (series)
termination resistor, plus the 50R parallel terminating resistor at the
other end of your 50R coaxial cable. Adding a 100R resistor to ground
in parallel with the cable load doesn't do anything useful, and
overlaods your output stage.

Whoever used it to kill and oscillation didn't know what they were
doing - the extra load may kill the amplifier's gain enough to
stabilise the loop, but at the expene of crippling the amplifier.

Get rid of it.It may be indirectly contributing to your noise
problems.


The wider gain bandwidth of the OPA657 is going to mean that there is
more wide-band noise on its output than there is on the output of the
OPA656. This means more wide-band current flowing into and out of the
power supply pins, and more opportunity for high -frequency noise
components to mix and generate low frequency difference products that
may add to your Johnson noise sources.

If you are over-loading the output stage, you may be forcing that to
operate in a voltage current region where it isn't all that lnear, and
the non-linearity can allow high-frequency components to mix and
generate extra low frequency noise..

We removed the 100 ohm resistor to ground today and the noise spectrum
didn't change.

I borrowed an rf spectrum analyzer so I could look at frequencies beyond 100
kHz and didn't see any noise peaks or oscillations up to 1 GHz. I did find
that from about 200 kHz to a couple of MHz the noise is flat and matches 5
nV/rt-Hz times a noise gain of 7 pF/0.15 pF with nothing connected to the
SMA connector soldered to the board. So it seems that I'm dealing with a
noise peak at about 60 kHz in addition to the Johnson noise of the feedback
resistor and the noise gain times the input voltage noise.

Bret Cannon

 

[Bret Cannon]

Bret Cannon wrote...

I'm pretty sure the 19kHz you see is coming from the radiated
magnetic field from the SRS analyzer's own CRT sweep circuits.
But there's very little energy in these narrow spikes and they
can be safely ignored.

Did you see my point about measuring the opamp's actual voltage
noise density, rather than relying on a datasheet typical spec?
That's an important step, as is checking the opamp's power-supply
high frequency rejection, and your power-supply's noise spectrum.

I saw your post and replied to it. The voltage noise density that I
measured is an excellent match to the data sheet in the range of the SRS
analyzer.

The noise on the power supply pins of the OPA657 on the board is about 35
nV/rt-hz at 60 kHz where the noise peak is, and is falling. Do you think I
need to check the power supply noise spectrum at higher frequencies than the
100 kHz upper limit of the SRS analyzer?

Perhaps I misunderstood your point about checking the op-amp's power supply
high frequency rejection. Do you think that I should measure these values
on the op-amp on the board? Is there an easy way to measure this without
pulling out all the power supply filtering on the board?

Thanks,
Bret Cannon

 

[Winfield Hill]

Bret Cannon wrote...

The voltage noise density that I measured is an excellent match
to the data sheet in the range of the SRS analyzer.

Well, by process of elimination is seems we're pointed back to
the circuit's capacitance. I don't see an explicit mention of
this noise source in your posts, except for "the model," which
you didn't specify. The noise we're talking about is what I call
"e_n-Cin" noise, an apparent noise current i_n = e_n Cin 2pi f,
where e_n is the effective opamp voltage noise density, including
PSRR stuff, etc., and Cin is the sum of all the capacitances at
the summing junction node, including the sensor, all the internal
input-stage capacitances of the opamp, etc. Note this noise source
rises with frequency, like the one you observed. The e_n-Cin noise
current is multiplied by the amplifier's trans-resistance and is
seen as a rising voltage-noise density on the output. The latter
assumes a flat transimpedance, of course, which may be incorrect
and has to be separately tested with calibrated input current
signals. I say calibrated, because you cannot necessarily simply
create small test currents with a resistor, due to the resistor's
parallel self capacitance, which is generally about 0.2pF or so.

We're looking for any peaking of the amplifier's transimpedance,
which will occur if the oamp's f_T is insufficient or the feedback
resistor has not been properly bypassed, f_T > fc^2 2pi Rf Cin,
where fc is 1/2pi Rf Cf. For Cin = 20pF+5pF (opamp) and Rf = 10M
and an fc of 60kHz we see f_T must be greater than 5MHz. Checking
on the OPA657 we see its spec is 1600MHz. OK, that doesn't look
like a problem. BTW, at 60kHz the 10M feedback resistor and
its strays must have much less than 0.26pF of capacitance to still
look like a resistor and not a capacitor. What have you done about
that? Did you tell us what the overall frequency response of your
transresistance amplifier is supposed to be? Have you measured it?
How, exactly?

 

[Bret Cannon]

Bret Cannon wrote...

Well, by process of elimination is seems we're pointed back to
the circuit's capacitance. I don't see an explicit mention of
this noise source in your posts, except for "the model," which
you didn't specify. The noise we're talking about is what I call
"e_n-Cin" noise, an apparent noise current i_n = e_n Cin 2pi f,
where e_n is the effective opamp voltage noise density, including
PSRR stuff, etc., and Cin is the sum of all the capacitances at
the summing junction node, including the sensor, all the internal
input-stage capacitances of the opamp, etc. Note this noise source
rises with frequency, like the one you observed. The e_n-Cin noise
current is multiplied by the amplifier's trans-resistance and is
seen as a rising voltage-noise density on the output. The latter
assumes a flat transimpedance, of course, which may be incorrect
and has to be separately tested with calibrated input current
signals. I say calibrated, because you cannot necessarily simply
create small test currents with a resistor, due to the resistor's
parallel self capacitance, which is generally about 0.2pF or so.

We're looking for any peaking of the amplifier's transimpedance,
which will occur if the oamp's f_T is insufficient or the feedback
resistor has not been properly bypassed, f_T > fc^2 2pi Rf Cin,
where fc is 1/2pi Rf Cf. For Cin = 20pF+5pF (opamp) and Rf = 10M
and an fc of 60kHz we see f_T must be greater than 5MHz. Checking
on the OPA657 we see its spec is 1600MHz. OK, that doesn't look
like a problem. BTW, at 60kHz the 10M feedback resistor and
its strays must have much less than 0.26pF of capacitance to still
look like a resistor and not a capacitor. What have you done about
that? Did you tell us what the overall frequency response of your
transresistance amplifier is supposed to be? Have you measured it?
How, exactly?

The noise model is the "e_n_Cin" noise added as the square root of the sum
of the squares to the Johnson noise of the feedback resistor. The Johnson
noise is rolled off by the stray capacitance of the feedback. In addition,
I've included the finite open loop gain as a function of frequency of the
OPA657 rather than taking the limiting case of infinite gain.

I measured the frequency response using the SRS analyzer, but with using an
old switched attenuator box to reduce the drive voltage to about 1 mV. This
drive voltage goes to channel 1 of the SRS analyzer and drives a thru-hole
resistor soldered into a Pomona box. I've used resistance values from 1 M
down to 25k, though most of the time I've used either a 91k or 37.4k
resistor, since this matches the series resistance of the quartz tuning
forks at the pressure of interest. The transimpedance gain with both 91k
and 37.4 k is down about 2.5 dB at 100 kHz. I calculated the 3 dB
frequency is about 106 kHz, from which I calculate the capacitance of the
feedback as 0.15 pF. The frequency response shows a little peaking, less
than 1 dB, before it starts to roll off.

BTW, I measure the transimpedance gain varying with source resistance, which
is very well fit with a model using a finite open loop gain for the op-amp.
The parameters for a best fit to my data are an open loop gain of 72 dB
versus 75 dB from the datasheet and the model feedback resistance matches
the actual feedback resistance to 1%.

Also, yesterday I borrowed an rf spectrum analyzer to look at the noise
above 100 kHz. I didn't see any noise bumps or oscillation up to 1 GHz. I
did see a flat noise density from about 200 kHz to a few MHz that matches (5
nV/rt-Hz) x (7 pF/0.15 pF). It seems to me that what I'm fighting is a
noise peak at about 60 kHz on top of the Johnson noise and the "e_n_Cin"
noise.

Did you get my last post from last night asking about measuring PSRR on the
board? Terry Givens' posting of having seen 10 dB of gain at 100 kHz in an
op-amp has me thinking that 35 nV/rt-Hz noise on the power pins at 60 kHz
may not be enough to eliminate that noise source.

Thanks,
Bret Cannon

 

[Phil Hobbs]

Also, yesterday I borrowed an rf spectrum analyzer to look at the noise
above 100 kHz. I didn't see any noise bumps or oscillation up to 1 GHz. I
did see a flat noise density from about 200 kHz to a few MHz that matches (5
nV/rt-Hz) x (7 pF/0.15 pF). It seems to me that what I'm fighting is a
noise peak at about 60 kHz on top of the Johnson noise and the "e_n_Cin"
noise.

Did you get my last post from last night asking about measuring PSRR on the
board? Terry Givens' posting of having seen 10 dB of gain at 100 kHz in an
op-amp has me thinking that 35 nV/rt-Hz noise on the power pins at 60 kHz
may not be enough to eliminate that noise source.

I'm too buried to go through all the posts on this topic, so apologies
if this has been suggested before....but a 60 kHz peak can be caused by
a linear voltage regulator with the wrong bypass cap. The output of a
7805 looks inductive, and the bigger the bypass, the bigger (and lower
in frequency) the noise peak is.

You might try putting a 5-ohm resistor in series with the regulator
output and see if that makes life better.

Cheers,

Phil Hobbs

 

[[email protected]]

Bret Cannon wrote...

Well, by process of elimination is seems we're pointed back to
the circuit's capacitance. I don't see an explicit mention of
this noise source in your posts, except for "the model," which
you didn't specify. The noise we're talking about is what I call
"e_n-Cin" noise, an apparent noise current i_n = e_n Cin 2pi f,
where e_n is the effective opamp voltage noise density, including
PSRR stuff, etc., and Cin is the sum of all the capacitances at
the summing junction node, including the sensor, all the internal
input-stage capacitances of the opamp, etc. Note this noise source
rises with frequency, like the one you observed. The e_n-Cin noise
current is multiplied by the amplifier's trans-resistance and is
seen as a rising voltage-noise density on the output. The latter
assumes a flat transimpedance, of course, which may be incorrect
and has to be separately tested with calibrated input current
signals. I say calibrated, because you cannot necessarily simply
create small test currents with a resistor, due to the resistor's
parallel self capacitance, which is generally about 0.2pF or so.

We're looking for any peaking of the amplifier's transimpedance,
which will occur if the oamp's f_T is insufficient or the feedback
resistor has not been properly bypassed, f_T > fc^2 2pi Rf Cin,
where fc is 1/2pi Rf Cf. For Cin = 20pF+5pF (opamp) and Rf = 10M
and an fc of 60kHz we see f_T must be greater than 5MHz. Checking
on the OPA657 we see its spec is 1600MHz. OK, that doesn't look
like a problem. BTW, at 60kHz the 10M feedback resistor and
its strays must have much less than 0.26pF of capacitance to still
look like a resistor and not a capacitor.

Come on Win - pay attention. In his original post Bret had already
assigned a value of 0.15pF to the parallel capacitance of his 10M
resistor, based on the frequency dependence of the closed loop gain of
the op amp.

Bret does seem to know his physics, though he isn't particularly
sophiistcated about electronics. At the moment he seems to be happy
with the noise figures for the circuit built around an OPA656, but the
nominally quieter OPA657 produces more noise than the OPA656, and he
can't find an explanation..

The only explanation that I've been able to come up with involves
handwaving about the difference between the 5mV peak-to-peak broad-band
noise he ought to be seeing at the output of the OPA656, and the 15mV
ptp I'd expect at ithe output of an OPS657.

The fact that the removal of the 100R to ground at the output didn't
make any difference to the noise level suggests that this explanation
doesn't work.

 

[Terry Given]

The noise model is the "e_n_Cin" noise added as the square root of the sum
of the squares to the Johnson noise of the feedback resistor. The Johnson
noise is rolled off by the stray capacitance of the feedback. In addition,
I've included the finite open loop gain as a function of frequency of the
OPA657 rather than taking the limiting case of infinite gain.

I measured the frequency response using the SRS analyzer, but with using an
old switched attenuator box to reduce the drive voltage to about 1 mV. This
drive voltage goes to channel 1 of the SRS analyzer and drives a thru-hole
resistor soldered into a Pomona box. I've used resistance values from 1 M
down to 25k, though most of the time I've used either a 91k or 37.4k
resistor, since this matches the series resistance of the quartz tuning
forks at the pressure of interest. The transimpedance gain with both 91k
and 37.4 k is down about 2.5 dB at 100 kHz. I calculated the 3 dB
frequency is about 106 kHz, from which I calculate the capacitance of the
feedback as 0.15 pF. The frequency response shows a little peaking, less
than 1 dB, before it starts to roll off.

BTW, I measure the transimpedance gain varying with source resistance, which
is very well fit with a model using a finite open loop gain for the op-amp.
The parameters for a best fit to my data are an open loop gain of 72 dB
versus 75 dB from the datasheet and the model feedback resistance matches
the actual feedback resistance to 1%.

Also, yesterday I borrowed an rf spectrum analyzer to look at the noise
above 100 kHz. I didn't see any noise bumps or oscillation up to 1 GHz. I
did see a flat noise density from about 200 kHz to a few MHz that matches (5
nV/rt-Hz) x (7 pF/0.15 pF). It seems to me that what I'm fighting is a
noise peak at about 60 kHz on top of the Johnson noise and the "e_n_Cin"
noise.

Did you get my last post from last night asking about measuring PSRR on the
board? Terry Givens' posting of having seen 10 dB of gain at 100 kHz in an
op-amp has me thinking that 35 nV/rt-Hz noise on the power pins at 60 kHz
may not be enough to eliminate that noise source.

Thanks,
Bret Cannon

it was a TL064, 1MHz GBW from ST, very cheap and very low power. I think
your opamp will be a bit better

Cheers
Terry

 

[Winfield Hill]

Bret Cannon wrote...

Also, yesterday I borrowed an rf spectrum analyzer to look at the noise
above 100 kHz. I didn't see any noise bumps or oscillation up to 1 GHz.
I did see a flat noise density from about 200 kHz to a few MHz that
matches (5 nV/rt-Hz) x (7 pF/0.15 pF). It seems to me that what I'm
fighting is a noise peak at about 60 kHz on top of the Johnson noise
and the "e_n_Cin" noise.

Just to clarify this "flat noise" bit. Once you're above the e_n-Cin
noise threshold, the noise density goes up with frequency. But above
100kHz you're also above the transimpedance rolloff frequency, where
the gain is dropping with frequency, thanks to the feedback resistor's
self capacitance. These two cancel and you see a flat output spectrum.

* But, this assumes the opamp has plenty of extra bandwidth, per the
formula I posted, so it doesn't also contribute to the rolloff. For
folks who have chosen a barely adequate opamp GBW, the rolloff will
be -12dB/octave rather than -6dB, or perhaps even more. Generally
this sharp rollof is a good thing. Certainly it helps to reduce the
wideband rms noise.