Differential mode analog active filter?

[Vladimir Vassilevsky]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

What would be the best way to do this circuit without sacrificing the
SNR and CMRR performance?

If we make the two identical analog filters for /+/ and /-/ paths, then
the CMRR will suffer because of the limited tolerance of the components.
I would expect problems mainly due to the accuracy of the capacitors.

Making a filter with fully differential opamps does not seem to be very
different from the two separate filters. There is the same limitation
because of the tolerances.

It is possible to collapse the input differential signal into the single
ended using an inamp, then filter it and convert it back to the
differential in front of the ADC. The CMRR will be limited to that of
the inamp, and the SNR is in danger because of the conversion from the
differential to ground referenced and back.

I tried to design a schematic similar to the conventional filter but
with the floating ground. A simulation shows problems with the CMRR and
the frequency response; the bootstrapping of the power rails of the
opamps towards the floating ground helps somewhat, however it rather
complex and doesn't look as a very reliable solution.

What topology should be the good approach to the problem? Can you
suggest a book, an appnote or any other information?

VLV

 

[Tim Wescott]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

What would be the best way to do this circuit without sacrificing the
SNR and CMRR performance?

If we make the two identical analog filters for /+/ and /-/ paths, then
the CMRR will suffer because of the limited tolerance of the components.
I would expect problems mainly due to the accuracy of the capacitors.

Making a filter with fully differential opamps does not seem to be very
different from the two separate filters. There is the same limitation
because of the tolerances.

It is possible to collapse the input differential signal into the single
ended using an inamp, then filter it and convert it back to the
differential in front of the ADC. The CMRR will be limited to that of
the inamp, and the SNR is in danger because of the conversion from the
differential to ground referenced and back.

I tried to design a schematic similar to the conventional filter but
with the floating ground. A simulation shows problems with the CMRR and
the frequency response; the bootstrapping of the power rails of the
opamps towards the floating ground helps somewhat, however it rather
complex and doesn't look as a very reliable solution.

What topology should be the good approach to the problem? Can you
suggest a book, an appnote or any other information?

VLV

The only concrete suggestions I have are negative -- if it's a
sampling converter, watch out that whatever solution you adopt is low
impedance on the output, and damps transient currents well. The
high-performance, differential input SARs that I have worked around (but
not on) inject a bit of charge back into their input pins when they
sample; this charge is more or less random, and if it's effects persist
until the next sample then it's just another thing that degrades accuracy.

I'd help you more, but on the projects that I've done using high
performance ADCs I've had analog circuit designers who were eager to
interpose themselves between me and the signal acquisition chain, so I
haven't had the opportunity to design one yet.

--
Tim Wescott
Control systems and communications consulting
http://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes, http://www.wescottdesign.com/actfes/actfes.html

 

[John Popelish]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

What would be the best way to do this circuit without sacrificing the
SNR and CMRR performance?

If we make the two identical analog filters for /+/ and /-/ paths, then
the CMRR will suffer because of the limited tolerance of the components.
I would expect problems mainly due to the accuracy of the capacitors.

Making a filter with fully differential opamps does not seem to be very
different from the two separate filters. There is the same limitation
because of the tolerances.

It is possible to collapse the input differential signal into the single
ended using an inamp, then filter it and convert it back to the
differential in front of the ADC. The CMRR will be limited to that of
the inamp, and the SNR is in danger because of the conversion from the
differential to ground referenced and back.

I tried to design a schematic similar to the conventional filter but
with the floating ground. A simulation shows problems with the CMRR and
the frequency response; the bootstrapping of the power rails of the
opamps towards the floating ground helps somewhat, however it rather
complex and doesn't look as a very reliable solution.

What topology should be the good approach to the problem? Can you
suggest a book, an appnote or any other information?

Have you looked at passive LC filters with symmetry and no
ground connections? I think the matching problem with this
sort of design involves the stray capacitances to the
common, not the tolerance of the parts. Of course, the
filter will not remove common mode energy, so you are
relying completely on the differential A/D input to remove
the common mode signals.

 

[[email protected]]

The only concrete suggestions I have are negative -- if it's a
sampling converter, watch out that whatever solution you adopt is low
impedance on the output, and damps transient currents well. The
high-performance, differential input SARs that I have worked around (but
not on) inject a bit of charge back into their input pins when they
sample; this charge is more or less random, and if it's effects persist
until the next sample then it's just another thing that degrades accuracy.

I'd help you more, but on the projects that I've done using high
performance ADCs I've had analog circuit designers who were eager to
interpose themselves between me and the signal acquisition chain, so I
haven't had the opportunity to design one yet.

--
Tim Wescott
Control systems and communications consultinghttp://www.wescottdesign.com

Need to learn how to apply control theory in your embedded system?
"Applied Control Theory for Embedded Systems" by Tim Wescott
Elsevier/Newnes,http://www.wescottdesign.com/actfes/actfes.html

When I was at IDT, we had a flash converter with performance very
dependent on the op amp driving it. My recollection is on the DUT
board we used some damn expensive Comlinear part which cost multiples
of the DUT itself.

I'd design the filter single ended using leapfrog design. Such designs
tend to have the lowest THD due to the op amp being used with one
input at ground.

I've done fully differential leapfrog designs in switchcap
implementation, where I have full control over the op amp. Obvious to
most but just to be sure everyone is on the same page, the fully
differential designs need to set the output common mode of each op
amp.

 

[colin]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of the
components is not a big issue.

What would be the best way to do this circuit without sacrificing the SNR
and CMRR performance?

If we make the two identical analog filters for /+/ and /-/ paths, then
the CMRR will suffer because of the limited tolerance of the components. I
would expect problems mainly due to the accuracy of the capacitors.

Making a filter with fully differential opamps does not seem to be very
different from the two separate filters. There is the same limitation
because of the tolerances.

It is possible to collapse the input differential signal into the single
ended using an inamp, then filter it and convert it back to the
differential in front of the ADC. The CMRR will be limited to that of the
inamp, and the SNR is in danger because of the conversion from the
differential to ground referenced and back.

I tried to design a schematic similar to the conventional filter but with
the floating ground. A simulation shows problems with the CMRR and the
frequency response; the bootstrapping of the power rails of the opamps
towards the floating ground helps somewhat, however it rather complex and
doesn't look as a very reliable solution.

What topology should be the good approach to the problem? Can you suggest
a book, an appnote or any other information?

If all you want is an anti aliasing filter then if you make a filter with no
reference to ground then you should not introduce any CMRR problems. maybe
just a differential mode torroid choke and a capacitor and some resistance
for damping.

However if you want to reject common mode noise to improve CMRR this might
be more difficult,
but ordinary LP filters on each input should help and not affect the CMRR
below their cutoff frequency.

Colin =^.^=

 

[Fred Bloggs]

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

It takes more to make a high performance design than just throwing high
performance and expensive parts at it. You seem to have no concept of
SINAD degradation of SNR, nor is it important to you to mention
bandwidth. You either decouple the 4th order Bessel requirement from the
anti-alias function or drown under the weight of the competent invention
needed.

 

[Vladimir Vassilevsky]

John Popelish wrote:

Have you looked at passive LC filters with symmetry and no ground
connections?

The required inductance is at the order of several mH. Besides the
inevitable problems with the mismatch, the inductors of that kind will
act as antennas picking up noise.

I think the matching problem with this sort of design
involves the stray capacitances to the common, not the tolerance of the
parts.

That's is exactly what happens when the conventional single ended active
filter is used for the differential operation with the floating ground.
The bootstrapping of the power rails helps, however it is cumbersome and
the performance is limited.

Of course, the filter will not remove common mode energy, so you
are relying completely on the differential A/D input to remove the
common mode signals.

Straight to the point!
Actually, it is a bit more complex. In the short, the analog filter has
to suppress only the critical bands near the multiples of the sample
rate; the rest of the filtering is done by the DSP. In the "low" bands,
I can rely on the common mode suppression of the ADC. The "high" bands
are attenuated by the other filters in /+/ and /-/ paths; the cutoff is
high enough not to worry about the CMRR degradation in the band of interest.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

 

[John Popelish]

John Popelish wrote:




The required inductance is at the order of several mH.

This is little problem.

Besides the
inevitable problems with the mismatch, the inductors of that kind will
act as antennas picking up noise.

Toroidal designs are pretty good in that respect.
That you need a Bessel design and not an elliptical or
Chebychev (lower Q resonators) helps a lot.

That's is exactly what happens when the conventional single ended active
filter is used for the differential operation with the floating ground.
The bootstrapping of the power rails helps, however it is cumbersome and
the performance is limited.

With a floating differential filter, I would enclose it in a
shield driven by the common mode voltage.

 

[colin]

The required inductance is at the order of several mH. Besides the
inevitable problems with the mismatch, the inductors of that kind will act
as antennas picking up noise.

You can have a torroid with two equal windings wich should not present any
problems.
1mh isnt hard to acheive. (bit like a common mode choke but with one winding
the other way round).

Straight to the point!

but above the cuttoff frequency of the differential filter the CMRR will not
be so badly affected by the mismatch of two low pass filters before it.

Colin =^.^=

 

[Kadir Solid Gold Suleyman]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

What topology should be the good approach to the problem? Can you
suggest a book, an appnote or any other information?

VLV

Hello Vladmir,

This is an interesting problem! I recommend you read James Karki's
app note SLOA54D especially the section about Multi Feed Back topology
and enhanced CMRR. (Maybe you have read this already?) By the way,
the THS4509 kicks ass!!!* It is an amazing fully differential
amplifier. Look at the GBW product on that baby. I used it for an
application and was extremely impressed with its performance. Oh
yeah, use the most accurate caps and resistors you can find! but you
know that. Good luck.

regards,
Kadir "Solid Gold" Suleyman
*No I don't work for TI.

 

[Fred Bloggs]

Oh
yeah, use the most accurate caps and resistors you can find!

Lessee 0.1% match preserves 60dB so ummm...0.001% match for 100dB CMRR?

 

[Kadir Solid Gold Suleyman]

Lessee 0.1% match preserves 60dB so ummm...0.001% match for 100dB CMRR?

Yeah good luck measuring the capacitors that accurately, right? I
didn't read that he needed 100dB CMRR just that his ADC was capable of
this and he didn't want to 'sacrifice' the performance of his ADC.
The graph of CMRR performance using the test circuit of the 4509 data
sheet shows 90dB @ 100kHz using 5V supplies. At higher frequencies
it is worse but it sounds like for his application the higher
frequencies are of less concern. He never specified bandwidth. For a
part that costs less than $4 in 1k quantity, I think the 4509 has
pretty darn good performance but he's not going to get 100dB CMRR
performance out of it without a having board manufacturable only by
thaumaturgic obeissance.

 

[[email protected]]

John Popelish wrote: (snip)

The required inductance is at the order of several mH. Besides the
inevitable problems with the mismatch, the inductors of that kind will
act as antennas picking up noise.

(snip)

Here is an example of the kind of inductors I am talking about:
http://www.vishay.com/docs/34078/tetd.pdf
20 H max inductance.

 

[Jim Thompson]

Hello All,

There is a high precision ADC with the differential input (SNR ~ CMRR ~
100dB). The incoming analog signal is differential also. However there
should be the antialiasing filter in front of the ADC. The filter is
Bessel 4th order. This is the high performance design, and the cost of
the components is not a big issue.

What would be the best way to do this circuit without sacrificing the
SNR and CMRR performance?

If we make the two identical analog filters for /+/ and /-/ paths, then
the CMRR will suffer because of the limited tolerance of the components.
I would expect problems mainly due to the accuracy of the capacitors.

Making a filter with fully differential opamps does not seem to be very
different from the two separate filters. There is the same limitation
because of the tolerances.

It is possible to collapse the input differential signal into the single
ended using an inamp, then filter it and convert it back to the
differential in front of the ADC. The CMRR will be limited to that of
the inamp, and the SNR is in danger because of the conversion from the
differential to ground referenced and back.

I tried to design a schematic similar to the conventional filter but
with the floating ground. A simulation shows problems with the CMRR and
the frequency response; the bootstrapping of the power rails of the
opamps towards the floating ground helps somewhat, however it rather
complex and doesn't look as a very reliable solution.

What topology should be the good approach to the problem? Can you
suggest a book, an appnote or any other information?

VLV

Take a look at the gyrator-based filter configurations on the SED page
of my website. You should be able to easily extend these to
differential.

...Jim Thompson

 

[Vladimir Vassilevsky]

Jim Thompson wrote:

Take a look at the gyrator-based filter configurations on the SED page
of my website. You should be able to easily extend these to
differential.

Thank you very much for the good advice. Yes, the gyrator topology works
well for the differential filter.

VLV

 

[Jim Thompson]

Jim Thompson wrote:



Thank you very much for the good advice. Yes, the gyrator topology works
well for the differential filter.

VLV

And you can synthesize BIG inductors ;-)

...Jim Thompson

 

[John Popelish]

I've been thinking about your differential input A/D
converter and wondered if you might be able to make use of a
differential amplifier with a differential output as a front
end, something like this:


___ ___
in+ -|___|-+----|___|-+
| |
| |\| |
+----|-\ |
| | >--+ out-
| +-|+/
| | |/|
| |
| |
___ | | ___
in- -|___|----+-|___|-+
| | |
| | |\| |
| +-|-\ |
| | >--+ out+
+----|+/
|/|

This circuit removes the common mode part of the signal to
the extent that the two resistor pairs match in ratio. You
can turn this into a one pole low pass differential filter
by paralleling each of the feedback resistors with
capacitors, or extend the concept into a pair of MFB 2 pole
Bessel filters.

I came up with this variation on the single opamp
subtractor, today, but I don't remember ever seeing it
before. Surely, something so simple and useful has a name.
Does anyone recognize it?

 

[Rich Grise]

I've been thinking about your differential input A/D
converter and wondered if you might be able to make use of a
differential amplifier with a differential output as a front
end, something like this:
___ ___
in+ -|___|-+----|___|-+
| |
| |\| |
+----|-\ |
| | >--+ out-
| +-|+/
| | |/|
| |
| |
___ | | ___
in- -|___|----+-|___|-+
| | |
| | |\| |
| +-|-\ |
| | >--+ out+
+----|+/
|/|

This circuit removes the common mode part of the signal to
the extent that the two resistor pairs match in ratio. You
can turn this into a one pole low pass differential filter
by paralleling each of the feedback resistors with
capacitors, or extend the concept into a pair of MFB 2 pole
Bessel filters.

I came up with this variation on the single opamp
subtractor, today, but I don't remember ever seeing it
before. Surely, something so simple and useful has a name.
Does anyone recognize it?

Well, if it is new, you only have 364 days left to file for a patent. ;-)

Cheers!
Rich

 

[john jardine]

I've been thinking about your differential input A/D
converter and wondered if you might be able to make use of a
differential amplifier with a differential output as a front
end, something like this:


___ ___
in+ -|___|-+----|___|-+
| |
| |\| |
+----|-\ |
| | >--+ out-
| +-|+/
| | |/|
| |
| |
___ | | ___
in- -|___|----+-|___|-+
| | |
| | |\| |
| +-|-\ |
| | >--+ out+
+----|+/
|/|

This circuit removes the common mode part of the signal to
the extent that the two resistor pairs match in ratio. You
can turn this into a one pole low pass differential filter
by paralleling each of the feedback resistors with
capacitors, or extend the concept into a pair of MFB 2 pole
Bessel filters.

I came up with this variation on the single opamp
subtractor, today, but I don't remember ever seeing it
before. Surely, something so simple and useful has a name.
Does anyone recognize it?

AD have started pushing diff in, diff out, amps (eg ADA4938) for their fast
ADCs but it's anyones guess what's inside.
In LTspice I'm seeing curious results both in .AC and .TRAN, if the Rs are
in the 10 to 1000 ohm area.
The circuit is perfect using a couple of VCVS' (1Vpk inputs) but LTspice
"ideal" single pole opamps, 25ma limit, Vos=0, seem to develop an offset
output signal and 10% second harmonic. I've set something wrong but don't
know what.

 

[Jim Thompson]

AD have started pushing diff in, diff out, amps (eg ADA4938) for their fast
ADCs but it's anyones guess what's inside.
In LTspice I'm seeing curious results both in .AC and .TRAN, if the Rs are
in the 10 to 1000 ohm area.
The circuit is perfect using a couple of VCVS' (1Vpk inputs) but LTspice
"ideal" single pole opamps, 25ma limit, Vos=0, seem to develop an offset
output signal and 10% second harmonic. I've set something wrong but don't
know what.

I see a positive feed back loop.

...Jim Thompson