distortion in current-mode opamps

[John Larkin]

Jim Williams' advice for achieving very low distortion in opamps is
"always invert." The idea here is that even if an opamp may have
screaming bandwidth and good open-loop linearity, it may well have
rotten common-mode performance, especially at high frequencies. And if
neither input pin has any substantial voltage on it, the various
nonlinear parasitic capacitances won't be pumped.

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

We wonder why.

John

 

[Martin Griffith]

Jim Williams' advice for achieving very low distortion in opamps is
"always invert." The idea here is that even if an opamp may have
screaming bandwidth and good open-loop linearity, it may well have
rotten common-mode performance, especially at high frequencies. And if
neither input pin has any substantial voltage on it, the various
nonlinear parasitic capacitances won't be pumped.

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

We wonder why.

Johnd

Yep, I saw that from JW as well.

Just been looking at some TI headphone amps, to drive 32 ohm stuff,
and it's spec'd to 100MHz+.
That is stupid, the wider you open the window, the more shit you let
in.

Good for marketing......


Martin

 

[Tom Bruhns]

Jim Williams' advice for achieving very low distortion in opamps is
"always invert." The idea here is that even if an opamp may have
screaming bandwidth and good open-loop linearity, it may well have
rotten common-mode performance, especially at high frequencies. And if
neither input pin has any substantial voltage on it, the various
nonlinear parasitic capacitances won't be pumped.

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

We wonder why.

John

Normally I find Jim Williams' advice to be pretty good, but this one I
gotta disagree with. You need to evaluate each op amp carefully to
determine what actually works best in a particular application. "One
size most certainly does NOT fit all."

Even at low frequencies, we've used some tricks to get rid of
distortion from nonlinear capacitive effects, so we can use non-
inverting (and thus high impedance) inputs. Some of the tricks we've
used I've seen built into some especially good low distortion
monolithic op amps.

As a counter-example of a current-feedback amp that shows a graph, at
least, of inverting configuration distortion, check out the OPA695.

I can also give you examples of voltage-feedback amps with very low
distortion at tens of MHz...

Cheers,
Tom

 

[Phil Hobbs]

Jim Williams' advice for achieving very low distortion in opamps is
"always invert." The idea here is that even if an opamp may have
screaming bandwidth and good open-loop linearity, it may well have
rotten common-mode performance, especially at high frequencies. And if
neither input pin has any substantial voltage on it, the various
nonlinear parasitic capacitances won't be pumped.

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

We wonder why.

John

Williams' law is "Always invert, except when you can't." Besides
distortion, this also saves startup problems caused by exceeding
somebody's CM voltage range on power-up. (That one bit me in the days
of my youth, when I wasn't used to op amps whose output swing was wider
than their CM range.)

Cheers,

Phil Hobbs

 

[Fred Bloggs]

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

The non-inverting gain of x2 would be apply to driving a controlled low
impedance transmission line with series termination, as in video and
baseband A/D applications. It would then make sense to publish a graph
pertaining to the most common application. Don't they specify an output
amplitude with that?
And I'm not sure that the non-linear capacitance effect is avoided
entirely by the inverting configuration. The most effective technique is
to keep internal signal swings small.

 

[Vladimir Vassilevsky]

Jim Williams' advice for achieving very low distortion in opamps is
"always invert."

Although it is generally true statement, I'd rather say "it depends".

The idea here is that even if an opamp may have
screaming bandwidth and good open-loop linearity, it may well have
rotten common-mode performance, especially at high frequencies. And if
neither input pin has any substantial voltage on it, the various
nonlinear parasitic capacitances won't be pumped.

The inverting configuration advantage is because of no CM swing in the input
stage. It may or may not be important depending on the topology.

And the best opamps for fast apps are current-mode parts. They have
the speeds and slew rates to get really low (below -60 dB) THDs at
frequencies upwards of 30 MHz or so. But *every* datasheet we've seen
on current-mode amps shows THD specs/curves only in non-inverting
mode, almost always with a gain of +2.

We wonder why.

The optimal feedback resistance in CFB opamps is given for every particular
opamp; you can't choose it arbitrarily. Usually this resistance is from
several hundred Ohm to several KOhm. This is a serious limitation for the
CFB in the inverting configuration; the non-inverting mode is more flexible.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

 

[John Larkin]

Normally I find Jim Williams' advice to be pretty good, but this one I
gotta disagree with. You need to evaluate each op amp carefully to
determine what actually works best in a particular application. "One
size most certainly does NOT fit all."

Even at low frequencies, we've used some tricks to get rid of
distortion from nonlinear capacitive effects, so we can use non-
inverting (and thus high impedance) inputs. Some of the tricks we've
used I've seen built into some especially good low distortion
monolithic op amps.

Tricks? Tell if you can.

As a counter-example of a current-feedback amp that shows a graph, at
least, of inverting configuration distortion, check out the OPA695.

Thanks, that's interesting. THD is actually a bit less in non-inv
mode. Since you get an extra free gain-of-one in that mode, that may
account for a bit of the difference.

John

 

[Winfield Hill]

Even at low frequencies, we've used some tricks to get rid
of distortion from nonlinear capacitive effects, so we can
use non-inverting (and thus high impedance) inputs. Some
of the tricks we've used I've seen built into some especially
good low distortion monolithic op amps.

I assume you're referring to cascode input stages, etc.
Anyway, Tom, we're all ears, or should I say, all eyes.

 

[Tom Bruhns]

I assume you're referring to cascode input stages, etc.
Anyway, Tom, we're all ears, or should I say, all eyes.

Well, I can't give away all our secrets, but one that I think
Analog Devices has actually mentioned in ap notes (Walt Jung...) is
that in a lot of amplifiers, the main parasitic capacitance is to the
negative-rail substrate. Thus, you can bootstrap the negative supply
to follow along with the input terminals and significantly lower
distortion.

Some of the lowest distortion discrete audio circuits I know of were
made by a friend from the distant past. He offered me a couple rules
of thumb: don't let the voltage change across the junctions of the
input device, and don't let the current through the input device
change. Although you must, of course, let them change _some_, you can
reduce the changes a whole lot: run the input not only cascode, but
bootstrapped cascode, where the control element (base; gate) of the
upper device is driven to follow the input voltage. Operate the
cascode into a current source load, and buffer its output with a
follower (or other circuit whose input current change is small
compared with the standing current magnitude). Note that the OPA627
input stage does these things...in addition, it uses a dielectrically
isolated process. It's capable of really low distortion operating as
a follower even when the source impedance is high. It has been a
while since I played with the test circuits, but as I recall, I was
able to get audio-range harmonics well below -100dBc with a 100k ohm
source resistance; that is not easy to do with other monolithic op
amps that I have tried.

But that doesn't help necessarily with high frequency stuff. My
earlier comment about the need to try amps in your application comes
from personal experience about four years ago. We had been using a
hybrid op amp that went obsolete. I spent quite a long time and tried
LOTS of different amps that all looked good on paper, in several
different topologies, to come up with a two-op-amp composite that
would do as good as the old hybrid in that particular application
(requiring very low distortion through about 40MHz). Interestingly,
the output stage amplifier did not do nearly as well when operating at
a gain of 2, as it did at unity gain, but at unity gain, it was
devilish to make stable. And the input stage, operating at pretty
high voltage gain, had very low distortion, but only when its output
was very lightly loaded (thus the buffer output stage).

I'm sure the op amp manufacturers are well aware of the fact that
certain amps shine when the conditions are just right. The OPA847
data sheet shows a couple circuits that have been carefully optimized,
and boy are they good. And that's a voltage-feedback amp.

Also, if you let the manufacturer include the feedback resistors in
the package with the amplifier, you'll generally get better
performance. Have a look at the LTC6400-20. It is particularly easy
to bypass, in my experience. TI also has some high frequency amps
with built-in resistors that set the gain. AD has packages that bring
the feedback in through a second pin, close to the inputs.

Beware how you place bypass caps! Realize that if the output stage is
not class A, there is a lot of harmonic current in the power supply
leads, and if the signal path shares even a very tiny resistance in
common with the current through the bypass caps, the distortion may be
much higher than a more optimum bypass topology.

OK...enough for now. ;-)

Cheers,
Tom

 

[Winfield]

Well, I can't give away all our secrets, but one ...

[ snip various secrets ]

OK...enough for now. ;-)

More, more!

 

[John Larkin]

Well, I can't give away all our secrets, but one that I think
Analog Devices has actually mentioned in ap notes (Walt Jung...) is
that in a lot of amplifiers, the main parasitic capacitance is to the
negative-rail substrate. Thus, you can bootstrap the negative supply
to follow along with the input terminals and significantly lower
distortion.

Some of the lowest distortion discrete audio circuits I know of were
made by a friend from the distant past. He offered me a couple rules
of thumb: don't let the voltage change across the junctions of the
input device, and don't let the current through the input device
change. Although you must, of course, let them change _some_, you can
reduce the changes a whole lot: run the input not only cascode, but
bootstrapped cascode, where the control element (base; gate) of the
upper device is driven to follow the input voltage. Operate the
cascode into a current source load, and buffer its output with a
follower (or other circuit whose input current change is small
compared with the standing current magnitude). Note that the OPA627
input stage does these things...in addition, it uses a dielectrically
isolated process. It's capable of really low distortion operating as
a follower even when the source impedance is high. It has been a
while since I played with the test circuits, but as I recall, I was
able to get audio-range harmonics well below -100dBc with a 100k ohm
source resistance; that is not easy to do with other monolithic op
amps that I have tried.

But that doesn't help necessarily with high frequency stuff. My
earlier comment about the need to try amps in your application comes
from personal experience about four years ago. We had been using a
hybrid op amp that went obsolete. I spent quite a long time and tried
LOTS of different amps that all looked good on paper, in several
different topologies, to come up with a two-op-amp composite that
would do as good as the old hybrid in that particular application
(requiring very low distortion through about 40MHz). Interestingly,
the output stage amplifier did not do nearly as well when operating at
a gain of 2, as it did at unity gain, but at unity gain, it was
devilish to make stable. And the input stage, operating at pretty
high voltage gain, had very low distortion, but only when its output
was very lightly loaded (thus the buffer output stage).

I'm sure the op amp manufacturers are well aware of the fact that
certain amps shine when the conditions are just right. The OPA847
data sheet shows a couple circuits that have been carefully optimized,
and boy are they good. And that's a voltage-feedback amp.

Nice part, but a little more slew rate would be nice.

Also, if you let the manufacturer include the feedback resistors in
the package with the amplifier, you'll generally get better
performance. Have a look at the LTC6400-20. It is particularly easy
to bypass, in my experience. TI also has some high frequency amps
with built-in resistors that set the gain. AD has packages that bring
the feedback in through a second pin, close to the inputs.

Beware how you place bypass caps! Realize that if the output stage is
not class A, there is a lot of harmonic current in the power supply
leads, and if the signal path shares even a very tiny resistance in
common with the current through the bypass caps, the distortion may be
much higher than a more optimum bypass topology.

Yup; a lot of opamps have nearly 0 dB of PSRR as the frequency goes
up. If you bead+cap filter their supplies, to avoid crosstalk from
other gadgets, you open up another bag-o-worms.

What we have is a differential-current-output DAC that has to be
converted to a single-ended signal; a discrete LC lowpass filter; and
an output amplifier. The target is 20 volts p-p out (10 p-p into 50
ohms), 30 MHz bandwidth, low distortion (goal -60 dB, under
unspecified circumstances), flat to 30 MHz (for certain values of
"flat") and good DC accuracy. The interesting things we've noted...

There's no obviously-best way to convert the DAC outputs to
single-ended. The DAC datasheets usually just use transformers. The
new gen of ADC-driver difference amps don't swing much if you use just
one of their outputs.

LC filter design has always been a PITA, and still is.

There aren't many opamps that will swing this far at this bandwidth,
much less into a hefty load. At higher output currents and big swings,
all sorts of things go to hell, especially the GBW of the output
transistors.

The output stage may wind up being a closed-loop composite, which
gives me bad dreams at these sorts of speeds.

If you look at a variety of name-brand ARBS and RF generators, output
swings like this are rare, whereas even at low signal levels THD
numbers can be shocking. ARBS typically run -30 or -40 dB THD, and we
have one late-model RF signal generator that runs about 1.5%
distortion!

Thanks for the tips.

After this is wrapped up, we may try it at 100 MHz.


John

 

[john jardine]

[...]

an output amplifier. The target is 20 volts p-p out (10 p-p into 50
ohms), 30 MHz bandwidth, low distortion (goal -60 dB, under
unspecified circumstances), flat to 30 MHz (for certain values of
"flat") and good DC accuracy. The interesting things we've noted...

[...]

The old Wavetek 166 function generator, runs 50MHz, 15Vpp, into 50ohm.
They use 19 of those transistor thingies in a conventional two stage
arrangement. Their #178 'waveform synth' does 50MHz at 10Vpp into 50ohms and
benefits from an unusual output amp' design.
I have their #145 and # 164, 30Vpp at 20MHz and 20Vpp at 30Mhz and know the
bare-bone output amp's run flat and at less than 1% distortion but would
have thought 0.1%! was decidedly uneconomic, given the reducing choice of
R.F. power transistors.

 

[John Larkin]

[...]

an output amplifier. The target is 20 volts p-p out (10 p-p into 50
ohms), 30 MHz bandwidth, low distortion (goal -60 dB, under
unspecified circumstances), flat to 30 MHz (for certain values of
"flat") and good DC accuracy. The interesting things we've noted...

[...]

The old Wavetek 166 function generator, runs 50MHz, 15Vpp, into 50ohm.
They use 19 of those transistor thingies in a conventional two stage
arrangement. Their #178 'waveform synth' does 50MHz at 10Vpp into 50ohms and
benefits from an unusual output amp' design.
I have their #145 and # 164, 30Vpp at 20MHz and 20Vpp at 30Mhz and know the
bare-bone output amp's run flat and at less than 1% distortion but would
have thought 0.1%! was decidedly uneconomic, given the reducing choice of
R.F. power transistors.

We need to do this with ICs, because we want 8 ARB channels (with
AM/FM/PM/PWM/Noise/Summing on a 6U VME card. It looks possible, but
optimizing the distortion looks to be a labor-intensive procedure.

The other problem is measurement: our spiffy new spectrum analyzer has
harmonic distortion spikes about 65 dB down, so we'll probably have to
feed it through highpass or notch filters or something. I *knew* there
was a reason I've spent my life avoiding RF designs.


John

 

[Fred Bloggs]

Nice part, but a little more slew rate would be nice.




Yup; a lot of opamps have nearly 0 dB of PSRR as the frequency goes
up. If you bead+cap filter their supplies, to avoid crosstalk from
other gadgets, you open up another bag-o-worms.

What we have is a differential-current-output DAC that has to be
converted to a single-ended signal; a discrete LC lowpass filter; and
an output amplifier. The target is 20 volts p-p out (10 p-p into 50
ohms), 30 MHz bandwidth, low distortion (goal -60 dB, under
unspecified circumstances), flat to 30 MHz (for certain values of
"flat") and good DC accuracy. The interesting things we've noted...

There's no obviously-best way to convert the DAC outputs to
single-ended. The DAC datasheets usually just use transformers. The
new gen of ADC-driver difference amps don't swing much if you use just
one of their outputs.

LC filter design has always been a PITA, and still is.

There aren't many opamps that will swing this far at this bandwidth,
much less into a hefty load. At higher output currents and big swings,
all sorts of things go to hell, especially the GBW of the output
transistors.

The output stage may wind up being a closed-loop composite, which
gives me bad dreams at these sorts of speeds.

If you look at a variety of name-brand ARBS and RF generators, output
swings like this are rare, whereas even at low signal levels THD
numbers can be shocking. ARBS typically run -30 or -40 dB THD, and we
have one late-model RF signal generator that runs about 1.5%
distortion!

Thanks for the tips.

After this is wrapped up, we may try it at 100 MHz.


John

MCL had a high frequency hybrid that would easily meet that as OIP3/2 at
your 24dBm, it used two amps differentially to eliminated the 3rd order
stuff very well and it was a power amp. Problem: the transformer
coupling high passed in at 5MHz or something. The little board was $40
at the time I used them.