Would an integrated amplifier have more high order harmonics than a discrete amp

[ELS122]

I learned that adding negative feedback decreases low order harmonics, but it also boosts high order harmonics (5th and 6th), up to a point.
And since op amp based amps rely on negative feedback to control gain. Would that make them inherently more distortive in the high order harmonics?
I guess it's like comparing apples to oranges. But does it?

[TimFox]

The effect of negative feedback reducing lower-order distortion but increasing higher-order distortion is well known.
A good description can be found in articles by Norman Crowhurst, who analyzed various circuits using push-pull 5881s.
An example from my library that you can find at  http://www.tubebooks.org/Books/crowhurst_basic_3.pdf
But, what are you comparing to the op-amp-based amplifier?
Most solid-state AB amplifiers use very similar circuits to integrated op-amps, with substantial negative feedback to obtain reasonable distortion and gain.
Typically, one starts with a differential input, follows with appropriate gain stages, continues with power output stage (often a glorified emitter follower pair), and applies a frequency compensation network somewhere in the middle, just like in an op amp.
A single-ended triode amplifier without negative feedback will have substantial 2nd-order distortion, which some people like.

[langwadt]

The effect of negative feedback reducing lower-order distortion but increasing higher-order distortion is well known.
A good description can be found in articles by Norman Crowhurst, who analyzed various circuits using push-pull 5881s.
An example from my library that you can find at  http://www.tubebooks.org/Books/crowhurst_basic_3.pdf
But, what are you comparing to the op-amp-based amplifier?
Most solid-state AB amplifiers use very similar circuits to integrated op-amps, with substantial negative feedback to obtain reasonable distortion and gain.
Typically, one starts with a differential input, follows with appropriate gain stages, continues with power output stage (often a glorified emitter follower pair), and applies a frequency compensation network somewhere in the middle, just like in an op amp.
A single-ended triode amplifier without negative feedback will have substantial 2nd-order distortion, which some people like.

asymmetric even, symmetric odd

[ELS122]

The effect of negative feedback reducing lower-order distortion but increasing higher-order distortion is well known.
A good description can be found in articles by Norman Crowhurst, who analyzed various circuits using push-pull 5881s.
An example from my library that you can find at  http://www.tubebooks.org/Books/crowhurst_basic_3.pdf
But, what are you comparing to the op-amp-based amplifier?
Most solid-state AB amplifiers use very similar circuits to integrated op-amps, with substantial negative feedback to obtain reasonable distortion and gain.
Typically, one starts with a differential input, follows with appropriate gain stages, continues with power output stage (often a glorified emitter follower pair), and applies a frequency compensation network somewhere in the middle, just like in an op amp.
A single-ended triode amplifier without negative feedback will have substantial 2nd-order distortion, which some people like.

Well an amplifier where the preamp is all op amps vs a preamp that is discrete.

[TimFox]

A simple discrete preamplifier might contain two CE stages followed by a CC (follower) stage, with feedback from the output emitter to the input emitter.
Without feedback, it will have predominantly 2nd order distortion at reasonable levels (due to asymmetry, as mentioned above), with 3rd order predominant at high levels (it grows faster than the 2nd order).
So, quantitatively, there will be a difference from an op amp with differential stages, where the 2nd order distortion is reduced by the symmetry, before the feedback is applied.
(Of course, real op amps often have nasty crossover distortion in the output stage, which should be avoided in the design.)

The interesting question, which involves perception as well as measurement, is why high-order distortion products are "nasty", as commonly believed.
My personal opinion, based on various readings and measurements on RF systems, is that the odd-order distortion is bad, since when you take a real musical waveform (with a very complex spectrum) and mess it up with 3rd-order intermodulation, distortion products occur between the fundamental frequencies of the musical notes and sound like "mud" (to use a technical term).
Harmonics themselves are relatively innocuous, since the musical notes are not pure tones (e.g., 440 Hz), but the fundamental and many harmonics at integer multiples thereof.
Think of a solo flute, whose notes are rich in harmonics, and the relative magnitude of the harmonics is under some control of the skilled player.
IM and THD are measurement techniques, but both distortions result from the same non-linear behavior of the amplifier under test.

I once mentioned in an audio forum that third-order distortion was bad, and someone agreed, since he had looked up the 3rd harmonic (1320 Hz) of 440 Hz and couldn't find it in a table of equal-tempered scale frequencies, the closest being a tempered E at 1318.51 Hz.  Of course, this is because the equal-tempered scale is an engineering approximation applied to music to fit everything into the black and white keys on a piano.

[ELS122]

Well you could just make the discrete amplifier simple without much negative feedback, and double it up for the other phase, like is done in a lot of studio gear.
That would cancel all even harmonics, if you want you could change the gain of one of the sides to bring back some even harmonics for a nicer sound.
And since there's minimal negative feedback, the high order harmonics should be that high, if the design is descent in the first place.
Then all you're really left with is the 3rd harmonic.

While with a standard op amp design, even if you double it up, it relies on negative feedback throughout, so I'm asking if that would add a lot more high order harmonics than the previous example of a discrete design with little feedback.
So the 5th and the 7th harmonic being of concern in that case, if you make it balanced.

[TimFox]

That is a quantitative question, and needs to be calculated, simulated, or measured.
General statement:  a push-pull or differential amplifier, properly balanced, will only have odd-order distortion.
It's late tonight;  I shall look at my best reference for harmonics produced by bjts (as a function of signal) and send you more information in the next couple of days.
(Short answer:  a bjt without feedback makes a lot of harmonics, and the harmonic level increases as a function of drive until clipping is reached.)

[strawberry]

While with a standard op amp design, even if you double it up, it relies on negative feedback throughout, so I'm asking if that would add a lot more high order harmonics than the previous example of a discrete design with little feedback.
So the 5th and the 7th harmonic being of concern in that case, if you make it balanced.

if someone can hear 7TH harmonic witch usually should be below -80dB then something is wrong indeed
amplifiers dont amplify harmonics but they generate them if there is lack of negative feedback
sound is harmonics not 1 harmonic. that 7th harmonic will apply to all recorded harmonics

[ELS122]

(Of course, real op amps often have nasty crossover distortion in the output stage, which should be avoided in the design.)

So you should bias the op amp away from the crossover region and pull the output externally towards one power rail to make it run in class A?
But I haven't seen a single preamp do this...

Looking into op amp preamps got me into a deep rabbit hole... for example this circuit:


Which I'm still trying to wrap my head around 
And also composite op amp circuits with high open loop gain leading to THD levels in the -180db range... a bit op.

But I'm still curious if an op amp circuit with low open loop gain would have inherently high high-order harmonics because of the negative feedback, iirc the high order harmonics peak around 20db of negative feedback, more than that and they decrease.

[ELS122]

I found the chart I mentioned:


I also found this quote:
"It's easy to identify an amplifier that works with a lot of feedback to cover up its basic flaws: it has a distortion curve that rises with frequency and has foolishly low output source impedances (high damping factors), as most Crown amplifiers do"
So I guess they would have inherently more high order harmonics, relative to the low order ones at least.
And the higher open gain op amps are used, the high order harmonics would be more relative to low order. while the total harmonic distortion would be a lot less.
Right?

I guess we shouldn't be so negative all the time

[langwadt]

a design from the 90's? that claimed superiority with not feedback, http://web.archive.org/web/20000305114233/www.lcaudio.dk/milldia.pdf

[Circlotron]

You might be interested in nested differentiating feedback loops.

Part 1 https://www.linearaudio.net/sites/linearaudio.net/files/cherry%20ndfl.pdf

Part 2 https://linearaudio.nl/sites/linearaudio.net/files/cherry%20ndfl%20amp.pdf

[David Hess]

"It's easy to identify an amplifier that works with a lot of feedback to cover up its basic flaws: it has a distortion curve that rises with frequency and has foolishly low output source impedances (high damping factors), as most Crown amplifiers do"

So I guess they would have inherently more high order harmonics, relative to the low order ones at least.
And the higher open gain op amps are used, the high order harmonics would be more relative to low order. while the total harmonic distortion would be a lot less.

The discussion here is consistent with that statement:

https://www.diyaudio.com/community/threads/thd-spectrum-and-negative-feedback.179416/

What is going on is that feedback increases higher order distortion *when it is used to try and make up for extremely non-linear intermediate stages*, like class-b stages with significant dead zones yielding crossover distortion.  Otherwise increasing negative feedback reduces all distortion within its passband.  So a properly biased class-ab amplifier, or at least an amplifier which lacks dead zones in its transfer curve, has reduced distortion at all frequencies as the feedback is increased.

As a practical matter, careful power amplifier designs with feedback can yield better than 0.01%, or even 0.001%, total harmonic distortion.  Careful small signal designs can yield distortion levels on the order of parts-per-billion, which is what is required to test 24-bit ADCs, but this requires using composite operational amplifiers and common mode suppression throughout the signal chain; the best operational amplifiers alone cannot achieve this.  Check out page 62 of Linear Technology application note 67 for this type of design.

[David Hess]

You might be interested in nested differentiating feedback loops.

Part 1 https://www.linearaudio.net/sites/linearaudio.net/files/cherry%20ndfl.pdf

Part 2 https://linearaudio.nl/sites/linearaudio.net/files/cherry%20ndfl%20amp.pdf

I think the explanation is simpler than he gives.  Local feedback is necessary to control gain and phase of a stage when that stage has a gain and phase which varies significantly with operating point.  You can see that in the above example that I show where each of the additional operational amplifiers has its own feedback network giving each stage a very predictable gain and phase response.

Without that local feedback, gain and phase response within the global feedback loop can vary wildly with operating point making frequency compensation impossibly difficult.  Then dominant mode compensation has to be used which severely limits bandwidth.  Dominant mode compensation has its place where bandwidth is low anyway, like with voltage regulators where a large amount of output capacitance is present.

[TimFox]

On the audibility of 7th harmonics:
As I said above, harmonics themselves are not the problem, since any musical note (from a real instrument) is already rich in harmonics.
However, the same mechanisms that produce a 7th harmonic tone from a clean sine wave input also produce 7th-order intermodulation, which is not up in the dog-hearing range.
For example, assume two test tones at 1000 and 1100 Hz.
The 2nd-order IM products are at 2100 Hz and 100 Hz, along with the 2nd harmonics at 2000 and 2200 Hz, all far from the fundamentals.
The 3rd-order products are at 900 and 1200 Hz, near the fundamentals, along with the 3rd harmonics at 3000 and 3300 Hz.
The 7th order products are more numerous, and include 700 and 1400 Hz, which are (4x1000)-(3x1100) and (4x1100)-(3x1000), respectively.
Also: 2800, 3500, 4900, and 5600 Hz to clutter further the spectral landscape.
THD and IM are practicable measurements, and correlate to amplifier linearity.

[David Hess]

The role of intermodulation distortion is under appreciated.  It is what really matters.

What often does not matter is distortion, including intermodulation distortion, in the amplifier because the speakers contribute considerably to all types of distortion.  The chief advantage of bass reflex speaker designs is that minimizing cone movement at low frequencies also minimizes intermodulation distortion.  Horn speakers have the same advantage for the same reason.

[TimFox]

The point made above that applying negative feedback to a seriously non-linear amplifier is not a good thing is quite valid.
In extreme cases, feedback does nothing good when the amplifier clips or encounters the dead zone in crossover distortion.
In a less extreme, yet common, case when a traditional op amp produces an output still below the slew-rate limit, the input stage is pushing its maximum current into the compensation capacitor, and the differential amplifier enters non-linear operation, even though the output has not yet reached its maximum dV/dt.
What can one do?
Op-amp solution:  use a good op amp, with decent linearity, not necessarily good DC performance, such as the (older) 5532 and 5534, or the (last-gasp of National) LM4562 and LME49720 that were designed for audio.
Discrete solution:  for a preamp, you maybe need 1 V rms "wiggle" around the quiescent point.  Therefore, use a higher supply voltage than practical for an integrated amplifier so that the AC component is much less than the available range, perhaps 80 VDC (> 70 V pk-pk range).  Discrete transistors are available with higher voltage ratings.  Individual stages can use local feedback:  emitter resistor (series feedback) or collector-base resistor (shunt feedback) to linearize each stage, before applying overall feedback.
Caveat:  for a preamplifier, distortion is not the only problem.  Adding an emitter resistor, for example, will increase the noise of the first stage, which may be important for low-level inputs (e.g., phono).

[magic]

You might be interested in nested differentiating feedback loops.

Part 1 https://www.linearaudio.net/sites/linearaudio.net/files/cherry%20ndfl.pdf

Part 2 https://linearaudio.nl/sites/linearaudio.net/files/cherry%20ndfl%20amp.pdf

I find Cherry's writings quite messy and confusing. Do you happen to understand what's the supposed connection between the theory given in part 1 and the amplifier described in part 2?

It just doesn't seem to check out if you take it at face value and simply look at the voltage gain of each stage. There is a few inverting integrators, where noise gain (relevant to stability of the local loop) is not the same as signal gain (relevant to stability of the global loop), and frankly the signal gain only applies when the stage is considered together with transconductance of the preceding stage. I think one would have to get fairly liberal about what to accept as "gain" (V→V, I→V, V→I, I→I) and/or what to accept as "stage" (one transistor/pair with its load, a sequence of transistors with their local feedback loops) in order to fit this circuit into the theoretical framework... Maybe that was the point?

I like the clear cut simplicity of IC design texts, where gain is usually understood as voltage gain and for each stage you divide collector voltage by base voltage and that's pretty much it.


And I don't see how in this amplifier Cherry makes attempt to keep constant gain over audio band and place individual stage poles exactly near 20kHz. Yet somehow it is shown that way in the theoretical part.

[ELS122]

FYI, I found this in a datasheet for a TDA2030..:

Transient inter-modulation distortion(TIM)
Transient inter-modulation distortion is an unfortunate phenomena associated with negative-feedback amplifiers.
When a feedback amplifier receives an input signal which rises very steeply, i.e. contains high-frequency
components, the feedback can arrive too late so that the amplifiers overloads and a burst of inter-modulation
distortion will be produced as in Fig.22.Since transients occur frequently in music this obviously a problem for the
designed of audio amplifiers. Unfortunately, heavy negative feedback is frequency used to reduce the total
harmonic distortion of an amplifier, which tends to aggravate the transient inter-modulation(TIM situation.)The best
known method for the measurement of TIM consicts of feeding sine waves superimposed onto square wavers,into the
amplifier under test.The output spectrum is then examined using a spectrum analyser and compared to the
input.This method suffers from serious disadvantages:the accuracy is limited, the measurement is a tather delicate
operation and an expensive spectrum analyser is essential.A new approach (see Technical Note 143(Applied by
SGS to monolithic amplifiers measurement is fast cheap,it requires nothing more sophisticated than an
oscilloscope-and sensitive-and it can be used down to the values as low as 0.002% in high power amplifiers.
The "inverting-sawtooth" method of measurement is based on the response of an amplifier to a 20KHz sawtooth
waveform.The amplifier has no difficulty following the slow ramp but it cannot follow the fast edge.The output will
follow the upper line in Fig.23 cutting of the shade area and thus increasing the mean level.If this output signal is
filtered to remove the sawtooth,direct voltage remains which indicates the amount of tIM distortion, although it is
difficult to measure because it is indistingishable from the DC offset of the amplifier.This problem os neatly avoided
in the IS-TIM method by periodically inverting the sawtooth waveform at a low audio frequency as shown in
Fig.24.Inthe case of the sawtooth in Fig. 25 the means level was increased by the TIM distortion, for a sawtooth in
the other direction the opposite is ture

Is it the input circuitry slew rate the problem in the op amp itself?
Would setting up the amp as a non-inverting amp fix the problem?

Is the thread topic related to this? If the negative feedback sort of "adds on" the slew rate of the amp itself, I can imagine why it would decrease low order harmonics more than high order ones

[TimFox]

In real feedback amplifiers, including the interior of integrated op amps, there is a network to control the high-frequency phase and amplitude response so that the amplifier is stable when the feedback is closed (doesn't oscillate).
In simple op amps, this is usually a capacitor providing local feedback around the second state, typically a current-in (from a current mirror), voltage-out (to an output follower), so that the phase shift is less than 180^o when the voltage gain has fallen to unity.
Since the input stage has a finite output-current capability, that current through the compensation capacitor imposes a slew-rate limit on the device.
If a transient drives the amplifier into slew-rate limit, then the feedback cannot work, and the amplifier may take a finite time to recover from that:  this will definitely distort the reasonable musical signal that you are trying to amplifier.
One way to avoid this is to limit the slew rate of the signal before the input to the feedback amplifier, with a suitable low-pass filter, while still allowing the frequency range of interest to be amplified.
TIM is distinguished from IMD since the former is unwanted modulation of the desired signal by a useless transient, while the latter is an unwanted interaction between signals within the passband.

[David Hess]

Is it the input circuitry slew rate the problem in the op amp itself?

TimFox gives a general description of the problem when it involves the slew rate limit, aka full power bandwidth, however it does also apply to the input stage which has its own slew rate limit of a different kind.  During large common mode swings, the input stage can be driven into cutoff or saturation.  This would produce the problem described, however I am only aware of this affecting comparators where fast common mode swings are more likely.

Naively designed audio power amplifiers have problems meeting the required full power bandwidth (slew rate) requirements.  Designs feature things like decompensation and transconductance reduction in order to improve the output slew rate.

This is also why the feedback network used to control frequency compensation should *not* be used to control output bandwidth; doing so reduces the slew rate.  This might be where the advice about not using too much negative feedback comes from.

Would setting up the amp as a non-inverting amp fix the problem?

It would if it was the input stage which is slew rate limiting, but this would not solve the problem described by TimFox.

[TimFox]

Technically, the common-mode effect described by David Hess is not a slew-rate limit (dV/dt) but common-mode-induced distortion, which is another real thing.
It depends more on the actual voltages at the input stage, not their time derivative.
That effect can be avoided by using an inverting configuration, which should have no common-mode input voltage, since one terminal is grounded.
Preamplifiers usually avoid the inverting connection (shunt feedback resistor and an input resistor) since the input resistor adds noise to the signal before the amplifier.
An interesting variation sometimes used with transducers such as phono cartridges is to do the negative feedback with a low-resistance voltage divider (like the ones normally used for non-inverting amplifiers), connecting the (ungrounded) transducer between the divider and the inverting input, with the non-inverting input grounded.
David Hess' suggestions of decompensation and transconductance reduction (such as degeneration resistors in series with both emitters of an input long-tailed pair) are useful to improve the slew rate problem.
Also, controlling the output bandwidth with the overall feedback network's capacitor is much better than overcompensating the open-loop gain of the op amp to limit the high-frequency response.

Another cause of common-mode distortion at the input is the Early effect in the input BJTs.
The input base current of a BJT is a function of the base-collector voltage, which in an op-amp or similar circuit varies with the base input voltage.
If there is a substantial source impedance (such as a MM phono cartridge), this real current variation flows through the source impedance and induces a voltage at the input node that is really there, so linearizing the amplifier after that point does not remove the distortion from that non-linear current.
This can be avoided either with an inverting input (with its other problems) or bootstrapping the collector of the input transistor(s) (cascode) to a second transistor (or pair), whose base(s) is(are) driven through a Zener or capacitor from the emitter(s) of the first transistor(s).