Hi all,
This question pertains just as well I presume to all types of amps be they SS, tube or hybrids but I am posting it here in Tubes because this is what interests me more. I've never given this issue much thought up until now but I would like to know from the seasoned veterans in what ways if at all possible, we can design for low IM distortion.
I understand that when designing an amp we can exert some sort of control on the THD products of the circuit by the choice of the load line, bias point, number of stages weather the amp is SE or PP etc. I am talking open loop here and I'm excluding the use of NFB since (at least where THD is concerned) application of feedback alone has a great effect in reducing distortion.
What I would like to know is if there are similar ways that can be used to effect a desired IM distortion profile to a particular design. In other words, what are the mechanisms that govern how much IM distortion is there in a particular circuit? If someone wanted to design an amp where the overriding concern above all else was low IM distortion how would a good designer go about his business? I hope this is not too much of a loaded question.
This question pertains just as well I presume to all types of amps be they SS, tube or hybrids but I am posting it here in Tubes because this is what interests me more. I've never given this issue much thought up until now but I would like to know from the seasoned veterans in what ways if at all possible, we can design for low IM distortion.
I understand that when designing an amp we can exert some sort of control on the THD products of the circuit by the choice of the load line, bias point, number of stages weather the amp is SE or PP etc. I am talking open loop here and I'm excluding the use of NFB since (at least where THD is concerned) application of feedback alone has a great effect in reducing distortion.
What I would like to know is if there are similar ways that can be used to effect a desired IM distortion profile to a particular design. In other words, what are the mechanisms that govern how much IM distortion is there in a particular circuit? If someone wanted to design an amp where the overriding concern above all else was low IM distortion how would a good designer go about his business? I hope this is not too much of a loaded question.
Last edited:
IMD is pretty much caused by the same mechanisms that cause THD
it is possible to construct a circuit with low THD but generates high IMD - but the methods are not much used in amps intended to be linear
Prof Cherry gave his 3 amplifier distortion reduction options:
reduce signal to bias ratio - use less of the nonlinear curve by wasting lots of power in heavy bias
cancellation - as in even order distortion reduction in well balanced diff pair, or push-pull output
apply negative feedback, locally as in degeneration or more globally - and in most situations global feedback can be shown to be more effective when not at least equal to local feedback in distortion reduction
selecting "more linear" amplifying devices is often suggested but somewhat dubious - we basically have the choice of "square law" or exponential gm - and hugely more gain in the exponential gm devices give more linear results with local feedback that reduce them to the same gain as what are often touted as "more linear" devices
it is possible to construct a circuit with low THD but generates high IMD - but the methods are not much used in amps intended to be linear
Prof Cherry gave his 3 amplifier distortion reduction options:
reduce signal to bias ratio - use less of the nonlinear curve by wasting lots of power in heavy bias
cancellation - as in even order distortion reduction in well balanced diff pair, or push-pull output
apply negative feedback, locally as in degeneration or more globally - and in most situations global feedback can be shown to be more effective when not at least equal to local feedback in distortion reduction
selecting "more linear" amplifying devices is often suggested but somewhat dubious - we basically have the choice of "square law" or exponential gm - and hugely more gain in the exponential gm devices give more linear results with local feedback that reduce them to the same gain as what are often touted as "more linear" devices
You get rid of IMD the same way you get rid of THD: improve linearity. Once you have a linear design, the only other way to reduce IMD is by not dirtying up the signal chain by using good star grounding techniques to keep ripple harmonics out of the signal chain, by isolating heater wiring by twisting the leads and keeping them well away from low level signal lines, by shielding low level lines, and the gNFB feedback connections to the summing node (I like RF coax for this, with the shield braid grounded at the summing node, but not at the other end, so that it acts as a Faraday cage to minimize elctrostatic coupling. You could also include a Zobel at the output to reduce the RF that rides in on the speaker leads. If you use DC heater power, make certain you filter it very well.
The less unnecessary frequencies you introduce, the fewer frequencies there are to mix with desired frequencies to make IMD distortion products.
The less unnecessary frequencies you introduce, the fewer frequencies there are to mix with desired frequencies to make IMD distortion products.
For reduced bandwidth to work you would have to filter both input to and output from the non-linear section. Even then you will still get odd-order products in band.
You just need to be careful that the cable capacitance in parallel with the summing node does not disturb the frequency response or degrade HF stability. Given that the feedback pickup point is usually low impedance a simple homemade twisted pair should be good enough. To eliminate induction, the ground wire of the pair can double as the ground reference for the output - this means that the output signal seen by the load and the output signal seen by the feedback loop are the same and referenced to the same ground as the input signal.Miles Prower said:
where's your evidence for that claim? - are you talking a "straw man" example of poorly applied negative feedback - it really can't help an cirucit with deadzone or hysterisis - the circuit must be able to drive the load without "gaps" before feedback
if a circuit has smooth nonlinearities, sufficient slew rate, dynamic capabilities then high (global) negative feedback reduces all orders of distortion
if a circuit has smooth nonlinearities, sufficient slew rate, dynamic capabilities then high (global) negative feedback reduces all orders of distortion
THD and IMD may exist at different frequencies. By definition, the harmonics in THD are higher in frequency than the fundamental. IMD may be higher or lower in frequency. This can create a difference if the circuit has different gain or different feedback at different frequencies.
By IMD I mean intermodulation between signal components. You can also get IM between signal and other things such as hum, RF, PSU noise etc. If this gets lumped in with normal IMD then the result is confusion, as the cause and hence solution is different. To get rid of normal IMD you improve linearity. To get rid of PSU IM you improve PSRR.
By IMD I mean intermodulation between signal components. You can also get IM between signal and other things such as hum, RF, PSU noise etc. If this gets lumped in with normal IMD then the result is confusion, as the cause and hence solution is different. To get rid of normal IMD you improve linearity. To get rid of PSU IM you improve PSRR.
THD is a poor choice as a comparison base - a 1% THD from pure 2nd order nonlinearity will have fewer IMD products than "the same THD" that is pure 3rd order distortion
but once you have good basic circuit linearity over the full load and signal range then high negative feedback can reduce distortion, including IMD
in "5-pin" op amp topology circuits psrr will be better than or at least equal excess loop gain == high feedback improves psrr
amps with gnd reference available to internal stages can have psrr much higher than loop gain
a nonlinearity not readily addressed by negative feedback is input common mode nonlinearity in positive gain amps
but once you have good basic circuit linearity over the full load and signal range then high negative feedback can reduce distortion, including IMD
in "5-pin" op amp topology circuits psrr will be better than or at least equal excess loop gain == high feedback improves psrr
amps with gnd reference available to internal stages can have psrr much higher than loop gain
a nonlinearity not readily addressed by negative feedback is input common mode nonlinearity in positive gain amps
>For reduced bandwidth to work you would have to filter both input to and output from the non-linear section. Even then you will still get odd-order products in band.
Output filtering might help, but it isn't necessary to get results. If an amplifier is only playing from 6kHz and up, there will not be any 1kHz+9kHz intermodulation products, for example.
The harmonic distortion isn't helped from this other than the effect of reduced per-band power. Most amplifers have reduced distortion at a couple watts compared to half or full power, especially at high frequencies.
Output filtering might help, but it isn't necessary to get results. If an amplifier is only playing from 6kHz and up, there will not be any 1kHz+9kHz intermodulation products, for example.
The harmonic distortion isn't helped from this other than the effect of reduced per-band power. Most amplifers have reduced distortion at a couple watts compared to half or full power, especially at high frequencies.
They never are. Any sort of balanced topology is going to null out the even harmonics. Getting rid of those harmonics reduces the THD. However, it does nothing if the whole thing is nonlinear enough to generate lots of IMD. That's basically what happens with a doubly balanced mixer: the balance nulls out undesirable harmonics of the carrier and local oscillator, but leaves the upper and lower sidebands.
You would see the same effect with an LTP splitter that used nastily nonlinear VTs (like 12AV7s or 6SF5s). No even order harmonics that would make the THD figure look pretty good, but it would also produce the sidebands of IMD, leaving you to wonder why it sounded not-so-good.
One thing I have found that really helps is to keep noise out of the power supply. Most of the amps I build have a separate power transformer and supply for the driver section. So no matter what is happening in the output section, there will be no voltage pertubations in the driver power supply.
I'm not completely clear on this, but I suspect that the source of distortion (like nonlineaity) will produce more THD and IM, but feedback loops can only reduce THD. But the fed back IM just introduces additional crud that isn't subtracted out or reduced. As it looks like part of the input signal when the feedback signal is merged with the input. That it produces yet more IM products, but at lower levels.
If the above is even vaguely correct (!), it would seem that the best approach to amp design is to create the best amp you can with no feedback loops, and only after that, use a little feedback to make it more perfect. IOW, don't expect to clean up a crappy design with feedback.
For more fun, consider that the human ear seems to like a small dose of 2nd harmonic, around 45dB down. But be careful that you don't induce too much IM, which the ear hates.
If the above is even vaguely correct (!), it would seem that the best approach to amp design is to create the best amp you can with no feedback loops, and only after that, use a little feedback to make it more perfect. IOW, don't expect to clean up a crappy design with feedback.
For more fun, consider that the human ear seems to like a small dose of 2nd harmonic, around 45dB down. But be careful that you don't induce too much IM, which the ear hates.
fortunately the above is not vaguely correct
it is a popular and heavily promoted misunderstanding http://www.diyaudio.com/forums/solid-state/94676-bob-cordell-interview-negative-feedback.html
high negative feedback reduces IM and PIM, AM, THD phase shift - any measurable errors - when "correctly" applied to suitable circuits high loop gain linearizes the early in the signal chain circuit stages by reducing the amplitude of the signal needed to move the output
I do wonder at the recent "IM" interest - has there been some popular "audiophile" rag/blog/pundit pontificating on IMD?
pretty much most of the "popular" "audiophile press" has made an industry of deprecating “negative feedback amplifiers” and promoting tweak “High End” concepts that actual EE and Audio engineers find laughable
sorry if it sounds harsh but if you’re binging the ideas you see in most audiophile rags here you need to recognize that you have been severely misserved by that marketing press
it is a popular and heavily promoted misunderstanding http://www.diyaudio.com/forums/solid-state/94676-bob-cordell-interview-negative-feedback.html
high negative feedback reduces IM and PIM, AM, THD phase shift - any measurable errors - when "correctly" applied to suitable circuits high loop gain linearizes the early in the signal chain circuit stages by reducing the amplitude of the signal needed to move the output
I do wonder at the recent "IM" interest - has there been some popular "audiophile" rag/blog/pundit pontificating on IMD?
pretty much most of the "popular" "audiophile press" has made an industry of deprecating “negative feedback amplifiers” and promoting tweak “High End” concepts that actual EE and Audio engineers find laughable
sorry if it sounds harsh but if you’re binging the ideas you see in most audiophile rags here you need to recognize that you have been severely misserved by that marketing press
Last edited:
Hi,
here you are incorrect: THD and IM are the same insofar that signals of certain frequencies are present at the output which were not there at the input. By feeding those signals back to a second --inverting-- input their amplitude can indeed be reduced. This is NFB.
However, what have we done?: We are now applying more signals of different frequencies at once to the input(s) of our amplifier -- which lead the way for an even more diverse spectrum of new THD and IM product frequencies to be generated. But, their amplitude will be far lower because of the summing action of the original input and the inverting NFB input of the amplifier.
Thus we have to choose between having a few "alien" frequencies which may be large in magnitude (low NFB case) or a plethora of alien frequencies which are all of tiny amplitude. For the ear to decide...
Kenneth
here you are incorrect: THD and IM are the same insofar that signals of certain frequencies are present at the output which were not there at the input. By feeding those signals back to a second --inverting-- input their amplitude can indeed be reduced. This is NFB.
However, what have we done?: We are now applying more signals of different frequencies at once to the input(s) of our amplifier -- which lead the way for an even more diverse spectrum of new THD and IM product frequencies to be generated. But, their amplitude will be far lower because of the summing action of the original input and the inverting NFB input of the amplifier.
Thus we have to choose between having a few "alien" frequencies which may be large in magnitude (low NFB case) or a plethora of alien frequencies which are all of tiny amplitude. For the ear to decide...
Kenneth
Yes, correctly applied feedback can reduce anything in the output which is not in the input: harmonics, IM, noise, hum. Two complications: common-mode distortion at the summing point is not reduced by feedback, and low order stuff gets turned into smaller amounts of high order stuff. This means you need to use enough feedback, or none. A little bit of feedback is only a good idea if it is not needed for distortion reasons, but only to set gain or reduce output impedance.
I.M. distortion occurs when gain vs amplitude is not linear, and the result is both sum and difference frequency products. I recommend using the most linear tubes you can easily get (6SN7, 6SL7, 6922) and then using current sources or sinks (depending on topology) as plate loads. In voltage gain stages the Mu Follower topology is apparently quite good. Using topologies that cancel even harmonic distortion products (the diff amp for ex.) may cut I.M. some but also might kill the even harmonic distortion products, thereby making remaining distortion products (mostly odd) less psycho-acoustically comfortable.
That's okay, as I suspected that it was in fact wrong. Duh. I'm a EE, but I haven't done any professional analog circuit work in at least 30 years. I've mostly have done system level digital signal handling. Now if I get confused, imagine what a civilian ends up with... Why op-amps spec good but some can sound bad.
Speaking on a personal level, I have been thinking about IMD for quite a while now and what made me pose the question here is that lately I've noticed a few designs in passing (don't ask me specifics, just don't remember) some of which had relatively low THD and high IMD while the others were the opposite.
I personally haven't seen any recent spike in "IM" interest in the audiophile sphere nor any recent mention in the audiophile "press" that would have sparked a renewed interest in this type of distortion. I will disagree with the statement about the industry deprecating negative feedback amplifiers. The overwhelming majority of amps reviewed by the "big mags" are NFB designs be they tube or SS. Solid state amps because they pretty much have to have NFB and tube amps because (outside of the DHT niche) just about all commercial offerings are based on the the Williamson circuit.
I am not one to blindly dismiss NFB but I do view it with a bit of suspicion. On the face of it NFB is an ingenious concept. Just return a small portion of the output to the input and voila! all sorts of great things happen: lower distortion by a significant amount, better linearity and lower Z-out. What's the cost? a piece of wire a resistor and/or cap some solder and some sacrifice in gain. It seems that we can have our cake and eat it too. Is it really that simple?
I don't want to sidetrack the discussion and make it into one about NFB. I will only say that I view IMD as more important than Harmonic Distortion and as far as THD is concerned I subscribe to Pete Millett's idea that the profile of the diminishing harmonics is more important than their absolute values. All this is of course in my humble opinion.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.