This is true.
But it is also true that the usual measurements are just a poor approximation of how an amplifier will sound.
The best microphones are no way near to hear what the human ears do.
What's the solution? Improve the damn measuring methods.
Use the impulse response first and foremost. Use multitones and not sinusoids for distortion measurements. Measure IMD, measure milliwatt THD, measure transient distortion.
But with 0,1% 2nd harmonic it is inevitable to obtain much more IMD products, and from ultrasonic disturbancies especially. You find "natural" those combinations products?
Try to send in "musical" sounding amplifier with 0,1% 2. and 3. harmonics signal e. g 50+51kHz and observe level of 1kHz. and combination products in audioband... You get very "rich" sound. With complex signal it is even much worse. It is not time to say, you like IMD distortion?
Last edited:
You have a quote from Paul Klipsch in your signature line, with his name mis-spelled.
Maybe I’m missing something here, but do you think that an amp with 0,01% distortion on all orders up to 6th will have a THD of 0,01%?
Can you explain me why (honest question)?
(my reference are no GNFB class A amps btw).
Personally I think that 0,1% 2nd or 3rd armonic are clearly audible, the threshold being somewhere between 0,01% and 0,05% (educated guess). Point is that it is extremely difficult to achieve this without GNFB.
My idea is specialization, i.e. to multiamp dividing the bandwith at least in 3 ranges.
Definitely not, I say that it should be specifically measured. And i agree with Pass' paper on it.
Sorry, the first figure in my sentence should have been deleted. (cant edit now the post).
Because circuit with higher nonlinearity (even low order) will produce more IMD products.
For sure you will not hear 0,1% 2. or3. harmonic with single sine ton. But you will hear IMD products with two tone signal or complex signal, e,g music.
Example for class A with about 0,2% 3.rd harmonics
Stereophile: Pass Labs Aleph 3 power amplifier
Member
Joined 2009
Paid Member
Not really, every musical instrument produces them.
Anyhow, my thinking is that if these higher harmonics are small enough, below the level of audibility then do they matter ?
Does not a middle level of nfb increase all harmonics and therefore produce the worse result.
Is it not true that it's best to have zero nfb or lots of nfb but avoid the middle ?
Hi,
"feedback is bad" was not my wording. Transistors are unusable without feedback. There´s, however, a distinctive difference between local feedback, feedback through two devices and feedback through multiple devices. The problems grow explosively with increasing number of stages, not just because of propagation delay and phase errors. For instance, each stage effects the feedback loop and is effected by the feedback loop, creating a mess.
"feedback is bad" was not my wording. Transistors are unusable without feedback. There´s, however, a distinctive difference between local feedback, feedback through two devices and feedback through multiple devices. The problems grow explosively with increasing number of stages, not just because of propagation delay and phase errors. For instance, each stage effects the feedback loop and is effected by the feedback loop, creating a mess.
Lumba,
You are absolutely right, local is more stable than interstage than global, no question. However, there are other areas fb can go wrong. The extraction of the error signal using an LTP is flawed since differential operation of the conventional LTP creates dynamically varying Vbe, more on one, less on the other, giving an S shaped transfer function. This creates errors in the output.
If the diff input to the LTP is kept very small, say under 10mV by adopting very high OLG, then these errors are tiny and distortion is very low. But not always.... and even on a high loop gain amp a CFP LTP makes a profound difference to SQ.
I enjoy your little heresies.... keep 'em coming.
You are absolutely right, local is more stable than interstage than global, no question. However, there are other areas fb can go wrong. The extraction of the error signal using an LTP is flawed since differential operation of the conventional LTP creates dynamically varying Vbe, more on one, less on the other, giving an S shaped transfer function. This creates errors in the output.
If the diff input to the LTP is kept very small, say under 10mV by adopting very high OLG, then these errors are tiny and distortion is very low. But not always.... and even on a high loop gain amp a CFP LTP makes a profound difference to SQ.
I enjoy your little heresies.... keep 'em coming.
Hi Bigun,
The points you have made are not technically incorrect, but it is a matter of degree.
It is true that phase delay is not the friend of negative feedback. It is fundamental that feedback loops must be properly compensated to be stable with the amount of delay in the loop. But this is done routinely by those who work with feedback circuits. At the same time, the phase delay does not render the NFB useless in the audio band. Bear in mind that we are usually at least a decade above the audio band before the phase delay reaches 135 degrees. It is easy to show by simulation that NFB is very effective at reducing distortion at frequencies well below the gain crossover frequency. The suggestion that NFB "cannot correct a problem after it happens" may have some intuitive appeal, but it is generally wrong. With phase margin of 45 degrees at a gain crossover frequency of 1 MHz, the "delay" in the loop is less than 0.5 us. This is only 1% of the period of a 20 kHz sinusoid. This largely describes the degree of imperfection due to the time delay. It simply must be put in context. Bear in mind, in all of this, that NFB is not perfect, but that does not mean it is ineffective.
It is technically correct that the application of NFB around a low-order nonlinearity will create some higher-order products. However, the generalization that NFB will increase the level of high-order products while reducing low order products is not valid. It has been shown in other threads here that Baxandall's results were correct as far as they went, but that Baxandall's results cannot be generalized to statements like "applying negative feedback increases the high order products." As NFB is increased around a low-order nonlinearity, it will increase the amplitudes of the high order products at low levels of NFB, but at levels beyond 15-20 dB it will decrease them. More importantly, vitually no real amplifier creates only second-order products in the open loop. This is because the exponential nonlinearities of BJTs create high order products in the first place, and because the cascade of multiple stages adds to the oder of the harmonics created in the open loop. Finally, crossover distortion in the output stage creates high-order products. The point is that the process of recirculation of products in a feedback loop generally creates high-order products of much lower level than are there in the first place from other mechanisms. Moreover, the NFB actually serves to reduce these other high-order products that were pre-existing in the open loop.
A few posts back, Andy_c posted a link to some earlier posts in another thread that shows the point with distortion plots much better than I can explain it verbally.
The main point is that we always must be careful of generalizing something that started out with a grain of truth.
Cheers,
Bob