Achieving something which is unnecessary (unity gain stability in an amp not used for unity gain) almost always means losing something else, which may be necessary. You might compromise slew rate, noise etc. Overengineering is almost always poor engineering.stocktrader200 said:
Wavewhipper might like to do some reading about common-mode distortion, which can be introduced when the input signal and feedback are not summed at a single point but at two inputs to a single stage (e.g. two bases of a BJT LTP, grid and cathode of a valve stage). Common-mode distortion cannot be removed by feedback. It is something to be considered, but it is usually small - so small that many designers seem to get away with not even being aware of it.
In a -50C to 100C environment gain of a 2N3904 can swing from 150 to 400, this is +/-63%, +/-4.2dB.
In the old days transistors were expensive (like 10~100 resistors), incidentally object of cult too, but a single transistor was better than nothing. ICs were even more expensive (like 10~100 transistors), and harsh at zero cross in audio circuits (this is the worst place to have a discontinuity!!) Designers were trying to make consumer equipment with whatever was available and affordable, to cover the needs of a society growing in size and complexity. It was 2nd generation equipment. Previous sound recorder approach was registering vibration mechanically in cylinders or discs. Figure out what gain accuracy did that have.
Imagine a recorder without AGC and with +/-25% (or more) gain change depending on weather, and taking one or more minutes to settle upon power up. Now imagine yourself trying to gather evidence to prove a crime with such a recorder (such that the voice of the criminal has to be recognized by a jury), or trying to record something in nature in extreme hot or cold conditions, or noise analysis in a factory to catch a problem. These circuits were abandoned for several reasons. A dedicated operator was required to continuously compensate the flaws of the machine.
In the end price leveled out, so in SMD it costs about the same to place a resistor, a MMDT3904 (dual NPN) or a LM358 (dual cheap op-amp), with part cost being almost negligible for the resistor and comparable to placement cots for the MMDT and the LM358. All values around 10 cents.
The environmental conditions for using single transistor stages are no longer met. What remains? Cult to something that original designers were not giving cult to? The appearance? Cult to appearance can govern human tastes when environmental conditions are not opposing, but this is a side effect of seduction. Humans not under seduction are not vulnerable to cult to appearance.
I remember looking at an LM723 series pass power supply of a few hundred watts and figured I could make it function as an audio power amp by tapping in at a Vref input.
It would have been a Class A amp and who knows what bandwidth as it was designed.
If a DC audio amp were modified for bipolar adjustable power supply duty, the power rating would need a good bit of de-rating. They are meant for pulse duty.
It would have been a Class A amp and who knows what bandwidth as it was designed.
If a DC audio amp were modified for bipolar adjustable power supply duty, the power rating would need a good bit of de-rating. They are meant for pulse duty.
Last edited:
I recently forced an amp with unbalanced output transistors, I forced it to sound good with negative feedback. The amp didn't tell me what components to use, I showed that amp that it will accept the components I made it with, forced it to work, and I like that.
Yes I'm beginning to like that very much.
Yes I'm beginning to like that very much.
There are limits imposed by physics on how much OLG can be obtained. Each gain device introduces phase shift and propagation delay.
The following rule of thumb applies: OLG*UGF/NGD=nearly_constant
OLG=open-loop gain
UGF=unity gain frequency
NGD=number of gain devices, cost
So cost is roughly proportional to gain bandwidth product (cost ~=GBW=OLG*UGF).
In practice it is more convenient to consider how much GBW can be obtained from a fixed number of stages (considering other constraints), as a design problem with wider scope is difficult to close.
However, when the distortion originates from self disturbance within the loop, such as capacitive or inductive coupling, or summing node non-linearity, increased OLG does not reduce THD (not proportionally):
- The magnitude of the error (drive) signal becomes reduced by the same amount of dB the open loop gain is increased.
- The magnitude of disturbance signal, relative to error signal, increases by same amount of dB (for 1st stage, progressively less for intermediate stages).
- Thus, past some point, X more dB of open loop gain result in: X more dB of open loop distortion, attenuated by X more dB of NFB gain.
So additionally: there is an optimum range in the number of stages, more stages would not allow to reduce THD arbitrarily, as non-NFB-correctable sources of THD become dominant.
The following rule of thumb applies: OLG*UGF/NGD=nearly_constant
OLG=open-loop gain
UGF=unity gain frequency
NGD=number of gain devices, cost
So cost is roughly proportional to gain bandwidth product (cost ~=GBW=OLG*UGF).
In practice it is more convenient to consider how much GBW can be obtained from a fixed number of stages (considering other constraints), as a design problem with wider scope is difficult to close.
However, when the distortion originates from self disturbance within the loop, such as capacitive or inductive coupling, or summing node non-linearity, increased OLG does not reduce THD (not proportionally):
- The magnitude of the error (drive) signal becomes reduced by the same amount of dB the open loop gain is increased.
- The magnitude of disturbance signal, relative to error signal, increases by same amount of dB (for 1st stage, progressively less for intermediate stages).
- Thus, past some point, X more dB of open loop gain result in: X more dB of open loop distortion, attenuated by X more dB of NFB gain.
So additionally: there is an optimum range in the number of stages, more stages would not allow to reduce THD arbitrarily, as non-NFB-correctable sources of THD become dominant.
Last edited:
The optimum number of stages for a class AB audio amplifier is between 4 and 6. Less than 4 stages: more HF THD. More than 6 stages: no improvement in HF THD.
THD sources not correctable by NFB:
- In class AB amplifiers: Distorted inductive coupling from: halves of output current waveform flowing through different inductive paths. Inductive coupling from distorted base current waveform. Inductive coupling from base current switching peaks at current waveform zero cross.
- In BJT transistor stages with miller capacitor: Logarithmic nature of Vbe=log(Ic).
The field where this knowledge is obtained is multi-channel class D, when the analog audio guru learns RF.
Attached picture shows a 50ppm THD @ 50W class D minimalist balanced modulator with single integrator, "a piece of analog audio electronics". NFB at 20khz is around 38dB. Open loop THD is below 1% typically. The limiting factor is: about 20dB stronger magnetic fields than in class AB, permanent magnetic fields at any volume, square wave and periodic glitches in nature.
But there is a technique to bury the THD of inter-channel disturbance under the THD of the own channel, and under noise floor at low volume (about 120dB-A). The SO-8 is balanced comparator. The SO-14 is quad op-amp.
THD sources not correctable by NFB:
- In class AB amplifiers: Distorted inductive coupling from: halves of output current waveform flowing through different inductive paths. Inductive coupling from distorted base current waveform. Inductive coupling from base current switching peaks at current waveform zero cross.
- In BJT transistor stages with miller capacitor: Logarithmic nature of Vbe=log(Ic).
The field where this knowledge is obtained is multi-channel class D, when the analog audio guru learns RF.
Attached picture shows a 50ppm THD @ 50W class D minimalist balanced modulator with single integrator, "a piece of analog audio electronics". NFB at 20khz is around 38dB. Open loop THD is below 1% typically. The limiting factor is: about 20dB stronger magnetic fields than in class AB, permanent magnetic fields at any volume, square wave and periodic glitches in nature.
But there is a technique to bury the THD of inter-channel disturbance under the THD of the own channel, and under noise floor at low volume (about 120dB-A). The SO-8 is balanced comparator. The SO-14 is quad op-amp.
Attachments
Last edited:
A question.
If a high OLG resulting in a lot of negative feedback is so good, why does almost every good commercial power amplifier use to have a fairly low amount of NFB?
Furthermore, cheaper amps use to have higher NFB ( typical ~50db ) than expensive amps. A typical figure among audiophile amps use to be 0 to 30db.
Can the reason be that amplifiers that are forced to sound good by the means of NFB tend to be a bit grumpy and will output a slight squeak of objections?
Low NFB amps are glad and natural and will make their owners glad.
If a high OLG resulting in a lot of negative feedback is so good, why does almost every good commercial power amplifier use to have a fairly low amount of NFB?
Furthermore, cheaper amps use to have higher NFB ( typical ~50db ) than expensive amps. A typical figure among audiophile amps use to be 0 to 30db.
Can the reason be that amplifiers that are forced to sound good by the means of NFB tend to be a bit grumpy and will output a slight squeak of objections?
Low NFB amps are glad and natural and will make their owners glad.
Audio manufacturers make whatever the market asks for and negative feedback has had bad press for decades due to a mistake in a 1966 article by Daugherty and Greiner that got repeated many times by Otala's group in Finland. See the following links for a more elaborate explanation:
https://linearaudio.net/sites/linearaudio.net/files/volume1ltemvdg.pdf
https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
https://linearaudio.net/sites/linearaudio.net/files/volume1ltemvdg.pdf
https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
Marcel, do you really think that mistake has biased an entire industry for several decades? We humans are stupid but not that stupid.
If there is smoke there usually is fire.
But I never get tired of this issue since it's probably the one factor that's dividing audio people.
I see a clear pattern:
Low NFB, simple designs, single ended class A in one end.
Complex amplifiers with a lot of NFB in the other.
I think the position in that gliding scale tells something about who you are.
The former; poeple who thinks subjective impressions are more important than objective data.
The latter; rationalists.
But I think Diyaudio has a light problem since the audiophiles usually goes to the Passlabs section, so the rationalists has to dwell here in the "solid state" section. So the "inbetween" people really don't have anywhere to go.
If there is smoke there usually is fire.
But I never get tired of this issue since it's probably the one factor that's dividing audio people.
I see a clear pattern:
Low NFB, simple designs, single ended class A in one end.
Complex amplifiers with a lot of NFB in the other.
I think the position in that gliding scale tells something about who you are.
The former; poeple who thinks subjective impressions are more important than objective data.
The latter; rationalists.
But I think Diyaudio has a light problem since the audiophiles usually goes to the Passlabs section, so the rationalists has to dwell here in the "solid state" section. So the "inbetween" people really don't have anywhere to go.
Bass and guitar amps are perceived as twice as loud with some even and second order harmonics. Most people think they sound better. Funny, at proper levels
of distortion the distortion isn't heard, but only a pleasant tone and much louder sonic output. Lots of stuff on the web.
A clip :
Even-Order Harmonic Distortion
Tube amplifiers have much more distortion than solid-state amplifiers, but most of it is second-order, which is quite musical. That's why it's called "harmonic" distortion.
Second-harmonic distortion is exactly the same note, an octave above. Ditto for higher-order even harmonics; they are also the same note more octaves above.
Even-order harmonic distortion can be so pleasant that back in the 1970s the Aphex Aural Exciter was very popular in recording and broadcast specifically because it was designed to generate and add these harmonic distortions! You can still buy it today.
Why Tubes Sound Better
Yes, or at least a small part of the audio industry. Most of the amplifiers, the normal ones that are designed for non-audiophiles, have lots of negative feedback.
Don't forget also that amps with little feedback are a lot easier to design and build with crappy tagboard-style construction methods. More sophisticated amps built with cheap and nasty construction tend to oscillate.
That's the key reason that kitset amps tend towards modest or even mediocre. If you've got zero control over how the design is going to be built, then you've got to cater to the lowest common denominator.
That's the key reason that kitset amps tend towards modest or even mediocre. If you've got zero control over how the design is going to be built, then you've got to cater to the lowest common denominator.
Svitjod, if you read my letter, you'll know that I did my literature study in the 1990's, so already decades after 1966, and you'll know that I had an open mind about the issue and just wanted to know the truth.
Back then there were articles published in various Dutch electronics hobby magazines stating that you should not use too much overall feedback or no overall feedback at all because of TIM. When there were literature references, they always referred to articles by Otala's group, and when any arguments were given, they were always similar to those in the articles of Otala's group. The oldest articles of Otala's group referred to Daugherty and Greiner.
On the other hand there was the early 1980's PhD thesis of Ernst Nordholt, Design of high-performance negative-feedback amplifiers, which recommended using as much overall negative feedback as stability considerations would allow, which had a very short comment about TIM and which referred to the articles of Peter Garde.
Critically reading the articles of Daugherty and Greiner, Otala and Garde it became quite clear who was right: Peter Garde.
Back then there were articles published in various Dutch electronics hobby magazines stating that you should not use too much overall feedback or no overall feedback at all because of TIM. When there were literature references, they always referred to articles by Otala's group, and when any arguments were given, they were always similar to those in the articles of Otala's group. The oldest articles of Otala's group referred to Daugherty and Greiner.
On the other hand there was the early 1980's PhD thesis of Ernst Nordholt, Design of high-performance negative-feedback amplifiers, which recommended using as much overall negative feedback as stability considerations would allow, which had a very short comment about TIM and which referred to the articles of Peter Garde.
Critically reading the articles of Daugherty and Greiner, Otala and Garde it became quite clear who was right: Peter Garde.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.