This isn't a strictly tube question but since I'm asking in the context of tube literature I decided to post here. I've attached a copy of an explanation of the effects of negative feedback from a 60s RCA Tube Manual.
My question is this: As I understand from the manual's explanation, the reverse phase signal is fed back to the amp input cancelling the peak as shown. This seems like a paradox to me thoughif the unwanted signal is being cancelled constantly, how can it appear at the amps output in the first place?
thanks
Mitch
My question is this: As I understand from the manual's explanation, the reverse phase signal is fed back to the amp input cancelling the peak as shown. This seems like a paradox to me thoughif the unwanted signal is being cancelled constantly, how can it appear at the amps output in the first place?
thanks
Mitch
It's not cancelled entirely. Assume you have an amp that amplifes 1000x without feedback.
That means that if you want, say, 10V at it's output, you need 10mV at is input terminals
If you want a gain of 10, an input of 1V for 10V output, you must somehow 'cancel' 990mV of that 1V input before it gets to the amp input terminals.
That way, the 10mV will give 10V output from the 1000x amp gain.
So you input 1V, and feed back about 1/10th of the output (a bit less, 990mV) to the input where it gets subtracted from the 1V input signal giving 10mV at the amp input termninals and thus the required 10V output. Easy ;-)
The other thing is the trap to think of it as something going around and being corrected 'before it can happen'. That doesn't happen, it is a continuous process. If the output starts to rise, the feedback also starts to rise and the cancellation is taking place as a continuous process. It's not like the feedback 'waits' for the output before it sends the signal beack to the input.
It's like riding a bike. There is a continuous process of negative feedback* from bike angle to your brain, a continuous process that keeps the bike upright.
Jan
* want to experience positve feedback? Try to ride your bike with your hands crossed, left hand to the right handle bar and vice versa. Make sure you have medical insurance before trying it.
That means that if you want, say, 10V at it's output, you need 10mV at is input terminals
If you want a gain of 10, an input of 1V for 10V output, you must somehow 'cancel' 990mV of that 1V input before it gets to the amp input terminals.
That way, the 10mV will give 10V output from the 1000x amp gain.
So you input 1V, and feed back about 1/10th of the output (a bit less, 990mV) to the input where it gets subtracted from the 1V input signal giving 10mV at the amp input termninals and thus the required 10V output. Easy ;-)
The other thing is the trap to think of it as something going around and being corrected 'before it can happen'. That doesn't happen, it is a continuous process. If the output starts to rise, the feedback also starts to rise and the cancellation is taking place as a continuous process. It's not like the feedback 'waits' for the output before it sends the signal beack to the input.
It's like riding a bike. There is a continuous process of negative feedback* from bike angle to your brain, a continuous process that keeps the bike upright.
Jan
* want to experience positve feedback? Try to ride your bike with your hands crossed, left hand to the right handle bar and vice versa. Make sure you have medical insurance before trying it.
The signal and the distortion are reduced (not cancelled) in the same amount when NFB is used.
To really understand how it works you have to understand the speed of electricity through the amp and the signal's wavelength and its time of existance in the entire amp circuit. If you can imagine a 20kHz signal beginning to rise at the input, the speed of the electricity is so fast through the amp that the feedback signal is already back to the input as the input signal is still just rising and then is being controlled by the NFB. Electricity is a liitle slower that the speed of light, but not significantly so in an audio amplifier. How fast will electricity travel through 3 feet of wire? At 186,000 miles per second, it's really quick. Figure it out for fun. And don't get caught up by those who will say there is a lag in the signal do to caps and inductors causing a timing lag. Those components cause a phase shift which is not a time delay. It is an instantaneous voltage difference propogated through the amp but still has its affect when it arrives back at the input.
Last edited:
186,000 miles per second. Right? . . . Right!
Now, let us apply that speed to a typical simple single ended 2 stage vacuum tube amplifier.
The rise time and the fall time of the amplifier are affected at least these 3 things:
Miller Effect Capacitance,
Output transformer primary's distributed capacitance
Output transformer's Leakage Inductance
Here is an example of a very good, simple, non-fed back single ended 2 stage amplifier:
The rise time and the fall time of the amplifier is 7usec.
Let us use a Gaussian response factor of 0.35 (which gives a well behaved square wave response).
Then the amplifier bandwidth (-3dB) = 0.35 / 7 usec = 50kHz, pretty good, Right?
With a 7us rise time and a 7usec fall time (nicely symmetrical), the time delay through the amplifier is about 7usec.
360 degrees of a 20kHz sine wave is 1/20kHz = 50usec
7usec / 50 usec = 0.14
0.14 x 360 degrees = 50.4 degrees.
As you can see, the negative feedback of a 20kHz sine wave, from the output transformer's secondary is delayed by about 50 degrees,
or about one seventh of a complete 20kHz sine wave.
The single ended 2nd harmonic distortion of 20kHz is 40kHz, that is within the 50kHz -3dB bandwidth of the amplifier.
The single ended 3rd harmonic distortion of 20kHz is 60kHz, the amplifier is only about -6dB at 60kHz.
But the phase of the 40kHz and 60kHz harmonic distortion products have higher numbers of degrees of delay.
40kHz is 100.8 degrees delay; 60kHz is 151.2 degrees delay.
Try and think about negative feedback feeding those distortions back to the input. Delay, Delay.
So much for perfect negative feedback at 20kHz.
Well, let us consider lower frequency music tones and music instrument's natural harmonics:
25 degrees at 10kHz; 12.5 degrees at 5kHz, and 6.25 degrees at 2.5khz. Not perfect either.
How many of you ever thought about global negative feedback that way?
Please let me know what you think about the above statements.
In other words, give me some 'negative feedback' about the ideas I presented (pun intended).
Enjoy the Music.
Worry less.
Now, let us apply that speed to a typical simple single ended 2 stage vacuum tube amplifier.
The rise time and the fall time of the amplifier are affected at least these 3 things:
Miller Effect Capacitance,
Output transformer primary's distributed capacitance
Output transformer's Leakage Inductance
Here is an example of a very good, simple, non-fed back single ended 2 stage amplifier:
The rise time and the fall time of the amplifier is 7usec.
Let us use a Gaussian response factor of 0.35 (which gives a well behaved square wave response).
Then the amplifier bandwidth (-3dB) = 0.35 / 7 usec = 50kHz, pretty good, Right?
With a 7us rise time and a 7usec fall time (nicely symmetrical), the time delay through the amplifier is about 7usec.
360 degrees of a 20kHz sine wave is 1/20kHz = 50usec
7usec / 50 usec = 0.14
0.14 x 360 degrees = 50.4 degrees.
As you can see, the negative feedback of a 20kHz sine wave, from the output transformer's secondary is delayed by about 50 degrees,
or about one seventh of a complete 20kHz sine wave.
The single ended 2nd harmonic distortion of 20kHz is 40kHz, that is within the 50kHz -3dB bandwidth of the amplifier.
The single ended 3rd harmonic distortion of 20kHz is 60kHz, the amplifier is only about -6dB at 60kHz.
But the phase of the 40kHz and 60kHz harmonic distortion products have higher numbers of degrees of delay.
40kHz is 100.8 degrees delay; 60kHz is 151.2 degrees delay.
Try and think about negative feedback feeding those distortions back to the input. Delay, Delay.
So much for perfect negative feedback at 20kHz.
Well, let us consider lower frequency music tones and music instrument's natural harmonics:
25 degrees at 10kHz; 12.5 degrees at 5kHz, and 6.25 degrees at 2.5khz. Not perfect either.
How many of you ever thought about global negative feedback that way?
Please let me know what you think about the above statements.
In other words, give me some 'negative feedback' about the ideas I presented (pun intended).
Enjoy the Music.
Worry less.
Last edited:
Again, the pitfall here is that the feedback doesn't wait for the signal to get to the output before reacting. It is a continuous process.
While your input signal is going through it's rise time, so does the feedback, at the same time.
Rise time is not propagation time!
Jan
While your input signal is going through it's rise time, so does the feedback, at the same time.
Rise time is not propagation time!
Jan
Jan,
I think it is an issue of continuous delay from amplifier input to amplifier output.
RC low pass filters have a delay time.
An impedance driving the Miller Effect capacitance causes delay of the signal.
RC low pass filters have a delay time.
An impedance driving the primary distributed capacitance of an output transformer causes delay of the signal.
RL low pass filters have a delay time.
An impedance driving a primary, which is driving leakage inductance, which is driving the secondary of an output transformer; that is driving a load impedance, causes delay of the signal.
I believe all of the above do have some delay time, however small.
I think the velocity of an e-field in air is approximately 1 nsec per foot.
A 1 usec delay would be about 1000 feet, much longer than akk the wiring in the amplifier (except perhaps the output transformer's primary winding).
The real world analog low pass filter examples I gave above are not the CD player's digital low pass filters that "create" pre transition 'ringing' and post transition 'ringing' of a CD's square wave test signal.
Why did those Tektronix 545B oscilloscopes need delay lines in the Vertical Channel?
And who of you has ever seen the differential Lumped CLCLCLC . . . delay lines of a non-b 545.
I am just wondering if any of the above is true.
I guess I have to find the time to measure some amplifier input to output delay times.
I think it is an issue of continuous delay from amplifier input to amplifier output.
RC low pass filters have a delay time.
An impedance driving the Miller Effect capacitance causes delay of the signal.
RC low pass filters have a delay time.
An impedance driving the primary distributed capacitance of an output transformer causes delay of the signal.
RL low pass filters have a delay time.
An impedance driving a primary, which is driving leakage inductance, which is driving the secondary of an output transformer; that is driving a load impedance, causes delay of the signal.
I believe all of the above do have some delay time, however small.
I think the velocity of an e-field in air is approximately 1 nsec per foot.
A 1 usec delay would be about 1000 feet, much longer than akk the wiring in the amplifier (except perhaps the output transformer's primary winding).
The real world analog low pass filter examples I gave above are not the CD player's digital low pass filters that "create" pre transition 'ringing' and post transition 'ringing' of a CD's square wave test signal.
Why did those Tektronix 545B oscilloscopes need delay lines in the Vertical Channel?
And who of you has ever seen the differential Lumped CLCLCLC . . . delay lines of a non-b 545.
I am just wondering if any of the above is true.
I guess I have to find the time to measure some amplifier input to output delay times.
Last edited:
Be sure that you don't conflate propagation delay with rise time. Output signal from an amplifier begins at near-speed-of-light delay after input signal. It then rises at a much slower rate linearly related to phase shift. Note that these things are completely unrelated.
All good fortune,
Chris
All good fortune,
Chris
Also, phase shift is definitely NOT delay. Even with a phase shift, input and output signals react to each others in basically zero time (except for a small transit time).
I've posted this several times so I'm not going to do it again, but try this in your simulator:
Feed a square wave current into a cap. Look at the voltage acros the cap.
Totally different waveforms because of the phase shifts. yet, at the very instant the input currewnt changes direction, the voltage across the cap changes direction. No delay.
Jan
I've posted this several times so I'm not going to do it again, but try this in your simulator:
Feed a square wave current into a cap. Look at the voltage acros the cap.
Totally different waveforms because of the phase shifts. yet, at the very instant the input currewnt changes direction, the voltage across the cap changes direction. No delay.
Jan
This is an example of a real time delay. There is no 'pre transient ringing'. The transient is delayed, the ringing starts at the start of the input transient.
Jan
An analog filter wouldn’t totally delay it - some very low effect would come through the filter immediately. Not much, but some. A digital filter would have a true delay, but it’s “frequency“ response extends to infinity. Maybe nothing useful happening because of aliasing, but it does “respond” to an infinite frequency, ie, there is an output when there is an input.
Because the peak in the response doesn’t happen immediately with an analog filter it “looks” like a delay is occurring.
Because the peak in the response doesn’t happen immediately with an analog filter it “looks” like a delay is occurring.
Yes. The problem (for understanding) is that a sine phase shifted still looks like a sine and thus gives the impression it is delayed.
That is why this sort of things is much better demonstrated with squares or triagle waves. Then you clearly see a phase shift but no delay.
jan
That is why this sort of things is much better demonstrated with squares or triagle waves. Then you clearly see a phase shift but no delay.
jan
Exactly. If it took the rise time and fall time to create a 180 degree phase shift then a cathodyne phase splitter would have to wait for that to occur for the inverted signal to be on the plate, but they are synchronous. It is a characteristic of the device, tube, cap, inductor, the energy is conducted at the speed of light but the immediate voltage value is different, so the phase of the signal is important but it is instantaneously everywhere in the amp. Way faster than an audio signal cycles.
Actually, a voltage propagates in copper around .64 of the speed of light.
But that's still very fast!
But that's still very fast!
I am so glad I started this discussion.
Thank all of you for noticing, and for chiming in!
First of all, I am not trying to support/criticize: negative feedback, or non-negative feedback.
I like local negative feedback, global negative feedback, and non-negative feedback (when they are implemented properly).
Now, consider the thing nobody has mentioned in this thread so far . . .
Lets take an often seen, typical global negative feedback circuit. We have a resistor that connects from the output transformer secondary to a second resistor to ground. That feedback resistor where it connects to the resistor to ground is a negative feedback node.
That negative feedback node, is connected to the bottom of the cathode's self bias resistor and bypass capacitor.
I am quite sure almost everybody has not only seen this, but a large percentage of you have owned such a circuit, or at least heard it at a friends, or at a Hi Fi / Stereo dealer.
But what did you forget? As often than not, there is a capacitor connected in parallel with the negative feedback resistor that comes from the output transformer secondary to the negative feedback node.
Wow! That looks like a high pass filter to me. Lead network, anyone?
That is a high pass filter that compensates for all the low pass filtering of the rest of the amplifier circuitry that is within the global negative feedback loop.
Perhaps we should open a thread to talk about group delay, but that is a completely different discussion, and generally understood by fewer people.
There are probably more threads about it in the loudspeaker posts.
Thanks again for your analysis and attention to these concepts.
Thank all of you for noticing, and for chiming in!
First of all, I am not trying to support/criticize: negative feedback, or non-negative feedback.
I like local negative feedback, global negative feedback, and non-negative feedback (when they are implemented properly).
Now, consider the thing nobody has mentioned in this thread so far . . .
Lets take an often seen, typical global negative feedback circuit. We have a resistor that connects from the output transformer secondary to a second resistor to ground. That feedback resistor where it connects to the resistor to ground is a negative feedback node.
That negative feedback node, is connected to the bottom of the cathode's self bias resistor and bypass capacitor.
I am quite sure almost everybody has not only seen this, but a large percentage of you have owned such a circuit, or at least heard it at a friends, or at a Hi Fi / Stereo dealer.
But what did you forget? As often than not, there is a capacitor connected in parallel with the negative feedback resistor that comes from the output transformer secondary to the negative feedback node.
Wow! That looks like a high pass filter to me. Lead network, anyone?
That is a high pass filter that compensates for all the low pass filtering of the rest of the amplifier circuitry that is within the global negative feedback loop.
Perhaps we should open a thread to talk about group delay, but that is a completely different discussion, and generally understood by fewer people.
There are probably more threads about it in the loudspeaker posts.
Thanks again for your analysis and attention to these concepts.
Last edited:
That nfb cap is sized just large enough to give a little more phase margin, but not to change the amplitude response.
Yes, it is called a lead network, since its phase leads, not lags. See p.172 here, from MIT open coursework book Roberge.
Especially see equation 5.6 and figure 5.7, which show possible amounts of phase shift from the component values.
Compare methods at table 5.1.There are also some nice lab photos, showing the response of the compensation methods.
Yes, it is called a lead network, since its phase leads, not lags. See p.172 here, from MIT open coursework book Roberge.
Especially see equation 5.6 and figure 5.7, which show possible amounts of phase shift from the component values.
Compare methods at table 5.1.There are also some nice lab photos, showing the response of the compensation methods.
Last edited:
A more interesting case for me is imperfect error correction. Given two otherwise identical two-stage amplifiers with matching gain structures, output stages and dB of feedback from the secondary winding, what difference is expected at the output if the front end of amplifier A has 0% thd and amplifier B 10% thd? It's temping to see it in mechanical terms as a decaying trail of uncorrected error while acknowledging the near instantaneous transit delay. Can't yet wrap my head around it.
The industry appeared to go through a period of 'let feedback sort it out' design philosophy, relying on very high gain front ends of mediocre linearity. 7199s, small plate loads on 12ax7s, etc.
No it isn't. It is a shelf filter whose purpose is to reduce the gain to zero before phase shift reaches 180 degrees. Fixing frequency response has nothing to do with it.