OK, I misread you, when you said "feedback delay" I thought you referred to the feedback network. I think we agree after all.
Jan Didden
Jan Didden
Like Jan, I do not entirely agree with Sajti's way of finding the "proper" input filter cut-off frequency. You can only completely remove the overshoot on the error signal of a negative-feedback amplifier by using an input filter with a bandwidth smaller than or equal to the open-loop bandwidth. But if you have a sufficient headroom in all stages before the one causing the dominant pole, there is no reason to avoid the overshoot entirely. You just need to reduce it to a value that the stages in front of the one causing the dominant pole can handle. This means that an input filter bandwidth much greater than the open-loop bandwidth can be sufficient to avoid hard TIM, or slew rate limiting, or whatever you want to call it, if you have a sufficient headroom.
As I found the low pass filter's frequency can be higher than the pole frequency. Of course there will be some overshoot, but few % is no problem.
I found some amplifiers, with 100% overshoot with input filter...
Sajti
I found some amplifiers, with 100% overshoot with input filter...
Sajti
That's remarkable. If we are talking about the same signal (the difference between the filtered input signal and the signal from the feedback network or something proportional to this), then you should get overshoot as soon as the filter bandwidth exceeds the open-loop bandwidth. You can easily get overshoots of 10000% on this signal without any hard TIM or other problems, if the amplifier has a small open loop bandwidth, a high low-frequency loop gain and ample headroom in the first stage(s).
MarcelvdG said:That's remarkable. If we are talking about the same signal (the difference between the filtered input signal and the signal from the feedback network or something proportional to this), then you should get overshoot as soon as the filter bandwidth exceeds the open-loop bandwidth. You can easily get overshoots of 10000% on this signal without any hard TIM or other problems, if the amplifier has a small open loop bandwidth, a high low-frequency loop gain and ample headroom in the first stage(s).
Yes. Maybe 741 can be good example for this.
I read some recommendation to avoid it:
1, Use low feedback max. 20dB
2, Build Your amplifier with large bandwidht (100kHz, or higher)
3, Use LP filter at the input set to the open loop gain frequency pole (0% overshoot)
4, Not use the typical Ccb compensation in the VAS. Local feedback, or series RC network works better.
Sajti
In an amplifier there are two kinds of distortion:
1. linear distortion
2. nonlinear distortion
The frequency response of an amp is linear distortion. The RIAA emphasis for example is some kind of linear distortion. Not the amount of overshoot is the question. The question is:
Does any stage saturate?
If it does you get large amounts of TIM. Not only VAS can saturate, the LTP without emitter degeneration resistors can handle only about 50mV without large amounts of distortion. If this overshoot saturates any stage the performace degrades rapidly. This might be one reason why amps without an overall feedback sounds better more often.
So take a look how large the signal between the inverting and non-inverting input of an amp or opamp ist. If the signal is larger than the max. nonlinear distortionless input signal (about 50mV + 2 x voltage across emitter degeneration voltage) you run out of the linear region of the amp.
1. linear distortion
2. nonlinear distortion
The frequency response of an amp is linear distortion. The RIAA emphasis for example is some kind of linear distortion. Not the amount of overshoot is the question. The question is:
Does any stage saturate?
If it does you get large amounts of TIM. Not only VAS can saturate, the LTP without emitter degeneration resistors can handle only about 50mV without large amounts of distortion. If this overshoot saturates any stage the performace degrades rapidly. This might be one reason why amps without an overall feedback sounds better more often.
So take a look how large the signal between the inverting and non-inverting input of an amp or opamp ist. If the signal is larger than the max. nonlinear distortionless input signal (about 50mV + 2 x voltage across emitter degeneration voltage) you run out of the linear region of the amp.
In an amplifier there are two kinds of distortion:
1. linear distortion
2. nonlinear distortion
Rubbish!
Linear, by definition, means NO distortion. An ideal (passive) filter is a linear network, whose amplitude and phase response are not constant with frequency.
All distortion is nonlinear behaviour.
Nonlinear means getting out frequency components that you didn't put in:
The people I deal with recognize the distinction between
linear and nonlinear distortion, subtle as it might be. A
variation in frequency response is not typically accompanied
by substantial harmonics and intermodulation sidebands.
linear and nonlinear distortion, subtle as it might be. A
variation in frequency response is not typically accompanied
by substantial harmonics and intermodulation sidebands.
Hi,
Agreed.
Ultimately any reproduction of a medium represents a mix of linear and non-linear distortions.
And there are much worse distortion mechanisms than deviating from a flat frequency response...not that I'd want to make sweeping generalisations one way or the other though.
Cheers,😉
The people I deal with recognize the distinction between
linear and nonlinear distortion, subtle as it might be.
Agreed.
Ultimately any reproduction of a medium represents a mix of linear and non-linear distortions.
And there are much worse distortion mechanisms than deviating from a flat frequency response...not that I'd want to make sweeping generalisations one way or the other though.
Cheers,😉
I spend my professional life analyzing and modeling distortion mechanisms at RF.
A linear network, BY DEFINITION, does not introduce any distortion products.
Vout(f) = H(f)*Vin(f)
where f is frequency, H is the transfer function - at a given frequncy, it is a number. As you can see, all this happens at a single fequency: there are no harmonics or any other frequency products generated: therefore there is NO distortion.
Any component, passive or active, that has a "nonlinear transfer function", which in simple terms might be expressed as
Vout = a0 + a1*Vin + a2*Vin^2 + a3*Vin^3 + ...
will clearly produce harmonics of the input signal Vin. If Vin is a pair of tones, this will give a bunch of intermodulation products.
More complicated distortion mechanisms include taking the history of the input signal into account - "memory effects" - which can, under certain conditions be modeled by a modified power series (Volterra series).
For a more thoroughgoing explanation, see for example,
Bob Spence's excellent book "Linear Active Networks"
Wilson Rugh "Nonlinear System Theory"
-- John
A linear network, BY DEFINITION, does not introduce any distortion products.
Vout(f) = H(f)*Vin(f)
where f is frequency, H is the transfer function - at a given frequncy, it is a number. As you can see, all this happens at a single fequency: there are no harmonics or any other frequency products generated: therefore there is NO distortion.
Any component, passive or active, that has a "nonlinear transfer function", which in simple terms might be expressed as
Vout = a0 + a1*Vin + a2*Vin^2 + a3*Vin^3 + ...
will clearly produce harmonics of the input signal Vin. If Vin is a pair of tones, this will give a bunch of intermodulation products.
More complicated distortion mechanisms include taking the history of the input signal into account - "memory effects" - which can, under certain conditions be modeled by a modified power series (Volterra series).
For a more thoroughgoing explanation, see for example,
Bob Spence's excellent book "Linear Active Networks"
Wilson Rugh "Nonlinear System Theory"
-- John
I guess once more people give different meanings to the words and it depends on the context. Here is my take on it:
A (mathematically) linear function would be one whereby:
f(a+b+...n)=f(a)+f(b)+... f👎
In audio amplifiers for most people distortion means: "undesirable behaviour" - any deviation from the "straight wire with gain", meaning same gain at any amplitude and frequency in the audio band.
So, people call the products of an undesirable linear transfer function (such as your frequency response example) "linear distortion" and the products of an undesirable nonlinear transfer function"nonlinear distortion".
A (mathematically) linear function would be one whereby:
f(a+b+...n)=f(a)+f(b)+... f👎
In audio amplifiers for most people distortion means: "undesirable behaviour" - any deviation from the "straight wire with gain", meaning same gain at any amplitude and frequency in the audio band.
So, people call the products of an undesirable linear transfer function (such as your frequency response example) "linear distortion" and the products of an undesirable nonlinear transfer function"nonlinear distortion".
Hi PGW
ANY deviation of an output signal from input signal (apart from a constant gain factor of course) is called distortion in communications theory (where audio is a part of).
What you describe is non-linear distortion. When non-linear distortion is present there will be spectral components in the output signal that were not there at the input.
Every element that has a frequency dependant transfer function (i.e. generally everything) generates linear distortion. Here you won't have any new spectral content that wasn't there at the input (which in turn doesn't mean that no spectral content is lost during the process !).
In many RF applications you can even trade one kind of distortion against the other. Since the relative bandwiths of RF applications are usually ridiculously small compared to audio you can get away with harmonics (non-linear distortion) by filtering it out (i.e. introduce linear distortion instead).
Regards
Charles
ANY deviation of an output signal from input signal (apart from a constant gain factor of course) is called distortion in communications theory (where audio is a part of).
What you describe is non-linear distortion. When non-linear distortion is present there will be spectral components in the output signal that were not there at the input.
Every element that has a frequency dependant transfer function (i.e. generally everything) generates linear distortion. Here you won't have any new spectral content that wasn't there at the input (which in turn doesn't mean that no spectral content is lost during the process !).
In many RF applications you can even trade one kind of distortion against the other. Since the relative bandwiths of RF applications are usually ridiculously small compared to audio you can get away with harmonics (non-linear distortion) by filtering it out (i.e. introduce linear distortion instead).
Regards
Charles
Will someone give me a definition of "linear distortion", so that I can understand what is happening to the input signal.
I have to disagree with the statement that any network with a frequency-depndent transfer function generates distortion. This implies that all passive filter networks generate distortion, and I am unable to see how this is physically possible.
-- John
BTW, the RF applications I deal with span the frequency range DC-110 GHz.
I have to disagree with the statement that any network with a frequency-depndent transfer function generates distortion. This implies that all passive filter networks generate distortion, and I am unable to see how this is physically possible.
-- John
BTW, the RF applications I deal with span the frequency range DC-110 GHz.
PGW said:Will someone give me a definition of "linear distortion", so that I can understand what is happening to the input signal.
I have to disagree with the statement that any network with a frequency-depndent transfer function generates distortion. This implies that all passive filter networks generate distortion, and I am unable to see how this is physically possible.
-- John
BTW, the RF applications I deal with span the frequency range DC-110 GHz.
But it is true! The definition of linear distortion is a deviation caused by a linear element. A freq dependent network generates deviations from a linear freq response, and that by definition is linear distortion.
Jan Didden
/Anything is possible with the right definition!
(Hey, I just made this up. Pretty neat huh?)
One of the more common misconceptions among freshman undergraduates in Electronics is that a frequency-dependent transfer function is a source of distortion - i.e. a nonlinear effect.
This was generally disabused by a course in Linear System Theory, which shows that freq-dependent transfer functions (of the passive filter type, for instance) can be described by linear methods.
What is the point of calling something 'linear', if it introduces distortions = nonlinearities.
It just doesn't make sense.
This was generally disabused by a course in Linear System Theory, which shows that freq-dependent transfer functions (of the passive filter type, for instance) can be described by linear methods.
What is the point of calling something 'linear', if it introduces distortions = nonlinearities.
It just doesn't make sense.
Hi PGW,
as you pointed out
Y(s) = H(s)*X(s)
where s is the complex frequency, H(s) the transfer function, X(s) the input signal and Y(s) the output signal.
Well in the signal theory linear distortion is the same as the transfer function. Nothing to worry about. Maybe a less physical explanation is a unwanted deviation from an ideal behavior:
if H(s) != constant (frequency dependant) this is called linear distortion. So if you have the inverse transfer function H*(s) you can completely restore your original signal X(s) given by:
X(s) = H*(s) * Y(s) = H*(s) * H(s) * X(s) = X(s),
because H*(s) * H(s) is 1 by definition.
as you pointed out
Y(s) = H(s)*X(s)
where s is the complex frequency, H(s) the transfer function, X(s) the input signal and Y(s) the output signal.
Well in the signal theory linear distortion is the same as the transfer function. Nothing to worry about. Maybe a less physical explanation is a unwanted deviation from an ideal behavior:
if H(s) != constant (frequency dependant) this is called linear distortion. So if you have the inverse transfer function H*(s) you can completely restore your original signal X(s) given by:
X(s) = H*(s) * Y(s) = H*(s) * H(s) * X(s) = X(s),
because H*(s) * H(s) is 1 by definition.
Or as seen from the Voltage gain, any distortion that is not part of the gain function A:
Vout=A*Vin+Dl1*Vin+Dl2*Vin+.... + Dn1 + Dn2...
=Vin(A+Dl1+Dl2+...) +Dn1+Dn2...
A=gain
Dl: linear distortion functions
Dn:nonlinear distortion functions
So for the linear part:
Vout1+Vout2=Vin1(A+Dl1+Dl2+...) + Vin2(A+Dl1+Dl2+...)
= (Vin1+Vin2)*(A+Dl1+Dl2+...)
For the nonlinear part:
Vout1+Vout2=Dn1+Dn2+.... <> (Vin1+Vin2)*(Dn1+Dn2+.... )
Example:
Harmonic distortion: linear (is a linear function of input)
Johnson noise, energy storage: non-linear, is independent of input.
Ahem. For lack of an "official" definition I inferred this myself from common usage of the terms...
Vout=A*Vin+Dl1*Vin+Dl2*Vin+.... + Dn1 + Dn2...
=Vin(A+Dl1+Dl2+...) +Dn1+Dn2...
A=gain
Dl: linear distortion functions
Dn:nonlinear distortion functions
So for the linear part:
Vout1+Vout2=Vin1(A+Dl1+Dl2+...) + Vin2(A+Dl1+Dl2+...)
= (Vin1+Vin2)*(A+Dl1+Dl2+...)
For the nonlinear part:
Vout1+Vout2=Dn1+Dn2+.... <> (Vin1+Vin2)*(Dn1+Dn2+.... )
Example:
Harmonic distortion: linear (is a linear function of input)
Johnson noise, energy storage: non-linear, is independent of input.
Ahem. For lack of an "official" definition I inferred this myself from common usage of the terms...
- Status
- Not open for further replies.
- Home
- Amplifiers
- Solid State
- Amplifier topology subjective effects