An impulse is the wave of choice to have a clue about the open
loop response.
ALL is said on it, the cut off frequency and the open loop
gain figure.
If we stay in small signal conditions, the higher the output
voltage for a given input , the higher the open loop gain.
And the faster the rising, the higher the frequency of roll off.
loop response.
ALL is said on it, the cut off frequency and the open loop
gain figure.
If we stay in small signal conditions, the higher the output
voltage for a given input , the higher the open loop gain.
And the faster the rising, the higher the frequency of roll off.
In your case, it doesn't say anything about the ol response because you push it into slew rate limiting, so you can't even see the bw.
But I think our opinions are clear, I'll let it go.
jd
If we take the graph in post19 , we can see that the outputs
levels are 1.9V and 3.8V for an input of 500uV , thus apparent
gains are 3800 and 7600 in magnitude.
This is adequate with the delta of Cdom by a factor of two.
Rising time is 50uS in both case, yielding 38 and 76mV/uS respectively.
I think you wont have any trouble deducting the resulting bandwiths..
levels are 1.9V and 3.8V for an input of 500uV , thus apparent
gains are 3800 and 7600 in magnitude.
This is adequate with the delta of Cdom by a factor of two.
Rising time is 50uS in both case, yielding 38 and 76mV/uS respectively.
I think you wont have any trouble deducting the resulting bandwiths..
janneman, why a frequency sweep ?..
Let s throw a full bucket of frequencies (ad infinitum!) at the input
and let s see which are the ones that manage to find the exit..
basically, this is the purpose of a step function.
Hi Jan, the loop gain BW and amounts/locations of frequency regions of negative and positive feedback are "explained" for those with enough math/engineering background in:
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
look at fig 3 and surrounding discussion
pretty much the same material:
http://web.archive.org/web/20071215014821/www.luriecontrol.com/CFC/ch4/C4.htm
(now have to surf Luri'e site in the Wayback Machine - try section 4.3 )
basically below the loop gain intercept frequency (essentially the amp response corner frequency with "single degree of freedom" error feedback) we have (frequency dependant) negative feedback which reduces nonlinear errors by ~ the feedback factor
above the loop gain intercept (~amp corner frequency) the feedback becomes positive and we have some response peaking and boosting of the amplifier’s nonlinearities
so we have a "conservation law" which says we have to trade a (large in linear frequency, typ MHz) region of (small magnitude) positive feedback above the corner frequency for the advantages of (large magnitude) negative feedback over a (smaller) low frequency range of interest for audio amplification
the larger the "distance" in frequency between the "working" frequency range and the loop gain intercept, the more negative feedback we can apply
so those understanding negative feedback look for the fastest devices that can do the job and try for the highest stable loop gain intercept frequency that can be safely used - an then band limit the input signal with a separate input fliter to only apply signals that will be in the region of high negative feedback and to keep from "exciting" the positive feedback region
http://trs-new.jpl.nasa.gov/dspace/bitstream/2014/19495/1/98-0905.pdf
look at fig 3 and surrounding discussion
pretty much the same material:
http://web.archive.org/web/20071215014821/www.luriecontrol.com/CFC/ch4/C4.htm
(now have to surf Luri'e site in the Wayback Machine - try section 4.3 )
basically below the loop gain intercept frequency (essentially the amp response corner frequency with "single degree of freedom" error feedback) we have (frequency dependant) negative feedback which reduces nonlinear errors by ~ the feedback factor
above the loop gain intercept (~amp corner frequency) the feedback becomes positive and we have some response peaking and boosting of the amplifier’s nonlinearities
so we have a "conservation law" which says we have to trade a (large in linear frequency, typ MHz) region of (small magnitude) positive feedback above the corner frequency for the advantages of (large magnitude) negative feedback over a (smaller) low frequency range of interest for audio amplification
the larger the "distance" in frequency between the "working" frequency range and the loop gain intercept, the more negative feedback we can apply
so those understanding negative feedback look for the fastest devices that can do the job and try for the highest stable loop gain intercept frequency that can be safely used - an then band limit the input signal with a separate input fliter to only apply signals that will be in the region of high negative feedback and to keep from "exciting" the positive feedback region
Last edited:
but useful amps mostly define openloop BW via local feedback loops
you need to refine the conditions of your question considerably to separate the conditions
the most nearly no feedback OL gain stage is a common emitter amplifier - without feedback your only BW limiting choices are input vs output filters
you need to refine the conditions of your question considerably to separate the conditions
the most nearly no feedback OL gain stage is a common emitter amplifier - without feedback your only BW limiting choices are input vs output filters
Perhaps these statistics will help.
The same amp is used with two different frequency roll off..
measurement of the distorsion is made at 1KHZ , loop closed,
at a frequency where the OL gain of the two amps is still very close.
Clearly, the amp with wider open loop (and also CL) bandwith has less
distorsion, wich is logical , since there s more NF at high frequencies.
You can transpose the logic to your exemple with 10 KHZ and 20KHZ
open loop bandwiths..
The same amp is used with two different frequency roll off..
measurement of the distorsion is made at 1KHZ , loop closed,
at a frequency where the OL gain of the two amps is still very close.
Clearly, the amp with wider open loop (and also CL) bandwith has less
distorsion, wich is logical , since there s more NF at high frequencies.
You can transpose the logic to your exemple with 10 KHZ and 20KHZ
open loop bandwiths..
Thd
Hello jd
Was this idea about amp bandwidth and distortion inspired by your Audio Precision , and the observable effect of reducing measurement bandwidth and the improvement on THD.
Regards
Arthur
Hello jd
Was this idea about amp bandwidth and distortion inspired by your Audio Precision , and the observable effect of reducing measurement bandwidth and the improvement on THD.
Regards
Arthur
Member
Joined 2009
Paid Member
That's an interesting statement. But if 20kHz in the highest audible frequency, would you not want to extend the bandwidth to 20kHz so that these frequencies would be reproduced ?
Not directly, but the effect I wonder about is the same (I think). But it is true, if you see an AP plot of THD vs freq, and it does NOT rise above 10kHz, it's a save bet that they use 20kHz measurement bandwidth, which attenuates all higher harmonics.
For example, take Bob Cordell's mosfet amp with the error correction output stage. Most of the distortion in that amp (as little as there is of it) is probably generated before the output stage. You then could get even lower distortion by limiting the bw of the output stage.
Not sure if it is a viable idea implementation-wise, but I was just turning over the concept in my head.
jd
Assuming a topology LTP + VAS + OS.
If the distorsion occur mainly in the two first stage , "low passing" the signal at a frequency of 10KHZ after the second one will reduce the distorsion of the said two stages in open loop conditions.
But, then, when the loop is closed ,you ll have less negative feedback
available at 20KHZ.
So the question that ultimately arise is :
" does the THD reduction due to the lower frequency roll off balances positively
the THD increasement due to the induced lower feedback ratio ?"
Since the gain will be reduced by 6db at 20khz, THD in closed loop condition
will double at that frequency.
The reduction of distorsion brought by limiting the first two stages frequency
response will not compensate for this THD increasement.
If the distorsion occur mainly in the two first stage , "low passing" the signal at a frequency of 10KHZ after the second one will reduce the distorsion of the said two stages in open loop conditions.
But, then, when the loop is closed ,you ll have less negative feedback
available at 20KHZ.
So the question that ultimately arise is :
" does the THD reduction due to the lower frequency roll off balances positively
the THD increasement due to the induced lower feedback ratio ?"
Since the gain will be reduced by 6db at 20khz, THD in closed loop condition
will double at that frequency.
The reduction of distorsion brought by limiting the first two stages frequency
response will not compensate for this THD increasement.
Bob´s amp generates most of its distorsion in the output stage eventhough the EC network reduces it by a factor of about 10 according to Bob himself. This still leaves lots of distorsion to be reduced by the NFB loop.
What happens with other types of distorsion such as TIM, SID, DIM and SMPTE (and probably yet a bunch of horrors) when reducing the bandwidth?
The damping factor is also likely to be reduced if the high frequency open loop gain is reduced.
Hi Rikard,
You're right, most of the distortion is in the output stage, even with EC. This is not to say the input stage could not be improved, but with good design it is not difficult to drive input/VAS distortion pretty much into the ground. The input stage in my MOSFET EC amp can be improved by using a driven cascode. For example, the cascode bases can be driven by a replica of the feedback signal. This reduces common mode distortion in the JFET LTP.
The output stage is really where it is at in a well-designed amplifier. It is the most difficult source of static and dynamic distortions. It is also the place where distortions arising from program-dependent, temperature-dependent bias errors originate. These distortions do not always show up on conventional bench testing.
In general, all of those distortions mentioned tend to go down if the amount of NFB at 20 kHz is increased while maintaining good stability.
Cheers,
Bob
Guess you're right. Ahh well, another good idea down the drain.
Thanks for thinking along.
jd
Yes. Slew rate is "clipping" in the time domain and bandwidth is a frequency domain effect. Certain circuits have slew rate and bandwidth controlled separately.
Sorry, this is a very popular myth. Time domain and frequency domain can only be statistically mapped to one another and cannot be directly calculated numerically from inpulse. See article in Journal or AES by D. Pries.
Results of impulse testing can be very accurate or completely wrong depending upon very many factors.
Yes to this five times over. Wide band amps are made from single gain cells which are completely stable and predictable acting as "a filter with gain" long before they are strung together into a circuit with global feedback or any other global function. At least this is the easy way to do it so the results can be modeled equally easy.
Yes, Ididn't pursue this further as it was overtaken in the discussion, but testing an amp by driving into slew rate limiting gives no info about the bandwidth. A good example are early opamps that had a certain bw, say 100kHz, but that could not deliver full output beyond 10 or 20kHz due to slew rate limiting. But they could amplify a signal up to 100kHz flat, but not full level. The difference between freq response and power bandwidth.
jd
- Status
- Not open for further replies.