Less 1khz signal is due to higher NFB available at 20khz , thus increasing
the linearity at high frequency...
IMD first order distorsion increase as a square law of the input signal,
while third order will increase at cubic rate...
We cant make such a deduction since no miller compensation is used,
so the available NFB will inherently correct any stage distorsion with the
same available ratio.
It s not like a sizeable part of the OLG is used solely for VAS compensation
and linearization , the available global feedback is the same wether this the VAS
or the OS that are in need of correction.
There will be less differential voltage, that is, less 19/20khz differential
voltage , thus , the input stage will generate less distorsion related
to theses frequencies, including IMD sub frequencies by products.
As already pointed , since IMD increase rapidly with increasing
input amplitude , a slight increase in high frequency linearity
will in principle yield a more than proportionnal reduction in IMD
related products.
More of the same...
Using the topology displayed a few post earlier , an insight
about IMD , wich is quite surprising.
The same amp is used, albeit with OLG dramatically increased ,
as displayed by the graph below.
Amp B display about 5db more OLG than amp A at 20khz,
but strictly the same OLG at 1khz.
The two variants are then compensated using a light TMC.
The resulting IMD 1khz by product with a 19/20khz input signal
is quite astonishing when comparing the two amplifiers.
Variant B display about 30db less 1khz residual , despite having
only 5db better OLG at 20khz and no OLG increasement at 1khz.
Perhaps JCX , who obtained simular results but with 30 db OLG
increase at 1 khz and 5.3db at 20khz , can shed some light about
such an unexpected improvement.....😉
Anyway, there is more to be done to reduce IMD low frequency
residuals than increasing the OLG at these frequencies.
Using the topology displayed a few post earlier , an insight
about IMD , wich is quite surprising.
The same amp is used, albeit with OLG dramatically increased ,
as displayed by the graph below.
Amp B display about 5db more OLG than amp A at 20khz,
but strictly the same OLG at 1khz.
The two variants are then compensated using a light TMC.
The resulting IMD 1khz by product with a 19/20khz input signal
is quite astonishing when comparing the two amplifiers.
Variant B display about 30db less 1khz residual , despite having
only 5db better OLG at 20khz and no OLG increasement at 1khz.
Perhaps JCX , who obtained simular results but with 30 db OLG
increase at 1 khz and 5.3db at 20khz , can shed some light about
such an unexpected improvement.....😉
Anyway, there is more to be done to reduce IMD low frequency
residuals than increasing the OLG at these frequencies.
Attachments
Last edited:
Wahab
Again interesting simulations.
I'm never sure what it is meant by TMC implementation but if this is a local loop around the vas and output stage, your simulation may suggest that the dominate distortion source is vas+output stage. As you say the higher loop gain around this loop at these frequencies would account for the reduced lower distortion at the output. The earlier post is consistent with this but it is difficult to separate effects.
Thanks
Again interesting simulations.
I'm never sure what it is meant by TMC implementation but if this is a local loop around the vas and output stage, your simulation may suggest that the dominate distortion source is vas+output stage. As you say the higher loop gain around this loop at these frequencies would account for the reduced lower distortion at the output. The earlier post is consistent with this but it is difficult to separate effects.
Thanks
Mmmm.
Interesting finding Wahab, and I think it is very important that we get to the bottom of this. We need a robust explanation as to why this is the case.
Since 1kHz is slap bang in the midband, we may well be looking at why (and this may be one of a number or sources) amplifiers have different sonic signatures.
Interesting finding Wahab, and I think it is very important that we get to the bottom of this. We need a robust explanation as to why this is the case.
Since 1kHz is slap bang in the midband, we may well be looking at why (and this may be one of a number or sources) amplifiers have different sonic signatures.
Last edited:
Something is creating an HF rolloff. This is unlikely to be in the output stage, as followers have good HF performance (limited usually by stability, not frequency response). So whether the rolloff is caused by deliberate Miller compensation, or intrinsic capacitance, is quite irrelevant. The signal level in the output stage is the same. The signal level in the earlier stages is lower, so less distortion produced.wahab said:
The relevant signal level in each case is not the input signal, but the signal level in each stage as that is what determines the distortion generated in each stage. Given that, it is difficult to compare like with like.
In a balanced circuit, even-order distortion (which is what we are looking at) depends critically on circuit balance as you are subtracting two big numbers from each other and looking at the difference. The only imbalance in your simulated circuit will be differences between PNP and NPN BJTs, which presumably have been chosen to be near complements. If your change has somehow improved the AC balance of the circuit then this will drastically reduce 1kHz IMD.
For example, if the distortion is being generated primarily in the VAS (quite likely) then local feedback around it (from TMC) will reduce distortion. Increasing the internal gain of this inner loop (by bypassing the VAS emitter resistor) will reduce distortion, and it will improve AC balance as the gain of this stage is then more closely determined by passive components (TMC) rather than BJT parameters. In real life it won't be quite so simple as you don't get perfect matching.
As I said, it is very difficult to compare like with like. Hence false conclusions can be drawn, which appear at first to conflict with feedback theory.
Often simulation models for complementary pairs are inherently exact complements, i.e. they use the same parameters except for the polarity. Amplifiers using such models would tend to be very well balanced, which in real life you will never be able to reproduce.
Not saying that is the case here but it is another trap to be aware of.
jan didden
Not saying that is the case here but it is another trap to be aware of.
jan didden
Last edited:
As the circuit appears to be fully balanced, identical complementary BJTs would give zero even-order distortion. Simulations can be very useful, as long as you ask the right questions and don't always believe the answer.
Hi,
it must be a wonderful feeling to have the deceptive imagination that just by applying some feedback, distortions get magically cut out with surgical precision and swept out like trash.
In an environment characterized by continual rapid changes, when the distorted output signal, after having traveled through various PN-junction diodes and capacitances arrives to the summing node for straightening, has a rather negligible validity. What it can do is mess things up even more.
it must be a wonderful feeling to have the deceptive imagination that just by applying some feedback, distortions get magically cut out with surgical precision and swept out like trash.
In an environment characterized by continual rapid changes, when the distorted output signal, after having traveled through various PN-junction diodes and capacitances arrives to the summing node for straightening, has a rather negligible validity. What it can do is mess things up even more.
wuyit: Handwaving beats rigorous analysis and data every time, is that the idea?
Still waiting for the other data from you.
Still waiting for the other data from you.
WuYit, thank you for your 'helpful contribution' to a discussion about how feedback affects IMD. We were getting along just fine, talking and thinking about real differences of opinion, and even teaching some people something.
Wahab
Indeed it does seem difficult (or impossible) to attribute the observations to a single parameter especially without knowing all the effects of the changes.
But it is fun and instructive to try and figure out whats going on, it seems reasonable that as has been mentioned, if the distortion is dominated by a source within the VAS+OS then the local TMC loop would reduce the resultant 1Khz product as it has significant loop gain at 1KHz. However it also seems likely that the TMC implementation may have significantly changed the VAS and or OS input impedance and output impedances. The effects of this may very well be significant and of course there may be others.
I think that part of your point is, if the above is correct then why did the first set of plots show a reduction in the 1KHz product when only the OLG at 20KHz was increased. But again, bypassing the degeneration resistors has more effects than just to increase the gain. The impedance changes mentioned above certainly apply here as well.
It would be interesting for me to see an ac sweep plot of both the input and output VAS currents for the different circuits.
Thanks, hope you post more data soon.
-Antonio
Indeed it does seem difficult (or impossible) to attribute the observations to a single parameter especially without knowing all the effects of the changes.
But it is fun and instructive to try and figure out whats going on, it seems reasonable that as has been mentioned, if the distortion is dominated by a source within the VAS+OS then the local TMC loop would reduce the resultant 1Khz product as it has significant loop gain at 1KHz. However it also seems likely that the TMC implementation may have significantly changed the VAS and or OS input impedance and output impedances. The effects of this may very well be significant and of course there may be others.
I think that part of your point is, if the above is correct then why did the first set of plots show a reduction in the 1KHz product when only the OLG at 20KHz was increased. But again, bypassing the degeneration resistors has more effects than just to increase the gain. The impedance changes mentioned above certainly apply here as well.
It would be interesting for me to see an ac sweep plot of both the input and output VAS currents for the different circuits.
Thanks, hope you post more data soon.
-Antonio
I for one really enjoy and appreciate WuYit's posts.
They make me think along a path I would not have ever taken.
Thanks
-Antonio
They make me think along a path I would not have ever taken.
Thanks
-Antonio
SY,
I clearly pointed at the time aspect now and many times before. It is very difficult to comprehend how people can solely review the amplitude, completely disregarding the time axis. The obnoxious distortions are not at all amplitude distortion, but time based.
The switching action depends on both the input voltage and output current, an accurate control (in time) of both halves would be needed, one problem is just the time delay in the sense mechanism. Uncountable attempts have been made...
I clearly pointed at the time aspect now and many times before. It is very difficult to comprehend how people can solely review the amplitude, completely disregarding the time axis. The obnoxious distortions are not at all amplitude distortion, but time based.
The switching action depends on both the input voltage and output current, an accurate control (in time) of both halves would be needed, one problem is just the time delay in the sense mechanism. Uncountable attempts have been made...
No, you've provided zero analysis or data. Handwaving only, and when asked for data, you've ignored these requests.
Your style is familiar...
Your style is familiar...
SY,
I am sorry. In absence of requisite evidence, most of the statements here are statements of opinion. Evidence could be found elsewhere for those who are interested.
I am sorry. In absence of requisite evidence, most of the statements here are statements of opinion. Evidence could be found elsewhere for those who are interested.
To my understanding there is no such a thing as "distorted output signal" in an NFB amp when no stages are over-driven, as long as one takes the input of the amp being the subtraction between the base of the input transistor pair.
How do you make a sound reproduction chain without using negative feedback? No followers, so high output impedance on all items therefore only very short cables allowed otherwise you get HF loss from interconnects. Almost all existing loudspeakers would be unusable, as they are designed for voltage drive. No stabilised power supplies anywhere, either. This is what you would have to do to avoid "bad sound". I guess it's back to live acoustic concerts only.
- Status
- Not open for further replies.