I think that I am going to withdraw what I wrote in post #524 about the language barrier !!! And Bruno has a lot of reason for being grumpy.
I don't know your topology but I guess it is a natural-sampling (i.e. the classic carrier -based PWM) amp.
There are some things that can go wrong when doing NFB in such a topology. And there is something called "ripple aliasing" that doesn't allow the same amount of distortion reduction from NFB - even when applied in very high amounts - as is achievable with a self-oscillating amp - unless special measures are taken. Maybe that is te reason for using no or low NFB factors combined with a regulated PSU. Just guesssing .......
Regards
Charles
I don't know your topology but I guess it is a natural-sampling (i.e. the classic carrier -based PWM) amp.
There are some things that can go wrong when doing NFB in such a topology. And there is something called "ripple aliasing" that doesn't allow the same amount of distortion reduction from NFB - even when applied in very high amounts - as is achievable with a self-oscillating amp - unless special measures are taken. Maybe that is te reason for using no or low NFB factors combined with a regulated PSU. Just guesssing .......
Regards
Charles
Hi,
DXA is selfoscillant ( pre-filter primary FB) after, use nfb as AB class.
scheme di modulator is absolute new. I do not want to discuss dxa ,if better or not (sorry for this).
I think that my last message have clare english,not absolute right, but very clean.
Independent of all, power supply is the soul of what you hear.
only the amp, should modulate the current. No Amp + PSU, especially at low power 40 50w.
Yes, entry-level use fixed carrier pwm (SDA-400). maybe if you listen this..understand all.
Last edited:
The PSU will be the soul of what you hear if the the amplifier is incapable of isolating itself from the supply (poor PSRR)
A more stable, higher reserve supply will make the task easier for the amp and make the most of its PSRR, but the end performance will be whatever it shown by the (measured) results.
I can't see why you have a problem - which is what your posts imply - with SMPSs and why you have a problem with the ncore producing its results when supplied by an SMPS.
An SMPS inherently contains regulation (the final specs. and performance will tell you how good it is).
It has to, because its job is to regulate a mains voltage directly down to a DC voltage
I, as an external reader of the specs and performance results that I have seen here - not as a 'Bruno fan' - have no problem in observing (from the results) and believing (from concept and theory) that an appropriately designed SMPS will match a linear PSU.
An SMPS is a form of class D amplifier. It uses switching topology to produce an output signal that tracks the input signal - a static DC value. It regulates AC mains down to a DC output signal.
A class D amplifier uses switching topology to produce an output signal that tracks the input signal - a varying value. It regulates the DC supply voltage to provide an accurate alternating voltage.
If you can accurately produce AC by switching DC then you can accurately produce DC by switching AC
Bruno, although you somewhat covered it already (you basically said that good IMD performance for the input below 20 kHz range most likely results in good-ish IMD performance for the input within the power bandwidth), it would be really interesting to see the results of actual measurements.
what measures? continuous signal with fft?
apart from that, listening to the problem is immediately apparent.
I remind you that we are talking about a high end amp, not a PA.
an audiophile feel immediately when the voltage moves. this is perfectly measurable by other measures (dynamic).
In this range, typically using a linear PSU up to 100.000 uF capacitors. NuForce use a good smps clean regulated psu and adjust the dynamic response with a table of capacitors.
I have developed a regulated power supply (SMPS) especially for this.
What are you talking about the PSRR?
I have nothing against the NCore. everyone is free to do and say whatever he wants.
I do not agree that over 20Khz does not make sense.(i start from this)
yes, if you extend the band, everything becomes complicated, I know. drivers, dt, bias, phase shift, delay, comparator and buffers. For this, the project is worth much more.
Last edited:
Why you have capacitors oversized in your smps? for ripple? well, what is frequency of ripple that you adjusted with it?
I have no doubt, only 5 transistors for good amp, this amp Can be Amp or regulator. then your amp solves a problem of the regulator. response time?
yes,i know B&O con mixed FB...
DXA not use mixed FB. use two fb and other circuit for obtain a linear response in range 10Hz to 60Khz.
behavior is also different in audio measurements refer to an class D.
I thought I'd just give a short run-down of the AES paper for those who wonder if they should read it:
The DC transfer of self-oscillating modulators is non-linear. This is inherent in any self-oscillating amplifier that is not an ideal first-order hysteresis modulator. Any amount of post-filter feedback gives rise to this distortion so it must be understood in order to be able to optimize for it.
The usual methods of estimating the oscillation frequency and small-signal gain do not predict this distortion, so they cannot be used as an optimization tool. The first step therefore is to establish an exact oscillation criterium that is valid for all types of self-oscillating amplifiers and which exactly predicts the oscillation frequency with duty cycle as a parameter. The second step is to determine the DC transfer from input to duty cycle. This then allows us to optimize the loop transfer function for the lowest open-loop distortion.
1) Any self-oscillating amplifier, regardless of whether it is "hysteresis controlled" or "phase shift controlled" can be rearranged as a linear transfer function H(s) and a delay-free sign detector. The propagation delay from the comparator input all the way to the power FETs is modeled as a factor exp(-s*tdelay) in H(s).
The usual approximation is to say that the amplifier will oscillate at the frequency where the phase shift is 360 degrees, and that the small signal gain is determined by the slope of the ripple at the comparator inputs. This is clearly incorrect on three counts: the 360 degree criterium assumes a sinusoidal oscillation (we're building a square wave oscillator!), duty cycle is completely ignored though we're all familiar with the frequency modulation of self-oscillating circuits and the gain approximation presumes that the ripple is independent from the signal.
To find the correct criterium, we take the fourier expansion of a square wave of frequency f and duty cycle h and multiply all fourier terms by their respective value of the H(s) at that frequency to obtain the output after the loop filter. If the sign of that output (minus the DC term!) is the same as the square wave we started off with, f and h are a valid oscillating condition. Solving the equation can't be done algebraically but the numerical solution converges very rapidly.
2) Having worked out f for each h, we can calculate the difference between the DC output of H and the instantaneous output at the crossings. This yields the DC transfer of the modulation.
What we find is quite interesting. While frequency modulates all over the place, the DC transfer can be made remarkably linear. This becomes the target of a numerical optimization which in practice only affects the location of a few high-frequency zeros. It doesn't get in the way of making lots of loop gain.
3) Even more interesting still, we find no indication that the DC transfer is frequency dependent. The model reliably predicts modulator distortion across the bandwidth of the amplifier. This is one of the chief reasons why HF IMD spectrum corresponds so well with the THD spectrum on a product-by-product basis. While it is never possible to make an amplifier with IMD that is better than THD, this finding shows at least that IMD is going to be exactly as good as THD.
I'd like to stress again that H(s) contains the amplitude and phase response of the output filter, the loop filter and the propagation delay. To the uninitiated it might seem that these various factors, listed in quick succession, make an impressively daunting list of issues but this analysis neatly pulls everything together into a fairly concise and very useful mathematical expression.
Something that isn't mentioned in the paper, but which you can easily work out for yourself, is that you can rewrite the criterium for unequal delays and solve the equation with the effects of dead time, output current and ripple current taken into account. This allows you to compute quite precise output spectra very rapidly without actually having to simulate the circuit, should one be so inclined.
I'm glossing over a lot but at least I hope to give an idea of whether you want to enrich the AES by a tenner or not. By the way I realised that I'm entitled to send people the paper by e-mail so you could give a shout (pm) if you're seriously interested.
The DC transfer of self-oscillating modulators is non-linear. This is inherent in any self-oscillating amplifier that is not an ideal first-order hysteresis modulator. Any amount of post-filter feedback gives rise to this distortion so it must be understood in order to be able to optimize for it.
The usual methods of estimating the oscillation frequency and small-signal gain do not predict this distortion, so they cannot be used as an optimization tool. The first step therefore is to establish an exact oscillation criterium that is valid for all types of self-oscillating amplifiers and which exactly predicts the oscillation frequency with duty cycle as a parameter. The second step is to determine the DC transfer from input to duty cycle. This then allows us to optimize the loop transfer function for the lowest open-loop distortion.
1) Any self-oscillating amplifier, regardless of whether it is "hysteresis controlled" or "phase shift controlled" can be rearranged as a linear transfer function H(s) and a delay-free sign detector. The propagation delay from the comparator input all the way to the power FETs is modeled as a factor exp(-s*tdelay) in H(s).
The usual approximation is to say that the amplifier will oscillate at the frequency where the phase shift is 360 degrees, and that the small signal gain is determined by the slope of the ripple at the comparator inputs. This is clearly incorrect on three counts: the 360 degree criterium assumes a sinusoidal oscillation (we're building a square wave oscillator!), duty cycle is completely ignored though we're all familiar with the frequency modulation of self-oscillating circuits and the gain approximation presumes that the ripple is independent from the signal.
To find the correct criterium, we take the fourier expansion of a square wave of frequency f and duty cycle h and multiply all fourier terms by their respective value of the H(s) at that frequency to obtain the output after the loop filter. If the sign of that output (minus the DC term!) is the same as the square wave we started off with, f and h are a valid oscillating condition. Solving the equation can't be done algebraically but the numerical solution converges very rapidly.
2) Having worked out f for each h, we can calculate the difference between the DC output of H and the instantaneous output at the crossings. This yields the DC transfer of the modulation.
What we find is quite interesting. While frequency modulates all over the place, the DC transfer can be made remarkably linear. This becomes the target of a numerical optimization which in practice only affects the location of a few high-frequency zeros. It doesn't get in the way of making lots of loop gain.
3) Even more interesting still, we find no indication that the DC transfer is frequency dependent. The model reliably predicts modulator distortion across the bandwidth of the amplifier. This is one of the chief reasons why HF IMD spectrum corresponds so well with the THD spectrum on a product-by-product basis. While it is never possible to make an amplifier with IMD that is better than THD, this finding shows at least that IMD is going to be exactly as good as THD.
I'd like to stress again that H(s) contains the amplitude and phase response of the output filter, the loop filter and the propagation delay. To the uninitiated it might seem that these various factors, listed in quick succession, make an impressively daunting list of issues but this analysis neatly pulls everything together into a fairly concise and very useful mathematical expression.
Something that isn't mentioned in the paper, but which you can easily work out for yourself, is that you can rewrite the criterium for unequal delays and solve the equation with the effects of dead time, output current and ripple current taken into account. This allows you to compute quite precise output spectra very rapidly without actually having to simulate the circuit, should one be so inclined.
I'm glossing over a lot but at least I hope to give an idea of whether you want to enrich the AES by a tenner or not. By the way I realised that I'm entitled to send people the paper by e-mail so you could give a shout (pm) if you're seriously interested.
Last edited:
Certainly as the prestige of the measures, there is a difference.
But if Bruno can say that above 20Khz is not important, then I can tell you that you do not listen difference between 0.001% and 0.01%. then is good amplifier.
the ear is more sensitive to intermodulation distortion because it attempts to change the sound. our brain has stored parameters for reconstruct a snare is a snare. (to recognize)
Thank Bruno,
during the development of my amp, I have noted only a few steps of this paper.(today i read) eg. 360 ° is wrong (Obvious). and transfer function H (s) contains all info for me.
interesting the part .. indefinite.
(remember that i start DXA in 2006)
I can not explain how I used, I certainly have obtained a different result. exactly what I wanted.
I agree to the PSRR. I do not agree that this solves the timbre of some instruments fast "fast bass called" orchestral and other behaviors that require a strong and steady power. obvious that the amp has to be fast and with very low output impedance.
Regards
Who's said "over 20khz does not make sense"?
Who's said that the band shouldn't be extended because it's too complicated?
What Bruno has said several times is that
- in the first instance, performance under 20k is more important than that above 20k
- improving performance above 20k at the expense of performance below 20k doesn't make sense
- improving performance below 20k typically means improving the innate performance of the circuit and therefore typically inherently provides improved performance above 20k
Your implication is that the high frequency design & capabilities of the ucd & ncore have been deliberately compromised for of reasons of design complexity & cost. The opposite is the case. The designed performance of the ucd & ncore at and above 20k are excellent as can be seen by the IM performance at 20k
Why should an energy storage device only be part of a tuned filter?
Why does frequency matter?
Why am I bothering to write this reply?
You do seem to missing the point of several clearly explained design concepts when the rest of us here seem to be able to fill in the detail for ourselves
To all:
Bruno's F-word document is interesting reading.
How many other power amps with audiophile pretensions (or PA for that matter) have that much negative feedback?
One can certainly infer that good IMD performance readins just below 20k most likely would translate into reasonably similar performance above 20k simply based on common sense and on the fact the 20k boundary is in fact arbitrary, but the actual measured numbers would be cool too.
People talking of fast bass and taking a snare as an example when discussing THD and IMD belong more to the highend charlatan camp than to the engineering camp IMO.
Keep in mind that most amps - switching and linear - don't have as much feedback as the nCore and almost never does it stay at these levels within the audible band. Most amp's perfomance starts to worsen way below 20 kHz. So accusing Bruno of optimising the sub 20 kHz range at the cost of > 20 kHz is definitely quite naughty (to say the least).
Regards
Charles
Keep in mind that most amps - switching and linear - don't have as much feedback as the nCore and almost never does it stay at these levels within the audible band. Most amp's perfomance starts to worsen way below 20 kHz. So accusing Bruno of optimising the sub 20 kHz range at the cost of > 20 kHz is definitely quite naughty (to say the least).
Regards
Charles
Tell me, how would you achieve low IMD if you do not have low THD?
Must you really go on to demonstrate to all how deep a hole you can dig for yourself?
I'm not sure if you spotted it but onlookers are giggling in the rafters. Stop making a spectacle of yourself. Attached is the IMD plot of Ncore which you say is so much more important, and which everyone knew would be the answer to youe above question. Well, it's put up or shut up time. Either you show that you know what you're talking about by posting a better result, or you stop wasting everybody's time.
Earlier I described to StigErik the game you are playing. You offer no technical arguments. You're also using the classical disruptive technique of incoherent language even though you have accidentally demonstrated to speak perfectly acceptable English. In fact, you have consistently proved to understand my English flawlessly. In other words: you are putting up a show just to unhinge the discussion.
Attachments
At this point, the question changes constantly.
I never heard a company that presents a new amp (better than class AB), and then says that it is better to have good parameters in the 20Khz range, than compromise if the range is extended.
And, without a power supply capable of responding in real time to requests for the current (if it is regulated SMPS) or large capacitors (if linear psu). all know that the timbre is changed without this.
PSRR has nothing to do with this behavior.
I understand that Bruno, has made a quantum leap compared to UCD400.
If the current knowledge on selfoscillante not make it easy to achieve a prestigious class AB, this is not my fault.
I apologize to DIY, but I will not comment on other posts in this thread.
When you want to get my attention, show in full measure in the 50Khz range, show the dynamic behavior Unregulated SMPS. spectral analysis shows the SMPS. (this is also not easy to get). Or you just want things easy and say that they are .. "state of art"?
This is better, what about me. "I must go and study papers Bruno"
Regards
I never heard a company that presents a new amp (better than class AB), and then says that it is better to have good parameters in the 20Khz range, than compromise if the range is extended.
And, without a power supply capable of responding in real time to requests for the current (if it is regulated SMPS) or large capacitors (if linear psu). all know that the timbre is changed without this.
PSRR has nothing to do with this behavior.
I understand that Bruno, has made a quantum leap compared to UCD400.
If the current knowledge on selfoscillante not make it easy to achieve a prestigious class AB, this is not my fault.
I apologize to DIY, but I will not comment on other posts in this thread.
When you want to get my attention, show in full measure in the 50Khz range, show the dynamic behavior Unregulated SMPS. spectral analysis shows the SMPS. (this is also not easy to get). Or you just want things easy and say that they are .. "state of art"?
This is better, what about me. "I must go and study papers Bruno"
Regards
- Status
- Not open for further replies.