Hi,
combining some ideas I've seen in Roender, ULD, Blameless and Krill, what about doing a 3pair NJL output stage?
Incorporate all six integrated diodes into the bias string.
Add a parallel variable resistor to two of these diodes.
When VR=0r0 there are effectively four series diodes in the bias string and we know the output is tempco undercompensated.
When VR=1k0 there is ~1.3mA passing the VR and all the remainder of the bias current passing the two diodes. This will probably be overcompensated.
At some adjustment between 0r0 and 1k0 the compensation "may" be about right. Maybe the VR should be 500r or 2k or 5k, I don't know!
Does this provide room for further thoughts?
Can we save the Vbe multiplier for a "Blameless extreme" thread?
Keep the 2 or 4 or 6 NJLs for the "ULD extreme" thread.
AKSA,
once again thanks for your thoughts (close to your heart) on improving the extremes.
combining some ideas I've seen in Roender, ULD, Blameless and Krill, what about doing a 3pair NJL output stage?
Incorporate all six integrated diodes into the bias string.
Add a parallel variable resistor to two of these diodes.
When VR=0r0 there are effectively four series diodes in the bias string and we know the output is tempco undercompensated.
When VR=1k0 there is ~1.3mA passing the VR and all the remainder of the bias current passing the two diodes. This will probably be overcompensated.
At some adjustment between 0r0 and 1k0 the compensation "may" be about right. Maybe the VR should be 500r or 2k or 5k, I don't know!
Does this provide room for further thoughts?
Can we save the Vbe multiplier for a "Blameless extreme" thread?
Keep the 2 or 4 or 6 NJLs for the "ULD extreme" thread.
AKSA,
once again thanks for your thoughts (close to your heart) on improving the extremes.
why is a resistor//diode string "a bit risky"?AKSA said:
If 300r were in parallel to the 6diodes it would pass ~12mA. The remaining 2mA would pass through the diodes. Is there some capacitance and/or inductance in these routes that creates a resonance/oscillation risk?
NB.
as the other VR across just two of these diodes is adjusted, the 300r current will move a bit away from 12mA.
At 11mA through the 300r, the diodes string will have 3.3V across it.
At 10mA through the 300r, the diodes string will have 3.0V across it.
Maybe the 300r should be a VR=500r??
The setting up procedure becomes ever more complicated as more VR are added. A 400mW 500r can pass a maximum ~28mA. 12mA is <=42% of max rating.
Regarding Hugh's item '8. Replace all output device emitter resistors with 0.22R.': I switched from 0R22's to 0R1's this morning, and couldn't hear any difference. However I did notice that the stability of the voltage across the emitter resistors was, after this change, less well matched, and less stable. I also know from my earlier mistakes (heatsink too small, and scratched heatsink) that the 0R1's are much less forgiving of thermal issues. This means that with 0R1's the builder has to be more careful and experienced, and also can't drive the bias up as high.
On the other hand, Self's treatise that led to the blameless design notes that larger emitter resistors do add distortion.
So, there may be some philosophical decisions - do we make it a little less 'blameless' if that increases robustness? In other words: Is this project about making the most blameless of blameless amps, or is it about using the blameless amp as inspiration, but making pragmatic compromises if required? I suspect the two approaches could lead to quite different designs...
On the other hand, Self's treatise that led to the blameless design notes that larger emitter resistors do add distortion.
So, there may be some philosophical decisions - do we make it a little less 'blameless' if that increases robustness? In other words: Is this project about making the most blameless of blameless amps, or is it about using the blameless amp as inspiration, but making pragmatic compromises if required? I suspect the two approaches could lead to quite different designs...
Jeremy,
The distortion benefits of 0.1R emitter resistors are really only evident with four ohm and less loading. The idea is to reduce waveform compression at peaks with less voltage drop on the emitter resistors. Accordingly, if you are using low impedance loads, the 0.1R is OK, though for high current loads you'd be well advised to go three pairs of output devices for 56V rails.
Another interesting fact is that the higher this resistor, the lower can be set the quiescent current. The math, based on crossover considerations, indicates that 26mV (derived from the Boltzmann constant) should be dropped across this resistor regardless, so 0.22R is passing 120mA and 0.1R SHOULD strictly be passing 260mA. It isn't in this design, so immediately there is one compromise....
The benefit of 0.22R, OTOH, is that you won't notice any degradation with an 8R load, and quiescent current will match far better, and it's great insurance for the outputs in the event of an accident, which as you realise, is inevitable.....
In my commercial designs, I go whole hog and use 0.47R. Quite ridiculous really, however, they never blow up, the bias is always super stable, and since these resistors are inside the global feedback loop I still have a damping factor around 80, indicating source impedance is actually around 0.1 ohm.
Hugh
The distortion benefits of 0.1R emitter resistors are really only evident with four ohm and less loading. The idea is to reduce waveform compression at peaks with less voltage drop on the emitter resistors. Accordingly, if you are using low impedance loads, the 0.1R is OK, though for high current loads you'd be well advised to go three pairs of output devices for 56V rails.
Another interesting fact is that the higher this resistor, the lower can be set the quiescent current. The math, based on crossover considerations, indicates that 26mV (derived from the Boltzmann constant) should be dropped across this resistor regardless, so 0.22R is passing 120mA and 0.1R SHOULD strictly be passing 260mA. It isn't in this design, so immediately there is one compromise....
The benefit of 0.22R, OTOH, is that you won't notice any degradation with an 8R load, and quiescent current will match far better, and it's great insurance for the outputs in the event of an accident, which as you realise, is inevitable.....
In my commercial designs, I go whole hog and use 0.47R. Quite ridiculous really, however, they never blow up, the bias is always super stable, and since these resistors are inside the global feedback loop I still have a damping factor around 80, indicating source impedance is actually around 0.1 ohm.
Hugh
Andrew,
My pleasure, enjoy the dialogue.
I agree with you about three pairs. Should people use this amp with <4R loads, three pairs is, in my opinion, almost mandatory for 56V rails. With 50V rails, two pairs would be fine, but SOAR considerations lead me to agree emphatically.
If you have the VAS current splitting into two current paths, one down five diodes and a pot at 2mA, and the other down a 210R resistor at 12mA, then since the diode voltage drop is controlling the outputs, we want that control rigid and low impedance.
2mA controlling 12mA without current gain is not a good look. It's not my preferred way of doing it. I really would like to use a Vbe multiplier, however, others may differ on this and that's cool. I hark back to Mihai's comments about the mismatched tempcos; I really think these devices need more development work, and they are a bitch to lay out on the pcb with their five leads.
Hugh
My pleasure, enjoy the dialogue.
I agree with you about three pairs. Should people use this amp with <4R loads, three pairs is, in my opinion, almost mandatory for 56V rails. With 50V rails, two pairs would be fine, but SOAR considerations lead me to agree emphatically.
If you have the VAS current splitting into two current paths, one down five diodes and a pot at 2mA, and the other down a 210R resistor at 12mA, then since the diode voltage drop is controlling the outputs, we want that control rigid and low impedance.
2mA controlling 12mA without current gain is not a good look. It's not my preferred way of doing it. I really would like to use a Vbe multiplier, however, others may differ on this and that's cool. I hark back to Mihai's comments about the mismatched tempcos; I really think these devices need more development work, and they are a bitch to lay out on the pcb with their five leads.
Hugh
sandyK said:
Hi Sandy,
Loudspeakers add more distortion than any modern reasonable amp. If runningmultiple drivers in isobarick reduces this distortion then I can put up with a bit more in the amp. I woul like to see this thread with a few options in power output. As to the 80s Home theartre came along and stuffed up the good years
Terry
Terry,
Speakers introduce mostly low order distortions (H2-H5) and phase shift, which is more problematic, and requires great care on the crossover design. Amps introduce many high order harmonics, particularly push pull, and specifically Class AB, and these are NOT benign like a speaker, and sound terrible.
It is difficult to compare qualitatively; the distortion mechanisms of speakers and amps differ hugely.
Poorly designed speaker crossovers produce serious phase shift which plays merry hell with global feedback amps, threatening stability by seriously reducing phase margin. The performance of many speaker crossovers I've seen is far from satisfactory, and with modern software there is no excuse for this. Oddly, some of the better speakers are the worst offenders. The speaker/amp/source is a system, and often the components are designed in isolation, which highly compromises the synergy.
Cheers,
Hugh
Speakers introduce mostly low order distortions (H2-H5) and phase shift, which is more problematic, and requires great care on the crossover design. Amps introduce many high order harmonics, particularly push pull, and specifically Class AB, and these are NOT benign like a speaker, and sound terrible.
It is difficult to compare qualitatively; the distortion mechanisms of speakers and amps differ hugely.
Poorly designed speaker crossovers produce serious phase shift which plays merry hell with global feedback amps, threatening stability by seriously reducing phase margin. The performance of many speaker crossovers I've seen is far from satisfactory, and with modern software there is no excuse for this. Oddly, some of the better speakers are the worst offenders. The speaker/amp/source is a system, and often the components are designed in isolation, which highly compromises the synergy.
Cheers,
Hugh
Regarding the SQ of different harmonics - it has occurred to me that Self's design is only "blameless" in terms of minimising THD. However, Lee and Geddes (Auditory Perception of Nonlinear Distortion, 2003 ) found that THD could actually be negatively correlated with SQ in some cases. Instead, they proposed a metric based on the 2nd derivative (i.e. the slope) of the transfer function, weighted such that distortion close the the 0V crossover was higher. They found that this metric was very highly correlated with SQ. This is the 'GedLee' metric. (I'm sure most of this is well known already to the folks on this forum - I'm just trying to provide some background for anyone who hasn't seen this research).
So I would have thought that a truly 'blameless' amp would minimise GedLee, not THD. In general, decreases to THD should decrease GedLee too, but this isn't guaranteed to be the case.
I have no idea what, if any, implication this would have on the actual design of the amp...
So I would have thought that a truly 'blameless' amp would minimise GedLee, not THD. In general, decreases to THD should decrease GedLee too, but this isn't guaranteed to be the case.
I have no idea what, if any, implication this would have on the actual design of the amp...
jp_howard said:
How did you figure that out?
From the introductory abstract:
“No significant relationships were observed when comparing the subjective ratings with THD and IMD metrics”
No evidence there that, in any case, a low THD amp = a bad sounding amp whatsoever.
Also, I would love to see anyone apply the Gm metric to the evaluation of an audio power amplifier.
Cheers,
Glen
Have a look at the graph, or stick the data into a pearson correlation test yourself. You'll see that in their data there is a negative correlation between SQ and THD (by which I mean "worse SQ correlates with 'better' THD". It is however, not statistically significant at p=0.05.G.Kleinschmidt said:
True. However it does show that in some cases better THD can be associated with worse SQ.
As would I. According to the GL results, we should find no correlation, because all modern solid state amps (AFAIK) would have GL metric <1.0, which should mean differences are not audible. This is of course a conclusion which many would disagree with - personally I don't have enough expertise or experience to have an opinion yet, besides which it's a can of worms I'd rather not open!
It's very difficult to do proper comparative testing of amps, which I guess is one reason we see so many debates about amp SQ differences. There has been tests like http://www.matrixhifi.com/contenedor_ppec_eng.htm and of course Richard Carver's stuff, but AFAIK there are not peer-reviewed articles based on fully documented, replicable, scientifically designed tests.
But I don't think that there's any need to resolve any of these long-standing questions in order to develop this particular project - if we agree it's about building a "blameless" amp, rather than "an amp everyone agrees is good", then there's a more objective goal that all can work towards. Hopefully as a result we end up with something we like listening to as well! (I'm certainly enjoying listening to the results of my playing with the amp each day...)
jp_howard said:
jp_howard said:
Well you have to be very careful with your wording here. You cannot reach a conclusion or state that a negative correlation between THD and sound quality was "found" if the results are not statistically significant.
Sure, there are instances where 20% THD can sound less objectionable than 10% THD, but that is a different thing.
Cheers,
Glen
G.Kleinschmidt said:I take that back. Don't look at their graph, because they've rather amateurishly attempted to draw a linear relationship through data that clearly isn't linear. Since SQ's variance increases with size, we need to use ln(SQ) in the graph. Here's the data with this correction made:jp_howard said:
(x-axis is THD, y-axis is ln(SQ)).
Doesn't change the conclusion (no significant relationship, but some negative correlation) mind you.
OK, so here's why this conclusion is important for this project: since there are instances where lower THD may not lead to better SQ, this means that an amp with low THD is not necessarily 'blameless'.
The validity of the GedLee metric to amp distortion is not yet well understood (AFAIK). However, since (unlike THD) no-one has yet shown a type of distortion that measures better on GedLee but has worse SQ (in a significantly large sample), maybe an amp with the lowest possible GedLee really would be 'blameless'. This is purely speculation, of course, but I think it's an interesting thought...
The validity of the GedLee metric to amp distortion is not yet well understood (AFAIK). However, since (unlike THD) no-one has yet shown a type of distortion that measures better on GedLee but has worse SQ (in a significantly large sample), maybe an amp with the lowest possible GedLee really would be 'blameless'. This is purely speculation, of course, but I think it's an interesting thought...
I think some different terminology is required here.
As far as I can tell, "Blameless", as coined by D. Self, means an amplifier that doesn’t distort the audio signal in any demonstrably audible or significant way. So in this regard, an amplifier with low THD throughout the audio spectrum is indeed more likely to be “Blameless” than one with high THD.
Cheers,
Glen
As far as I can tell, "Blameless", as coined by D. Self, means an amplifier that doesn’t distort the audio signal in any demonstrably audible or significant way. So in this regard, an amplifier with low THD throughout the audio spectrum is indeed more likely to be “Blameless” than one with high THD.
Cheers,
Glen
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.