Ah, yes Danny, the bias resistor on input transistor is indeed R1, my mistake.
And you could change that C6 value from 22p to 33p but it need not be doubled because the 100k nfb resistor has pulled back the loop gain so compensation can be relaxed a little.
BTW, the change of the nested fb to 100k will load down the VAS a little more and pull back the loop gain. This will improve the depth of image; there is a reaction here, and always more of this is less of that......... the downside is higher THD and higher Zout. But we are building an audio amp, not an instrument, and it must be tailored to the ear, not the camera.
HD
And you could change that C6 value from 22p to 33p but it need not be doubled because the 100k nfb resistor has pulled back the loop gain so compensation can be relaxed a little.
BTW, the change of the nested fb to 100k will load down the VAS a little more and pull back the loop gain. This will improve the depth of image; there is a reaction here, and always more of this is less of that......... the downside is higher THD and higher Zout. But we are building an audio amp, not an instrument, and it must be tailored to the ear, not the camera.
HD
Last edited:
HI Danny,
Tricky..... the global fb is set by the ratio of nested fb to fb series resistors. 220k/22k is 10, which is 20dB. If you reduce the nfb to 100k but remain R7 at 22k, the ratio more than halves, and this is too low; global fb at 13.1dB, too low for this amp. And if you halve R6, you should halve R1 too, to maintain offset. Hmmm, a conundrum.......
Interesting, you comment on depth of image. 90% of designers are horrified about this, because they see higher THD as BAD without realising it's related to the depth of image, which adds realism and 'engagement', which you cannot measure......
There is much pleasure in preaching to the choir, Danny.... you catch on very quickly.
Hugh
Tricky..... the global fb is set by the ratio of nested fb to fb series resistors. 220k/22k is 10, which is 20dB. If you reduce the nfb to 100k but remain R7 at 22k, the ratio more than halves, and this is too low; global fb at 13.1dB, too low for this amp. And if you halve R6, you should halve R1 too, to maintain offset. Hmmm, a conundrum.......
Interesting, you comment on depth of image. 90% of designers are horrified about this, because they see higher THD as BAD without realising it's related to the depth of image, which adds realism and 'engagement', which you cannot measure......
There is much pleasure in preaching to the choir, Danny.... you catch on very quickly.
Hugh
If the nfb is reduced to 100k, R1 and R7 remains at 22k,
but the shunt 820r FB resistor is set a little higher like 1.2k for some extra nfb ?
Will that work?
Depth of image is unrelated to THD for me, once the THD is low enough.
One of my first amplifier was a KT88 PP amp in which it was easy to adjust the FB,
without global FB I got the best 3D sound.
Loudspeakers at best have around -50db, so why aim for -120db in amplifiers ?
Better to transform those unneeded low THD in something useful like 3D depth of image and transparency
Grts,
Danny
but the shunt 820r FB resistor is set a little higher like 1.2k for some extra nfb ?
Will that work?
Depth of image is unrelated to THD for me, once the THD is low enough.
One of my first amplifier was a KT88 PP amp in which it was easy to adjust the FB,
without global FB I got the best 3D sound.
Loudspeakers at best have around -50db, so why aim for -120db in amplifiers ?
Better to transform those unneeded low THD in something useful like 3D depth of image and transparency
Grts,
Danny
Last edited:
Not quite, Danny. The ratio of global fb is still set by 100k/22k, but by reducing the shunt fb resistor to 820R you merely increase the linearity of the first and second stage. The gfb drives the performance of the output stage, and to keep Zout and bandwidth we keep the ratio around 20dB. In fact you can tolerate up to about 30dB of global feedback and maintain good depth of image, so going too low is not needed. 20dB is low in this situation because the output stage is highly linear since the quiescent is high and no device turns off.
You can reduce it and see what happens. I have designed amps with gfb down to 17dB, no lower, but using bipolars which have more transconductance than mosfets.
Your last observation give me the original idea of reducing gfb about twenty years ago and set me on the pursuit of depth of image, which as you know is the forte of tube amps. Seems obvious to me, but not well accepted. Purists think the lowest THD is the best, but the THD Japanese wars of the eighties and nineties did not convince me.
HD
You can reduce it and see what happens. I have designed amps with gfb down to 17dB, no lower, but using bipolars which have more transconductance than mosfets.
Your last observation give me the original idea of reducing gfb about twenty years ago and set me on the pursuit of depth of image, which as you know is the forte of tube amps. Seems obvious to me, but not well accepted. Purists think the lowest THD is the best, but the THD Japanese wars of the eighties and nineties did not convince me.
HD
Last edited:
OK, I get it, so the shunt fb of 820r is for the first and second stage,
but the gfb ratio of 100k/10K or 220k/22k is for the output stage.
In the first line I suppose you meant :
"but by augmenting the shunt fb resistor to 1.2k you merely increase the linearity of the first and second stage"
Danny
but the gfb ratio of 100k/10K or 220k/22k is for the output stage.
In the first line I suppose you meant :
"but by augmenting the shunt fb resistor to 1.2k you merely increase the linearity of the first and second stage"
Danny
Yes, well expressed. Augmented is quite correct, my mistake. But this is awkward to understand in any language........
You can see why once the amp is designed for 8R, setting it up for 4R does set up compromises and a new set of simulations has to be carefully weighed to get it right. Keeping global fb pretty much the same is important to maintain depth of image, but you'd expect THD to rise and DF to halve. Compensation is changed too, of course, because these three resistors bear hugely on the stability of the amp.
With lower impedance loading the distortion will rise inevitably, particularly by expanding the higher harmonics of the profile. There is an argument for increasing the nfb resistor by double to maintain the DF, which would be around 200k with the fb series resistor at the same 10k. That would be a gfb ratio of 26dB, 20 times and would relieve the VAS collector of its heavy load. But I would be quite sure that the depth of image would go out the window.........
HD
You can see why once the amp is designed for 8R, setting it up for 4R does set up compromises and a new set of simulations has to be carefully weighed to get it right. Keeping global fb pretty much the same is important to maintain depth of image, but you'd expect THD to rise and DF to halve. Compensation is changed too, of course, because these three resistors bear hugely on the stability of the amp.
With lower impedance loading the distortion will rise inevitably, particularly by expanding the higher harmonics of the profile. There is an argument for increasing the nfb resistor by double to maintain the DF, which would be around 200k with the fb series resistor at the same 10k. That would be a gfb ratio of 26dB, 20 times and would relieve the VAS collector of its heavy load. But I would be quite sure that the depth of image would go out the window.........
HD
Last edited:
Yep, better to keep the depth of image as priority one and sacrifice some THD for it
But for the DF there's your trick of adding a voltage divider around the sense resistor,
this will add also some FB but more for the bass,
does this divider also has an influence on the depth of image ?
Here's the update schematic for my next iteration of the Alpha:
10k input impedance + 10k/100k for gfb
33p for gfb
820r for shunt FB
66r-66r voltage divider around sense resistor for higher DF
Danny
But for the DF there's your trick of adding a voltage divider around the sense resistor,
this will add also some FB but more for the bass,
does this divider also has an influence on the depth of image ?
Here's the update schematic for my next iteration of the Alpha:
10k input impedance + 10k/100k for gfb
33p for gfb
820r for shunt FB
66r-66r voltage divider around sense resistor for higher DF
Danny
Attachments
Danny, that looks perfect, all values you have suggested check out nicely. I don't think the sense voltage divider will alter the depth of image.
HOWEVER:
When I designed the ALPHA, and then X built it in a week and had results to me a few days later (if not sooner!) in the intervening days I tried to figure out how to get more power from less rails. Efficiency is rather important on Class A because it is almost brute force, and keeping it as high as possible is necessary.
The output stage on the ALPHA is pure genius Pass, but in fact both devices were nmos. The upper nmos controlled the current - and the bottom device controlled the voltage on the load. I saw that the Aleph J relied on strong global fb to set the voltage at the load. This meant driving the controlling signal into the lower nmos gate then taking it out at the drain. I have always disliked common source; I really like source follower because with the signal driving from the source the intrinsic impedance is very low, as much as Rdson and typically 1/gm.
So for the ALPHA I substituted a pmos in place of the lower nmos. It worked like a charm, BUT it clips early on the negative half cycle because of the high Vgs, around 4.5V below the ouput 0V point. The VAS saturates and cannot bring the source voltage close the negative rail.
How to fix this?
The Nirvana! If you upend the output stage, sent the signal to the upper nmos and arrange the active CCS around the lower pmos, it works well and clips another four volts higher, giving you a symmetrical clip and greatly higher efficiency. Because the AKSA/Lender is set a few volts higher than the negative rail, if we drive the gate of the upper nmos which is 4.5V above the 0V output point, it should give wonderful clip, and the advantage is that the VAS never, ever saturates because at clip the collector is always about three voltages higher than the emitter. A saturated VAS is a disaster in an audio amp.......
This brings 50Vpp from a bipolar rail of 27V. VERY efficient.
This is a roundabout way of suggesting to you to build the Nirvana ALPHA, rather than the ALPHA 4R. I changed the output stage on the Nirvana too, it's a little simpler, and far easier to configure for a 4R load.
Ciao,
Hugh
HOWEVER:
When I designed the ALPHA, and then X built it in a week and had results to me a few days later (if not sooner!) in the intervening days I tried to figure out how to get more power from less rails. Efficiency is rather important on Class A because it is almost brute force, and keeping it as high as possible is necessary.
The output stage on the ALPHA is pure genius Pass, but in fact both devices were nmos. The upper nmos controlled the current - and the bottom device controlled the voltage on the load. I saw that the Aleph J relied on strong global fb to set the voltage at the load. This meant driving the controlling signal into the lower nmos gate then taking it out at the drain. I have always disliked common source; I really like source follower because with the signal driving from the source the intrinsic impedance is very low, as much as Rdson and typically 1/gm.
So for the ALPHA I substituted a pmos in place of the lower nmos. It worked like a charm, BUT it clips early on the negative half cycle because of the high Vgs, around 4.5V below the ouput 0V point. The VAS saturates and cannot bring the source voltage close the negative rail.
How to fix this?
The Nirvana! If you upend the output stage, sent the signal to the upper nmos and arrange the active CCS around the lower pmos, it works well and clips another four volts higher, giving you a symmetrical clip and greatly higher efficiency. Because the AKSA/Lender is set a few volts higher than the negative rail, if we drive the gate of the upper nmos which is 4.5V above the 0V output point, it should give wonderful clip, and the advantage is that the VAS never, ever saturates because at clip the collector is always about three voltages higher than the emitter. A saturated VAS is a disaster in an audio amp.......
This brings 50Vpp from a bipolar rail of 27V. VERY efficient.
This is a roundabout way of suggesting to you to build the Nirvana ALPHA, rather than the ALPHA 4R. I changed the output stage on the Nirvana too, it's a little simpler, and far easier to configure for a 4R load.
Ciao,
Hugh
Last edited:
Hi Hugh,
All good arguments.
I've also been thinking in that direction,
why not build the Nirvana instead of this Alpha updated ?
Especially since the Nirvana design is less compromised when configured for 4R.
But with my 93db speakers 4vpp is already loud, so I'll probably never use those extra vpp.
So what's left ?
The subjective sound quality of the Nirvana against the Alpha4R,
is the Nirvana better when played at 4R and 4vpp with 25v rails?
Maybe X can chime in ?
Regards,
Danny
All good arguments.
I've also been thinking in that direction,
why not build the Nirvana instead of this Alpha updated ?
Especially since the Nirvana design is less compromised when configured for 4R.
But with my 93db speakers 4vpp is already loud, so I'll probably never use those extra vpp.
So what's left ?
The subjective sound quality of the Nirvana against the Alpha4R,
is the Nirvana better when played at 4R and 4vpp with 25v rails?
Maybe X can chime in ?
Regards,
Danny
Or maybe using low Vgs lateral mosfets made for audio? (2SK1058/2SJ162 or equivalent types made by Exicon nowadays).
Not meant as criticism, but I am interested in the considerations of designers why they use bipolars, vertical or lateral mosfets in class A output stages.
There are several circuits wandering around this forum, all around 20-25 watt class A.
Bonsai opts for bipolars; Fab does it with audio latfets; Pass and Hugh pick vertical mosfets made for switching.
A choice for one of the options will more or less have implications on VAS stages as well, so on the amplifier design as a whole (and with possible differences in how the amp behaves sonically).
But maybe this is something for another thread.
There are several circuits wandering around this forum, all around 20-25 watt class A.
Bonsai opts for bipolars; Fab does it with audio latfets; Pass and Hugh pick vertical mosfets made for switching.
A choice for one of the options will more or less have implications on VAS stages as well, so on the amplifier design as a whole (and with possible differences in how the amp behaves sonically).
But maybe this is something for another thread.
Hi Danny,
My opinion on sound quality is often “the newest and latest amp sounds the best!”
But what I do is switch out and come back months later. Switch again and come back weeks or months later. So on the Alpha Nirvana I have listened to it over 2 cycles so far. The last tine comprehensively as an AB test with other amps designed by Hugh: the Deltic, a 65w Class AB; the Glass Harmony driven by the Melbourne- my best SE Class A zero global feedback and with inductive reactive load MOT; and compared to Hugh’s Omega - a souped up 39w Alpha 20; is also compared to an LM3886 chip amp. I don’t have 4 ohm speakers though. My speakers however, are very low sensitivity (82.5dB) true 8ohms. They need a lot of power and can clip 25w amps easily.
My impression was that the Alpha Nirvana is perhaps the best amp beneath Hugh’s pinnacle of amp design in the Glass Harmony. It’s close but the GH is the best amp I have heard and this has been confirmed time and time again with long term testing and swapping in and out of rotation. Each time GH comes back to play, I still get the WOW! That’s a fantastic sounding kick @ss amp and it’s not even DC coupled, it’s cap coupled! The grip on bass transients and high-level detail burried in the top crest of a crescendo can be resolved clearly. Along with an absolute blacker than black noise floor - I don’t think many amps DIY or commercial can touch it. I actually feel really lucky to have one to listen to.
Now getting back to the Alpha Nirvana. It doesn’t have a dual monoblock configuration so may not be fair. Also, the Glass Harmony has zero global feedback. It’s phase information is pristine up to Mhz but we filter it off at 300kHz.
The Alpha has about 40w vs the GH’s 50w. Not much different and despite not having dual monoblocks, sounds surprisingly close - maybe 90% of the way to the GH on most material. However, the GH being a true single ended (single rail) that is AC-coupled, will always sound smoother even played loud (past 25wrms). Like a glassy surface of a lake undisturbed by wind or waves. The name is aptly appropriate.
But in summary, the Alpha Nirvana is probably the best sounding public open source design DIY accessible Class A amp in my opinion. It doesn’t need CPU coolers or fans when using a 4U x 300mm chassis either. Very accessible construction.
Last edited:
My opinion on sound quality is often “the newest and latest amp sounds the best!”
I was afraid you were going to say this
Your speakers are at 82.5db, that is really low,
when playing the same db level, 4vpp on mine is like 48vpp on yours.
Suppose if I go for the Alpha Nirvana,
the rails of my PS are now at 25vdc, is this OK for the Nirvana 4R/8R?
Although I think that the question of 4R or 8R Nirvana is not so relevant when using high sensitive speakers that play loud with only 1-4vpp.
I think it's time to do some Nirvana simulation in LTspice to see the behavior at different loads, vpp and rail voltage
Grts,
Danny
Last edited:
You state "probably", and "in my opinion", which makes sense
.Did you compare with the other class A designs around here (Hifisonix kx; USSA-5)?