I was meaning input isolation for my K-multiplier, not PSRR.
The K-multiplier cannot equal a really huge capacitor, no...
EDIT: If big capacitors aren't a problem at low voltage though, maybe the K-multiplier has a use for high voltage supplies for tube amps. I wonder how much 300V, 22F caps go for these days?
- keantoken
The K-multiplier cannot equal a really huge capacitor, no...
EDIT: If big capacitors aren't a problem at low voltage though, maybe the K-multiplier has a use for high voltage supplies for tube amps. I wonder how much 300V, 22F caps go for these days?
- keantoken
Last edited:
There's not much argument that larger and ever larger RC filtering of the rails to the input stage will reduce rail ripple. As long as the design will then tolerate lower rail voltages, it seems a good idea - up to the point Kean has made. i.e. input connections allow noise in that swamps the best efforts at cleaning and polishing power supplies.
Many real amps even use a small resistor at the input, raising ground to limit earth currents into the amp, which can be a real problem for unbalanced inputs. It really becomes a little academic improving PSRR without serious consideration of what and how sources are connected and hoping to benefit from 'gilding the lily' at the rails.
Many real amps even use a small resistor at the input, raising ground to limit earth currents into the amp, which can be a real problem for unbalanced inputs. It really becomes a little academic improving PSRR without serious consideration of what and how sources are connected and hoping to benefit from 'gilding the lily' at the rails.
Ian, you are talking about noise and interference, but I don't think this is what we're dealing with.
To me, the mod works because it isolates the Jfet from the rail variations, which because of Early effect, induce second harmonics. Thus I think that the filtering is reducing H2 distortion, and this might seem to explain the subjective observations of Mikel. I recall in simulation the H2 was very sensitive to rail filtering.
- keantoken
To me, the mod works because it isolates the Jfet from the rail variations, which because of Early effect, induce second harmonics. Thus I think that the filtering is reducing H2 distortion, and this might seem to explain the subjective observations of Mikel. I recall in simulation the H2 was very sensitive to rail filtering.
- keantoken
Sure, I should have just stayed with "input" rather than dwell on external devices. I stick to my point about the inherent contribution of real input devices, however.
The PS noise arrives at the gate indiscriminately, rails or input, because the earth is the PS ground and the ripple relative to the source ground will reflect as input too. I don't think you can separate them in a system without shorting the input and that is pointless.
In any case, a cascode transistor removes the modulation of the input stage bias current and early effect. I think ~10dB improvement could be expected. That's about half the problem there and a buffer would help to maintain a low source impedance and again, minimise PS and external noise injection.
The PS noise arrives at the gate indiscriminately, rails or input, because the earth is the PS ground and the ripple relative to the source ground will reflect as input too. I don't think you can separate them in a system without shorting the input and that is pointless.
In any case, a cascode transistor removes the modulation of the input stage bias current and early effect. I think ~10dB improvement could be expected. That's about half the problem there and a buffer would help to maintain a low source impedance and again, minimise PS and external noise injection.
Just to clarify one point.
I am using choke regulation in my main PSU. Differential chokes to keep the rails clean and common mode chokes to keep the ground clean - so I already have very quiet supplies.
However, the power rails will still be modulated by the signal and the lower the frequency the more that the rails are susceptible to this modulation.
Unfortunately the 20R + 470uF filter originally proposed ( by me I think ) only really works well at high frequencies so although it clearly brought some gains it still left a lot of room for improvement some of which I noticed when I improved it !
If applied to a system without choke regulation I would imagine that there is still further room for improvement.
However, if a PSU is spewing RF + HF noise out onto the ground line and thus polluting it, as most PSU's do, as Ian points out, differential filtering of the rails is not going to clean up the ground.
Common mode chokes can help in this respect.
I know not everyone agrees with this but for me low noise is a prerequisite for good audio reproduction.
I am using choke regulation in my main PSU. Differential chokes to keep the rails clean and common mode chokes to keep the ground clean - so I already have very quiet supplies.
However, the power rails will still be modulated by the signal and the lower the frequency the more that the rails are susceptible to this modulation.
Unfortunately the 20R + 470uF filter originally proposed ( by me I think ) only really works well at high frequencies so although it clearly brought some gains it still left a lot of room for improvement some of which I noticed when I improved it !
If applied to a system without choke regulation I would imagine that there is still further room for improvement.
However, if a PSU is spewing RF + HF noise out onto the ground line and thus polluting it, as most PSU's do, as Ian points out, differential filtering of the rails is not going to clean up the ground.
Common mode chokes can help in this respect.
I know not everyone agrees with this but for me low noise is a prerequisite for good audio reproduction.
Last edited:
is choke regulation the same as choke input PSU?
Capcitor input PSU is what most use because it is cheaper and more consistent in output voltage with load.
Choke input can increase output voltage by ~50% when the load current momentarily drops.
progress . . . .
Please see below the circuit I've been listening to for the last few days.
I have made 4 significant changes over what I regard as the "Vanilla" version of the DC linked design.
1) Reduced drive impedance to 22 ohms
2) Added i/p stage CCS
3) Increased filter caps to 22,000uF
4) Put 3310a's in front of O/P devices to create MF "darlingtons"
( the diagram show 3306's but that actual amp has 3310's throughout. )
Each of these design changes brought a significant sonic improvement and together they have brought an already good design to a level of reproduction that I am very happy with.
As you can see I have been able to dispense with all gate stopper resistors and the bias spreader cap - I measured and heard no worsening of the sound with any of these omissions.
The amp appears to have an identical level of stability to the vanilla version and also appears to be unconditionally stable - Although I did not do exhaustive testing on this yet.
The i/p stage CCS has helped quite a lot with DC stability but I did not do a formal test on this yet either - need to buy a logger.
I cannot imagine there is anything more that I can now do with the design of this amp but would be interested if there are any further suggestions. It does not need less distortion but more speed would always be welcome.
mike
Please see below the circuit I've been listening to for the last few days.
I have made 4 significant changes over what I regard as the "Vanilla" version of the DC linked design.
1) Reduced drive impedance to 22 ohms
2) Added i/p stage CCS
3) Increased filter caps to 22,000uF
4) Put 3310a's in front of O/P devices to create MF "darlingtons"
( the diagram show 3306's but that actual amp has 3310's throughout. )
Each of these design changes brought a significant sonic improvement and together they have brought an already good design to a level of reproduction that I am very happy with.
As you can see I have been able to dispense with all gate stopper resistors and the bias spreader cap - I measured and heard no worsening of the sound with any of these omissions.
The amp appears to have an identical level of stability to the vanilla version and also appears to be unconditionally stable - Although I did not do exhaustive testing on this yet.
The i/p stage CCS has helped quite a lot with DC stability but I did not do a formal test on this yet either - need to buy a logger.
I cannot imagine there is anything more that I can now do with the design of this amp but would be interested if there are any further suggestions. It does not need less distortion but more speed would always be welcome.
mike
Attachments
Last edited:
If the .01R resistor is for Rds balancing, then you will need to wrap the 100R driver resistor around it as well to balance the current in the driver FET. Or you could use a single resistor across output MOSFET gates to bias both drivers, but I assume you haven't done this for a reason or it isn't significant to the sound.
Just for kicks you could also connect the VAS gate to the Jfet CCS source, so it gets some extra drive from the CCS...
- keantoken
Just for kicks you could also connect the VAS gate to the Jfet CCS source, so it gets some extra drive from the CCS...
- keantoken
Hi KT,
At present the 0.01R is simply to accurately measure the o/p stage bias current. I may play around later with a 0.1R to try and shape the HD signature but I wonder if I will hear it - the HD is pretty low now.
The composite output devices gave best stability in SPICE with the arrangement shown and I also liked it because it meant I could have very short leads connecting from the 3310's to the o/p MF's which I thought would be needed to successfully remove the o/p device gate stoppers. I soldered the 3310's directly onto the legs of the o/p devices.
Did u mean a kind of bootstrap between CCS & VAS i/p ?
If you did it might be odd as I have no cap across the bias spreader R and the bias would probably change with the signal - also my guess is that it would oscillate . . . but I'll try it in spice . . . . I can't blow that up!
Nice design, Mike.
Using a CFP output stage with mosfet driver and lateral slave is a very good idea. This would make it far less tetchy to set up the bias, probably nice and progressive on adjustment. My only concern, even in deep Class AB, is the switchoff profile of the outputs - is there any ringing? I doubt there is, in fact, since when the outputs switch off the opposite number would be passing a lot of current, so any switching events would be buried in the high output current at the switching.
One of the advantages of the bootstrap is a better treatment of vocals, but at the slight expense of bass. Using a CCS, as you have done here, gives you marvellous FR but at the top end where the bootstrap is starting to tail off you still have very good impedance, and this must be tamed with lag compensation. The bootstrap circuit can actually avoid using LC, which is a neat trick.
I was never happy with an additional CCS on the input device, but you have made it work well and from what I can gather you think the amp actually sounds better for having one less cap. Greg's and my approach was to bias up the gate of the input stage and run the stage current through the fb series resistor, a bit simpler, and since the input cap is now polarised it sounds good anyways.
It's a good and purist circuit, and you developed it largely on your own. Hats off..... and thank you. Does it have front to back depth in the sound field?
Cheers,
Hugh
Using a CFP output stage with mosfet driver and lateral slave is a very good idea. This would make it far less tetchy to set up the bias, probably nice and progressive on adjustment. My only concern, even in deep Class AB, is the switchoff profile of the outputs - is there any ringing? I doubt there is, in fact, since when the outputs switch off the opposite number would be passing a lot of current, so any switching events would be buried in the high output current at the switching.
One of the advantages of the bootstrap is a better treatment of vocals, but at the slight expense of bass. Using a CCS, as you have done here, gives you marvellous FR but at the top end where the bootstrap is starting to tail off you still have very good impedance, and this must be tamed with lag compensation. The bootstrap circuit can actually avoid using LC, which is a neat trick.
I was never happy with an additional CCS on the input device, but you have made it work well and from what I can gather you think the amp actually sounds better for having one less cap. Greg's and my approach was to bias up the gate of the input stage and run the stage current through the fb series resistor, a bit simpler, and since the input cap is now polarised it sounds good anyways.
It's a good and purist circuit, and you developed it largely on your own. Hats off..... and thank you. Does it have front to back depth in the sound field?
Cheers,
Hugh
Last edited:
Thanks for you nice comments Hugh, I appreciate that.
Actually the o/p's are MOSFET darlington pairs rather than CFP's.
I'm always tempted with CFP's because at the volumes I use, they would probably very rarely switch, but in this design the darlingtons seem to be just fine - perhaps I'll try the CFPs in spice and see if they offer any advantages
Yes, I tried both lead & lag compensation and I preferred lag - it just seems to sound more musical and I have a few tricks I am going to try with that soon.
I'm currently trying the amp with different speakers and so far I am delighted with how it has worked out - All the speaker have been very close to a rear wall but the new transmission lines give very nice sound staging within the available space.
Many thanks for all of your help with this Hugh, I feel like I learn't more about amp design in the last 6 months than I did in the last 10 years ! I think each of us, in our own way, got this design to a very good place.
cheers
mike
Actually the o/p's are MOSFET darlington pairs rather than CFP's.
I'm always tempted with CFP's because at the volumes I use, they would probably very rarely switch, but in this design the darlingtons seem to be just fine - perhaps I'll try the CFPs in spice and see if they offer any advantages
Yes, I tried both lead & lag compensation and I preferred lag - it just seems to sound more musical and I have a few tricks I am going to try with that soon.
I'm currently trying the amp with different speakers and so far I am delighted with how it has worked out - All the speaker have been very close to a rear wall but the new transmission lines give very nice sound staging within the available space.
Many thanks for all of your help with this Hugh, I feel like I learn't more about amp design in the last 6 months than I did in the last 10 years ! I think each of us, in our own way, got this design to a very good place.
cheers
mike
I just twigged what you meant here Hugh.
Actually I didn't remove the bias spreader cap because I wanted one less cap in the cct !
Rather, I discovered ( by accident ) that a 1V 44meg hz oscillation that appeared when I made o/p darlingtons and removed the o/p device gate stopper resistors, disappeared when the 4.7uF bias spreader cap was removed.
So removing this cap was an expedient measure that allowed me to use my preferred o/p arrangement ( which did sound better ! )
Also I just tried CFP o/p's in spice and they do indeed make the amp even faster and more stable. So provided there are not local oscillation problems when they switch they might bring another improvement.
cheers
mike
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.