Yes, the servo must be responsible for the spike. And since it is there in the real circuit, I don't see why I should remove it for simulation purposes. I haven't built anything yet, but why should the sim predict a spike that isn't there in real life? I'd bet that the real amp will show the gain peak, just like the simulation does. Anyone with a Mooly Amp and a suitable signal generator at hand?
That'll work, too. The advantage of a smaller C6 would be that the feedback resistors could be reduced by a factor of 10 as well, improving the noise performance. Increasing the servo cap would necessitate another electrolytic though.
When you say C6, are you referring to the numbering from your post or the original schematic? They are different which can be confusing. If you are referring to the NFB blocking cap (C6 on yours, C3 on the original), I don't follow your reasoning about reducing the feedback resistors. If the resistors are smaller, the cap must be larger to maintain the same roll-off frequency.
Regarding the Servo cap, you can stick with a polyester cap. For example the WIMA MKS2 series with a 5mm pitch and 7.2mm x 8.5mm footprint (Digikey 1928-1630-ND). Larger than a electrolytic cap, but still reasonably sized.
I would think addressing this at the servo would be preferred. Reducing the feedback cap would increase the -3db corner frequency. 470Ω & 470µF is 0.72Hz where 470Ω & 47µF is 7.2Hz. We generally see most designs targeting for somewhere around 2Hz.
Regarding the Servo cap, you can stick with a polyester cap. For example the WIMA MKS2 series with a 5mm pitch and 7.2mm x 8.5mm footprint (Digikey 1928-1630-ND). Larger than a electrolytic cap, but still reasonably sized.
I would think addressing this at the servo would be preferred. Reducing the feedback cap would increase the -3db corner frequency. 470Ω & 470µF is 0.72Hz where 470Ω & 47µF is 7.2Hz. We generally see most designs targeting for somewhere around 2Hz.
Yes, I'm referring to C6 on my screenshots. That's from the original simulation file supplied by Mooly; I did not notice that the numbering is different from the original schematic.
Instead of reducing C6 from 470uF to 47uF to get rid of the gain peak, you could leave it at 470uF but reduce the 22k/470R feedback resistors to 2k2/47R. That will flatten the frequency response in the same manner, but reduce the noise contribution of R19 at the same time.
Give it a try in the sim. Your calculation does not apply in this case, since the servo response is messing it up - that's the gain peak.
Instead of reducing C6 from 470uF to 47uF to get rid of the gain peak, you could leave it at 470uF but reduce the 22k/470R feedback resistors to 2k2/47R. That will flatten the frequency response in the same manner, but reduce the noise contribution of R19 at the same time.
Give it a try in the sim. Your calculation does not apply in this case, since the servo response is messing it up - that's the gain peak.
Here is a guide to Prasi's PCBs. Maybe Mooly can add this to Post #1.
Post #694: Prasi's 2 pair PCB (135.59mm x 91.44mm )
Post #678: Nested Feedback described (by Hugh Dean aka: AKSA)
Post #699: Prasi's 2 pair PCB with nested feedback option via R9 (135.59mm x 91.44mm )
Post #723: Prasi's 1 pair PCB with nested feedback option via R9 (100.00mm x 91.44mm )
Post #739: Prasi's BOM
Post #749: Prasi's updated BOM
Post #755: Prasi's drill template
Post #904: Prasi's 2 pair compact PCB with nested feedback option via R9 (100.08mm x 61.47mm ) with nested FB option (R9)
Altering the feedback resistors has been discussed several times earlier in this thread. Mooly has cautioned against it as it can disturb the DC operating points. Not trying to dissuade you, just proceed with caution.
Maybe Mooly can chime in with his thoughts.
It is interesting that the approach you're suggesting flattens the response. Does this increase the DC offset beyond the servo's authority? Or put it close to the edge if it's authority?
The latter. DC offset is now closer to -1V instead of -5V, which might be too close to the positive rail of the opamp (i.e. ground) to work reliably. Good catch. So better leave the resistors as-is and just reduce C6 instead. Or increase C7.
Tinkering with this in simulations, I found that reducing the feedback cap C3 from 470µF to 220µF and increasing the servo cap C8 from 0.1µF to 2.2µF removed the bump and seems to maintain the stability of the design. The change in the feedback corner frequency is 0.72Hz to 1.54Hz. This seem reasonable to me.
Looking at Prasi's PCBs, the only issue I see is that C8 allows for a 5mm pitch and 2.5mm width. Looking at available polyester Film caps, it looks like the smallest are 7.2mm x 5mm. Looking at his designs in Posts 699 (2 pair) & 723 (1 pair), these could likely be tweaked to accommodate a wider cap.
Looking at Prasi's PCBs, the only issue I see is that C8 allows for a 5mm pitch and 2.5mm width. Looking at available polyester Film caps, it looks like the smallest are 7.2mm x 5mm. Looking at his designs in Posts 699 (2 pair) & 723 (1 pair), these could likely be tweaked to accommodate a wider cap.
But that certainly increases the settling time for the output offset. The sim still shows almost 400mV after 10 seconds now, whereas the original circuit and my proposal are well below 20mV by then.
Hi Brian,
I think many polyester film caps are available in 2.5mm wide 5mm pitch 7.2mm length 0.1uf 63V/100V.
https://www.mouser.in/c/passive-com...-capacitors|~Capacitance|~Width|~Lead Spacing
I think many polyester film caps are available in 2.5mm wide 5mm pitch 7.2mm length 0.1uf 63V/100V.
https://www.mouser.in/c/passive-com...-capacitors|~Capacitance|~Width|~Lead Spacing
It is a real issue and not just a sim thing. I think the reason is that the servo output is used to derive the total input bias voltage needed (about -5.6 volts in the sim and in the real builds). Any AC signal voltage appearing at the servo output is added (or subtracted depending on phase) and that AC voltage increases as frequency decreases. There must also be as you say some other interaction with the various time constants in the circuit.
Years later I did a sim with a 'real' power supply to look at the switch on behaviour and increasing the servo cap value may something that could be tolerated, particularly with a longer speaker delay.
Standard values. Its instructional to run the sim also with a stop time of say 30 seconds and a start point of say 10 seconds in to see the settling time.
And with a 0.47u for the servo.
Its around twenty years since I put this together and there are undoubtedly things I would do differently today but what I can say is the design certainly delivers the sonics and it is possible that ironing out any perceived issues might actually work against sound quality... or not... it is an unknown.
In many ways it is more than just a servo because as mentioned earlier, it provides the high DC bias voltage to set the operating point of the input stage. Most amps you see with a servo will operate quite happily without it as they use conventional input stages with a differential pair and if the design is good they have low DC offset anyway.
This is the file with a transformer supply to show settling time. You'll have to do the necessary to suit your own model file locations.
Thanks Mooly! So if one wanted to tinker, it looks like increasing C8 up to perhaps a max of 0.47u might be a reasonable place to start. A cap of this size would be 3mm wide and likely fit within Prasi's boards.
Along with the value of the feedback cap C3 - dropping from 470 to 220. These would reduce the bump and likely keep the settling time reasonable.
It would be interesting to see if the perceived sonics change at all. Or if this is simply an academic exercise. Either way, it's been interesting. Seems like I'm always learning something new from this thread.
Along with the value of the feedback cap C3 - dropping from 470 to 220. These would reduce the bump and likely keep the settling time reasonable.
It would be interesting to see if the perceived sonics change at all. Or if this is simply an academic exercise. Either way, it's been interesting. Seems like I'm always learning something new from this thread.
Yes, that's certainly possible and remains to be seen, errm, heard. Again, I am not after modifying this design willy-nilly, but am concerned that it might have some serious side effects unless a good subsonic filter is used in front of it. My intention was merely to try and fix what appears to me as a shortcoming.
You should also be able to increase the 1meg given that the opamp is a FET type.
The more you look into this and the more involved it becomes. The source impedance driving the preamp seems to be a major player here. Use the original values and add 5k source impedance and you get this:
Using your modified values and the source impedance effect is much less pronounced. Interesting.
C1 and R2 seem to also be at play. I was trying different values here to accommodate Prasi's boards. His boards have provisions for 22.5mm pitch for C1. If I want to use WIMA MKP4 series in this position, the largest value I can use it 2.2u at that pitch. R2 needs to increase to 150K to keep the same -3db frequency. When I do this with all the other values being original, I notice that the circuit is less sensitive to higher source impendence values.
If you also throw in the C8 change from 0.1u to 0.47u and reduce C3 from 470u to 220u, the low frequency hump drops from +11.8dB at 6Hz to +3.7dB at 3Hz.
I do note that settling time increases a few seconds. Something that likely needs to be accounted for with a mute delay provision.
If you also throw in the C8 change from 0.1u to 0.47u and reduce C3 from 470u to 220u, the low frequency hump drops from +11.8dB at 6Hz to +3.7dB at 3Hz.
I do note that settling time increases a few seconds. Something that likely needs to be accounted for with a mute delay provision.