You have to earth the enclosure of the amplifier for safety. It's not necessary to connect the star ground directly to the earth. 10ohm+100nF is good enough.
I use almost same grounding schema, the only difference, that I not connect the input connector's gnd to the star ground.
Sajti
Please post a pic of how you would earth Toni's amp.
You can bodge my pic if you like.
I mention twisting transformer wires as just ONE of the things that ISN'T shown in my pic. Self has loads more.
Suckout caps etc
There is a subtle difference between how 'pure Cherry', with its Holy HiZ VAS collector, behaves with suckout / bias spreader caps. ... and lesser compensation schemes like plain Miller, MIC, TPC, TMC ... everything else.
These lesser, evil schemes all take some feedback from the VAS collector and make it sorta LoZ. But that means the VAS MUST be coupled tightly to BOTH drivers else they will be driven differently.
Hence the need for loadsa big bias spreader, suckout caps etc. for good THD.
'Pure Cherry' however, has Holy HiZ at that point. It sorta doesn't matter if the current output of the VAS goes to the +ve or -ve driver and hence what's between the driver bases is of less importance. The caps. are chosen for stability and only have small effect on distortion.
In most cases, they can be much smaller or even removed completely. eg #4 of tpc-vs-tmc-vs-pure-cherry
I'm guilty of not pointing out this evil behaviour when I put out my challenge to TMC & other evil pundits to beat 'pure Cherry'. Mea maxima culpa
Toni, for TMC, you need a big C11
____________________
But each case needs to be looked at individually .. in SPICE and also in 'real life'. eg
In #271, I was looking at V(outa) to see some of the stuff that Toni saw in real life. You clearly see the wriggles on the top of the squares with different C16. But I couldn't get the wriggles on the upward slope ...
.. until I looked at V(outa) without C16 on a 8R load. NoC16.gif clearly shows the wriggles on the upward slope. C16-470n+10.gif shows none. Yes I know this isn't what Toni found.
Conclusion: more work to be done but neither 'no C16' or 'C16 470n' are optimum.
___________________
'Pure Cherry' always uses smaller (sometimes no) spreader, suckout caps compared to lesser, evil compensation schemes. That's one reason why I like it. (Another reason is less & smoother 20kHz THD)
Small caps mean less changes on overload and quicker recovery. That's why #4 of tpc-vs-tmc-vs-pure-cherry has excellent overload & recovery without extra clamping diodes.
There is a subtle difference between how 'pure Cherry', with its Holy HiZ VAS collector, behaves with suckout / bias spreader caps. ... and lesser compensation schemes like plain Miller, MIC, TPC, TMC ... everything else.
These lesser, evil schemes all take some feedback from the VAS collector and make it sorta LoZ. But that means the VAS MUST be coupled tightly to BOTH drivers else they will be driven differently.
Hence the need for loadsa big bias spreader, suckout caps etc. for good THD.
'Pure Cherry' however, has Holy HiZ at that point. It sorta doesn't matter if the current output of the VAS goes to the +ve or -ve driver and hence what's between the driver bases is of less importance. The caps. are chosen for stability and only have small effect on distortion.
In most cases, they can be much smaller or even removed completely. eg #4 of tpc-vs-tmc-vs-pure-cherry
I'm guilty of not pointing out this evil behaviour when I put out my challenge to TMC & other evil pundits to beat 'pure Cherry'. Mea maxima culpa
Toni, for TMC, you need a big C11
____________________
But each case needs to be looked at individually .. in SPICE and also in 'real life'. eg
In #271, I was looking at V(outa) to see some of the stuff that Toni saw in real life. You clearly see the wriggles on the top of the squares with different C16. But I couldn't get the wriggles on the upward slope ...
.. until I looked at V(outa) without C16 on a 8R load. NoC16.gif clearly shows the wriggles on the upward slope. C16-470n+10.gif shows none. Yes I know this isn't what Toni found.
Conclusion: more work to be done but neither 'no C16' or 'C16 470n' are optimum.
___________________
'Pure Cherry' always uses smaller (sometimes no) spreader, suckout caps compared to lesser, evil compensation schemes. That's one reason why I like it. (Another reason is less & smoother 20kHz THD)
Small caps mean less changes on overload and quicker recovery. That's why #4 of tpc-vs-tmc-vs-pure-cherry has excellent overload & recovery without extra clamping diodes.
Attachments
Last edited:
kgrlee: very nice drawing. It correctly demonstrates my grounding schema. Thanks for the RFI hint.
See my attached picture which explains the [?].
The [?] contains a 35A bridge-rectifier used as antiparallel double diodes and 100nF/10R. In Austria every device with metal case has to be grounded/earthed for safety and you would easly get multiple groundloops. If we can lift the RCA shielding above earth/gnd by 0.7V we can effectively fight the hum. As 2 parallel diodes of this bridge rectifier are capable to drive 70A and peak more as 800A every fuse will be blown by this combination. sajti's recommendation only to use 10R without diodes can be letal - think of a defective toroid transformer where 235V find it's way to PowerGND.
Regarding "two monoblocks in one case": yes, all is doubled. There are two toroids, two power supplies, two amplifiers in one case.
Attachments
Last edited:
Thanks for contributing the plots. The residual looks nearly the same as in real life on scope. The higher level at crossover from minus to plus is also viewable. May be one of the effects caused by suckout cap C16 if it's too big? Will try in real life.
There is a subtle difference between how 'pure Cherry', with its Holy HiZ VAS collector, behaves with suckout / bias spreader caps. ... and lesser compensation schemes like plain Miller, MIC, TPC, TMC ... everything else.
These lesser, evil schemes all take some feedback from the VAS collector and make it sorta LoZ. But that means the VAS MUST be coupled tightly to BOTH drivers else they will be driven differently.
Hence the need for loadsa big bias spreader, suckout caps etc. for good THD.
'Pure Cherry' however, has Holy HiZ at that point. It sorta doesn't matter if the current output of the VAS goes to the +ve or -ve driver and hence what's between the driver bases is of less importance. The caps. are chosen for stability and only have small effect on distortion.
In most cases, they can be much smaller or even removed completely. eg #4 of tpc-vs-tmc-vs-pure-cherry
I'm guilty of not pointing out this evil behaviour when I put out my challenge to TMC & other evil pundits to beat 'pure Cherry'. Mea maxima culpa
Toni, for TMC, you need a big C11
____________________
But each case needs to be looked at individually .. in SPICE and also in 'real life'. eg
In #271, I was looking at V(outa) to see some of the stuff that Toni saw in real life. You clearly see the wriggles on the top of the squares with different C16. But I couldn't get the wriggles on the upward slope ...
.. until I looked at V(outa) without C16 on a 8R load. NoC16.gif clearly shows the wriggles on the upward slope. C16-470n+10.gif shows none. Yes I know this isn't what Toni found.
Conclusion: more work to be done but neither 'no C16' or 'C16 470n' are optimum.
___________________
'Pure Cherry' always uses smaller (sometimes no) spreader, suckout caps compared to lesser, evil compensation schemes. That's one reason why I like it. (Another reason is less & smoother 20kHz THD)
Small caps mean less changes on overload and quicker recovery. That's why #4 of tpc-vs-tmc-vs-pure-cherry has excellent overload & recovery without extra clamping diodes.
Current test is using 100nF/33R as suckout combination. Big enough to smooth out the "wriggles" running square wave tests. Only small effect on THD at high freqency. As the fc is > 48kHz should have no impact on audio.
Real life THD+N@40kHz@200W@8R: 0.004/0.008% (Bandwidth80kHz/no bandwidthlimit)
Wow!
BR, Toni
Last edited:
post297.
I think H.Ott will tell you that all 3 are required.
RF attenuate at EVERY cable entry into the RF shielding box.
If the cables are two wire, then both cables need to have the RF attenuation of the common mode interference.
The two wire connection/s can still have separate differential mode attenuation of the non audio signals.
I think H.Ott will tell you that all 3 are required.
RF attenuate at EVERY cable entry into the RF shielding box.
If the cables are two wire, then both cables need to have the RF attenuation of the common mode interference.
The two wire connection/s can still have separate differential mode attenuation of the non audio signals.
Last edited:
There is a subtle difference between how 'pure Cherry', with its Holy HiZ VAS collector, behaves with suckout / bias spreader caps. ... and lesser compensation schemes like plain Miller, MIC, TPC, TMC ... everything else.
These lesser, evil schemes all take some feedback from the VAS collector and make it sorta LoZ. But that means the VAS MUST be coupled tightly to BOTH drivers else they will be driven differently.
Hence the need for loadsa big bias spreader, suckout caps etc. for good THD.
'Pure Cherry' however, has Holy HiZ at that point. It sorta doesn't matter if the current output of the VAS goes to the +ve or -ve driver and hence what's between the driver bases is of less importance. The caps. are chosen for stability and only have small effect on distortion.
In most cases, they can be much smaller or even removed completely. eg #4 of tpc-vs-tmc-vs-pure-cherry
I'm guilty of not pointing out this evil behaviour when I put out my challenge to TMC & other evil pundits to beat 'pure Cherry'. Mea maxima culpa
Toni, for TMC, you need a big C11
____________________
But each case needs to be looked at individually .. in SPICE and also in 'real life'. eg
In #271, I was looking at V(outa) to see some of the stuff that Toni saw in real life. You clearly see the wriggles on the top of the squares with different C16. But I couldn't get the wriggles on the upward slope ...
.. until I looked at V(outa) without C16 on a 8R load. NoC16.gif clearly shows the wriggles on the upward slope. C16-470n+10.gif shows none. Yes I know this isn't what Toni found.
Conclusion: more work to be done but neither 'no C16' or 'C16 470n' are optimum.
___________________
'Pure Cherry' always uses smaller (sometimes no) spreader, suckout caps compared to lesser, evil compensation schemes. That's one reason why I like it. (Another reason is less & smoother 20kHz THD)
Small caps mean less changes on overload and quicker recovery. That's why #4 of tpc-vs-tmc-vs-pure-cherry has excellent overload & recovery without extra clamping diodes.
No bias spreader cap will lead to higher bias spreader impedance which one would not want as it is neccessary for stable driver bias and so important for crossover distortion. Id say this is a bad compromise. Without a cap here could mean oscilation which is unwanted. Ive never found this to have any effect on overload or recovery
There is also no need to use high capacitance here, what is more important is either the use of high hfe parts or darlington or cfp. There are other methods for lowering the impedance which I ll not mention now.
Yes. Big bias spreader cap MUST be used for plain Miller, MIC, TPC, TMC etc. 'Pure Cherry' gets away with much smaller, or sometimes none, due to its Holy HiZNo bias spreader cap will lead to higher bias spreader impedance which one would not want as it is neccessary for stable driver bias and so important for crossover distortion. Id say this is a bad compromise. Without a cap here could mean oscilation which is unwanted. Ive never found this to have any effect on overload or recovery
There is also no need to use high capacitance here, what is more important is either the use of high hfe parts or darlington or cfp. There are other methods for lowering the impedance which I ll not mention now.
For 'pure Cherry', the cap is selected for stability, not distortion.Yes. #297 is entirely in line with Ott.
(1), (2) & (3) are ALL required. But ONE (and only ONE) of them should be a direct connection. In Power Amps and other physically big stuff with multiple in/outs (eg a mixer), the Star Point is what I've used in the past.
For eg a simple 2 channel mike preamp, the Output XLR pin 1. This is in line with Jensen, Whitlock, Self & other gurus recommendations on line level connections. For these, screen should have a direct connection at the Send end.
I'm thinking for a Power Amp, it should be the RCA input screens but I've not tried it before. (Or likely, I tried it in Jurassic times en it dun wuk! Dis is all more than 2 decades ago.)
But ALL other In/Outs that aren't directly connected to Chassis need the 100n Ceramic with SHORT leads to Chassis AT THE TERMINALS/SOCKETS.
The +ve speaker lead is a problem. Some amps object to hanging 100n there. Toni's 2nd Zobel AT THE TERMINALS is a good compromise.
But will all of you who have opinions on this subject please post a little pic of how you would do it. It's an important issue which is poorly understood.
'Real life' please but pontificatin' OK if backed up by 'real life' experience
Here, a pic is definitely worth 1k words.
Last edited:
Grounding/Earthing
There should be one, & ONLY one Direct SIGNAL connection to Ground, in the Whole system ! And that should @ the Signal Source, NOT the Amplifier/s.
Most people will have multiple sources, so the next available point will be @ the PreAmp, or Mixer.
All conducting cases MUST be connected Directly to Ground for safety. These should be wired individually to the Mains Earth terminal in the plug.
Ground lift devices, if fitted, should be engaged to ALL items, Except the Source.
A point never mentioned, as far as i've Ever seen is, the more GL's in place = more in Parallel. Which means that the time constants of the individual R/C's will equate to "maybe" non ideal performance/protection.
Ideally, each ground return on Each component, should be taken Individually to the Star Point. Or at least, bunched together on the PCB's as seperate High/Med/Low current returns, & then taken as such to the SP. ALL signal GR's should be taken individually to the SP, or @ least bunched together as a SP on the PCB, & then wired to the Main SP.
There should be one, & ONLY one Direct SIGNAL connection to Ground, in the Whole system ! And that should @ the Signal Source, NOT the Amplifier/s.
Most people will have multiple sources, so the next available point will be @ the PreAmp, or Mixer.
All conducting cases MUST be connected Directly to Ground for safety. These should be wired individually to the Mains Earth terminal in the plug.
Ground lift devices, if fitted, should be engaged to ALL items, Except the Source.
A point never mentioned, as far as i've Ever seen is, the more GL's in place = more in Parallel. Which means that the time constants of the individual R/C's will equate to "maybe" non ideal performance/protection.
Ideally, each ground return on Each component, should be taken Individually to the Star Point. Or at least, bunched together on the PCB's as seperate High/Med/Low current returns, & then taken as such to the SP. ALL signal GR's should be taken individually to the SP, or @ least bunched together as a SP on the PCB, & then wired to the Main SP.
Last edited:
This is what I show as the IEC to Chassis connection marked Class I Safety in #297
This is NOT NEGOTIABLE and leaving it out is illegal in many countries.
I take it you really mean 'Ground lift devices should be fitted to ALL items.'
This used to be (pre 1980?) common thinking, even in professional circles. There is no excuse for requiring this in a modern balanced line environment.
For evil domestic RCA stuff, I submit that the earthing design I propose in #297 will work (no hum loops) in most (all?) situations except the most idiotic situations. I will humbly admit there are lots of idiots around. Examples of these are warmly invited
The #297 pic is what I remember of an Engineering Memo I wrote about designing stuff to meet 'both' Class I & Class II safety standards Internationally and the implications re Hum Loops etc.
Actually this is of great concern in Medical applications. I think there is at least one EN (IEC standard) laying down the 'law' on this.
Zero, please post a pic of how you would earth Toni's or your own amp. You can botch my pic if that's more convenient.
Ahemm: Ground lift a device is strictly forbidden in europe.
Grounding generally: If one has more special GR, GL and SP knowledge please open a new thread about "correct grounding". This thread discusses only designing/optimizing a
"2stageEF high performance class AB power amplilfier"
Cabling, power supply, case etc. is not part of the discussion except there is a need to clarify specific problem situations.
BR, Toni
Grounding generally: If one has more special GR, GL and SP knowledge please open a new thread about "correct grounding". This thread discusses only designing/optimizing a
"2stageEF high performance class AB power amplilfier"
Cabling, power supply, case etc. is not part of the discussion except there is a need to clarify specific problem situations.
BR, Toni
Hi "manso"
I have had a think about this and wonder why you think the gain will be lower?
Does not seem consistent with my analysis or Toni's result below.
Hi Toni
Nice that this recommendation resulted in simple and successful improvement
Best wishes
David
Hi Toni,
I've been watching this thread since you opened it and right from the beginning I thought of building this amp - "geiles Teil", a true beauty with promising sim and real life figures. I'm gonna join the 2 boards to one and will give the idea of discrete darlingtons (2SC3503 + 2SC5359 and their complementaries) a try
2 questions since I'm rather unexperienced in designing analogue things:
- does it make sense to repeat R48/C41 for each of the discretes?
- would a 10n BE-cap on every driver improve sound/stability?
Coming back to the quote, did you keep track of all the mods you tried out and did you preserve the best results in an up-to-date-best-of-all-schematic?
Thanks in advance and keep going, I'll keep watching
BR,
Holgi
Hi David
BJT theory, any emitter degeneration leads to lower gain. Lower THD results may be seen but this is due to better matching of the current mirror and henceforth the LTP which will cancel not only 2nd harmonic but also 3rd. The better matched the LTP the lower the distortion.
Hi Toni,
I've been watching this thread since you opened it and right from the beginning I thought of building this amp - "geiles Teil", a true beauty with promising sim and real life figures. I'm gonna join the 2 boards to one and will give the idea of discrete darlingtons (2SC3503 + 2SC5359 and their complementaries) a try
2 questions since I'm rather unexperienced in designing analogue things:
- does it make sense to repeat R48/C41 for each of the discretes?
- would a 10n BE-cap on every driver improve sound/stability?
Coming back to the quote, did you keep track of all the mods you tried out and did you preserve the best results in an up-to-date-best-of-all-schematic?
Thanks in advance and keep going, I'll keep watching
BR,
Holgi
Dear Holgi,
thank you for joining this thread!
From time to time the latest simulation has been posted. Maybe in a few days we can present the final revision 2 schematics and pcb boards.
Both boards to join will be a huge pcb.
The decision to make 2 boards conencted by ribbon cable was primarily to be able to place the input stage as near as possible to the input jacks and to keep this pcb as far as possible away from all big currents and powersupply toroids.
The discrete darlington version as well as the mosfet version are only drafts and therefore totally unfinished and needs a lot of simulation work to get all things right. Maybe some day...
BR, Toni
As said, I'll keep watching...
Current idea is to use a 3U 400mm-depth case with side-panel heat sinks. The input stage will be very close to the back panel and have the RCA jack connected via a few cm coax cable. The input stage rails are decoupled using keantoken's rail filter (thanks for this handy little circuit
), we'll see how this performs. I attached a sneak peek on the current layout state.
I rather build-and-try than dive into sims, once I have finished the layout I'll flip a coin and decide whether to build it or to keep watching and learning for some more days/weeks
BR,
Holgi
Hi ?
This is true for a common emitter. Here the current mirror merely acts as a load for the LTP. I think the main impact is that the VAS EF buffer can have increased emitter resistor and so the LTP sees an increased load and that gain increases. The emitter resistor of the the VAS may need to increase too and that will decrease gain there. So it is possible to use this to redistribute the gain. Many interactions and I wasn't sure which one you meant.
Best wishes
David