The two shunt compensation RC networks impedances are too low to ensure
adequate gain/phase response so stability is almost entirely on the
shoulder of the lead compensation cap.
Also , a single pair of outputs devices is quite short given
the high PS voltage , two pairs would be mandatory.
Not sure that this design was well thought in these respects despite
its designer s recognized authority in audio....
I use a late-model but not latest-model Circuitmaker. I do type models in from time to time, with some of them easier to enter than others.
The program is buggy as can be, but considering where it started it did grow up a bit. And it's the devil I know, as the saying goes. Someone mentioned that someone has started supporting it again, but said the emphasis was on its use as a tutorial tool.
I will get LT Spice installed one of these days. It's becoming a nearly de facto standard.
Brad
Hi,
I used to be big on P-Spice in the 90's and had a full version. Since I moved to Win 7 X64 it no longer works, so I'm back to cocktail napkins and Tina-Ti for easy simple stuff.
In some ways the return to the cocktail napkin was liberating, in other ways I miss my simulator (mama!)...
I am working on this one. I am sadly again a complete novice and while LT-Spice has good community support, I seem to lack time and any incentive to burn the midnight oil learning it.
Ciao T
I used to be big on P-Spice in the 90's and had a full version. Since I moved to Win 7 X64 it no longer works, so I'm back to cocktail napkins and Tina-Ti for easy simple stuff.
In some ways the return to the cocktail napkin was liberating, in other ways I miss my simulator (mama!)...
I am working on this one. I am sadly again a complete novice and while LT-Spice has good community support, I seem to lack time and any incentive to burn the midnight oil learning it.
Ciao T
I'd concur with this advice. Although there is probably another way to lower steady-state distortion numbers, the compensation now works pretty well. And there are parasitic inductances that haven't been included in the models that may make some of the tweakier compensation networks important.
I have to remind myself about the title of this thread. But, it might be fun to see/hear if one could detect the difference if the amp was compensated in a more conventional way, for example turning the second stage into a feedback integrator to realize the dominant pole.
I recall reading that EB came to prefer moving the dominant pole more-or-less to the input stage, at least in connection with his small-signal designs. Rich May also preferred this, and I believe it was in the interest of better slewing performance. So games could be played in simulation, but I'd not risk it with the actual silicon right off the bat, particularly with parts that are hard to get. But I'm getting more cautious as I get older, and your mileage may vary.
I guess, in addition to the suggestions made for better and more robust components, one could ask what single topological change would result in a significant performance "improvement"?
Brad
Hi,
Possibly not, check where the poles and zero's fall.
As it is an existing, physical amplifier this may less than feasible. Hence I recommended instead using the double-die TO-3 Parts from Exicon etc., which does mostly the same, but without having to change the existing mechanics.
I suspect that the use of a single pair was down to the same reasons I found in many early Fet Amp's. Cost.
I acquired a healthy dislike for these Amplifiers mainly from PA Amplifiers that simply implemented the datasheet circuit without any adjustment and without enough pairs for the load/rail voltage combinations.
The lack of drivers caused what I and others have called "Mosfet Mist" (it is funnier in german with the double entendre) as the amplifiers where slewing and had insufficient bandwidth and "soggy bass" which was down to the lack of current capability.
We incidentally had bought a few such Fet Amp's and ended up using them only on the upper mids (they where crap everywhere else) which used 2" Compression drivers on copy JBL Horns.
The lower mids (turbosound copy horns) and bass (dual 18" W-Bins) where amped by my "several Killerwatt" Borg cubes (I designed and build them before I knew of the Borg). These used 380V three phase switching darlington modules (these where intended to switch together with rectifiers 380V three phase at several 100 Ampere!) in a circlotron on 100V rails (that's around 550W into 8 Ohm and these could easily drive eight 18"drivers in parallel if the mains power was up to it.
The treble used some old Peavey CS series Amp that sounded better there than those newly fangled mosfet jokes.
Given the DH-120 uses drivers and a good GBWP they should have done well for us in the mids and treble, with a single pair of fets, forget bass...
In domestic application i suspect the problems a bit less...
Ciao T
Possibly not, check where the poles and zero's fall.
As it is an existing, physical amplifier this may less than feasible. Hence I recommended instead using the double-die TO-3 Parts from Exicon etc., which does mostly the same, but without having to change the existing mechanics.
I suspect that the use of a single pair was down to the same reasons I found in many early Fet Amp's. Cost.
I acquired a healthy dislike for these Amplifiers mainly from PA Amplifiers that simply implemented the datasheet circuit without any adjustment and without enough pairs for the load/rail voltage combinations.
The lack of drivers caused what I and others have called "Mosfet Mist" (it is funnier in german with the double entendre) as the amplifiers where slewing and had insufficient bandwidth and "soggy bass" which was down to the lack of current capability.
We incidentally had bought a few such Fet Amp's and ended up using them only on the upper mids (they where crap everywhere else) which used 2" Compression drivers on copy JBL Horns.
The lower mids (turbosound copy horns) and bass (dual 18" W-Bins) where amped by my "several Killerwatt" Borg cubes (I designed and build them before I knew of the Borg). These used 380V three phase switching darlington modules (these where intended to switch together with rectifiers 380V three phase at several 100 Ampere!) in a circlotron on 100V rails (that's around 550W into 8 Ohm and these could easily drive eight 18"drivers in parallel if the mains power was up to it.
The treble used some old Peavey CS series Amp that sounded better there than those newly fangled mosfet jokes.
Given the DH-120 uses drivers and a good GBWP they should have done well for us in the mids and treble, with a single pair of fets, forget bass...
In domestic application i suspect the problems a bit less...
Ciao T
Hi,
1) Double Die output devices (by far) at double the Iq.
2) Mosfet or high GM J-Fet with lowish capacitance for VAS Transistor (see Nelson Pass's Threshold FET series - 3N172 was used)
3) J-Fet followers before the input differential bases or J-Fet IPS (2SK246 or similar low Gm/C devices would do well) ibid..
Ciao T
1) Double Die output devices (by far) at double the Iq.
2) Mosfet or high GM J-Fet with lowish capacitance for VAS Transistor (see Nelson Pass's Threshold FET series - 3N172 was used)
3) J-Fet followers before the input differential bases or J-Fet IPS (2SK246 or similar low Gm/C devices would do well) ibid..
Ciao T
Thanks Jan, not I did not know of that distro. That was going to be my research this weekend. I saw about 50 listed on the Berkeley page. Saves me some time. I never knew everything that has come out of UCB besides student rights until I visited earlier this year. Now it seems every other science article I read has some link to UCB or LBL.
D4 and D6 are there to allow quicker recovery if the transistors are saturated? So that is why one might want faster lower C diodes? Only matters if over-driven, right?
Still totally lost on the zero portion of R12/C5.
Or how C6 is calculated. I came up with 72K but that is a total guess that it is the 220p and 10K of R5.
D4 and D6 are there to allow quicker recovery if the transistors are saturated? So that is why one might want faster lower C diodes? Only matters if over-driven, right?
Still totally lost on the zero portion of R12/C5.
Or how C6 is calculated. I came up with 72K but that is a total guess that it is the 220p and 10K of R5.
D4 and D6 prevent the saturation of Q7. The drive to Q6 is limited by the clamp on the current mirror output, as Thorsten noted, so it doesn't saturate either.
The distortion associated with the capacitance change with voltage on D6 is mighty small when considered as being in parallel with C3, so although low capacitance is generally good, I wouldn't worry about it at this point. One small caveat is that the specified 1N4148 is seeing some fairly high reverse voltage at very high levels, and the rating is 100V, but I've rarely seen any with quite that breakdown. Being gold-doped for speed iirc (and they are plenty fast for this application), the leakage current is not super-low, but again compared to 3mA it's not that much, especially at reasonable temperatures. There are some higher-voltage lower-leakage parts around, some old Fairchild parts (Fsomething-333?) but I'm not sure which are preferred and still in production.
Edit: found the part: FDH333. 150V rating, 1nA leakage at 125V.
Last edited:
It is inherent to the shunt compensation on VAS collector.
Increasing the CCS current is limited as this will increase
dramaticaly the thermal dissipation of both the CCS and the common
base transistor of the cascode wich are low PWR trannies.
As already said stability is ensured by the output LR network.
Without the lead compensation cap the amp oscillate even
without capacitive load.
Increasing the CCS current is limited as this will increase
dramaticaly the thermal dissipation of both the CCS and the common
base transistor of the cascode wich are low PWR trannies.
As already said stability is ensured by the output LR network.
Without the lead compensation cap the amp oscillate even
without capacitive load.
If I take this correctly, I would be better off with something like 100p and 73K for C3/ R11 so I am not dumping a couple miliamps down R11 and I have enough current to actually charge and discharge C3?
Looking for "The complete idiots guide to Spice" Are all the distros about the same GUI wise so I can use the users guide from UCB?
Looking for "The complete idiots guide to Spice" Are all the distros about the same GUI wise so I can use the users guide from UCB?
I dont use LTspice although i did install it.
It s extremely user unfriendly for my liking so i took Simetrix in its free version
wich has limited circuit size but still enough fo quite complexe schematics.
As for the amp , the sims i made so far just convinced me that a simpler
compensation and schematic would have yielded better results with
a better balance between stability and distorsion.
It s extremely user unfriendly for my liking so i took Simetrix in its free version
wich has limited circuit size but still enough fo quite complexe schematics.
As for the amp , the sims i made so far just convinced me that a simpler
compensation and schematic would have yielded better results with
a better balance between stability and distorsion.
Attachments
Last edited:
On his later designs, he eliminated R11 completely.
The input LP depends entirely on P201, 660K to as little as 20K . I used this as an"integrated" amp and replaced P201 with an Alps pot, a couple sets of inputs etc. I now understand two reasons to put in a nice buffer in between. For my learning experience, we can assume a low Z input.
On the multiple outputs, it would not be difficult to drill the heat sink to use two outputs. Some rethink of the board design could help the lengths of the leads if switching to TO-3P type packages. I would have bought them this time except they did not have the in-die zieners. D10-13 are a long way away. I hate taking current through the mounting hardware on TO-3 packages. Learned that back in failure analysis. I'll leave that until I have more than a bit more experience. Baby steps.
The input LP depends entirely on P201, 660K to as little as 20K . I used this as an"integrated" amp and replaced P201 with an Alps pot, a couple sets of inputs etc. I now understand two reasons to put in a nice buffer in between. For my learning experience, we can assume a low Z input.
On the multiple outputs, it would not be difficult to drill the heat sink to use two outputs. Some rethink of the board design could help the lengths of the leads if switching to TO-3P type packages. I would have bought them this time except they did not have the in-die zieners. D10-13 are a long way away. I hate taking current through the mounting hardware on TO-3 packages. Learned that back in failure analysis. I'll leave that until I have more than a bit more experience. Baby steps.
Your version is easier for me to get my head around, that's for sure. As yo have done this sim, I will try to copy that to learn the tool and the main reason, to learn how these puppies work. As I have mentioned, reading textbooks is a great step one. Nothing beats being able to ask a question. I really appreciate your efforts and the patience of the many on this forum.
What would the sim say if you slowed down the outputs and changed R18 (23 on the mod) to 680 and R21 (R12 on themod) to 470? I am switching to the Exicon EC-10N16, P16 pair. They have half the gate cap. from the Hitachi. So the above gives about 1M for f3. Would I not risk some output stability with the 100 & 68? I am going to fly the gate resistors right off the socket.
150 on the CM emitters. OK, I was guessing 100. Actually, I happen to have some. How about a .1u across Q8?, or Q12 in the mod.
THose 2SC1775's sure do have a LOT more gain.
I gather the output network is a "typical" ?
You have moved the compensation to the output of the first and second stages into the feedback loop. No more big DC load on the VAS Do I look at them, C8/R13 and C5/R13 as if they were separate?
Simetrix. Found it. Simple is good as I don't intend to do this for a living. I am already on my third career. This is far more fun as a hobby.
What would the sim say if you slowed down the outputs and changed R18 (23 on the mod) to 680 and R21 (R12 on themod) to 470? I am switching to the Exicon EC-10N16, P16 pair. They have half the gate cap. from the Hitachi. So the above gives about 1M for f3. Would I not risk some output stability with the 100 & 68? I am going to fly the gate resistors right off the socket.
150 on the CM emitters. OK, I was guessing 100. Actually, I happen to have some. How about a .1u across Q8?, or Q12 in the mod.
THose 2SC1775's sure do have a LOT more gain.
I gather the output network is a "typical" ?
You have moved the compensation to the output of the first and second stages into the feedback loop. No more big DC load on the VAS Do I look at them, C8/R13 and C5/R13 as if they were separate?
Simetrix. Found it. Simple is good as I don't intend to do this for a living. I am already on my third career. This is far more fun as a hobby.
Thorsten,
Now I know the cause of the "Mosfet mist". Clearely audible. Neither the DH-220 or my B&K 140 did very well on my subs but I liked the B&K mids. Believe or not, the 50W RB 951 sounded better. This is on sealed 12 inch Peerless boxes. Music levels, not special effects so I am sure that 50W would die real quick with the first fighter jet fly-over or dino footfall. ( It beat them for the top too). I put the HCA 1200 there for now and I was very surprised that I heard a difference. I just did a real quick level match, but there is more to it than that. I only use them below 100, but it seemed , here we go with unsubstantiated subjective terms, "tighter". It already convinced me to look for something bigger so I can put the 1200 back on the top. It still has to pass my wife's ultra ear. 2200 or something.
Now I know the cause of the "Mosfet mist". Clearely audible. Neither the DH-220 or my B&K 140 did very well on my subs but I liked the B&K mids. Believe or not, the 50W RB 951 sounded better. This is on sealed 12 inch Peerless boxes. Music levels, not special effects so I am sure that 50W would die real quick with the first fighter jet fly-over or dino footfall. ( It beat them for the top too). I put the HCA 1200 there for now and I was very surprised that I heard a difference. I just did a real quick level match, but there is more to it than that. I only use them below 100, but it seemed , here we go with unsubstantiated subjective terms, "tighter". It already convinced me to look for something bigger so I can put the 1200 back on the top. It still has to pass my wife's ultra ear. 2200 or something.
- Status
- Not open for further replies.