Folded Cascode in the Making

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
As a respite from working on tubey things, I'm piecing together this folded cascode power amp. I'm in the advanced assembly stage, and trying to figure out the most expedient way of mating boards to heat spreader/heat sink. Pick away, but I'm building the thing this way for the time being, though I'll probably be adjusting some resistor values for proper output stage quiescent current. Compensation and zobel values will be determined when the thing is up and running.
There is no servo, and it shouldn't need one, as the input FETs are pretty much perfectly matched. I have a single sex differential amp running in my living room the same way, and the output offset is quite low and stable.
 

Attachments

  • folcasc.gif
    folcasc.gif
    32.4 KB · Views: 619
Member
Joined 2009
Paid Member
Not bad, but a couple of ideas anyway :)

1) Use real zeners instead of Q1/Q2
2) You don't really need R22/C5
3) Use LED's as references for current sources, match 2 if you really want to. Best types are red low Z ones, those can be feed with just a resistor instead of that fancy Q22 current source. That will save a lot of parts, incl. C3/C4
4) This can be one fast sucker, you ARE going to need a single compensation capacitor from Q16/17 collector to gnd
5) Offset should not be a problem with AC coupled feedback
6) But BIAS is uncontrolled, are you going to hand match all those parts ?
7) I'll recommend to lower the current in the input stage a little, will solve any heat problems and you don't need R7/R8
8) And also lower current in drivers, instead put a capacitor from Q20/Q25 to add charge out transistors

But with mosfets I really prefer a regular source follower instead of the "diamond" buffer type

Have fun.

Søren
 
Comments -

1) Transistor BE juctions make very tight low current zeners, and the 7.5-8V you get is just right for the 2SK170 and 2SJ74, whose gate leakage goes up into exponential cloud la-la land for drain voltage much higher than that.

2) As for R22, C5 - I'll be the judge of that when the amp is up. The places are there on the board. I've used those guys before to fine tweak square wave response.

3) I like one central current source being the boss, instead of having to mess with a bunch of rugged individualists. I considered using LEDs to bias the folded cascode stages (I did it before on an earlier version of this amp ~ 10 years ago), but the Rs and Cs offer greater voltage flexibility and take less room, as I'd probably want to use 3 LEDs in series for bias. The folded cascode stages respond well to having some voltage headroom.

4) I've used that sort of compensation before on an amp similar to this, and I'll cross that bridge when I come to it. Last time I tried it, a gimmick was enough to quiet things down. A teflon wire gimmick makes for a pretty high quality quality conpensation cap, and some heat shrin around the finished gimmick helps keep it twisted.

5) I agree, offset is no problem as long as the input fets are well matched. The cascode actually helps, as the increased gate leakage combined with the 1M DC feedback resistor is enough to start to unbalance the input stage a wee bit for higher collector voltages - it even shows up in a PSPICE simulation..

6) The output bias is controlled - the driver fets are on the same heatsinks as the mosfet. The controlled bias is one motivation for doing the diamond setup at the output. The ratio of resistor pairs R16/R33 and R28/R34 roughly set the bias current. It's not quite as simple as that, as I also used much smaller drivers than outputs (size 1 die vs size 4 die) so that the 20mA of bias I'm pushing through the drivers is more effective at jazzing the output stage. I used PSPICE to set up the initial bias resistor values, and I'll tweak the values when the heat sink interface is finalized. I had a smaller version of this amp (actually, not much smaller than this one) running for years with no hoohoo.

7) Lowering the current in the input reduces the loop gain. - I'll keep the resistors.
 
Member
Joined 2009
Paid Member
I can follow you, but in generally like things that are reproduceable and can be manufactured in volume. So in my opinion:

1) Zeners are specified and if needed you can change the voltage
2) In my experience, once I reach 1 Mhz power bandwith I don't care too much is the square wave is completely square :)
3) I have done quote some experiments (not simulations), and much admit that using just one led in a folded cascode is pretty perfect, assuming it's the correct LED. Just too many parts otherwise and a jfet to measure.
4) eeeh, a twisted teflon wire in production.... But I assume that it will just be a one off, right ?
5) Yeah, you can easily match jfets, I have even matched sot323 jfets to about 1mV offset....
6) I don't mean thermally, the mosfets will of course track, but those power mosfets are specified for 2-4V Vgs(th). Not good for manufacturing, but again, a one off....
7) I think you will have plenty of gain, you will probably need to slow it down anyway....
8) I was a little unclear there.... I meant to say add a capacitor between Q20/Q25 collectors, to help charge the Vgs in the output mosfets, in order to speed it up. Otherwise you will only have the 20 mA in the current circuit to charge the mosfets.

Btw, why C8 ?? To me it will just lower the positive PSRR, although C3 helps.

Søren
 
Soekris -

Thanks for the comments. I'm going with the amp as is to see what happens, and correct problems as I encounter them. I'll report on progress as it happens. I'm interested in leaving the drive for the output stage as it is for the present, though I understand what you're saying. It'll be interesting to see if the output stage is the limiting pole in the frequency response, with 20 mA of available drive. In part I'm doing this to relearn what I did with the earlier incarnation of this amp, as that thing is in pieces, raped for parts (the Panasonic drivers I used aren't made any more, so they went into other things), and I didn't keep a lab notebook. Next time I'm down in the basement, I'll check and see if one of those old boards has a gimmick pasted on the back...

As far as the voltage margin for the folded cascode is concerned, I wanted to keep some voltage across the mirror transistor, as I'm dropping about a volt on its emitter resistor. We'll see about that when the thing is actually pasted on the heat sink and powered up. I may just cheat and mount both amp modules on a heat spreader and bolt it to a heavy-duty pentium fan/heat sink combo. I have plenty of them around (a steal at $7 each), and they're good for lots of watts.

I have a similar design in the works (without the diamond output stage) that will be mostly MOS. That will definitely be a one-off design, as Supertex doesn't offer their small signal mosfets (the VN/VP01 size die, 200V) in the TO-220 package any more. I have enough hoarded so I can try matching some pairs for the inputs. With the TO-220 package, I have more freedom of choice for bias current, thermal coupling is easy, and I can ditch the cascode transistors I used to protect the JFETs in this amp.
 
Member
Joined 2009
Paid Member
Go ahead, I'll like to see how things turns out, as that might give me some input to the design I'm currently working on.... It will also be folded cascode with all fets in the signal chain, but otherwise pretty different from yours as I'm going for no NFB and using a uC to adjust and monitor things....

I can't recommend many parts to you as I do SMD nowadays, but I do like the Zetex 250V mosfets, the ZVP4525/ZVN4525, haven't found anything better for the drivers, low Vgs and Crss....

I also like the Toshiba output power mosfets, but they're hard to get and expensive although I do have a couple of 2SJ201/2SK1530 sets. I REALLY like the new 2SJ618/2SK3497, but they seems to be impossible to get, I have had an order in with Digikey for 6 months now without much happening....

Søren
 
The Zetex parts are very nice, but the E-line packages are limited to 1W (I'd pump a lot less than that through them), and I really don't want to do surface mount at home, as I have enough headaches with the stuff at work.

I'm going ahead with the MOS design using the Supertex parts. I've been hoarding them since about 1990 or so, and they do no good to anybody just sitting there in the bags. I bought them at the time because I was not satsified with the capacitance of even the smallest IR-type parts. I have a little itch just like you concerning manufacturability, since I have to think about it all the time at work, but it's DIY after all. It will be a unique amp.
 
Member
Joined 2009
Paid Member
The Zetex ZVP4525/ZVN4525 in sot223 are specified at 2W, with the nice "surfboards" adapters from Digikey they are as easy to use as a leaded part, can be put directly down into a TO-126 or TO-220 footprint. Maybe you can even find a small clip-on heatsink to fit....

Don't know the Supertex VP01/VN01, don't have many of my old paper databooks anymore, but the Zetex parts are the best small complementary mosfets I have been able to find....

But of course it's alway good to use what you have laying around, unless something better is available....

Søren
 
Zetex, when they were still Ferranti, supposedly got their mosfet process from Supertex back in the day. At least, that's what the Ferranti rep told me back around 1980-81 or so when I was looking at their parts. No doubt their processes have diverged in the almost 30 years since then, but they share a common ancestry. I doubt their small signal parts are all that different except for things like QC. As for dealing with surface mount (even SOT-223, which is actually kinda reasonable), I'll decline. If I actually wanted to manufacture something like this in quantity, it may be a different question.
 
I was looking at my Supertex stash last night, and I may not have enough TO-220 devices to get the matched pairs I need. If that turns out to be the case, I'd consider getting a bunch of the little Zetex E-line devices and glueing the matched pairs to a heat spreader with Arctic Silver. That will solve both dissipation and thermal matching. I figure that way I can get a full watt of dissipation without big problems. I wouldn't try pushing the dissipation much over that, even with heat sinking. This should also fit the PC boards I currently have with a little bit of lead gymnastics.
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.