Group Build: Dynafet

[swt61]

I simply can't hold my tongue any longer! I'm here to caution against Dyna Fett! I'm not yet calling for a Fett ban, but there are at least 7 galaxies that are very worrisome.

As yet there is no absolute proof that Dyna shares his cousin Boba's views, but again, caution is in order!

[Pars]

More worrisome are the actions and belief system of the stepmom, Dyna Mo Hum...

[Pars]

On 2/8/2017 at 8:35 PM, Pars said:

Promising... actually listened to single channel on some crap phones tonight. My scope isn't showing anything bad (though I think my scope may need recapping). More testing, etc. necessary. One thing I do have is some hum/buzz; this increases with volume until the pot gets past 90%, then decreases to nothing. No buzz all the way down either. Not sure if this indicates noise/oscillation or not, but what was on the scope wasn't conclusive.

One other change not mentioned above was increasing the feedback resistors from 1K/100R to 100K/10K (R56/55). I noticed that on the schematic for the SuSy DynaFET that feedback was shown as 200K/10K; I assume this is mostly to increase input impedance, but figured going higher might not hurt. I'll switch it back and see what effect that has.

 

 

 

Getting back on subject, I decided to try to evaluate the noise I was hearing. I hooked up one of my dynahi boards for the other channel and took a look at things on the scope. Both channels (FET/BJT) had some noise on them, so inconclusive. Since I don't have a function generator, I was using a portable CD player with a disc of hi quality test tones. When I unplugged the AC adapter, the noise diminished noticeably, but not completely gone. From what I recall, the dynahi was quiet when I was testing it. I'll need to recheck this with both channels to verify and make sure I'm not chasing my tail.

Since my previous (4 years ago or whenever) testing had shown that the FET amp board went nuts with the crappy headphones I have in the shop for testing, I have had these plugged in most of the time.

I replaced the feedback resistors with the original 1K/100R combo. Now things got interesting. Without the source plugged in (RCAs), I was getting a sine wave on the outputs (both boards!), at all volume positions other than all the way down (full attenuation). I did not determine what the frequency was, but strangely it wasn't really audible on the phones.

I am theorizing that the FET board is oscillating and this is feeding back thru the PSU? Not sure how rational that is, but that is all I have to go on right now as there was no input plugged in, it remained with or without the headphones plugged in, and the only thing common besides the pot is the PSU.

Next time I look at it I will verify that it is behaving consistently, and then go back to the 100K/10K feedback resistors. Maybe see what 10K/1K does as well.

Any other suggestions?

 

EDIT: I asked Todd to move this thread from the Casino to the DIY forum, and he obliged in doing so. If there are objections to this let him know (or me).

Edited February 10, 2017 by Pars

[Pars]

Took a look tonight at it, and the oscillation is repeatable. If I am remembering how to do these things correctly, it is at ~5.2MHz (3.85 divisions at 0.05 uS/div). It is also at around 15V p-p, but a really nice sine wave. Inverted in one channel (the DynaFET seems to be non-inverting from what I saw; the Dynahi board must be inverting).

I'm going to try some ferrite beads; I have some surface mount. Working on this, do I ever dislike the black soldermask now

[Pars]

The ferrite beads seem to have no effect. I decreased the feedback resistors to 10K/1K (still the same 10pf compensation cap). Oscillation is still there, though I now have to turn it up a bit further before it begins. Waveform looks slightly different as well, but I didn't recheck frequency.

The spice simulation doesn't behave exactly like the real world circuit (big surprise), but does seem quite picky regarding the compensation cap. The 10pf works well for the std. 1K/100R feedback network. 5.1pf seems to work well for the 10K/1K network, and 1pf up to about 4pf works for the 100K/10K network. I'll play around with the compensation cap in the actual circuit.

Now for the interesting part: since this only occurs with some crappy phones plugged in, I tossed a zobel together with a series 20R MF resistor and a Wima 0.047uf cap. This completely killed the oscillation. I'll get some more appropriate parts per Amb's b22 parts list and install.

I don't know how to do Bode plots nor take the measurements necessary for one in order to try to figure out exactly what is going on in the amp. At least that is my guess as to what it would take to really nail down the actual cause. Any advice here appreciated!

[Pars]

Ran some Bode plots in my spice simulation. Not completely sure how to interpret these yet. Haven't installed the zobels yet either (in the actual amp). I have them in the simulation; I'll have to run some of these with/without them and see what differences I can see. At any rate, some below:

Standard 1K/100R feedback loop, 10pf compensation cap

Same 1K/100R feedback, 33pf compensation

10K/1K feedback loop, 10pf comp.

10K/1K feedback loop, 5.1pf comp

[Pars]

One other question... looking at the board layout for these, I noticed that the 220uf/100V caps up by the FETs are run from V+ to V-, not V+ to GND and V- to GND. Is this correct?

[luvdunhill]

Yeah, that was done that way by the designer on purpose. Nothing to add yet - how many output devices are you using?

[Pars]

Hmmm, re: caps. Does that actually do anything? Thinking of changing these to ground?

I'm running 4 devices (2 pairs).

 

 

Sent from my iPhone using Tapatalk

 

[Pars]

More interesting. I pulled 1 pair of devices.

Still oscillating. Actually, now it was oscillating with the volume all the way down, would quit when turned up slightly, then back to oscillating as volume went up.

Pulled the replacement 2SK214 and put one of the originals back in.

Still oscillating, same as before.

Pulled the ferrite chips I had put in and replaced with a couple of these:

http://www.digikey.com/products/en?keywords=240-2437-1-nd

Not oscillating! Hmmm. Of course Digikey doesn't have those anymore and I don't have 8 of them (5). Shit. Looks like I ordered 10. Will have to see what I used them in... flea clock boards probably.

[luvdunhill]

1K gate stopper is probably too high. Can you cut that in half?

 

You might read through this paper:

 

http://www.cordellaudio.com/papers/MOSFET_Power_Amp.pdf

[Pars]

Sure, I'll give it a try. I noticed that Borbely was using 100R or 221R on these devices.

Any thought on the zobel fixing this? Band-aid? Cheating?

[luvdunhill]

The gate stopper will form a low pass filter with the mosfet input capacitance, which changes with the number of paralleled devices. Since this value differs between n and p channel values, the goal would be to pick a frequency and then calculate different gate stopper values for p and n ch devices.

 

Gate stopper should slow things down enough to be sufficient.

[kevin gilmore]

i'm looking at my board which is labeled 2009 Jacob potter and all the electrolytics go from power rails to ground

 

[Pars]

^ I'll have to check the actual board. The Eagle files I have, which are not quite the final board. show the two 220uf/100V as going from V+ to V-. I've been trying to update the Eagle file so it more or less matches. The combo SMD/TH resistors he created are a massive PITA in this though, requiring 2 connections on each end.

[Pars]

Checked the actual board (marked the same as yours Kevin, from the group buy, lovely black soldermask), and the 2 electros up by the MOSFETs run from V+ to V-. I'd have to refresh my memory on bypassing, as I recall this being done with opamps before, but I would think to gnd would be better?

Still running 1 pair of devices, removed the 1K gate stopper and put sockets in. With 221R everything looks fine, other than it is resembling lampizator construction technique.

Marc: thanks for the Cordell link. I read quite a bit of that, plus took a look at his book on amplifier design. It appears that he always uses gate zobels consisting of 39pf->100R->gnd on his amps. He seemed to indicate that he always uses them.

[Pars]

Replaced the 221R gate stoppers with the original 47R. I had previously switched to some ferrite beads I got from Arrow (Fair-Rite 2773009112) and still no sign of oscillation.

I put in the other set of FETs (4 total now) with beads and 47R gate stoppers. Still looked fine.

On listening to it, I decided I wanted to isolate the R channel (DynaFET R, Dynahi L) and discovered that both channels were still playing with only the R input connected, so I need to figure out what is messed up in the input wiring. Probably why when it was oscillating, it appeared on both outputs. Oops.

 EDIT: Biased at ~150mA per device pair. Bias starts at 320mV and drops as amp warms up. 2R ballast resistors.

 

 

Sent from my iPhone using Tapatalk

 

 

 

 

[kevin gilmore]

yep those 2 caps go from +vcc to -vcc

not sure why

[luvdunhill]

Cool, so are the ferrite beads needed, or have you tried to remove them?

[Pars]

Yes they're needed. But haven't tried removing them.

Knock on wood, I think I have a working board. I can get it to oscillate, but only with phones plugged in AND meter(s) connected to the ballast resistors. Pull the meters and it quits.

I am going to put the gate zobels ala Cordell in but need some parts.

1) tried the mid feedback point and really bad oscillation.

2) the mid feedback point doesn't control gain (rail to rail). My spice sim confirms this unless I am configuring it wrong.

3) I am going to run RMAA on it w/Dynahi for other channel. That will hopefully give me a better idea if there are bad things going on that I am not seeing on the scope.

4) I had the input wiring reversed. Works much better now [emoji4]. Boards aren't marked for shit. The @!?# black solder mask makes it very difficult to see the traces as well.

5) note to anyone messing with these: the power connector is reversed from Dynahi.

 

 

Sent from my iPhone using Tapatalk

 

[Pars]

Just a bit more regarding the first paragraph:
I haven't removed the ferrites as they were one of two things which killed the oscillation that I have tried, the other being output zobels as Amb describes in his B22 write up.


Sent from my iPhone using Tapatalk

[Pars]

One other thing, some of the transistors in the diamond buffer section are running quite warm: Q26, Q23, Q18 and Q17 (purple ovals). Q26 is running ~60*C, and this is open air. I have some of the EDM sinks, but they push the transistor leads closer together than I like them. I also have some To92 heat sinks that I might try.

DynaFET_schematic_marked.pdf

Edited February 24, 2017 by Pars

[justin]

i'm late here but your experience with the oscillation is very similar to what i tried >10 years ago with a mosfet output version of a balanced dynalo front end. It only oscillated with K1000 plugged in

[Pars]

Brave man plugging good phones in [emoji4] The ones I am using are real crap Sony something or other. $5 and you overpaid...

I would bet that the output zobel would take care of that for sure, the gate zobel (Cordell) probably, and the ferrites probably. Some people do quite involved zobels with inductors etc. from gate to drain.

 

 

Sent from my iPhone using Tapatalk

 

[Pars]

Started working on the other board last night, re-configuring it to match the first. I plan on bringing this up without the ferrite beads, but add the gate zobels to it, which I haven't tried yet. These seem more like a snubber than a zobel (cap -> resistor -> gnd). Cordell uses 39pf -> 100R; I don't have any 39pf but do have 33pf, so will probably increase the resistor a bit (120 or 150). The one Cordell uses seems to sim out at -3dB of > 10MHz. I may increase the cap to 100pf, but I'll try it with the 33pf first. I'll report back on how this behaves without the ferrites.

I also plan on running RMAA on the first board, with a Dynahi board in the other channel. If things look amiss, or even if not, I will re-run this with the Amb zobel on the output and see what that looks like. If that looks better/different, I may put the Cordell zobels on this board and see what that looks like.