Group Build: Dynafet

[Pars]

Ran RMAA on the 1st board (R ch). Everything looked fine. Re-ran it with the output zobel on, looked about the same, so my conclusion is it is working. I think for safety I will put on zobels ala Cordell. Need to find a near enough ground on the bottom.

Set up the 2nd board with one pair of output devices. It biased right up, offset adjusted and looks good. I could not get this one to oscillate. No zobels or anything other than the stock configuration that I've settled on. I'll document what that is later. I'll add the 2nd pair of output devices and see what happens then.

RightMark Audio Analyzer test : Dynahi-L DynaFET-R.pdf

[luvdunhill]

What load are you testing at? I would assume both High and Low impedance would be preferable.

[Pars]

Not running any load right now, just straight into an maudio fw box. I do have a load box from tangent's plans (https://tangentsoft.net/audio/hp-dummy.html ) that I could toss on there. It has 33, 330 and I can't recall what else on it.

EDIT: it has input and output jacks on it. Its been so long since I built it and used it that I forgot what all is on it

Edited February 26, 2017 by Pars

[Pars]

Of course the 2nd board was too good to be true. Put the 2nd pair of outputs in, oscillating to beat hell now without volume up, nothing plugged in. 70+MHz (0.7 div @ 0.2us/div). Board was acting strange even before I powered it (wasn't even plugged into AC). When I put the scope probe on the R output, it showed some fuzz on the trace. Board was plugged into all I/O. Strange. When powered, the oscillation was independent of volume setting, and with no phones plugged in. Varied at times by just approaching the bench, moving hands, etc. Fun.

I'm going to try Cordell's zobels on this one first. Probably use some 100pf Wima FKPs I have with 90.9R or 120 or 150 ohms to gnd. Or I have some 33pf Mica caps.

Also had noticed that I hadn't gotten the devices all the way onto the angle brackets (drilled like my dynahi ones), and recalled getting pissed at not being able to accurately hit a good center punch (Starrett) dimple with my drill press, so bought some cheap center bits from Harbor Freight. Redrilled the angle bracket for the good channel. Much better. I'll redo the other one presently.

Also discovered my load box is broken; Lorlin rotary switch wasn't working correctly.

[Pars]

Put in the zobels (100pf-90.9R) in which seemed to cure this problem. I haven't done much testing yet, but didn't find any oscillation. Hard to find any ground points close by on the bottom (gnd plane is on top). Closest is about 2". Put the cap right on the gate pin. Resistor and teflon tubing later (x4) and done.

That is the only change (other than Vbe multiplier resistor values) that I've done to this board from std. config.

[luvdunhill]

Time to add a load

[Pars]

Yeah, I'll rig something up. Any suggestions? The 33 and 330 ohm that I have good?

[luvdunhill]

Yup!

[Pars]

Put a toggle switch in the load box (33/330) and ran RMAA on the #2 board (Cordell gate zobels only).

Still looks good. I had a scope on the outputs and saw a touch of oscillation when I first fired it up. None after.

So the 3 "cures" for the oscillation that I have found:

1. Ferrite beads
2. Cordell gate zobels
3. Amb output zobels

Board #1 has the ferrite beads. #2 has the Cordell zobels. I think I will put the beads on #2 also and the zobels on #1 and be done.

Dynafet2_33_3.pdf

Dynafet2_330_4.pdf

[Pars]

I am planning on using 1 and 2 above on both boards (ferrites and gate zobels).
Bias is currently at ~150mA. I will decrease bias to 75mA and maybe 100mA and take a look at things including RMAA results before deciding where to leave it.
The outputs seem to be fine at 150mA temp wise. The buffer transistors previously mentioned run hot but that appears to be unrelated to the bias level. The sim is only showing around 8mA or so running thru any of these, so not quite sure why they are getting toasty. Because of the layout tightness, it is hard to get in to take measurements.


Sent from my iPhone using Tapatalk

[Pars]

Got both boards up and running with mods 1 and 2 above. Ran RMAA on it at 150mA, 100mA and 75mA bias. I think 150mA is probably the best.

See what any of you think of the plots attached.

Also brought my T50RPs down and listened to some tracks on it. I think this may be my favorite amp! I'll have to do more listening though.

Dynafet_both_75_330L.pdf

Dynafet_both_150_330L.pdf

Dynafet_both_150_33L.pdf

[Kerry]

Congrats.  That took alot of effort in your part.  Glad it paid off for you and thanks for sharing

[Pars]

Documenting how these are configured:

Vbe multiplier:

R17-R18: 3.3K

R62: 1K

VR3: 20K

R19-R20: 2K

Mods:

Cordell zobels: 100pf / 90.9R from gate to ground (Cordell uses 39pf / 100R)

Fair-rite ferrites (2773009112)) in series with gate resistor (cut gate traces)

Gate stoppers: 47R

Current feedback loop: 10K / 1K / 10pf (R55 / R56 / C1). I will try going back to the stock 1K / 100R on this.

I would not even consider using the mid feedback point; last time I tried it (after board was working), it oscillated badly with no phones/output plugged in. I am going to solder the jumpers in place on this.

 

Edited March 5, 2017 by Pars

[Pars]

Failure finally; switched to the std. 1K/100R feedback loop, and getting oscillation with the crap phones plugged in only when the volume is all the way down (or nearly so). Turn it up and it goes away. This looked nasty though (nice sine wave), and appears to be the right channel only causing it (unplug right output and it goes away). About 4.5MHz, 17Vpp. Also, plugged in my load box. 33R looked ok, but different oscillation with the 330R.

I guess I'll just stick with the 10K/1K feedback, and figure out what to decrease the compensation to.

Kevin: any issues with increasing the actual feedback resistor values 10x over your design? In poking around, I didn't get a sense of how you determine the actual values, just the ratio. The balanced dynafet design went way above this (200K/10K), though I suspect that was because of the THAT340 input.

[kevin gilmore]

Don't remember if there is a 100 ohm input series resistor on that board.

the ss design need lowish input resistors because of the offset currents in the that 340 and the gain resistor is calculated from there

[Pars]

Thanks Kevin!
The original DynaFET used a 100R input resistor (plus 100k to gnd), with a 1K/100R feedback resistors. I've changed these to 10K/1K (left the 100R input alone).
The SS balanced DynaFET showed 200K/10K for the feedback resistors.


Sent from my iPhone using Tapatalk

[Pars]

Went back to the 10K/1K feedback resistors, and all is well again.

I've been reading online (Cordell, IRC, others) regarding using parallel mosfets with no source ballast resistors. Nelson Pass has been talking about it as well (Burning Amp talk). I know the Counterpoint amps used to do this (and were criticized for it). This apparently is without really tightly matching the FETs, but the gate voltage/current must be matched well and the Vgs should be as well. I measured Vgs on this, and while it is matched well between pairs of devices (N-ch or P-ch), it is not between the two types.

I don't see any example schematics around either. I wonder what would happen if I decreased the ballast resistors to something like 0.1 ohms...

[luvdunhill]

It is nearly impossible to match n to p with these. Even the ones I sent were from multiple tubes of each.

[Pars]

I figured that. Same with bipolars, normally.

If someone had a matched set of these I might be interested. Should be a bunch of them out there.

I know I got the set I am using from you, and had been going back thru the GB stuff and my email, but didn't find where I had talked to you about them.

[luvdunhill]

Decreasing the ballast resistors could happen as long as Vgs is really close. You could maybe experiment with unequal ballast resistors on the n and p side. I wouldn't mess with it though.

[Pars]

Agreed, not worth messing with.

 

EDIT: And I found out last night what you were talking about regarding the two DC offset pots Marc. I didn't realize they would effect the bias as well, and had one board biased >300 mA for a short time. Pulled the board and found I had one side at around 45 ohms (pot || 620 ohm R). I may replace one side with a 500 ohm resistor.

 

This is the most stable wrt DC offset of any KG design I've built.

 

 

Sent from my iPhone using Tapatalk

 

[Pars]

Did some work on mocking up the chassis and getting into a more listenable state. I will need to get aluminum top/bottom and rear panels for this (not using the steel ones that it came with). I'm also still a bit concerned with how warm the buffer transistors run, and this is open air.

Using a modified amb e12 protector board (with muting). I tested this to see if HF oscillation would trip it, and it did (12V ~70MHz with the feedback set to the mid FB point).

Listening-wise, it is likely my favorite amp, even being single-ended. This thing has balls, and seems to open up the recording more than any other amp I have built or used. Since my SE DynaHi boards are a drop-in, I may take another listen to those as well, but from memory, the FET version stomps it.

I am going to build up a GRLV for this, and will likely get a 50 or 100VA Antek trafo for it instead of the 120 or 160VA SumR I have in here (a bit in the HC overdo it fashion ).

Edited September 2, 2017 by Pars

[Pars]

I've been enjoying listening to this. I decided I wanted to lower the gain, since on more modern hot recordings, the pot was in the 1st quarter of rotation.

I changed the 10K resistors to 4.7K, and changed the compensation cap from 10pf to 4pf. Looked fine on the bench including with my load box.

I brought it upstairs along with the pair of dynahi boards which just pop right in, and listened to those first. They sounded good. Intending to do a comparison, I put the dynafet boards in. When I plugged the headphones in (10' mogami extension and 10' mogami headphone cables), the protection circuit tripped, and would just keep tripping. Hmm. I determined that the headphone cable and extension alone were enough to do it, without the headphones (T50RP) connected.

I took it downstairs along with the cables, and sure enough, oscillation (guess of 70MHz or so based on scope timebase). Amb's zobel networks would keep it from occurring, but lacking suitable resistors, I decided to change how I was reducing the gain. I put the 10K resistors back in, and replaced the 1K with 2K. Now it was fine, so back upstairs it went. I haven't done a proper comparison yet, but still think the FET version sounds better to me.

I am going to install the Amb zobels on the board outputs just for safety (47nf cap->22 ohm 2W resistor from output to gnd).

Takeaway from this is

1. use a protection circuit (I'm using an Amb e12 modified)
2. use Amb's zobels on the output just in case...

Edited April 22, 2017 by Pars

[Pars]

Just built and fired up a +/-30V GRLV for this. In conjunction, I never had checked the power draw of the amp boards. My spice simulation showed 366mA. Actual was 415mA or so at initial startup, dropping to around 375mA when kind of warmed up. This is at bias of 150mA and 2 pairs of outputs. Tested using the sigma22 I already had in here.

I had thought that I might see a higher draw than the model due to the buffers/Vbe mult. running warmish, but the actual was fairly close to the model.

Edited May 2, 2017 by Pars

[Pars]

Continued working on this.

1. I added the zobels aka Amb on the board outputs. Because I have a protection board installed, I put these on the amp boards rather than at the output jack.

2. I cut traces and reworked the two 220uf caps from V+ to V- to instead run from either V+ or V- to GND.

3. Redrilled and tapped to place both amp boards on the right heatsink, with the left reserved for the GRLV when implemented.

 

While doing (2) above, I ran the amp without these caps installed for a few days. Without the caps installed, the amp lost some of its bass slam (one of the things I liked about it). Reinstalling the caps brought this back. I couldn't really pinpoint a sound difference between having caps rail to rail vs. rail to Gnd.

Another artifact of this seems to be better highs response. Not sure if that had to do with the cap change or the zobels (output->22R->0.047uf cap->Gnd).

I'm quite satisfied with the sound of this after these changes. Seemed to bring a slight bit more clarity and finesse to the sound. Soundstage depth and height seem to follow what is on the recording accurately.

I may at some point try running this with a single set of output devices.

 

 

Sent from my iPhone using Tapatalk