Group Build: Dynafet

[DigiPete]

I see oscillation in your input waveform in the second picture.

Your signal generator may be injecting the noisey waveform, which is amplified and output.

maybe I am not interpreting the image correctly?

[Pars]

I'm using a portable CDP with a test CD. As you turn the volume up, you can see it on both the input and the output. I believe it is being fed back via the feedback. The input waveform by itself is clean at any volume setting. It did the same thing feeding it with a computer generator via my M-Audio FW. Yes, a real signal generator would be nice, but I don't have one

[Pars]

After doing some more reading (and a PM from digger945), I think I will try increasing the gate stoppers from 47 ohms to 100-120 ohms. From the device datasheets, the J device has a typ. input capacitance of 90pf, whereas the K device is 120pf. Since I am only seeing this on the tops of the waveforms, it would appear that the K devices are marginally unstable. I tried putting a 33pf from the gate to drain on the K devices, but mis-connected them to the gate-source pins. They didn't seem too happy with that, and remained unhappy once I removed the caps

Not sure if I toasted them or not...

[DigiPete]

ouch! Keep forging on! Let me know if I can help.

Does Kevin have any advice?

Pete

[Pars]

Replaced the gate stoppers with 120 ohm. Seems to have helped quite a bit, although I may have gone too far for the J devices, as at the lower gain, on a couple of load / pot rotation combinations, I saw oscillation "sidebands" on the lower portion of the waveform. At default gain of 11, and bias set at 125mA, everything is clean at all pot rotations except for the pot all the way down (or almost all the way) with no load.

The J and K devices may wind up needing different values. I found some good articles, but the one with a good formula required source inductance values, etc. that are not in the datasheet. And also stated that many of the datasheet values were at questionable temperatures and test circuits. I don't have the equipment nor expertise to obtain these values, so winging it I go. Some posts also advocate putting the gate stopper right on the FET pin. The ones on this board are about 1cm away. Ferrite beads were also deemed useful on source or drain pins.

I may muddle through this yet

EDIT: and another data point on current draw. Biased at 125mA with 4 devices, draw is 267mA.

Edited February 8, 2011 by Pars

[luvdunhill]

Still cannot use this new website, none of these posts showed up new for me.

Can you drive more current through the output stage? I'd guess this to be the problem. I'll be caught up with you in a few days and we can compare notes. Stuffing basic stuff tonite hopefully.

[Pars]

How much more current? I've driven 300mA and no joy. The devices are only rated for 500mA. My observations are that the more current the worse things get with oscillation.

Per the T2 thread, I tried upping the gate stoppers to 626 ohm. About the same as the 120 ohm, maybe a bit worse on the negative swing of the wave. I switched back to 47 ohms on the P channel devices and 91 ohms on the N channel. Things are fine EXCEPT for the pot all the way down (i.e., shorted input), then it goes nuts. This only on no load. At 33 or 330 ohm load, it is fine at any pot rotation.

So then I thought I would try listening to it. With the crap set of phones I use for testing, it oscillated with the pot all the way down (and not just a small one, but unintelligible waveform). Unusable. I'll have to check to see what I have gain at right now, think at 5.5 (1K/221R).

I may move the gate stoppers to right on the FET pins, but barring that, about ready to throw in the towel.

EDIT: Gain was set at 11 (1K/100R). Bad as normally it behaves better at the higher gain. Haven't tried moving the gate stoppers yet. I'd be happy to ship the board if someone wants to look at it themselves.

Edited February 9, 2011 by Pars

[luvdunhill]

installed a socket for R55

what is R55 again?

I'm getting closer to getting a new board stuffed. If you don't mind, what equivalent resistance do you have across R22 || TR2 and R21 || TR1? I'd like to put a fixed resistor there and was thinking 560 ohms (assuming that the design point was at the center of the pot).

[Pars]

what is R55 again?

R55 is the FB resistor, normally 1K; R56 is the 100R to ground.. After rereading the build threads, I decided that changing out R56 was the better way to change the gain, so put R55 back in the board and have R56 socketed and at 100R currently. Have also used 221R here.

I'm getting closer to getting a new board stuffed. If you don't mind, what equivalent resistance do you have across R22 || TR2 and R21 || TR1? I'd like to put a fixed resistor there and was thinking 560 ohms (assuming that the design point was at the center of the pot).

R21 and R22 are 626 ohm. Pots are 10K. Equivalent resistance is currently set at 500R, which is the design goal (I believe).

BTW, do you have the Eagle files and any documentation for BOTH board types? I have Eagle files for the through hole, but must have been for the prototype boards as they don't exactly match the GB boards. I emailed jacob, but no response. Kevin doesn't have them either.

Naaman was asking for a BOM/documentation as there are unlabeled pads on the bottom. It would be good to have the board files for both as well, for the sake of all participants.

[luvdunhill]

500 ohms assumes a 1.6V LED, which results in 2mA for each half of the front end (1mA per FET). I plan on choosing LEDs an adjusting this value to result in the final 2mA. Seems easier to me.

I don't have any EAGLE files. I'd recommend PMing Dan C and see what he can do to help.

[Pars]

I'm using FLV-110s (IIRC) which all measured ~1.62V. Dan C? Icarium?

[luvdunhill]

I'm using FLV-110s (IIRC) which all measured ~1.62V. Dan C? Icarium?

Also, Vbe of the transistors in the DynaFET are .1V or so higher than those in the other Dyna- designs, IIRC.

Yes, that's the man!

[Pars]

Same transistors as the Dynahi on the CCS (2SA1145/2SC2705). Dynalo used 2SA1015/2SC1815.

[Pars]

Put the gate stoppers right on the FET pins, on the bottom of the board. Success? I think perhaps.

With the scope hanging off the output (and POS HP), and a DMM to watch the bias current, I initially could not get things to go south at volume all the way down (~80mA bias). Taking the bias up to 150mA, I could occasionally get it to misbehave. Unplugging the phones and then plugging them back in was a good trigger. With the DMM and scope off, at 80mA I could not get it to do it. Still don't trust it though.

Using 90.9 ohms on the 2SK devices and 47 ohms on the 2SJ. Looking forward to your observations Marc.I'd be quite happy to admit I was wrong about anything I've done on this

Edited February 10, 2011 by Pars

[luvdunhill]

well, for what it's worth (I just measured this), with a Vf of 1.838V I get 595 ohms for the top (R21) and 560 ohms for the bottom (R22) to draw 2.000mA through the front end.

Hfe of both polarity (yeah I know obsessive) is 136 and the Mouser part for the LEDs are 859-LTL-4213.

So, perhaps turn it up a bit.

[Pars]

So what your saying is there is an ~0.07V difference in Vbe between Q7 and Q8? And by turn it up a bit, you mean the effective resistances across R21/R22? How did you measure this? Measure Vbe and then calculate what you needed for resistance?

EDIT: BTW, Dan C. doesn't have anything.

Edited February 10, 2011 by Pars

[luvdunhill]

So what your saying is there is an ~0.07V difference in Vbe between Q7 and Q8? And by turn it up a bit, you mean the effective resistances across R21/R22? How did you measure this? Measure Vbe and then calculate what you needed for resistance?

EDIT: BTW, Dan C. doesn't have anything.

I constructed both CCSes on a breadboard and chose a resistance that resulted in 2mA current draw. I'll use the same devices on the actual board. I'll do this for the other CCSes as well.

Looks like I'm lacking one resistance (figures). I'm going to have to take stock of what I have and make one order to shore things up.

[Pars]

I'm not sure that getting exactly 2mA into the input differential pair is worth messing with, as you will need to adjust it anyhow to balance out offset. You will have matching issues in the 2SJ/2SK pair, as well as the 2SA1349/2SC3381s (or whatever) feeding them. And then matching issues out to the outputs. I personally try to get the output offset as low as possible before resorting to the servo. Sounds like you want to let the servo handle any offset?

The 50 ea. of the 2SA1145/2SC2705 O grade that I got from Mouser at least was pretty consistent in hfe, though not between NPN / PNP. My P devices were up around 140-150 range, while the N devices were ~100-110. You probably have Y devices and hopefully better balance between the two.

[luvdunhill]

no, good matches are very hard to get between the devices 2SA1145/2SC2705.

I in fact have nearly perfect matches, the front end, and even the MOSFETs.

In general, low hfe devices are more consistant. For example, try matching the 2SA1349/2SC3381, it's very hard. That's why I used single devices... and then matched a crap load of them.

This is neither here or there though, the circuit shouldn't really care that much about this anyways...

[Pars]

Just for the hell of it, I measured the voltage drop across R21 and R22. I was kind of surprised that adjusting the pot changed this voltage? I assumed that with the LED biasing the transistor at a constant voltage, the voltage across the resistor would remain constant also. Guess not. Not sure if the Vbe changes or what.

I have a pair of 2SA1349/2SC3381 in right now, but was using closely matched pairs of 2SA970/2SC2240. Might go back to those although the duals are somewhat easier to use.

I put a set of 1 ohm metal oxide resistors in for the source resistors. Seems to be behaving better, only oscillation I could get was with the phones in, pot all the way down, but then only when amp was cold. 10 secs after turn on, it wouldn't do it. Also behaved across all 3 loads on my load box (none, 33 and 330 ohms) at all pot positions. I noticed that amb only recommends metal oxide for these in the B22.

[luvdunhill]

Has anyone else had problems with PRP mismarking resistors? I found some, and then ran across this post:

http://www.audiocircle.com/index.php?topic=67078.0

PcX is providing replacements, so I'm on hold for a bit.

[spritzer]

The only issues I've had were with uneven epoxy coatings. It's not very comforting to see the resistive element on something meant for high voltage use...

[luvdunhill]

agreed, I think that's one of the reasons they seem to mysteriously fail open. The incorrect marking is even more disconcerting. Props to PcX for taking care of me though and not being total assholes about it (other than asking what I measured them with, and it seems that the word Keithley was sufficient). That sort of makes me think that this has happened before... then I ran into that post via Google...

[Pars]

Try the other feedback setting as well, once you get accustomed to this version. Need to recruit more advocates for adding a switch here

As the testing progresses, one of my observations is a DC offset delta of ~160mV between the two feedback settings. This obviously won't work for anyone wanting to switch between the two (at least without some readjustment). I'd be curious to see what others note on this as well as any adjustments to parts values that would close this gap to 0 ideally.

I'm using 2K for R19-R20, 3.3K for R17-R18 and 1K/100R R55 / R56. I think it was about the same with 1K/221R R55/R56 as well.

EDIT: didn't see the above before posting. Yes, I'm not a fan of PRPs coating, but haven't gotten any mismarked (yet).Coating seems more like vinyl than epoxy, at least in it's durability

Edited February 11, 2011 by Pars

[luvdunhill]

Why do you think 160mV would be too much for the servo to handle? The original circuit as posted by Kevin (also, the Dynalo) didn't even have a method for offset adjustment, and he felt that was adequate.