Randall RG200G3 oscillation woes

[Khron]

So, i bought the (faulty) electronics out of a Randall RG200G3 guitar combo a few years ago, and at the time, determined that one of the output transistors had failed shorted.

Here's the kicker: the power amp employs lateral MOSFETs (@SK1058 & 2SJ162) 
You can find the schematics over here.

Finding replacements that 1) can be had, 2) aren't fakes, and 3) don't cost an arm and a d*ck was about as successful as finding hen's teeth, as you might imagine.

So recently i got a devious idea - retrofit a more conventional BJT output stage. I "mocked up" the circuit in LTspice, and got a seemingly working option easily enough. I then whipped up a board in Eagle, then drilled & scored the PCB in an afternoon.

I went with some free OnSemi samples i got ages ago - MJE15032/33 drivers, and MJW21195/6 output transistors. All was well, until i got to the biasing stage. As soon as i approach the "optimal bias" area (ie. around 2V across the Vbe multiplier), the thing breaks into oscillation  :-[

Preemptively, i had already added a 100pF drain-to-gate cap across the MOSFET source-follower that drives the power stage (since "even" the simulation broke into oscillation without that), but that doesn't seem to have helped much.

Attached you'll find the schematic of the power section (screenshot; LTspice file is HERE)

I'd dearly appreciate some thoughts and ideas  :-\

 

[PRR]

Link to Randall plan is munged. Try this:
View: https://drive.google.com/file/d/1cBQsfZ82C4eShJsMC8TJGkDnogkhvSCG/view
Download: https://doc-00-c0-docs.googleusercontent.com/docs/securesc/pt0hb3mc4veu3u4a9iq1us0fqc9v8mqc/lduia0vlo6an6r97fcbc9dgh6ag3ip8b/1533578400000/17350957661598369830/09325432396681178760/1cBQsfZ82C4eShJsMC8TJGkDnogkhvSCG?e=download

I do not know why it oscillates. (Except power amps love to do that.)

I am sure that if it even comes up near 200 Watts in 8 Ohms (even as self-oscillation), two large BJTs is not enough. 75W/pair is often an upper guide. For a stage-amp I would aim 50W/pr even 40W/pr..... 8 pair.

And BJTs *need* protection circuits!!

That driver actually won't force that output to full power (but maybe enough to blow stuff). The driver MOSFET pull-down is 100K. The original output stage bias was 2*470K or 235K, a higher impedance; because the driver was higher voltage it could swing the 70V peaks needed. Your BJT plan has 2*10K for 5K load, on 100K pull-down, and can probably swing about 10V peak.

200W amps are not easy. 200W amps that survive stage-use are harder. Obviously Randall didn't get it 100% right, and his guy knew more than you and me both.

I do not see why it needs to be Lateral MOS. There's reasonable NFB so the exact shape of the power device isn't real important. Yes, lat-fans will claim supple string sound in the soft passages.... but this is a LOUD guitar amp, not a Hi-Fi. Different MOSFETS *will* need different bias. But that's heaps easier than designing a whole-different building block to drop into the Randall topology.

A minor note: this resistor-divider center-point bias will not give low output DC offset. Randall has a servo to control that.

 

[Khron]

Link to schematics fixed.

Well, on a hunch, i stuck on a 4.5ohm load on the output, and that seems to have settled it down. I was able to then bias it to roughly where one would want a class-B amp to be (40mV or so across both emitter resistors).

Regarding the output pairs, i only put one in the LTspice schematic / simulation to not crowd the thing more than necessary. I used that as a "proof of concept" that the whole thing works as intended. The actual physical circuit does indeed have three pairs of MJW21195/6 output transistors. I might need to tack on some heatsinks onto the drivers, though...

And yes, i did reuse / reconnect the existing servo in the circuit, and it does its job as intended.

 

[CJ]

if it sucks for music you can make a VFD out of it, just wire in a signal generator and you can go down to 4 Hz ,

control irrigation pumps,  build a grain elevator or test current sensors for true rms output at 10 Hz, 
or see how quick you can burn up a Sears box fan,

 

[JohnRoberts]

Link to schematics fixed.

Well, on a hunch, i stuck on a 4.5ohm load on the output, and that seems to have settled it down. I was able to then bias it to roughly where one would want a class-B amp to be (40mV or so across both emitter resistors).

You are learning about amp design by jumping into the deep end.

A load on the output can improve stability and some amp designs use a RC network to ground at the output to provide that load only at HF (when oscillation happens).  These components need to be robust enough to survive HF oscillation or HF signal content.

Regarding the output pairs, i only put one in the LTspice schematic / simulation to not crowd the thing more than necessary. I used that as a "proof of concept" that the whole thing works as intended. The actual physical circuit does indeed have three pairs of MJW21195/6 output transistors. I might need to tack on some heatsinks onto the drivers, though...

And yes, i did reuse / reconnect the existing servo in the circuit, and it does its job as intended.

Lateral mosfets were popular for a while because they were relatively easy to drive (mostly a capacitive load), unlike bipolar power devices, so you may need to beef up the driver/pre-driver stage.  Lateral mosfet were also attractive for their temperature stability.

IMO they received a little more credit than they deserved sonically. Not all mosfet amps were created the same.

Good luck

JR

 

[Khron]

Going on PRR's idea, i did tweak the driver stage as well - the two 10K's upped to 100K, and the 100K in the source-follower got a second 100K in parallel with it. At least in simulation, that did push a pretty clean 220W or so into an 8ohm load.

Today, after the above-mentioned mods in the real-life circuit, biasing with the dummy-load connected went on without a hitch. One interesting thing though - it might've been just a probing artefact, but on the output i noticed a periodic (every 20ms) but brief (2ms) dip to -1V or so, on the output.

Sinewave tests to follow...