JFET input, MOSFET VAS, LATERAL output = Perfect!!

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Mmm - as I mentioned before, as DIYers, we have it much easier because we can make the amp tailor made for our own speakers - doesn't have to be stable into any load. My current version is still stable into a resistive load ( in spice ) with no compensation which swordfishy says sound amazing but manufacturers don't really have this option available because for them the load is an unknown.

Ringing may be the a natural function of a good amp but it can still roughen the sound. The ringing I am seeing with tricky loads is in the 100- 500khz range and amps don't have much PSRR at that frequency and once you start generating it - with ringing - it can get everywhere and really spoil the sound.

So I think it is best for us to make our amps at least stable with our own speakers and JLH would then suggest to add a little more compensation for good measure. This will probably make the sound a bit more smooth and a bit less exciting - but this comes down to a personal preference.
 
Last edited:
More compensation will not decrease the output inductance of the amplifier - in fact compensation itself increases the effective inductance. A more effective method would be to add series resistance to the amp output, which would dampen the resonance. That is what I think, anyways.

Regulating the input stage is one way of getting rid of major PSRR troubles, because it eliminates feedback loops on the rails. Probably something simple like my K-multiplier is sufficient for the vast majority of amps:

http://www.diyaudio.com/forums/powe...ntokens-cfp-cap-multiplier-2.html#post2371837

- keantoken
 
The discussion is becoming more speculative and academic. SWF is struggling valiantly to make it stable, and I don't think it's layout because he is using a ratsnest build which normally has very low parasitics. No, I think it's topology, component choice and dimensioning.

I would normally stabilise it with lag comp, then add some phase lead. More clearly is required, and this could compromise sound quality. Greg, if the issue is the highly capacitive gates on the outputs - unlikely in my view as I have a stable amp with a vertical mosfet gate and only running the VAS at 8mA - then you could always add 'drivers', EFs to drive the gates. You would insert a 'stay alive' resistor between the driver emitters, like Self's EFII.

Notable in all this discussion is the very tenuous link between sound quality and measurement. This is really important, and drives us to consider
the measurement system suggested by Earle Geddes on this forum.

Hope this helps,

Hugh
 
Last edited:
The discussion is becoming more speculative and academic. SWF is struggling valiantly to make it stable, and I don't think it's layout because he is using a ratsnest build which normally has very low parasitics. No, I think it's topology, component choice and dimensioning.

I am a little perplexed by this. Previously I was struck how closely Greg's real life square wave ringing matched what I saw in spice under similar loading.

Now in spice my 2x 3310 design is still stable without any compensation with a resistive load and also with 1nF in parallal and even without an o/p snubber so I can only conclude that either it is a necessary to have 2 3310's in parallel and 20mA, or that it is a layout issue.

sorry I can't help more. I might get one channel working today or tomorrow but my first build will be with 9610s.

mike
 
Yes Hugh, I think you're right in that it is probably not the unequal gate capacitance - I have tried compensating for it using both asymmetrical gate resistors and adding capacitance to the N channel fet. Neither seemed to help.

My new depletion mode fets have arrived. I go on holiday early Sunday morning but I'm hoping very much to get a version working with them before then...

Going to stick with the ZVP3310A for at least one more attempt, as I am convinced it has sonic advantages. If all else fails it's back to old faithful - the 9610.
 
Mikelm,

I think you're doing the right thing. The 9610 version seems rather bulletproof. It is tempting to stick it to a proper heatsink and try even more current in the VAS - like 100mA! The transconductance improves substantially with more current for that device - but then it might become unstable, who knows?

My layout is very bad and it is perfectly reasonable that it is the culprit. The signal stages are quite nicely tied together, but I do have long leads to the output fets and also the feedback path. That said, I do have it oscillating in spice with the same circuit. I really need to try it in spice with more current in the vas. I think there is something to that idea as all of my stable versions have had over 15mA.

Look forward to your impressions. My thoughts are that the 9610 is "valvey" sounding, but lacks the punch, drive and general dynamics of the ZVP devices.

Both sound good, but the flavour is slightly different!
 
Last edited:
MikelM, have you added parasitics to your simulation? Say 12nH for each film or ceramic, 10mR series resistance. Give 25nH for electrolytics, and give them no less than 50mR resistance.

This will help "reality-check" you simulation. Then try adding trace inductance in sensitive parts of the circuit.

- keantoken
 
Greg,

I think more current on the 9610 will radically increase its transconductance and yield much higher OLG and therefore, loop gain. It would need more degeneration to pull back loop gain assuming you are gunning for no more than 60dB.

If you operate a fet at around 2/3 Idss you see a much more consistent tranconductance, but even in the case of the DN2530 this would mean up around 125mA, which is absurd...

All this tends to point you back towards a bipolar VAS, doesn't it?

By all means stay with the jfet front end, but consider the VAS, and WHY the mosfet VAS sounds better. I put it to you that the wildly rising transconductance at these very low currents gives a commensurate asymmetry to the waveform, which feedback obviously ameliorates, but whichever way you look at it, this is H2 and H3 and you seem to like it.

Accordingly, see if you can insert H2 some other way.

Unequal output device source resistors?

Hugh
 
Hugh,

Fair points and probably true. What can I say? It makes sense.

I was actually thinking it might be worth leaving out one source resistor altogether for this purpose.

Mikelm,

For an input filter I am currently using a 1k series resistor followed by a 220pF cap to ground. Maybe not ideal.
 
SWF, why not try setting the OLG -3db point at over 20KHz?

I would do this by Jfet degeneration first, which seems to be the most benign way to decrease OLG.

The bootstrap idea adds a good dose of H2, but this is because the lower output MOSFET is modulating the rail which in turn modulates the bootstrap. Some may not like the idea of adding the supply filters to the sound characteristic of the amp.

- keantoken
 
Intricate's of all Fet circuit

SWF, why not try setting the OLG -3db point at over 20KHz?

I would do this by Jfet degeneration first, which seems to be the most benign way to decrease OLG.

The bootstrap idea adds a good dose of H2, but this is because the lower output MOSFET is modulating the rail which in turn modulates the bootstrap. Some may not like the idea of adding the supply filters to the sound characteristic of the amp.

- keantoken

Uhhh - 'great minds think alike'?:cool:

I like the idea of the DN2520? for current source and also think it a bit essential to 'reign in' the 1st stage Fet - but then again - I'm currently just in that geographical position IE: 'The reign in the middle'/Svenzhen, China...;)
Likewise the use of a ZVP device 'is right up my alley';)

All the best Greg
DocO:cool:
 
Edit: I'm sorry there is an error in the schematic below. I have a 221R gate stopper on the ZVP3310A

Doco, Kean, thanks for your comments. I will try decreasing OLG even further tomorrow, though it's already degenerated quite a lot!

In the mean time I have a new circuit with amazing potential I think, though I have not yet listened to it. It uses some pretty neat devices that Hugh came across - DN2530 depletion mode mosfets. The beauty of these is they can self bias and hence make a very simple and fairly latch proof VAS current source.

It looks very promising. Stability with a load is fair. Bandwith is excellent. So far no compensation.

Problem is, I get the exact same oscillation as before on the negative half cycle when the amp is unloaded. Anyway, here are some pictures.

From left:

1) Schematic.

2) 20kHz square wave into 7.5R

3) 20kHz square wave into 7.5R + 100nF.

4) 100kHz square wave into 7.5R

5) 20kHz square wave unloaded.


Now, a question for you. Given that it seems reasonably stable into a normal load, do I need to worry about the oscillation at all? It does not occur until about half power or so, and I imagine would not occur with a speaker attached.

100pF of phase lead did not help.

100pF of miller capacitance did not help.

47pf and 100pF of capacitance in parallel with the feedback resistor did not help.

Will try other values and listen tomorrow. Will also play around with gate resistors too, just in case, though I don't think that's the cause.

Thoughts?
 

Attachments

  • Image 030.jpg
    Image 030.jpg
    87 KB · Views: 502
  • 20k.gif
    20k.gif
    7.3 KB · Views: 492
  • 20k_100n.gif
    20k_100n.gif
    7.6 KB · Views: 485
  • 100k.gif
    100k.gif
    7.2 KB · Views: 473
  • oscil.gif
    oscil.gif
    9.7 KB · Views: 471
Last edited:
One annoying thing about the schematic I posted above is that a simple 1nH + 10R zobel on the output would clear that square wave right up and make it behave perfectly well with speakers attached...problem is it's oscillating with no load attached!

Hmm, what on earth is going on?

Is it really an issue anyway given that it won't oscillate with a reasonable load attached?

It's tempting to leave it as is and add a zobel to clean up that ringing and not worry about it!
 
One annoying thing about the schematic I posted above is that a simple 1nH + 10R zobel on the output would clear that square wave right up and make it behave perfectly well with speakers attached...problem is it's oscillating with no load attached!

Leave the zobel in place.

Is it really an issue anyway given that it won't oscillate with a reasonable load attached?

Yes, it's marginally stable. You don't want that.

Increase your Gate Resistor to 330 ohms or 410 ohms. Should help to stablise the amp.
 
Does it oscilate with something like 100 ohm to gnd....?? If not then that could be what the doctor would order....(they seem to treat symptoms, not curing the disease)...:)

would it be possible for you to post the depletionmode mosfet model..?

One thing that comes to mind.. is how robust is this circuit for LSP generated back emf...??

I have seen amplifier with two local feedback blocks that totally separate the voltage amplification from the current output stage....
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.