Possible improvements for Aleph?

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
I've simmed with Aleph on the underbelly. And while the AB
current crossings do look reasonably smooth, the distortion
overall is relatively bad. Certainly not up to the standards
of your original circuit.

Seems the Aleph holds its own side very close to the input
swing, but the load has to be tied of the far side of the AB
current sensing resistor, which doesn't make for linearity...

So I wish to nullify my suggestion. I think you already have
an optimum lower half to cancel out your upper's distortion.
There is no "upgrade" to Aleph under these circumstances.

I'm still interested to see how or if you might get rail to rail.
I'll be watching your thread to see how that develops.
 
The one and only
Joined 2001
Paid Member
kenpeter said:
OK, you don't care for Aleph's bootstrap caps, I can dig that...

I think I've already shown more than one way how to get rid
of Aleph's caps (if its still an Aleph afterward?). Would get you
3V closer to your neg rail without upsetting how your upper
circuit derives class AB.

It's sometimes inconvenient, but not particularly difficult to get rid
of the capacitors that you solder in. It is my opinion, however, that
the worst capacitors are the ones in the gain devices themselves.

Just check out the nonlinear capacitance of Mosfets which result in
easily measured THD, much worse than a decent film cap or silk
electrolytics.

As to swinging rail to rail, this is dependent on the D-S saturation
voltage of the gain devices and resistive losses. Swinging rail to rail
also exacerbates the nonlinear capacitance issue, as this becomes
much larger at low D-S voltage.

Personally, I like to leave a few extra volts on the table to minimize
this.

:cool:
 
True, True....

But the circuit in question was clipping 4.5V from the negative rail.
Is this clipping "more linear" than tolerating the capacitive effects
of letting the output swing to an extremely low dropout? Wouldn't
strong local error correction largely hide those effects anyway?

I have no problem with bootstraps. They increase impedance and
decrease complexity. I don't see them as being significantly in the
direct "signal path." Seeing if they can be got rid of was merely a
thought excercize. In my simulations, I've seen no evidence that
doing so improves Aleph's performance. But was worthy of testing.

I merely read that our guest would not likely investigate the Aleph
option at all. Perhaps reminding that a version without caps was
possible, might change his mind?

Moot point, as it seems now that his unique AB circuit required a
complimentary drive. And Aleph's linear behavior wasn't helping.
It was beyond my skill to adapt Aleph to his circuit in a way that
didn't cause more problems than it solved. Mostly because the
load has to be tied to the far side of the AB sense resistor, and
referencing Aleph's error correction here turns it into a CCS.
 
Like the chicken and the egg, the whole taller rails solution
always kinda bugs me. In hand with overly stiff linearity and
damping factor that drive us a collision course with the wall.

You can never solve the problem entirely by moving the wall
further away. Always some dynamic load and transient that
don't get along with each other and there's clipping again...
Or some clown will hook up an even less efficient speaker.
Come on, tell me it isn't true!

Anyways, you may not see example of it yet in this thread,
but a soft clipping, slightly different top and bottom, with
linearity in the middle, and transient recovery prioritized
over linearity are my preferences.

In fact, I'd like to see an amp that identifies the root notes,
and adds or subtracts only harmonic wavelets to evade the
worst peaks. Thus sidestepping the whole clipping/IMD thing.
I mean proactively adding the same harmonics that clipping
inevitably would have, but without ever touching the rail.
The soft clipping default behavior would then only apply to
transients that don't have an identifiable root note.
 
The one and only
Joined 2001
Paid Member
When you to evaluate a topology, it's easiest to start with some
idealized components such as perfect voltage and/or current sources
so as to isolate the performance issues. You can always put real
components in later.

Eventually you might listen to the amp and see what you get.

:cool:
 
But before we risk ever getting getting back to the original plot:
Yet another glorious distraction! Like you didn't see one coming.

Quasi Complimentary Opto-Alephs?
 

Attachments

  • optoalephs20.gif
    optoalephs20.gif
    56.3 KB · Views: 419
Fatal Error: Could not open include file "irfp9140n.inc"

Uh yeah... Just delete my include command for the irf9140n.
Leftover from before it went full Quasi with the IRF530's.

4N28 optocoupler is also an LTSpice default. No need to go
scouring the web for any non-default libraries.

Here, I've saved you the trouble... Correction attached:
 

Attachments

  • optoalephs20.zip
    1.3 KB · Views: 42
Not the fastest way to abuse an optocoupler...
Been reading up on how to stretch opto bandwidth.

We can cascode the collectors to hide Miller, but there
ain't really enough voltage between gate and source
in this application to squeeze such a cascode.

We can use just collector-base junction as photodiode.
Thats very fast, but then you still have to amplify it...

Easiest cheat might be 47K ~ 470K bleeder from base
to emitter. Lose some gain, but shuts down quicker...
Letting bases float as I did, the sure ticket to slowville.

And obviously the whole thing needs some diamond
style buffering to get the input impedance up. Easy.
 
Another screwy idea: Did not exactly improve the thing I was
shooting to fix. (Drive bandwidth can't make 20KHz rail to rail),
but still might be of lame slackademic interessence or whatnot...
I'm still working on the original problem.

JFETs didn't work better than bipolars as comparators. But in
this case, matched sums (from matched pairs) make voltage
spanned across the input buffers equal. Toshiba 2sk389/2sk
109 matched pairs. I abuse the models for 2sk170/2sj74 as
JFET singles. It seemed plausible...

Equal, but not necessarily right. Got to jigger with resistors
to equalize thresholds at the MOSFET gates. Inequality is
almost assured. And mind you, the thresholds of the JFET
comparators have to be significantly less than thresholds
of the MOSFETs to be controlled.

Whatever else, the input impedance increase seems useful.
 

Attachments

  • jfork1.gif
    jfork1.gif
    55.8 KB · Views: 254
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.