YAP - Yet Another PowerAmp

Status
Not open for further replies.
Re: What could be done better...

syn08 said:
Glen,

As far as I recall, Andy has these models up and running for LTSpice.

Regarding the MOSFET matching, I'm not sure what the issue appears to be (E. is on my ignore list) but certainly something sounds fishy with the models (or models usage). What I was trying to explain is that if you are matching devices for Vgs = Vt+/-10mV @Id =150mA, then at Id = 5A the Id variation for +/- 10mV is much larger. I can proof this mathematically, but there's little reason to bother. Enough to recall that Id=f(Vgs) is parabolic. That is, the same variation of Vgs at low current has a much larger impact at large current.

I'm on your ignore list? So you are commenting on something you haven't even read? That's really smart.
Nevertheless I'm pretty sure you will sneakily look at my comments anyhow, just as you did in the past.

It can be easily shown that a Vgs variation of "e" volts creates a drain current variation of Ie=2*K*(Vgs-Vt)*e Now plug in some numbers from the data sheet (Vgs=2.8V at Id=5A and Vt=0.8V and a transconductance of 5S and you'll get Ie=400mA. If you consider +/-10mV you'll get a spread of (worst case) 800mA. Is this a lot, that is, having an about +/-10% current imbalance in the output devices acceptable? This I don't know for sure, but intuitively I believe it is not.


At 150mA we are still dealing with the weak inversion. In this region Id=f(Vgs) is NOT parabolic, rather logarithmic. Your math is wrong and as an expert on modeling, you should know better. Besides, you admitted that at Id=5A, a Vgs delta of 10mV results in Id delta of only 1%, see:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=128809&perpage=25&pagenumber=6
At 150mA I get an Id delta of 9.5mA or 6.3%. These figures are based on real measurements and in accordance with the data sheet.

Regarding the other question regarding C34. Even mr. Ignored acknowledged that it's the diamond buffer that helps discharging the charge (through the CC sources). Of course C34 helps as well! However, the same ignored omniscient claims that C34 helps turning on which I am still having some troubles understanding as of why. Just to figure it out, I removed C34 from one of the boards and the rise and fall times remained equal and increased from 150nS to about 250nS, rendering a slew rate of 160V/uS (comparing to 270 V/uS).

Clearly, there is something wrong with your measurements/observations/conclusions or whatever. Without C34 it's impossible that tf=tr, as the gate charging current is limited by the CCS (10mA), while the discharge current is determined by drive capabilities of the (diamond) emitter followers, at least 100mA, but probably more.

Anyhow, for HF signals C34 acts like a short between the two emitters of the driver stage and in terms of tr and tf, your diamond thing behaves exactly in the same manner as a normal driver stage (as used by anybody else).
The only difference consists of a bias shift if the OPS is abused by applying an unrealistic signal like a large sine of say 200kHz. C34 in your circuit will be discharged, resulting in under-bias, while in case of a normal driver, the cap will be charged, resulting in over-bias. Under normal conditions or even with a square wave of 20kHz or so, this phenomenon will not occur (provided that C34 is not much large than ~100nF).
 
TMC

Edmond Stuart said:
..............
The current gain of the front-end is simply insufficient to drive 3 pair of MOSTETs. Without 'excess' of open loop gain, FE-distortion is not well suppressed by NFB, let alone that TMC has a change to be effective.

You will need one more gain stage to reduces the front-end distortion. If done that, you will see that TMC is very effective.
BTW, in the PCP amp I've solved this issue without adding more transistors in the signal path.

Regrettable, YAP is another example of a poor implementation of my inimitable brilliant ideas. But don't blame me, rather that particular implementation of yours.

andy_c said:
(Re: Edmond's comments):

That seems like an apples-to-oranges comparison though. Sure, the things you're talking about are important for the overall amplifier. But the requirements for the output stage are somewhat different - namely, maximize the unity loop gain freq to allow the highest feedback possible when putting it in the global feedback loop. It's not clear whether or not the distortion-reducing techniques you're talking about would reduce the frequency of the non-dominant poles, thus requiring a smaller ULG freq. If they do, then it could be a six of one, half dozen of the other situation where the output stage would have lower distortion, but require a lower global ULG freq (for the amplifier as a whole).

Hi Andy,

After some thoughts I think that we are on cross purposes. What I mean is that the gain inside the compensation (Miller) loop must be increased. See also my previous comment on this topic:
http://www.diyaudio.com/forums/showthread.php?s=&postid=1160022&highlight=#post1160022
With more gain only inside the inner loop (along C2/C12), the ULG of the outer loop (along R52) will hardly be affected.

As for non-dominant poles, TMC will push them a little bit further. So, no need to lower the ULG.


Cheers,
Edmond.
 
Bonsai said:
. . . clearly you guys don't like each other. Pity, since PGP was a good collaborative effort.

Bonsai,

I promise this is my last message on this topic; in fact, I am seriously considering moving away from this virtual place. It's gone well beyond having fun and any kind of DIY spirit.

I cannot have warm feelings for anybody that insulted, in public and in private, my education, credentials, past and present jobs, DIY work, the country and even the continent I am living in. Perhaps I should not respond to any of these attacks (I know I'm not an angel myself) and understand from the very beginning that this is a war that can't be won; I cannot compete with someone that has all the time in the world, has an overinflated ego and is enjoying his own frustrations. Also, to me, an intelligent person ought to be doubtfull; somebody that believes only in his POW and refuses to look from other perspectives (the HEC debate is a very good example, there are others as well) is, to me, anything but.

As you mentioned the PGP... I was trying to avoid opening that file but perhaps a few things have to be told. Edmond's contribution to the PGP amp was designing and simulating the front end. Everything else, including the website content and deployment, is my work, on my time and dime (and well over a regular DIYer budget) and I guess anybody looking on the website could estimate the effort. Any common sense would tell that Edmond should mention this on his website, or at least acknowledge this in public.

Anyways, this week we are launching a new satellite and I'll be busy as hell. Then I'll leave for a month of vacation in Europe and be back sometimes in the second half of october. I'll keep an eye around...
 
Re: What could be done better...

syn08 said:
It is fair to reveal a few small issues with the OPS.
[snip]

I'll report back with other deficiencies as I discover them...
[snip]

In the mean time I've discovered a few:

1. The TPC resistors R54/R56 should be tied to ground instead of to the supply rails.

2. R15 (in the front-end with op-amps) should also be connected to ground instead of the non-inverting input of U6. What's even more striking is that this part of the schematic is blindly copied* from the Alexander amp (without giving him the credits btw). In the original amp this connection has a clear function: DC coupling of the input signal. In YAP however, DC is blocked by C10, so it's pointless to leave it connected to the non-inverting input. Even even worse, the output offset voltage is increased by an amount equal to the voltage drop across R11, i.e. 10k x 6uA = 60mV.

Of course there are more flaws, but enough for today.


(* except for the gain setting, Haksford cascodes and (TMC?) frequency compensation)
 
syn08 said:
................
I know I'm not an angel myself

Exactly!

and understand from the very beginning that this is a war that can't be won; I cannot compete with someone that has all the time in the world,

That's right, but for quite another reason.

Edmond's contribution to the PGP amp was designing and simulating the front end.

I have told that to many others.

Everything else, including the website content and deployment, is my work.........

History distortion.

Any common sense would tell that Edmond should mention this on his website, or at least acknowledge this in public.

Agreed, I will do that ( edit: has been done, see: http://home.tiscali.nl/data.odyssey/PGP.html ).
What about acknowledging me for the NFB-OPS and Alexander for the front-end?

Then I'll leave for a month of vacation in Europe ...

Meet you in Amsterdam. 🙂
 
Plot

The whole YAP closed loop frequency and phase responses. Market is at the 3dB point.

The measured 3dB bandwidth at full power (200W into 4ohm) is 537KHz and the phase shift at the 3dB point is 41 degrees. In the audio band, the phase shift is very small. Unfortunately, the resistive power load that I am using is not made out of anti-inductive resistors (as much as I was able to look after, such resistors are very expensive), so probably the phase measurement is a little off.

The bump, both in amplitude and phase at around 300KHz, is a little strange and needs more analysis. Could it be related to the Zoebel and/or the output inductor? Food for thought for when I'll be back...
 

Attachments

  • yap-freq.gif
    yap-freq.gif
    8.8 KB · Views: 591
anti-inductive load resistor
how bad :xeye:

because the figures, test data you will get
will not match your anti-inductive 8 Ohm speakers

well, not much you can do now, syn08
with your all screwed up test results
( i did not even bother look at your attachment .. sorry )

Linie
 
lineup said:
anti-inductive load resistor
how bad :xeye:

because the figures, test data you will get
will not match your anti-inductive 8 Ohm speakers

well, not much you can do now, syn08
with your all screwed up test results
( i did not even bother look at your attachment .. sorry )

Linie

You think such measurements should be done in a reactive speaker load? If so, which speaker model would you suggest to use?
 
andy_c said:
Maybe there is some kind of resonance that interacts with the output inductor-resistor combo. I'm assuming this is not at the feedback point, but at the other side of the output inductor?

Yes it's at the output, after Zoebel and coil. I think it's a resonance between the output coil and the Zoebel cap, strongly dampened (the bump is around 1dB) but I have to confirm this.

The 3dB corner is set by the input filter.
 
Re: Plot

syn08 said:
The whole YAP closed loop frequency and phase responses. Marker is at the 3dB point.[snip]


Syn08,

Just on the mechanics of the graphs, if I may: I guess this is done with your new HP network analyzer. Does it have an interface to a PC to get the graph, or do you use a separate program to import the graph from HPIB/GPIB? (I use 7470.exe for my 3577A).

Jan Didden
 
Re: Re: Plot

janneman said:

Just on the mechanics of the graphs, if I may: I guess this is done with your new HP network analyzer. Does it have an interface to a PC to get the graph, or do you use a separate program to import the graph from HPIB/GPIB? (I use 7470.exe for my 3577A).

It has the feature to capture a screenshot and save it on an internal disk (RAM drive or floppy) in TIFF format (about 180k/screenshot). Move the floppy to the PC and copy the file then convert to whatever format fits you.

It also has the option to download the data or screenshots via FTP. Unfortunately the network interface is AUI and 10BaseT only, you would need a AUI to RJ45 adapter ($20 on EBay, $60 brand new, mine is on order).

Of course, everything is available over HPIB as well.
 
janneman said:
Neat!
You *may* be interested in 7470.exe; it waits for you to press 'plot' on the test set (in listen mode), then reads in the data via the GPIB and draws the plot. See attached example.

Jan Didden

Edit: The is a problem to upload files. I'll try later.

I tried that... I am using a National Instruments GPIB ENET interface (to be able to control the instruments remotely, from upstairs, over TCP/IP, and that's because ultra sensitive measurements don't like humans around). For whatever reason, 7470.exe was never able to retrieve plots from my 3562A spectrum analyzer, and I tried any configuration that I was able to figure out. That's why I'm instead taking photos as screenshots 🙁
 
Status
Not open for further replies.