Current feedback high speed headphone amp with diamond output buffer - Page 9

pjacobi · 21st January 2004, 10:22 AM

Fingers faster than brain, sorry to all:

Quote:

Originally posted by pjacobi

Hi Jam,

The input offset current can be made small with diamond of darlingtons, e.g. 0.3uV in this test circuit:
http://www.linearaudio.de/scratch/dmnd-test-2a.pdf

Best Regards,
Peter Jacobi

That are 0.3uA (micro Amperes), as I was discussing the input
offset current here. This is caused by the beta mismatch, whereas the the input offset voltage is caused by Vbe mismatch.

With 47k at the input, that would give about 14mV offset caused by the beta mismatch. As some extent of this can be considered systematic and unchanging, it can be trimmed lower somewhat.

Per-Anders: Fortunately, when simulating diamonds, you already have some real life mismatches in your calculations, as the NPN and PNP devices aren't identical (if not somebody has 'synthesized' one one model out of the other).

Sorry again,
Peter

Christer · 21st January 2004, 10:40 AM

Peter,

Yes, models of complementary transistors usually are different.
However, that doesn't mean they are reliable. Take some Spice
models of BJTs and plot some of the usual diagrams (curve trace
of IC vs. Vce for different Ib, Hfe vs. IC, IC vs. Vbe etc.) and
compare with the datasheets. Then you will not trust absolute
figures from simulations as much I think.

pjacobi · 21st January 2004, 11:14 AM

Hi Christer, All,

Quote:

Originally posted by Christer
[...] Take some Spice
models of BJTs and plot some of the usual diagrams (curve trace
of IC vs. Vce for different Ib, Hfe vs. IC, IC vs. Vbe etc.) and

compare with the datasheets. Then you will not trust absolute
figures from simulations as much I think.

But if you stay in class A, and well below IKF, and well above Vce(sat) (which can be done in number of applications), you are in a overwhelmingly linear regime. Using the right BJTs of course.

Look for example at the datasheets of the 2SA872/2SC1775
(but lots of small and medium power transistors can be found for this):

- Ic(Vce) is strikingly linear and a ruler and pencil will give you a consistent VAF

- Ic(Vbe) is amazingly linear (on semilog plot), beatiful physics at work!

- hfe(Ic) is linear in technical quality, for some reasonable Ic range.

And for the simple example of the diamond transistor input offset current, the SPICE result can be straightforward manually verified:

Ic is 15mA
hfe of the darlingtons is better than 10000
=> Ib is less than 1.5uA
=> Ib mismatch better than 1uA,

But of course, and I hope that I've stated that myself in every applicable posting, very low SPICE THD results should always be treated as necessary but not sufficent for real life low distortion.

Regards,
Peter Jacobi

Christer · 21st January 2004, 12:03 PM

Peter,

sounds like you have found some unusually perfect BJTs then, if
what you said was taken from the datasheets. Most BJTs aren't
that perfect and are difficult to model. You are right though that
the model can work rather well in the "linear" region. My experience
after doing some SPice model DIYing is that it is impossible to
make a realistic all-around model. One has to make a lot of
compromises, or at the very best it is possible to tune the model
to behave decently for a specific working point.

peranders · 21st January 2004, 01:20 PM

Quote:

Originally posted by pjacobi
Hi peranders,

Thank you for posting the LTSpice .asc file (in the "True Current Feedback..." thread), so that I was able to easily play with it.

I've switched back to this thread, as it seems more fitting.

As said, as a first test I substituted the current source for the input diamond with a simple 11k resistor, with no noticable effect on distortion. Both version give 0.0068% THD from the".fourier 20k V(vout)" command.

Peter can you tell me how were your settings for achieving so good distortion figures?

When I make a FFT I don't get so exciting figures.

Christer · 21st January 2004, 02:04 PM

Quote:

Originally posted by PMA

The simulation results we have for the Jung's buffer are very promissing (THD lowered about 10x).

I just made some quick simulations (well, as quick as it can be to
run extensive transient analysis and FFT). The results agree with
yours, it seem. Using CFP outputs improved the 3rd order harmonic
by 20dB, the others improved but not so much, but the 3rd order
is the dominant one in both cases. It looks promising indeed.

pjacobi · 21st January 2004, 02:13 PM

Hi Per-Anders,

Quote:

Originally posted by peranders

Peter can you tell me how were your settings for achieving so good distortion figures?

When I make a FFT I don't get so exciting figures.

Very short answer:

Don't trust the numbers too much. Essentially they all read "very low".

Short answer:

a) Everytime I start LTSpice I disable compression in the Control Panel (once I will learn how to make this permanent)

b) Starting with your .asc, I only added the directive
.fourier 20k V(vout)

c) After running the transient analysis, the results are available via "View SPICE Error Log"

Long answer:

To get consistent result in this low THD region is difficult and best assisted by a fast CPU.

More tips:

- To get a better result from the graphical FFT, I'll suggest 65536 points and Hann window.

- Better results for both .fourier and graphical FFT are achieved by shorting input and output C (or using an insanely long "Time to start saving data").

- If you're really desperate to get most accurate numbers, use two E blocks to add output and -4.3 input and analyze the sum (i.e. difference signal). You can refine this technique to something like a software "Klirrfaktormeßbrücke" (don't know the english term).

Regards,
Peter Jacobi

pjacobi · 21st January 2004, 02:27 PM

Hi Christer,

Quote:

Originally posted by Christer

[...] Using CFP outputs improved the 3rd order harmonic
by 20dB, the others improved but not so much, but the 3rd order
is the dominant one in both cases. It looks promising indeed.

I hope we are talking about vaguely similiar circuits...

Getting most of current swing from the added transistors allow all transistors in the diamond to operate at fairly constant current.

This, IMHO, is the main source of improvement.

Depending on your output current needs, it also becomes possible, to have the same transistor pairs in both stages of diamond, and at the same Iq.

For a power output stage this may require actually using darlington outputs, like in this circuit:
http://www.double-diamond.de/scratch/DARL-BUFFER.pdf

Sorry, that one may look confusing, not only because of my lacking drafting talents, but also because the stage is designed to have small gain (4dB).

Regards,
Peter Jacobi

jam · 21st January 2004, 02:57 PM

Peter,

Thanks for your response. Your latest circuit is starting to look like Hiraga's Class A amplifier.

In your original circuit, I suppose the Darlingtons would help and replacing R5 and R6 with adjustable current sources would help adjust bias and offset.

Regards,
Jam

Christer · 21st January 2004, 02:58 PM

Peter,

Yes similar but not identical. I don't use Darlingtons and I didin't
add any resistors. I just replaced the output followers with CF
pairs. There should probably be a bias resistor, but I was to lazy
to put it in and it worked fine in the simulations without it.

I have always thought of a CF pair as a CE output stage with
a complementary driver. However, perhaps it is more appropriate
to view the first transistor as a follower output and the second
one just as a kind of current booster? That seems to make sense in
the Diamond buffer case.

Thread Tools	Search this Thread
Show Printable Version Email this Page	Search this Thread: Advanced Search

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
four transistor emitter follower (diamond buffer) as power output stage?	capslock	Solid State	101	15th March 2013 07:25 PM
FS:TI(Burr-Brown) BUF634T 250mA HIGH SPEED BUFFER	YFW	Swap Meet	2	3rd July 2008 02:15 AM
CFP diamond buffer output stage with vertical FETs, thermally stable?	Russ White	Solid State	5	1st January 2008 04:16 AM
Diamond Output buffer eats trannies (thermal)	blu_line	Solid State	50	13th February 2007 08:03 PM
sziklai-darlington diamond buffer (SDDB) for power output stages	mtlin12	Solid State	5	1st April 2005 06:55 AM

New To Site?	Need Help?
Register to Participate Search Privacy Statement Contact Us	Frequently Asked Questions Did you forget your password? Mark Forums Read


	Please consider donating to help us continue to serve you. Ads on/off / Custom Title / More PMs / More album space / Advanced printing & mass image saving

21st January 2004, 10:40 AM	#82
Christer diyAudio Member Join Date: Sep 2002 Location: Sweden	Peter, Yes, models of complementary transistors usually are different. However, that doesn't mean they are reliable. Take some Spice models of BJTs and plot some of the usual diagrams (curve trace of IC vs. Vce for different Ib, Hfe vs. IC, IC vs. Vbe etc.) and compare with the datasheets. Then you will not trust absolute figures from simulations as much I think.

21st January 2004, 12:03 PM	#84
Christer diyAudio Member Join Date: Sep 2002 Location: Sweden	Peter, sounds like you have found some unusually perfect BJTs then, if what you said was taken from the datasheets. Most BJTs aren't that perfect and are difficult to model. You are right though that the model can work rather well in the "linear" region. My experience after doing some SPice model DIYing is that it is impossible to make a realistic all-around model. One has to make a lot of compromises, or at the very best it is possible to tune the model to behave decently for a specific working point.

21st January 2004, 02:57 PM	#89
jam diyAudio Member Join Date: May 2001 Location: Charlotte,NC,USA	Peter, Thanks for your response. Your latest circuit is starting to look like Hiraga's Class A amplifier. In your original circuit, I suppose the Darlingtons would help and replacing R5 and R6 with adjustable current sources would help adjust bias and offset. Regards, Jam

21st January 2004, 02:58 PM	#90
Christer diyAudio Member Join Date: Sep 2002 Location: Sweden	Peter, Yes similar but not identical. I don't use Darlingtons and I didin't add any resistors. I just replaced the output followers with CF pairs. There should probably be a bias resistor, but I was to lazy to put it in and it worked fine in the simulations without it. I have always thought of a CF pair as a CE output stage with a complementary driver. However, perhaps it is more appropriate to view the first transistor as a follower output and the second one just as a kind of current booster? That seems to make sense in the Diamond buffer case.