Cascading diamond buffers - a cheap low THD 10W amp with TIP41C

Administrator
Joined 2007
Paid Member
I've just been trying the TIP41 version and the distortion result is remarkable. A square wave test needs investigation though.

Interesting :)
 

Attachments

  • Capture1.JPG
    Capture1.JPG
    293.7 KB · Views: 874
  • Capture2.JPG
    Capture2.JPG
    285.3 KB · Views: 841
  • Capture3.JPG
    Capture3.JPG
    230.4 KB · Views: 837
Administrator
Joined 2007
Paid Member
:cop:

carlthess40, I have deleted your posts as they are about as far off topic as they could be for a thread like this.

If you wish to dispose of the equipment and parts then please open a thread in Swap Meet area of the site.

Thanks for your understanding
 
Member
Joined 2011
Paid Member
Suzyj, why are the push-pull transistors [Q14-Q15 in post #11] chosen to be 2N3904/6 ? Doesn't their rather terrible output resistance (low Va) roll off the low frequency gain unnecessarily? Perhaps 2N5551/2N5401 might be an improvement in those positions. Here are a couple measurements I made a year ago.

Thanks for any insight.

_
 

Attachments

  • early.png
    early.png
    81.4 KB · Views: 785
  • 2N3906.PNG
    2N3906.PNG
    32.7 KB · Views: 725
  • 2N5401.PNG
    2N5401.PNG
    28 KB · Views: 211
I've just been trying the TIP41 version and the distortion result is remarkable. A square wave test needs investigation though.

Thanks! It's wide open to HF, and loop gain is nowhere near unity at 180, so it rings like a bell given the right stimulus. Adding a HF blocking cap at the input and Zobel to the output shuts it up nicely, cleaning up the square wave response pretty well, with no effect on THD. See attached image and sim.

Interesting, good idea squeezing out top performance from widely available and affordable parts for anyone to replicate, what's the OLG for the circuit?

OLG is perhaps 105dB. Most of the linearity improvement comes from linearising the final stages with diamonds rather than just chucking piles of gain at the problem.

Suzyj, why are the push-pull transistors [Q14-Q15 in post #11] chosen to be 2N3904/6 ? Doesn't their rather terrible output resistance (low Va) roll off the low frequency gain unnecessarily? Perhaps 2N5551/2N5401 might be an improvement in those positions. Here are a couple measurements I made a year ago.

There's not terribly much in it - I think the slightly worse output resistance of the 3904/3906 is ameliorated by the slightly better Hfe at the reasonably low 3mA operating current, not to mention lower input capacitance. 5551/5401 would definitely be the transistors of choice if I wanted to push the supplies beyond +/-20V.
 

Attachments

  • Screen Shot 2017-12-10 at 11.57.06 am.png
    Screen Shot 2017-12-10 at 11.57.06 am.png
    804.8 KB · Views: 305
  • 20W BD179 Amp with diamond output and npn input.asc
    15.2 KB · Views: 122
The amp is ridiculous complex for such low output power capability.
And it will not work as shown - it is oscillating even in simulation.

Thankyou! I pride myself on the complexity of my amplifiers. No, it's not oscillating in simulation.

Why not use a LM3886 or similar. It is cheaper and more adequate for
building your first amp.

Seems to me as you only need around 10W per channel you could make use of dirt-cheap car radio amp chips. One example being TDA1554 which goes on Taobao for around $0.50 and gives you stereo hence $0.25 per channel.

An LM3886 or TDA1554 both likely have many more transistors than this design - any monolithic part will. Just because they're on the same die doesn't mean they're not there. Plus being on the same die they're not able to mix and match processes like I can with discretes. That's why these parts are pushing to break 0.01% at 10KHz.
 
Administrator
Joined 2007
Paid Member
Thanks Suzy, that certainly looks better in your sim. It was your TIP41 version I was trying yesterday and I hadn't actually noticed there was no Zobel at the time. That one still seems unstable tbh, even with the filter and Zobel network added.

For your latest file I haven't got the BD179/180 models and so swapped in TIP41/42's.

It seems very critical on models used...... this is the .asc you just posted.

(sorry, it all sounds a bit negative... and I don't mean it to be :))
 

Attachments

  • Capture.JPG
    Capture.JPG
    328.9 KB · Views: 360
Hi,

Hmm, at least in my simulation it is oscillating wildly. See attached picture.
This is no suprise as the open loop gain at 10 kHz is over 100 dB - far too high.
To stabilize this crap, you need at leat 10x the feedback caps.
And you are generous with the number of transistors, but stingy with the bias current.
Look up a transistor datasheet and you will see that a discrete transistor
needs about 5 - 20 mA for best HF performance to overcome the plenty
of capacitance in a discrete design.

I am using LTSpice IV, and had to download the BD180/BD170 models from Onsemi.
The input signal is the green one, 1 Volt sinus with a 100 mV square wave.

Regarding complexity we have different opions. I try to make my circuits
as simple and as cheap as possible.
For a +- 18 Volt amplifier with about 10-20 Watt i recommend an integrated
chip or an Opamp with output buffering (a simple diamond buffer).

And i need not count the transistors in an LM3886 :)
They work and do what they should. In addition i get overcurrent and
overheating protection for free.

Regarding the hearing limit of THD, i bet that you cannot distinguish the
difference between 0.1% and 0.001% in a blind test.
Anyway the tweeter will have much worse THD.

Greetings,
Udo

Thankyou! I pride myself on the complexity of my amplifiers. No, it's not oscillating in simulation.

An LM3886 or TDA1554 both likely have many more transistors than this design - any monolithic part will. Just because they're on the same die doesn't mean they're not there. Plus being on the same die they're not able to mix and match processes like I can with discretes. That's why these parts are pushing to break 0.01% at 10KHz.
 

Attachments

  • Oscillation1.jpg
    Oscillation1.jpg
    182 KB · Views: 376
Regarding the hearing limit of THD, i bet that you cannot distinguish the
difference between 0.1% and 0.001% in a blind test.

Might want to be careful about that one.

Earl Geddes said he could provide and example of 0.1% THD nobody could hear and another example of 0.1% THD anybody could hear.
He also said it has been known for decades that THD as a metric correlates very poorly with hearing perception.

Also, some people can hear undithered 16/44, which is getting down close to -100dB.

And Paul Frindle found that some people can hear harmonic distortion components at 80 dB below signal level even when they are more than 10 dB below the noise floor, which was verified with ABX testing.
 
Last edited:
An LM3886 or TDA1554 both likely have many more transistors than this design - any monolithic part will. Just because they're on the same die doesn't mean they're not there. Plus being on the same die they're not able to mix and match processes like I can with discretes.

Sure the TDA1554 has more transistors, but its at least an order of magnitude cheaper. You had mentioned that $30 was too high and $10 was better so I figured another order of magnitude lower might be interesting - up until now you'd not mentioned that sheer number of transistors was a metric under consideration or I'd not have made the suggestion.
 
......................

Regarding the hearing limit of THD, i bet that you cannot distinguish the
difference between 0.1% and 0.001% in a blind test.
................

Might want to be careful about that one.

Earl Geddes said he could provide and example of 0.1% THD nobody could hear and another example of 0.1% THD anybody could hear.
He also said it has been known for decades that THD as a metric correlates very poorly with hearing perception.

Also, some people can hear undithered 16/44, which is getting down close to -100dB.

And Paul Frindle found that some people can hear harmonic distortion components at 80 dB below signal level even when they are more than 10 dB below the noise floor, which was verified with ABX testing.
It depends on what distortions are being generated.
An amp that has zero crossover distortion and ~0.1% of 2nd and 3rd THD will probably pass your test.

An amplifier that has crossover distortion and a very low proportion of 2nd & 3rd can sound terrible at the 0.1% level.

HF components of distortion from 5th to 25th that are very short term but high in amplitude will read low on a conventional THD meter. But it's the high amplitude (low duty cycle) that we can hear and dislike.
 
Hmm, at least in my simulation it is oscillating wildly. See attached picture.
This is no suprise as the open loop gain at 10 kHz is over 100 dB

...

The input signal is the green one, 1 Volt sinus with a 100 mV square wave.

See post 25 above regarding HF rejection. An amplifier must be seriously over-compensated to avoid oscillation if you apply a square wave without any HF filtering.

I’d suggest you try the sim from post 25. If you’re still keen on showing it oscillating, I suggest the best approach would be inserting a square wave current directly into the base of one of the VAS transistors.

OLG at 10KHz is a healthy 76dB. Not 100dB. It’d be a neat trick if I could manage 100dB of OLG at 10KHz.
 
Maybe it would be more clear if we were to say the amplifier would be ringing, rather than oscillating, with a non-band-limited square-wave input excitation?

Yes, absolutely. Apologies for the sloppy language.

Mooly, I've attached a reworked TIP version, with HF rejection. The Zobel network wasn't useful, so I've thrown a 470n inductor (12 turns of 22AWG wire on a 6mm former) on instead. The cap across the output (C3) isn't intended as an actual component, but rather is a test of how it goes driving capacitive loads (due to it not liking the Zobel).

I much prefer the BD179/180 version - it's significantly less fussy to compensate.
 

Attachments

  • Screen Shot 2017-12-10 at 9.33.30 pm.png
    Screen Shot 2017-12-10 at 9.33.30 pm.png
    138.8 KB · Views: 943
  • Screen Shot 2017-12-10 at 9.31.54 pm.png
    Screen Shot 2017-12-10 at 9.31.54 pm.png
    680.8 KB · Views: 945
  • 10W TIP41C Amp with diamond output and npn input.asc
    10.9 KB · Views: 123
Member
Joined 2011
Paid Member
There's not terribly much in [boosting Early voltage from 40v to 280v] ...
A mere factor of seven, which increases low frequency gain by only 16.9 dB

Perhaps extra LF gain is actively and vigorously spurned; perhaps one might actually prefer to decrease LF gain instead; this is easily accomplished by increasing the loading on the collectors of Q14-Q15.

~
 
Last edited:
I played with your later amplifier from post 25 and 34 and they all burst into
wild oscillation with HF excitation at the input shortly after simulation starts
- You definition of stability seems very strange to me.

This is a complex circuit with some nice ideas, but it will need more
work to make it useful.

Attached is your asc file where you could simulate it for yourself.
The BD140,BD139, TIP41 and TIP42 transistors are downloaded
from the Onsemi Website.

Of course with the removed silly 22nF input capacitor as you need
a fast risetime in Spice to excitate oscillation!

Don't forgett that in a real cirucit the thermal noise will excitate
your circuit from DC up to infrared and will trigger any instabilities...
 

Attachments

  • 10W TIP41C Amp with diamond output and npn input-Post_34.asc
    11 KB · Views: 99
See post 25 above regarding HF rejection. An amplifier must be seriously over-compensated to avoid oscillation if you apply a square wave without any HF filtering.

This is nonsense.

An ideal amplifer behaves like a R-C filter (with a constant gain facdtor)
with exactly NO ringing. This is independent of HF excitation at the input!

And it does not need "seriously over-compensation".
 
Hey
It seems all the people suggesting the use of "cheap car audio chps" did not notice the THD target of <1ppm. You will never get that with those garbage ICs - they defy good design practise on many fronts and the results are fully expected.

Is the TC9 actually just an 8R load? Some data for it would be nice.

Years ago I built an amp for a friend with an Infinity Refernece monitor system. That cabinet had eight tweeters, six mids, a mid-bass coupler and a woofer, and he kept blowing tweeters. We rewired each speaker group to be all-parallel and built a multi-channel amp to "bi-amp" it using an active cross-over. One range eneded up being 300mR and the other 500mR, with about 6R for the mid-bass and 8R for the woofer. he used a conventional amp for the woofer and we built dedicated amps for the remaining drivers. These ran of +/-25V and used simple fully-complimentary circuits: diff input, VAS, EF2 or EF3 depending on the range. Because of the low voltage, there was no worry about secondary break-down or other faults with the MJ15002/3 outputs. We tested everything with squarewaves and adjusted the compensations, then he packed it off and listened to it. Subjectively, he could now play the system louder than ever before and everything was clean. Didn't have a THD analyser at the time...

Point of the story is partly that to get great performance at low power, you have to go to the same effort you would for high power. Also, when you want truly good performance it does take some effort even if you are trying to stick to a budget.

It is obvious suzyj has her "preferred" input stage, as we all tend to, and she knows it well. The results using "obsolete" output devices is pretty amazing. Many of the TIPs and BDs have very little data available, unlike more modern devices aimed at audio. The 2N3904/6 was used by LT in their "damn fast" composite amplifier with a slew rate of 3kV/usec. This is a diamond-based circuit using two diamonds cascaded, with a DC servo IC and another to regulate bias. So, those 4-cent BJTs have bandwidth to burn.

Back when someone suggested mosfets for the output stage, the obvious thing to do is run them CS to accommodate the low rails.This works well but in these modern times I think most people recognise that BJTs are actually superior as an audio amplifier?
 
Hey
Regarding amplifier distortion: To me, electronic distortion is wholly "unnatural" inasmuch as we did not evolve with it and therefore have only recently had to wrap our brains around filtering it out of what we hear. Sometimes this can take a lot of energy and it becomes fatiguing. But... there are some electronic distortions that come closer to natural distortions of natural sounds than others.

The high-number low-order THD of some circuits, often tube amps or simple SS, can seem benign or even pleasing depending on the music. It is generally uncluttered music that people favour when played through such amplifiers, as a too-busy midrange shows up the inadequacy of the circuit and "warm" becomes "mud".

At the opposite extreme are the low-number-THD amps that have THD harmonics stretching out to high orders. Even though those harmonics are at extremely low amplitudes, we can still detect them as something that shouldn't be there, so our brains work hard to filter it out or justify its presence.

Even though I like lots of different sounds and can tolerate widely varying system performances, I prefer systems with the lowest THD possible, and where the harmonic profile is of a low-order. Yes, you can have both. Frankly, I can't stand all the MP3 and ipad noise - it has left out too much actual information and gone way beyond mere compression.

So, suzyj's objective of <1ppm at 10kHz is admirable! This should be the goal of every designer. In my feeble simulations, I aim for all-zeroes, which in LTspice is <10ppb. It is easy to get this for normal listening power levels at 20kHz, say up to 1W - way too loud to stay in the room - and what it does above that level I don't care too much about.

Suzyj, have you made a sim at 1W? I think you'll find 10ppb or less. Always exciting when you see that!