LM1875 nested Gainclone

[danielwritesbac]

Question:

When C9, C10 are 220uF//220uF, then, in that case, can R16, R17 (PI filter) be located upon the power supply board, which makes the power circuit for LM3886 look like this so that LM3886 can run cool enough to drive a 4 ohm speaker from a single chip:

AC
Transformer
Rectifier
2200uF per rail
0.1R per rail (R16, R17)
--optional power supply interconnect--
10,000uF per rail (or 5x 2200uF per rail)
100uF polyester cap per rail
3.3uF (or 4.7uF) high esr polyester cap V+ to V-
Star Ground, Speaker Ground, 0V
Cable (not long)
amp board
220uF//220uF per rail
100nF ceramic or polypropylene cap per rail
lm3886
4 ohm speaker (or 16 ohm speaker or 8 ohm speaker)

This works in practice to get more than enough thermal margin for TDA7294 and LM3886 to run 4 ohm speaker per each chip. Cleaner power makes for cooler heatsink--no extraneous heating like the 47 labs style amps.

Any suggestion for improvements would be quite welcome; but the question is about the location of R16, R17.
Is it correct?

[BrianL]

I've been following this with interest as I've had interest in trying this nested topology. However, stabilizing nested amps (SO MUCH open-loop gain) is very difficult, more than your simulations would indicate.

Also, how do you know that you are correctly modeling the power amp chip? I have yet to find a National-provided model. I have seen one or two via this web site, but they were created based on the simplified schematic provided by National. For my use I have tried to take a generic op-amp 'component' in SPICE (Tina-TI) and modify the gain/bandwidth parms based on extrapolation from the National datasheet plots. In any event, I don't think any of these are necessarily close to reality. Google did provide me a 'hit' to a company that seems to make and sell models, but I don't want to pay what they will probably charge.

Interestingly, Twisted Pear Audio sells an amp kit using a nested topology. I need to simulate that to get some clues as to how it might work.

Summary: I don't see anything that says that this thread is anything more than a thought/simulation exercise that won't work in practice. I wish it were not so, and would appreciate any comments from people who have actually built such nested power amps.

P.S., I have much experience in using such a nested topology in production test setups for IC op-amps, so an vouch for the difficulty of getting things to go right.

[linuxguru]

Originally Posted by danielwritesbac

Question:

When C9, C10 are 220uF//220uF, then, in that case, can R16, R17 (PI filter) be located upon the power supply board

AC
Transformer
Rectifier
2200uF per rail
0.1R per rail (R16, R17)
...
Any suggestion for improvements would be quite welcome; but the question is about the location of R16, R17.
Is it correct?

Yes, no problem. R16 & R17 are optional - they help a bit in quieting the 50 Hz harmonics from the rectifiers. They can be placed on the PSU board. For the rectifiers, you can consider using fast, soft recovery rectifiers like the BYV32-200 or similar. I've used them previously in a discrete amp, and they're noticeably quieter - the background 50 Hz harmonic hash is distinctly muted, even without the use of snubber capacitors.

[linuxguru]

Originally Posted by BrianL

However, stabilizing nested amps (SO MUCH open-loop gain) is very difficult, more than your simulations would indicate.

Agreed - it took a while to figure out a reliable way to stabilize a nested gainclone while still retaining favourable sonics, and it's still work in progress. The best bet is either the Jung or the nested Howland Current Pump (a la Penasa). The nested Howland will probably win on stability.

Interestingly, Twisted Pear Audio sells an amp kit using a nested topology. I need to simulate that to get some clues as to how it might work.

That's probably the MyRef Rev A or Rev C by Mauro Penasa. It's a transconductance stage (Howland Current Pump) wrapped within a traditional Voltage-Series feedback outer loop. The tricky thing there is the compensation for the LM318 front-end op-amp - Rev A and Rev C use different schemas, and are audibly different.

Summary: I don't see anything that says that this thread is anything more than a thought/simulation exercise that won't work in practice.

No better way to prove or disprove it, other than by prototyping and validating it. I'm fully confident about the stability of the 20W LM1875/NE5532 nested gainclone, but less so about the LM3886/NE5532 one. However, there's no harm in building a few prototypes and bringing them up at lower rails like +/- 24V. The worst that can happen is that some magic smoke may escape from the chipamp, but it still has SPiKe protection. We'll see...

[danielwritesbac]

Originally Posted by linuxguru

. . .No better way to prove or disprove it, other than by prototyping and validating it. I'm fully confident about the stability of the 20W LM1875/NE5532 nested gainclone, but less so about the LM3886/NE5532 one. However, there's no harm in building a few prototypes and bringing them up at lower rails like +/- 24V. The worst that can happen is that some magic smoke may escape from the chipamp, but it still has SPiKe protection. We'll see...

I have today ordered Philips NE5532's,
and also JRC4580 (M-audio audiophile 192 sound card) and 4560 (big current SIP size) for other projects,
and also the 15v zenier diodes.

It will take a week for arrival.

Currently on hand are NE5534's, JRC4556 (Audigy 2), and TL061 (jfet), along with 12v zenier diodes. So, if these won't do, then I'll just wait on the mailbox for the new parts to arrive.

[danielwritesbac]

Originally Posted by linuxguru

. . . The best bet is either the Jung or the nested Howland Current Pump (a la Penasa). The nested Howland will probably win on stability.. . .

In replacing LM1875 with LM3886, we have replaced a more-stable power op-amp with a less-stable power op-amp.
Unfortunately, LM3886 is inherently less-stable.
Inverting mode can help it somewhat, according to both empirical data and working examples of common usage that demonstrate inverting LM3886's working with out-of-spec gain settings.
This all the data I have on the LM3886 stability errata.

[linuxguru]

I haven't simulated it, but the JRC4556 should be usable in this application - it isn't terribly fast, which is an advantage here (for stability). No idea about the sonics.

[linuxguru]

If the LM3886 Howland Current Pump is used in inverting mode, then there are two alternatives for the NE5532:

i) Make the NE5532 inverting, with non-inverting voltage-series GNFB (a la Mauro Penasa MyRef); or
ii) Make the NE5532 non-inverting, with inverting voltage-shunt GNFB.

Both are feasible, but at the expense of more complicated compensation schemas, IMHO. I'll try them out if the present non-inverting topology doesn't stabilize easily or turns out to be audibly inferior.

[danielwritesbac]

Originally Posted by linuxguru

I haven't simulated it, but the JRC4556 should be usable in this application - it isn't terribly fast, which is an advantage here (for stability). No idea about the sonics.

Some historical data on that:

JRC4556, JRC4558 are frequently removed, discarded and replaced.

JRC4560, JRC4580 are in some of the favored sources and devices here at diyaudio.com.

These are simple facts, correlation is unlikely, and these facts might not apply. However, after consideration, I'd prefer to await the arrival of the NE5532 that is specific for this project.

[danielwritesbac]

My apologies! NE5532 has not yet arrived.

The reputable JRC NJM4560 did arrive and I could employ them if you think they are compatible with this project.
Opinion: So many people enjoy these that it seems a fun idea to make LM3886 "sound like" a really big NJM4560 by using nested topology.

However, I could certainly await the arrival of the NE5532's if that's the best option for success.