High-end chipamp build project
Posted 24th January 2013 at 08:21 AM by abraxalito
Updated 15th February 2013 at 06:12 AM by abraxalito
Updated 15th February 2013 at 06:12 AM by abraxalito
Go here : Possibly the most frugal high-end sounding amp?
If you have comments, feedback, critique or improvements and don't want to put it on that thread, feel free to leave it in the comments here.
Here are the first measurements - the Nitro amp being fed by the Ozone DAC with a six-tone multiple, peak digital level -20dB. Peak output power around 100mW in an 8R speaker (connected for this test). It looks from this that the datasheet is a little pessimistic on THD, though here I'm measuring the TDA8566, not the 8561. No measurement graphs are presented in the 8566 DS and the 8561 only has plots for 2R and 4R loads, not 8R.
Schematic now added. BOM here : https://docs.google.com/spreadsheet/...Hc&usp=sharing
Additional schematic showing how to use TDA8566Q in place of TDA8561Q as the latter isn't available from Mouser. Mouser only has fewer than 300 TDA8566Q in stock so best to get in quick - I doubt they'll be getting more as this chip is going obsolete. The pins shown unconnected in the mod schematic aren't unconnected - it means they're connected the same way as for the TDA8561Q.
If you have comments, feedback, critique or improvements and don't want to put it on that thread, feel free to leave it in the comments here.
Here are the first measurements - the Nitro amp being fed by the Ozone DAC with a six-tone multiple, peak digital level -20dB. Peak output power around 100mW in an 8R speaker (connected for this test). It looks from this that the datasheet is a little pessimistic on THD, though here I'm measuring the TDA8566, not the 8561. No measurement graphs are presented in the 8566 DS and the 8561 only has plots for 2R and 4R loads, not 8R.
Schematic now added. BOM here : https://docs.google.com/spreadsheet/...Hc&usp=sharing
Additional schematic showing how to use TDA8566Q in place of TDA8561Q as the latter isn't available from Mouser. Mouser only has fewer than 300 TDA8566Q in stock so best to get in quick - I doubt they'll be getting more as this chip is going obsolete. The pins shown unconnected in the mod schematic aren't unconnected - it means they're connected the same way as for the TDA8561Q.
Total Comments 72
Comments
-
What do they say again about idle hands and the devil ... :D
Already had a glance, interesting choice ... I'll give it to you, when you get a concept you really run hard with it. On one hand you appear to be making life hard for yourself, the THD figures are an order of magnitude worse than for the National chips, and the PSRR looks pretty dismal ...
But on the other, my hacked HT machine uses TDA7269A, which acquit themselves well. Lucky to get 15W, with THD inbetween the Nationals and your car radio unit; they work for me because the bass work is offloaded to a separate box of electronics, and the speaker load is simple. And your chip amp has the feedback circuit fixed, which can only help.
So, shall be interesting how it proceeds; if you do everything right it should put up a decent show ...
I wouldn't hold my breath for Steve to do one though ... ;)
FrankPosted 24th January 2013 at 09:56 AM by fas42 -
Thanks - the concept I am running hard here with is noise modulation. When I have a hypothesis I try my damndest to break it, flog it to death, test it in many ways to see how strong it is. This is one such test of the 'noise modulation is the missing SQ metric' hypothesis. So the high THD of this chip works in my favour
You seem to have misread the DS if you think the PSRR is dismal, I shall explain that more on the thread. Still you're in good company as Charles Hansen also misread it.Posted 24th January 2013 at 12:29 PM by abraxalito -
Good attitude, I likeee ... :D
Well, I've had another look, the RR seems fairly clear, -48dB at worst. And for the whole spectrum. The LM3875 goes off the boil on the -ve side and matches the TDA at the highest frequencies for a typical specimen. I'd be looking in the region of -100dB for the whole frequency range, to be happy ...
FrankPosted 24th January 2013 at 11:04 PM by fas42 -
Posted 25th January 2013 at 02:02 AM by abraxalito
Updated 25th January 2013 at 02:04 AM by abraxalito -
I've used LTspice to look at Scott Wurcer's new baby, the Discrete Opamp, and Bob Cordell's "final", high performance power amp from his book. That's the order of performance I was getting, and what I'm looking for personally, as a safety margin.
Agreed, the level of feedback and precise test circuit changes everything. But, that sheet says, or suggests nothing better than the figures I've mentioned. Not quite sure how you extrapolate to get something more, I'm scratching my head ... care to elaborate ... :confused:?
FrankPosted 25th January 2013 at 02:54 AM by fas42 -
Order of performance for what? THD is the metric that Scott's using, he, like Doug Self pays little or no attention to what (IME) really matters for SQ which is low level IMD performance. So you're of the school that -100dB THD is a non-negotiable requirement for high-end sound? If so then there are a whole lot of high-end amps (considered subjectively) which violate that by huge margins!
Are you aware that the PSRR of LM3886 chipamps is measured relative to the input? The Philips says its measured at the output - so the closed loop gain makes >20dB difference between the two figures.Posted 25th January 2013 at 03:48 AM by abraxalito -
Posted 25th January 2013 at 04:11 PM by buzzforb -
No, you haven't missed the schematic, it'll be along in due course. Its very simple - the only external components are capacitors and you'll need a decent CM choke if you want to power from a laptop PSU - probably you'll have to wind this yourself. Or otherwise use a linear PSU, preferably regulated.
Posted 26th January 2013 at 12:12 AM by abraxalito -
Posted 26th January 2013 at 01:32 AM by buzzforb -
[QUOTE=abraxalito;bt2112]Are you aware that the PSRR of LM3886 chipamps is measured relative to the input? The Philips says its measured at the output - so the closed loop gain makes >20dB difference between the two figures.[/QUOTE]
Would be nice if it worked that way ... :)
Sorry, took a little while to get back on this but I wanted to play with a sim to get a better feel on how all this hangs together. And, you can understand why the PSSR figures are published the way they are, because the "true", open loop performances look pretty dreadful. The NFB is what makes real world PSRR work, using the open loop gain to do the hard grunt of giving good performance in this area.
So, the LM38xx PSRR graph shows what's possible if you could use the device as a voltage follower. If you lop 20db from the flat tops of the PSRR curves that gives you the real world rejection for these chips, as normally used. Around 80dB low in the frequency range, falling to below 60 for the -ve side at the high end.
So this is still better than that car chip amp, which sits somewhere below 60dB all along the range ...
FrankPosted 27th January 2013 at 12:57 AM by fas42 -
On my copy of the LM3886 DS the -ve PSRR line crosses -50dB at 20kHz. Subtract the 26dB which is the min stable closed loop gain and you get 24dB for the PSRR at the output at 20kHz.
What am I missing?
Klaus (KSTR here) has realized the -ve PSRR problem and has a way to improve it - by re-referencing all the input signals to the -ve rail, rather than to ground. This improves the PSRR on -ve rail to the CMRR, he reports improvement subjectively.Posted 27th January 2013 at 01:33 AM by abraxalito
Updated 27th January 2013 at 01:37 AM by abraxalito -
In the real world it doesn't work this way, that 6dB fall off which typically crosses the 20kHz point at around 60dB down for most opamps, for gain graphs, etc, is due to the internal frequency compensation. Which also is a feedback mechanism for PSRR, etc; it perfectly "balances" the change in feedback set by the external resistors, so to speak ...
As I said, what happens is that the top of the PSRR graph is lopped off, [I]in a straight, horizontal line[/I]. The downward slope is not affected at all, not raised or lowered in any way.
I needed to play with a sim myself to be fully confident of what I'm saying here. Try it yourself, to fully satisfy yourself as to how it behaves ...
I agree the -ve PSRR always seems to be more of an issue, and as you say there are ways, probably many ways to counter it. The main thing is not to be trapped into believing that the circuit will behave better than it actually can ...
FrankPosted 27th January 2013 at 02:54 AM by fas42
Updated 27th January 2013 at 04:11 AM by fas42 -
Posted 27th January 2013 at 03:18 PM by abraxalito -
This has now become a very interesting exercise. The real situation is more complex than I thought, your reply has prompted me to try other amplifying circuits and the answer is, it depends! The precise nature of the internal topology actually dictates how the PSRR varies or not at a particular frequency as the feedback ratio changes, which then in turn depends on how accurate the sim model is!
So, at the moment it looks neither as simple as you believe it is, nor as straightforward as I thought it was. One either needs a very accurate model of the actual chip, or real measurements done on a specimen working at the required gain level ...
FrankPosted 27th January 2013 at 11:25 PM by fas42
Updated 27th January 2013 at 11:29 PM by fas42 -
I don't have beliefs about how simple it is. I've built a model of an LM3886 and as far as I recall it didn't have the correct PSRR and I didn't investigate further how to incorporate this. Nowadays I find its more fun to modify and listen, rather than run sims. The 'real measurements' are in the end listening impressions but I do like your idea of listening to what the power supply does. I propose to rig up my Sony recorder to record the signal in one channel and the power supply modulation in the other and have a listen to that. It is probably the best way to pick up resonances between the ceramic and electrolytic caps I'm using as decouplers.
Posted 28th January 2013 at 05:06 AM by abraxalito -
An idea of how the situation is more complex can be found in the datasheet of the OPA627/637. The premium chip, '637, beloved by many audiophiles, has rejection in the +ve line of the same order as the LM38xx units. The lesser cousin, '627, sacrifices a bit of open loop gain, and gains unity gain stability, and far better PSRR. Essentially the same chip, but with different frequency compensation, and is ranked second by the listeners.
Yes, it's quite intriguing "listening" to the supply: most enlightening to hear how the amp loses its composure at the very point where the modulation becomes dramatic. It was only a cheap amp I tried the exercise on, and you could hear the music track playing, a Bizarro version thereof ... :)
FrankPosted 28th January 2013 at 06:20 AM by fas42
Updated 28th January 2013 at 06:25 AM by fas42 -
Interesting you should raise the OPA627/637 DS. I rather suspect that's an error - the OPA637 should be better because its the comp capacitor screwing up the PSRR - that cap you can see next to the pin5 label on the schematic on page1. With a decomp opamp, the cap is smaller or even non-existent. Verify that the CMRR of the 637 is better than that of the 627 - so how is it possible that with a better CMRR, the PSRR actually gets worse?
Posted 28th January 2013 at 08:27 AM by abraxalito -
Makes sense. I looked up a couple of comparable AD chips, unity gain versus decomp versions and the pattern is as you indicate.
So, the mystery of the behaviour of LM38xxs remains; the only true decider will be to plug it into a test circuit.
If you go by listening tests, the ball game changes entirely, in my book. The opamp, whatever, is almost irrelevant; I can modify the "sound" purely by playing with the power supply, and wiring, that's where the real game is ...
FrankPosted 28th January 2013 at 11:28 AM by fas42 -
I agree that the power supply is the big issue, but only once the lower hanging fruit of the architecture (LTP or not LTP) has been plucked. That's low hanging because its easy to fix by just picking the right chip. Once the opamp is blameless then the next layer of that delicious yet eye-watering onion beckons...
Posted 28th January 2013 at 11:48 AM by abraxalito -
Some nice design decisions made ... the ceramics are doing precisely what such a device, and in fact what all amplifiers relying heavily on FB need: very low power supply impedance in the 100k to 1Meg frequency range. As you say, this makes or breaks the SQ.
And you're going to some effort to clean up crud coming in via the raw power feed ...
FrankPosted 5th February 2013 at 09:51 AM by fas42