My "audiophile" LM3886 approach

Jump to Latest
#2,821
I have used this regulator, the "TeddyReg", on low power circuits:

http://www.pinkfishmedia.net/forum/showthread.php?t=39990&highlight=teddyreg

(BC550/560 instead of fet)
...with great performance.

I have used this regulator, "The PowerReg", from the same inventor, on a chipamp for my daughter 🙂 :

http://www.pinkfishmedia.net/forum/showthread.php?t=39306

Also with great performance (not compared with unregulated supply because I trust the guy 😀 ).

Maybe regulating both chips will be too much work, but I plan to power the LM318 with a +/-12V TeddyReg as time allows. I know it will sound better than any monolytic reg. 😉 Some say it sounds even better than ALWSR. HF noise is not apparent.

I hope someone will get interested...
M
 
#2,822
UnixMan said:


I'm not the most qualified person to answer (Mauro would be the one... 🙂 ), but I don't see any reason why a regulated PSU shouldn't work. 🙄 Generally speaking, the only reasonable concerns w/r to a regulated PSU that come to my mind are:

1) regulator(s) bandwidth:

do they have low impedance and good enough regulation up to the full bandwidth of the amplifier? (i.e. at least up to ~50-100KHz?)

(mmmh, an interesting idea could be to design and build a wide-bandwidth PSU regulator using... another 3886! 😎 ).

2) noise:

regulators may introduce higher freq. noise which may be much worse than the simple PS ripple (at LF the 3886 have a quite high PSRR, but that falls down very quickly with increasing frequency...).

If these issues are properly worked out, that may be quite an interesting mod... I would be quite curious to see (hear...) which effects it may have on the sound.

BTW: in his new Evolution, also Mauro have adopted something better than the simple Zener that was used for the 318 in the original my_ref.
Click to expand...

Ok.
I have no information about the frequency response of LT1083. The only data about that matter is about ripple rejection ratio vs. frequency and the datasheet says, it has 30dB RRR at 100Khz, however as I know its all about the input ripples.
And concerning the noise figure of 1083, they say %0.003 of Vout.. And I am not sure about the contents of that "noise". But if you add my
- Output capacitor of LT1083 (47n)
- Snubber of regulator (0R47+100u)
- Onboard filter + reservoir capacitors of GC (1500u+100n)
- Onboard snubber of GC (0R1+3n3)
- PSRR of LM3886,
Are you sure; the output noise of LT1083 is that much important? Just a question? Because I am not sure..

AndrewT said:
+ve feedback?
Click to expand...

+V and feedback lines are not connected together..

In fact (as you can understand) the psu idea was based on CarlosFMs regulated snubberised GC.. And I dont have enoughly information of his results..
 
#2,827
AndrewT said:
positive feedback = +ve feedback.
Normally we use negative feedback (-ve FB, or NFB)
Click to expand...

AndrewT said:
post 2818 says this is project modified by you.
When do we believe you?
Click to expand...

Although he have re-drawn the diagram in a way-less-than-obvious way, 😀 it looks like 🙂att'n: I haven't checked carefully all the details...) Dxvideo have only modified the PSU parts. The rest of the circuit is basically still that of Mauro's my_ref design.

The 3886 is configured as a "Howland Current Pump", i.e. a VCCS (Voltage Controlled Current Source).

Indeed, you can see it as a mixed (negative+positive) feedback circuit. Or if you prefer you may think of it as a form of "current NFB" (the 3886 inputs are feed-back by a voltage proportional to the output current, with reversed phase).

For more details look back at earlier posts and/or at the other thread with Mauro's technical report on the project.
 
#2,829
Anyway,

I saw the Rev.C and decided to make it. Of course with my mods (regulated PSU and 78/79 regulators for LM318)
However, I couldnt find any 22K0 and 47K0 (non SMD) resistors in %0.1 tolerance.. I dont like to work with SMDs..
I can find Welwyn E96 series precision resistor (%0.1) from Farnell but theyre;
21K5, 22K1, 22K6 and
46K4, 47K5, 48K7..
- Do you think I must go with 22K1 / 47K5?
- Or purhase some 22K0 and 47K0 %1 s and try to match them in %0.1?
- Or purchase %0.1 805 SMDs with 22K0 / 47K0 (I dont like)
- Or any alternative idea...

Thanks in advance..
 
#2,832
AndrewT said:
matching 1% smds should be possible with a steady hand and good (aided) eyesight.
You should be able to achieve better than 0.05% with care.
Click to expand...

Dear Andrew,
I can find %0.1 SMD resistors but I cannot able to solder them and PCB design will be too complex, so I dont want them.
In other hand; to make normal %1 resistors match,
I guess my multimeter has a problem with especially 40-49K range! Normally it shows 4,5 digits however after 40K it drops to 3,5 digits (I dont understand why). I tried to make parallel two 47K s just for measuring and matching but in this case I coulnd find a method to find which is which value..
I think I will buy a new multimeter may be 5 digits or more!
Thx..

Nordic said:
I know it is frowned upon, but I used standard 1% resistors for over a year, until I could afford to get the right stuff from TP...
Click to expand...

Dear Nordic,
Do you know why Mauro keep these values in %0.1 standard? Because of theyre used for both positive and negative feedback lines? If so I can use 47K5 and 22K1 s in %0.1.
But if these values have a relation with global feedback and total performance of the amp then your way is better (with some matchings of course)..
In fact I coulndt understand even the working principle of that amplifier but everybody says "its good" than I decide to make it. And if a 5 cent resistor will spoil everything, then I can use SMD resistors...
What do you say?
 
#2,836
Dxvideo said:
Anyway,

However, I couldnt find any 22K0 and 47K0 (non SMD) resistors in %0.1 tolerance.. I dont like to work with SMDs..
I can find Welwyn E96 series precision resistor (%0.1) from Farnell but theyre;
21K5, 22K1, 22K6 and
46K4, 47K5, 48K7..
- Do you think I must go with 22K1 / 47K5?
Click to expand...

that's fine, IMHO. What's critical about tolerances is the precision of the bridge balance (that is, the equally valued resistors must be closely matched). A small tolerance of the absolute values does have little influence.

With the resistors values you've found, you'll just get a tiny bit more transconductance from the current pump... but (IMHO) not so much to significatively change the circuit behavior. I would try those without much worry.

:att'n: BTW: the only note is that this circuit (by itself, regardless of your mods) use nested NFB, and A LOT of it! It's very critical with respect to stability, and very sensible to any change, including circuit layout. This is not a trivial gainclone you can modify and/or assemble more or less "randomly". The layout must be designed with the maximum care for EVERY detail. Both signal and GROUND paths must be designed with extreme care; their length must be minimized, as must be any kind of capacitive and/or inductive coupling. That is, you must threat the circuit just about as if it is an RF circuit rather than an AF one.

And of course you'll need some real instruments (at least an oscilloscope with >= 20MHz bandwidth) to verify functioning and stability.

As long as you take care of all the details and if you know what you're doing, probably it can be built also point-to-point over a copper sheet or a piece of bare PCB used as mechanical support and ground plane (in this case you may use wood drilling tips to create insulated posts). :smash: Using a perfboard IMHO is not recommended unless you use some copper tape to create a suitable ground plane - using a bare PCB sheet then may be easier and cheaper.

In any case, any attempt to build it in a "rough" way (including a poorly designed PCB) will result in a spectacular failure... :dead:

edit: ops, my answer arrived late... you're too fast, guys! 😀
 
#2,838
UnixMan said:


that's fine, IMHO. What's critical about tolerances is the precision of the bridge balance (that is, the equally valued resistors must be closely matched). A small tolerance of the absolute values does have little influence.

With the resistors values you've found, you'll just get a tiny bit more transconductance from the current pump... but (IMHO) not so much to significatively change the circuit behavior. I would try those without much worry.

:att'n: BTW: the only note is that this circuit (by itself, regardless of your mods) use nested NFB, and A LOT of it! It's very critical with respect to stability, and very sensible to any change, including circuit layout. This is not a trivial gainclone you can modify and/or assemble more or less "randomly". The layout must be designed with the maximum care for EVERY detail. Both signal and GROUND paths must be designed with extreme care; their length must be minimized, as must be any kind of capacitive and/or inductive coupling. That is, you must threat the circuit just about as if it is an RF circuit rather than an AF one.

And of course you'll need some real instruments (at least an oscilloscope with >= 20MHz bandwidth) to verify functioning and stability.

As long as you take care of all the details and if you know what you're doing, probably it can be built also point-to-point over a copper sheet or a piece of bare PCB used as mechanical support and ground plane (in this case you may use wood drilling tips to create insulated posts). :smash: Using a perfboard IMHO is not recommended unless you use some copper tape to create a suitable ground plane - using a bare PCB sheet then may be easier and cheaper.

In any case, any attempt to build it in a "rough" way (including a poorly designed PCB) will result in a spectacular failure... :dead:

edit: ops, my answer arrived late... you're too fast, guys! 😀
Click to expand...

So you wrote just before my PCB design posting?
Then if you examined my layout, what can you say?
I applied two seperated gnd plate one is signal GND the other is dirty..
I'v tried to keep signal lines as short as possible and feeback lines are also very short (as I could make of course).
The only possible problem that I can see is, my input capacitor is very big (but its a really good cap and I dont want to leave him!) Do you think it causes any problem?
Thx..
 
#2,839
Dxvideo,

Don't forget the startup protection, there might be turnon thumps.
Will not be the dissipation on the 78L12 [which, at this point, could be 78L15] a bit too much for that kind of package? Ok, You have less current in the device [than in the original] but still, it drops from 35V to 12V..

Ciao, George
 
#2,840
Joseph K said:
Dxvideo,

Don't forget the startup protection, there might be turnon thumps.
Will not be the dissipation on the 78L12 [which, at this point, could be 78L15] a bit too much for that kind of package? Ok, You have less current in the device [than in the original] but still, it drops from 35V to 12V..

Ciao, George
Click to expand...

In fact, I was worrying about that also..
However, I will use a regulated supply (will be posted in minutes..) and adjusted to 30v (or 33), so just 😀 18v will be dropped on 78/79L s.. If you take; 10mA will be sinked from each rail by the opamp then total power dissipation will be 18mW for each regulator.
Do you think its a bit high? I coulnt see any SOA curve in the 78LXX datasheet. But I guess 18mW is not out of SOA.
And concerning the startup protection. My exist regulated design has also a turn on delay circuit (but my own desing and without a DC protection). Does Mauro's design have a DC protection also?
 
You must log in or register to reply here.
Top Bottom