Improving the Non-Inverting chipamp

[carlosfm]

Basically, you should once and for all forget the datasheet circuit.

There are advantages in using very low value resistors, and also with (consistently) manageable DC-Offset, so that the Ci cap can be discarded.

It works, it's simple, it sounds very good.
Even without matching impedances (because I use very low values, and the amp would need a buffer), DC-offset is around 30mv.

Also, regarding the use as a minimalist integrated amp, there's a very important point that it seems few care about: using a (good) input coupling cap after the pot, DC-offset will not change at all depending on the position of the pot.

This is the way to go:

[apassgear]

Hi Carlos,

Nice to see you onece again pushing the envelope!!!

Will it work also with 3875's?

[Dr. Photon]

what is the point of the resistors in series with the output? was one of them supposed to be an inductor instead?

And what is the point of shielding the output from the amp-chip? is that to reduce interference that it might contaminate the input with?

WHy not add a capacitor to the mute pin to give a delay to the turn-on (to remove any possible thumping, especially if part of an integrated amp, when the preamp might not have stabilized yet).

depeding upon the preamp being used in an integrated amplifier, the coupling cap from the wiper of the pot to the amplifier input would not be needed. I would just remove the input resistor to ground and the input capacitor and connect the + input directly to the wiper of the pot, making sure that there will be no DC on the pot. If the pot is driven by a typical transistor/FET stage, the coupling capacitor from the stage to the top of the pot should take care of the DC. IF driven by op-amps, there should be a coupling capacitor in the amp somewhere to take it off. adding a capacitor to the feedback network to bring the gain down to unity at DC is also helpful, too.

I think the zobels snubbing the power supply raisl wouldn't be needed. I would install some smallish film capacitors right at the pins of the IC, though.

[GregGC]

Originally posted by Dr. Photon
I would just remove the input resistor to ground and the input capacitor and connect the + input directly to the wiper of the pot, making sure that there will be no DC on the pot. If the pot is driven by a typical transistor/FET stage, the coupling capacitor from the stage to the top of the pot should take care of the DC. IF driven by op-amps, there should be a coupling capacitor in the amp somewhere to take it off. adding a capacitor to the feedback network to bring the gain down to unity at DC is also helpful, too.

I think the zobels snubbing the power supply raisl wouldn't be needed. I would install some smallish film capacitors right at the pins of the IC, though.

You absolutely need the res. to GND in case the pot (with time and depending on its quality) decides not to conduct. You'll get a nice rail DC on the speakers or as a min. nasty cracking (DC/ spikes).

The res in the output are in addition to the RC network and improve the stability of the amp in case the impedance of your cables/speakers have high capacitance component.

I'm 100% convinced that you need the snubbers on the PS pins. It made a huge diff. on my amp. Just the smallish caps wont do anything (personally tried it on my GC). How much difference it'll make depends on the type of the PS caps and how revealing your speakers are.

Greg

[GregGC]

Originally posted by carlosfm
Basically, you should once and for all forget the datasheet circuit.

There are advantages in using very low value resistors, and also with (consistently) manageable DC-Offset, so that the Ci cap can be discarded.

It works, it's simple, it sounds very good.
Even without matching impedances (because I use very low values, and the amp would need a buffer), DC-offset is around 30mv.

Also, regarding the use as a minimalist integrated amp, there's a very important point that it seems few care about: using a (good) input coupling cap after the pot, DC-offset will not change at all depending on the position of the pot.

This is the way to go:

Carlos,

The only worry I have is the DC offset. I went the same course at one point and got offset more than 65-70mV, so had to increase the res. to GND. It would be OK if no input cap is used, so that at moderate volume levels the DC offset is fairly low. But I wanted to not have DC going through the pot (noise issue) and not to depend on the pot for the DC offset. With low out impedance preamp that kind of configuration will rock... Of course servo would solve the DC offset too, but I have never tried it and no time for now.

I tried the T-network, which actually does exact opposite (increases the ser res. to -IN) and to be honest didn't like the sound too much. It created a bump in the highest frequencies. My speakers are quite revealing and didn't sound as open as the low -in res. classic version. Also I found it's good to connect the FB res from -In to GND to the same star GND where the speaker is connected. The +in to GND res should also go there. And the input GND should go to the same GND-star. The lower the res. values of the NFB the more important that becomes.

Nice work!

Greg

[Nuuk]

I'll give it a go Carlos.

My 'findings' to date are that I prefer the single feedback resistor to the T-network and I prefer the inverting circuit to the non-inverted.

It may not be obvious to newbies but we should add that this circuit would be used with the snubbered PSU.

I'll be ordrering parts today (so if you have any more ideas please post them before noon )!

[pinkmouse]

Very interesting, as always, Carlos.

I have found that DC offset does seem to make a difference to the sound, and I usually spend a little time fiddling to get it as low as possible. However, this is mostly with IGCs, like Nuuk, I prefer them overall.

[peranders]

Originally posted by carlosfm
Basically, you should once and for all forget the datasheet circuit.

It's no harm to have it in mind though.

Originally posted by carlosfm
It works, it's simple, it sounds very good.
Even without matching impedances (because I use very low values, and the amp would need a buffer), DC-offset is around 30mv.

Carlos, if you _do_ read the datasheet you'll realize that your claim isn't right. If you are unlucky you may get a LM3886 with 10 mV and that gives you 300 mV offset regardless of resistor values.

Once again: C2, C3 and C4, C5 does form a parallel resonance circiut so you should get rid of C3 and C5. Why don't you use _only_ C2 tight together with C7+R4?

The IC have it's return currents to ground so I'll suspect that C6 doesn't do much good but not harm either.

Why don't you use mute, pretty convinient and has no drawbacks.

What is the advantage of using 0.25 ohms as output impedance compared to have much lower (using an inductor)?

[Tor M]

Nice woork Carlos!
Do you prefer this amp over your buffered inverted amp?


Tor Martin

[Upupa Epops]

All, what Carlos " develope " now, you can find on my PA 03 ( and my version is still better ). Not a long time ago, Carlos had said about my version, that it is s...

[carlosfm]

Originally posted by Nuuk
I'll be ordrering parts today (so if you have any more ideas please post them before noon )!

Still before noon.

Sorry, on my yesterday late schematicing I forgot to include two components on the input.
Here's the revised schematic.

[carlosfm]

Originally posted by peranders
Carlos, if you _do_ read the datasheet you'll realize that your claim isn't right. If you are unlucky you may get a LM3886 with 10 mV and that gives you 300 mV offset regardless of resistor values.

Notice that I'm not claiming.
30mv is what I have on this working amp, that is
in-in on my main system right now.
Also, if you change R1 (15k) for a 10k resistor, DC-offset will be a little lower, at around 25mv.
I used 15k because I had some good 2.2uf caps I wanted to use.

Originally posted by Upupa Epops
All, what Carlos " develope " now, you can find on my PA 03 ( and my version is still better ).

I wouldn't say that, but you are free to sell your fish.

Originally posted by Upupa Epops
Not a long time ago, Carlos had said about my version, that it is s...

Yes, because of the PSU.

[carlosfm]

As I say on the schematic, this is a power amp.
And for that role, it doesn't need an input buffer.
What it does need is a good active pre.
Notice that I give the option for a "minimalist integrated amp", which is just using a pot and increasing the gain of the amp.
I thought I should mention this option, even if it's not my favourite.

For an integrated amp I would use a gainstage at 2~3x gain and keep the gain on the LM3886 the same.
This is the way to go.

I have a preamp, with a gain of 3x.
This NI power amp doesn't need an input buffer if it's used as a power amp.
Even as a power amp, with the Inverting topology, it is recommended (and beneficial) to use an input buffer.

So, in conclusion, using the chip as a dedicated power amp as I do, I'm having very good results with this amp (exactly as on the schematic), better than what I tested before with the NI topology and the commonly used values.

[carlosfm]

Originally posted by peranders
Why don't you use mute, pretty convinient and has no drawbacks.

Aha!
At last a good question.

I just use a 10k resistor to the negative rail, never used the cap.
And there's absolutely no turn-on thump, with the high cap. snubberized PSU.

Why don't I use the cap on the muting circuit?
I don't feel very well inserting more capacitance on the negative rail, even it it's after a 10k resistor.
Maby it's just me, but I don't do that.

[GregGC]

Carlos,

You know, I've been looking at the 0.27 Ohm res at the output for so long and never actually implemented it until now (didn't believe it would make a big diff.). So, looking at your schematic I decided to connect it today. I used an ugly ceramic 5W resistor because it has some extra inductance and I think it's perfect for that place. Well, It did improve the HF. It looks like the amp has had some slight instability problem and that resistor made it sound more balanced, not as forward in the HF area. I did have the Zobel network before, but I guess it's not enough in some cases. Now I have to revisit some other variations of GC including the t-network (the advantage of much lower DC offset). One never knows....

Greg

[carlosfm]

2 // 0.47R = 0.235R.

Yes Greg, you did well.
MY suggestion for the zobel includes these.

[shusha]

I used an ugly ceramic 5W resistor because it has some extra inductance and I think it's perfect for that place. Well, It did improve the HF.

Well, I've done the same thing. Also put some 0R22 ugly ceramics (5W) and I think that sound in HF changed somehow, but will know more after few days/weeks.

BTW, just finished snubber this weekend, had no chance to implement it yet. I have listened it while I was working. LM3875 NI was connected to my mini monitors in workshop (Visonik David) and there were certainly difference in sound between my 338 reg PSU and snubber. Will try next week on main system.

[Vix]

Hi

Carlos, your Gc now gets me thinking into more options, as:

1. Use it with a Bride of Zen preamp, it should be more than enough,

2. Try to use a tube, but not only as a buffer, but also with some gain??

3. Since 0,22 ohm resistor has been added , maybe Zobel can be removed.

If I am not mistaken, Peter Daniel does not use a Zobel...

Regards

Vix

[cjd]

Carlos:

Have you experimented with the output resistor value? Or more specifically, why ~0R22?

I am still gathering parts for that SUSY amp but you now have me pondering component values again. Fortunately, a good selection of the small wattage resistors in various values is cheap relative to the other parts so I can easily afford to experiment. The larger output resistors are a tad more, so I'm curious. Since my amp will use the LM4780 with the amps paralleled I already had plans for 0R1.

And, of course, I have to know - how do you think this compares to the inverted amp?

C

[carlosfm]

Originally posted by cjd
Have you experimented with the output resistor value? Or more specifically, why ~0R22?

I didn't test 0.1R, but it seems very small to have any effect.
A resistor in series with the output, as with any op-amp, makes the chip stable driving capacitive loads.
The amp is much less picky with cables.

Originally posted by cjd
And, of course, I have to know - how do you think this compares to the inverted amp?

Right now the only inverted LM3886 I have has a regulated LM338 PSU.
The non-inverted, non-buffered LM3886 as per this schematic is sounding better, but the PSU is also different, so I can't make a direct comparison.

[GregGC]

Last night I tried replacing the 0.2 Ohm res with L||R network in ser. with the output. The res is 10 ohms/1W and I put 8 turns of wire (24AWG from a network cable) on top of it. My GC is even more civilized now. So I do recommend using it. It protects from instability in the HF range much better that the 0.2 res no matter the type of speakers. Also it allows me to go higher with the FB resistors for better DC offset (I'm using 4.7k/100k) without ill effects.
So I recommend using the L||R with high cap PSU.
Just my observation.
Greg

[analog_sa]

tried replacing the 0.2 Ohm res with L||R network in ser. with the output

This may or may not be due to the inductor as it really has very little inductance. What you are hearing is IMO the effect of removing some nasty ceramic resistors (which really sound like sh**) and replacing them which a much better sounding, slightly inductive wirewound resistor. I have been using similar resitors but wound to cancel the inductance for many years and they sound better than even boutique high power resistors.

[carlosfm]

Originally posted by analog_sa
I have been using similar resitors but wound to cancel the inductance for many years and they sound better than even boutique high power resistors.

Interesting.
Do you know the internal contruction of the wirewound resistors?
How do you cancel the inductance externaly?
Winding in the opposite direction?

Here we go again, discussing the resistors' polarity.

[Upupa Epops]

To Carlos : Not neccessary, look at Caddock resistors, typ MV 234.

[analog_sa]

Hi Carlos

How do you cancel the inductance externaly?

A good discussion of winding methods can be found here

I have only used conventional insulated copper (wire wrap or magnet wire) bifiliar wound and the ends connected in reverse phase. Of course you get only half the resistance this way. It seems like a more economical solution to follow Sully's suggestion from the quoted thread.

As i intend rebuilding my GC with 3886 following all your recommendations i guess it's winding time again

[carlosfm]

Originally posted by analog_sa
A good discussion of winding methods can be found here

I remember that thread, it's very interesting.
I never considered building my own resistors.

"If you want to make an apple pie from scratch,you must first create the universe."
(c) Carl Sagan

[analog_sa]

I never considered building my own resistors.

If you seriously consider buying high-power Caddocks per example, spending 10 min to wind your own may actually seem appealing

Tom probably spends longer but he is really a perfectionist.

[jackinnj]

Originally posted by analog_sa


If you seriously consider buying high-power Caddocks per example, spending 10 min to wind your own may actually seem appealing

Tom probably spends longer but he is really a perfectionist.

there was an item in the Linear Tech magazine (I think, correct me if I am wrong) on calculating R for current shunts by designing them on the PCB.

what are you using for resistance wire.

and another completely irrelevant sidebar discussion on resistance wire -- the reason that "toasters" in the U.S. in the 1950's lasted so long was that GE and Westinghouse bought up large stocks of government surplus nichrome wire -- the kind which was used in flying suits (and tested to a high standard) -- you couldn't kill one of these things -- us cynics believe that in order to get the market for small appliances moving they substituted inferior wire as part of programmed obsolescence. my mother uses the same toaster she received as a wedding present in 1949.

[Shoog]

Programmed obsolescence is very real and accelerating. We got a basic steam iron about 10years ago. It lasted about 7years. Since then we have had to replace one every year.
Every time I go into my local repair shop I see huge flat screen TVs piling up, with about one or two years on the clock. Our ex rental TV is still going strong after 10years.

Quite frankly it disgusts me.

Shoog

[Nuuk]

I was told by somebody in the trade that washing machines are designed to last just two years.

Given the amount of energy involved in making these items, and then diposing of them and then recyling the parts, this is nothing short of environmental suicide!

[analog_sa]

what are you using for resistance wire

As i already mentioned in this thread i use wire wrap kynar insulated wire or just magnet wire. Nichrome seems way too non-linear to expect anything nice and soldering it is pretty much impossible. That's the stuff that 'white coffin' resistors use and they all sound atrocious to my ears.

I first started winding sub one ohm resistors at the time i was into discrete SS amps and noticed that the addition of emitter resistors always brought the sound down. With home made resistors the difference was barely perceptible.

[analog_sa]

washing machines are designed to last just two years.

Apparently mine has a design fault Or it's just too old.

What's wrong with that anyway? If i was a manufacturer i'd make sure all products self destructed a day after the warranty expired.

[Nuuk]

What's wrong with that anyway? If i was a manufacturer i'd make sure all products self destructed a day after the warranty expired.

I sincerely hope that you are joking!

[analog_sa]

Relax Nuke. I'm either joking or not a manufacturer Still, i guess, many are tempted.

[Nuuk]

Relax Nuke. I'm either joking or not a manufacturer Still, i guess, many are tempted.

I'm relaxed now thanks! And recycling some more electronic parts!

[carlosfm]

Originally posted by jackinnj
the reason that "toasters" in the U.S. in the 1950's lasted so long...

Everything lasted longer and there were people to repair and parts available for replacing.
In 1989~1990 I repaired computer motherboards and monitors.
Now you don't repair. You trash it.

Talking about lasting longer, look at modern cars.
Older cars with decades old had the original carb in shape.
On new cars it is very common to have problems with the electronic injection/ignition in a question on 2~3 years.
Specially when warranty expires.
And they don't repair it.
They just replace the "black box" and ask you a huge amount of money.

[jeff mai]

Originally posted by carlosfm
using very low value resistors

I built up a quickie amp today using LM3875 with 2Kohm and 100 ohm feedback resistors as previously shown:

[carlosfm]

Originally posted by jeff mai
I built up a quickie amp today using LM3875 with 2Kohm and 100 ohm feedback resistors as previously shown:

How much dc-offset?

[jeff mai]

Originally posted by carlosfm
How much dc-offset?

About 4mV on both channels. This is less than the other chip amps I've built by quite a bit. I think all the others were 10 - 30mV

[carlosfm]

Originally posted by jeff mai
About 4mV on both channels. This is less than the other chip amps I've built by quite a bit. I think all the others were 10 - 30mV

Great.
And what value did you use from NI to ground?
R1 on my schematic (15k).

[jeff mai]

Originally posted by carlosfm
And what value did you use from NI to ground?
R1 on my schematic (15k).

I used what I had: 22K

It doesn't much matter what is used here in my set-up because the driving source impedance is always much lower (a few hundred ohms.)

[carlosfm]

And how does it sound?

[jeff mai]

Originally posted by carlosfm
And how does it sound?

I'm terrible at describing what I hear, but I'll give it a shot.

Compared to the "standard" gainclone values (22K and 680ohm) I noticed a shift away from the HF, which I always thought was a little exaggerated before. I also noticed some separation and textures that I'd not heard before - and this was on music with which I'm quite familiar.

This was with only a couple of hours on the amp. It's certainly worth trying! I think I'm going to prefer it in the long run.

[carlosfm]

Thanks, I just had to ask, because I thought I was dreaming.

Let it -in.

PS: notice the wall-to-wall soundstage.

[jeff mai]

FYI - R3 in Carlos' schematic should be at least 1/4W and 1/2W rating would be preferred. 1/8W would be exceeded at full power.

[carlosfm]

Btw I use 0.6w 1% film resistors, from Philips.

[jeff mai]

Originally posted by carlosfm
Btw I use 0.6w 1% film resistors, from Philips.

I figured you had it sorted Carlos.

I just wanted to point this out to others who might build. I saw some people using 1/8W resistors in their photos of other projects.

I used a .6W Beyschlag metal film. It's the perfect size to fit between the pins on the underside of the board.

Listening some more this morning, I remain very impressed. Good work!

[carlosfm]

With the second snubber near the chip, there are advantages in using bigger caps than 100uf.
I started testing with 470uf and yes, extreme bass gets tighter.
Then I tested 2,200uf and...


This is it!
Midband and treble are gorgeous, as ever, but the bass is really on another league.

PS: If you are wondering, I do not test amps with choral music.
There's a good recording from the beginning of the 90's (!) that always was a good test: Counting Crows - "August and everything after".
Believe me, very few amps can drive my Epos speakers with this disc.
This one does it:

[carlosfm]

The PSU is, of course, the high-cap. snubberized.
3x4,700uf per rail, MBR16100 diodes, heatsinked.
The amp, with the new caps:

[carlosfm]

No, I'm not joking.
I have this at home.