A little R&D from myself.
I made two modules with LM3886, which have been playing on my bench.
I did it to test these chips, and because this will end up as an amp for my father.
First, I didn't use any resistor in the non-inverting input (direct wire).
I had 28mv on one channel and around 8 mv on the other.
Everything else being equal (I even hand-select and measure the 1% resistors), it's a fact that these chips have a huge production drift.
They are so alike as Michael Jackson is to Ben Harper.
Some days before, and playing with no problems, I decided to mount some 50k multi-turn pots to see if I could lower the DC offset to around 0mv, as I do with the LM3875.
I selected the channel with more DC offset (28mv) to test this.
Unfortunately, with the 50k pot, I couldn't get lower than around 24mv.
I removed the pot and tested with resistors.
As I increased resistance, DC lowered.
With the resistors I've got to the conclusion that a 250k multi-turn pot would do it.
Next day I bought some 500k multi-turn pots.
They didn't have anything near 200k.
Well, I mounted them, adjusted, and... 0.0mv!
Then I measured the pot and... 216k!
Hey Nuuk, are you there?
Then I turned it on and there was hiss on this channel.
Not hum, hiss.
I tried to connect the ground pin on the chip to the signal star ground formed by the pot, and I heard a loud noise.
Then I removed everything and back to the original form, just a wire.
The conclusions I get from this are:
1 - Forget the resistor or pot on the LM3886, because the ground pin on the chip doesn't let you do it. Use a direct connection, as Peter Daniel does, and hope for the best.
You won't have this problem with the LM3875.
2 - The pot you will use to null the DC-offset is related to the DC offset you originally have.
I'm sure that on the other channel (8mv, same circuit) the 50k pot would do fine.
Some guys, like master Kuei, will instantly calculate the pot you need for a given circuit, but that may not be sufficient because of the production drift of these chips.
3 - This applyes only to the LM3875: a 250~300k multi-turn pot seams to be the ideal for any situation.
4 - I wonder how could we hand-select two chips before we use them, so that we don't have huge differences in DC-offset from one channel to the other...
I hope this helps.
I made two modules with LM3886, which have been playing on my bench.
I did it to test these chips, and because this will end up as an amp for my father.
First, I didn't use any resistor in the non-inverting input (direct wire).
I had 28mv on one channel and around 8 mv on the other.
Everything else being equal (I even hand-select and measure the 1% resistors), it's a fact that these chips have a huge production drift.
They are so alike as Michael Jackson is to Ben Harper.
Some days before, and playing with no problems, I decided to mount some 50k multi-turn pots to see if I could lower the DC offset to around 0mv, as I do with the LM3875.
I selected the channel with more DC offset (28mv) to test this.
Unfortunately, with the 50k pot, I couldn't get lower than around 24mv.
I removed the pot and tested with resistors.
As I increased resistance, DC lowered.
With the resistors I've got to the conclusion that a 250k multi-turn pot would do it.
Next day I bought some 500k multi-turn pots.
They didn't have anything near 200k.
Well, I mounted them, adjusted, and... 0.0mv!
Then I measured the pot and... 216k!
Hey Nuuk, are you there?
Then I turned it on and there was hiss on this channel.
Not hum, hiss.
I tried to connect the ground pin on the chip to the signal star ground formed by the pot, and I heard a loud noise.
Then I removed everything and back to the original form, just a wire.
The conclusions I get from this are:
1 - Forget the resistor or pot on the LM3886, because the ground pin on the chip doesn't let you do it. Use a direct connection, as Peter Daniel does, and hope for the best.
You won't have this problem with the LM3875.
2 - The pot you will use to null the DC-offset is related to the DC offset you originally have.
I'm sure that on the other channel (8mv, same circuit) the 50k pot would do fine.
Some guys, like master Kuei, will instantly calculate the pot you need for a given circuit, but that may not be sufficient because of the production drift of these chips.
3 - This applyes only to the LM3875: a 250~300k multi-turn pot seams to be the ideal for any situation.
4 - I wonder how could we hand-select two chips before we use them, so that we don't have huge differences in DC-offset from one channel to the other...
I hope this helps.
Good morning Carlos. Well we could all build the minimal inverted GC with the wire from non-inverting input to ground and measure the DC offset. Then mark the LM3875 that we used with that offset figure and post the results here. Then we could look for somebody else's chip with exactly the same offset and try a swap!
When I built my GC's with the wire to ground, both my amps measured 28mV so I must have been lucky! Or was I? Because when I used the 18K/22K combination, I got 0mV on one and 4mV on the other!
With the 216K from non-inverting to ground I get 0mV on one amp, and 0.3mV on the other. So it seems to depend on which method that you use to reduce the offset as to how much difference there is between channels, ie there is more of a difference with both inputs connected to ground.
Nuuk said:
You were lucky, because you had two channels with the same DC offset.
But a high value, unfortunately.
You could have had better luck still.
When I built my first GC, with 18k resistor, I had 10mv on one channel and 10.5 mv on the other.
A 50k multi-turn pot managed to cure that.
The point here is that a multi-turn pot of 250~300k will do for any situation.
And the point here too is that this doesn't work with the LM3886.
And when I said "hand-select" a chip I mean a way to measure with a multimeter on the pins, before mounting the chips.
Maby with some tests we can get to a conclusion, but well... it's not that important.
It's much more easier to test an op-amp on a socket.
I built an amp using LM3886 with resistor to ground on the non-inverting input. There was no audible noise on the output. This resistor is going to raise the noise floor itself, but the big problem tends to come from induced noise due to the high input impedance. This is going to be determined by layout, shielding, and your environment.
I don't generally recommend using this resistor due to the possibility of noise problems, but it IS possible to get LM3886 working in the same manner as LM3875. (i.e. no hum or hiss with input impedances matched)
I don't generally recommend using this resistor due to the possibility of noise problems, but it IS possible to get LM3886 working in the same manner as LM3875. (i.e. no hum or hiss with input impedances matched)
tiroth said:
How?
What about the ground pin on the LM3886?
If you put a resistor on the non-inverting input, instantly you have a ground loop.
The amount of noise is related to the value of the resistor, the bigger it is, more noise you get.
If you connect the ground pin to that resistor, you'll have noise too.
If you don't connect the ground pin you'd better stay away, because it makes a big noise.
The problem here IS the ground pin of the LM3886, which the LM3875 doesn't have.
What's the solution for this other than no resistor on the non-inverting input?
tiroth said:
No, it's because the ground pin is connected to the star ground and the non-inv. input has a resistor.
That gives problems. It's a ground loop.
If you connect the ground pin to the signal star ground (the resistor) you may have problems if you need around 200k, for instance, to minimize DC offset.
Maby with 18k or lower you may not notice.
On one channel I have 8mv DC without a resistor, and on this one I may not notice that problem if I put a pot, because a 50k pot will do, and the final resistance will be lower than 18k.
But on the other channel, with 28mv DC, I need 216kohms to eliminate DC offset, and then there's noise.
This is lottery, because the chips are not matched.
That's what I'm trying to say, don't bother with the resistor or pot with the LM3886 if you have around 20~30mv DC offset, because it will not work.
But if you have less than 10mv, you can try to optimize that, you'll use a small value resistor and you may not have noise.
The LM3875 is different, you won't have these problems.
There are no two equal things in the world
Just to exemplify how the chips are different, on a preamp I made I measured the DC offset on the OPA627s.
On one I had 1mv, on the other, 0.2mv.
It's easy to test, because I have sockets.
I picked another one and now I have 0.2mv on one channel and 0.3mv on the other.
Then I removed the input cap, because it was doing nothing, only mucking up the sound.
No cap is better than any cap.
So, the circuit is the same, only the chips are changed and I could easilly optimize and match the op-amps.
With the LM chips in your GC you may have very different results depending on the chip.
Believe me, if you do another GC you may have very different results.
Just to exemplify how the chips are different, on a preamp I made I measured the DC offset on the OPA627s.
On one I had 1mv, on the other, 0.2mv.
It's easy to test, because I have sockets.
I picked another one and now I have 0.2mv on one channel and 0.3mv on the other.
Then I removed the input cap, because it was doing nothing, only mucking up the sound.
No cap is better than any cap.
So, the circuit is the same, only the chips are changed and I could easilly optimize and match the op-amps.
With the LM chips in your GC you may have very different results depending on the chip.
Believe me, if you do another GC you may have very different results.
If we aren't talking about the substrate then I don't see where the loop is.
GND pin ------\
+in -- 220k --\
Cap1 ---------- GND
Cap2 ---------/
Supply -------/
Every ground node has its own line to ground. ???
On the GC I am talking about there is 220k to ground at the noninverting input. With external power supply there is no noise.
GND pin ------\
+in -- 220k --\
Cap1 ---------- GND
Cap2 ---------/
Supply -------/
Every ground node has its own line to ground. ???
On the GC I am talking about there is 220k to ground at the noninverting input. With external power supply there is no noise.
carlmart said:
I don't have a scope.
Anyway, I built a GC with LM3875 and no problems at all.
And no, no cap bypassing the offset resistor.
tiroth said:
Well, that was in the original Thorsten schematic.
He now recommends a 18k resistor in place of the 220k and 0.1uf cap.
Anyway, or I did something wrong with the LM3886 circuit, or I don't know what's happening.
Anyway, it's playing beautiful sounds for two weeks now, without a resistor in the non-inverting input, and I don't see why I would have oscillations.
I have the 1000uf caps and the 0.1uf bypass caps very near the chip's pins.
What I can tell you is that with the LM3875 I tested with 18k, 50k multi-turn pots and no problem at all.
It stayed with the pots in place, and I've got 0.0mv on both channels.
When I tried to do the same with the LM3886, there was noise.
What bugs me is that misterious ground pin that mucks it all.
The LM3875 behaves exactly as a normal op-amp.
The LM3886 doesn't, because an op-amp doesn't have a ground pin.
Make no mistake, I'm not talking big noise.
It's that I'm used to the silence of my LM3875 GC.
I take the volume to max and absolutely nothing, no noise, even with my ear on the speaker.
It isn’t a ground loop... It is noise generated by the non inverting input current (only fets and tubes dont generate current noise) the higher the resistor, higher the noise!
Think why there is a cap in parallel with that resistor
Obviously this is completely wrong...
The Lm3886 is an opampCheers
Sometimes I sit down and read...
...and I found this on the AN-1192 appnote from National, on page 5:
"The non-inverting input resistance, Rb, is used to create a
voltage drop at the non-inverting terminal to offset the
voltage at the inverting input terminal due to the input
bias current flowing from the output to the inverting input.
Generally, the value of this resistor equals the value of
the feedback resistor so that the output offset voltage will
be minimized close to zero. However, if this value is too
large, noise can easily be picked up which will be amplified
and seriously affect the THD+N performance. If the
resistor is eliminated and the terminal is grounded, the
THD+N performance will be much better, but it will not
necessarily be optimized. By connecting the
non-inverting input directly to a ground reference, any
noise on that ground will be directly injected into the
amplifier, amplified and thus will also affect the THD+N
performance. The best solution is to use a value of
resistance not too large to pick up stray noise and not too
small as to be affected by ground noise fluctuations. The
value used in the previous plots was a 3.32k resistor. It
should be noted that this is not necessarily the optimized
value and can change with varying circuit layouts."
It seams that my conclusions are correct.
If you can't remove DC offset with a low value resistor (say... until 50k), then don't bother with it.
You'll have noise with a high value resistor.
Or use a low value resistor or a wire (if DC is low).
...and I found this on the AN-1192 appnote from National, on page 5:
"The non-inverting input resistance, Rb, is used to create a
voltage drop at the non-inverting terminal to offset the
voltage at the inverting input terminal due to the input
bias current flowing from the output to the inverting input.
Generally, the value of this resistor equals the value of
the feedback resistor so that the output offset voltage will
be minimized close to zero. However, if this value is too
large, noise can easily be picked up which will be amplified
and seriously affect the THD+N performance. If the
resistor is eliminated and the terminal is grounded, the
THD+N performance will be much better, but it will not
necessarily be optimized. By connecting the
non-inverting input directly to a ground reference, any
noise on that ground will be directly injected into the
amplifier, amplified and thus will also affect the THD+N
performance. The best solution is to use a value of
resistance not too large to pick up stray noise and not too
small as to be affected by ground noise fluctuations. The
value used in the previous plots was a 3.32k resistor. It
should be noted that this is not necessarily the optimized
value and can change with varying circuit layouts."
It seams that my conclusions are correct.
If you can't remove DC offset with a low value resistor (say... until 50k), then don't bother with it.
You'll have noise with a high value resistor.
Or use a low value resistor or a wire (if DC is low).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.