Problem with parallel LM3886 (PA 100)

[infinia]

Well bridging is a whole new level up from the 2x parallel discussed here.
After all you can't just slap these together and work without careful thought and checkout. It's more difficult to do on separate boards and not being able to accurately measure within few mV of DC will hamper things I'm sure.

I think a combination of AC and DC offsets can quickly bring things out of control thermally.

 

[Sebastiaan]

Well bridging is a whole new level up from the 2x parallel discussed here.
After all you can't just slap these together and work without careful thought and checkout. It's more difficult to do on separate boards and not being able to accurately measure within few mV of DC will hamper things I'm sure.

I think a combination of AC and DC offsets can quickly bring things out of control thermally.

Truth. Makes me wonder if the problems of the OP are solved already? But I am almost confident, that if he remove the 220p, AC couple each input like Andrew suggested and change the resistors like you suggested, that it must work properly.

With kind regards,
Bas

 

[infinia]

The 200pf are needed more, unless you use separate input terminations with twisted wires, look at the LM3886 data sheet regarding their need ie long in/out wires. I don't see them causing oscillations unless grounding is really wonked.

 

[DJ4MC]

Parallel woe's myself

Hey I am in the process of building a paralleled LM3886 amp. I am getting really weird AC oscillations. I am using the boards form DIY Chip Amplifier Kits, PCB's, Components and Information. and and matched the components for Ri and Rf closely. I also have a OPA2134 stage buffer I built that drives both amps. My DC offset is in the 3millivolt range. As soon as I connect the outputs together (via a .1ohm 1% dale 3W resistor on each board, they oscillate around 240Khz and have 1.5 amps flowing between them. I am using 22.6K for Rf and 680ohm for Ri values so the gains should be within normal for the chips. Any thoughts anyone?

Jules

 

[Sebastiaan]

Dear DJ4MC,

3 mV. offset is to much. Do you have a schematic of your setup?

With kind regards,
Bas

 

[infinia]

did you understand this thread so far? Buffers may make this harder if DC coupled.

I'm not sure these board support this arrangement without advance checkout and guidance!@
I think you should start a new thread in chipamp.com forum. click here

 

[DJ4MC]

I understand the thread so far. My buffers are coupled with .47uf cap with a 47K resistor to ground to the non inverting input on a OPA2134 with the output directly connected to the inverting input. The output then drives the non inverting LM3886 input on the chipamp circuit board. Sebastian is correct though I probably should start a new thread. I need to draw out the full schematic then post it in a new topic. I think that having separate boards each with its own PS decoupling caps may be the issue. I am using a start ground for everything but each board has about 8 inches of wire to the common Power Supply +/- rails and ground. I am running about +/-35 volts for the rails. I may make it easy on myself and just run one LM3886 per channel. i have been messing with this thing way too long! Here is the buffer, essentially taken from the National Ap note for paralleling these things. I am not using the 220pf cap on the inputs. I did see another schematic that had a 4.7K resistor in series with a 47pf cap dropping the overall bandwidth.

Thanks Guys!

Jules

 

[infinia]

DJ4MC another option for you> PCB designed for parallel here

Not sure what test gear you have but you need good accuracy on small DC voltages. And know how to test for really close AC gain matching. Just because your meter shows 0 mV doesn't automatically make it true!

 

[Sebastiaan]

I understand the thread so far. My buffers are coupled with .47uf cap with a 47K resistor to ground to the non inverting input on a OPA2134 with the output directly connected to the inverting input. The output then drives the non inverting LM3886 input on the chipamp circuit board. Sebastian is correct though I probably should start a new thread. I need to draw out the full schematic then post it in a new topic. I think that having separate boards each with its own PS decoupling caps may be the issue. I am using a start ground for everything but each board has about 8 inches of wire to the common Power Supply +/- rails and ground. I am running about +/-35 volts for the rails. I may make it easy on myself and just run one LM3886 per channel. i have been messing with this thing way too long! Here is the buffer, essentially taken from the National Ap note for paralleling these things. I am not using the 220pf cap on the inputs. I did see another schematic that had a 4.7K resistor in series with a 47pf cap dropping the overall bandwidth.

Thanks Guys!

Jules

Dear Jules,

It is my experience that any mismatch between the input arms of the chips can cause instability once parallel. Try to swap the values for the ones advised in the PA100 data sheet.

Using the pot meters from to null DC offset as done by others and Jeff Rowland cause other problems, and can give instability problems.

I keep repeating myself, but if you want a safe and reliable parallel chipamp, go with the servo's as showed in the application sheet.

With kind regards,
Bas

 

[Sebastiaan]

DJ4MC another option for you> PCB designed for parallel here

Seems to be a good design. However I don't like the offset DC potmeter approach. And after some searching I am not alone. See here for the problems I am talking about: http://www.diyaudio.com/forums/chip-amps/155713-parallel-lm3886-problem.html

Second, Even though (power opamps) have better specs in inverting mode, I don't like this approach in the parallel chipamps. The resistor values, especially the one in the feedback become very high, and gives more noise and distortion. Go non inverted and use low value resistors, then go with the DC servo approach and you are in heaven

Third, The Zobel resistor is very low value (I know that National recommended a extraordinary low 2.7 Ohm value). This value should be equal to the serial impedance of the loudspeaker. I recommended something between 6 and 10 Ohm. Eventually this makes the amplifier more stable.

With kind regards,
Bas

 

[infinia]

Hi Bas
but 3 servos inject much more 1/f noise than inverting LM3888's.
I think you need to do some testing of both aproaches. You should test to see that open loop DC offset once adjusted at temp doesn't really move too much. I would modify the schemo in that inverting to non inverting and use a single pot.

 

[Sebastiaan]

Hi Bas
but 3 servos inject much more 1/f noise than inverting LM3888's.
I think you need to do some testing of both aproaches. You should test to see that open loop DC offset once adjusted at temp doesn't really move too much. I would modify the schemo in that inverting to non inverting and use a single pot.

Dear Infinia,

I beg to disagree, and I have measurements to confirm that. The servo's noise is minimal above 20Hz because of the huge cap between output and negative input. It is basically an integrator and doesn't do a lot above 20Hz. and can be neglected. Admitted below 20Hz. it add's up in noise but not that much. If you can maintain a stable servo with a good time constant with even bigger cap's up 1 uF, noise almost completely disappears. The huge resistor values in the inverted schematic however pick up RFI noise, which gives worser THD+N in the result. Further more, impedance imbalance between the positive and negative input add's more noise as well.

There is really a way to add a servo with mostly benefits, lower noise and better sound quality. I can give you a slight hint. Look at the superbal balanced configuration here: Balanced Line Technology

then imagine building the same circuit, but where A2 in fig12 is a servo.

This way the input (if used with a balanced source) have a near ideal equal input impedance, so DC is offset is already better, even without servo. The opamp output may contribute a bit noise, but is often still lower then the noise coming from ground to where the circuit would be connected otherwise.

My problem with the pot meters to null offset isn't as much that there is drift with temperature changing, but more that those pot meters cause imbalance in the input stage, and give ringing and/or other instabilities. Ideally impedance seen by both input pins should be equal.

Tomorrow I will post some schematics we can brainstorm about

With kind regards,
Bas

 

[infinia]

You don't need to sell me on differential interfaces, I'm there.
yea use a very low open loop gain diff amp integrator in that figure

Re servos really there are so many issues with servos it's not funny. The integrators big caps are problems and costly too. Servos can be easy to go wacky with leakage currents, dirty, dust, PCB etc. Extra costly opamps, caps, and problems what's to like?

 

[Sebastiaan]

Dear Infinia,

Here some simulations.

Of course simulations are not the real world, but it gives an indication, at least about the self noise of the used components. Also, the simulated chips are not the LM3886's but the OPA549's since there isn't a spice model for the LM3886's.

In real world the inverted high value resistor setup will give much more noise due RFI pick up enz.

In those simulations the servo solution is 1,5 dB more silent then the inverted one, and there isn't any noticeable difference in noise between servo or non servo. I designed multiple chip amps in real life, and never gave a servo me more noise. Using the servo as virtual ground as in the superbal configuration, the noise floor even lowered.

I don't get it why some people are so highly against DC servo's, I never encountered any problems with it, all turn out to be very reliable, good sounding without a real drawback or compromise. Designing parallel chip amps with servo sound in my opinion much better then without, due the better stability with servo.

The 4th attachment contains my simple idea of how the bridged chip amp should be. In this case each chip has equal impedances for inverted and non inverted. This really contributes to a better sound quality and a more stabile chip (from experience not from simulation). It get driven by a NE5534, which get DC corrected in one of the "comp" pins. If anyone ever built such a circuit, care must be taken of a DC protection output circuit, just in case of..

(before I get flamed... the NE5534 really isn't that bad at all, and offer still one of the lowest noise in the industry, for the money a bargain opamp! )

With kind regards,
Bas

 

[Sebastiaan]

You don't need to sell me on differential interfaces, I'm there.
yea use a very low open loop gain diff amp integrator in that figure

Re servos really there are so many issues with servos it's not funny. The integrators big caps are problems and costly too. Servos can be easy to go wacky with leakage currents, dirty, dust, PCB etc. Extra costly opamps, caps, and problems what's to like?

I am there with you, that the extra servo opamps add's extra cost. A low noise fet opamp isn't cheap. That is a real set-back. The caps's aren't really a problem. I know how much ceramics get hated here by audiophiles, but I use cheap SMD ceramics with very good results. It measures good, and it sounds good.. good enough for me.

With kind regards,
Bas

 

[abraxalito]

Guys - some insights into paralleling problems might be found on this thread:

lm3886-x-2-parallel

 

[abraxalito]

Also, the simulated chips are not the LM3886's but the OPA549's since there isn't a spice model for the LM3886's.

I've made a fairly simple macromodel in LTSpice for the 3886, using the AD Alexander/Bowers appnote as a guide. Drop me a PM if you'd like to beta-test it.

(before I get flamed... the NE5534 really isn't that bad at all, and offer still one of the lowest noise in the industry, for the money a bargain opamp! )

Totally agree (dons asbestos vest quickly)

 

[Sebastiaan]

Dear,
Sorry for going slightly off topic again, but all issues raised here are far to interesting to ignore

Here an idea of my briged/paralleled chipamp.

Some explanation.

-The servo's feeding the positive inputs, serves a double function. 1: it nulls DC offset. 2: it serves a virtual low impedance reference.

-Non of the input legs from the chips are with reference to ground. The cips are driven balanced with a differential signal. This avoids the chip's have a reference to a often dirty ground, and avoid groundloops/noise.

-Being driven with a differential signal, each chips maintains a good common mode noise reduction, hence lower noise and distortion

-The driver stage is totally inspired by Mr Walt Jung. The servo for this stage is fedback to one of the two drivers, so that the DC offset seen by the input of the chips is near zero.

-Amplifier voltage gain is 29 V/V

-Half of the gain is thrown away in the input stage in order to keep the chipamps in their safety gain margin with maintaing a total amplifier gain not higher then 30dB.

This circuit is build in real life, and is for a big part the same as illustrated here, but with LM3886's. The noise is the lowest I've ever achieved with any bridged/parallel chip-amp. This configuration is super reliable and stable. DC offset per chip is as low as 6 uV! total output DC offset is below 1uV. (you read it good no mV but uV! )

The sound is big, tight and fast and black silent. I hope this circuit inspire some of us to do something with it and/or improve it.

With kind regards,
Bas

 

[Sebastiaan]

I've made a fairly simple macromodel in LTSpice for the 3886, using the AD Alexander/Bowers appnote as a guide. Drop me a PM if you'd like to beta-test it.



Totally agree (dons asbestos vest quickly)

Wow! that sounds good! The problem is, I am not really a software whizz, and LTSpice is still a hell for me to work with.. with other words I can't really get used yet to the user interface. Also laziness from my part to really take the time for it to learn it.

With kind regards,
Bas

 

[infinia]

I am there with you, that the extra servo opamps add's extra cost. A low noise fet opamp isn't cheap. That is a real set-back. The caps's aren't really a problem. I know how much ceramics get hated here by audiophiles, but I use cheap SMD ceramics with very good results. It measures good, and it sounds good.. good enough for me.

With kind regards,
Bas

No No I have nothing against ceramics used in the right places In fact I use them all the time.. But you have found the most dreadful place to put one. In a 80dB gain loop!
Oh boy you stepped on a land mine. I been blown up before by this same one too! Ive designed PLL's with a loop filter that had FM modulation sidebands by acoustic vibration. In fact also had clock recovery circuits causing bit errors big time by tapping fingers on the bench. Been there done that. Classic stuff here my Dutch friend and your fingers are in the dike.

OK ceramics are a piezoelectric generator (microphone if you will) X7R is barium titanate its one of the worst ones. Put a scope on the output of the servo and tap the PCB, you know use really slow sweep rates. you'll see 10-100s of mV.