Oscillation in BTL LM3876/LM3886 Amp

Hi all,

I am hoping for some advice.

I have designed and built a LM3876/LM3886 amp. The basic design is:

Balanced input > Lundahl LL1544A (4:1) > dual inverting IC

Upon reflection it is a little like Jeff Rowlands of old, but I wasn't thinking of that at the time. The rational is to use the IT as a 10K > 600R step down buffer, utilising the inherent balancing and bandwidth qualities that a transformer brings (whilst aware of the higher SNR that 100x gain brings, I think the trade off is acceptable (?) ). It has provision for some Belleson power regulators, but they are bypassed at present for testing.

Here is the schematic:

View attachment AS Amp 1.3 Schematic.pdf

I got some cheap prototype PCBs made in China for the first time which was fun - they are four layers:

Layer 1 is signal.

AS Amp 1.3 Layer 1 (Signal).jpg


Layer 2 is Ground

AS Amp 1.3 Layer 2 (Ground).jpg


Layer 3 is V+ (24V+)

AS Amp 1.3 Layer 3 (24V+).jpg


Layer 4 is V- (24V-)

AS Amp 1.3 Layer 4 (24V-).jpg


Here it is in all it's glory:

2016-05-24 10.44.23.jpg


Now, I know it would be smarter to have built a hard wired version first to test the circuit, but getting some PCBs made was part of the learning experience - so happy with that, and when I play them they sound pretty good after I played around with the grounding jumpers that I installed.

I have tried to be as careful as possible with grounding and layout, but when I got a friend to help me test it yesterday it shows severe oscillation when shown continuous 1kHz signal both at the input and the output after about a second:

2016-06-14 11.07.47.jpg

2016-06-14 11.07.54.jpg

2016-06-14 22.59.33.jpg

Looking at the scope the chips both begin to oscillate and then the SPIKE protection pulls it down and the oscillation begins again, ad infinitum. Installing a 0.1uF across the inputs reduces/eliminates the behaviour, but only below -20dB referenced to a 1V input, and affects bandwidth as well.

So, now that I have created Build 1, it's time to debug this puppy to stop it eating grass. Any advice here would be appreciated.

1/ Have I correctly loaded the transformer? The spec sheets say it's 10K>600R but I guess I should measure it.... additionally the specs state a 220k self resonant point. Time to make a filter? Should I measure this in situ?

2/ Is the basic typology sound? I think so - but I am unsure whether connecting the non-inverting inputs together to SIG GND with reference to the IT is best, but I reasoned that the IT might not be perfectly balanced, whereas the resistor can be (I have also inserted a trimpot for additional adjustment)

3/ No Zobel - I am loading the amp with an 8 ohm load should this be the next step?

4/ Other ideas re circuit?

5/ PCB layout? - I really tried hard here to optimise layout, but perhaps I have done wrong. Bypass caps right at the chip pins, thick power rails routed away from signals, etc. Spot any glaring mistakes?


Just trying to create a plan of attack that seems sensible. I have been given an oscilloscope, signal generator and distortion analyser to work with.


Thanks,

Ben

PS, I live in NZ, so if I don't reply promptly it's just a time difference....
 
Your schematic is showing inverting operation, but your transformer's going to look fairly high impedance at HF, meaning you're most likely violating the stability criteria (noise gain >= 20dB). If I were you I'd reconfigure for non-inverting operation.

Thanks for this insight - I was hoping/thinking that the low impedance drive offered by the IT would power the chip nicely - I was unaware of a rising impedance load. I read that Rin (300R) defined the input impedance, but of course this is for the circuit .... is there a calculation for this? I keep finding equations for the circuit rather than the IC itself.
 
The secondary of the transformer might look lowish impedance at audio frequencies but only provided there's a low impedance output connected to the amp's input. The stability of the LM3886 though can only be guaranteed if the source impedance doesn't rise above 10% of the feedback resistor at frequencies in the MHz range. That's (meaning the impedance of the trafo above 1MHz) probably not going to be defined in the trafo datasheet so stability will be something of a gamble in inverting mode.
 
OK. Needed to read that a few times.

I'm still having a little trouble understanding why/how the transformer will couple to the input of the LM3886 in the MHz range when it obviously will not receive/transmit these frequencies from a source. Is it that the 'innate' impedance of the trafo is most likely high at these frequencies?

Lastly, is the highest frequency possible for oscillation defined by the Gain Bandwidth Product, or something else?

Happy to move to a non-inverting input, but I'm just trying to understand this problem first before I move on. I guess I could test the circuit without the trafo first, and establish whether this is in fact the source of oscillation?

Learning a bunch! Thanks.
 
Last edited:
Considering the standard models of transformers, they normally have series inductance ('leakage inductance') which means from the point of view of the LM3886, the trafo looks high impedance at high frequencies. Its only going to see that series L, nothing beyond it matters. But they also have shunt capacitance which helps counteract the leakage inductance.

The highest frequency of oscillation depends on the gain of the loop - output via feedback back to input. When the gain's gone below unity there's no hope of any oscillation due to feedback, so yeah the open loop bandwidth defines the max freq.

You can test if your oscillation is due to the high impedance (at HF) of the trafo by simply shorting out its secondary. If your oscillation ceases, that pretty much nails down the problem.
 
The RF attenuating capacitor placed across the input Rin resistor often provides that low impedance at HF/VHF/UHF

Thanks Andrew, I will try that. When we put a cap across the primary it cleaned things up considerably....across the secondary it might work a lot better. I wonder if one might combine this with a filter for the self resonance point of the trafo itself?
 
Considering the standard models of transformers, they normally have series inductance ('leakage inductance') which means from the point of view of the LM3886, the trafo looks high impedance at high frequencies. Its only going to see that series L, nothing beyond it matters. But they also have shunt capacitance which helps counteract the leakage inductance.

Yes, this makes sense. Thanks for joining those dots .... it seems obvious with hindsight :shhh:

The highest frequency of oscillation depends on the gain of the loop - output via feedback back to input. When the gain's gone below unity there's no hope of any oscillation due to feedback, so yeah the open loop bandwidth defines the max freq.

Good. At least I am not a total idiot......

You can test if your oscillation is due to the high impedance (at HF) of the trafo by simply shorting out its secondary. If your oscillation ceases, that pretty much nails down the problem.

My job for tomorrow! I hope it is, because that would mean that my PCB layout wasn't so bad after all for a first attempt. Hopefully I can measure the transformer too.

Thanks for the explanation abraxalito - it's been very helpful.

More soon,
 
The RF attenuating capacitor placed across the input Rin resistor often provides that low impedance at HF/VHF/UHF

Thanks Andrew, I will try that. When we put a cap across the primary it cleaned things up considerably....across the secondary it might work a lot better. I wonder if one might combine this with a filter for the self resonance point of the trafo itself?
but cap attached to the -IN pin creates conditions for instability.
You need to understand opamp and -IN compensation.
 
but cap attached to the -IN pin creates conditions for instability.
You need to understand opamp and -IN compensation.


Learning slowly.

The thing I love about these Power ICs is all that onboard protection means we can make a dog's breakfast and still get away with it :) It struck me that I am unsure whether to place a small cap across the Vin+/Vin- pins of each cap, of across the Vin- pins of both chips - time to see.....

I'll also read up on inverting input and cap instability if I can find something. Thanks!