Amplifier loses stability on loading

Hello everyone!

I decided to design a 30W RMS 2 channel amplifier using smd components for the input,the vas, the driver stage and the mje15028/29 bjts for the class ab output stage. For the input stage transistors i have used the dual bjt in package transistors BC847/57BS, the bc847/57 for the vas, and the BCX53/56 transistors for the driver stage. I have set the gain to 27db and the miller cap to 120p. Tail current is set to around 2mA and the vas current to around 6mA. I have set the vbe multiplier to around 1V at the point before output transistors start conducting. Pi filters at the rails, the resistors are shorted because i didnt have the component on hand but no big deal.

I have soldered all the component by hand, and i have measured the current source resistors and they seem to be within spec. I also have soldered 2 boards to see if i have made a soldering mistake but the same issue.

The issue that i have is that when the output is not loaded the amplifier is outputting a nice and clean signal and within spec. Once i add a resistive load at the output , even a 100omhs resistors the amp loses stability and starts oscilating.

I also noticed that when signal is 0 and i have a load connected i have a stable few milivolts noise each second. Noted!

What i have tested:
1) tried to slightly increase the miller cap. no noticeable difference.
2) i have disoldered the output bjts and drive the output with the drivers as i can do that and see if it is mje issue. no difference.
3) applied load directly at output resistors to see if its feedback or an output trace inductance issue. no difference
4) i have added 100p caps by hand between output and various nodes. no difference.

I am struggling to find the issue for days and i have soldered/disoldered to test different stuff, so that's why the board looks like frankenstein!

Anyway, i have attached scope meaurements, pcb layout/schematic.
If i have made an obvious mistake or you have encountered a similar instability issue it would realy help me.

Thanks in advance!

Best regards
Patsaoglou P. GR
 

Attachments

  • pcb.PNG
    pcb.PNG
    42.8 KB · Views: 181
  • 20240131_160209.jpg
    20240131_160209.jpg
    679.9 KB · Views: 112
  • 20240131_160231.jpg
    20240131_160231.jpg
    367 KB · Views: 112
  • 20240131_160452.jpg
    20240131_160452.jpg
    334.6 KB · Views: 111
  • 20240131_160523.jpg
    20240131_160523.jpg
    609.2 KB · Views: 115
  • 20240131_161255.jpg
    20240131_161255.jpg
    348.8 KB · Views: 173
  • sch.PNG
    sch.PNG
    22.8 KB · Views: 174
Perhaps including the emitter follower in the VAS Miller loop is the cause.

The output RL simply prevents capacitive loads from causing instability,
it does not improve the amplifier's inherent stability, and is more of a bandaid.
 
Is there any other thing that could cause such an issue?
Instability generally comes from feedback loops when the 'phase margin' is not quite enough to keep it together, and it wants to turn into an oscillator.

I'm currently playing with simulations that avoid feedback on the output devices...

The easy way is to keep everything the same, unloaded, but drive a duplicate pair of (heatsunk) output devices that have no GNFB connection, and are simply there for driving the speakers.

So the amplifier feedback all works nicely with no interference from the speaker, with 'dummy' devices, just there to linearise them with the NFB, but the actual speaker drivers don't panic when the cone whizzes about etc. This is just my thoughts - but it's an idea, and it's DIY 😀
Or better still, just use a linear VAS and drive the followers from that, no duplicate transistors required. They are only followers anyway, they don't really need any special drive, they will just follow the VAS, according to their gain (consider darlingtons).

This little trick (zero feedback) also allows you to completely decouple the speaker ground from the signal ground, because there's no longer a GNFB loop from the speaker output to signal ground, to scold you.

I haven't tried this in practice yet, but it should result in a very stable amp 🙂
 
Looks similar with what I'm building.
No expert here but no zobel on the output asks for trouble.
Also your Vas if you clip the amp will behave weird and may also latch to rail. One diode in parallel with the Cdom would help.
4.7n in the input filter looks weird to me, maybe I misunderstand it.
- Bruno.
 

Attachments

  • received_3608461986033970.jpeg
    received_3608461986033970.jpeg
    53.3 KB · Views: 86
  • orca-image--668000175.jpeg
    orca-image--668000175.jpeg
    88.3 KB · Views: 75
Perhaps including the emitter follower in the VAS Miller loop is the cause.
Tested driving vas directly from input stage same issue.
o expert here but no zobel on the output asks for trouble.
Its logical but shouldnt it work without it good enough!?
4.7n in the input filter looks weird to me, maybe I misunderstand it.
It is low pass @330MHZ
Instability generally comes from feedback loops when the 'phase margin' is not quite enough to keep it together, and it wants to turn into an oscillator.
Probably thats the case!
Also your Vas if you clip the amp will behave weird and may also latch to rail
I alsocame across that rail latching but shouldnt that be an issue at clipping

Thank you guys for the responses!!!
 
Instability generally comes from feedback loops when the 'phase margin' is not quite enough to keep it together, and it wants to turn into an oscillator.

I'm currently playing with simulations that avoid feedback on the output devices...


The problem is most of the distortion in a 3-stage amp is from the output devices and is very load-dependent, which is why feedback is needed in the first place - typical output stages are -70dB to -80dB distortion simulated and worse in real life (into loads), but with feedback -110dB is relatively easily obtainable at 1kHz (much harder at 10kHz). Various compensation schemes can help, which effectively reduce the influence of the output devices at higher frequencies which is kind of what you want - check out two pole and inclusive compensation schemes.

The Zobel and inductor are standard output networks for any such amp BTW. You can omit the inductor for short cables (for instance testing with a dummy load), but its a wise thing to include for generic use.
 
  • Like
Reactions: Patsaoglou
The issue that i have is that when the output is not loaded the amplifier is outputting a nice and clean signal and within spec. Once i add a resistive load at the output , even a 100omhs resistors the amp loses stability and starts oscilating.
Those resistors looks like wire wound and therefore have additional small inductance, it looks like the resistors for the output stage are of the same dreaded types, use only film resistors here, which is even more important when having a CFP output stage.
This plus adding an RC zobel before the RL would be some of the first things to do.