IRS2092 customizable

[DarkOne]

Hi All,
When I realized that switching amp can be better than most linear amps, I started thinking about building my own Class-D amp. It's gonna be my first Class-D project after few Class-A / AB amps projects that I've designed. I still don't know whether pre- or post-filter feedback design is better, so in this design I can try both (I hope so).
As it's my first class-d amp, I decided to use reference design from IRS2092 datasheet and add opportunity of selecting post-filter feedback by jumpers. Maybe it's possible to use both feedback paths, but now I'm only in the beginning.

Could you please comment it as it's my first switching amp project.
Every opinion is welcome.

Please see attached schematic and PCB layout.

[DarkOne]

And the PCB layout

[Ouroboros]

The IRS2092 reference design works well (I have the dual-channel evaluation board from IR in front of me as I type), but I doubt if you will be able to make a stable post-filter NFB design using the 2092. The internal op-amp and comparator have internal connections that assume a self-oscillating sigma-delta design, with pre-filter NFB. When I first saw the internals of the 2092 I wondered if a UCD-style amp could be made using it, but this would require access to certain internal nodes that are not brought out from the chip. (I'd love some-one to prove me wrong about the inability to make a UCD topology using this chip though!)

[DarkOne]

Ouroboros,
I thought that COMP pin is output of comparator. It just depends on design whether it's sigma-delta or ucd like amp. The comparator has the same function as ucd input comparator, I believe

[mag]

Hi Darkone,
I have built an amplifier with IRS2092
(Class-D Amp with IRS2092)
It works very well and it sounds very good.
Now I am planning a newer version with lower power and your idea of post-filter feedback is very interesting.

Your pcb layout looks good, BTW which voltage rails will you use on that amp?

[MOER]

Post filter feedback is impossible with this chip. The inegrator is not accessible.

The irs20955 is possible as the integrator is outboard

Stephen Mantz
Zed Audio Corp.

[Ouroboros]

Originally posted by DarkOne
Ouroboros,
I thought that COMP pin is output of comparator. It just depends on design whether it's sigma-delta or ucd like amp. The comparator has the same function as ucd input comparator, I believe

No!, the comp pin is the output of the transconductance amplifier used as the input stage in the chip. The output of this amp feeds the internal comparator. If the output of the transconductance amplifier and the input of the comparator were on separate pins, then a UCD amplifer might have been possible, but not as the chip stands at the moment (unfortunately).

[Pafi]

I don't really understand you, guys! Why would be impossible to make a simple fast amplifier with OTA?

[Ouroboros]

For the normal topology UCD amplifier the post-filter NFB has to feed into the input of the high-speed comparator stage (not the op-amp) and be summed with the audio signal from the input stage. It is possible to split the feedback route in a UCD by having part going directly into the comparator, and part going via the input op-amp to provide an active pole which increases the loop-gain at low frequencies, but you cannot have a single NFB connection into the input of the op-amp. This is different from a sigma-delta amplifier where the input op-amp is essentially an integrator.
The reason that I said earlier that it would be possible if the output of the OTA was brought out to a pin, and the input to the comparator was also on a separate pin, is that we could put a 1k (say) resistor between the output of the OTA and the input of the comparator and feed the NFB into the comparator input.

When I first saw the internals of this chip, and realised that the input op-amp was an OTA, having a high-impedance current output, I thought that it might be possible to connect the post-filter NFB to the COMP pin, but unfortunately, once you apply the required local NFB around the OTA, it decreases its output impedance which messes up the idea. (At least that is what SPICE tells me, using a simple model of the 2092).

[Pafi]

What's the problem with this?