BLS doodle

[mirlo]

What do you guys think of this?

The idea is that the current sources force the Vgs's of M0/M1 to be constant, so that the current in Rgm is Vin/Rgm. Since this current is the difference of the drain currents of M2 and M5, given sufficent matching between ( M2 and M3 ) and (M4 and M5 ) their drain currents differ by the same amount, so the difference between the outputs is linearly dependent on the input. Gain is nominally about 2Ro/Rgm.

Again, as with my last doodling (linearized source follower) the feedback, biasing, stability, etc are all potential problms.

Also, I evidently have a one track mind.

I suppose there is not much benefit from using FETs everywhere; it is probably better to replace M2 through M5 with degenerated BJTs.

Same for the cascodes; they could also be NPNs.

Another idea would be to use multiples of M3 and M4 to get current gain before going to the output.

Finally, one could add some frequency shaping for use as a filter / equalizer / phono stage by doing interesting stuff with the Rgm impedance.

On the other hand, the signal path is probably too complicated for it to be noise-free enough to make a phono stage

[Nelson Pass]

I don't know if it's good, but it certainly is imaginative.

[AudioFreak]

I think you may need to add some loss into the cross coupling to avoid excessive phase shift.