Yea, you have to take a close look at exactly how the internal offset null circuit is configured. For the LT1028, the offset null pins have a resistance of about 130 ohms to the positive supply, so anything done to drive them looks like a current in or out, and that works just fine. When I made my measurements, I got a scale factor of microamps in or out of the offset null pin for a given change in input offset voltage. Then I designed the external network to drive that current in or out, but it comes down to just a resistor or resistor divider.
I understood your concept. Retaking the old non-inverting schematics (attachment, still the one with wrong values): in this case the only thing that doesn't convince me is the following: let's say that I change R4, R5, R6 and increase their value. Respecting some proportion with AppNote21 (without any calculations for now), I guess that I should go with R4,R6~15/20 kOhm at least, while for R5 I guess I would go for 100's of kOhm, if not approaching MOhm (to respect some proportion). I imagine that this is about the expected offset range that I want to correct. Now, even before doing calculations my doubt is: if such high values for the R5 value are involved, doesn't this somehow interfere with one of the two opamps (either the OP07 output, or the NE5532 input)?
I ask this because, in an almost stupid way, I know these old opamps don't work very well when they have such high resistors around, but I don't know if that is only when talking about the "commonly used pins" (+,-,out). Usually one doesn't touch the offset correction pins with "strange" configurations like these.
It's also true that when comparing the pure LT1028 datasheet and the correction offered by AN21 we fall in a similar situation (we want ~100's Ohms, but we are also putting 30k there), hence it shouldn't matter that much.
EDIT: For now, in all of this I'm quite neglecting the fact that, noisewise, I'm somehow introducing a 100 kOhm/MOhm resistor, but maybe it can be reused somewhere else.
Sorry if my questions here are quite convoluted, I see that there are also a lot of side-things in all of this.
Also, I think this is important because if I can go higher with R5 values without major drawbacks I could reconsider entirely this configuration.
Circuit simulators can be very helpful in such situations.
For instance, I found that with your value of R1 there was additional noise at all frequencies, which I suppose makes sense because U1 voltage noise present at its inverting input mixes with the input signal through R1. Simply increasing R1 solves this problem. At low frequencies I found it helpful to increase C1, but it appears to be true that one can decrease R3 instead and the effect is almost the same (not shown, download LTspice and try it yourself ).
BTW, regarding this simulation:
LT1001 is Linear's OP07, very similar noise.
U2 is an ideal opamp, so noiseless.
The blue plot shows everything minus resistor noise, because simply plotting V(U1) doesn't work
This is total output noise, so divide by 100 to get input-referred.
At high frequencies the main feedback divider is the dominant factor, about on par with NE5532.
Don't exactely know what to say but thank you. I didn't actually noticed that LTSpice had a noise analysis, but mind that I **barely** learned how to do an AC analysis, so..