Walt Jung’s Op Amp Audio series [ see the Walt Jung Archive:
http://www.elecdesign.com/Articles/Index.cfm ] doesn’t make claims about inner loop feedback with respect to loop oscillation stability, Walt was writing for engineers in a technical magazine and must have assumed the issues would be clear to his audience. It is only in these forums that I have seen the claim that (high values of) the inner feedback loop gain (~=R6/R5) improve stability, in fact the inner feedback resistor R6 value is not very important for stability, its parasitic capacitance is much more important. R5 should be set to match the impedance of the feedback network as seen at the minus input to the source&attenuator impedance at the plus input.
I believe that stability against loop oscillation is not usefully improved by the Jung multiloop inner feedback loop connection for the commonly recommended resistive feedback values. The stability (phase margin) can never be better than that of the input op amp running at the same overall closed loop gain (with the load isolation of the output buffer)
Start with a voltage mode op amp driving a unity gain buffer stage and consider the inner and outer feedback separately:
Either all the feedback is inner loop feedback from the op amp output setting our desired closed loop gain and we simply have a well understood positive gain op amp circuit followed by a open loop unity gain buffer, or we can place all of the feedback in the outer loop, ie from the buffer’s output and now have added any phase shift from the buffer (and op amp out/buffer input series RC) to the op amp phase shift, the added phase shift around the loop will reduce stability.
The two cases above are the limiting cases and since the multiloop feedback network linearly combines op amp and buffer outputs we can say that the general multiloop amplifier response is the linear combination of the two cases above for the purposes for which linear system analysis and the superposition principle applies such as frequency response and loop oscillation stability (phase and gain margin calculation)
This shows that more inner loop feedback does reduce the buffer phase shift added to the loop, but only to the extent that the inner feedback loop controls the overall closed loop gain, in the PPA circuit (typical of Jung multiloop values) the inner/outer loop resistive divider values only reduces the added buffer phase shift contribution by ~4%. Unless the inner loop feedback is so high (= low inner loop gain) that the (low) inner loop gain strongly affects (reduces) the closed loop gain, the inner loop doesn’t usefully effect or improve loop stability, which can never be better than the loop stability of the input op amp operating at the same overall closed loop gain setting.
Capacitive effects are larger than inner loop resistance at the closed loop corner frequency in most Jung multiloop circuits on these forums. The input capacitance of the op amp (large for fet op amps, common&diff mode input capacitance 7&8 pF and 15&8pF for OPA637, AD8610) must be considered with layout and component parasitics to determine the actual feedback conditions above a few MHz, in the case of the OPA637 in a gain +2 circuit the closed loop corner frequency is ~30MHz and the feedback conditions are dominated by real world circuit capacitance that didn’t make it on to Tangent’s schematic – Z(7 pF)@30 MHz ~ 750 ohms. The input capacitance issues are discussed on nearly all higher speed op amp data sheets and app notes, the OPA637 data sheet fig 5 pulse response shows a 4 pF lead cap to match capacitive to resistive effects at a gain of +5. This suggests that the counter examples to my analysis offered by Tangent of METAs using OPA637s at a gain of +2 are in fact noise gain compensated by these capacitive effects to a higher and more stable closed loop gain.
I believe that virtually all Jung multiloops on these forums with unity gain output buffers will be equally stable with or without the inner feedback resistor R6, some may require a 1-2 pF cap in the inner loop feedback resistor position to provide capacitive feedback equivalent to the missing resistor’s parasitic capacitance. The measured distortion will of course be less without the inner loop resistive feedback as well.