ee484,

  Quoting from "Striving for Small-Signal Stability", M. Tian, et al., Circuits
and Devices Magazine, Jan. 2001, page 31-41

    The return-loop model used in Middlebrook’s approach implicitly
    assumes that signals flow through the feedback loop unilaterally.
    This is a reasonable assumption for most low-frequency
    applications.

The issue is that at high frequencies the loading and the reverse return
ratio may effect the results. If you look at section 3.4.1.4, Feedback
Parameters of Real Circuits, in Ken's book, it discusses using four parameter
characterization of blocks[z-parameter, g-parameter, ...] to isolate the
contribution to the loop gain from the op-amp from the contribution to the
loop gain due to loading, ... This approach correctly accounts for high frequency
effects and should produce results consistent with the results from the approach
implemented in the Spectre's stability analysis.

                                                                   Best Regards,

                                                                      Sheldon

ee484,

  First just to be clear, the loading effect refers to parasitic capacitances "loading"
the feedback network and modifying the return ratio as a function of frequency.
The classic example is the board capacitance at the inverting input of the op-amp
and the gain setting resistor in the feedback network. At high frequencies the stray
capacitance shorts the gain setting resistor, creating a zero in the return ratio at high
frequencies. Don't know if it is still in print but the Analog Devices High Speed Seminar
used to have a detailed discussion of this issue. If you want to eliminate this effect
then use a the ideal component method of setting gain described in Ken's book.
For example, by replacing resistor feedback with a voltage-controlled, voltage-source
to set the return ratio creates a unilateral feedback path.

So if you use an ideal component to feedback the output, then you can create a
unitlateral return ratio. In this case, the Middlebrook results should correlate
better with Spectre's stability analysis.

                                                                  Best Case,

                                                                      Art Schaldenbrand