Está en la página 1de 7

Loop gain measurements

Sergio Franco - September 13, 2014

There is no doubt that the loop gain T(jf) plays a central role in negative feedback, so it is important
to have this parameter at hand during the phases of a design process. If the circuit of interest is
available in textbook form, with the error amplifier modeled by a dependent source of gain av in the
case of an op amp, or of gain gm in the case of a transistor, then we find T(jf) as the return ratio of
said source [1]. But what if the error amplifier is shown at the transistor level (as when we wish to
simulate it via SPICE), or is available in its physical form (as when we wish to perform bench
measurements)? We no longer have access to any internal dependent-source model, so we need
some other means for finding T(jf).

Among the various methods reported in the literature, [2], [3], [4], I will discuss the elegant method
pioneered by R. D. Middlebrook in 1975, which is suited both to SPICE simulation and to testing in
the lab. As depicted in Fig. 1, this method requires breaking the loop to inject two stimuli, a series
voltage vt, and a shunt current it. The signal input(s) must be set to zero to put the circuit in its
dormant state, and each stimulus must have zero dc component to avoid perturbing the dc
conditions of the circuit, and must be kept suitably small to guarantee small-signal operation. The
test voltage vt of Fig. 1a causes a voltage vf to propagate in the forward direction, to which the
system responds with the return voltage vr. Likewise, the test current it of Fig. 1b results in a
forward current if, in turn eliciting the return current ir. If we let

then the loop gain T is obtained by exploiting the condition [2]

Solving for T gives

Alternatively, using Eq. (1), we have


Let us apply Middlebrook’s method to the circuit of Fig. 2, popularly known as the feedback bias
type. There are three points where we can break the loop, namely, X, Y, and Z; however, T(jf) is an
intrinsic loop characteristic that must be independent of the particular point. To gain better insight,
let us investigate each point separately.

The results, shown in Figs. 3-5, confirm that even though the makeup of Tv and Ti may vary
significantly with the point of signal injection, T(jf) remains always the same.

Equation (2) indicates that the terms (1 + Tv) and (1 + Ti) combine like parallel resistances, so if one
of the terms happens to be much smaller than the other, the smaller will dominate. Sure enough, for
sufficiently high frequencies we have T →Ti in Fig. 3, and T →Tv in Fig. 4. On the other hand, in
Fig. 5 we have T →Ti at low frequencies and T →Tv at high frequencies. We wonder whether it is
possible to break the loop at a point where only one signal injection will suffice. To this end,
consider again Fig. 1, where Zf and Zr are the impedances seen looking in the forward and in the
return directions relative to the point of signal injection. If we can identify a point where Zf >> Zr
over the frequency range of interest, then we expect if << ir and vr << vf, implying Ti >> Tv, or (1 +
Tv) << (1 + Ti). Consequently, only a voltage injection will suffice in this case. The opposite occurs
if Zf >> Zr, when only a current injection suffices (however, in practice, voltage injection is
preferable because we can conveniently generate vt by means of a transformer). The choice of the
injection point is facilitated by the condition [2]
As an example, consider the circuit of Fig. 6, whose loop can be broken at four different points, X1
through X4.
A convenient point is X2 because it sees a forward impedance Zf of at least 100 kΩ, whichis much
higher than the return impedance Zr, consisting of the op amp’s internal resistance ro in parallel with
CL. Fig. 7 confirms that T ≅ Tv over the frequency range of interest, so only voltage injection suffices
(for convenience, also current injection is shown to confirm that we indeed have Ti >> Tv).

The Phase Margin φm

Fig. 8 shows the frequency and transient responses of the circuit of Fig. 6.
As discussed previously [5], the presence of Cn and CL tends to cause a significant amount of peaking
in the frequency domain and ringing in the time domain. Both peaking and ringing are in turn
determined indirectly by the loop gain T(jf) via the phase margin, defined as

where Fx is the phase of T(jfx), with fx being the frequency at which |T| drops to 0 dB, or unity. Since
T = aß, we can visualize fx also as the frequency at which the |a| and |ß| curves intersect each other
[6], this being the reason why fx is called the crossover frequency. Fig. 9 depicts the relationship
between peaking/ringing and the phase margin for the case of a second-order, all-pole system (if the
system at hand is of a higher order, the graphs of Fig. 9 can still be used for initial estimates).
Peaking starts as φm drops below about 65°, and ringing starts as φm drops below about 75°. The
smaller φm, the larger the amount of peaking and ringing. As φm reaches 0°, the circuit becomes
oscillatory. Evidently it pays to ensure that φm exceeds 0° by a sufficient margin (hence the name).
Popular margin specifications are φm = 90°, 75°, 60°, or in some cases even 45°.

Examination of the traces of Figs. 3-5 reveals that |T| intersects the 0-dB line at fx ≅ 6.2 MHz, where
the phase plot of T (not shown) reveals that φx ≅ –94°. Consequently, φm ≅ 180 – 94 = 86°, indicating
a very stable circuit. By contrast, the op amp circuit of Fig. 6 does not have an adequate phase
margin. Using the voltage injection method alone, we obtain the magnitude and phase plots of Fig.
10 (#1 traces), which reveal fx ≅ 338 kHz, where φx ≅ –164°. Consequently, φm ≅ 180 – 164 = 16°, a
poor phase margin indicative of a circuit in need of compensation, as discussed in Ref. [5]. For
comparison, also the case Cn = CL = 0 is shown (#2 traces), for which we have fx ≅ 459 kHz, where φx
≅ –105°. Consequently, φm ≅ 180 – 105 = 75°, indicating the absence of both peaking and ringing, as
confirmed by the #2 traces of Fig. 8. Evidently, Cn and CL conspire to lower φm from 75° to 16°!
The interested reader can rerun the voltage-injection test first with only Cn in place, and then with
only CL in place, to investigate how each erodes φm. Do the individual phase-margin erosions add up
linearly? Can you justify?

References

[1] Two-Port vs. Return-Ratio Analysis

[2] R. D. Middlebrook, “Measurement of Loop Gain in Feedback Systems,” Int. J. Electronics,


Vol. 38, no. 4, pp. 485-512, 1975.

[3] S. Rosenstark, Feedback Amplifiers Principles, MacMillan, New York, 1986.

[4] M. Tian, V. Visvanathan, J. Hantgan, and K. Kundert, "Striving for Small-Signal Stability",
IEEE Circuits and Devices Magazine, Vol. 17, No. 1, January 2001, pp. 31-41.

[5] Feedback and Impedances

[6] Inverting vs. Noninverting

También podría gustarte