I~T. J. Er~ECl'RONIC8, 1975, VOL. 38, ~O.

+, +85-512

Measurement of loop gain in feedback systems+

R, D. MIDDLEBROOKt

In the design of a· feedbeok system it is desirable to make experimental mousuromonts of t.he loop gain as a function of frequency to ensure that, the physical. system operates as analytically predicted or, if not, to supply information upon which a design correct.ion can be based. In high loop-gain systems it is desirable that the loop-gnin measurement he made without opening the loop. This paper discusses practical methods of measuring and in tcrprcting the results for loop gain of the olosod-Ioop system by a voltage injection or a currerrt-inject.ion technique; extension to the case in which the measurement can be made even though the system is unstable; and extension to t.he case in wh ich neither t.he volt.age nor current- injection technique alone is adequate, hut in which a combination of both permits the true loop gain to be derived. These techniques han, been found useful not only in linear feedback systems but also ill dcsceihing-Iunction analysis of switching-mode converters and regulators.

I. Introduction

In approaching a system design problem, the usual procedure is to make a preliminary paper design and analysis, then to build a breadboard and make experimental tests of the performance. If discrepancies are found between the predicted and observed properties of the system, the analysis model is corrected in a manner suggested by the nature of the observed discrepancies, and the modified predicted performance again compared with the experimental results. Several iterations of the analysis- .. measurel1lent- .. correction sequence may be required before final adoption of the system design.

When the system being designed incorporates a negative feedback loop, one of the important performance parameters to he predicted analytically and experimentally verified is the loop gain. This paper is concerned with experimental methods of making such measurements with emphasis on practical problems of accuracy and proper interpretation of the results. These techniques have been found useful not only in linear feedback systems, but also in the describing-function analysis of many types of switching-mode converters and regulators.

The method of measuring loop gain l' by injection of either a test voltage or a test current into the loop is first reviewed. The important feature of this method is that the loop remains closed, so that operating points arc not disturbed. Usc of a narrow-band voltmeter permits loop-gain measurements to be made in high-gain systems, and also in systems in which there is a large amount of noise as, for example, in a switching-mode regulator. A technique for determination of phase as well as magnitude of the loop gain, with usc of

Received 23 September 1974,

t This work is an extension of material presented at the ESTEC Spacecraft Power Conditioning Electronics Seminar, ESRIK Centre, Frascati. Italy, 20-22 May 1974.

t Profcsaor of Electrical Engineering, California Institute of Technology, Pasadena, California.

Copyright©2001;AII Rights Reseved.

48(:i

H. D. Middlebrook

magnitude measurements only, is discussed and elimination of possible illconditioning of the phase formula by means of suitably scaled magnitude measurernents is presented.

Several extensions of this basic method of loop-gain measurement. are introduced, and it. is first shown how an unst.able loop gain can be measured directly. An example is given in which the phase margin is measured to be ._- 6°.

For the voltage injection form of the loop-gain measurement. method to give the correct. result, it is necessary that. the injection be performed at a point in the loop at which the impedance Z2 looking' backward' from the injection point is sufficiently smaller than the impedance Zl looking' forward' from the injection point. The opposite condition, Z2 ~ Z1, is necessary for t.he current injection technique to give H correct result. Since in a practical system it may not be possible to find an injection point that satisfies either of these extreme conditions, at least over the entire frequency range of interest, it is desirable to extend the loop-gain-measurement method to be applicable at an injection point where the impedance ratio Z2/Z1 is arbitrary.

It is shown that the true loop gain T can be derived from measurements of ratios 'If" and T, obtained respectively by successive voltage and current injection at it point of arbitrary impedance ratio. The method is illustrated by a practical example, which is also used to demonstrate an inherent accuracy defect in the method, namely, that poor accuracy in the derived T is obtained at frequcnoies beyond loop-gain crossover when I T I < 1.

The inherent accuracy defect. is eliminated in an improved method of loop-gain measurement by simultaneous voltage and current injection at a point of arbitrary impedance ratio. 1 n this' null double-injection ' method, the true loop gain T is derived from measurements of ratios 'I',,'" and T;" obtained by adjustment of the relative magnitude and phase of the injected voltage and current to null out the current looking backward from the doubleinjection point, for '1\,", and the voltage looking backward, for Tin. The null double-injection method is illustrated by RIl example in which accurate results are obtained for a loop gain T that has an additional pole beyond the crossover frequency, an example which is particularly poorly conditioned for application of the successive voltage and current-injection method,

Finally, reconsideration is given to t.hc method of loop-gain measurement by simple voltage 01' current injection alone, and the conditions to be satisfied by the im pedance ratio Z2/ Zl at the signal injection point. are discussed in further detail.

2. Measurement of loop gain in a closed loop by voltage injection or by current injection

I n principle, loop gain T call be measured by opening <L feedback loop at an appropriate point, application of a test signal in the' forward." direction at the opened point, and measurement of t.he resulting loop-transmitted signal that appears looking' backward' at. the opened point.

!Vf are specifically, one such 'a.ppropriate point· is at the output of a dependent. voltage generator within the feedback loop, as indicated in Fig. [(0). The closed loop is defined by a proportionality between the voltage

....... - .. -~~---------- -- -. - --~.copyr.ight.@. 200 1- ;. All Rig htsReseved. ,·c··· _ .

jll casurement. of loop gain in feedback systern.8

487

developed by the dependent voltage generator and a signal (either voltage or current) at the impedance Z. (Notation : independent generators are represented by circles, dependent generators by squares.] If the loop is opened and a test voltage Vx applied at point A ,., and the resulting voltage Vy of the dependent generator measured, then, by definition, the loop gain is given by T = vl/!v", Since the loop is open, a voltage Vz = Vx + V11 appears across the break, as shown in Fig. 1 ((1).

feedback

(0)

feedback

(b)

Figure 1. Meaeuremcnt of loop gain T by opening the loop, (a) hy voltage ratio, (b) by current ratio.

Another "appropriate point' is at the output of a dependent current generator, as indicated in Fig. 1 (b). In this case, ' opening the loop' implies short-circuiting the output of the dependent current generator, and the loop gain, by definition, is T = illfix, where iy is the dependent generator current resulting from the test current ix applied at point A i: Since the loop is open, a current iz=ix+iy flows in the common connection as shown in Fig. I (b).

In many practical situations it is inconvenient to open the feedback loop in order to make such loop-gain measurements because, particularly in highgain systems, it is difficult to maintain proper d.c. operating conditions, and the system may saturate on noise. However, the voltage conditions of .Fig. I (a) can be maintained without opening the loop by the simple expedient of making vz, instead of v"" the independent test signal: as shown in Fig. 2 (a), the original closed-loop circuit topology is retained, and although Vx and vlI are now both dependent quantities, the loop gain is still T = vy!vx. Similarly, in the dual method shown in Fig. 2 (b), substitution of iz as the independent test signal permits the original circuit topology to be retained and the loop

Copyright©2001. All Rights Reseved.

48S

R. lJ.Middlebrook

gain is still 71 = i,)i:c- Thus, in either method, injection of no test signal enables a direct measurement of loop gain to be made without opening the loop.

In a convenient practical implementation of the current-injection method, a test oscillator voltage output is converted to the current iz by a high resistance and is injected via a bloeking capacitor, and the resulting currents i.r

Vz
Vy
T~~
VX
,... <,
I \
I j
\ + I
-, /
<, ;"
......... -4-- __ ------- ---
(a)
iy i y
T ~..,-
I~
....
/ \
I \
I J
\ /
<, ..._ /'
..-
-_ ..._---____.-__'~
(b) Figure 2. Mccsurerncnt of T by injection into the closed loop, (a) by voltage ratio,

(b) by current rat.in. .

and i!1 arc measured by a clip-on current probe attached to a current-tovoltage converter amplifier connected to a voltmeter input. For the voltage. injection method, a clip-on current probe is connected to thc oscillator output and used 'backwards' as a one-turn secondary transformer to inject the voltage Vz, and the resulting voltages v~, and Vy are read directly by the voltmeter. Although other methods are possible (Electron. Design, 196,5), the usc of the current probe as a signal-injection transformer is a very convenient way of obtaining the necessary 'floating' voltage vz. In the voltage-injection method, larger Injection voltages can be obtained by wrapping the lead more than once around the clip-on probe, thus increasing the Bum bel' of' Recondary' turns.

Since the voltage-injection and current-injection methods are duals of each other, it is convenient. to refer to the three signals as UT, nil and 110, where nis to he interpreted as cither v or i as appropriate. Thus, for either method,

(l)

'1' = UII 1J",

(i)

M easuremeni of loop gain in feedback systems

489

The signals H.r' 1(1/ and U2 are phasors, and so use of an ordinary voltmeter gives directly only the magnitude I T I of the loop gain. Often, it is known that the system being measured is minimum-phase, and so a measurement of loop-gain magnitude is sufficient. However, if phase measurements are desired, a phase-reading voltmeter such as the Princeton Applied Research Two-Phase/Vector Lock-In Amplifier Model 129 may be used. This instrument has the additional advantage that it is phase-locked to tho test oscillator signal, and so functions as a narrow-band tracking voltmeter capable of reading very small signals that would otherwise be buried in noise. Indeed, the narrow-band property may be essential in order to make even the magnitude measurements, because in a system with high loop gain, nJl~u.r and the allowable magnitude of uiJ is limited by system overloading; consequently, u; is very small and may only be extracted from noise by use of a narrow-band voltmeter. Also, the narrow-band property is especially advantageous in making loop-gain measurements all switching regulators, since otherwise the switching noise 'would be likely to swamp out at Icast the smaller of the loop-gain test signals.

load

1 I r

+

r·-----, I i, Ai iy I

L J

current injection

L J

voltooe injection

Figure :~. Appropriate poi fl."; for voltage and for currunt injection to measure loop gain of a sirn plc switching regulator.

An example of the application of the signal-injection technique for measuremcnt of the loop gain of a simple switching regulator is shown in Fig. 3. As indicated, suitable injection points A ,. and A i for voltage and current injection respectively can be found. The location A" is suitable for voltage injection because the impedance Zl looking forward around the loop is much greater than the impedance Z2 looking backward, so that Vy approximates the voltage of an ideal voltage generator, as required for the model of Fig. 2 (a) to be valid. Conversely, the location Ai is suitable for current injection because here the opposite condition Z2~Zl obtains, so that iy approximates the

Copyright ©2001. All Rights Reseved.

4!J()

R. 1).M iddlebrool:

current of an ideal current generatC)f (namely, the collector of a. t.rn.nsist.or-), l1fl required for the model of Fig. 2 (b) to (w valid.

Although it provides directly only maguitudc information, a wave analyser such I1S one of the Hewlet.t-Packard :302A series is a very convenient instrumerit for lise in the signal-injection method of loop-gain measurement (Spohn 11)63, Hewlett-Packard Application. Note, 1965r In the ' BFO' mode the instrument operates as an adjustable-Irequenov oscillator with an automatically tracking narrow-band voltmeter, and thus provides both the test. signal and the required narrow -band voltmeter. Moreover, phaseinformation can also be obtained by a. simple indirect method. At [my frequency, one measures HOt. onlv the magnitudes lu ... 1 and InNI to give

but H.lSO the magnitude I Hz I of thc third phasor, so that by trigonometric solution of the phasor triangle one obtains

(4 a)

or, equivalently,

LTc~ + 2 cos:'! J(L!~12-(IUxl--11~!l1)2)

- 41uxllnJlI

Although the sign givcn by eqn. (4) is ambiguous, propel' choice is usually obvious from the qualitative nature of the magnit.ude response and known properties of tho loop. The above forms apply when LT is in the range 0° < ± I. T < 1800; for the range 1 HO" < ± 1_ T < 3600, t.ho appropriate form is L 7' = ± (3fiOO- 0), where e is the first or second quadrant angle given by eqn. (4),

In the current-injection method, measurement. of the third phasor 7:: is straightforward, but in the voltage-injection method measurement of v" directly is inconvenient because of the requirement to float the voltmeter, Instead (in either method) thc phasor sum I"l/.:I = iUr+nyl can be react from the addition of the signals u , and1l.!f passing through unity (or equal] gain amplifiers, as indicated in Fig. 4 (a).

Equation (4) gives accurate results for the loop-gain phase angle in the important frequency range in the neighbourhood of loop-gain crossover, when ITI~l. However, it is ill-conditioued when either I'T!p] or 1'l'I<;,1. since then either lulIl ~ In,,1 and 111:1 ~ Inlll, or Iu..rl ~ 111'11 and 111,1:::::: 111.ri· This condition is immediately obvious in graphiml1 terms from the phasor diagram illustrated in Fig. 4 (a), drawn for I T I > 1: clearly, smal I errors in the mensurornonf of the magnitude lUll I or luz I can lead to large errors in IT.

The ill-conditioning defect in the phase measurement of Fig. 4 (0) (;,1Il easily be overcomel by use of Ilneq ua I gains for H" and ny, as shown in Fig.

(4 b)

-----------~-

t Usef'ul suggest-ions relating to t.his technique were made by Dr. Yuan Yu of TRW Systems. Inc .. ami D. ,J. Packard of the Califrn-nia .L nstitut« of Technology and Hughes Aircraft Co.

Copyright'@-2001, All Rights Reseved:

Measurement of loop gain in feedback systems

491

U Y

U. (0)

Au y y

u zz

A U ••

[ I uzzi ] L2IA•U.1 J

lAy Uyl : I Ax uxl

when

(b)

Figure 4. Determination of T, (a) by measurement of the three magnitudes 111",1, Iuyl, IUzl; (h) by measurement of the scaled magnitudes I A",u", 1 = I Alluy 1 and IUzzl.

4 (b). '1'0 the extent that the gains A", and All have zero (or the same) phase, the angle between A)/1);I/ and Axu;r. determined from the three magnitudes IAxu"l, IAI/uyl and lu/I=IAxux+AlIulIl is the same as the angle between uy and UX' namely LT; thus eqn. (4 b), for example, becomes

LT= + 2 cos ? J(luz'12_(IAxUxl-IA!lu!/1)2)

- 4lAxu",IIAy1Jyl

and can be made well-conditioned at all frequencies by adjustment of the gains Ax and Al1 so that all three measured magnitudes are of comparable sir-e. In fact, an additional condition simplifies the procedure considerably: oue merely adjusts Ax and Ay so that the magnitudes I Axu" I and I Al/ull I are equal. Let the corresponding magnitude of the summed signal be IUzzl = Iuz' I when I A xu", I = IAyutll; then eqn. (5) reduces to

LT = + 2 eos-1 [ I uz" I J

- 21Axuxl

(0)

The practical procedure is therefore to adjust A.r and All until equal voltmeter readings I A,11-.r I and ! A !iuill are obtained, and then to read the corresponding phasor sum IUzzl = IAxux+Alluyl for usc in cqn. (6). It is not necessary to know either Ax or All' As shown ill Fig. 4 (b), this procedure scales the two largest phasor magnitudes until they are comparable with the smallest phasor magnitudes, without changing the required loop-gain angle, LT. The required loop-gain magnitude I T I is of course obtained by setting the two

Copyright©2001. All Rights Reseved.

u. n. MirldldmJUI.:

gainH equal, A£=Ay=A, and measuring I A·If.r I and l.hl.!,1 :,;0 that. 11'1~· IAnl/I/IAn.rl = Inyl/lncl·

As already mentioned, the origjna.l ex pression for the phase angle /_ T,

eqn. (4), gives accurate results in the moighbourhood of the loop-gain CI.'OMS· over frequency because 111,;,.1 and 1','Jili a.re already comparable in magnitude without. the necessity of Healing by means of the amplifier gains A.r and "I II" In partioular-, at the crossover frequency, IU.rl = luyl, so the conditions established hy the implementation of Fig. 4 (b) exist already without tho amplifiers. Therefore, the implementation of Fig. 4 (a) may be used directly, and the phase angle at the crossover frequency may be obtained from cqn. (H) with -A.r = I and -/I,," ~~ Hz, or

(7)

I rI manv applicat ions a measurement of the phaRP angle solely at the crossover frequency (to determine the phase lllargin) may be sufficient, so that the amplifier ga.ins A,. aud All arc not. needed at all.

To demonstrate the application of the loop-gain-measurement techniques so far discussed, the circuit of l"ig . .'5 was constructed. The objective was to obtain experimental measurements of the magnitude and phase of the loop Wlin, without opening the loop. A preliminary oxpectat.ion is that the loop gain T contai ns two poleK, The point A I' should be suitable for measurement I))" the voltage-injection method since it satisfies the condition Z2~Zl' where Zu the impedanco looking forward 11)1I1Ul the loo-p, is greater than 20 k, and Z2' the imperlance looking backwards, is the output impedance of the 7tH) opemtional amplifier, which if' on tho order of 100 U .

.-----------~~~--------~

10k

B

43k

Figllre.'i. Fp('dbaek amplifier oirr-uit for dr-monst rat ion of loop-gain moa-urcmont with voltage iujoct.ion o.t point AI"'

The test set-up is shown in Fig, (i. Voltage injection was accomplished through a. clip-on currcuf probe (' one-turn secondary'), operated 'bnckwarda ' from the oscillator output, of a Hewlett=Packard ~{590A Wave Analyser. Voltages v.r and VII were measured with 10:.1 voltage probes at.tached respectively to channels 1 and 2 of a Tektronix Type lA 1 Preamplifier in an oscilloscope. The two channels have separately adjustable gains, corresponding to Ar and A" of Fig. 4 (b), Channel 1 polarity is set at , normal' to read + Axvp and channel 2 polarity is set at ' inverted' so that

Copyright© 200LAII Rights Reseved.

211 easuremeni of loop gain in feedback systems

493

-Vy is converted to read + Ayvi/" The oscilloscope' vertical out' is connected to the analyser voltage input, and the IAI selector switch directs channel I, channel 2, or the sum of channels 1 and 2, to the voltmeter. Thus, the configuration of Fig. 4 is implemented.

10: I voltage probe

r
10;1 voltage v
probe y
_v 1
x
+
0
Type IAI
Preamp
veri.
Ch I Ch2 auf - "4-- 'oJ_______"'+ Z

A + v

v •

Hewlett - Packard 3590A Wave An a Iyser

voltage osc.

in out

Figul'e n. l ust I'U nu-nt at ion .~('( -up fol' voltage inject ion and voltage ratio measurement at point A ,. ill the circuit of Fig. 0.

2°1-- '-----,-----, ----"

10~ dB!

O~I _-

-IO~

-2ol

1

-301

-40l

-5J-, _- ---- _!_ .. _l_

0.1 0.5 I kHz

I

1 l J I

I

---~--,~, ,-'" _"_-"-

;. ~~3.4kHZ

I • a I

f = 1.4 kHz· • '

a • .

ITI m.~~'" ~ ; .

f"ol~

_J_ .. 1 .. 10

_[_, J

50 100

5

Figure 7. Loop-gain magnitude vurxus Irr-quuncv plot obtained hy voltage ratio rneasurementx at point A" in the circuit. of Fig. 5.

With the channel I and channel 2 gains equal (A,r=A,1=A), readings of I A 11". 1 and I AVlIl were taken, leading to data points for the magnitude I TI of the loop gain as shown in Fig. 7. Best-fit straight-line asymptotes are drawn through the data points, from which it is seen that there is a pole at some low off-scale frequency causing an asymptote zero-d B crossover frequency

Copyright © 2001. All Rights Reseved.

49J

R, n. Jfiddlebroo1.·

at /(1 = 1·4 kH v., another pole at I"~ = :3-i1~ k l-lz , and a zero at It, = 12 kHz. 'I'hc ]lJ'cHellcc of the zero in addition to the two expected poles must be accounted for: presumably, it is due to the gain of the 70!-) levelling off as the () 1 IJet' compensation capacitor becomes essentially a short circuit.

If the loop gain T of the circuit of Fig. 5 were assumed to be a rninimurnphase function, the phase response would be implicit in the magnitude response, with a noo lag from each of the two poles and a nOD lead from tho zero, giving a total phase shift at high frequencies asymptotic to - 900• lndependent measurement of the loop-gain phase angle I 'T by the method described in connection with Pig. 4 (b) shows that, on the contrary, the loop gain is not a minimum-phase function. The dat.fL points of !T obtained by usc of cqn. (6) arc shown in Fig. H, and it is seen that ! ']' approaches not - 900, hut . ... ::270D at high frequencies. These data are well fitted by asymptotes (MiD pCI' decade) that correspond to the two expected left-half plane poles and to a Tight-half plane zero at. 12 kH z : consequently, reconsideration of the origin of the zero is required. Examination of the internal circuit of the 709 operational amplifier reveals that the compensation capacitor is in a collectorto-base position, and so the low-frequency phase inversion of the commonemitter amplifier stage is removed when the capacitor becomes an a.c. shortcircuit.

O°,.---------.::r::-_____ ------...._ -'T' -- ---- ,--- -r-- l

. 3:-~.~b =:12kHz asymptote for "

-45'i. asymptotes for left .-:-.-. _ ~ rrqht-holf plone 1

_90o'-I Pla~t:.::tes ~3.4·~~:~Z-i

l· .;--- ....... ~ I

-135".

• •• r--LT measured

l · I

totot asymptote" I

,180" l

-225ol

-2l0oL

0.1

____ .1. L___ ..

0.5 I kl-lz

._l._._ . __L_

5 10

~_ 1

~

~50 100

Figure 8, Loop gain phasr- versus frequency plot obtained for the circuit of Fig. 5.

it iR concluded, therefore, that the loop gain T of the circuit of Fig. ;) can he expressed as

8 1--

1'=---.- w"

8 ( 8 )

- 1 +---

wI) w"

(R)

where

10 = wo/2rr-.' 1·4 k Hz

(U)

(10)

(II )

Copyright @·2001. AIlRights Reseved.

M co surement 0/ loop gain in feedback systems

495

The example has demonstrated the importance of reconciling all features of the measurements with understanding of the physical sources of the effects. The erroneous assumption of a minimum-phase function for T that fitted only the magnitude data points of Fig. 7 would have predicted a greater stability margin than is in fact the case, as follows. Because of the proximity of t, to /1)' the actual crossover frequency at which ITI = 1 is 1·3 kHz, slightly less than t« The true phase angle at the crossover frequency, from eqn. (8), IS

LT11T1=1 = ~[900+tan-l (l·3/3·4) + tan-l (1'3/12)] = ~[90D+2Io+6Dl

( 12)

(13)

glvmg a phase margin of 180o~1l7°=63°. If a minimum-phase function had been assumed for T, the phase contribution from the zero at / b = .12 kHz would have been a lead instead of a lag, and the phase angle at the crossover frequency would have heen

LTIITI~l =-[90o+'21o~6°J

(14)

giving a phase margin of 1800 - 10.')° = 75°, which is 12° greater than the true value of 68°.

3. Measurement of loop gain in unstable loops

] n pursuit of the analysis-·mcasllrement~correction procedure of system design, it sometimes happens that an actual feedback system unintentionally oscillates, so that the loop-gain measurement cannot be made because the oscillation builds up until limited by non-linearity. One way· of restoring the desired design iteration sequence is artificially to reduce the loop gain until oscillation ceases, by introduction of an impedance divider at some convenient point in the loop. The loop-gain measurement is then made, and the true original loop gain deduced by. taking account of the correction introduced by the impedance divider. This method is inconvenient, not only because of its indirectness, but also because it may he difficult to account properly for the effect of the impedance divider, since this must be done analytically and its correctness depends upon the system model being correct in the neighbourhood of the impedance divider.

A more satisfactory solution for temporarily extinguishing oscillation is to introduce the impedance divider at the same point where signal injection is made for measurement of loop gain. For the voltage-injection method, an impedance Z,. is introduced in series with the injected voltage, as shown in Fig. 9 (a). The actual loop gain has thereby been reduced by the factor Z/(Z" + Z), but it is clear thilt vll/vx is still the original, or true, loop gain T. Similarly, for the current-injection method, an impedance Zi is introduced in parallel with the injected current as shown in Fig. 9 (b), so that the actual loop gain is reduced by the ratio Z,./(Z.i + Z) but iy/i.c is still the true loop gain T. Therefore, if the original feedback system oscillates, Z,. can be made large enough, or Z; small enough, to eliminate the oscillation, yet direct.

Copyright©2001.AII Rights Reseved.

4!)(i

R. D. M£ddlruroo/;

measurement. of the gain of the origillal (unstable) loop ('an be made Hn!1 knowledge of the i In pC'da.])('p divide-r rat io is not ncerled.

- ...... --_. Vz________.,._ t-

-, \

)

/

./

__ /

---- . ...---_ .....

(b)

Figur,; !). Measurement of T by inject ion into the closed loop from a. non-ideal source, (a) by voltag« ratio, (0) by current ratio.

As an example of this technique, a feedback circuit was constructed as shown in Fig. 10. Since tho output impedance of the 7m) operational amplifier is very low compared to 10 k , point A,. is suitable for measurement of loop gain by the voltage-injection method. The system was purposely made to oscillate, at. about (j·7 kH s., by dORing switches 81 and 82, It was found that insertion of a series resistance great.er than 8·2 k at point. A,. eliminated the oscillation, thus permitting the loop gain to be measured by voltage injection as in Fig. 9 (0) with Z,. greater than 8·2 k. Data points for the magnitude I TI of the loop gain, with Z" arbit rnrily chosen as 10k. are shown in Fig. 11.

-~-~-~V'v\r

10k

B

10k

0.02 r'-~- 22k T

~_.__J2

Ik

Figure 10, Feedback amplifier circuit for dcmonst.rat.ion of unstable loop-gain measurement h.Y nJltage injection at A,. (switches 81 and 82 closed), and for demonstration of loop-gain rnr-asuromcut by successive volt.age> and current injection at R (switclw;;; open) .

..... Copyright © 2001; All R~ghts Reseved-.

111 easurement a/loop gain in. feedback systems

497

Loop-gain crossover occurs at 8·0 kHz, where IV:r1 = Ivyl, and measurement of the third phasor hi at this frequency gave Ivzlvyl =0·11. The corresponding angle given by eqn. (7) is ± 174°. Since the original system is known to oscillate, the actual phase angle is a lag exceeding 1800, so that the proper solution is LTIITI ~ 1= - (360° -174°) = - 1860 (see discussion following egn. (4)). As a check, the phase angle can be calculated from the measured I T I

10 o f----- _~ __ c~

-10

-20

-30

-40

-50

-60

- 70 L- _L __ _L_ • _ ____j ••• ~_.L__ __

50 100

18dE}"8va

40 30 20

12dE¥8vo

8kHz

I I I I

22kHz

5

10kHz

Figure J T. Magnitudn versus frequency plot of the unstable loop gain of the circuit of Fig. 10 with switches 81 and 82 closed.

versus frequency data of Fig. J 1 on the assumption that the loop gain is a minimum-phase function.'] There is a pole at some low off-scale frequency, one at 2·0 kHz, and another at 22 k Hz : hence, the phase angle of T at the crossover frequency of 8·0 kHz is given by

LTIITI~1 = -[900+tan-1 (8j2)+tan--l (8/22)1

(15)

in excellent agreement with the value directly measured. Hence, the phase margin is ~ 6°.

It is therefore demonstrated that loop gain in an unstable system may be directly measured. As a corollary, it may be noted that, even if the original loop is not unstable, the fact that the measured loop gain is independent of Z,. or Z; indicates that Z" could be considered part of the source of the injectedvoltage signal or Zj part of the injected current signal, and hence in general the injection can be performed from non-ideal voltage and current sources.

t The circuit under discussion (Fig. 10 with 8] and 82 closed) is similar to that. of Fig, 5, which was found in § 2 not to have a minimum-phase loop gain. However, the 709 compensation capacitor in Fig, 10 is about 1/20 of that in Fig, 5, so that the corresponding right-half plane zero in '1' is at about 20 x 12 = 240 kHz, above the range for which data point" are shown in Fig. ll.

Copyright©2001. All Rights Reseved.

498

R. D. Middlebrook

4. Measurement of loop gain at a point of arbitrary impedance by successive voltage and current injection

To measure loop gain by the methods so far described. an injection point in the loop must be found that is driven either by an ideal voltage generator or an ideal current generator. In general, in an actual system, it. ma.'-' not be possible to find an injection point that satisfies either of these extreme conditions" In the switching regulator of Fig. :J, for example. success of tho current-injection method at point Aj depends upon the inequality Z2;Y Zj : however, since Z2 represents the output impedance of Cl transistor, Z2 certainly declines in magnitude at increasing frequencies ow-ing to the collector capacitRIH:C component. In either the voltage 01' current-injection method, inaccuracy in rneasurement of T will occur if the appropriate impedance inequality does not hold.

'1'0 examine this more general ease, consider a point in the feedback loop at which the driving signal is represented neither by an ideal voltage source nor by lUI ideal current source. Nuch a driving signal can be represented either by a Thevenin equivalent or by a Norton equivalent; the Norton equivalent is arbitrarily chosen for discussion, as illustrated in Fig. 12. Current injection at point A would give the true loop gain 'l' as

'I' = (J;,,(/"IIIZ2) ( ]f))
A B
+
+ t
Gmv ZI V
i
T" Gm( Z I II Z2) Fip-ul'l' 12. Exrres~ioll for loop gain T obt.aillPd hv current iujr-ot ion at A.

Nim:c point A in general is not accessible, let voltage injection from n non-ideal voltage source of impedance Z,. be performed at the accessible point 13, representative of a point of arbitrary impedance ratio Z2/Z1' as shown in Fig> 1:1 (a). Measurement of the resulting primed voltages gives a rat.io T,. "" 1"/ /v,/. Analysis of the circuit shows

(L 7)

which is not the same as the true loop gain T. However, olimination of 0", between eqns. (10) and (17) allows the measured ratio 1',. to be expressed in terms of the true loop gain T as

( 18)

Next, let current injection from a non-idea.l current. source of impedance Z i be performed at the same accessible point B, a.:-l shown in Fig-. U~ (b) .

.. . Copyright© 2001. All Rights Reseved .....

~11('(J8nrpmenf oj loop gaitl in fppdhack sysfe'tns

499

+

t

v

!

(a)

+

t

v

!

(b)

Figure 13. Successive signal injection at the (accessible] point B of arbitrary impedance ratio Z2/Z1 in the circuit of Fig. 12, (a) voltage injection and nu-asurt-me-nt of 1'", (IJ) current. injection and measurement. of T;.

~feasuI'ement. of the resulting primed currents gives a ratio T', == i!/Ii/. Analysis of the circuit shows that

( 19)

which is not the same as the true loop gain T, but which can be expressed in terms of T as

(20)

As expected, the inequality Z2-1Z1 leads to T,.~T, so that voltage injection at point B gives the true loop gain, and the opposite inequality Z2>Zl leads to T;-t-T, so that current injection at point B gives the true loop gain.

In general, for arbitrary Zl and Z2' the true loop gain can be derived from measurement of either T,. or T', if Z2/Z1 is known, but it is clearly more convenient to make both sets of measurements (separately) and then to determine T and, if desired, Z2/Z1 from simultaneous solution of eqns. (IS) and (20) :

I I I

--=~--I---

1 -I- T I + T,. I -I- r,

(21 a)

4E2

Copyright © 2001. All Rights Reseved.

500

R. D.Ml:ddlebrool;

or

rf1 '1' 1

l' = _. _,.~.i.__~_

2+ '[',. + '1',

illlrl

Zz ] -[ 7',. Zl ='j'+'Ti

(22)

To illuHlrate this procedure, volb.ge and current-injection measurements were scpnru.telv taken at point H in the circuit of Fig. 10 (switches ;-;[ and ~2 open). Data points of 11',.1 and I T, I are shown in Fig. 14-. Nince the t» are phasora, the proper phase relations m ust be used in evaluation of T from cqn. (21). Although 'I',. and T', con lei be obtained by the phasor triangle measurement described in ~ 2, in the present case the transfer functions Hrc known to he minimum phase] and so the phases can be inferred from the magnitude responses. This is done implicitly, by inference, from the magnitude plots, of expressions for 1'" and '1', in pole-zero form. Thus, in Fig. 14-, best-fit

lor

dB

01

I

-IO~

I

-2°1

L

~ .. ____l_ _ •• __

5 10l\HI

FignH' 14-. Ma.gnittHjp versus frequency plot;; of T'; and 'l'; measured Imccc:-'.",ivel,v 'Lt B in the circuit of Fig. 10 (switches open), the resulting calculated 1'1'1- and the data points of I T I measured diroct.ly hy voltage injection at point A.

straight-line asymptotes are drawn through the data points for I P,·I and 11'i I, from which it is Roell that

r r (1 w,,)

I,! ;:..= "m -+- -;;

( w.)

T,.=T"Ir> 1+--;

(24- )

t The right- half plum: zero present in the loop gain of the circuit of Fig. 5 if; absent in that of .Fig. 10 with 81 open. because of the presence of the 1 {i k j n fleI'il''; wir.l: the compensation capacitor.

,."",.,., • .,o.~ ••• ."."",., •• > ••••••••••••• Cop.y.r.ight. © 2001. AI~'Hights Reseved: ·

JlI easuremenl of loop gain in feedback syslemli

501

where

Tvm.= + 2'5 dB

(25)

t, = w')2rr= 74 kHz

Tim= -2'5 dB

(26) (27) (28)

Substitution of eqns. (23) and (24) into eqn. (2 r) leads to

(29)

where

1 1 1

-=--+-~

Wo T"mwv TimWi

(30)

III -=-+Wl wi WI!

(31 )

1 TvmT'ivr-1

W22 TumTimwvWi

(32)

1 2+ T,.m+ Tim

Ws T"mw" + T·imw,:

(33)

Insertion of numerical values from egns. (25) to (28) gives

10 = wo/2rr = 30 kHz

(34)

(35)

(3G)

fa = w3(2rr = 40 kHz

(37)

Hence, since Wl = W3 and W2 = co, the result for T is

T=wo 8

(38)

The magnitude I T I of the true loop gain calculated from the measured I Ttl and I T', I is thus a straight line of - 6 d Bjoctave slope with croseover at 10 = 39 kHz, as shown in Fig. 14. Since, as has already been seen, the point A" in this circuit meets the condition for the voltage-injection method to give directly the true loop gain, data points for I T I were thus obtained and are also ShO\VB in Fig. 14. Thc calculated straight-line asymptote clearly agrees with the measured data points.

Copyright © 2001, All Rights Reseved.

502

R. D. Middlebrook

For complotcncss, the ratio Z2/Z1 may he determined by substitut.ion of eqns. (~:3) and (24) into eqn. (22) as

r.;». u=·_-

1f,.mw;

(40)

(41 )

(+2)

Insertion of numerical values giveH

( 4[))

A:,; a check, it can be seen by inspection of the circuit of Fig. 10 that ZI .'~ 2211;(10+!)=7-:3k and Z2=IOk, so that Z2iZl··~10/7';i=I·4. Thus, the predicted expression of eqn. (39) is i(Hwcurate both in low-f'requencv value and in that Z2i Xl should not be a function of frequency, that. is, the zero and pole should cancel,

Review of this method of determinat.ion of true loop gain T, from separate measurements of T,. and 'I', at an arbitrary injection point, reveals t.hat inaccuracy is an inherent defect. It is seen from eqn. (2l b) that if T is very small, the product T,.'I'i must approach unity. This can also be seen from eqns. (18) and (20), wherein rp,.· ... Z2/Z1 and T,,,,,X1/Z2 when T_(). Can· sequontly, it is concluded that beyond the loop-gain crossover frequency, when 1'1'1 declines below unity, T,. and 'I', each becomes dominated by' Z2/Z1 and insensitive to T, so that use of eqn. (21) to calculate T from T,. and T', necessarily gives inaccurate results. This effect is seen in the numerical example based on the circuit of Fig. 10, in which W2 .·_c XJ from eqn. (32) because 'l',.",Tfln = 1. Indeed, the numerical results of this example gave better accuracy than one has a right to expect, because a small error in the product T"",T"m would make W2 finite, thereby introducing a spurious zero into the predicted loop gain. Conversely, if thc t.rue loop gain were to have additional zero" or poles beyond the crossover frequency, experimental determination of ']1 from T,. and 'I', would predict them with very poor <lecumey, and might. not detect them at all.

An improved method to overcome this accuracy defect is introduced in the next section.

Copyright© 200·1-;·AII Rights Reseved.

Measurement of loop gain in feedback system8

503

5. Improved measurement of loop gain at a point of arbitrary impedance ratio by null double injection

The problem, restated from the previous section, is to derive the true loop gain of a feedback system by signal injection and measurements at an accessible point that satisfies neither of the extreme inequalities Z2 ~ Zl or Z2>Zl' The problem is represented in Fig. 12, in which the point B is accessible; the true loop gain is

( 40)

but cannot be directly measured if point A is inaccessible.

It may be shown as follows that the desired result is obtained if two sets of measurements are made, each involving simultaneous voltage and current injection at point B as indicated in Fig. Hi. By superposition, each voltage and current indicated in Fig. 15 can be represented by a linear sum of two terms, proportional respectively to the injected voltage and to the injected

----v~ +

f

v

~

Figure 15. .K ull double inject ion of voltagc and current at point B in the circuit of Fig. 12, and measurement of the Hull ratios T,," and T,:".

current. Each such voltage or current can be nulled to zero by a respectively unique ratio of injected voltage to injected current. If, in particular, the injected voltage and injected current are adjusted with respect to each other so that ill' is nulled, measurement of the corresponding values of vx' and VI,' gives a ' null ratio' T,!' which, by inspection of Fig. 15, is easily seen to be

V 'I

n- II _

T" = -:r - GmZ2

v,l: -iy'=O

(47)

Further, if the injected voltage and injected current arc instead adjusted with respect to each other so that vy' is nulled, measurement of the corresponding values of (c' and iy' gives a ' null ratio' Tin which is seen to be

( 48)

Copyright © 2001. All Rights Reseved.

R. 1). Middlr'bro()k

It follows immediately that

/.'t + T1in =l,~ (~l + ~J

and so, from eqll. (4ti),

OJ'

(49)

(50 a)

T nil'!' 'F = -~"~.-'-. 1""+ 7';"

(50 b)

and, directly from eqns. (47) and (48),

T n

1-'

(;31)

T·" ,

TIll' t.rue loop gain T', t.horeforo, can be determined by eqn. (50) from mcusurcrncnts of the null ratios T,." and 7',", where each null ratio is ostahliahr«] by a specific relation between simultanoously injected voltage and current. It is to be noted that it is not necessary to know what this specific relation is ; it is merely necessary to establish the relation by nulling the appropriate voltage or current.

I t is seen that eqn. (!)()) does not suffer from the inherent accuracy defect of eqn. (21), since eqn. (50) does not involve the "mall difference of two nearly equal nurn bers. C'ouscquentlv use of the null double-injection method and the associated eqn. (50) givl:'s the same accuracy whether T is small or lal'g(', and so useful results can be obtained to well beyond the loop-gain crossover frequency. The practical i rnplernentation of the nu II dou ble-injection method is, however, somewhat more complicated than the less accurate successive injection method of the previous section .

.-------~---_'Nv__.-----...,

10k

910n

20k

A

B

9in

Ik

43k

Figure IH. Feedback amplifier circuit for dcmonst.rat.ion of loop-ga.in meaxurvmcu; hy the null double-injection method at point E.

To demonstrate the improvedmethod, the circuit of Fig. 1 () was constructed am] null double injection was performed at point R with the inst.rumentation illustrated in Fig. 17. Voltage injection was accomplished through a clip-OIl current probe 'with 11 ten-turn secondnry, operated 'backv,'anL,,' from the oscillator output of a Hewlett-Packard ;)590A Wave Analyser. Simultaneolls

........ Copyright © 2001. All Rights Reseved.

Measurement oj loop gain in [eedback sy&tems

current . I
probe ly
-
- 1
10:1 vollage v I
L{f:- y
1
+ v/ +
i/
. I +
Iz r
T
v I
y
! magnitude and phose adjust

Hewlett ~ Packard 3590A Wove Analyser

voltage esc.

in out

Figure l7. l nstrumontnt.inn set-up fIJI' the null tlou ble-iuject.ion moaxurouu-nt s at point B in the circuit of Fig. 16.

buffer ompl ifier

C;UITent Injection was accomplished through a blocking ca.pacitor and series resistor. Nince all quantities of concern are phasors, nulling of it signal requires individual magnitude and phase adjustment of the injected current with respect to a given Injected voltage, so the injected current was derived via a phase-shifting network from the same source as the injected voltage. \" oltage and current measurements were taken with appropriate probes.

The experimental technique is as follows. To measure T,", the ana.lyecr input is connected to the current probe to read j;', and ill' is nulled out by appropriate settings of the magnitude and phase adjustments. Then, the

~---~--,----
dB
20
fi~2.okHz
10
0

-10
-20
-30
0.5 1kHz 5 10 50 Figure 18. Magnitude versus frequency plot" of T.." and ']';" measured by null double injection at B in the circuit of Fig. 16, the J'cOiulting calculated I 'I' I, and data points of ITI measured directly at point A.

Copyright©2001. All Rights Reseved.

i'jO()

if, u. Middlebrovk

ana.lvser input is switched to the voltage probe and 11),/ I and I v_/ I a.rc measured The corresponding ratio is I ']',." I, and the data points are shown in Fig. 18. Similarly, to measure Tin, the analyser input is connected to Ow voltage probe to read v!/', and 1)1/ is nulled out by appropriate magnitude and phasn adjustment. Then, the analyser input is switched to the current probe and I il,' I and Ii £' I are measured. The corresponding ratio is 1'1';" I, and data points are also shown in Fig. 18.

As described in ~ 2, independent phase infunnat ion could be obtained b~' measurement of the third phasor but, a.gain, since the transfer functions in this case are known to be minimum phasc.f t he- pole-zero forms for T," 'liHI T,n can he deduced from the magnitude plots. Thus, in Fig lH, best-fit st.raight-line asymptotes are drawn through the data points for IT,." I and 11'/' I, from which it is seen that.

I

7',."" = --;:-( ---;-)

- 1+-

WI· W"

]

(;'2 )

where

']'/'= 8 ( 8 )

- 1+-

Wi w"

f =W /21r= 3·4 kHz

" "

f,.,~w,./27T=4'9 k Hz

(5;3)

Ii = w,·fl7T = 2-0 kHz ~lIhKtitu1iOIl of eqns. (i>2) and (53) into eqn. (50) Iril(h~ to

(i'jU)

I

T = -~----- .. -

"( 8 )

- 1+-

Wo =:

( iii)

where

-:..~::.-.+~

0)0 WI· (0;

(;,)H)

and insertion of numerical values gives

The magnitude I T I of the true loop gain calcu luted from the measured I '/',." I and I Tin I, obtained from eqn, (fi7), is also shown in Fig. 1 R. As a cheek, since the point A in Fig. Hi is accessible and satisfies the condition for volt.ageinjection meaauremerits to give directly the true T, such measurements wen! made, and the data points also shown in Fig. 18 clearly verify the calculated

ITI,

'i'The circuit of Fig. 16 i~ simila.r to that of Fig. t:., and in fact has the same values of 10 and la. However, because t.he 709 compensation capacitor is only 1/10 that in Fig. 5, the corresponding right-half plane zero in T is at about 10 x 12= 120 kHz. well above the range for which data points arc shown in Fig. 18.

Copyright © 2001. All Rights Reseved.

ill easuremenl 01 loop gain in leedback systemti

507

Because In is close to 10' the actual crossover frequency is 1·3 kHz, slightly less than 10' Measurement of the third phasor Ivz I at this frequency gave Ivz/v,/1 = 1·1.5, and the corresponding angle given by eqn. (7) is LTI17'I=L =- 110°. This agrees with the result obtained from eqn. U>7), namely

(uO)

The impedance ratio Z2/Z1 mlly be found by subst.it.ut.ion of Cf[IlS. Ui2) and (53) into eqn. (51) as

(fn)

By inspection of the circuit of Fig. 16, it is seen that hI::::::; 4-:{ II (10 + I) = H·H k and Z2 = 20 k, so that Z2/Z1 = 20/8'8 = 2·3, in good agreement with the cxperimentally determined value.

It has thus been demonstrated that the null double-injection method permits accurate determination of the true loop gain T at a point where neither Z2<{ Zl nor Z2>h1·

The circuit of Fig. 16, chosen for illustration of the null double-injection method, was purposely designed to have a second pole in T beyond the loopgain crossover frequency in order to verify that accurate results could be obtained for this case even for frequencies well above crossover. The circuit is particularly poorly conditioned for measurement of T by the previous successive voltage and current-injection method. As a matter of interest, the ratios T,. and T,. that would be obtained by separate injection at point H can be determined by substitution of eqns. (57) and (6l) into eqns. (IS) and (20) as

I (WJl) (w/I)2

1+- - + -

T =Z2 Q,. s 8

,. hI W"

1+- 8

(Hi)

where

Q,,=Jw;

w"

(04)

and

I (w) (w)2

I + - __!l + __!l

T.=Zl Qi 8 8

. I Z2 W"

1+- 8

( (),3)

where

(U(j )

Qi=J(:J

(67)

Copyright © 2001 . All Rights Reseved ..

;"iOH

R. f). ~MiddlebTo()k

insertion of numerical values from cqns. (54) to (Mi) gives II' = w/J/2iT = 2·(j k Hz

Q,.=()·77

ti' = wq/ 217 = 3'!) k 1-1 Z Qi= 1·.1

( tiS) (tiD) (70) (7 !)

The corrcspondiug n.SYIll ptotcs for 1'1',·1 ,lnd 11' i I arc showu in Fig. I n, along with data points obtained by direct measurement at point B with sepamtc voltage and current injection. Good llgrccment is obtained, but obviously /lny attempt to work tho problem in the other direction, that is, to deduce T from the measured (latn points of l'1'l"! and IT, I, would be futile.

o

-10

66'

f[:..b l!. s»

I

fq ~ 3.9 kliz

__ _________i ... ~ __ •• __ • _ _L .. ___l~ .

0.5 1 kHz 5 10

-20

Figure 1 B. Data p()illt~ of IIl'Lgllitud(· VI·["~H.~ frequelle_\, pIli!" "f 'l',. aru l 'l'i IlW<LAll'P(l hy successi v« voltage a.nd current injection n.t B ill tit .. circuit of Fig. Hi, a.nr] a.~yrnpt()tcs caleu latod from th(· known loop ga.in T. Attcmpt« to deduce T from the measured data point" would lip fu t.i It'.

Finally, it. may be not.ed thn.t. t ho null douhle-injoct.iou method (und abo the suceessivo voltage and current-injection method) at a point B of arbitrary impedance ratio can also he used to measure unstable loop gains. In the derivation based upon the circuit of l"'g. 1;) (and also Fig. 13), tho results are independent of the source impedances Z" and hi respectively" of the uon-idcal voltage and current-injection sources. Consequently, hi' can be made large enough, or Zi small enough, to extinguish any oscillation originally present, in the same way as previously described for injection at a point A where Z2<;Zl or Z2~Zl'

6, Reconsideration of loop-gain measurement by voltage or by current injection

Although the null double-injection method described in the previous section permits the loop gain T to be measured at a point of arbitrary irn pedance ratio Z2/Z1' it. is obviously m uch simpler in practice if an injection point can he found at which the irn pedu.noc ratio iR sufficiently small or sufficiently large t.lmt. either the voltage-injection 01' cun-cnt-injection method alone (:'111 be used. In this section attention will be focused on conditions for which t.his simpler' procedure is acceptable.

Gopyright© 200L.AII Rights. Resevect

ill easuremeui of loop ga1:nin feedback systems

509

In § 2 the practical example represented by the circuit of Fig. 5 was discussed, and the loop gain was measured by voltage injection at point A,.. It. was concluded that tho magnitude and phase measurements of T were consistent with the expression for T given in eqn. (8).

It the measurements on the circuit of Fig. 5 arc extended an additional decade or so in frequency, it is found that, the magnitude response deviates substantially from tho - () dB/oct.ave final asymptotic slope shown in Fig. 7, whereas the phase re~,ponse docs not deviate from the - 2700 final asymptote shown in Fig. H. The extended results are shown in Figs. 20 and 21.

The question now arisos as to whether the magnitude deviation occurs because the actual loop gain doos indeed have such It characteristic, or because the measurement. is giv ing a false result. Further consideru.tion of the circuit gives no cause for modification of the loop gain form glven by eqn. (8), so

o .~.-.,----,~-----,---,,-----~--,-~--,

• fa ~ 3.4 kHz

-10 • ,

-20 dB -30

-40

-50

-60

IT'measured7

v. /

e • /. ••

I~~I measured1

'7

. .

K= -64dB_J Ifz= 8.3kHz

-70 ~ .. L____..-.1..-.~ ~~~~ __ ~ _ _j_____

1510kHz 50 100 500

Figur·c 20.I<:xtencied frcquonoy range magnitude versus frequency plot obtained by voltage ratio measurements at point A ,. in t.he circuit. of Fig. f), and IZ2/Zll data obtained by signal injr-ct.ion at point B with the 0·038 ftF capacitor xh ort . ci rcu itr-r].

! -- ~·······-···r--~

.90°1_

-1350:~~ e., ~~

I ". ~h

-180°·- •

I

-225°r

-270or

,----- -'-. ----,- .-- --I

J J J I

measured

..

. .

. ~. ___L___l ~~ _. __ ..... L_ ._I_ .. .. _ .. _. L_~~ J

510kHz 50 100 500

Fig-tuc 21. Extended frequency rangf' phase versus frequency pl()t ohtuinod for the circuit of Fig. 5.

Copyright©2001. All Rights Reseved.

;")10

If. D. Mhldlf:brook

attention is turned to the measurement. What is actually being measured is not 1', but 7',.: the relation between them is given in eitll. (IS):

(72)

The requirement for 'I',. to be essentially equal to '1' is not only that h2/hl <: l , but also that 152/Z1 <: T, and this second condition is much more restrictive than the first, beyond loop-gain crossover when 'I' <: 1. To determine whether violation of this second condition is responsible for the magnitude deviation observed in Fig. 20, an independent measurement of Z2/hl is needed, In principle, the ratio Z2iZ1 could he obtained from eqn. (22) as

Z2 J -i- :P" ZI = J + 'I';

III the frequency range of interest, 7'" <: J, l1R observed hom Fig, 20, and it, is expected that hZ/hI <: I, :-;0 that 'IT; ~ I. Consequcnt.ly, eqn, (7:{) reduces upproximately to

(74)

and 80 a measurement of T; by' current injection at point A" in the circuit would give the required ratio Z2/Z1' In practice, however, this measuremen L was not possible because of dynamic range limitations : the ratio Z2! Z 1 is so small (at low Irequenoios) that an injected signal capable of giving R readable i", (even with the narrow-band voltmeter of a wave analyser) required an ill that overloaded the output of the 70n, Therefore, recourse was taken to a different method in which the output impedance Z2 of the 709 was measured independently by application of a test, voltage t'J (through a blocking capacitor) to point B in the circuit of Fig, 5, with the 0·038 JA-F capacitor shor-t-circuited to disable the feedback, and measurement of the resulting voltage v2 at point A ". The impedance Z2 was then determined from

(7 fi)

T'h« impedance Zl may be calculated directly from the circuit as Zl = 20 + (4:'11'1 10) = 28 k (in the frequency range of interest, above tal, The resulting measurements of IZ2/Z11 are also shown in Fig, 20, and the best-fit straightline asymptotes indicate that

( 7G)

where

K = 6·3 X JO-4_ -- 64 d B

(77)

(78)

Copyright© 200,L,AII Rights, Reseved.

J·1 easurement o/loop gain in /eedbark systems

511

It is clear from Fig. 20 that the measured 1 T,.I changes from following 1 TI to following IZ2/Z11, within acceptable experimental error, when IZ2/Z11 exceeds 11'1, so that T,. is given by eqn. (72) with Z2/Z1 <{ I

(79)

The analytic expression for T,.) obtained by substitution of eqns. (8) and (76) into eqn. (79), is

where

(80)

(81 )

Substitution of the previously determined numbers gives /1:=72 kHz, in satisfactory agreement with the intersection of the ITI and iZ2/Z11 asymptotes in Fig. 20.

Since the right- and left-half plane zeroes in eqn. (80) give zero net contribution to the phase angle, LT,. given by eqn. (80) is the same as LT given by eqn. (8), in agreement with the measurements of Fig. 21.

It may be concluded, therefore, that the actual loop gain T of the circuit of Fig. 5 is indeed given by eqn. (8), and that the deviation from T observed in the measured T" at higher frequencies occurs because of the breakdown of the condition Z2/Z1 <{ T, in spite of the validity of the condition Z2/Z1 <{ L The actual loop gain IT 1 does in fact continue to decrease at - 6 dB/octave, and does not increase as do the I 'I'; 1 measurcments. This example illustrates the importance of understanding whether the magnitude measurements taken by signal injection are or are not equal to the actual loop-gain magnitude at frequencies beyond crossover, when 1 T I « 1.

7. Conclusions

Practical methods of measuring loop gain T in feedback systems, including switching-mode regulators, have been discussed. "Methods of measurement by voltage injection and by current injection into a closed-loop system are reviewed : if a point in the loop can be found at which the impedance Z2 looking backward is sufficiently smaller than the impedance Zl looking forward, voltage injection is appropriate; conversely, if a point can bc found at which Z2 is sufficiently greater than Zl' current injection is appropriate. Measurement of both magnitude and phase of T is discussed, and it. is shown how the phase can be determined indirectly by measurement of three phasor magnitudes. Possible ill-conditioning of the formula for phase can be eliminated by appropriate scaling of the three measured magnitudes.

A first extension of the method of measurement of loop gain without opening the loop is concerned with loop-gain measurement when the system is unstable. By lowering the loop gain at the point of signal injection, the

Copyright©2001. All Rights Reseved.

;i I:!

Meaeuremen: of loop qain in [eedbaclc ,~ystr'iln8

oscillation can he inhibited and yet a. direct measurement of the unstable loop ga.in 01m still be made. This permits the syst.em to be properly characterizerl so that appropriate correct.ive measures may be taken.

A second extension is concerned with loop-gain measurement when points at which ;1;2 is eufficicntly smaller or sufficiently greater than Z. are inaccessible. I t is shown that in principle the true loop gain T can he determined indirectly hom measurements 1',. and 1'; resulting separately from voltage injection and from current injection at an arbitrary point in the loop at which the ratio .22/7.1 m,LY have any value, However, it is found that this method gives inaccurate result.s above the loop-gain crossover frequency when I T I < 1_

A third extension is concerned with an improved method of determination of 'I' by signal injection at a point of arbitrary impedance ratio Z2/Z1' in the improved method. the null double-injection method, the true loop gain T iK determined from measurements T,." and T;n, each resulting from simult,UH'OUS injection of voltage and current with specific adjusted magnitude and phase relations. The ratio T,." is measured with the magnitude and ph,lfW adjusted to null the current looking backward from the injection point, and the ratio Ti" is measured with the magnitude and phase relation adjusted to nu ll the voltage looking backward from the injection poi ut. It is show II that this improved method avoids t.he inacouracy inherent in the successive voltage-injection and current-injection methods.

Finally, attcnt.ion is redirected to the measurement of loop gain hy the simple voltage OJ' current-injection method, and the oondit.ions to he satisfied by the impedance ratio Z2/Zj at the signnJ-injeetion point iU'C reconsidered.

R,J<:FF. HEN C ,"8 1!ltjfi, Electron. De8igl1, 13, 43.

SPOHN. l\W3. I1m!'!et/-Pachrrd J .. 14, fl.

l!lW'i. l-Iell'll'.tt-Parkard Ap plication Noif, ]\;(). !in, I:> .l anuarv.

·.·-.-.----~-·;_;v_.,.,.,..·-w-·~-·m~~7~;·m~;.:·:·:·:·:·.,.,;,:,:'··:·-·-·-·-G-GPYf1ght.@ 2001': All Rights"Res-eved:: .' .