Basic Principle
xs + xi xo

xf
Benefits of Feedback
Some of the main benefits of feedback are derived from the following observation. Without
feedback, the system gain is simply A. So, for a system with amplification, which is mostly the
case, this A has to be an amplifier, i.e. A > 1. It is hard to build an amplifier where the gain is
wellcontrolled. Often, even with the best designs and processing, amplifier gains can easily vary
by large amounts. Also, the gain is often nonlinear, meaning that it depends on the amplitude of
the input signal. This causes what is called nonlinear distortion.
With feedback, the gain not very dependent on A, but mainly on the feedback factor β. To get a
system that has an overall amplification, this means that β < 1. Circuits that reduce the signal
(gain < 1), such as this β, are much easier to make accurately, e.g. by using a resistive divider.
As a result, with feedback, the overall system gain can be much more accurate and linear. In
other words, some of the benefits of feedback are that it desensitizes the gain and reduces non
linear distortion.
y(t)
x(t)
Low ω0
t
y(t)
x(t)
High ω0
t
∠ A( j ω 0 ) − π −π T
∠ A( j ω 0 ) = π ⇔ τ= = = =−
− ω0 ω 0 2π ⋅ f 0 2
∠H(jω)
5.7°
90°
45°
Actual
0° 0.1·a a 10·a ω (log scale)
5.7°
2
Stability in general
A system is considered to be unstable if there is at least one frequency for which a finite input
sine wave leads to an infinite output (remember any arbitrary signal can always be represented
by a sum of sine waves).
N ( s ) a m ⋅ s m + a m −1 ⋅ s m −1 + .. + a 0
H (s) = =
D(s) s n + b n −1 ⋅ s n −1 + .. + b 0
( s − z1 ) ⋅ ( s − z2 )...( s − zm ) ⎡ ( s − z1 ) ⋅ ( s − z2 ) ⎤ ⎡ ( s − z3 ) ⋅ ( s − z4 ) ⎤
H ( s ) = am ⋅ = am ⋅ ⎢ ⎥⋅⎢ ⎥..
( s − p1 ) ⋅ ( s − p2 )...( s − pn ) ⎣ ( s − p1 ) ⋅ ( s − p 2 ⎦ ⎣
) ( s − p3 ) ⋅ ( s − p 4 ⎦
)
Therefore, to have a stable system, the poles have to lie in the left half plane (σ0 < 0).
3
Stability in a feedback system
A( s )
In a feedback system H ( s ) = A f ( s ) =
1 + A( s ) ⋅ β ( s )
Again, the poles of Af(s) have to lie in the left half plane in order to have a stable system.
By using the Nyquist stability criterion (we will not go into the details of this criterion in this
class), there is an equivalent condition that can be dirived for a feedback system. This equivalent
condition looks at the frequency response of the loop gain:
A( j ω ) ⋅ β ( j ω )
This Nyquist stability criterion (in a simplified form which captures most practical scenarios)
states that, in order to have a stable feedback system, the magnitude of the loop gain has to be
smaller than unity when the phase is equal to 180°. In other words:
∠Aβ = −π ⇒ Aβ < 1
To evaluate this criterion in practical systems, the designer can look at two metrics: the gain
margin and the phase margin.
A( j ω a ) β ( j ω a ) = 1 ⇒ ∠A( jω a ) β ( jω a ) > −π
Therefore, to have a stable system, the PM has to be strictly positive.
4
Aβ (dB)
0
GM > 0 ω (log scale)
<Aβ
0°
ωa ωb
ω (log scale)
90°
PM > 0
180°
Alternatively, the magnitude of A and 1/β can be plotted separately on the same graph. This is
especially useful when β ( jω ) is a constant. It provides a convenient way to evaluate the effect
of different feedback factors β on system stability.
⎛ ⎞
GM = −20 ⋅ log ( A( jω1 ) β ( jω1 ) ) = 20 ⋅ log ⎜ ⎟ − 20 ⋅ log ( A( jω1 ) )
1
⎜ β ( jω ) ⎟
⎝ 1 ⎠
Magnitude (dB)
1 GM > 0
β
0
ωa ωb ω (log scale)
Phase
0°
ω (log scale)
90°
PM > 0
180°
∠Aβ = ±∠ A
5
In summary:
A( s )
We can evaluate the stability of the feedback system A f ( s ) =
1 + A( s ) ⋅ β ( s )
by looking at the frequency response of the loop gain A( jω ) ⋅ β ( jω ) , where we have
defined the gain and phase margins.
Remark:
Often, the feedback system H(s) = Af(s) and the open loop component A(s) are not designed by
the same person. E.g. A(s) could be an op amp you designed, and which is then used by someone
else in their feedback system.
Therefore, op amps and other amplifiers are typically designed such that they are still stable even
with the worst case feedback. The larger the feedback factor β, the larger ωa becomes and
therefore the smaller the PM is. It is typically assumed that the feedback factor is realized with
passive components (resistors, capacitors, inductors), such that β ≤ 1. The worst case feedback,
i.e. the one which is least stable (smallest PM), is unity gain feedback (β =1) in this case.
As an op amp designer, you want to design your device such that the PM is sufficient, even when
β =1. In this case, ωa is the frequency where A( jω a ) = 1 (which is called the unity gain
frequency). When designing your op amp, one of the specifications is therefore often the
minimum phase allowed at unity gain frequency.
Magnitude (dB) 1 1
A >> A <<
β β
A 1
β
ω1 ω2
ω p1 ωa ω p2 ω (log scale)
6
Let us first try to use some intuition to find the closed loop A( jω )
response in this case. A f ( jω ) =
1 + A( jω ) ⋅ β ( jω )
In general, we can identify three regions:
1 A( jω ) 1
(a) A( jω ) >> A f ( jω ) ≈ =
β ( jω ) A( j ω ) ⋅ β ( j ω ) β ( j ω )
(both in magnitude and phase!!!!)
1 A( j ω )
(b) A( jω ) << A f ( jω ) ≈ = A( j ω )
β ( jω ) 1
(both in magnitude and phase!!!!!)
This third type of region is typically bordered by the other two regions. An
example is shown in the figure and marked there by the box. Consider the border
frequencies ω1 and ω2. They also belong to one of the first two types of regions,
and therefore one of the following must hold:
1
A f ( jω ) ≈ or A f ( jω ) ≈ A( jω )
β ( jω )
As a result, the slope of the curve (in dB/decade) at these border frequencies is
known, as it is the slope of either A or β at these frequencies. Let us denote this
slope as S1 and S2 at ω1 and ω2 respectively.
⎡ S1 − S 2 ⎤
Then, from Bode plots, we know that in this area, there must be ⎢ ⎥ poles
⎣ 20 ⎦
(or if negative, that number of zeros). In this example, S1 = 0 dB/decade, S2 = 20
dB/decade, and therefore there is 1 pole in the boxed area for the closed loop
response.
A0
A( s ) = β (s) = β0
⎛ ⎞ ⎛ ⎞
⎜1 + s ⎟ ⋅ ⎜1 + s ⎟
⎜ ω ⎟ ⎜ ω ⎟
⎝ p1 ⎠ ⎝ p2 ⎠
7
In principle, the closed loop response can be calculated mathematically by plugging these
transfer functions into the generic feedback equation:
A( s )
Af (s) =
1 + A( s ) ⋅ β ( s )
We can also obtain the same result more intuitively by using the methodology just described:
1 1
ω < ω1 A >> (e.g. by 20 dB) Af (s) ≈
β β (s)
1
ω > ω2 A << (e.g. by 20 dB) A f ( s ) ≈ A( s )
β
ω1 < ω < ω2 As discussed before, there is one pole in this area. From Bode
plots, we know that this pole is at the intersection of the two
previous line segments on the magnitude plot.
Based on these observations, we can draw the Bode plot of the magnitude of Af, as marked with
the thick line.
Magnitude (dB)
A 1
β
GBW
ω p1 ωa ω p2 ω (log scale)
Af
If you would have calculated this transfer function mathematically, the result would be the same
after appropriate simplifications (i.e. neglecting some insignificant terms).
8
Note that the low frequency gain of the closed loop response is
1 ⎛ 1 ⎞
Af 0 = = A0 ⋅ ⎜⎜ ⎟⎟ .
β0 A ⋅
⎝ 0 0⎠β
Or in other words, compared to the open loop system, feedback reduces the low frequency gain.
It reduces it by a factor corresponding to the loop gain.
Another metric of interest is the bandwidth. Although you may find slightly varying definitions
sometimes, typically, the bandwidth is defined as the frequency at which the gain drops by 3 dB
below the low frequency gain. In this case, this is at the first pole. For the open loop system, the
bandwidth is ωp1. For the closed loop system, the bandwidth is ωa, where
A0
ω a = ω p1 ⋅ = ω p1 ⋅ ( A0 ⋅ β 0 ) .
1 β0
Or in other words, compared to the open loop system, feedback increases the bandwidth. It
increases it by a factor corresponding to the loop gain. Feedback thus results in bandwidth
expansion, at the cost of a proportional reduction in low frequency gain. This is a common
observation for most feedback systems, not just for this particular example.
This proportional gainbandwidth tradeoff is captured in a specific system design metric that is
called the gainbandwidth (GBW) or gainbandwidth product. As the name suggests, it is defined
as the product of the low frequency gain and the bandwidth.
GBWopen = A0 ⋅ ω p1
⎡1⎤
[ ]
GBWclosed = A f 0 ⋅ ω a = ⎢ ⎥ ⋅ ω p1 ⋅ A0 ⋅ β 0 = A0 ⋅ ω p1
⎣ β0 ⎦
We observe that the GBW of the open loop is the same as that of the closed loop. It therefore
represents an intrinsic parameter of the system: although the gain and bandwidth can be traded
off with each other by varying the amount of feedback (i.e. varying β), the GBW will always
remain the same. This is generally true for feedback systems.
Also note that the GBW can be found on a Bode plot by looking at the frequency where the 20
dB portion of the curve intersects with the 0 dB line, as indicated in the figure. This follows
directly from Bode plots.
9
Practical Systems
R2
R1

_ G vout
vin + Rout
xs + xo
A

β xobs
xtest
We have seen that the loop gain is essential in determining stability. Thus far, we have basically
analyzed abstract systems, dealing with signals in general. How do we find the loop gain in
practical electrical circuits, where we have two types of signals: currents and voltages?
A solution is the following methodology: cut the loop, inject a test signal and observe the result:
xobs
= −A⋅β
xtest
However, it is important to terminate the loop with the correct impedance. This will probably
become clearer when we discuss op amps as an example of this general technique, in the next
chapter.
xtest
ZT ZT
10