CHAPTER 3

Closing the Loop — How Feedback Changes Everything

3.1 From Open-Loop to Closed-Loop: A Fundamental Transformation

The Lonely Control Signal

Imagine you're an engineer tasked with controlling a heater to maintain a room at 70°F. The simplest
approach—and the one that seems most intuitive—is open-loop control: measure nothing, assume
everything, and hope for the best. Here's how open-loop control works: you calculate, based on the
room size and heater capacity that setting the heater to "medium" should produce 70°F. You set the
dial, walk away, and cross your fingers.

What could go wrong? Everything:

 It's colder outside than you assumed—the room never reaches 70°F
 It's warmer outside—the room overheats to 85°F
 Someone opens a window—heat pours out, temperature drops
 The heater ages and loses efficiency—same setting, less heat the fundamental problem? You set
an input and hoped for an output, with no way to know if you achieved your goal. This is the
essence of open-loop control: the controller is blind to the results of its actions.

The Problem with "Fire and Forget"

Let's examine the open-loop heater more carefully. The controller sets a power level based on the
desired temperature, but the actual room temperature depends on many factors the controller knows
nothing about.
The open-loop controller faces an impossible task: it must predict the exact relationship between heater
power and room temperature under all conditions. This relationship depends on outdoor temperature,
wind, insulation quality, how many people are in the room, whether windows are open—countless
factors that vary constantly. This is the open-loop curse: to achieve perfect control, you need a perfect
model of everything that affects your system. Since perfect models don't exist, open-loop control is
always imperfect.

The Revolutionary Idea: Watch and Correct

Now consider a different approach. Instead of calculating the "right" heater setting and hoping it works,
what if we simply measured the room temperature and adjusted accordingly? This is the essence of
closed-loop control—also called feedback control. The key insight is disarmingly simple: Don't try to
predict the perfect input. Instead, watch the output, compare it to what you want, and correct any
difference.

With a thermostat (a temperature sensor plus a simple controller), the heater system becomes closed-
loop. Now the controller has eyes—it can see the actual temperature and respond to reality rather than
assumptions.
Look at the dramatic difference. The same physical heater, facing the same varying conditions, now
maintains nearly perfect temperature. Cold day? The controller sees the temperature dropping and
increases heater power. Warm day? It backs off. Window opened? It compensates automatically. The
closed-loop system achieves something remarkable: it doesn't need to know about outdoor
temperature, wind, or insulation quality. It only needs to measure the one thing that matters—the
actual room temperature—and act on the difference between reality and desire.

What "Closing the Loop" Actually Means

The phrase "closing the loop" refers to creating a circular path for information flow. In an openloop
system, information flows in one direction only: from input to output. In a closed-loop system, output
information flows back to influence the input. This circular flow—this loop—is what gives feedback
control its power. Let's trace the information flow in our thermostat system: 1. You set the desired
temperature (the reference or setpoint): 70°F 2. The controller compares the setpoint to the measured
temperature 3. The difference between them—the error—tells the controller what to do 4. If the room
is too cold (positive error), increase heater power 5. If the room is too warm (negative error), decrease
heater power 6. The changed heater power affects the room temperature 7. The new temperature is
measured, and we return to step 2 This cycle repeats continuously. The system is constantly asking: "Am
I at my goal? If not, what must I do to get there?" The loop never truly "closes" in the sense of ending—
it's continuously circulating, endlessly correcting.

The Error Signal: The Gap Between Desire and Reality At the heart of every feedback system is a
comparison: what do I want versus what do I have? The difference between these two quantities is
called the error signal, and it's the lifeblood of feedback control.

Error = Reference − Output

e (t) = r(t) − y(t)

The error signal tells the controller everything it needs to know:

• Positive error (r > y): Output is below target. Action needed to increase output.
• Negative error (r < y): Output is above target. Action needed to decrease output.
• Zero error (r = y): Output equals target. Goal achieved—for now.
Notice something profound: the controller doesn't need to know anything about the plant's internal
workings. It doesn't need to understand thermodynamics, heat transfer, or any physics whatsoever. All
it needs to know is whether the output is above or below the target and by how much. This abstraction
—reducing complex system behavior to a simple error signal—is one of feedback's greatest conceptual
contributions.

Why Feedback Is Revolutionary

At this point, you might think feedback is just a clever trick—a practical workaround for imperfect
models. But feedback is far more than that. It represents a fundamentally different philosophy of
control.

Open-loop philosophy: "If I understand the system perfectly and calculate the exact right input, I'll get
the desired output."

Closed-loop philosophy: "I don't need to understand everything. I'll watch what happens, compare to
what I want, and adjust continuously." The closed-loop philosophy is humbler—it admits we can't
predict everything—but precisely because of that humility, it's more powerful. It works even when our
models are wrong, when conditions change, when components age, when disturbances strike. Consider
the implications:

• Robustness: Closed-loop systems can handle variations and uncertainties that would defeat
open-loop control
• Simplicity of design: We don't need a perfect system model—just "good enough"
• Automatic compensation: Changes in plant behavior are automatically corrected without
redesign
• Disturbance rejection: External disruptions are countered without explicit prediction
 This is why feedback control has become ubiquitous in engineering. From the thermostat in your home
to the cruise control in your car to the voltage regulators in your phone, feedback systems are
everywhere—quietly measuring, comparing, and correcting billions of times per second.

The Transformation in System Character

When you close a feedback loop, you don't just improve a system—you create a new system with
fundamentally different characteristics. The closed-loop system has its own transfer function, its own
poles, and its own dynamic behavior. These can be dramatically different from the original open-loop
system. Consider these transformations that feedback can achieve:

• A sluggish system can become fast

• An oscillatory system can become well-damped
• An imprecise system can become accurate
• A sensitive system can become robust But feedback is not magic—it's a trade-off. To achieve
these benefits, we typically sacrifice some open-loop gain, add complexity, and introduce the
possibility of instability. Understanding these trade-offs is the art of control system design.

3.2 The Closed-Loop Transfer Function

From Block Diagram to Mathematical Relationship

In Section 3.1, we built intuition for why feedback transforms system behavior. Now we need
mathematical tools to quantify that transformation precisely. How does the closed-loop behavior relate
to the open-loop components? The answer is elegant, universal, and deeply useful. Let's begin with the
standard feedback block diagram. Every element in this diagram has a specific meaning and role.
Let's carefully define each element:

 Reference r(t): What you want—the setpoint, target, or command. In our heater example, this is
the desired temperature (70°F).
 Output y(t): What you have—the actual controlled variable. This is the measured room
temperature. Error e(t): The difference e = r − Hy, or for unity feedback (H=1), simply e = r − y.
This is the gap between desire and reality.
 Controller Gc(s): The decision-maker. It takes the error signal and produces a control action. For
now, think of it as an amplifier with gain K.
 Plant Gp(s): The system being controlled—the heater, motor, or whatever physical process
responds to the control signal.
 Sensor H(s): Measures the output and converts it to a signal the controller can compare with the
reference. Often H(s) = 1 ("unity feedback"), meaning the sensor is ideal.
 Forward path G(s): The combined effect of controller and plant in the forward direction: G(s) =
Gc(s) × Gp(s).
 Loop gain L(s): The total gain around the entire loop: L(s) = G(s) × H(s). This quantity will prove
crucial.

Deriving the Closed-Loop Transfer Function Now for the key derivation. We want to find the relationship
between the input r(s) and the output y(s)—the closed-loop transfer function T(s) = Y(s)/R(s).

We'll proceed step by step, starting from what each block does and combining them algebraically.
Let's walk through this derivation in detail, because understanding it deeply pays dividends throughout
control engineering.

Step 1: Write the equations. Each block in the diagram represents a simple input-output relationship:

• The summing junction produces: E = R − HY (error is reference minus feedback)

• The forward path produces: Y = GE (output is forward gain times error)

Step 2: Substitute to eliminate the internal variable E. We want a relationship between R and Y only:

Y = G · E = G · (R − HY)

Step 3: Expand and collect terms with Y on one side:

Y = GR − GHY

Y + GHY = GR

Y(1 + GH) = GR

Step 4: Solve for the transfer function:

T(s) = Y(s)/R(s) = G(s) / [1 + G(s)H(s)])

This is the fundamental equation of feedback control.

The "1 + GH" That Appears Everywhere

Look at the denominator of our closed-loop transfer function: 1 + G(s)H(s), or equivalently, 1 + L(s)
where L is the loop gain. This expression appears throughout feedback theory—in transfer functions,
stability criteria, sensitivity analysis, and more. What does it mean?

The "1" represents the direct path from reference to output if there were no feedback at all. It's the
"baseline" before feedback acts.

The "GH" represents the effect of feedback—signals traveling around the loop and returning. Adding
them (1 + GH) accounts for the combined effect of the direct path and the feedback path.

Here's a physical interpretation: imagine a signal traveling from input to output. Part of it goes directly
through (the "1"). Another part circulates around the loop and comes back (the "GH"). The denominator
(1 + GH) accounts for how these effects combine.

What the Loop Gain Tells You

The loop gain L(s) = G(s)H(s) is the most important quantity in feedback system analysis. It tells you how
strongly feedback acts—how much the system "cares" about errors. Consider three regimes:

Small loop gain (|GH| << 1): The "1" dominates the denominator. T(s) ≈ G(s). The system barely notices
the feedback—it behaves almost like the open-loop system. Feedback is too weak to matter.

Moderate loop gain (|GH| ~ 1 to 10): Both the "1" and "GH" contribute. The closed-loop system shares
characteristics of both the forward path and the feedback path. This is the transition region where
design trade-offs are most delicate.

Large loop gain (|GH| >> 1): The "GH" dominates. T(s) ≈ G(s)/G(s)H(s) = 1/H(s). The closed-loop behavior
is determined almost entirely by the feedback network H(s), not by the forward path G(s). This is the
regime of robust, predictable behavior. The large loop gain case deserves special attention.

When GH >> 1:
 T(s) = G/(1+GH) ≈ G/(GH) = 1/H

With high loop gain, the closed-loop transfer function depends only on the feedback network—not on
the plant. This is the foundation of precision feedback systems.

The Op-Amp Example: Why 1/H Matters

This result explains the mystery from our Engineering Prelude. In an op-amp inverting amplifier, the
open-loop gain A can vary from 20,000 to 500,000 depending on the specific device and temperature.
Yet the closed-loop gain is rock-solid. How?

Let's analyze this carefully:

• The forward gain is G = A (the op-amp's open-loop gain, say 100,000)

• The feedback factor is H = Rin/(Rin + Rf) (set by resistors, say 1/11)
• The loop gain is GH = 100,000 × (1/11) ≈ 9,000
• Since GH >> 1, we have T ≈ 1/H = (Rin + Rf)/Rin = 11

The closed-loop gain of 11 (or −10 for the inverting configuration, accounting for the sign) depends only
on the resistor ratio—not on the op-amp's gain. If the op-amp gain drifts from 100,000 to 50,000, the
loop gain drops from 9,000 to 4,500, but both are still >> 1, so T still equals 1/H. This is the power of high
loop gain: it makes the system insensitive to forward-path variations. The feedback "doesn't care" if the
op-amp gain varies, because it's high enough that the closed loop behavior is set by H alone.

Unity Feedback: The Common Special Case

In many practical systems, the feedback sensor is essentially perfect—it measures the output directly
without any scaling or dynamics. This case, where H(s) = 1, is called unity feedback. For unity feedback,
the closed-loop transfer function simplifies to:

T(s) = G(s) / [1 + G(s)]

And the high-gain limit becomes:

If G >> 1: T(s) ≈ 1

This means a unity feedback system with high loop gain approaches an ideal behavior: output equals
reference. The system tracks the setpoint perfectly. Unity feedback is common because it's conceptually
simple (the sensor just measures the output) and because the resulting equations are cleaner.
Throughout this course, unless otherwise specified, we'll often assume unity feedback.

The Closed-Loop Poles: A Preview

Here's a crucial insight that will occupy us for much of this course: the denominator 1 + GH determine
the closed-loop system's poles.
Recall from Chapter 2 that poles determine dynamic behavior—stability, speed, oscillation. The closed-
loop transfer function has the form:

T(s) = G(s) / [1 + G(s)H(s)]

The poles of T(s) are the values of s that make the denominator zero:

1 + G(s)H(s) = 0

This is called the characteristic equation of the closed-loop system. Its roots are the closedloop poles—
and they're generally different from the open-loop poles of G(s) and H(s).

Closing the loop moves the poles. The closed-loop poles are NOT the same as the open-loop poles. This
is why feedback can transform a sluggish system into a fast one, or an oscillatory system into a well-
damped one.

Verification: Does the Formula Make Sense?

Before trusting any formula, good engineers check that it makes sense in limiting cases:

Check 1: No feedback (H = 0). If there's no feedback path, we should get the open-loop behavior.

T = G/(1 + G·0) = G/1 = G ✓

Yes! With no feedback, closed-loop equals open-loop. The formula works.

Check 2: Infinite forward gain (G → ∞) with unity feedback. This should give perfect tracking. T = G/(1 +
G) = 1/(1/G + 1) → 1/(0 + 1) = 1 ✓

Yes! With infinite loop gain, output equals reference exactly.

Check 3: Loop gain equals −1 (GH = −1). This is the boundary of stability.

T = G/(1 + (−1)) = G/0 → ∞

The system "blows up"—infinite gain at this point. This is the onset of instability,

Why This Equation Is Universal

The closed-loop transfer function T = G/(1+GH) isn't just for heaters or op-amps. It applies to virtually
every feedback system in existence:

• Audio amplifiers with negative feedback

• Voltage regulators
• Motor speed and position controllers
• Phase-locked loops
• Biological homeostasis (body temperature, blood sugar)
• Economic feedback systems (supply and demand)
 • Social feedback loops (popularity, trends)

Wherever there's a loop—wherever output information circles back to influence input—this

fundamental equation describes the resulting behavior. The specific G(s) and H(s) may differ, but the
structure T = G/(1+GH) is universal.

3.3 How Feedback Moves the Poles

The Poles Are Not Where You Left Them

Here's a fact that surprises many students: when you close a feedback loop, the system's poles
move. The closed-loop poles are not the same as the open-loop poles. They're in completely
different locations.

Consider the simplest possible scenario: a first-order plant controlled by a simple gain (proportional
controller), with unity feedback. The plant transfer function is:

Gp(s) = 1/(s + 2)

This plant has a single open-loop pole at s = −2, this means a time constant τ = 0.5 seconds. The
system responds to inputs with a characteristic time of about half a second.

Now add proportional controllers with gain K: Gc(s) = K

The forward path is: G(s) = Gc(s) · Gp(s) = K/(s + 2)

With unity feedback (H = 1), the closed-loop transfer function is:

T(s) = G/(1 + G) = [K/(s+2)] / [1 + K/(s+2)] = K/(s + 2 + K)

Look at the denominator: s + 2 + K.

The closed-loop pole is at: s = −(2 + K)

This is not at s = −2 anymore! The closed-loop pole location depends on the controller gain K.

The pattern is clear: increasing K pushes the pole further into the left half-plane, making the system
faster. With K = 98, the system is 50 times faster than the original plant! This is the magic of feedback:
by adjusting a single parameter (the controller gain), we've fundamentally changed the system's
dynamic character.

The General Principle: Characteristic Equation

The pole movement we just saw isn't a coincidence—it follows directly from the closed-loop transfer
function structure. For any feedback system:

T(s) = G(s) / [1 + G(s)H(s)]

The closed-loop poles are the roots of the characteristic equation: 1 + G(s)H(s) = 0

Or equivalently:
 G(s)H(s) = −1

This equation tells us: the closed-loop poles are located wherever the loop gain equals exactly −1. These
locations generally have nothing to do with where the open-loop poles were.

The characteristic equation 1 + GH = 0 is the master key to closed-loop dynamics. Its roots determine
stability, speed, and oscillation behavior.

Example: Second-Order Plant — Changing Both Speed and Damping

First-order systems can only get faster or slower. Second-order systems are more interesting: feedback
can change both the natural frequency and the damping ratio. Consider an underdamped second-order
plant (perhaps a lightly-damped mechanical system):

Gp(s) = 4 / (s² + 0.4s + 4)

This has ωn = 2 rad/s and ζ = 0.1 (very underdamped—it oscillates significantly). The open-loop poles are
at

s = −0.2 ± j1.99

These poles are close to the imaginary axis (small real part) and have significant imaginary component
(oscillation).

Now add proportional control with gain K and unity feedback:

T(s) = 4K / [s² + 0.4s + 4 + 4K] = 4K / [s² + 0.4s + 4(1+K)]

The closed-loop characteristic equation is:

s² + 0.4s + 4(1+K) = 0

Comparing to the standard form s² + 2ζωₙs + ωₙ²,

we get: ωₙ_CL = 2√(1+K)

ζ_CL = 0.2/√(1+K)
This illustrates a critical lesson: simple proportional control doesn't always improve things. For this
system, we've traded damping for speed. To improve both, we need smarter control strategies (which
we'll explore in later chapters with PD and PID controllers).

Visualizing Pole Movement: The Root Locus Preview

The examples above show poles moving as we change the gain K. If we plot all possible pole locations as
K varies from 0 to ∞, we get a picture called the root locus.

The root locus is a map showing where the closed-loop poles can go as we adjust gain. It's one of the
most powerful tools in classical control design, and we'll study it extensively in Chapter 5. For now,
here's the key intuition:

• At K = 0: closed-loop poles equal open-loop poles (no feedback effect)

• As K increases: poles migrate along specific paths determined by the system structure
• At K → ∞: poles approach the open-loop zeros or go to infinity along asymptotes
The Fundamental Trade-off: You Can't Get Everything

Our second-order example revealed a fundamental truth: pole placement involves trade-offs. You
typically can't optimize everything simultaneously with a single parameter.

Common trade-offs include:

• Speed vs. damping: Faster response often means more overshoot

• Speed vs. stability margin: Pushing poles far left may reduce robustness
• Low-frequency accuracy vs. high-frequency noise: High gain helps tracking but amplifies noise
• Disturbance rejection vs. sensor noise sensitivity: The same feedback that rejects disturbances
also amplifies sensor imperfections The art of control design is navigating these trade-offs—
choosing pole locations that give acceptable performance across all requirements, not optimal
performance in any single dimension.

What Determines Where Poles Can Go?

A natural question arises: can we put poles anywhere we want? The answer is "it depends." For a
system with n poles, we generally need n adjustable parameters to place all n poles at arbitrary
locations. With only one parameter (a proportional gain K), the poles are constrained to move along
specific paths—we can't put them just anywhere. This motivates the use of more sophisticated
controllers:

• Proportional (P) control: One parameter → poles move along root locus
• Proportional-Derivative (PD) control: Two parameters → more freedom in pole placement
• Proportional-Integral-Derivative (PID) control: Three parameters → even more flexibility
• State feedback: n parameters → arbitrary pole placement (in theory)

Closed-Loop Zeros: The Other Half of the Story

We've focused on poles, but closed-loop systems also have zeros. The closed-loop zeros come from two
sources:
1. Zeros of the forward path G(s): These appear directly in the closed-loop transfer function
numerator
2. Poles of the feedback path H(s): If H(s) has dynamics, its poles become closed-loop zeros

For our unity feedback examples (H = 1), the closed-loop zeros are simply the zeros of G(s). But for
more complex feedback structures, zeros can appear in unexpected places. Zeros affect the
transient response shape—particularly overshoot and the initial direction of response. A right-half-
plane zero, for instance, can cause the system to initially move in the wrong direction before
correcting (non-minimum phase behavior).

3.4 The Four Gifts of Feedback

Gift #1: Reduced Sensitivity to Plant Variations Components drift. Resistors age. Motors wear out.
Transistor gains vary with temperature. In the real world, nothing stays constant. An open-loop
system designed for nominal conditions will perform poorly when conditions change.

Feedback fights this problem by automatically compensating for variations in the plant

The Sensitivity Function Consider our closed-loop transfer function:

T(s) = G(s) / [1 + G(s)H(s)]

Suppose the forward gain G changes by a small amount ΔG due to component drift. How much does
the closed-loop gain T change? Using calculus (specifically, the definition of sensitivity), we can
show:

(ΔT/T) / (ΔG/G) = 1 / (1 + GH)

This ratio is called the sensitivity function S:

S = 1 / (1 + GH) (3.2)

The sensitivity function tells us: if G changes by 1%, how much does T change in percentage terms?

Gift #2: Disturbance Rejection Real systems don't operate in isolation. Disturbances constantly
assault them: load changes on motors, wind gusts on aircraft, power supply ripple on voltage
regulators, thermal fluctuations on precision instruments. An open-loop system has no defense
against disturbances—they pass directly to the output. A closed-loop system fights back.

Analyzing Disturbance Response Let's add a disturbance D(s) entering at the plant output (a
common scenario):
With this configuration, the output becomes:

Y(s) = [G(s)/(1 + G(s)H(s))]R(s) + [1/(1 + G(s)H(s))]D(s)

The first term is the familiar response to the reference.

The second term shows how disturbances affect the output. Notice the same (1 + GH) factor in the
denominator!

The disturbance rejection ratio is: Control Systems for Electronics Engineering

Y_from_D / D = 1/(1 + GH) = S

It's the same sensitivity function! Feedback attenuates disturbances by the same factor it reduces
sensitivity to plant variations. This isn't a coincidence—both effects arise from the same mechanism.
Gift #3: Improved Transient Response

We've already seen that feedback moves poles, changing the system's dynamic character. Let's quantify
exactly how feedback improves transient response.

Speeding Up a Sluggish System For a first-order system with proportional control:

Gp(s) = a/(s + a) (time constant τ = 1/a)

T(s) = Ka/[s + a + Ka] = Ka/[s + a(1+K)]

The closed-loop time constant is:

Adding Damping to an Oscillatory System

Second-order systems present more nuanced opportunities. While simple proportional control may not
improve damping (as we saw in Section 3.3), more sophisticated control can. Consider adding derivative
action—a controller that responds to the rate of change of error:
 Gc(s) = Kp + Kd·s

This proportional-derivative (PD) controller adds a term proportional to velocity. In the closed loop
characteristic equation, the Kd term adds to the damping coefficient, allowing us to increase ζ while also
increasing ωn.

Gift #4: Predictable Behavior from Unpredictable Components This gift is perhaps the most profound,
and we've already glimpsed it in the op-amp discussion. With sufficient loop gain:

T(s) ≈ 1/H(s)

The closed-loop behavior depends only on the feedback network, not on the forward path. This means:
• The plant can be imprecise, nonlinear, or poorly characterized

• The plant can vary with temperature, age, or manufacturing lot

• As long as the gain is "high enough," none of this matters

The feedback network H(s) is typically made of precision passive components (resistors, capacitors) that
are stable, linear, and well-characterized. By making T depend only on H, we've traded unpredictable
active components for predictable passive ones.

The Op-Amp Golden Rule Revisited In an op-amp circuit with large A (open-loop gain):

• Closed-loop gain ≈ 1/β where β is the feedback fraction

• For an inverting amplifier: Gain ≈ −Rf/Rin
• For a non-inverting amplifier: Gain ≈ 1 + Rf/Rin

These depend only on resistor ratios! Modern resistors can achieve 0.1% tolerance easily and 0.01%
with modest cost. The op-amp's wild variations (25:1 gain spread, temperature coefficients,
nonlinearity) become irrelevant.
Design Philosophy: "Make G Large, Make H Precise" This leads to a powerful design philosophy for
precision systems:

1. Make G as large as possible: Use high-gain amplifiers, don't worry about their exact gain
2. Make H precise: Use precision components in the feedback path
3. Let feedback do the rest: The closed-loop behavior will inherit H's precision

This is why op-amps are designed for maximum open-loop gain, not for precise open-loop gain. The
gain doesn't need to be accurate—it just needs to be big.

The Four Gifts: A Unified View

All four benefits of feedback trace back to the same mathematical structure—the (1 + GH) term in the
denominator:
How Much Feedback Is Enough?

The benefits scale with loop gain: more GH means better sensitivity, better disturbance rejection, more
pole movement freedom. So why not make GH enormous? Several factors limit practical loop gain:

1. Stability constraints: Very high gain can push the system toward instability

2. Bandwidth limitations: Gain decreases at high frequencies for physical systems

3. Noise amplification: High feedback gain also amplifies sensor noise

4. Actuator saturation: Large corrections require large control signals, which may exceed physical limits

Gift or Curse? When Feedback Backfires Feedback's gifts come with conditions.
When those conditions aren't met, feedback can actually make things worse:

• At frequencies where GH ≈ −1: Feedback amplifies rather than attenuates (resonance,

instability)
• When sensor noise is significant: Feedback injects noise into the system
• When modeling errors are large: Feedback may fight the real system instead of helping it
• When delay is present: Feedback corrections arrive too late and overshoot.

3.5 The Price of Feedback

Nothing Is Free We've spent the last two sections celebrating feedback's gifts: reduced sensitivity,
disturbance rejection, improved dynamics, predictable behavior. These are real benefits that make
feedback indispensable in modern engineering. But feedback is not a free lunch. Every benefit
comes with a cost. Understanding these costs is essential for knowing when feedback is worth the
complexity—and when simpler approaches might be better.

Cost #1: Reduced DC Gain

Here's a fact that surprises many students: closing a feedback loop reduces the overall gain
compared to the forward path alone.

Consider a unity feedback system:

T = G / (1 + G)

If G = 100, then T = 100/101 ≈ 0.99.

We started with a gain of 100 and ended with a gain of approximately 1!

Where did all that gain go? Into fighting errors. The high forward gain is "used up" creating the large
loop gain that provides sensitivity reduction and disturbance rejection. The gain doesn't disappear—
it's redirected from amplification to error correction.
This isn't necessarily a problem—we often want a gain of 1 (perfect tracking). But if you need high
gain and high loop gain, you need an enormous forward gain to have "enough left over" after
feedback takes its share.

For non-unity feedback systems:

T = G/(1 + GH) ≈ 1/H (for large GH)

The closed-loop gain is set by 1/H, regardless of how large G is. To get a closed-loop gain of 100, you
need H = 0.01. Then to have high loop gain (say GH = 1000), you need G = 100,000. The required
forward gain escalates quickly.

Cost #2: Bandwidth-Gain Trade-off

Real amplifiers and controllers don't maintain constant gain forever—their gain decreases at high
frequencies. A typical op-amp might have DC gain of 100,000 but only gain of 10 at 1 MHz. This
frequency-dependent gain creates a fundamental trade-off. The loop gain GH—which determines all
our feedback benefits—also varies with frequency. At frequencies where |GH| drops below 1,
feedback essentially stops working.

The gain-bandwidth product is approximately constant for many amplifiers:

 Gain × Bandwidth ≈ constant

Want more gain? You get less bandwidth. Want more bandwidth? Accept less gain. This is a
fundamental limitation of the amplifier, not something feedback can circumvent—feedback just lets you
choose where on this trade-off curve you want to operate.

For an op-amp with gain-bandwidth product of 1 MHz:

 Closed-loop gain of 1 → Bandwidth ≈ 1 MHz

 Closed-loop gain of 10 → Bandwidth ≈ 100 kHz
 Closed-loop gain of 100 → Bandwidth ≈ 10 kHz
 Closed-loop gain of 1000 → Bandwidth ≈ 1 kHz

Cost #3: Added Complexity and Components An open-loop system is simple: input goes in, output comes
out. A closed-loop system requires additional components:

• Sensor: Must measure the output accurately and quickly

• Reference source: Must provide a stable, accurate setpoint
• Summing circuit: Must compute the error signal
• Controller: May require additional active components
• Wiring/interconnections: More points of potential failure

Each added component introduces cost, space, power consumption, failure modes, and design
complexity. For simple applications, this overhead may not be justified.

Cost #4: The Specter of Instability

This is the big one—the cost that keeps control engineers employed. Feedback can make systems
unstable. A system that was perfectly well-behaved open-loop can oscillate wildly, saturate its actuators,
or destroy itself when the feedback loop is closed. How can "negative" feedback—which is supposed to
correct errors—cause oscillation? The problem is phase shift. As signals travel through the forward path
G(s) and feedback path H(s), they experience frequency-dependent phase shifts. At some frequency, the
total phase shift around the loop may reach 180°. At that frequency:

 What was negative feedback becomes positive feedback

 Instead of correcting errors, the system reinforces them
 If the loop gain exceeds 1 at this frequency, oscillations grow

Instability occurs when the loop gain equals −1: magnitude of 1 and phase of 180°. At this condition,
errors are not corrected—they are perfectly reinforced.

Cost #5: Noise Amplification

Feedback loops don't just process reference signals and disturbances—they also process sensor noise.
And they process it in an unfortunate way.

The output becomes:

Y = [G/(1+GH)]R − [GH/(1+GH)]N

Look at the noise term: for large GH, the noise transfer function approaches:

Y_from_N / N = −GH/(1+GH) ≈ −1
The sensor noise passes directly to the output with magnitude ≈ 1! The same high loop gain that rejects
disturbances entering at the plant does nothing to reject noise at the sensor. In fact:

• Disturbances at the plant are attenuated by 1/(1+GH)

• Noise at the sensor is passed through with gain GH/(1+GH) ≈ 1

This asymmetry has profound implications. A noisy sensor will inject that noise directly into the
controlled variable. If your temperature sensor has 0.1°C of noise, your temperature control will have
0.1°C of noise—no matter how good your controller.

Cost #6: Actuator Saturation

When disturbances strike or references change rapidly, the controller demands large corrective actions.
But real actuators have limits: motors have maximum torque and speed, amplifiers have maximum
voltage swing, valves can only open so far, heaters have maximum power. When the controller demands
more than the actuator can deliver, the actuator saturates—it gives its maximum, regardless of what's
requested.
When Is Feedback Worth It?
Feedback is clearly worth it when:

• Plant parameters vary significantly or unpredictably

• Disturbances are substantial and must be rejected
• Precise output is needed despite imprecise components
• Dynamics must be modified (faster response, better damping)

Feedback may not be worth it when:

• Plant is inherently precise and stable

• Disturbances are negligible
• Open-loop accuracy is sufficient for the application
• Simplicity and reliability are paramount

A resistor voltage divider doesn't need feedback to produce 2.5V from 5V—it's simple, reliable, and
accurate enough for many purposes. A power supply delivering 5V ± 1% to a microprocessor while
rejecting 120Hz ripple and load transients absolutely needs feedback.
3.6 When Feedback Fights Back: The Stability Problem
The Shower Temperature Problem You step into a shower. The water is cold. You turn the hot water
knob. Still cold. You turn it more. Still cold! Frustrated, you crank it all the way up. Suddenly, scalding
water hits you. You leap back, turn the knob down hard. Now it's freezing again. You adjust up, then
down, then up—oscillating between extremes, never finding comfort. What went wrong? The
shower has a delay between when you turn the knob and when the temperature changes. You were
adjusting based on the current temperature, but your adjustments affected the future temperature.
By the time you felt the scalding water, you had already made it too hot.
This is instability in its most familiar form: corrections arrive too late and overshoot, causing
oscillation. The system isn't broken—you're just fighting the physics.

The Positive Feedback Trap In normal negative feedback:

 Temperature too low → increase hot water → temperature rises

 Temperature too high → decrease hot water → temperature falls
 Error produces correction in the opposite direction.

But with delay, the relationship shifts. At the critical frequency—where the delay equals half a
period—negative feedback has become positive feedback. Instead of fighting errors, the system
amplifies them.

When the phase shift around the loop reaches 180°, negative feedback becomes positive feedback.
If the loop gain is greater than 1 at this frequency, oscillations grow.

Understanding Phase Shift

Where does phase shift come from? Every dynamic element in the loop contributes:

First-order systems (like RC circuits, thermal masses):

G(s) = 1/(τs + 1)

• Magnitude decreases at high frequencies (−20 dB/decade roll-off)

• Phase shifts from 0° toward −90°

Second-order systems (like RLC circuits, mechanical resonances):

G(s) = ωₙ² / (s² + 2ζωₙs + ωₙ²)

• Magnitude has resonant peak (if underdamped)

• Phase shifts from 0° through −90° toward −180°

Pure time delay (like transport processes, digital sampling):

G(s) = e^(−sT)

Phase decreases linearly: ∠e^(−jωT) = −ωT

• Magnitude is constant (|e^(−jωT)| = 1)
•
 The Critical Point: GH = −1
Let's state the stability criterion precisely:
A feedback system becomes unstable when the loop gain GH = −1.
This condition means:

• Phase: ∠GH = −180° (or equivalently, +180°)

• Magnitude: |GH| = 1

At this condition, a signal going around the loop comes back with the same magnitude and inverted
sign. But since we're already subtracting (negative feedback), the inversion makes it adding. The signal
reinforces itself exactly—perfect positive feedback.

Physical Intuition: Why Does It Oscillate?

Let's trace what happens in an unstable system, step by step:

1. Small disturbance: Noise creates a tiny deviation from equilibrium

2. Error detected: The feedback path senses this error
3. Correction generated: The controller produces a corrective signal
4. Phase-shifted correction: Due to system dynamics, this correction is delayed
5. At critical frequency: The correction arrives 180° out of phase—as reinforcement
6. Amplification: If |GH| > 1, the reinforcement is larger than the original disturbance 7. Growing
oscillation: The cycle repeats, with each pass making the oscillation larger

Example: The Unstable Position Controller

Consider a simple position control system: a motor driving a mass with proportional control. The
plant (motor + mass) has transfer function:
Gp(s) = Km / [s(Js + b)]
This is a second-order system with two poles: one at s = 0 (the integrator—position is integral of
velocity) and one at s = −b/J. The open-loop system already contributes −180° at high frequency
(−90° from each pole). Add proportional control and unity feedback: L(s) = GH = K·Km / [s(Js + b)]
At ω → ∞, the phase approaches −180°. But there's also a frequency where |L| = 1 while phase
= −180°. If our proportional gain K is too high, we push |L| above 1 at this frequency—and the
system oscillates.

Example: The Unstable Position Controller Consider a simple position control system: a motor
driving a mass with proportional control. The plant (motor + mass) has transfer function:

Gp(s) = Km / [s(Js + b)]

This is a second-order system with two poles: one at s = 0 (the integrator—position is integral of
velocity) and one at s = −b/J. The open-loop system already contributes −180° at high frequency
(−90° from each pole).
Add proportional control and unity feedback:

L(s) = GH = K·Km / [s(Js + b)]

At ω → ∞, the phase approaches −180°. But there's also a frequency where |L| = 1 while phase
= −180°. If our proportional gain K is too high, we push |L| above 1 at this frequency—and the
system oscillates.
Stability Margins: How Close to the Edge?
In practice, we don't just want stability—we want robust stability with margin for error. Two
margins are commonly used:
Gain Margin (GM): How much can we increase the gain before the system goes unstable? •
Measured at the frequency where phase = −180° • GM = 1/|GH| at that frequency • Typically
expressed in dB: GM(dB) = −20 log|GH|

the frequency where |GH| = 1 (the "gain crossover") • PM = 180° + ∠GH at that frequency
Phase Margin (PM): How much additional phase lag can the system tolerate? • Measured at

Common Causes of Instability

In practice, stability problems usually arise from:
1. Excessive gain — The most common cause. Fix: Reduce controller gain.
2. Unmodeled dynamics — Your model said the plant was first-order; it's actually fourth-order.
Fix: Include all significant dynamics in analysis.
 3. Time delays — Transport delays, computational delays, sampling delays. Fix: Account for
delay; reduce gain; use Smith predictor.
4. Right-half-plane zeros — These add phase lag unexpectedly. Fix: Accept performance
limitations; specialized control techniques.
5. Resonant modes — Mechanical resonances, LC oscillations. Fix: Notch filters, or gain
scheduling to avoid resonance.

3.7 Feedback in Electronic Circuits

Where Theory Meets the Soldering Iron We've developed feedback theory using abstract block diagrams
—G(s), H(s), transfer functions, [Link] it's time to see these concepts in real electronic circuits. The
mathematics is identical, but seeing feedback "in the flesh" deepens understanding and prepares you for
practical design.

3.7.1 The Inverting Amplifier: Feedback's Signature Circuit

The op-amp inverting amplifier is perhaps the most important circuit in analog electronics. It transforms
a component with wildly variable gain into a precision amplifier with gain determined entirely by passive
resistors.
Circuit Analysis Using Feedback Theory
Let's map this circuit to our feedback framework:

The forward path G: The op-amp itself, with open-loop gain A (typically 10⁵ to 10⁶). G = −A

The feedback network H: The voltage divider formed by Rin and Rf: H = Rin/(Rin + Rf)

The loop gain: GH = −A · Rin/(Rin + Rf), magnitude very large For large A, the closed-loop gain simplifies
to:

T = −Rf / Rin

This is the famous inverting amplifier formula: gain equals minus the ratio of resistors.

The Virtual Ground Concept

Feedback theory explains the "virtual ground" at the inverting input. With loop gain GH >> 1:

• The error signal (voltage at the inverting input) must be nearly zero
• If it weren't zero, the huge gain A would produce an enormous output
• Feedback forces the inverting input to stay at the same potential as ground
The Feedback Analysis

Reference: A zener diode or bandgap reference provides stable Vref.

Error amplifier: An op-amp compares the reference to a fraction of the output. Pass element: The
transistor controls current flow from input to output.

For large loop gain: Vout ≈ Vref / β = Vref · (R1 + R2) / R2 (3.6) The output voltage depends only on the
reference and resistor ratio—not on input voltage, load current, or transistor characteristics!
3.7.4 The Phase-Locked Loop
The phase-locked loop (PLL) is a feedback system that locks an oscillator's phase to a reference signal.
PLLs are ubiquitous in communications, clock generation, frequency synthesis, and motor control.
3.8 Chapter Summary
The Big Picture

This chapter has been about one of engineering's most powerful ideas: closing the loop transforms
system behavior. By feeding output information back to the input, we create systems with
capabilities far beyond their open-loop components.

The Fundamental Equation

Everything flows from the closed-loop transfer function:

T(s) = G(s) / [1 + G(s)H(s)] (3.7)

This single equation encapsulates the mathematics of feedback. The numerator G(s) is the forward
path. The denominator 1 + GH adds the feedback effect. The roots of 1 + GH = 0 determine stability
and dynamic behavior.
The Stability Constraint
Feedback's benefits depend entirely on stability. The critical condition:

1 + G(s)H(s) = 0 ⟺ GH = −1

When loop gain equals −1 (magnitude 1, phase −180°), the system is at the boundary of instability.
Stability margins provide safety buffer:

• Gain margin ≥ 6 dB: System tolerates 2× gain increase

• Phase margin ≥ 45°: System tolerates 45° additional lag