Transmission Zeros

In the world of control systems and signal processing, some concepts are intuitively grasped while others remain shrouded in mystery. Transmission zeros fall firmly into the latter category. While system poles, representing natural frequencies, are a familiar part of an engineer's vocabulary, zeros represent something more subtle and profound: an inherent property that can block signals, dictate performance limits, and even cause a system to initially move in the wrong direction. This article addresses the knowledge gap by demystifying transmission zeros, transforming them from an abstract mathematical curiosity into a tangible engineering concept. In the chapters that follow, we will first delve into the "Principles and Mechanisms," exploring what transmission zeros are from both an input-output and an internal state-space perspective. Subsequently, under "Applications and Interdisciplinary Connections," we will see how these zeros manifest as both fundamental limitations and powerful design tools across various fields, from aerospace engineering to audio filter design.

Now that we’ve been introduced to the curious idea of a transmission zero, let’s take a journey to understand what it really is. Like any good exploration in physics or engineering, we’ll start with a simple, familiar idea and then see how it blossoms into something deep and powerful when we look at it from different angles. We’re not just learning a definition; we’re trying to build an intuition for the machinery of the universe as it's reflected in the systems we build.

Let’s start with a simple system, one with a single input and a single output (SISO). You might have seen its behavior described by a transfer function, which is just a fancy fraction of polynomials in a complex variable $s$: $G(s) = N(s)/P(s)$. The roots of the denominator $P(s)$ are the ​​poles​​, which you can think of as the system’s natural "ringing" frequencies—the frequencies at which it wants to vibrate on its own.

But what about the roots of the numerator, $N(s)$? These are the ​​zeros​​. If you feed the system a signal oscillating at a frequency $s_0$ that happens to be a zero, something remarkable happens: the output is zero. The system, at that specific frequency, transmits nothing. It completely blocks the signal. This is the origin of the name ​​transmission zero​​. For example, a system with a transfer function like $G(s) = (s-3)/(s^2+5s+6)$ has a transmission zero at $s=3$.

This is easy enough for one input and one output. But what about a more complex machine, like a chemical reactor with multiple valves (inputs) and multiple sensors (outputs)? Now we have a whole matrix of transfer functions, $G(s)$. What does it mean for a matrix to "block" a signal?

The idea becomes more subtle and beautiful. A transmission zero is no longer a frequency where all outputs are zero for any input. Instead, it’s a frequency $s_0$ where the system's ability to map inputs to outputs becomes compromised. It's a frequency where you can find a special input direction—a specific combination of input signals—that results in zero output. The system isn't completely dead, but it has a blind spot. Mathematically, we say the transfer matrix $G(s_0)$ ​​loses rank​​.

For a square matrix that is normally invertible, this loss of rank happens precisely when its determinant is zero. This gives us a handy computational tool. For a $2 \times 2$ system, for instance, we can find the transmission zeros by solving $\det(G(s)) = 0$. But this is just a special case! The fundamental idea is the drop in rank, which is a much more general concept that works even for non-square systems (say, 3 inputs and 2 outputs) where a determinant isn't even defined. The true definition of a transmission zero is this loss of transmission capability, this emergence of a "null" direction.

So far, we've treated our system as a "black box," only caring about the inputs and outputs. This is the transfer function perspective. But modern control theory invites us to open the box and look at the machinery inside. This is the ​​state-space​​ view, where we describe the system's internal state, $\mathbf{x}(t)$, with a set of first-order differential equations:

Here, the matrix $A$ governs the internal dynamics, $B$ shows how inputs drive the state, $C$ shows how the state produces the output, and $D$ is a direct "feedthrough" from input to output.

How do we find zeros in this picture? A wonderful mathematical object called the ​​Rosenbrock System Matrix​​ comes to our aid:

Look at this! In one elegant package, we have the internal dynamics ($A$), the input mapping ($B$), the output mapping ($C$), and the direct path ($D$). This matrix represents the entire system. And what are the transmission zeros? They are the values of $s$ for which this grand system matrix itself loses rank!

For a "minimal" system (one with no redundant, hidden parts), the zeros found using this internal state-space method are exactly the same as the transmission zeros we found from the external input-output transfer function. This is a beautiful piece of unity. It tells us that our external observation of signal blocking is a direct consequence of a specific degeneracy in the system's internal structure at that frequency. Sometimes, a system might not have any such frequency where it blocks a signal; in that case, the determinant of its Rosenbrock matrix is a constant, and it has no finite transmission zeros.

Now we come to a point that is truly profound. As a control engineer, you have a powerful toolkit. One of the crown jewels of control theory is ​​pole placement​​. If a system is "controllable," you can use state feedback—measuring the internal state $\mathbf{x}$ and feeding it back into the input—to move the system's poles anywhere you like. You can take an unstable system and make it stable. You can make a sluggish system respond lightning-fast. The poles are like puppets on your strings.

But what about the zeros?

Here is the kicker: ​​Transmission zeros are invariant under state feedback​​. You can't move them. No matter how clever your feedback law $u = -Kx + r$, the zeros of the system from the new reference input $r$ to the output $y$ remain exactly where they were in the original open-loop system. They are a fundamental, baked-in property of how the actuators and sensors are connected to the system's dynamics, defined by the matrices $A$, $B$, and $C$.

Think about that. Zeros represent an intrinsic limitation on the performance of a system. If you have a "bad" zero, you are stuck with it. It is an unmovable obstacle, a fundamental constraint imposed by the physics of the system's construction. You can't just "control" your way out of it. This tells us that designing a system is not just about the controller; it's about the physical plant itself. The placement of sensors and actuators fundamentally determines these performance-limiting zeros.

So, what do these unmovable zeros actually do? What is their physical meaning? This is where the story gets really interesting. Let's ask a strange question: What would it take to force the system's output to be identically zero for all time, $y(t) \equiv 0$? To do this, you'd need to supply a very special input $u(t)$ that continuously works to cancel out any output.

The dynamics of the system's internal states $\mathbf{x}(t)$ while this zero-output condition is being maintained are called the ​​zero dynamics​​. And the remarkable connection is this: the eigenvalues that govern the stability of these internal zero dynamics are precisely the system's transmission zeros.

Now we can see why some zeros are "bad." If a system has a zero in the right-half of the complex plane (RHP)—say, at $s=3$—it means its zero dynamics are unstable. To force the output to zero, the internal states of the system (and the input you need to supply) must grow exponentially! The system is fighting you every step of the way. Such systems are called ​​non-minimum phase​​.

The most famous calling card of a non-minimum phase system is ​​initial undershoot​​ or ​​inverse response​​. Imagine you want the system's output to go from 0 to a positive value of 1. You give it a step input. Because of the RHP zero, the output first moves in the wrong direction—it goes negative—before eventually turning around and settling at the correct final value.

Why? Think of parking a very long fire truck. To get the back end of the truck to swing right into the parking space, you first have to turn the steering wheel to the left, causing the front cab to move left. You must initiate a move in the opposite direction to achieve your final goal. The RHP zero imposes a similar dynamic. The system must internally reconfigure itself in a way that initially seems counterproductive to produce the desired final output. This behavior is a direct, physical manifestation of those unstable zero dynamics—the ghost in the machine dictated by the location of the transmission zero.

Now that we have grappled with the mathematical machinery of transmission zeros, it is time to ask the question that truly matters: What are they for? Are they merely ghosts in the machine, phantoms of our linear algebra, or do they correspond to something real and tangible? As is so often the case in science, the answer is a resounding "both." Transmission zeros are at once the source of profound, unbreakable limitations and the secret ingredient in some of our most elegant engineering designs. They are a unifying thread that ties together the design of an audio filter, the flight dynamics of an advanced aircraft, and the fundamental rules of stability itself.

Perhaps the most direct and intuitive way to understand a transmission zero is to see it as a point of perfect deafness. Imagine you are piloting a sophisticated Vertical Take-Off and Landing (VTOL) aircraft, a machine with multiple inputs (like the thrust of different rotors) and multiple outputs (like its vertical speed and pitch angle). You would expect that pushing the control sticks would, well, do something. But what if you found a peculiar combination of commands—a specific wiggle of the controls at a particular frequency—that produced absolutely no response from the aircraft? The engines would whine, but the craft would remain stubbornly unperturbed.

This is not a malfunction. It is a transmission zero in action. For a multi-input, multi-output (MIMO) system, a transmission zero at a frequency $s=z$ corresponds to the existence of a special input direction that the system completely blocks. If an input signal of that frequency and direction is applied, the output is identically zero. Mathematically, the transfer function matrix $G(s)$ loses rank at $s=z$, meaning there is a non-zero input vector $u$ for which $G(z)u = 0$. In our VTOL example, a specific ratio of sinusoidal commands to the port and starboard rotors at the zero's frequency would result in their effects on vertical velocity and pitch rate perfectly cancelling each other out, leading to no net motion. This is a beautiful, physical manifestation of a zero: it is a specific question you can ask the system to which its answer is complete silence.

Nature may present us with these blind spots, but engineers have learned to create them by design. If a zero is a frequency where a signal is perfectly blocked, why not place these "perfect nulls" exactly where we want them? This is the central idea behind some of the most powerful signal processing filters.

Consider the challenge of designing a high-fidelity audio system. You want to separate different parts of the sound spectrum with surgical precision. A lowpass filter, for instance, should allow all the bass frequencies through unharmed while completely eliminating high-frequency hiss. A simple filter might roll off the high frequencies gradually, but an ideal "brick-wall" filter would have an infinitely sharp cutoff. How can one achieve such a thing?

The answer lies in building a filter that has transmission zeros right in the frequency band you want to eliminate (the "stopband"). Elliptic (Cauer) filters and Chebyshev Type II (inverse Chebyshev) filters are masterpieces of this kind of engineering. Their transfer functions are explicitly constructed to have zeros on the $j\Omega$ axis, corresponding to real frequencies. At each of these zero-locations, say $\Omega_i$, the magnitude of the filter's frequency response $|H(j\Omega_i)|$ is precisely zero. The attenuation is theoretically infinite.

The genius of this design is how the locations of these zeros are determined. They are not placed randomly. For an $n$-th order Chebyshev Type II filter, for example, the zero locations are derived directly from the roots of the $n$-th order Chebyshev polynomial, a famous mathematical object known for its unique "equiripple" properties. By applying a clever frequency transformation, $x = \Omega_s/\Omega$, the stopband frequency range $[\Omega_s, \infty)$ is mapped onto the interval $(0, 1]$, where the Chebyshev polynomial's zeros live. This allows an engineer to calculate the exact frequencies where the filter response will be annihilated, creating the steep, equiripple stopband that is the filter's hallmark. Here, the transmission zero is not a curiosity to be analyzed; it is a brick to be laid in the foundation of the design.

So far, we have seen zeros as either natural phenomena or useful design tools. But they also play a more severe role: they are the arbiters of what is fundamentally possible in feedback control.

The goal of feedback control is to make a system behave as we wish. We measure the output, compare it to a desired value, and use the error to calculate a corrective input. A central question is "pole placement": can we, through feedback, place the poles of the closed-loop system anywhere we want to achieve a desired stability and response speed?

The answer depends critically on what information the controller has. If a controller has access to the full internal state of the system ($x$), then for a controllable system, the answer is a resounding yes. We can place the poles anywhere. However, we rarely have such a luxury. We can typically only measure the system's outputs ($y$). And here, transmission zeros reveal their restrictive power.

A transmission zero represents an inherent loss of information between the internal state and the output. If a system has a transmission zero at $s=z$, it is impossible to place a closed-loop pole at $s=z$ using only output feedback. The zero acts as an impenetrable shield. At that specific complex frequency, the output is blind to a certain mode of the system's internal dynamics, and no amount of feedback based on that output can influence that mode. This is a profound limitation. It tells us that the very act of measurement can place fundamental constraints on our ability to control.

If we cannot break the rules imposed by zeros, can we at least bend them? The answer, happily, is yes. The location of a transmission zero is not always an immutable property of the physical plant alone; it is a property of the plant and the way we observe it.

Imagine a system with two available sensors. We can choose to use one, the other, or a weighted combination of both. It turns out that this choice of "sensor selection" can change the location of the system's transmission zeros! By simply adjusting how we combine our measurements, we can take a problematic zero—for instance, a zero at $s=0$ that might interfere with steady-state performance—and actively move it to a less troublesome location in the complex plane. This transforms the zero from a fixed constraint into a design parameter we can manipulate.

This idea becomes even more critical when we encounter "non-minimum phase" systems. These are systems that possess transmission zeros in the unstable right-half of the complex plane (RHP). The presence of even a single RHP zero has dramatic consequences. It often introduces an unnerving "wrong-way" effect: to get the output to go up, the system must first dip down. Think of trying to move the top of a balanced broomstick to the right; you must first move your hand to the left. This initial inverse response makes control exceptionally difficult and sluggish.

Sometimes, a system that is perfectly well-behaved can be pushed into this non-minimum phase regime by a simple change in a physical parameter. For instance, increasing the coupling gain between two channels in a process can cause a stable transmission zero to cross the imaginary axis and enter the RHP, fundamentally changing the system's character from easy to difficult to control.

Why are these RHP zeros so pernicious? The deepest reason lies in their effect on system stability, which can be visualized through the lens of the generalized Nyquist stability criterion. Think of this criterion as tracking the phase of a special function, $f(s) = \det(I+L(s))$, as $s$ travels around a contour enclosing the entire RHP. The number of times the plot of $f(s)$ encircles the origin tells us if the closed-loop system is stable.

An RHP zero in the plant $G(s)$ imparts a non-minimum-phase character to the loop transfer matrix $L(s)$. This character manifests as an additional, unavoidable phase lag that grows with frequency. This extra lag bends and twists the Nyquist plot of $f(s)$, making it far more likely to loop around the origin in the clockwise ("unstable") direction.

Furthermore, there is a fatal attraction between high-gain feedback and RHP zeros. As one increases the feedback gain, the closed-loop poles of the system are forced to migrate towards the open-loop transmission zeros. If one of these zeros is in the RHP, a closed-loop pole will inevitably be drawn toward it. A pole in the RHP means the system is unstable. Therefore, a system with an RHP zero cannot be controlled with arbitrarily high gain; it is guaranteed to become unstable. This is not a failure of controller design; it is a fundamental law, a direct consequence of the zero's location.

From the silent spot in a drone's control to the brick-wall of an audio filter, from the subtle limits of feedback to the violent onset of instability, the transmission zero reveals itself as a concept of remarkable power and unity. It shows us, in sharp relief, the beautiful and intricate dance between what a system is, how we see it, and the ultimate boundaries of what we can make it do.