Bilinear Transform

Bridging the gap between the continuous world of analog systems and the discrete logic of digital computers is a central challenge in modern engineering. How can we translate a time-tested analog filter or controller design into a digital algorithm without losing its essential characteristics, most importantly, its stability? The bilinear transform emerges as a powerful and elegant solution to this problem, providing a robust mathematical bridge between these two domains. This article delves into the core of the bilinear transform. The following chapters will uncover its mathematical origins, explore its unparalleled ability to preserve stability, and dissect its inherent side effect—frequency warping. We will then see this theory put into practice, examining how engineers leverage the transform and the clever technique of pre-warping to design precise digital filters and high-performance digital control systems.

Imagine you are trying to describe a flowing river to a friend who can only perceive the world in a series of snapshots. You can't show them the continuous motion, only still frames. How can you capture the essence of the flow—its speed, its direction, its accumulation over time—using just these discrete moments? This is the fundamental challenge faced by engineers and scientists when they try to implement ideas from the continuous world of physics and analog electronics onto the discrete, step-by-step logic of a digital computer. We need a bridge, a translator, between the smooth language of calculus and the staccato rhythm of computation. The ​​bilinear transform​​ is one of the most elegant and powerful of these bridges.

Let's start with one of the most fundamental ideas in calculus: integration. In the world of signals and systems, an integrator is a device that accumulates a quantity over time. Its output at any moment is the total sum of everything that has gone in up to that point. In the continuous domain, described by the Laplace transform, this simple concept is represented by the transfer function $H_a(s) = \frac{1}{s}$.

How do we teach a computer to do this? A computer takes snapshots at regular intervals, a time $T$ apart. A beautifully simple and effective idea is the ​​trapezoidal rule​​. To approximate the total accumulated area, at each step we add a small trapezoid whose area is based on the average of the current input value and the previous one, multiplied by the time step $T$. This simple geometric idea translates into a precise computational recipe, a ​​difference equation​​:

$y[n] = y[n-1] + \frac{T}{2} \left( x[n] + x[n-1] \right)$

Here, $y[n]$ is the new accumulated value (the output), $y[n-1]$ is the previous value, and $x[n]$ and $x[n-1]$ are the current and previous input values. This equation is something a computer can execute perfectly. It's a digital recipe for integration.

By applying the Z-transform, the discrete-time equivalent of the Laplace transform, to this recipe, we get the digital transfer function for our integrator:

$H(z) = \frac{\frac{T}{2}(1 + z^{-1})}{1 - z^{-1}} = \frac{\frac{T}{2}(z+1)}{z-1}$

This is the digital blueprint for an integrator. If we now say that our analog blueprint ($1/s$) must be equivalent to our digital blueprint ($H(z)$), we uncover the secret of the bilinear transform. It's the substitution rule that connects the two worlds:

$s = \frac{2}{T} \frac{z-1}{z+1}$

This is our universal translator, a dictionary that lets us convert any design from the continuous $s$-plane to the discrete $z$-plane.

A good translation must preserve not just the words, but the essential meaning. In engineering, perhaps the most critical meaning of all is ​​stability​​. A stable system is one that doesn't "blow up"—its response to a finite input remains bounded. An unstable system is a disaster waiting to happen. If our translation from analog to digital turns a stable design into an unstable one, it's worse than useless.

This is where the true beauty of the bilinear transform shines. In the continuous world, a system is stable if all its characteristic "modes," or poles, lie in the left half of the complex $s$-plane, where their real part is negative ($\Re(s) \lt 0$). In the digital world, stability requires all poles to lie inside the unit circle in the complex $z$-plane, where their magnitude is less than one ($|z| \lt 1$).

The bilinear transform performs a remarkable feat: it maps the entire left-half of the $s$-plane perfectly and exclusively into the interior of the unit circle in the $z$-plane. It’s not an approximation; it's a complete, one-to-one conformal mapping. Any stable analog pole you start with is guaranteed to land inside the unit circle, resulting in a stable digital system. For instance, a crucial stabilizing pole in an analog controller at $s_p = -2000$ will be mapped, with a sampling period of $T = 0.5 \text{ ms}$, to a digital pole at $z_p = 1/3$. This is comfortably inside the unit circle, and stability is preserved.

This property is not to be taken for granted. Other "simpler" methods can fail spectacularly. Consider a pure oscillator, like a perfect pendulum, whose poles lie exactly on the imaginary axis of the $s$-plane—the very boundary of stability. If we use a different method like the Forward Euler transform, the resulting digital poles are pushed outside the unit circle, turning a perfectly balanced oscillator into an unstable, exponentially growing one. The bilinear transform, however, correctly maps the imaginary axis onto the unit circle itself, preserving the marginal stability of the oscillator. It understands the profound importance of the stability boundary and respects it perfectly.

This stability-preserving property is so fundamental that it extends beyond simple transfer functions to the general state-space representation of systems. The matrix transformation, $A_d = (I - A\frac{T}{2})^{-1}(I + A\frac{T}{2})$, is a famous mathematical operation known as the ​​Cayley transform​​. For centuries, mathematicians have known that the Cayley transform maps the left-half complex plane to the open unit disk. Here we see a beautiful piece of pure mathematics providing the rigorous foundation for a critical engineering tool, guaranteeing that the stability of a complex system, captured by the eigenvalues of its state matrix $A$, is preserved in its digital counterpart $A_d$. This is the unity of science and engineering at its finest.

This perfect mapping of stability regions, however, comes at a curious price. The translation of frequency is not linear. Imagine you have an infinitely long rubber band representing the analog frequency axis, from $\Omega = -\infty$ to $\Omega = +\infty$. The bilinear transform forces you to map this entire infinite length onto a finite loop: the unit circle, which represents the digital frequency range from $\omega = -\pi$ to $\omega = \pi$.

To achieve this, you have to stretch the rubber band non-uniformly. Near the center (low frequencies), the mapping is almost one-to-one, and the relationship is nearly linear: $\Omega \approx \omega/T$. But as you move toward the ends of the digital frequency range ($\omega \to \pm\pi$), you have to stretch the rubber band more and more violently to cover the rest of the infinite analog axis. This non-linear stretching is called ​​frequency warping​​.

The exact relationship, derived by substituting $s = j\Omega$ and $z = \exp(j\omega)$ into our transformation rule, is:

$\Omega = \frac{2}{T} \tan\left(\frac{\omega}{2}\right)$

The tangent function reveals the non-linear stretch. As the digital frequency $\omega$ approaches the Nyquist limit of $\pi$, $\omega/2$ approaches $\pi/2$, and its tangent shoots off to infinity, cramming all high analog frequencies into a small region of the digital spectrum.

This is not just a mathematical curiosity; it has real-world consequences. Suppose you design a simple analog RC low-pass filter to have a cutoff frequency at $\Omega_c = 1/(RC)$. If you convert this to a digital filter using the bilinear transform, the new digital cutoff frequency $\omega_c'$ will not be at the location you might naively expect. Instead, it will be at $\omega_c' = 2\arctan(\frac{T}{2RC})$. The frequency has been "warped" to a new location.

So, must we simply live with this warped reality? No. Clever engineers have turned this problem on its head. If the transformation is going to warp our frequencies, we can simply plan for it in advance. This technique is called ​​frequency pre-warping​​.

The logic is beautifully simple. Suppose we want our final digital filter to have a critical frequency (like a passband edge) at a specific digital frequency, $\omega_{desired}$. We know the bilinear transform will distort any analog frequency we start with. So, instead of starting with the analog frequency that seems to correspond to our target, we ask: what analog frequency, $\Omega_{pre-warped}$, when warped by the transform, will land exactly at our desired $\omega_{desired}$?

We can find this by running our warping formula in reverse:

$\Omega_{pre-warped} = \frac{2}{T} \tan\left(\frac{\omega_{desired}}{2}\right)$

We then design our initial analog filter prototype using this calculated "pre-warped" frequency. Now, when we apply the bilinear transform, the inherent non-linear warping bends our carefully chosen frequency right onto the target we wanted all along. It’s like an archer aiming high to compensate for gravity's pull on the arrow. We don't aim directly at the target; we aim where we need to for the natural laws of the process to guide our projectile to the bullseye.

This final, clever step completes the story of the bilinear transform. It is a journey from an intuitive approximation of a continuous process to a mathematically profound tool that offers a "perfect" translation of stability, but with a peculiar twist in its handling of frequency. By understanding that twist, we learn to master it, allowing us to design digital systems that meet our specifications with uncanny precision.

Now that we have acquainted ourselves with the machinery of the bilinear transform, we might ask, "What is it good for?" It is one thing to admire the cleverness of a mathematical substitution, but it is another thing entirely to see it in action, shaping the world around us. The true beauty of the bilinear transform lies not in its definition, but in its role as a master translator, a robust and elegant bridge between two fundamentally different realms: the continuous world of analog electronics and physical systems, and the discrete world of digital computers. Its applications are vast, touching everything from the music you stream to the cruise control in your car. Let us embark on a journey to explore some of these connections.

Perhaps the most common and direct application of the bilinear transform is in the design of digital filters. For decades, engineers perfected the art of analog filter design, creating a rich catalog of "prototypes"—like the Butterworth, Chebyshev, and Elliptic filters—each with specific, well-understood characteristics. When the digital revolution began, a natural question arose: must we reinvent the wheel? Or can we somehow leverage this vast analog knowledge to build their digital counterparts?

The bilinear transform provides a resounding "yes!" It allows us to take an analog filter design, described by a transfer function $H_a(s)$, and convert it into a digital filter, $H(z)$. However, this is not a simple plug-and-chug process. We must be mindful of the transform's signature effect: frequency warping. As we've seen, the transform non-linearly squeezes the infinite frequency axis of the analog world ($j\Omega$) onto the finite unit circle of the digital world. If we are not careful, a filter designed to have a cutoff at a certain frequency in the analog domain will, after transformation, have its cutoff shifted to an entirely different frequency in the digital domain.

This is where the true art of the design process reveals itself. The key insight is to work backward. Instead of designing an analog filter and seeing where the frequencies land, we start with our desired digital frequency specifications (e.g., a 3,000 Hz cutoff for an audio filter). Then, using the inverse of the warping equation, we calculate what analog frequency would warp into our desired digital frequency. This crucial first step is called ​​pre-warping​​. Once we have these pre-warped analog frequencies, we can proceed to design a standard analog filter that meets these new, adjusted specifications. Finally, we apply the bilinear transform. Because we pre-compensated for the warp, the critical frequencies of our final digital filter land exactly where we intended them to be.

The importance of this pre-warping step cannot be overstated. Imagine building a digital filter to mimic a classic analog resonant system—perhaps a synthesizer filter that gives an instrument its characteristic "voice". Such systems often have a sharp peak in their frequency response at a specific resonant frequency. If we were to naively apply the bilinear transform without pre-warping, the frequency warping effect could drastically alter this peak. The resonant frequency might shift, and the peak's sharpness and height could be reduced, fundamentally changing the character and sound of the filter. The digital version would be a pale, distorted imitation of the original. Pre-warping allows us to preserve these critical features with surgical precision. We can even calculate the exact phase deviation introduced by the warping at any frequency, giving us full command over the process.

This entire procedure uncovers a beautiful connection to a seemingly unrelated field: numerical analysis. Why this particular substitution, $s \leftrightarrow \frac{2}{T}\frac{z-1}{z+1}$? It turns out that applying the bilinear transform to a system's differential equation is mathematically equivalent to solving that differential equation using a classic numerical method known as the ​​trapezoidal rule for integration​​. This is a profound unification. It tells us that our frequency-domain tool for filter design is, from another perspective, a time-domain tool for approximating the evolution of a system from one moment to the next.

Filtering signals is a relatively passive act. But what if we want to actively control a physical system? The bilinear transform is a cornerstone of modern digital control, the technology that enables everything from factory robots to aircraft autopilots.

The workhorse of the control world is the Proportional-Integral-Derivative (PID) controller. For over a century, this analog control law has been the go-to solution for making systems behave as desired. When a PID controller needs to be implemented on a microprocessor, its continuous-time equation, rich with integrals and derivatives, must be converted into a discrete-time algorithm—a difference equation that can be executed in a loop. The bilinear transform is the perfect tool for this job. By applying the transform to the standard PID transfer function, we can systematically derive the exact coefficients for the digital implementation. This allows engineers to take time-tested tuning rules, like the famous Ziegler-Nichols method, and directly translate them into code that runs on a digital chip.

Here again, the magic of pre-warping shines, ensuring that the digital controller's performance faithfully matches its analog blueprint, especially where it counts. In control systems, a critical measure of stability and performance is the ​​phase margin​​, typically measured at the system's gain crossover frequency. A low phase margin means the system is close to instability—think of a wobbly, oscillating robot arm. An engineer might painstakingly design an analog compensator to achieve a healthy phase margin. What happens when this is converted to a digital controller? If we use the bilinear transform and pre-warp the frequency axis to match at that exact gain crossover frequency, the resulting digital controller will have the exact same phase margin as the analog design. This is a remarkable result. It means we can guarantee that our digital implementation preserves the stability characteristics of the well-analyzed analog original.

This preservation of performance is not just a frequency-domain curiosity. It has direct consequences for the system's time-domain behavior. A system's "rise time"—how quickly it responds to a command—is a key performance metric. By using pre-warping to match the frequency response at the critical crossover frequency, we also create a digital system whose step response rise time more accurately matches the original continuous-time system. In other words, matching the system's behavior in the frequency domain makes it act more like the original in the time domain.

The reach of the bilinear transform extends into the most advanced areas of control theory. Modern control often uses a "state-space" representation, a more powerful framework than simple transfer functions. When designing sophisticated controllers like the Linear Quadratic Regulator (LQR), one must discretize these state-space models. The bilinear transform provides a way to do this that is both numerically stable and highly accurate, with an error that decreases with the square of the sampling period—a property known as second-order accuracy.

Perhaps the most surprising and profound property of the bilinear transform relates to a system's steady-state behavior. A control system's "type" determines its ability to follow different kinds of commands without long-term error. For example, a "Type 1" system can perfectly track a constant velocity command, like a cruise control system maintaining a set speed on a flat road. This ability is quantified by steady-state error constants, such as the position constant ($K_p$) and velocity constant ($K_v$). One might reasonably assume that a digital approximation would introduce some small error, degrading this perfect tracking ability.

Astonishingly, this is not the case. The bilinear transform ​​perfectly preserves the steady-state error constants​​ of the original system. A digital controller derived via the bilinear transform will have the exact same $K_p$, $K_v$, and $K_a$ as its analog parent. This means that the digital system's ability to track position, velocity, and acceleration commands in the long run is identical to the original. This reveals that the transform does an exceptionally good job of capturing the low-frequency, or DC, essence of a system—the very essence that governs its steady-state performance. In this crucial aspect, the transform is not an approximation at all; it is an exact match.

This journey from filters to controllers, from simple mimicry to the preservation of profound system properties, shows the bilinear transform for what it is: an uncommonly elegant and powerful tool. Its one supposed "flaw," the frequency warping, is in fact a well-understood feature that we can master and use to our advantage. It is a beautiful compromise, a testament to the ingenuity that allows us to command the continuous, physical world using the discrete, logical language of machines.