Describing Function Analysis

While linear systems offer mathematical elegance, the real world—from aircraft controls to biological cells—is inherently nonlinear. A critical challenge in these systems is understanding and predicting self-sustained oscillations, or limit cycles, which can range from a minor nuisance to a cause of catastrophic failure. How can we analyze the stability of a system that defies simple linear description? Describing Function Analysis provides a powerful and intuitive engineering approach to tackle this problem. It is a heuristic method that cleverly approximates a system's nonlinear behavior, making it tractable for analysis.

This article delves into this essential technique across two main chapters. First, we will explore the ​​Principles and Mechanisms​​, uncovering the core idea of harmonic balance, the crucial filter hypothesis, and the graphical method that allows us to predict the amplitude and frequency of limit cycles. Then, we will journey through its diverse ​​Applications and Interdisciplinary Connections​​, seeing how the method is used to diagnose issues in control engineering, design robust systems, and even model the rhythmic behavior of biological circuits.

The world is stubbornly, beautifully, and often frustratingly nonlinear. While our textbooks are filled with the elegant mathematics of linear systems—where cause and effect are neatly proportional—the reality of swinging pendulums, saturating amplifiers, and sticky valves defies such simple description. So, how do we analyze a system that contains a mix of the linear and the nonlinear? How do we predict, for instance, the persistent, unwanted oscillations known as ​​limit cycles​​ that can plague everything from aircraft control systems to robotic arms?

We do it with a clever, powerful, and deeply intuitive piece of engineering reasoning: the ​​describing function method​​. It's not an exact mathematical proof in the way a mathematician would demand, but rather a brilliant piece of physical intuition that allows us to make astonishingly accurate predictions. It's the art of making a calculated, justifiable approximation—of pretending the nonlinear world is a bit more linear than it really is.

Imagine a simple feedback loop, the workhorse of control engineering. It consists of a well-behaved linear component, like a motor with its dynamics described by a transfer function $G(s)$, and a "wild" nonlinear component, like an amplifier that hits a limit or a valve that sticks. This setup, a linear system in feedback with a single static nonlinearity, is known as a ​​Lur'e system​​.

The full dynamics of this loop are often impossible to solve with pen and paper. The key idea of describing function analysis is this: what if we could find an approximate linear "gain" for the nonlinear part? If we could do that, the whole loop would become approximately linear, and we could use all the powerful tools of linear systems analysis to understand its behavior.

But how can you assign a single gain to a component whose very nature is that its "gain" changes depending on the input? The answer lies in focusing on the specific phenomenon we're interested in: a steady, sustained oscillation.

Let's assume such an oscillation exists. This means the signal flowing into our nonlinear element is periodic. For many systems, this oscillation will be reasonably smooth and sinusoidal. Let's say the input to the nonlinearity is $e(t) = A \sin(\omega t)$. The output will also be periodic with the same frequency $\omega$, but it will be distorted. A sine wave goes in; a square wave, a clipped sine wave, or some other complex shape comes out.

This distorted output signal can be decomposed, using Fourier's magical series, into a sum of sine waves: a ​​fundamental​​ component at the original frequency $\omega$, and a series of ​​higher harmonics​​ at frequencies $2\omega$, $3\omega$, $5\omega$, and so on.

Here is the crucial leap of faith, the "trick" that makes the whole method work. The distorted signal from the nonlinearity's output is fed back into the linear part of our system, $G(s)$. Most physical systems that we want to control—systems with mass, inertia, or thermal capacity—naturally act as ​​low-pass filters​​. They respond readily to low-frequency inputs but become sluggish and unresponsive to high-frequency inputs.

This means that as the signal travels through the linear block $G(s)$, the higher harmonics ($3\omega$, $5\omega$, etc.) are much more attenuated than the fundamental frequency $\omega$.

Let's make this concrete. Imagine our nonlinearity is an ideal relay that outputs $+M$ or $-M$, and our linear plant is $G(s) = K / (s(Ts+1))$. If a sine wave $e(t) = E \sin(\omega t)$ enters the relay, the output is a square wave. The Fourier series for this square wave tells us that the amplitude of the third harmonic ($3\omega$) is exactly one-third the amplitude of the fundamental ($\omega$) at the relay's output. However, after passing through the plant, the situation changes. For typical values like $\omega = 1$ rad/s and $T = 0.5$ s, the plant's gain at $3\omega$ is much smaller than at $\omega$. A direct calculation shows that the ratio of the third harmonic's amplitude to the fundamental's amplitude at the plant's output is only about $0.0689$. The third harmonic has been suppressed by over 93%! The fifth harmonic would be suppressed even more.

This is the ​​filter hypothesis​​ in action. We are justified in ignoring the higher harmonics because the linear system itself washes them out. The signal that emerges from the linear block and feeds back to the input of the nonlinearity is, once again, almost a pure sine wave. This creates a self-consistent picture: a sine wave goes in, a distorted wave comes out, the linear system filters it, and a sine wave emerges. We are analyzing the behavior of the fundamental frequency and assuming the rest is negligible noise.

By agreeing to ignore the higher harmonics, we can now characterize our nonlinear element in a wonderfully simple way. We only care about what it does to the fundamental frequency component. For a sinusoidal input $e(t) = A \sin(\omega t)$, we look at the fundamental component of the output. This component will have some amplitude and some phase shift relative to the input. The ratio of the complex phasor of the output's fundamental to the phasor of the input is what we call the ​​describing function​​, $N(A)$.

$N(A) = \frac{\text{Phasor of Output Fundamental}}{\text{Phasor of Input Sinusoid}}$

It's a "gain," but it's a special kind of gain: it depends on the amplitude $A$ of the input signal. For a small input signal, the describing function might have one value; for a large signal, it will have another.

• For an ​​ideal relay​​ that switches between $\pm M$, the output's fundamental is always in phase with the input, and its amplitude is inversely proportional to the input amplitude $A$. Its describing function is real and positive: $N(A) = \frac{4M}{\pi A}$.
• For a nonlinearity with a ​​dead zone​​, like a relay that only activates when the input exceeds a certain threshold $d$, the describing function only becomes non-zero for $A > d$. Its formula might be $N(A) = \frac{4M}{\pi A} \sqrt{1 - (d/A)^2}$.

The key is that for any given nonlinearity, we can compute its describing function $N(A)$. We have effectively "linearized" the nonlinear element into a block with an amplitude-dependent gain.

Now our feedback loop looks beautifully simple. We have a linear block $G(s)$ in a negative feedback loop with another block that has a gain $N(A)$. From linear systems theory, we know that such a loop will be on the verge of instability—capable of sustaining a pure oscillation—if its loop gain is equal to $-1$.

The loop gain of our approximated system is $G(j\omega)N(A)$. So, the condition for a self-sustained oscillation, or limit cycle, is:

$G(j\omega)N(A) = -1$

This beautifully simple equation is the heart of the method. It's called the ​​harmonic balance equation​​. It's really two equations in one, one for the magnitude and one for the phase:

1. ​​Phase Condition:​​ $\arg(G(j\omega)) + \arg(N(A)) = -180^\circ$
2. ​​Magnitude Condition:​​ $|G(j\omega)| |N(A)| = 1$

These two equations give us two unknowns to solve for: the limit cycle's amplitude $A$ and its frequency $\omega$. For many common nonlinearities (like relays and saturation), the describing function $N(A)$ is real and positive, so $\arg(N(A)) = 0$. In this common case, the phase condition simplifies to finding the frequency $\omega_{lc}$ where the linear plant's phase shift is exactly $-180^\circ$. Once we have that frequency, we plug it into the magnitude condition to solve for the amplitude $A_{lc}$ that makes the loop gain exactly one.

The harmonic balance equation $G(j\omega) = -1/N(A)$ gives rise to a powerful graphical interpretation.

1. We can draw the ​​Nyquist plot​​ of our linear system, $G(j\omega)$. This is a curve in the complex plane that shows the gain and phase shift of the linear system at every frequency $\omega$. This curve represents the left-hand side of our equation.

2. We can also plot the term $-1/N(A)$ on the same plane. Since $N(A)$ depends on amplitude $A$, this term is not a single point (like the classic critical point $-1$ in linear stability analysis) but a ​​locus​​ or a curve traced out as $A$ varies from $0$ to $\infty$. This is the "critical point" for our nonlinear system, and it moves!

A limit cycle is predicted to exist if and only if these two curves intersect.

​​An intersection of $G(j\omega)$ and $-1/N(A)$ means we have found a pair $(\omega, A)$ that satisfies the harmonic balance equation.​​ The frequency of the limit cycle is the value of $\omega$ on the Nyquist curve at the intersection point, and the amplitude of the limit cycle is the value of $A$ corresponding to that point on the $-1/N(A)$ locus.

This graphical method is incredibly insightful. For some systems, the curves might intersect at more than one point, predicting the existence of multiple limit cycles with different amplitudes and frequencies. Or they might not intersect at all, suggesting no limit cycle will form.

Finding an intersection is not the end of the story. A predicted limit cycle can be ​​stable​​ (if perturbed, the system returns to the oscillation) or ​​unstable​​ (if perturbed, the system either collapses to a stable point or flies off to another state). An unstable limit cycle acts as a "watershed" in the state space and is not typically observed in practice.

There is a simple graphical rule of thumb (related to the Loeb criterion) to assess stability. Look at the direction of increasing amplitude $A$ along the $-1/N(A)$ locus. A limit cycle is typically stable if, at the intersection point, the $-1/N(A)$ curve crosses the $G(j\omega)$ curve from the "unstable" region (the region encircled by the Nyquist plot) to the "stable" region (the non-encircled region) as amplitude $A$ increases. This gives us a way not just to predict oscillations, but to predict which ones we are likely to actually see.

The describing function method is a powerful engineering tool, but it is an approximation, a map of a territory. And like any map, it has its limitations. It's crucial to know where the map is no longer reliable.

• ​​Complex Behaviors:​​ The method's very foundation is the assumption of a single-frequency sinusoidal oscillation. It is therefore fundamentally blind to more complex dynamics. It cannot predict ​​subharmonic oscillations​​ (where the system oscillates at a fraction, like $1/2$ or $1/3$, of some internal driving frequency) or the intricate, non-repeating dance of ​​quasi-periodic​​ and ​​chaotic​​ motions.

• ​​Strongly Nonlinear Regimes:​​ The method can also be misleading when analyzing the stability of an equilibrium point where the nonlinearity is "flat," i.e., its derivative is zero. In such a "strongly nonlinear" case, a naive application of the describing function can yield an incorrect prediction about the system's stability near the equilibrium.

Perhaps the most important lesson is understanding the place of the describing function method in the engineer's toolkit. It is a ​​heuristic​​, not a mathematical proof. It provides candidates for limit cycles, not guarantees.

This becomes crystal clear when we compare it to rigorous mathematical tools like the ​​Popov criterion​​ or the ​​circle criterion​​. These methods provide sufficient conditions for absolute stability. If the Popov criterion is satisfied for a system, it proves, with mathematical certainty, that the system is globally asymptotically stable. This means all trajectories converge to the origin, and therefore, no limit cycles can possibly exist.

What if you have a system where the describing function method predicts a limit cycle, but the Popov criterion proves the system is absolutely stable? Who do you believe? You always believe the rigorous proof. The Popov criterion's conclusion trumps the heuristic prediction. The limit cycle predicted by the describing function is a ​​"false positive"​​—an artifact of neglecting the higher harmonics which, in this case, were essential for ensuring stability.

This doesn't make the describing function method useless. Far from it. Rigorous stability tests are often very conservative; they may fail to prove stability for a system that is, in fact, perfectly stable. In these inconclusive cases, the describing function method shines. It acts as an invaluable investigative tool. It provides a concrete hypothesis—a potential limit cycle at a specific amplitude and frequency—that can guide further analysis, simulation, and physical experiments. It provides a workflow for design when rigorous proofs are silent.

In the end, understanding the describing function method is to understand a core tenet of engineering thought: the intelligent use of approximation. It's about knowing how to simplify a problem to make it tractable, understanding the assumptions behind that simplification, and respecting the boundaries where the approximation breaks down. It is a tool that, when used with wisdom, transforms the intimidating complexity of the nonlinear world into a landscape we can navigate and engineer.

Having journeyed through the principles and mechanisms of describing function analysis, we have, in essence, learned the grammar of a new language. It is a language that allows us to have a conversation with nonlinear systems—systems that are notoriously stubborn and often refuse to speak the simple, linear tongue we are most comfortable with. But learning grammar is one thing; reading poetry is another. Where does this new language come to life? Where does describing function analysis move from a blackboard exercise to a powerful tool for discovery and invention?

The answer, it turns out, is everywhere. It finds its home in the humming of our machines, the precision of our robots, and even in the silent, intricate dance of life itself. In this chapter, we will explore this vast landscape, seeing how our analytical tool becomes a detective's magnifying glass, an architect's blueprint, and a biologist's microscope.

Most often, the engineer's first encounter with a limit cycle is as an unwanted guest. It's the mysterious hum in an audio amplifier, the jitter in a robotic arm, or the "chatter" of a valve in a chemical plant. These self-sustaining oscillations are born from the marriage of feedback and nonlinearity. Describing function analysis is our primary method for predicting their arrival and characterizing their nature. Let's meet some of the usual suspects.

• ​​The All-or-Nothing World of Relays:​​ The simplest form of control is a switch: on or off. Your home thermostat is a classic example. When it's too cold, the heat is fully on; when it's warm enough, it's fully off. These "relay" controllers are simple, robust, and inexpensive, making them ubiquitous in industrial control. However, this simplicity comes at a cost. The abrupt switching action is a harsh nonlinearity that can easily provoke a system into a sustained oscillation, or limit cycle. Using describing function analysis, an engineer can look at the linear part of their system—say, a motor or a heating element represented by its transfer function $G(s)$—and predict with remarkable accuracy whether a relay controller will cause it to oscillate, and if so, at what frequency and amplitude. This foreknowledge is crucial; it's the difference between designing a stable temperature controller and a system that endlessly cycles on and off, wasting energy and wearing out components.

• ​​Hitting the Wall: Saturation:​​ Nothing in the real world is infinite. Amplifiers have a maximum voltage, motors have a maximum torque, and valves can only open so far. This fundamental limitation is known as "saturation." When we command a system to perform an action that exceeds its physical limits, the actuator saturates, and the feedback loop's behavior changes dramatically. The system temporarily stops responding as commanded. This nonlinearity, like the relay, can lead to limit cycles, especially in high-performance systems that are pushed close to their limits. Describing function analysis allows an engineer to quantify this risk. By analyzing the system, they can predict the conditions under which the unavoidable saturation of a component will trigger an unwanted, high-frequency oscillation.

• ​​The Slop and Stickiness of the Mechanical World: Backlash and Dead-Zones:​​ In the world of precision mechanics—robotics, servomechanisms, machine tools—our enemies are often friction and looseness. A "dead-zone" is a region of unresponsiveness; imagine a valve that requires a certain amount of pressure before it even begins to open. "Backlash" is the familiar "slop" or "play" in a set of gears; when the driving gear changes direction, it turns for a moment before it re-engages the driven gear. Both are insidious nonlinearities. Backlash is particularly interesting because it not only reduces the effective gain but also introduces a phase lag—it takes time to cross the gap. The describing function for backlash is therefore a complex number, beautifully capturing both the gain and phase effects in a single mathematical object. By plotting the frequency response of the linear system against the negative reciprocal of this complex describing function, an engineer can predict the precise frequency and amplitude of the "jitter" that backlash will introduce into a high-precision positioning system.

Prediction is powerful, but the true goal of engineering is design. Describing function analysis elevates from a diagnostic tool to a creative one, allowing us to build systems that are not only functional but also robust to the inescapable nonlinearities of the real world.

Imagine we are designing a control system and we know a saturation element is present. The describing function method gives us a "keep-out zone" on our analysis plots, represented by the locus of $-1/N(A)$. Our job as a designer is to shape the linear part of our system, $G(s)$, so that its frequency response, the Nyquist plot, gives this dangerous region a wide berth. This principle transforms how we approach controller design. For instance, when adding a compensator to improve tracking accuracy, we are in a trade-off. We can increase the gain to make the system more responsive, but this pushes the Nyquist plot outwards, closer to the $-1/N(A)$ locus and the risk of a limit cycle. Describing function analysis allows us to calculate the maximum achievable performance right up to the boundary of instability, letting us squeeze every drop of performance out of a system without pushing it over the edge.

Furthermore, it allows us to foresee the subtle, sometimes counter-intuitive, interactions between our linear design choices and the system's nonlinear behavior. Suppose we have a servomechanism plagued by backlash-induced oscillations. We might decide to add a "lead compensator," a standard technique to make the system respond faster. But what does this do to the limit cycle? The compensator reshapes the system's entire frequency response. This change alters the point of intersection with the $-1/N(A)$ locus, meaning our well-intentioned modification could change the limit cycle's frequency, increase its amplitude, or even create one where none existed before! Describing function analysis is the tool that lets us anticipate these consequences before we ever build the hardware.

In a fascinating turn of the tables, sometimes an oscillation is not a bug, but a feature. Self-oscillating circuits are the heart of signal generators, and a deliberately induced high-frequency oscillation, known as "dither," can be used to overcome static friction in mechanical systems. In such cases, the goal is not to eliminate the limit cycle, but to create one with a specific, desired amplitude and frequency. Here, describing function analysis becomes a true synthesis tool. We can use it to choose the parameters of our controller—for example, the gains of a PI controller—to force the intersection of $G(j\omega)$ and $-1/N(A)$ to occur at exactly the point that gives us the oscillation we want.

Perhaps the most profound testament to a scientific principle is its universality—its ability to describe phenomena in vastly different fields. The logic of feedback and nonlinearity is not confined to machines built of metal and silicon; it is the logic of life itself.

Consider the field of synthetic biology, where scientists engineer new biological circuits inside living cells. A common motif is a "transcriptional cascade," where one gene produces a protein that, in turn, activates or represses another gene. The response of a gene to its activating protein is not linear; as the protein concentration increases, the rate of gene expression eventually saturates, much like an amplifier hitting its voltage limit. This is a fundamental nonlinearity baked into the machinery of life.

If a biologist engineers a feedback loop—say, a protein that ultimately represses its own production—they have built a nonlinear feedback system. Will this circuit be stable, or will it oscillate, producing proteins in rhythmic pulses? This is not an academic question; such oscillations are the basis for biological clocks, like our own circadian rhythm. Remarkably, the very same describing function analysis we use for servomechanisms can be adapted to analyze these genetic circuits. By modeling each gene's saturating response with a describing function and the protein production and degradation as a linear filter, a systems biologist can predict whether their engineered genetic circuit will oscillate. The mathematics that explains the chatter of a relay can also shed light on the rhythm of a cell.

As with any powerful tool, it is essential to understand its limitations. Describing function analysis is a brilliant approximation, not an exact science. Its derivation rests on the assumption that the input to the nonlinearity is roughly sinusoidal and that the system effectively filters out higher harmonics. When these assumptions break down, so can the accuracy of our predictions.

A classic example is the problem of "integrator windup" in control systems. When a controller with integral action faces a saturated actuator, the integrator, unaware that its commands are being ignored, can accumulate a massive error value. This "windup" can lead to huge overshoots and poor performance once the system comes out of saturation. This is a dynamic effect, a form of "memory" in the controller state, that the static, memoryless gain of the describing function cannot fully capture. Understanding this limitation has led to the development of sophisticated "anti-windup" techniques that go beyond the describing function framework.

But this limitation does not diminish the value of our tool. It simply delineates its boundaries. Describing function analysis provides an unparalleled intuitive window into the behavior of a vast class of nonlinear systems. It gives us a "first-order" understanding, a brilliant glimpse into a complex world. It stands as a testament to the power of a good approximation, reminding us, in the spirit of Feynman, that it is often through simple, elegant ideas that we gain the deepest insights into the workings of the universe, from the machines we build to the very cells we are made of.