M-circles

In control engineering, a fundamental challenge lies in predicting the final behavior of a system after feedback is applied. We design and analyze the open-loop system, but our ultimate interest is in the performance of the closed-loop system. This creates a knowledge gap: how can we intuitively and graphically understand the effect of the feedback loop without resorting to tedious point-by-point recalculations? We need a visual tool that translates the language of open-loop frequency response into the reality of closed-loop performance.

This article introduces M-circles, an elegant graphical method that serves as this very translator. It provides a direct visual link between the open-loop plot we can easily generate and the closed-loop characteristics, like gain and resonance, that we care about. You will learn how these graphical contours empower engineers to both analyze and design feedback control systems with greater insight. The discussion is structured to first build a strong theoretical foundation before moving to practical application.

The journey begins in the "Principles and Mechanisms" chapter, where we will uncover the mathematical origin of M-circles as Apollonian circles and explore how they are represented on both Nyquist and Nichols plots. Following this, the "Applications and Interdisciplinary Connections" chapter will demonstrate how this concept becomes a powerful tool for finding resonant peaks, estimating system damping, and guiding the design of controllers to meet specific performance criteria.

In the world of control systems, we live a life of indirect influence. We design and build what is called the ​​open-loop system​​—an amplifier, a motor, a chemical reactor—which we can describe with a transfer function, let's call it $L(s)$. However, our ultimate goal is to understand and predict the behavior of the final, complete system after we've wrapped a feedback loop around it. This is the ​​closed-loop system​​, whose transfer function for a standard unity feedback setup is $T(s) = \frac{L(s)}{1 + L(s)}$.

The relationship between what we build, $L(s)$, and what we get, $T(s)$, is captured in that simple fraction. But don't be fooled by its algebraic simplicity. The crucial question is: how does the behavior of the open-loop system across a range of frequencies, which we can plot and visualize, translate into the behavior of the closed-loop system? If we have a plot of the open-loop frequency response, $L(j\omega)$, can we somehow "see" the resulting closed-loop response, $T(j\omega)$, without the tedious task of recalculating everything from scratch for every frequency? We want a graphical tool, a kind of decoder ring, that lets us look at the open-loop world and immediately grasp the closed-loop reality. This is the quest that leads us to the elegant concept of M-circles.

Let's begin our journey with a simple, almost naive question. Imagine the ​​Nyquist plot​​, which is simply a drawing of the complex numbers $L(j\omega)$ in a plane for all frequencies $\omega$. Now, let's pick a number, say 2. Where in this plane are all the possible points $L(j\omega)$ that would result in a closed-loop magnitude $|T(j\omega)|$ of exactly 2? Or 0.5? Or any constant value $M$?

By definition, every point on such a contour must share the property of yielding the same closed-loop magnitude. The condition is simply $|T(j\omega)| = M$, where $M$ is a positive constant. Writing this out, we get:

Let's drop the $(j\omega)$ for a moment and just think about the complex numbers. The equation is $|L| = M |1+L|$. We can rewrite the terms inside the absolute values to make them look like distances between points:

This is a startlingly beautiful result! It says that any point $L$ that satisfies our condition must be such that its distance from the origin (the point $0$) is exactly $M$ times its distance from the "critical point" $-1$. This is a purely geometric condition.

The ancient Greeks, particularly Apollonius of Perga, studied this very problem. The locus of points for which the ratio of the distances to two fixed points (called foci) is constant is a circle, now famously known as a ​​Circle of Apollonius​​. Our two foci are the origin, $0$, and the critical point for stability, $-1$.

So, the answer to our simple question is a profound geometric truth: the loci of constant closed-loop magnitude in the open-loop plane are circles! Because they correspond to a constant magnitude $M$, these are aptly named ​​M-circles​​.

What does this family of Apollonian circles look like?

Let's consider the special case where $M=1$. Our condition becomes $|L| = |1+L|$, meaning the point $L$ must be equidistant from the origin $0$ and the point $-1$. The set of all such points is the perpendicular bisector of the line segment connecting them. This is simply the vertical line at $\text{Re}\{L\} = -1/2$.

What if $M$ gets very, very large? A huge closed-loop magnitude $|T|$ means the denominator, $|1+L|$, must be very, very small. This happens when $L$ gets extremely close to the point $-1$. As $M \to \infty$, our Apollonian circle shrinks and collapses right onto the critical point $-1$. This makes perfect physical sense; an infinite closed-loop gain corresponds to instability, which occurs when the Nyquist plot passes through $-1$.

Conversely, if $M$ is very small, approaching zero, our condition $|L| \approx 0 \cdot |1+L|$ implies that $|L|$ must be tiny. The M-circle for $M=0$ is just the origin itself.

For any other value of $M$, we get a non-degenerate circle. A bit of algebra on the equation $(x+jy) = L$ shows that the circle for a given $M \gt 0$ and $M \neq 1$ has its center at $(-\frac{M^2}{M^2-1}, 0)$ and a radius of $R = |\frac{M}{M^2-1}|$. So if we were to measure the radius $R$ of an M-circle on a Nyquist plot, we could solve for the magnitude $M$ it represents. For instance, if a system's plot was tangent to a circle of radius $R=1.8$, we could deduce the peak closed-loop gain is $M_p \approx 1.32$.

These contours give us a new way to interpret the open-loop plane. Let's say we are interested in a closed-loop gain of $+2.0$ dB. In linear terms, this is a magnitude of $M = 10^{2.0/20} \approx 1.26$. We can use the formulas above to draw a specific circle on the Nyquist plot. Any frequency for which the $L(j\omega)$ plot falls on this circle will have a closed-loop gain of exactly $+2.0$ dB. The entire plane is now filled with a family of these M-circles, each one a contour line for a specific closed-loop gain.

While the Nyquist plot and its M-circles are conceptually beautiful, engineers often prefer to work with gain in decibels (a logarithmic scale) and phase in degrees. This leads us to the ​​Nichols chart​​.

The Nichols chart is a clever transformation of the Nyquist plane. Instead of plotting the real part of $L$ versus its imaginary part, it plots the magnitude of $L$ in decibels ($20\log_{10}|L|$) on the vertical axis against the phase of $L$ in degrees ($\angle L$) on the horizontal axis. The origin of the Nyquist plot ($|L|=0$) is sent to negative infinity on the vertical axis. The unit circle of the Nyquist plot ($|L|=1$) becomes the horizontal axis (0 dB) of the Nichols chart. And, most importantly, the critical point $L=-1$ (which has magnitude 1 and phase -180°) maps to the point (0 dB, -180°) on the Nichols chart.

What happens to our family of M-circles under this transformation? They are no longer circles. Instead, they become a set of nested, closed contours looping around the critical (0 dB, -180°) point. The line for $M=1$ (the vertical line at $\text{Re}\{L\}=-1/2$) becomes a specific open curve. The circle for $M=\infty$ (the point $L=-1$) becomes the point (0 dB, -180°) itself. Modern software or old-fashioned chart paper can come with these M-contours pre-drawn, creating a kind of topographical map of the closed-loop gain, overlaid directly onto the open-loop canvas.

Herein lies the true power of this tool. An engineer plots the open-loop response $L(j\omega)$ of their system on a Nichols chart. This plot is a path snaking through the chart's landscape. As the path crosses the M-contours, we can instantly read off the closed-loop gain at each frequency.

The most important feature is often the ​​resonant peak​​, $M_p$, which is the maximum magnitude the closed-loop system ever reaches. It's a measure of how "peaky" or oscillatory the system's response is. Finding this peak is now incredibly simple. On our topographical map of M-contours, the resonant peak is simply the "highest" contour our system's path touches. More precisely, the open-loop plot $L(j\omega)$ will be ​​tangent​​ to one of the M-contours, and the value of this contour is the resonant peak $M_p$.

Imagine a system whose open-loop response on a Nichols chart just grazes an M-contour at a point where the open-loop gain is $G_{OL} = 4.20$ dB and the phase is $\phi_{OL} = -125^\circ$. At this point of tangency, the closed-loop gain is at its maximum. We don't even need to know which M-contour it is beforehand! We can just use the definition of $|T|$ with the open-loop values at that point to find the resonant peak. A quick calculation shows that these open-loop values correspond to a closed-loop magnitude of approximately $1.72$ dB. We have found a critical closed-loop performance metric just by observing a tangency point on the open-loop plot.

The story doesn't end there. Just as we can draw contours of constant magnitude (M-circles), we can also draw contours of constant phase for the closed-loop system, $\angle T(j\omega) = \text{constant}$. These are called ​​N-circles​​, and it turns out they also form a family of circles in the Nyquist plane.

What is truly remarkable, a sign of a deep underlying mathematical structure, is that the family of M-circles and the family of N-circles are mutually ​​orthogonal​​. Wherever a circle of constant magnitude intersects a circle of constant phase, they do so at a perfect right angle.

This property of orthogonality is a hallmark of a special class of functions known as ​​conformal maps​​, which have the amazing property of preserving angles locally. The simple-looking transformation $T = L/(1+L)$ is one such map. It takes a simple rectangular grid in the closed-loop $T$-plane (concentric circles for constant magnitude and radial lines for constant phase) and transforms it into the beautiful, orthogonal, curvilinear grid of M- and N-circles in the open-loop $L$-plane. It's a glimpse into the hidden geometric harmony that governs the world of feedback, turning a practical engineering problem into a journey of mathematical discovery.

In our previous discussion, we laid out the mathematical foundations of M-circles, those elegant loci of constant closed-loop magnitude. It might have felt like a purely abstract geometric exercise. But now, we arrive at the fun part. We get to see how these circles, when overlaid on the frequency response plots of real systems, become a kind of "magical decoder ring." They allow us to peer directly into the soul of a closed-loop system—its performance, its stability, its very character—by looking at its much simpler, open-loop behavior. This is not just a mathematical convenience; it's a profound tool for engineering insight and design.

The central challenge in control engineering is this: we design and build the "open-loop" part of a system, let's call its response $G(j\omega)$. This could be a motor, a chemical reactor, or the flight dynamics of a rocket. But what our user ultimately experiences, and what we truly care about, is the "closed-loop" behavior, when we wrap a feedback controller around it. The response of this complete system is a more complicated beast, $T(j\omega) = G(j\omega) / (1 + G(j\omega))$. How do we know if our choice of $G$ will lead to a good $T$?

This is where the M-circles come to the rescue. By plotting the open-loop response $G(j\omega)$ on a Nyquist or Nichols chart and overlaying the M-contours, we can instantly see the magnitude of the closed-loop response, $|T(j\omega)|$, for every single frequency. At any point where the curve of $G(j\omega)$ intersects an M-contour, the value of that contour is precisely the closed-loop gain at that frequency. We can trace the open-loop path and immediately read off the closed-loop story. For instance, we can determine the closed-loop system's response to very high-frequency noise by observing which M-contour the plot settles on as the frequency $\omega$ goes to infinity. Or, we can find the closed-loop gain at a critical frequency, like the point where the system's phase shift hits a full $-180$ degrees, just by looking at the M-contour passing through that point on the chart.

While knowing the gain at any frequency is useful, an engineer is often most interested in the maximum gain. This is the ​​resonant peak​​, denoted $M_p$. Think of it like pushing a child on a swing. If you push at just the right frequency—the resonant frequency—the amplitude of the swing gets very large. For a control system, a large resonant peak in its frequency response often signals trouble. It corresponds to excessively oscillatory, "ringing" behavior in the time domain. If you tell a robotic arm with a high $M_p$ to move to a new position, it will overshoot dramatically and wobble back and forth before settling down.

Graphically, finding this peak is astonishingly simple. As we trace the frequency response locus $G(j\omega)$ across the chart, we watch which M-circles it crosses. The resonant peak $M_p$ is simply the value of the highest-valued M-circle that the locus just touches. The locus is tangent to this one special M-circle. The frequency at which this tangency occurs is the resonant frequency, $\omega_r$. In one beautiful geometric picture, we have found the worst-case amplification of the system and the frequency at which it occurs.

The beauty of this method deepens when we realize that the resonant peak isn't just an abstract number; it's a direct window into the system's fundamental characteristics. For a vast number of systems, the behavior is dominated by a pair of poles, much like a simple mass-spring-damper. The character of such a system is captured by two parameters: its natural frequency, $\omega_n$, and its damping ratio, $\zeta$. The damping ratio is particularly crucial: it tells us how "sluggish" ($\zeta > 1$) or "bouncy" ($\zeta < 1$) the system is.

Now, here is the marvelous connection. The resonant peak $M_p$ that we find from our graphical chart is directly related to the damping ratio $\zeta$. For a standard second-order system, this relationship is given by a precise formula:

This equation is a Rosetta Stone, translating the language of frequency domain peaks into the language of time domain damping. It means we can perform a frequency sweep on a real-world system, plot its Nichols chart, find the tangent M-contour to get $M_p$, and then use this formula to estimate the system's effective damping ratio $\zeta$. This is an incredibly practical tool for system identification, allowing us to diagnose the health of a system from the outside, without having to know its internal guts.

The geometry holds even deeper secrets. In a remarkable piece of mathematical elegance, it can be shown that the damping ratio $\zeta$ is related to the M-circle diagram in another way. If you find the resonant M-circle (the one tangent to the Nyquist plot), and then draw a line from the origin that is tangent to that circle, the angle $\psi$ this line makes with the negative real axis gives you the resonant peak $M_p$ directly through the simple relation $M_p = 1/\sin(\psi)$. The physical property of damping is encoded right there in the geometry of the plot!

So far, we have used the charts to analyze a given system. But the true power of an engineering tool lies in design. M-circles are not just for diagnostics; they are a canvas for creation. Suppose our system has too high a resonant peak—it's too oscillatory. We need to reshape its frequency response curve, $G(j\omega)$, to pull it away from the high-value M-contours.

The simplest way to do this is with a proportional gain controller, $K$. This gain simply multiplies the open-loop response, which on a Nichols chart (with its logarithmic gain axis) corresponds to shifting the entire $G(j\omega)$ curve vertically up or down. We can literally slide the curve up or down until it becomes tangent to our desired target M-contour, say $M_p = 1.3$ (about $2.3$ dB). This allows us to find the exact gain $K$ that balances performance requirements, like tracking accuracy, against the need for a well-damped response.

Often, a simple gain change isn't enough. We might need to bend and twist the locus in more complex ways. This is where compensators, like lead or lag networks, come in. A lead compensator can provide a "phase boost" at a critical frequency, effectively pushing the locus curve to the right on a Nichols chart, away from the troublesome high-peak region. The design process becomes a beautiful geometric puzzle: identify a point on the uncompensated plot that is causing trouble, calculate the required phase shift and gain boost needed to move it to a desirable location (like a tangency point on the target M-contour), and then design a compensator that achieves this transformation at the right frequency. This graphical method turns the abstract algebra of compensator design into a tangible act of sculpting the system's response.

Finally, it's crucial to understand that the Nichols chart, armed with M-contours, is more than just a tool for finding one or two numbers. It is a holistic dashboard for assessing a control system's overall quality. An experienced engineer sees many things at once on this chart.

The proximity of the locus to the critical point $(-180^\circ, 0 \text{ dB})$ immediately reveals the system's stability margins—its buffer against unforeseen changes. The tangency to the highest M-contour tells us about its transient performance and tendency to oscillate (its $M_p$ value). But there are other families of contours we can draw, too. For instance, contours of the sensitivity function, $|S(j\omega)| = |1/(1+G(j\omega))|$, tell us how sensitive the system is to noise and variations in the plant itself. The peak sensitivity, $M_s$, is a key measure of robustness.

A great design, when viewed on this chart, tells a clear story. Its locus gives the critical point a wide berth (indicating healthy gain and phase margins), and it gently kisses low-value $M_p$ and $M_s$ contours. A poor design, by contrast, will swing perilously close to the critical point, resulting in small margins and frighteningly large peaks in both resonant gain and sensitivity. This graphical representation allows the designer to see the intricate trade-offs between stability, performance, and robustness all on a single page. It transforms the complex interplay of feedback into a visual narrative, allowing for design choices based not on blind computation, but on deep, practiced intuition.