Control Systems in Practice, Part 10: Nichols Chart, Nyquist Plot, and Bode Plot
From the series: Control Systems in Practice
Explore three popular methods to visualize the frequency response of a linear time-invariant (LTI) system: the Nichols chart, the Nyquist plot, and the Bode plot. Learn about each method, including their strengths and weaknesses, and why you may choose one over another.
Find out how each plot presents the gain and phase shift of an LTI system across frequency, and discover how the different ways to present the information can help with system identification and closed-loop controller design.
In this video, we’re going to compare three popular methods for visualizing the frequency response of a linear, time-invariant system; the Bode plot, the Nyquist plot, and the Nichols plot. I want to explain what each of them are, what some of their strengths and weaknesses are and why you may choose one over another. Now, to keep this video a reasonable length, I’m going to mention, but gloss over a lot of concepts like linear systems, transfer functions, controller design, and stability criteria. I want to mention them so that we understand these three plots a bit better, but if you want more information on those other concepts, I’ve left a bunch of resources in the description of this video. Ok, let’s get to it. I’m Brian, and welcome to a MATLAB Tech Talk.
Let’s begin with a short overview of LTI systems. As the name suggests, LTI systems are linear - which means that they have the properties of superposition and homogeneity, and they are time-invariant - which means that their behavior does not change over time. The reason LTI systems are important is because they are mathematically easy to work with and to understand. The tools that we have to analyze an LTI system are much more capable than what we have to analyze and understand non-LTI systems. So, if the thing that you are working on can be approximated and modeled as LTI, then that can make design and analysis much easier.
And the main thing that we need to realize with LTI systems for this video is how the linear dynamics affect the signals that are fed into them. If the input is a sine wave with frequency omega, amplitude A, and phase phi, then what comes out on the other side is still a sine wave at the exact same frequency, but with a potentially different amplitude and phase. So, an LTI system won’t affect the frequency or change the shape of a sinusoid, but it will affect the gain and the phase of the signal. And the amount of gain and phase depends on the frequency of the input.
Therefore, we can fully describe an LTI system, by looking at how it changes the gain and phase of every single frequency across the spectrum. So, there are three parameters that we are concerned with; gain, phase and frequency.
And those are what Bode plots, Nyquist plots, and Nichols plots are showing. They all display the gain and the phase shift for every single frequency, it’s just presented differently depending on how you want to see the data.
Now, in addition to displaying these parameters neatly in graphical form, these plots are also helpful for system identification and closed loop controller design.
Let’s start with system identification. Let’s say our system can be modeled by this transfer function. To generate a Bode, Nyquist, or Nichols plot we could solve the transfer function mathematically and draw the curves using the equations that we come up with. However, this method assumes that we already have the transfer function to start with.
Let’s assume we don’t have this model. We’re working with the hardware directly, like a physical electrical circuit. We know that this can be modeled as an LTI system but we don’t yet have a model - this is just a black box to us were we can apply different inputs and measure the outputs.
For instance, we could apply a voltage as a sine wave at, say 1 rad/s and with a known amplitude, and compare it to the measured output voltage once all of the initial transients die down. By overlaying these two we could calculate the resulting gain, or change in amplitude, and the resulting phase shift. In this case, for a frequency of 1 rad/s, the gain increased by a factor of two and the signal was delayed by 90 degrees of phase. Now, since gain can span such a large range, we sometimes like to represent it on a log scale. So, instead of reporting a gain of 2, we would report it in decibels, with the following equation. So, the gain here is about +6 dB.
Let’s bring up two graphs. On the first, we can plot the gain or the magnitude in dB against frequency, and on the second we can plot phase shift in degrees against frequency. So, that’s our one point so far for this system. Not a whole lot we can do with that so let’s get another one. Now I’ll feed through an input voltage with a frequency of 0.5 rad/s. And check this out. At this slower frequency, the gain is nearly 1 or 0 dB, and the phase was delayed by only about 18 degrees. So, overall, less of an impact. And we can add these two points on our graph.
And we could choose another point at the faster 3 rad/s. This higher frequency is attenuated by more than 18 dB, and it’s hard to see exactly, but the phase is delayed by about 170 degrees.
And instead of one frequency at a time, if we input a sine sweep into this system, that starts at some super low frequency and slowly ramps up to high frequency inputs, we can generate the frequency response for this system. And we’ve created a Bode plot.
Remember, we don’t have the model of the system, we got this information by measuring the behavior of the physical circuit. But from this information, we can fit a model to this result. For example, we can see that this system has a low pass filter-like behavior since the low frequency signals are passed nearly unaltered and the high frequency signals are attenuated. We can also tell that there are two more poles than zeros, since the magnitude at higher frequencies slopes down at -40 dB per decade. And we can also see that there is a lightly damped mode around 1 rad/s. And by looking at this Bode plot, we could fit a transfer function, or a model to this data, that approximates this response.
But more than just representing the system and identifying the system with data, we may also want to use this plot for design. For example, we often want to know what would happen if we wrapped a feedback controller around this plan. It turns out that we can know a lot about the closed loop behavior by looking at the frequency response of the open loop system.
For instance, if there are any frequencies where the open loop system produces a gain of 0 dB and a phase of -180 deg, this will cause an oscillation in the closed loop system. So, without actually having to close the loop and test it, we know that an open loop system with this behavior will be a bad closed loop design.
So, we can look across the Bode plot and check every frequency that had a magnitude of 0 dB and ensure the phase isn’t -180 degrees. And if it’s not, then we’re good to go, the closed loop system won’t have any undamped oscillations. Of course, oscillation isn’t the only thing we’re concerned with, we also want to know if the closed loop system is stable.
But here’s where things can get a bit confusing, and I apologize if my super fast explanation adds to the confusion, but let me give it a shot anyway. If the open loop system is minimum phase and stable, so both its poles and zeros are in the left half plane, then we can determine closed loop stability easily from a Bode plot.
This is a Bode plot for a stable, minimum phase open loop system. As long as the magnitude is not above the 0 dB when the phase is -180 degrees, or the phase is lower than -180 when the magnitude is 0 dB, then the closed loop system will be stable. And you can see that the closed loop step response is in fact stable. So, when we’re dealing with these types of open loop systems, we often see gain and phase margin displayed on a frequency response plot. This is how much the open loop system can increase in gain or decrease in phase before the closed loop system goes unstable - before the curve goes beyond the critical point.
And, loop shaping, is the act of adding gain, and poles and zeros to the open loop system in the form of a controller to get the gain and phase margin and the performance that you want for the closed loop system. And we can use a Bode plot to see where we need to shape the response to get the stability margins we need. And we can see how close our design is to becoming unstable and we can watch the step response blow up when we cross that point.
Now, with that being said, checking for closed loop stability with a Bode plot is not as straightforward if the open loop system is nonminumum phase or is unstable. Gu is an unstable open loop system. And if we check out the Bode plot for this system, it doesn’t look like the response crosses either 0 dB or -180 degrees. So you might expect the closed loop system to be stable. However, you can clearly see from this step response that it is unstable. So, with non minimum phase or unstable plants, a Bode plot can be misleading.
And so this is where the Nyquist plot is useful.
A Nyquist plot still displays gain, phase, and frequency, but it does so all on a single chart - and with polar coordinates. Gain is the radial component, or the distance from the origin, on a linear scale, so we’re not dealing with decibels here, and phase is the angular component, or distance around the origin. Frequency isn’t shown directly, but as we progress along this curve, the frequency is changing.
So, if we compare this to the Bode plot, we can see that this point here has a phase shift of 0 degrees and a gain of 1 (which is 0 dB, but again, we’re not using decibels for a Nyquist plot), and as we can see on the Bode plot, corresponds to a very low frequency. And if we increase frequency until we get to 1 rad/s, we can see this is the max gain, or furthest from the origin, with a phase shift of about -90 degrees. And at the higher frequencies where this system has a very low gain, the Nyquist plot tends toward the origin.
Now, the upper half of the Nyquist plot is a mirror of the lower half and shows the response for negative frequencies, which are typically not shown on the Bode plot.
Also, notice, the critical point, the one we’re trying to avoid for oscillations, is just a single point on a Nyquist plot and it corresponds to -180 degrees and a gain of 1. So, this is how the Nyquist plot packages the same data that we have in a Bode plot, but the question now might be, why display the data in this form?
And there are some good reasons. One, is that we can determine closed loop stability much easier with this format. If the open loop system is minimum phase and stable, then, just like with a Bode plot, we can claim that the closed loop system is stable as long as the curve doesn’t cross over this point. And since all of the information is presented in a single graph, and there is only a single critical point, we can see very clearly how close the curve is to this point, even if there are multiple close passes. And, gain and phase margin can still be read directly off of this chart just like we can with a Bode plot.
Also, the benefit of a single plot like this is that we can more easily see how combinations of gain and phase impact the stability of the system. In fact, disk margin, which is arguably a better way to think about margin in your system is built on this kind of representation.
Furthermore, per the Nyquist stability criterion, we can also determine closed loop stability for unstable open loop systems - which is something that was very difficult to do on a Bode plot. We can compare the number and direction of encirclements around the critical point with the number of right half plane poles for the open loop system and determine closed loop stability. So, for certain systems, it is easier to check for stability and design a closed loop controller with a Nyquist plot than it is with a Bode plot.
The downside of Nyquist is that it tends to be harder to read since frequency is not explicitly called out, and I find that it’s not terribly intuitive to look at a Nyquist plot without a lot of practice. Of course, that’s entirely a preference thing and not a statement of fact. But one of the reasons I think it’s hard to read is that it’s harder to visualize an angular distance than it is a straight line distance.
For example, how much phase margin is there in this design? Well, it’s the angle between the critical point and where the curve crosses the unit circle. So, whatever this is. Maybe around 45 degrees or so. Well, to find out exactly, let’s go over to MATLAB to analyze it.
Turns out it’s about 41.4 degrees. So, you can read phase from this plot for sure, but my personal preference is to be able to view these margins on a cartesian graph rather than a polar graph. Plus, on a polar graph, we can create some pretty gnarly results as the response curve loops around for systems with a lot going on.
So, enter the Nichols plot. Nichols is a single graph like Nyquist, but it plots the information on a cartesian grid. Magnitude is the vertical axis, in dB, and phase is the horizontal axis, in deg. So, it’s the exact same units as the Bode plot.
In fact, it is the Bode plot information just package onto a single chart. And I think it’s easier to visualize if we align the magnitude and phase plots with the Nichols chart. As we trace through the frequencies, you can see how the Bode plot is mapped into this single graph.
A Nichols plot has some of the same benefits as the Nyquist plot, in that the information is on single chart so we can see visually how combinations of gain and phase will impact stability. Plus, it is easier to determine stability of nonminimum phase and unstable open loop plants than it is for a Bode plot. We can apply the Nyquist stability criterion to this chart as well by counting how many times the curve crosses above the critical frequency and in which direction.
Also, when we’re talking about closed loop design, we can obtain the values of the closed-loop transfer function from the open loop frequency response by overlaying a grid that show the so-called M and N circles. From this grid, we can read the gain and phase of the closed-loop system directly. And since adjusting the open loop gain causes Nichols plot to move purely up or down, with the grid, it’s really easy to see how open loop gain will affect close-loop response.
But, a Nichols plot has some of the same downsides as the Nyquist plot in that frequency is still not explicitly called out.
So Nichols and Nyquist seem pretty similar, however, the difference is that I think it can be easier to read off gain and phase on a cartesian grid rather than a polar grid. And for open loop systems that have a lot of phase lag, rather than looping around a bunch like the Nyquist plot, the Nichols plot just keeps moving to the left indefinitely - adding additional critical points every 360 degrees.
So, is the Nichols plot the best? Yes it is! Well, maybe not for everyone. The plot that you rely on the most is more than likely the one that you’re most familiar with. Like I said, they all show the same information. But there are some situations where it might be better to use one representation over another. For example, I prefer to do loop shaping with a Bode plot, assuming the open loop plant is stable and minimum phase. Whereas, I tend to look at a Nyquist or Nichols plot to visually see sensitivity and stability margins.
Alright, I hope this has helped you understand the differences between these three frequency response plots. Let me know in the comments which one you use most often and for what purpose. I’d love to see what everyone comes up with.
Don’t forget that if you want to learn more about anything I glossed over, like the details of the Bode plot or Nyquist Stability Criterion, I’ve left links to many great resources in the description.
If you don’t want to miss any other future Tech Talk videos, don’t forget to subscribe to this channel. Also, check out my channel, Control System Lectures if you are interested in other control theory topics as well. Thanks for watching and I’ll see you next time.
You can also select a web site from the following list:
How to Get Best Site Performance
Select the China site (in Chinese or English) for best site performance. Other MathWorks country sites are not optimized for visits from your location.