Please Log
In for full access to the web site. Note that this link will take you to
an external site (https://shimmer.mit.edu) to authenticate, and then you will be redirected
back to this page.
(Adapted by notes from T.F. Weiss)
Introduction and Outline
Previously we showed that for systems described sets of linear differential equations, we can easily derive the system's transfer function, H(s), in the form of a rational function in s (a ratio of polynomials). That is,
where L is the order of the system (assuming a_L \ne 0), the L values of s for which \sum_{l=0}^L a_l s^l = d_h(s) =0 are its poles or natural frequencies, and the \tilde{L} values of s for which \sum_{\tilde{l}=0}^{\tilde{L}} b_{\tilde{l}} s^{\tilde{l}} = n_h(s) = 0 are its zeros.
We also showed that given an input x(t) = X e^{s_0t} and a set of initial conditions, the output, \mathbf{y}, is given by
where the L coefficients \{A_1,A_2,\ldots,A_l\} are determined from
L initial conditions, and the \lambda_l's are the roots of the denominator polynomial, or the natural frequencies, or the poles, of H(s).
Our focus so far has been on the the roots of the denominator polynomial, the poles or natural frequencies, \lambda_1,...\lambda_L. They are important because they determine a system's stability. That is, if
\Re(\lambda_i) < 0\;\;\;\; i \in \{1,...,L\}
then given any bounded input of the form x(t) = X e^{s_0 t}, \Re(s_0) < 0, the output y(t) remains bounded. But what happens when H(s) is used in a control system?
Suppose H(s) = \frac{n_h(s)}{d_h{s}} is used in a feedback system with a gain, K_0. Then from Black's formula, the closed-loop system function, G(s), would be
The stability of G(s) depends on its natural frequencies, which are the roots of d_g(s), given by
d_g(s) = d_h(s) + K_0 n_h(s).
So, the stability of H(s) depends only on its denominator polynomial, d_h(s), but the stability of a feedback system that includes H(s) depends on BOTH H(s)'s denominator AND numerator polynomial. How much of each depends on the feedback gain, K_0.
If we design a more complicated controller,
K_0 K(s) = K_0 \frac{n_k(s)}{d_k(s)},
and then use K_0K(s)H(s) in a feedback system, the natural frequencies, and hence the stability, of that feedback system will depend on the roots of
d_k(s)d_h(s) + K_0 n_k(s)n_h(s).
If we try to design K(s) so that all the natural frequencies of the associated closed-loop transfer function are dominated by very negative real parts, we are designing by "placing the poles" of the closed-loop transfer function. This is exactly the strategy we used in discrete time, with differences. In CT, we pick \lambda_i so that e^{\lambda_i t} goes to zero as fast as possible, but in the DT, we pick \lambda_i so that \lambda_i^n goes to zero as quickly as possible. In CT, \lambda_i's dominated by very negative real parts correspond to fast decay, and in DT, \lambda_i's with magnitudes close to zero correspond to fast decay.
Rather than placing poles, a process that often defies intuition, we will look at a different approach based on using an alternative representation of our system functions. Thanks to a result in complex analysis (whose formal proof goes beyond the scope of this class), we can characterize H(s), a complex function of complex variable, by examining H for purely imaginary values of s. That is,
Since s is a complex number, its real and imaginary parts form a two-dimensional plane over which one evaluates H. But we are now saying you can learn everything you need to know about H by sampling s on a SINGLE LINE in the two-dimensional plane. Kind of amazing, until you think about the form of H(s) as a rational function with a finite set of real coefficients.
As we will see below, H(j \omega) is the "frequency response" of our system, meaning it tells us about the response to input sinusoids as a function of sinusoid frequency \omega = 2 \pi f, where f is frequency in Hertz. We can also characterize a feedback system's frequency response the same way,
If G(j \omega) tells us something about the response of our feedback system to input sinusoids, we might be particularly concerned if there is a frequency \omega such that
K_0 K(j \omega) H(j \omega) = -1,
because then G(j \omega) = \infty. Perhaps we should design K(s) to avoid this circumstance.
While it is simple to substitute j \omega for s in formulas for H(s), the meaning of that substitution bears more study. In what follows, we try to do just that.
y(t) = X|H(j\omega)| \cos(\omega t + \angle H(j\omega)) .
H(s) along the imaginary axis
We shall examine H(s) for s = j\omega or H(j\omega) where \omega is
the radian frequency.
H(s) evaluated along the imaginary axis is called the frequency response.
Radian frequency and frequency
\omega is the radian frequency in units of radians/second.
\omega can be expressed as \omega = 2\pi f.
f is the frequency in units of cycles per second which is called Hertz (Hz).
The frequency response H(j2\pi f) is often plotted versus f.
Measurement of H(j\omega)
To measure H(j\omega) of a test system we can use the system shown below.
However, when the sinusoid is turned on the response contains both a transient and a steady-state component. For example, suppose
H(s) = {5s \over (s+1)^2 + 10^2} .
The response of this system, y(t), to the input x(t) = \cos(2 \pi t)\,u(t) is shown below.
Note the transient at the onset which damps out after a few
cycles of the sinusoid so that the response approaches the particular solution, i.e., this is the steady-state response.
Therefore, to measure H(j\omega) we turn on the oscillator and wait till steady state is established and measure the input and output sinusoid. At each frequency, the ratio of the magnitude of the output to the magnitude of the input sinusoid defines the magnitude of the frequency response. The angle of the
output minus that of the input defines the angle of the frequency response.
More elaborate systems are available for measuring the frequency response rapidly and automatically.
Relation of time waveforms, vector diagrams, and frequency response
Consider and LTI system with system function
H(s) = {1 \over s + 1},
which has the frequency response
H(j\omega) = {1 \over j\omega + 1} .
The input is
x(t) = \cos(\omega t) ,
and the response is
y(t) = |H(j\omega)|\cos(\omega t +\angle H(j\omega)) ,
y(t) = {1 \over \sqrt{\omega^2 + 1}}\cos(\omega t - \tan^{-1}\omega).
Use MATLAB software (sysfun.m) to relate time waveforms, vector diagrams, and frequency response.
Different ways of plotting the frequency response
The frequency response
H(j\omega) = 1/(j\omega + 1),
is plotted in linear coordinates on the left and in mostly log coordinates (the Bode diagram) on the right
Bode diagrams
Definition and rationale
Frequency responses are commonly plotted as Bode diagrams which are plots of
20 \log_{10} |H(j\omega)| is plotted versus \log_{10} \omega (log-log).
\angle H(j\omega) is plotted versus \log_{10} \omega (log x - linear y).
The reasons are:
Logarithmic coordinates are useful when the range of |H(j\omega)| and/or \omega is large.
Asymptotes to the frequency response are easily plotted.
When the poles and zeros are on the real axis, the asymptotes are excellent
approximations to the frequency response.
Pole-zero and time-constant form
Consider an LTI system whose system function has poles and zeros on the negative, real axis.
It can be displayed in pole-zero form as follows
H(s) = C {(s + z_1)(s+z_2)\cdots(s+z_M) \over (s + p_1)(s+p_2)\cdots(s+p_N)} .
This system function can be evaluated along the j\omega axis to yield
Therefore, to plot the frequency response we need to add terms of the above form.
Decibels
It is common to plot frequency responses as Bode diagrams whose magnitude is expressed in decibels.
The decibel, denoted by dB, is defined as 20 \log_{10} |H| .
Decibel equivalents for a few quantities.
|H|\rightarrow20\log_{10}|H| dB
10^n\rightarrow20n dB
2\rightarrow\approx 6.02 \approx 6 dB
5\rightarrow\approx (20 - 6) =14 dB
\sqrt{2}\rightarrow\approx 6/2 = 3 dB
1.12\rightarrow\approx 1 dB
How many decibels correspond to |H| = 50? Express |H| = 100/2. Then
20\log_{10}(100/2) = 20\log_{10}100 - 20\log_{10}2 \approx 40 -6 = 34 dB.
Corner frequency
At \omega = \omega_c = 1/\tau, called the corner or cut-off frequency,
so that both frequency responses are first-order lowpass filters with a corner frequency of 1 rad/sec.
Magnitude
For
H(j\omega) = {1 \over j\omega +1} ,
the low-frequency asymptote
has a slope of 0 and an intercept of 0 dB and the high-frequency asymptote
has a slope of -20 dB/decade and an intercept of 0 dB at the corner frequency.
The corner frequency is 1 rad/sec and the bandwidth is 1 rad/sec.
The two asymptotes intersect at \omega = 1 where 20 \log_{10}|H(j\omega)| = -3 dB.}
Angle
For
H(j\omega) = {1 \over j\omega +1} ,
the low- and high-frequency asymptotes of the angle of the frequency response
are 0 and -90^{\circ} (-\pi/2 radians), respectively. The angle is -45^{\circ} at the corner frequency (1 rad/sec).
A line drawn from the low frequency asymptote a decade below
the corner frequency to the high frequency
asymptote a decade above the corner frequency approximates the angle of the frequency response.
At low frequencies, |H(j\omega)| \rightarrow 1 and
\angle H(j\omega) \rightarrow 0.
The inertia of the mass is negligible, and the damping force dominates so that the external force is proportional to velocity.
At high frequencies, |H(j\omega)| \rightarrow \frac{1}{\omega} and \angle H(j\omega) \rightarrow -90^{\circ}.
The inertia of the mass dominates so that the acceleration is proportional to external force and the velocity decreases as frequency increases.
At low frequencies, |H(j\omega)| \rightarrow 1 and \angle H(j\omega) \rightarrow 0.
The impedance of the capacitance is large so that all the input voltage appears at the output.
At high frequencies, |H(j\omega)| \rightarrow 1/\omega and \angle H(j\omega) \rightarrow -90^{\circ}.
The impedance of the capacitance is small so that the current is determined by the resistance and the
output voltage is determined by the impedance of the capacitance which decreases as frequency increases.
Miniproblem
Given
H(s) = {s \over (s+1)(s+10)} .
Determine the magnitude of the frequency response in the form of a Bode diagram.