## Calculus Without Tears## CWT Volume 4 - Good Vibrations, Fourier Analysis and the Laplace Transform## The Fourier PhilharmonicThe 1812 Overture by Tchaikovsky lasts for 14 minutes and 40 seconds. It is scored for a philharmonic orchestra having up to eighty musicians, playing a wide variety of instruments. The instruments include strings: violins, cellos, basses, brass: trumpets, trombones, French horns, tubas, woodwinds: clarinets, saxophones, flutes, piccolos, as well as percussion: drums, cymbals, bells, and triangles. The music for each instrument consists of multiple pages, each containing hundreds of notes. Each instrument gets a full workout in the course of the overture, which contains loud bombastic sections, as well as tranquil interludes. When Joseph Fourier decided to form an orchestra to play the Overture, he had only flute players (fortuitous, as the flute plays what is close to a pure sine wave) at his disposal. And, the players were severly limited, each player could play only one note, at one volume, for the duration of the piece. So that, when Fourier's baton dropped on the initial downbeat, each player played his/her one note, and held it at a constant volume, for 14 minutes and 40 seconds, through the climatic finale. Fourier claimed that he could re-score the piece (see how in the last section) so that his orchestra's version of the 1812 Overture would sound exactly like the full orchestra version, as played by the Berlin philharmonic, for example. Let me suggest that you carefully re-read the previous two paragraphs a few times, to realize how absolutely unbelievable the theorem demonstrated below really is. Each player plays one note, at one volume, for the duration of the piece. Not a misprint. Think of it this way: right before the final climatic chord, there is a second of silence. During this silence every player is playing the same note he/she has played since the beginning of the piece, at the same volume, and yet the listener hears silence; each player continues playing his/her note at the same volume through the finale, and the listener hears crashing symbols, blaring horns, strumming basses, etc. Clearly, this is not possible. Yet, it is, as we will see shortly. Famous mathematicians of the day, (and among the greatest of all time), including Poisson, Laplace and Lagrange, thought Fourier was wrong, and told him so. But Fourier was right. He couldn't prove it though, and the controversy raged for twenty years until Johann Dirichlet proved that Fourier's orchestra could play the 1812 Overture, and any other piece of music, and sound just like the Berlin philhamonic. Fourier's technique of decomposing any signal into its sine wave components has become the foundation of linear analysis and engineering math, with applications in all branches of science. ## Fourier SeriesHere's how Fourier assembled his orchestra. Let Each pair of sin and cos terms above represents a note played by one of Fourier's flute players, with j cycles per the length of the piece the frequency of the note, and A Fourier claimed that as n increases, the series above approximates
Sawtooth waveform Sines and cosines One half cosine with period 0.5/T
## The Fourier Transform and the Fourier Integral TheoremPeriodic functions can be approximated by Fourier series. This result can be extended to represent any function as an integral of sine and cosine components. Let The Fourier transform consists of the Fourier cosine transform and the Fourier sine transform. The Fourier cosine transform of The Fourier Integral Theorem (FIT) states
Impulse ## Fourier Theory using Complex Cosine-sinesWe need to shift gears slightly here. Typically, a presentation of Fourier theory begins as we have begun and then at some point, 'for convenience', introduces the complex exponential form of the theory as a change in notation. The effect is to rewrite the Fourier series as a sum of complex cosine-sines. We will avoid complex exponentials, but we still need to rewrite the series as a sum of complex cosine-sines. Why is this necessary? A sine wave is generated by a projection, say the height, of a point on a revolving wheel. But a sine wave doesn't do a good job of representing the physical process that generates it, that is, the motion of the wheel. There is a constancy to the motion of a wheel, but the sine wave speeds up and slows down. Rotation is well represented by a complex cosine-sine. By using complex cosine-sines, which initially appear more complicated, the mathematics will actually be simpler because they accurately represent the underlying process.
Yet, that is exactly the way 'complex numbers' are defined and constructed. Here is a number - 2343, here is a different unrelated number, 5656, and here is a complex number (2343, 5656), 2343 is the real part of the complex number, and 5656 is the imaginary part! It is impossible to overestimate the confusion this terrible terminology has caused math students at all levels. Then the confusion is compounded by raising the irrational number OK, if an imaginary number is not imaginary and is not a number, what is it? An imaginary number is a type of complex number. However, a complex number is not really a number, any more than two apples are an apple. A complex number is an ordered pair of numbers. So, an 'imaginary number' is really an ordered pair of numbers where the first number is 0.
Let Similarly for damped oscillation, if 3. a complex cosine-sine fully represents the process generating the vibration.
By considering the second term above as the the contribution at the frequency -jω, we can rewrite the Fourier series as Similarly, we can write the Fourier transform using complex cosine-sines. Now, the value of the Fourier transform at a particular frequency is a complex number that multiplies the complex cosine-sine at that frequency in the Fourier integral. The big payoff is that the derivative rule, which follows, takes a very simple form. ## The Laplace TransformFourier series can be used to approximate periodic functions but not aperiodic ones. Fourier integrals can be used to approximate aperiodic functions but not periodic ones. Sort of a Jack Spratt situation. We need a transform that can be used to represent both aperiodic and periodic functions. That would be the Laplace transform. The reason that a function So we multiply The complex valued Laplace transform is defined by
The crucial property of the Laplace transform (it's also true for the Fourier transform) is that it transforms differentiation into multiplication by s, that is, for s = (σ ,ω) Recall the product rule for differentiation (uv)' = u'v + v'u,
The Laplace transform maps the complex plane (σ ,ω) to the complex plane (a, b). A point in the domain represents the damping factor of the exponential used to modify the function, and the and frequency of the Fourier transform, a point in the range represents the Fourier transform cosine and sine components at the frequency. Thus, 'complex numbers' are being used to represent different things for the domain and range of the Laplace transform. We can take the Laplace transform of a damped complex cosine-sine, which is easy (see the derivative of damped oscillation above):
We can use MATLAB/FREEMAT to plot Laplace transform magnitudes, as shown below, this is a plot of the Laplace transform of the circuit output for the second example below
## Solving Differential Equations with the Laplace TransformWe can use the Laplace transform properties and library from the previous section to solve problems where the functions involved have rational polynomial Laplace transforms. This is a wide class of problems in science and engineering. A cookbook method is given for solving them.
We need two results, the first is the Fundamental Theorem of Algebra states that a polynomial equation of degree n has n roots (solutions) and can be written as a product of n factors of degree one. The second result we need is that a set of n linear equations with n unknowns has a solution. This leads directly to the partial fraction expansion for a rational polynomial. The theory has been elegant and we've obtained some amazing theorems with a modest effort. First, the bad news: the algebra required to solved even a simple problem, factoring the polynomials and writing the partial fraction expansion, however, can be intimidating. Now the good news: OCTAVE can do the algebra automatically.
Solution for a step input Solution for a pulse input Solution for a damped sine wave input
## Linear Systems Analysis
Many systems can be analyzed as input/output systems, where the input is a function, always of time in our examples, and the output is also a function of time. The system is characterized by its transfer function, which is the ratio of the Laplace transform of the output to the Laplace transform of the input. Given that the system differential equation is linear with constant coefficients, the resulting transfer function is a rational polynomial, and hence invertible using our table of Laplace transforms (and a little algebra).
Given that a system, or a component of a system, can be represented by its transfer function, then we can represent the system or component by a box, and the only property of the box is its transfer function. This is a block diagram representation of the system/component. We can connect block diagrams to create larger more complex systems.
We can characterize a system by its transfer function, its impulse response, and its frequency response. The system impulse response is the inverse Fourier transform of the system frequency response. The transfer function is the Laplace transform of the system impulse response. The frequency response is transfer function evaluated at (0, ω).
The transfer function of a linear system is a rational polynomial, and a rational polynomial is determined by its poles and zeros. We'll see that its possible to get from the transfer function to the poles and zeros using OCTAVE, in about as much time as it takes to type it in, and from that it is possible to write down the impulse response and step response directly, no sweat whatever (note: CWT does a way better job of facilitating this than the other books I've seen, CWT makes it COMPLETELY trivial - the CWT Laplace transform table has only 2 entries !).
Suppose you want to design an automatic driver for a car. To make it simple we'll suppose that the car only travels in a straight line, and you only want to program point to point moves. Go to location 10 and stop. Go to location 25 and stop. And so on. Your only controls is the throttle, which we will make reversable. You could spend a lot of time trying to program the throttle controller, but even then, if you run 'open loop', without feedback, the performance of your automatic driver will probably not be very good. To give the automatic driver any degree of accuracy, you'll need feedback, information about the actual acceleration, velocity, or position of the automobile. You'll need to incorporate this information into your control algorithm. You'll need a system with feedback. A block diagram for a generic feedback control system is shown below. The uncontrolled plant is P. The feedback is provided by the loop back, and is compensated by H. The controller is C.
So the closed loop transfer function is
We'll examine a system like the automatic driver described above. This is the typical servo control problem that is the basis of machine control. We'll simplify the car and throttle, so that the 'plant' is a box sliding on a table, the input to the plant is force which we control directly, and the only other force acting on the box is friction which is proportional to velocity.
A proportional, integral, derivative (PID) controller has one term which is proportional to the error signal, one that is the integral of the error, and one that is the derivative of the error. Hence its transfer function is The differential equation for the box's motion is Now, using OCTAVE, we can run through our repetoire of linear system analysis tools: We can use the toys provided by OCTAVE/MATLAB to study the effects of the controller parameters on system responsiveness, accuracy, and stability. We can vary paramters and display the closed loop step response as easily as shown below: ## The Discrete Fourier Transform and the Fast Fourier TransformSuppose we'd like to write Fourier's score for Tchaikovsky's Overture, or ten seconds of the tune Bangarang plotted below: how can we do it? All the transforms we have seen thus far were defined by mathematical functions, and we don't have a function that calculates ten seconds of the Bangarang waveform. Instead, we have sampled data. The data for the graph was obtained by sampling microphone output at a rate of 44,100 samples per second. But, the sound amplitude was defined for every instant of the ten seconds of music, thus, there were an infinite number of possible sample values, and we only have 10*44,100, not even one percent of the total. Is that enough? (Yes, see below.) Would you like to hear Fourier's orchestra tackle the ten seconds of Bangarang shown? Of course if we assemble the full orchestra the result will be indistinguishable from the original. So, let's limit Fourier's band to the first 25000 players (we never said Fourier's band was small). We can use the DFT described below to compute the score (the amplitude and phase of each player's note) and the result can be heard here. The discrete Fourier transform (DFT) calculates DFT coefficients from sampled data. The formula is If Let Now that the correspondence between the sampled data DFT and the Fourier series has been shown, our job is done. The DFT is an essential tool used in digital signal processing, for example, in mp3 compression. The Fast Fourier Transform (FFT) is a fast algorithm for computing the DFT. |