Chapter 18: Fourier Transform

Welcome to Last Minute Lecture.

This free chapter overview is designed to help students review and understand key concepts.

These summaries supplement not replaced the original textbook and may not be redistributed or resold.

For complete coverage, always consult the official text.

You know, when we usually talk about engineering math, there's this comforting expectation of perfect repeating pattern.

Right, like everything's a nice, neat wave.

Yeah, exactly.

Think of a pendulum swinging endlessly or, I don't know, an AC generator humming along at a constant 60 hertz.

It feels safe.

You look at a wave and you know what it's going to do 10 minutes from now because it just repeats forever.

Well, that is the realm of periodic functions.

It's highly predictable and historically it's where we build our foundational mathematical tools.

But then you step into the real world.

I mean, think about the internet or your cell phone television broadcasts.

Suddenly, that predictable repetition is entirely gone.

Yeah, it completely vanishes.

The signals we actually care about, the ones carrying a human voice or a text message or you know, a streaming video, they don't repeat.

They are messy, isolated, non -periodic events.

Which puts engineers in a fascinating mathematical bind.

I mean, if our standard tools like the Fourier series from earlier studies only work for signals that repeat infinitely.

Right, the stuff we covered in the last chapter.

Exactly.

So how do we analyze the random one -off signals that actually run modern communication systems?

Welcome to this deep dive into the source material.

Today we're tackling a beast of a topic for you.

We're doing a dedicated one -on -one tutoring session based on chapter 18 of Fundamentals of Electric Circuits, the fifth edition.

It's a crucial chapter.

It really is.

And our mission is to decode the Fourier transform exactly as the text presents it.

No skipping steps.

We are going to master exactly how to turn messy real -world signals into clean math, understanding the why behind the equation so you can actually apply this to physical circuits.

And that application is vital.

As the chapter points out early on, the future of electrical engineering is deeply tied to communication systems.

Oh, absolutely.

You simply cannot understand how information travels through a circuit without the Fourier transform.

So we have a major conceptual hurdle right out of the gate here.

We need to jump from the periodic to the non -periodic.

How do we take a tool built for endless repeating waves and force it to work on a single isolated pulse of voltage?

We trick the map.

We use this really brilliant thought experiment from the text.

Okay.

Lay it on me.

Imagine a non -periodic function.

Let's visualize a single isolated rectangular voltage pulse on a graph.

The voltage spikes up, stays flat for maybe a millisecond, and drops back down to zero.

Just one solitary block sitting on the timeline?

Right.

Now, to use our old tools, we pretend this single pulse actually is periodic.

Wait, we just pretend?

Yes.

We imagine a clone of this pulse happening sometime later and another one before it, creating this endless train of pulses.

The time between these pulses is the period, which we call t.

Okay.

I'm picturing a repeating train of identical rectangular blocks.

Now what happens if we take that period t and we start stretching it?

We pull the adjacent pulses further and further away from our original pulse centered at zero.

So the gap of zero voltage between the pulses is just growing wider and wider.

Yes.

Now take it to the mathematical limit, push t all the way out to infinity.

Those clone pulses move infinitely far away in both directions.

What is left sitting in front of you?

Oh, I see.

Just the single original non -periodic pulse we started with.

We essentially erase the copies by hiding them at infinity.

That is the conceptual leap.

As the period approaches infinity,

our artificial periodic function simply becomes our actual non -periodic function.

That's a clever trick.

But the magic happens when we look at what this stretching does to the frequencies.

Right, the frequency spectrum.

In a standard Fourier series, the frequency spectrum is discrete.

You have fundamental frequencies and harmonics.

They look like distinct separate spikes on a bar graph.

Like a picket fence.

Exactly.

And the spacing between those spikes is 2 pi divided by the period t.

Wait, hold on.

If the spacing is divided by t and we just made t infinitely large, dividing by infinity means the spacing between the frequencies gets infinitely small.

Yes.

The discrete harmonic spikes pack closer and closer together until there's no space left between them.

The incremental spacing becomes a differential separation.

Wow.

As that happens, the summation formula we use for the Fourier series morphs into an integral.

The spiked bar graph blurs together and becomes a smooth continuous curve.

We have officially transitioned from the Fourier series to the Fourier transform.

I love visualizing this like a prism.

Like a Fourier series takes a repeating pattern and splits it into distinct separate colored laser beams.

Those are your discrete frequencies.

Uh huh.

But the Fourier transform takes a single isolated flash of white light and smears it out into a continuous seamless rainbow.

A very elegant way to picture it.

Mathematically, the textbook defines this continuous spectrum with a Fourier transform formula written as f of omega.

F of omega, right.

Now, looking at the raw calculus in the text can be intimidating, but let's break down what it physically means.

We take our time domain signal, let's call it f of t, and we multiply it by a term that looks like e to the negative j omega t.

Okay, that e term.

Think of that e term as a spinning mathematical probe.

We scan that probe across our signal from the beginning of time negative infinity all the way to the end of time positive infinity.

We are essentially checking to see if that specific frequency probe matches anything hiding inside our signal.

Exactly.

And because we scan across all time using that integral, the output, f of omega, tells us exactly how much of every single frequency is present.

And importantly, it works in reverse.

We have the inverse Fourier transform.

Right, the other half of the pair.

If you know the continuous frequency spectrum, you can integrate it across all frequencies to perfectly rebuild your original time domain signal.

They form a complete Fourier transform pair.

Let's ground this abstract math with a specific example from the chapter.

What happens if our signal is an impulse function, you know, a delta function?

Ah, delta of t.

In a circuit, that would be like an infinitely high, infinitely narrow spark of voltage right at time zero.

Okay, so a pure instantaneous spark.

If we run that infinitely narrow spark through our scanning mathematical probe, we use something called the sifting property.

The sifting property.

Because the impulse is zero everywhere, except at that exact instant of time zero, it sifts out the value of our probe right at that single moment.

Our probe term, evaluated at time zero, is just e to the power of zero, which is simply one.

Wait, so the Fourier transform of an impulse spark is just, it's just the number one.

A constant value of one stretching endlessly across all frequencies.

Think about the physical reality of that.

To make an event happen instantly in time, to create a perfectly sharp spike, you need an infinite amount of frequency bandwidth working together.

Yes.

Every single frequency, from direct current up to infinite gigahertz, must be represented with the exact same strength to build that one spark.

It highlights a fundamental law of nature.

The narrower a signal is in time, the wider it must be in frequency.

But let's be honest for a second.

Setting up scanning probes and doing complex integrals from negative to positive infinity every time we want to analyze a basic circuit sounds like a nightmare.

Oh, we'd be incredibly tedious.

Nobody wants to do that.

That's why the textbook immediately transitions into a toolkit of mathematical shortcuts,

the properties of the Fourier transform.

These rules allow us to bypass the calculus entirely when transforming complicated signals.

Let's see how these work in practice.

What if a signal isn't just sitting there, but is squashed or stretched, like playing a voice recording at double speed?

That introduces the time scaling property.

If you compress a signal in the time domain, so it happens twice as fast, the mathematics show that its frequency spectrum does the exact opposite.

Meaning what?

The spectrum expands outward, getting twice as wide, and its overall amplitude drops.

I can see that.

If I play a vinyl record at double speed, the song is over in half the time, but the singer's voice sounds like a chipmunk.

The frequency is shifted higher and spread out.

The textbook illustrates this by showing the transform of a rectangular pulse.

The frequency spectrum of a rectangle looks like a shape called a sinc function.

S -I -N -C, right?

Correct.

Imagine dropping a large rock in a pond, you get one huge central splash of water, and then smaller ripples dropping off on either side.

Got the picture.

When you time compress the rectangular pulse, making it narrower in time, that central splash in the frequency domain stretches outward, requiring higher and higher frequencies to exist.

Okay, so squashing time expands frequency.

Got it.

But what if the signal isn't squashed?

Just… just late, like a delayed text message.

Do the frequencies change if the pulse just happens a few seconds later?

That is the time shifting property, and no, the frequencies themselves do not change.

Oh, they don't.

Not the magnitudes, no.

If you delay a signal, the shape of the frequency spectrum's magnitude remains identical.

The delay in time simply manifests as a phase shift in the frequency domain.

So the ingredients are the same.

Exactly.

We just mathematically rotate the original spectrum by a phase angle.

Delaying a signal alters the alignment of its internal frequencies, not the ingredients themselves.

That makes intuitive sense.

Now what about calculus?

In a physical circuit, inductors and capacitors are governed by differential equations.

Taking derivatives and integrals of messy non -periodic signals sounds miserable.

And this is where the true power of the Fourier transform emerges for circuit analysis.

Through the time differentiation property, taking a derivative in the time domain is mathematically identical to simply multiplying your frequency spectrum by j omega.

Oh wow, because the derivative of our spinning e to the j omega t probe brings a j omega term down to the front.

Precisely.

You skipped the derivative entirely.

And the reverse is true for the time integration property.

Let me guess.

Integrating a function in the time domain translates to dividing its frequency spectrum by j omega.

You've got it.

Plus adding a DC impulse term if there's a constant offset, but basically yes.

So brutal calculus in the time domain transforms into middle school algebra basic multiplication and division in the frequency domain.

That's huge.

It gets even better when we look at the convolution property.

If you remember chapter 15, finding the output of a system required a convolution integral.

Oh, don't remind me.

Sliding one complex function over another and integrating the overlapping area.

It's the part of circuit analysis that makes students want to change majors.

The Fourier transform offers an escape hatch.

Convolution in the time domain corresponds perfectly to straight multiplication in the frequency domain.

Wait, seriously?

Seriously.

If you want to convolve an input signal with a system's impulse response, you just transform them both into frequency, multiply them together algebraically, and transform the result back.

So we can basically ignore time domain differential equations entirely.

How does this actually look when we are staring at a copper breadboard full of components?

We apply a frequency domain version of Ohm's law.

We swap out our physical time domain components for frequency domain impedances.

Okay, let's break that down.

Resistors.

Resistors just stay as their resistance value are.

Easy enough.

Inductors.

Inductors, instead of demanding differential equations, become an algebraic impedance of j omega l.

And capacitors.

Capacitors become an impedance of 1 over j omega c.

The relationship between voltage and current becomes a simple algebraic formula.

Voltage equals impedance times current, v equals z times i.

Which means all those standard circuit techniques we learned, voltage division, node analysis, mesh analysis, Thevenin equivalence, they all still work.

We just use these j omega terms instead of raw component values.

Exactly.

And this leads us to the concept of the transfer function, to note it as h of omega.

The transfer function is the defining characteristic of a circuit.

It is simply the ratio of the output response to the input excitation.

Okay, so if I have a wildly complex non -periodic input voltage, I don't need to do any convolution.

I just find the circuit's transfer function, multiply it by the Fourier transform of my messy input, and I immediately have the frequency spectrum of my output.

It turns a highly complex physics problem into a streamlined algebraic workflow.

But let me push back for a second.

This feels almost like a cheat code.

We are skipping so much math.

What is the blind spot here?

There has to be a catch.

There is a vital caveat, and the chapter stresses this rule strictly.

The Fourier transform, in this standard form, cannot handle circuits with initial conditions.

Meaning, uh, what exactly, pre -charged capacitors.

Exactly.

Or an inductor that already has current flowing through it before we start our analysis.

Ah.

Remember our mathematical probe.

It scans from negative infinity to positive infinity.

It fundamentally assumes the signal has been running for all of time.

Right.

If a capacitor on your breadboard is holding a 5 -volt charge at time zero, the Fourier transform doesn't have a built -in mechanism to handle that stored historical energy.

Speaking of energy, circuits aren't just abstract equations, they dissipate real power.

Like a resistor gets physically hot.

If we've transformed our entire analysis into this imaginary frequency space with j omega terms, how do we calculate the actual physical heat radiating off a resistor?

For that, we use Parseval's theorem.

This is the crucial bridge that connects abstract frequency math back to physical reality.

Okay, Parseval's theorem.

How does it work?

Parseval's theorem proves that the total energy delivered to a 1 -ohm resistor by a signal in the time domain is exactly equal to the total energy calculated in the frequency domain.

Let me make sure I'm visualizing this.

In the time domain, I know power is voltage squared over resistance.

If resistance is 1 ohm, power is just the voltage signal squared.

If I measure that power over time, I get the total energy, the total heat dissipated.

Yes, that's the time domain integral.

Parseval's theorem states you get the exact same number if you take the magnitude of the Fourier transform, square it, and integrate that across all frequencies dividing by 2 pi.

But why would I ever do that?

Why not just measure the voltage over time and be done with it?

What is the advantage of measuring energy in the frequency domain?

Because often, engineers only care about the energy within a specific frequency band.

The squared magnitude of the Fourier transform is literally the energy density of the signal.

Energy density.

Right.

Think of a graphic equalizer on a stereo.

Parseval's theorem lets you calculate exactly how much physical energy is packed into the heavy bass notes versus the high treble notes just by changing the limits of your frequency integral.

Oh, I like that.

It's like determining the total calories in a meal.

I could throw the entire plate of food into a calorimeter and burn it to measure the total heat released.

That's degrading in the time domain.

Sure.

Or I could take the recipe, calculate the exact calories of every single microscopic ingredient, the proteins, the carbs, the individual frequencies, and add those specific calorie counts up.

The total energy is identical either way, but the recipe method tells me exactly where the calories came from.

A highly accurate comparison.

It gives you incredible diagnostic insight.

Now, I'm putting myself in the shoes of a student working through this textbook.

Just a few chapters ago, in chapter 15, we spent weeks learning about Laplace transforms, which also moved us from the time domain into a frequency domain using a variable called S.

Yes, the S domain.

Why do we need both?

Seems redundant.

It's a very common point of confusion, and the textbook explicitly contrasts the two to clear it up.

Let's look at their distinct jobs.

The Laplace transform integrates from time zero to positive infinity.

It only looks forward.

So it only cares about positive time.

Because of that, it handles those initial conditions perfectly.

It is purpose -built for transient problems—the chaotic, bumpy moments right after you flip a switch and a pre -charge capacitor starts dumping its energy.

Furthermore, the Laplace variable is a complex number that encompasses an entire two -dimensional plane of possibilities.

And Fourier.

The Fourier transform applies to all time, past, and future.

It fails at initial conditions, as we discussed, but it provides far superior physical insight into the steady -state frequency characteristics of a signal.

So it's more specialized.

Yes.

Mathematically, it is essentially a constrained version of the Laplace transform, where we restrict our view purely to the imaginary axis.

We replace s with just j omega.

I see it like this.

Laplace is a heavy -duty tow truck.

You need it for those sudden, bumpy starts—the transient problems where you have initial conditions yanking on the circuit.

Painology.

But Fourier is a sleek sports car.

You use it when you're already cruising smoothly in the steady -state, the chaos of startup is over, and you really want to pop the hood and look closely at the pure frequency spectrum of the engine.

A great distinction—Laplace for the startup transients, Fourier for the steady -state spectrum.

Let's bring this entirely into the real world.

The chapter concludes by proving how these abstract, integral equations physically built modern technology, focusing on amplitude modulation and sampling.

Let's look at AM radio.

Okay, AM radio.

We know we want to send human voice through the air.

Voice frequencies are very low.

Why can't I just build an antenna and broadcast my voice directly?

Because transmitting low -frequency electromagnetic waves requires unimaginably massive antennas miles long and absurd amounts of power.

It's physically impractical.

To transmit efficiently, we need to hitch a ride on a much higher frequency.

So we use a high -frequency cosine wave called a carrier wave.

Let's say it's humming along at 1 megahertz.

The textbook says we multiply our slow voice signal by this extremely fast carrier signal in the time domain.

And according to our properties, multiplying by a cosine wave in time causes a frequency shift.

Wait, pause right there.

I know the rule says multiplication equals a frequency shift, but why?

Why does multiplying a voice by a carrier wave suddenly teleport my voice up to the radio band?

It comes down to how waves physically interact, which links back to basic trigonometry.

When you multiply two oscillating waves together, you create beat frequencies.

Beat frequencies, right.

You mathematically generate two brand new waves, one at the sum of the two original frequencies and one at the difference.

By multiplying your slow voice by a 1 megahertz carrier, you force the voice to ride the carrier, creating an upper side band and a lower side band centered perfectly around 1 megahertz.

Oh!

The entire low -frequency audio spectrum is mathematically picked up and neatly relocated to the high -frequency radio band.

And because of the math, the FCC knows exactly how wide those side bands are.

If an AM station's audio is 5 kHz wide,

the total bandwidth after modulation is 10 kHz.

The FCC can physically organize the invisible airwaves, spacing radio stations exactly 10 kHz apart so their frequency spectra never crash into each other.

You do algebra in the frequency domain to physically organize the airwaves.

The perfect application.

That is incredibly elegant.

Let's look at the final application from the chapter,

sampling, taking a continuous analog reality and chopping it up into discrete data points for a computer.

We model this process by multiplying our continuous real -world signal by a train of impulse functions, perfectly spaced apart by a set sampling interval.

So every time a spark fires, the computer captures exactly one instantaneous snapshot of the voltage and records zero everywhere else.

When we look at the Fourier transform of this chopped -up signal, we find something astonishing.

Sampling a signal in time causes its frequency spectrum to clone itself infinitely.

Wait, it clones itself?

Yes.

You get the original frequency spectrum, plus infinite copies of it shifting up and down the frequency axis.

So chopping up time creates endless echoes in frequency.

That leads to the most important rule in digital engineering, the Nyquist rate.

Because these spectral clones are repeating, we must ensure they don't overlap.

The textbook outlines the vital rule.

To completely recover an analog signal from chopped -up digital samples without losing a single drop of information,

you must sample at a frequency at least twice the highest frequency component of your signal.

Twice the highest frequency.

So if a telephone signal has a maximum frequency of 5 kHz, the Nyquist rate dictates you must take at least 10 ,000 snapshots per second.

Exactly.

But what if I try to cheat the system?

What if I want to save hard drive space and sample slower than that required rate?

If you sample too slowly, those repeat -e frequency clones get too close together and they begin to overlap.

The high frequencies of one copy physically bleed into the low frequencies of the adjacent copy.

And once they bleed into each other, can we filter them back apart?

You cannot.

It causes a permanent, unrecoverable corruption of the data known as aliasing.

Have you ever watched an old western movie and noticed that the wagon wheels look like they are spinning backwards?

Yeah, the stagecoach is moving forward, but the wooden spokes are slowly rotating in reverse.

It always looks so weird.

Yeah, that is visual aliasing.

The camera's frame rate, its sampling rate, was too slow to capture the true high frequency of the spinning spokes.

The overlapping frequencies created a false, lower frequency signal of a wheel spinning backward.

Aliasing is the ultimate consequence of ignoring the Fourier transform.

Wow.

This has been quite the journey today.

Let's recap what we've covered for you.

We started by taking a periodic pulse and stretching its period to infinity, watching as the distinct harmonic frequencies blurred together to form a continuous, seamless spectrum.

That fundamental shift.

Then we learned the mathematical properties that let us bypass differential calculus, allowing us to solve complex circuits using simple j omega impedances and transfer functions.

We explored Parseval's theorem, revealing how to calculate physical heat using energy density in the frequency domain.

Don't forget Laplace.

Right.

We pitted Laplace's heavy -duty transient analysis against Fourier's sleek, steady state power.

And finally, we saw how multiplying waves creates beat frequencies for AM radio, and how sampling at the Nyquist rate allows us to safely digitize the analog world.

It is a profound toolkit.

The Fourier transform acts as the fundamental bridge between the physical reality of changing voltages over time and the deeply analytical landscape of the frequency domain.

Which leaves me with a final slightly mind -bending thought for you to mull over.

We just talked about the Nyquist sampling theorem.

It proves mathematically that a continuous, infinitely smooth real -world analog signal can be perfectly reconstructed from discrete, chopped -up data points as long as you sample fast enough.

You don't need infinite points to capture reality, you just need enough of them.

If a continuous physical wear can be flawlessly represented by discrete, disconnected slices,

does that mean the smoothness of our physical world might just be an illusion?

Could the universe itself just be a series of discrete, incredibly high -frequency samples?

Taking circuit analysis to the level of quantum philosophy.

But it certainly demonstrates just how deeply the implications of this math resonate.

Something to ponder while you're grinding through those end -of -chapter problems.

Thank you so much for sending in your textbook and studying with us today.

From the Last Minute Lecture Team, this is the Deep Dive, signing off.