Introduction

Every engineering instrument that records, transmits, filters, compresses, or computes with changing quantities is, in some sense, asking a Fourier question.

A microphone converts air pressure variations into an electrical signal. An oscilloscope displays voltage as a function of time. A radio receiver separates one broadcast from another even though many electromagnetic waves arrive together at the antenna. A spectrum analyzer shows which frequency components are present in a measurement. A digital audio system samples a continuous waveform, stores numbers, and later reconstructs sound. A quantum computer, at a much smaller and more abstract scale, can apply a quantum Fourier transform to amplitudes spread across many computational basis states.

These examples look different, but they share one central idea: a complicated pattern can often be understood by decomposing it into simpler oscillatory patterns.

That is the guiding theme of this book.

The first idea: signals carry structure

A signal is a mathematical description of a quantity that varies with one or more independent variables. In electrical engineering, the independent variable is often time.

For example, if \(v(t)\) denotes the voltage at the output of a sensor at time \(t\), then \(v\) is a continuous-time signal. The phrase continuous-time means that the variable \(t\) is allowed to range over a continuum of real values, such as

\[ t = 0,\quad 0.001,\quad 0.0015,\quad \sqrt{2},\quad 10.7, \]

at least in the mathematical model.

If instead a digital system stores one voltage sample every \(T_s\) seconds, we may write the resulting sequence as

\[ x[n] = v(nT_s), \]

where \(n\) is an integer. This is a discrete-time signal. The variable \(n\) does not move continuously; it jumps from one integer to the next.

At first, it may seem natural to study signals mainly in the time domain: we draw \(v(t)\) or \(x[n]\), inspect its shape, and ask when events occur. Time-domain thinking is essential. But many engineering questions are not easiest in the time domain.

For example:

• A pure tone at \(440\ \text{Hz}\) may be hard to locate by staring at a short audio waveform, but it appears directly as a frequency component.
• A low-pass filter may have a complicated impulse response in time, but its design goal is simple in frequency: pass low frequencies and attenuate high frequencies.
• Noise in a measurement may look irregular in time, while its average power distribution over frequency can reveal whether it is broadband, narrowband, or shaped by the instrument.
• A sampled signal may appear plausible in time but contain aliasing, meaning that different continuous-time frequencies have become indistinguishable after sampling.

The Fourier transform gives us a disciplined way to move between these viewpoints.

The second idea: sinusoids are natural building blocks

A sinusoid is a waveform built from sine or cosine functions, such as

\[ \cos(2\pi f t) \]

or

\[ \sin(2\pi f t). \]

Here \(f\) is the frequency in hertz, meaning cycles per second. A frequency of \(5\ \text{Hz}\) means five complete oscillations per second.

Sinusoids are special in linear systems. A linear system is a system where scaling and addition behave predictably: if input \(x_1\) produces output \(y_1\), and input \(x_2\) produces output \(y_2\), then input \(a x_1 + b x_2\) produces output \(a y_1 + b y_2\), for scalars \(a\) and \(b\). A time-invariant system is one whose behavior does not change when the input is delayed in time. Together, these are called linear time-invariant systems, or LTI systems.

One of the most important facts in signals and systems is that complex exponentials behave especially simply when they pass through an LTI system: the output has the same exponential form, multiplied by a frequency-dependent complex number. Standard signals-and-systems texts make this property central because it turns many differential-equation and convolution problems into algebraic frequency-domain problems (Oppenheim, Willsky, and Nawab, 1997).

The complex exponential

\[ e^{j2\pi f t} \]

may look less familiar than sine and cosine, but Euler’s formula says

\[ e^{j\theta} = \cos \theta + j\sin \theta, \]

where \(j\) is the imaginary unit satisfying \(j^2=-1\). Thus a complex exponential is a compact way to carry both cosine and sine information at once. Electrical engineering uses this notation because it makes phase shifts, filtering, modulation, and spectral analysis much cleaner.

For example, suppose an LTI circuit receives a sinusoidal input at \(1\ \text{kHz}\). In steady state, the output is also at \(1\ \text{kHz}\), but its amplitude and phase may change. The circuit does not create a new frequency from a single sinusoidal input if the model is linear and time-invariant. That is why frequency response is so useful: it tells us how the system treats each sinusoidal component.

The third idea: a spectrum describes “how much of each frequency”

A spectrum is a representation of a signal in terms of frequency content. Informally, it answers the question:

How much of each frequency is present, and with what phase?

For a simple example, consider

\[ x(t)=3\cos(2\pi 100t)+\cos(2\pi 250t). \]

This signal contains two sinusoidal components: one at \(100\ \text{Hz}\) with amplitude \(3\), and one at \(250\ \text{Hz}\) with amplitude \(1\). Its time-domain graph may look like a combined oscillation, but its frequency-domain description is more direct: two frequency components, one stronger than the other.

A Fourier transform is a mathematical operation that maps a signal from a time or sample-index description into a frequency description. The inverse Fourier transform maps the frequency description back to the original signal, under the appropriate mathematical conditions.

This idea has a long history. Joseph Fourier’s work on heat conduction developed the idea that functions could be represented by trigonometric series, an idea that became foundational for Fourier analysis even though later mathematics refined the required conditions and meanings of convergence (Fourier, 1822). Today, Fourier methods appear throughout electrical engineering, applied mathematics, physics, imaging, communications, control, and quantum information.

This book will develop the idea carefully, because there is not only one Fourier transform. There are several related transforms, each suited to a different signal model:

• Fourier series for periodic continuous-time signals.
• Continuous-time Fourier transforms for aperiodic continuous-time signals.
• Discrete-time Fourier transforms for infinite discrete-time sequences.
• Discrete Fourier transforms for finite data records.
• Fast Fourier transform algorithms for computing DFTs efficiently.
• Quantum Fourier transforms for unitary operations on quantum states.

These are not separate tricks. They are related versions of one deep idea: represent a signal, vector, or state using oscillatory basis functions.

Why finite data needs special care

Real measurements are finite. A laboratory instrument does not observe a signal for all time. A microcontroller does not store an infinite sequence. A computer program usually receives \(N\) samples:

\[ x[0], x[1], \ldots, x[N-1]. \]

The discrete Fourier transform, or DFT, converts these \(N\) samples into \(N\) frequency-domain coefficients. These coefficients describe the finite data record as a combination of discrete complex exponentials.

The DFT is not merely a sampled version of the continuous-time Fourier transform. It has its own structure. It assumes periodic extension of the finite record, leads naturally to circular convolution, and requires careful interpretation of bin spacing, leakage, windowing, and zero padding.

For example, if a measurement contains a sinusoid whose frequency falls exactly on a DFT bin, the DFT can represent it cleanly in one pair of bins for real-valued data. If the sinusoid falls between bins, its energy spreads across many bins. This spreading is called spectral leakage. Leakage is not an error in the DFT; it is a consequence of observing a finite-length record and representing it using the DFT basis.

Computing the DFT directly requires on the order of \(N^2\) arithmetic operations. The fast Fourier transform, or FFT, is a family of algorithms that compute the same DFT much faster by exploiting symmetry and factorization. The famous Cooley–Tukey algorithm showed how to reduce the computational cost dramatically for composite lengths, especially powers of two, and helped make Fourier analysis a practical tool in digital signal processing (Cooley and Tukey, 1965).

For example, a direct DFT of \(N=1{,}048{,}576\) samples would require an enormous number of complex multiplications and additions. An FFT of the same data length is feasible on ordinary hardware. This difference is one reason FFTs are used in audio processing, vibration analysis, radar, medical imaging, communications, and numerical simulation.

Sampling connects the analog and digital worlds

Electrical engineers often begin with a physical voltage or current that varies continuously in time, then convert it into digital samples. This raises a precise question:

When do the samples contain enough information to reconstruct the original continuous-time signal?

The answer depends on bandwidth. A signal is bandlimited if its Fourier transform is zero outside some finite range of frequencies. In ideal sampling theory, a bandlimited signal can be reconstructed from sufficiently dense uniform samples. Shannon’s classic communication-theory treatment helped establish the modern sampling theorem formulation and its role in communication systems (Shannon, 1949).

A simple example shows the danger. Suppose a sensor samples at

\[ f_s = 1000\ \text{samples/second}. \]

A sinusoid at \(100\ \text{Hz}\) is represented by ten samples per cycle, so it is naturally distinguishable. But a sinusoid at \(900\ \text{Hz}\) produces the same sample values as a lower-frequency sinusoid at \(100\ \text{Hz}\), up to phase convention, when sampled at \(1000\ \text{Hz}\). After sampling, those two continuous-time frequencies cannot be distinguished from the samples alone. This is aliasing.

Aliasing is not caused by poor software. It is a mathematical consequence of sampling. Practical systems therefore use anti-alias filters before sampling to attenuate frequency components that would fold into the band of interest.

This book will treat sampling not as a memorized rule, but as a frequency-domain phenomenon. Once you see the spectral replicas created by sampling, the sampling theorem becomes easier to understand and harder to misuse.

From Fourier matrices to quantum circuits

The second half of this book moves toward quantum computation. That transition may seem surprising at first. What does a radio spectrum have to do with a quantum algorithm?

The bridge is linear algebra.

The DFT of a finite vector can be written as multiplication by a complex matrix. With the right normalization, this matrix is unitary. A unitary matrix preserves vector length and inner products. In quantum mechanics, the allowed evolution of a closed quantum state is represented by unitary operations, and quantum circuits are built from unitary gates acting on qubits. Standard quantum information texts place unitary evolution, tensor products, measurement, and circuit composition at the center of the mathematical model (Nielsen and Chuang, 2010).

A qubit is the basic unit of quantum information. Mathematically, its pure state is a unit vector in a two-dimensional complex vector space. One common notation writes a qubit state as

\[ \alpha\lvert 0\rangle + \beta\lvert 1\rangle, \]

where \(\alpha\) and \(\beta\) are complex amplitudes satisfying

\[ |\alpha|^2 + |\beta|^2 = 1. \]

The symbols \(\lvert 0\rangle\) and \(\lvert 1\rangle\) are basis states. The squared magnitudes \(|\alpha|^2\) and \(|\beta|^2\) give measurement probabilities in that basis.

For \(n\) qubits, the state vector has \(2^n\) complex amplitudes. The quantum Fourier transform, or QFT, is a unitary transformation on these amplitudes. It is closely related to the classical DFT matrix, but it appears inside a quantum circuit rather than as a classical list of arithmetic operations performed on stored data.

This distinction is crucial. The QFT does not simply give a magical faster FFT for ordinary data. A quantum computer does not let us read all \(2^n\) output amplitudes directly. Measurement gives limited classical information. The QFT becomes powerful when it is embedded in algorithms where the desired information is encoded in a way that measurement can reveal efficiently.

The most famous example is Shor’s factoring algorithm, which uses Fourier structure through order finding and showed that integer factoring and discrete logarithms can be solved in polynomial time on an ideal quantum computer (Shor, 1994). This result does not mean that every Fourier problem becomes exponentially faster on a quantum computer. It means that certain algebraic problems have hidden periodic structure that a quantum Fourier-based procedure can exploit.

What this book will build

The path of the book is deliberately gradual.

We begin with classical signals: continuous-time and discrete-time models, LTI systems, complex exponentials, and the meaning of frequency. Then we develop the mathematical foundation: complex numbers, inner products, orthogonality, basis expansions, and convergence ideas. These tools are necessary because Fourier analysis is not only a formula; it is a way of projecting signals onto structured building blocks.

Next, we study Fourier series and the continuous-time Fourier transform. You will learn how periodic and aperiodic signals are represented, why magnitude and phase both matter, how bandwidth is described, and how energy and power appear in spectral form.

Then we study transform properties. Linearity, shifting, scaling, modulation, convolution, and multiplication are not decorative identities. They are engineering tools. A difficult time-domain operation may become simple in frequency, and a confusing frequency-domain effect may become obvious in time.

After that, we address generalized signals such as impulses and impulse trains. The Dirac delta is not an ordinary function; it is an idealized object that must be handled carefully. But it is indispensable in circuit theory, sampling, and system modeling.

The middle chapters connect Fourier analysis with systems, sampling, the DTFT, the z-transform, the DFT, FFT algorithms, communications, circuits, and measurement. This is the engineering core of the book. The goal is not only to compute transforms, but to interpret them correctly.

Finally, we shift to the finite-dimensional linear algebra viewpoint. The DFT becomes a matrix. The matrix becomes a unitary change of basis. Tensor products lead naturally to multi-qubit state spaces. From there, we introduce quantum information, reversible computation, the QFT, its circuit decomposition, phase estimation, and quantum algorithms based on Fourier structure.

By the end, you should be able to explain both of these statements without confusing them:

1. The FFT is a fast classical algorithm for computing the DFT of explicitly stored data.
2. The QFT is a quantum unitary operation that transforms amplitudes in a quantum state and is useful inside certain quantum algorithms.

They are related, but they are not the same computational object.

How to study the equations

Fourier analysis contains many formulas. Some are short but conceptually dense. For example, a continuous-time Fourier transform is often written as

\[ X(f)=\int_{-\infty}^{\infty} x(t)e^{-j2\pi ft}\,dt. \]

At first glance, this may look like a definition to memorize. But each part has meaning:

• \(x(t)\) is the signal being analyzed.
• \(e^{-j2\pi ft}\) is a complex sinusoid used as a probing waveform.
• The integral combines the signal with that probing waveform across all time.
• The result \(X(f)\) is a complex number describing how strongly frequency \(f\) is present, including phase information.

A useful habit is to read equations as sentences. The formula above says:

To find the frequency-domain coefficient at frequency \(f\), compare the signal with a complex sinusoid of frequency \(f\) over time.

Likewise, when you later see the DFT,

\[ X[k]=\sum_{n=0}^{N-1} x[n]e^{-j2\pi kn/N}, \]

read it as:

To find the \(k\)-th frequency coefficient, compare the \(N\)-sample data record with the \(k\)-th discrete complex exponential pattern.

The quantum version will also become less mysterious when read this way. The QFT maps computational basis states into structured phase patterns. The formulas are compact, but their meaning is geometric: change the basis so that hidden periodic or phase information becomes measurable in a useful form.

The promise and the caution

Fourier methods are powerful because they reveal structure. But they also demand interpretation.

A spectrum is not automatically “the truth” about a signal. It depends on the model, the observation interval, the sampling rate, the window, the instrument, and the assumptions behind the transform being used. A sharp FFT plot can still be misleading if the data were undersampled, windowed poorly, or interpreted with the wrong frequency scale.

Similarly, a quantum Fourier transform is not automatically a speedup. It is a unitary operation inside a larger algorithm. Its usefulness depends on state preparation, interference, measurement, noise, and the structure of the problem.

The goal of this book is therefore twofold:

First, to make you technically fluent with Fourier transforms from classical signals through FFTs.

Second, to prepare you to understand the quantum Fourier transform without treating it as magic.

The same discipline will serve both goals: define the signal model, choose the correct transform, track normalization and phase, understand the basis, and interpret the result with care.

If you follow that discipline, Fourier analysis becomes more than a collection of formulas. It becomes a language for seeing patterns that are otherwise hidden.

References

Cooley, J. W., and Tukey, J. W. (1965). “An Algorithm for the Machine Calculation of Complex Fourier Series.” Mathematics of Computation, 19(90), 297–301.

Fourier, J. (1822). Théorie analytique de la chaleur. Paris: Firmin Didot.

Nielsen, M. A., and Chuang, I. L. (2010). Quantum Computation and Quantum Information: 10th Anniversary Edition. Cambridge: Cambridge University Press.

Oppenheim, A. V., Willsky, A. S., and Nawab, S. H. (1997). Signals and Systems (2nd ed.). Upper Saddle River, NJ: Prentice Hall.

Shannon, C. E. (1949). “Communication in the Presence of Noise.” Proceedings of the IRE, 37(1), 10–21.

Shor, P. W. (1994). “Algorithms for Quantum Computation: Discrete Logarithms and Factoring.” In Proceedings of the 35th Annual Symposium on Foundations of Computer Science, 124–134. IEEE.