iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Fourier analysis answers a practical question: what sinusoidal frequencies make up a signal, and how much of each is present? For a periodic signal, a Fourier series represents the waveform as harmonics of one fundamental frequency. Integration calculates the harmonic coefficients; sampling then connects continuous-time frequency to the digital representations used in DSP.
How a Fourier series describes a periodic signal
Consider a periodic waveform that repeats every T seconds. Its fundamental frequency is f0 = 1/T hertz. A Fourier series represents the waveform using sinusoids at that fundamental frequency and its integer multiples: f0, 2f0, 3f0, and so on. These multiples are called harmonics.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Discrete-Time Signal Processing (Prentice-Hall Signal Processing Series) | $262.52 | Buy on Amazon |
| 2 |
|
Digital Signal Processing | $144.98 | Buy on Amazon |
| 3 |
|
Digital Signal Processing: Principles and Applications | $85.58 | Buy on Amazon |
Using angular frequency, let ω0 = 2π/T radians per second. One common trigonometric form is:
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsx(t) = a0 + Σn=1∞ [an cos(nω0t) + bn sin(nω0t)]
#1 Best Overall
The coefficients an and bn describe the contribution of each harmonic; a0 represents the average, or DC, component in this convention. The series coefficients are the analysis: they measure the signal’s components. Adding the sinusoids with those coefficients is the synthesis: it reconstructs the signal. Fourier series are a standard tool for studying periodic signals and their spectra; see the DSP First publisher contents for coverage of series analysis, synthesis, operations, convergence, and approximation.
A square wave as an example
An ideal, zero-average square wave contains odd harmonics: its fundamental, third harmonic, fifth harmonic, and so on, with amplitudes that decrease as harmonic number rises. This makes the frequency-domain description more than a list of frequencies: the coefficients also capture each component’s size and, in a complex-exponential representation, phase.
A finite sum can approximate a waveform without reproducing it exactly. Near a discontinuity, partial Fourier sums can show ringing; the behavior reflects approximation by a limited set of harmonics, not an extra frequency in the original idealized signal.
Recommended Free Tools
Why integration finds the coefficients
Sinusoids at different integer harmonics are orthogonal over a full period: when one is multiplied by another and the product is integrated across the period, the average is zero for distinct harmonics. A matching sinusoid, by contrast, contributes a nonzero average. Integration therefore acts like a projection that isolates a component from the mixture.
For the trigonometric convention above, the coefficients can be calculated over any full period using:
a0 = (1/T) ∫t0t0+T x(t) dt
an = (2/T) ∫t0t0+T x(t) cos(nω0t) dt, bn = (2/T) ∫t0t0+T x(t) sin(nω0t) dt
These formulas use an average over one period, with the factor of 2 applying to the non-DC sine and cosine terms. Other Fourier conventions distribute scale factors differently, so coefficients should always be interpreted alongside the convention that defines them. Pearson’s DSP First contents also connect Fourier-series operations with the derivation of the Fourier integral.
Free tools Windows power users keep installed
One-click scans. No signup required.
Rank #2
From periodic harmonics to continuous frequency
A periodic signal has harmonics on a discrete frequency grid: every component lies at an integer multiple of the fundamental. For an aperiodic signal, one useful intuition is to imagine a periodic description whose period grows. The fundamental spacing, 1/T, becomes smaller, so the harmonic grid gets denser. In the limiting picture, frequency varies continuously and sums give way to integrals, producing a Fourier-integral representation.
This is a conceptual bridge, not a claim that every signal has an ordinary, convergent Fourier series. The appropriate transform and convergence conditions depend on the signal. The essential continuity is that both approaches describe a signal through its relationship to sinusoidal basis functions; the periodic case uses harmonics, while the aperiodic case uses a continuous frequency variable.
What regular sampling does to a spectrum
Digital signal processing works with sequences of samples rather than a continuously observed waveform. Suppose samples are taken at regular intervals Ts seconds. The sampling frequency is fs = 1/Ts samples per second. An idealized model represents this operation as multiplying the continuous signal by an impulse train in time.
In that model, the sampled signal’s spectrum consists of shifted copies of the original spectrum, repeated at integer multiples of the sampling frequency:
Xs(f) = Σk X(f − kfs), k an integer
This expression follows the normalization used in TU Delft’s 2025 MUDE textbook sampling section. It is an ideal sampling model, not a complete description of every physical converter. Real systems have front-end behavior, including filtering, and produce finite-valued samples.
Why overlap causes aliasing
If the repeated spectral copies overlap, components at different continuous-time frequencies can lead to the same sampled sequence. This is aliasing: after sampling, those components cannot in general be distinguished from the samples alone. The frequency-domain copies make the problem visible—overlap means the original spectral content is no longer uniquely separated.
A familiar sampling threshold is the Nyquist condition: for a signal band-limited to frequencies at or below B hertz, ideal reconstruction requires a sampling frequency greater than 2B samples per second, together with suitable reconstruction assumptions. This condition is about a band-limited signal and idealized sampling and reconstruction; it is not a guarantee for arbitrary signals. Practical systems use anti-alias filtering before conversion to limit out-of-band energy that could fold into the frequencies of interest. MIT’s Lecture 9 on sampling and aliasing covers these issues as part of DSP.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.DTFT, DFT, and FFT are related but different
The names are easy to confuse because they belong to the same family of frequency-domain tools. The distinction is whether the signal is discrete or finite, whether frequency is continuous or sampled into bins, and whether the term names a representation or an algorithm.
| Tool | Input and frequency view | Role |
|---|---|---|
| Fourier series | Periodic signal; discrete harmonics at integer multiples of the fundamental | Represents a periodic waveform through harmonic coefficients |
| DTFT | Discrete-time sequence; continuous digital-frequency variable | Describes a sequence’s frequency content; the result repeats every 2π radians per sample |
| DFT | Finite record; a finite set of frequency bins | Computes frequency samples for the observed data |
| FFT | Input to a DFT calculation | An efficient family of algorithms for computing the DFT, not a separate transform |
The discrete-time Fourier transform (DTFT) is a continuous function of digital angular frequency, conventionally written ω in radians per sample. It is periodic with period 2π. To relate digital frequency to hertz, use the sampling rate: for sampling frequency fs, digital angular frequency ω corresponds to f = ωfs/(2π) hertz, with the periodic interpretation of digital frequency understood.
The discrete Fourier transform (DFT) takes a finite set of samples and evaluates the frequency representation at a finite grid of bins. If the record contains N samples at sampling frequency fs, the bin spacing is Δf = fs/N hertz, equivalently 1/Trecord when the record duration is defined as Trecord = N/fs. The FFT computes those DFT values efficiently; it does not change what the DFT means.
What a finite record can and cannot tell you
A DFT describes a finite observation, not an infinitely long signal. If a sinusoid does not complete an integer number of cycles within the recorded interval, treating the record as one period creates a boundary mismatch. Its energy then spreads across multiple bins rather than appearing only at a single bin; this is spectral leakage. Window functions can reduce the impact of the boundary mismatch, but they also affect the spectrum and should be chosen with the measurement goal in mind.
Record duration constrains frequency resolution: a longer observation produces more closely spaced DFT bins at a fixed sample rate. Zero padding adds values to the finite record before calculating the DFT, making a plotted spectrum more finely sampled and often smoother-looking. It does not add measurements, reveal new signal information, or improve the underlying ability to distinguish nearby components. IIT Palakkad’s EE3020A outline treats DFT, FFT, leakage, and limits on spectral resolution as connected DSP topics in its course outline.
How the ideas fit together in DSP
- Fourier series: periodic signals are described by harmonically related sinusoids and their coefficients.
- Integration: weighted averages over a period project the signal onto those sinusoidal components.
- Fourier integral: as the period grows and harmonic spacing narrows, the description leads toward continuous frequency for aperiodic signals.
- Sampling: regularly spaced observations create repeated spectral copies in the ideal model; overlap creates aliasing.
- Digital transforms: the DTFT describes discrete-time frequency continuously, the DFT samples frequency for a finite record, and FFT algorithms compute the DFT.
This progression—from harmonics, to projections, to sampling and finite frequency bins—is also reflected in DSP course sequences such as MIT OpenCourseWare’s RES.6-008 and the University of Texas at Austin EE351M course page.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
