Signal Processing — Fourier Transform, Filters & Sampling

Signal processing means analyzing or changing a signal so you can extract useful information from it. A signal might be a changing voltage, a sound recording, a vibration, or light measured by a detector. In practice, the main questions are: which frequencies are present, which parts should be kept, and how fast must you sample to store the signal digitally without creating errors?

Three ideas do most of the work. The Fourier transform shows frequency content. Filters keep some parts of the signal and suppress others. Sampling turns a continuous signal into digital data, but only if the sampling rate is high enough for the frequencies you need.

Time domain and frequency domain

In the time domain, you look at how the signal changes moment by moment. That view is useful when timing matters, such as when a pulse arrives or when a sensor spikes.

In the frequency domain, you look at the same signal as a mix of frequency components. That view is useful when you want to separate slow drift from rapid oscillation, identify a dominant tone, or understand what a filter will do.

The two views describe the same signal. They answer different questions.

What the Fourier transform tells you

The Fourier transform rewrites a signal in terms of sinusoidal components. In plain language, it asks how much of each frequency is present.

For a continuous signal $x(t)$, the transform is written as

$$X(f) = \int_{-\infty}^{\infty} x(t)e^{-i 2\pi f t}\,dt$$

You do not need to compute that integral by hand to use the idea well. The main intuition is that a complicated waveform can often be understood as a combination of simpler oscillations.

If a microphone signal contains a $200\ \mathrm{Hz}$ hum and a $2000\ \mathrm{Hz}$ tone, the time-domain plot may look messy, but the frequency-domain view can show peaks near those frequencies. That is often the fastest way to see what is actually happening.

How filters change a signal

A filter changes how strongly different frequencies pass through a system.

• A low-pass filter keeps lower frequencies and reduces higher ones.
• A high-pass filter keeps higher frequencies and reduces lower ones.
• A band-pass filter keeps a selected range and reduces frequencies outside it.

That description depends on frequency. It does not mean the filter automatically knows what is "noise." A filter only helps if the part you want to remove is separated from the part you want to keep in frequency, or in some other property the filter is designed to use.

This is why filters appear in audio cleanup, sensor smoothing, communication systems, and measurement equipment.

Why sampling rate matters

Sampling is the process of recording a continuous-time signal at discrete times. If the highest frequency you need to preserve is $f_{\max}$, then the sampling rate must satisfy

$$f_s > 2f_{\max}$$

to avoid aliasing in the ideal sampling picture. This threshold, $2f_{\max}$, is the Nyquist rate.

Aliasing means higher-frequency content can masquerade as a lower frequency after sampling. Once that happens, the sampled data can point to the wrong signal.

In real systems, engineers usually sample above the bare minimum and often use an anti-alias filter before digitizing. The extra margin matters because real signals are not perfectly band-limited.

Worked example: removing high-frequency noise from a sensor

Suppose a temperature sensor should change slowly, but the measured voltage includes a rapid electrical ripple. A simple model is

$$x(t) = 2 + 0.3\sin(2\pi \cdot 1\, t) + 0.05\sin(2\pi \cdot 60\, t)$$

Here the constant term $2$ is the baseline level, the $1\ \mathrm{Hz}$ term is the slow physical change you care about, and the $60\ \mathrm{Hz}$ term is unwanted interference.

The Fourier-transform viewpoint says the signal has energy at three main frequency locations:

• $0\ \mathrm{Hz}$ from the constant offset
• $1\ \mathrm{Hz}$ from the slow temperature variation
• $60\ \mathrm{Hz}$ from the interference

If your goal is to keep the slow trend, a low-pass filter can help because the useful part is at low frequency and the interference is much higher. A cutoff somewhere well above $1\ \mathrm{Hz}$ but well below $60\ \mathrm{Hz}$ would reduce the ripple while leaving the slow change mostly intact.

Now consider sampling. If you sample at $f_s = 20\ \mathrm{Hz}$, then the Nyquist rate is $10\ \mathrm{Hz}$. The $60\ \mathrm{Hz}$ component is above that limit, so the sampled data cannot represent it correctly unless it is removed before sampling. In an ideal aliasing calculation, a $60\ \mathrm{Hz}$ sinusoid sampled at $20\ \mathrm{Hz}$ lands on sample points that repeat exactly, so it can even disappear from the samples. That does not mean the original interference was harmless. It means the sampling process has hidden it in a misleading way.

This example ties the three ideas together:

• the Fourier transform helps you identify the unwanted component
• the filter removes it because it sits in a separate frequency range
• the sampling rule tells you why pre-filtering matters before digitizing

Common mistakes in signal processing

Treating the Fourier transform as a complete picture

Real measurements are finite, noisy, and often changing over time. A frequency plot is useful, but it is still an interpretation of measured data, not a perfect fingerprint of reality.

Filtering without stating what must be preserved

A low-pass filter reduces high-frequency noise, but it also reduces any real high-frequency signal. A filter is only "good" relative to a clear goal.

Using the Nyquist rate as a comfortable design target

$f_s > 2f_{\max}$ is the ideal minimum condition. In practice, working too close to that limit leaves little room for non-ideal filters and real-world signal content.

Assuming sampling creates information

Sampling only records what is already there. If the rate is too low, digitization does not recover the missing detail later.

Where signal processing is used

Signal processing appears anywhere measurements or waves carry information. In physics and engineering, that includes microphones, seismometers, radio receivers, ECG and EEG instruments, optical detectors, and control systems.

It is especially useful when the raw signal is not directly readable. Instead of staring at a complicated waveform, you use transforms, filters, and sampling rules to turn it into something interpretable.

Try a similar signal processing problem

Take a signal like

$$x(t) = \sin(2\pi \cdot 5\, t) + 0.4\sin(2\pi \cdot 40\, t)$$

Then ask two practical questions: which frequency would a low-pass filter try to keep, and what sampling rate would be comfortably above the Nyquist limit for the full signal? A useful next step is to compare this idea with transfer functions and see how systems shape signals in the frequency domain.