What signal processing is

Digital signal processing (DSP) is turning real-world signals, like sound, radio waves and sensor readings, into numbers, and then using maths to clean them up, analyse them or change them.

When you speak into your phone, the microphone produces a wavy electrical signal. The phone measures it thousands of times a second and turns it into a list of numbers. DSP is what happens next: removing background noise, compressing it to send, and decoding what arrives on the other side. The same ideas run in music apps, noise-cancelling headphones, Wi-Fi, medical scanners, and the image-recognition layers of AI models.

Two ideas unlock the whole subject. First, any signal can be described as a mix of simple waves at different frequencies, like a chord made of notes, so removing noise often means removing certain frequencies. Second, convolution, a way of sliding one signal across another, is how filters work, and it’s the same operation at the heart of image-recognition AI.

What you’ll be able to do

By the end you will be able to:

  • explain how a smooth signal becomes a list of numbers, and how often you must measure to avoid distorting it;
  • break any signal into its frequencies, quickly, with the FFT;
  • explain convolution, and why it connects filters, signal processing and image AI;
  • design simple filters that remove unwanted frequencies;
  • analyse real recordings and read a spectrogram (a picture of frequencies over time);
  • explain how data is carried on radio waves, and receive real radio with a cheap USB receiver.

Where to start

Before you start: some trigonometry (sine waves) and basic programming in Python. Complex numbers help for later notes.

Read notes 01–05 in order; they’re the core. Notes 06–09 apply it.

  1. Signals and Systems: what a signal and a system are, and the special kind of system you can describe completely by how it responds to one short click.
  2. Sampling and Aliasing: turning a smooth signal into numbers, why you must measure more than twice as fast as the highest frequency, and the distortion you can never undo if you don’t.
  3. The Frequency Domain: the central idea: any signal is a sum of simple waves, and seeing it that way makes hard problems easy.
  4. The FFT: the fast method for breaking a signal into frequencies, which made real-time signal processing possible everywhere.
  5. Convolution and LTI Systems: sliding one signal across another, why it equals multiplying their frequencies, and why it’s the heart of image-recognition AI.
  6. Digital Filters: keeping some frequencies and removing others, the two main kinds of filter, and keeping them stable.
  7. Spectral Analysis: measuring the frequencies in real, messy recordings, and the trade-off between precision in time and in frequency.
  8. Modulation and SDR: how data rides on radio waves, and software-defined radio, where a computer does the work of a radio’s circuits.
  9. DSP in Practice: the tools, doing the maths with whole numbers on small chips, and how much everyday technology quietly relies on DSP.

Practice

  • Practice Exercises: twelve short Python scripts, such as watching a tone become a single spike, hearing aliasing, and proving the convolution rule, plus two exercises with a real radio receiver.
  • Solutions: worked answers. Try first.
  • Projects: hands-on tasks with a clear “done when”. A USB radio receiver (RTL-SDR, around $30) turns the whole course into a working radio.

Key ideas to remember

  1. Any signal is a mix of simple waves. Filtering is choosing which ones to keep. → 03
  2. Measure more than twice as fast as the highest frequency, or high frequencies disguise themselves as low ones, permanently. → 02
  3. Convolution in time equals multiplication in frequency. This one rule ties the subject together. → 05
  4. An image-recognition network’s convolution layer is DSP convolution with patterns it learned itself. → 05

About this course

  • Size: 9 notes plus practice, written in August 2026 to connect the hardware, radio and information theory courses.
  • Where it came from: reference sources, not yet tested by building.
↑ Back to Kingsley's Knowledge Base