Signal processing analyses and transforms signals: sound, images, radio, sensor readings, vibrations. It lets you filter noise, find a tone, compress data or recover a message. The central idea is that any signal can be seen both in time and in frequency, and each view makes some problems easy.
Key concepts
- Sampling: capturing a continuous signal at discrete times; the sampling rate must be above twice the highest frequency, or aliasing occurs.
- Linear time-invariant systems: described by an impulse response, and applied by convolution.
- Fourier transform: the spectrum of a signal; the FFT computes it quickly.
- Filters: FIR and IIR designs that keep wanted frequencies and reject others.
- Windowing and estimation: leakage, resolution and noise in real spectra.
Vocabulary
- Nyquist rate: twice the highest frequency present, the minimum safe sampling rate.
- Aliasing: a high frequency appearing as a false low one after sampling.
- Convolution: the operation that applies a filter to a signal.
- SNR: signal-to-noise ratio, usually in decibels.
What to study, in order
- Sines, complex exponentials and the idea of frequency.
- Sample a tone in code and see aliasing by lowering the rate.
- Implement a discrete Fourier transform, then use an FFT library and compare.
- Design and apply FIR and IIR filters to a noisy recording.
- Process a real signal end to end: audio, an accelerometer or a radio capture.