Blog
#216: The Fourier Transform β How to Hear the Shape of a Signal
Every signal can be expressed as a sum of sine waves. Exactly. Not as an approximation.
This makes operations that are complex in time domain trivial in frequency domain:
β’ Convolution β multiplication
β’ Differentiation β multiply by frequency
β’ Filtering β zero out coefficients
Full developer post covering:
π’ Discrete Fourier Transform β the math, O(nΒ²) naive implementation
β‘ Fast Fourier Transform β Cooley-Tukey 1965: DFT of n = two DFTs of n/2. O(n log n). For n=1M, factor 50,000Γ speedup.
π Convolution theorem β audio reverb, image blur, polynomial multiplication, all become O(n log n) via FFT
ποΈ Filtering β low/high/band pass in 3 lines of numpy. How JPEG uses DCT. How MRI raw data IS the Fourier transform.
π Parseval's theorem β energy preserved. Why lossy compression works: keep most energetic frequency components.
π΅ Nyquist theorem β sample rate must be > 2Γ max frequency. Why CD audio is 44.1kHz.
With working Python code throughout.
https://ai.jskitty.cat/blog.html#mathematics #fourier #signalprocessing #programming #python