Audio series · 16 free episodes
A walkthrough of Steven W. Smith’s The Scientist and Engineer’s Guide to Digital Signal Processing, rewritten for listening rather than reading. The book leans on its figures; these episodes describe every one of them out loud, so nothing requires a screen. Each episode ends with a numbered key-points recap.
Choose Detailed for the full lesson and worked calculations, or Driving for a shorter explanation built around concepts and intuition. Both versions are available to download.
The pages here add back what audio cannot carry. Where an idea rewards being played with — aliasing, dither, convolution, spectral leakage — the figure is interactive. Everywhere else it is a static plot.
What a signal is, why order matters, and the same four operations hiding inside telephones, radar, sonar, oil exploration and medical imaging.
02How to describe a mess numerically. Mean, standard deviation, histograms, the normal distribution, and why averaging only helps as the square root of N.
03The one mistake you can never undo. Aliasing, the sampling theorem, dither, antialias filter choice, and bandpass sampling.
04Fixed versus floating point, Q notation, catastrophic cancellation, denormals, and why memory rather than arithmetic is now the bottleneck.
The two conditions that make everything else possible, why sinusoids keep their identity, and what to do when a system is not linear.
06The single most important operation in the field, explained three ways — stamping, the output-side sum, and a weighted average.
07Built from correlation rather than from a formula. The four Fourier transforms, orthogonality, polar notation, and the polar nuisances.
08Spectral leakage and why it is not a defect, choosing a window, why zero padding is not resolution, and circular convolution.
09The rules that let you reason about the frequency domain directly, plus the transform pairs worth knowing by heart.
10A worked eight point transform, bit reversal arising rather than asserted, and why Gauss had the algorithm 160 years early.
Where does the information live? Plus decibels properly explained, group delay, and three worked filter-choice decisions.
12The provably optimal smoother that is also the worst frequency-domain filter, and how to get a genuinely sharp cutoff.
13Arbitrary responses, why deconvolution disappoints, the Wiener filter, and overlap-add for very long kernels.
14Sharp responses from a handful of coefficients, what stability costs, biquads, and every filter in the series head to head.
Rotation, not imagination. Euler's formula, why sinusoids are shadows of phasors, and why radio needs I and Q.
16Poles, zeros, and the geometric picture — why stability is a circle, and why aliasing is what happens when you bend an axis into a loop.