dsp-engineer
Think DSP-inspired signal-processing engineering guide for C++ implementations. Use when Codex needs to explain, design, debug, or implement DSP analysis/synthesis workflows involving Signal/Wave/Spectrum models, FFT/DFT/DCT, spectra, spectrograms, harmonics, aliasing, noise, autocorrelation and pitch estimation, convolution, LTI systems, impulse responses, filtering, modulation, sampling, interpolation, or translating Think DSP/Python/NumPy examples into C++.
How do I install this agent skill?
npx skills add https://github.com/shortwavlabs/agent-skills --skill dsp-engineerIs this agent skill safe to install?
- Gen Agent Trust Hubpass
The skill is a legitimate reference guide for Digital Signal Processing (DSP) concepts and C++ implementations based on the 'Think DSP' book. It contains no malicious code, prompt injections, or data exfiltration risks.
- Socketpass
No alerts
- Snykpass
Risk: LOW · No issues
What does this agent skill do?
DSP Engineer
Overview
Use this skill to reason about digital signal processing the way Think DSP teaches it: start with programmable signals, inspect the time and frequency domains, then turn the idea into clear C++.
This skill complements the dsp skill in this repository. Use dsp for production-ready audio algorithms and cookbook implementations; use dsp-engineer for concept-first analysis, derivations, experiment scaffolding, and translating Think DSP-style Python/NumPy examples into C++.
Source
This skill is based on a review of Think DSP by Allen B. Downey, Green Tea Press, copyright 2014, distributed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0. Treat the references here as transformed engineering notes, not as a replacement for the book. Attribute Think DSP when using its explanations, exercises, or example structure.
Workflow
- Classify the task as synthesis, analysis, transformation, comparison, resampling, or system modeling.
- Establish units first: sample rate in Hz, duration in seconds, frame count, bin spacing, Nyquist/folding frequency, amplitude convention, and whether the signal is real or complex.
- Choose the representation:
Signal: a mathematical signal that can be evaluated at times.Wave: sampled values plus sample rate and start time.Spectrum: complex frequency bins plus bin frequencies.Spectrogram: a time-indexed set of short-time spectra.
- Build the smallest C++ experiment that proves the idea. Prefer
std::vector<double>andstd::complex<double>for analysis code; usefloatand library FFTs only after the behavior is correct. - Validate with round trips and invariants: DFT then IDFT, DCT then inverse DCT, convolution direct versus frequency-domain multiplication, known harmonic peaks, expected alias folds, or known autocorrelation lag.
- Only then optimize for real-time constraints: preallocate, avoid locks and allocations in audio callbacks, smooth parameters, and replace direct DFT examples with an FFT library.
Reference Map
Read only the files needed for the current task:
| Reference | Read when |
|---|---|
references/thinkdsp-concepts.md | Need the Think DSP chapter map, concept summaries, equations, or task routing. |
references/cpp-patterns.md | Need C++ translations of the book's Python/NumPy examples. |
references/engineering-checks.md | Need implementation checks, scaling conventions, test cases, common DSP failure modes, or coherent-tone/sub-sample-phase error measurement. |
assets/thinkdsp.hpp | Need a compact educational C++ header to copy into a prototype or use as a reference implementation. |
C++ Defaults
- Use
doublefor offline analysis and examples; move tofloatin real-time paths when tests pass. - Use
std::complex<double>for DFT, DCT derivations, transfer functions, modulation, and phase-aware work. - Name sample rate
sampleRate, frame countn, time samplests, amplitudesys, complex binshs, and bin frequenciesfs. - Define DFT as
X[k] = sum_n x[n] * exp(-i * 2*pi*k*n/N)and inverse DFT asx[n] = (1/N) * sum_k X[k] * exp(i * 2*pi*k*n/N). - For a real sinusoid exactly on bin
k, estimate amplitude as2 * abs(X[k]) / N, except DC and the Nyquist bin. - Use a proven FFT for production: FFTW, KissFFT, pffft, JUCE
dsp::FFT, vDSP, IPP, or the host framework's FFT. The direct DFT examples in this skill are for clarity and small tests.
Engineering Rules
- Check Nyquist before synthesis, analysis, downsampling, or modulation. A component above
sampleRate / 2aliases unless removed before sampling. - Window finite segments before spectral analysis unless the segment is exactly periodic. Hamming is a good default; choose other windows deliberately.
- Distinguish circular convolution from linear convolution. Zero-pad before FFT convolution when the signal should not wrap.
- Treat bin 0 as DC/bias. Differentiation destroys it, and integration cannot recover it without a constant of integration.
- Treat frequency-domain filters as complex operations. Magnitude changes amplitude; angle changes phase.
- Use spectrogram segment length as a resolution tradeoff: time resolution is
N / sampleRate; frequency resolution issampleRate / N. - Validate noise by distribution, serial correlation, and log-log power slope. White is slope near 0, pink is near
-beta, Brownian/red is near -2.
How can the creator link this skill?
Add the canonical catalog link to the repository README so users can inspect current installs and available audits. The publishing guide covers the complete discovery path.
<a href="https://skillzs.dev/skills/shortwavlabs/agent-skills/dsp-engineer">View dsp-engineer on skillZs</a>