Signal Kernels

Analyze information flow with C++23 signal, entropy, causality, and forecasting kernels.

Signal Kernels is a header-only C++23 library for scientific signal processing and telemetry analytics. It includes entropy measures, divergence metrics, causal tests, change-point detection, FFT helpers, graph curvature, and forecasting primitives.

Why it matters

Large AI and research systems need reliable measurement kernels before a model interprets noisy data. This repo provides public, testable primitives that can feed receipt-backed scientific and operational workflows.

Try it

cmake -S . -B build -DSIGNAL_KERNELS_BUILD_TESTS=ON
cmake --build build --config Debug
ctest --test-dir build -C Debug --output-on-failure

What to test first

• Build the CMake project with tests enabled.
• Run CTest.
• Include algorithms/entropy.hpp or another module in a small C++ target.

Current status

Header-only C++23 library with unit tests. The bundled CMake setup targets Windows x64/MSVC, while the headers are standard-library-only.

Existing technical notes

Header-only C++23 signal/information-theory library: entropy, MI/transfer entropy, divergences, Granger, PELT, FFT, and forecasting.

Overview

signal-kernels implements a focused set of well-known, published analytics primitives for entropy, forecasting, causal analysis, change-point detection, graph curvature, and spectral analysis. It depends only on the C++ standard library and is intended as a reusable foundation for telemetry analytics and scientific signal processing.

Modules

• algorithms/entropy.hpp -- Shannon, Rényi, Tsallis, min, block, spectral, and permutation entropy.
• algorithms/information.hpp -- mutual information, transfer entropy, and KL / Jensen-Shannon / Hellinger / Wasserstein divergences.
• algorithms/causal.hpp -- Granger causality test.
• algorithms/changepoint.hpp -- PELT (Pruned Exact Linear Time) change-point detection.
• algorithms/forecast.hpp -- SARIMA and VAR time-series forecasting.
• algorithms/curvature.hpp -- Forman-Ricci and Ollivier-Ricci graph curvature.
• algorithms/_fft.hpp -- radix-2 Cooley-Tukey FFT (internal).
• algorithms/_numeric.hpp -- Welford variance, log-sum-exp, small-matrix linear algebra, autocorrelation, Yule-Walker (internal).

Why this is publishable

• No offensive-action primitives (no exploitation, command execution, persistence, lateral movement, or credential tooling).
• No secrets, credentials, keys, or operator-specific identifiers.
• Standalone build system and public API header set.
• Unit tests are included for all headers.

Usage

Add signal-kernels to your CMake project and link the signal-kernels INTERFACE target:

add_subdirectory(signal-kernels)
target_link_libraries(your_target PRIVATE signal-kernels)

Then include the headers you need:

#include "algorithms/entropy.hpp"

double h = signal_kernels::algorithms::shannon(probs);

See USAGE.md for a full walkthrough -- the public function/class list per header, worked examples with expected output, and build notes. A single end-to-end program lives at examples/demo_pipeline.cpp.

Platform

The bundled CMakeLists.txt targets Windows x64 / MSVC only and stops with a fatal error on other platforms. The library is header-only and uses only the C++23 standard library, so the headers can be compiled with other conforming toolchains if you bypass the bundled CMake configuration.

Building and testing

cmake -S . -B build -DSIGNAL_KERNELS_BUILD_TESTS=ON
cmake --build build --config Debug
ctest --test-dir build -C Debug --output-on-failure

Tests are gated on SIGNAL_KERNELS_BUILD_TESTS (defaults to ON only when signal-kernels is the top-level project). They use doctest: a vendored copy at tests/third_party/doctest/doctest.h is used if present, otherwise doctest v2.4.11 is fetched via FetchContent. This test-only dependency does not affect consumers of the header-only library.

License

MIT. See LICENSE.

---

Zain Dana Harper -- small tools with explicit edges. Portfolio · HarperZ9 _{Built with Claude Code; reviewed, tested, and owned by me.}

For developers

Keep the public README, build notes, and examples aligned with current behavior. Before opening a PR or pushing a release, run the local native verification path.

cmake -S . -B build
cmake --build build
ctest --test-dir build --output-on-failure

See AGENTS.md for the repo-specific operating boundary and CHANGELOG.md for current delivery status.

---

Zentropy Labs · order out of entropy. An independent lab building evidence-first tools that leave a re-checkable artifact behind. Built by Zain Dana Harper in Seattle. The full workbench is at Project Telos.