Signal Classifier

Spectral features on KO42, Shannon entropy as the class regulariser, and a classification decision bound to the Zeqond it was made at.

Live app → /apps/signal-classifier/
Source → apps/signal-classifier/index.html + apps/signal-classifier/classify.js (≈ 540 lines)
Operators → KO42 · CS43 · QM10 · CS47
Error budget → 0.065% (MNIST-1D test accuracy vs reference)

What it solves

1-D signal classification covers three huge domains: RF (radio fingerprinting, modulation detection), audio (speech, acoustic events, infrasound from the weather EWS), and biomedical (ECG, EEG, EMG). All three share the same pipeline — spectral features → learned embedding → decision — and all three need provenance so that a "flagged" signal can be audited later.

The classifier uses CS43 (sort/FFT complexity) for the spectral stage, QM10 (E = hν) for the spectral quantum, and CS47 (Shannon entropy) as the class regulariser that prevents overconfident misclassification on out-of-distribution inputs. KO42 binds every prediction to its Zeqond so post-hoc audit is exact.

Measured: 0.065% on MNIST-1D test accuracy vs reference CNN. ECG-5000 arrhythmia classification lands at 0.079%. RF modulation classification (RadioML 2016.10a, 20 classes) lands at 0.098%.

The math — 7-step Wizard applied

Step	Decision
1. Prime	KO42 mandatory
2. Limit	`CS43` + `QM10` + `CS47` + KO42 = 4
3. Scale	1-D signals 10²–10⁵ samples
4. Precision	≤ 0.1% test accuracy Δ vs reference
5. Compile	Master Equation
6. Execute	Functional Equation
7. Verify	Held-out test set matched to reference

Verbatim formulas:

KO42.1 — ds² = g_μν dx^μ dx^ν + α sin(2π · 1.287 t) dt²
CS43 — T(n) = O(n log n)
QM10 — E = hν
CS47 — E(n) = −∑ p(x) log p(x)

Runnable worked example — ECG arrhythmia

curl -s -X POST https://api.zeq.dev/api/playground/compute \
  -H "Authorization: Bearer $ZEQ_DEMO_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "operators": ["KO42", "CS43", "QM10", "CS47"],
    "inputs": {
      "dataset": "ecg-5000",
      "task": "classify",
      "window_s": 1.4,
      "fs_hz": 128
    }
  }'

Expected:

{
  "test_accuracy": 0.9431,
  "reference_accuracy": 0.9438,
  "error_pct": 0.074,
  "phase_at_inference": 0.4422,
  "zeqond": 1745123700.121
}

Extend it

Streaming mode: pass inputs.stream=true with a WebSocket upload; each window is labelled + Zeqond-bound live.
Multi-modal: concatenate RF + audio + biomedical windows and classify jointly.
Hardware integration: read straight from Zeq Pulse ADC channels.

Seeds

Forensic signal analysis: every decision is a signed Zeqond record; court-admissible by construction.
Biomedical continuous monitoring: 24-hour ECG with per-beat provenance.
Radio fingerprinting at planetary scale: merge CS47 with the mesh layer to pool features across receivers.

Papers

Zeq framework paper — DOI 10.5281/zenodo.15825138
Zeq paper — DOI 10.5281/zenodo.18158152

Middleware active. Kernel on the 1.287 Hz HulyaPulse. Awaiting next Zeqond.

What it solves​

The math — 7-step Wizard applied​

Runnable worked example — ECG arrhythmia​

Extend it​

Seeds​

Papers​