Signal Tools

Why this matters

Sensor streams, time series, and audio all hide their structure in the frequency domain, and an LLM cannot compute an FFT or design a filter. Signal tools give an agent an exact, O(n log n) FFT, a one-sided power spectrum with the dominant frequency picked out, and windowed-sinc FIR filters to isolate or remove frequency bands — all deterministic and computed natively, then handed back chart-ready.

Transforms

• ✓signal.fft — Forward FFT of a real or complex signal (native, rustfft). Returns the complex spectrum, magnitude, phase, and per-bin frequency; one_sided for real input
• ✓signal.ifft — Inverse FFT, 1/n-normalized so ifft(fft(x)) == x; recovers the time-domain signal from a spectrum
• ✓signal.spectral — One-sided power spectrum via the native FFT: per-bin magnitude, power, and phase, plus the top peaks and dominant frequency

Filtering

• ✓signal.filter — FIR windowed-sinc filter: lowpass, highpass, bandpass, bandstop. Cutoffs normalized to the sample rate (0.5 = Nyquist)
• ✓signal.convolve — Discrete convolution (full / same / valid) for smoothing, feature detection, and applying custom kernels
• ✓signal.correlate — Cross- and auto-correlation by lag for signal alignment and periodicity detection (distinct from the math.correlate coefficient)

Use cases

• ◆Sensor & IoT analysis — Find the dominant frequency of vibration, current, or temperature data with spectral
• ◆Time-series denoising — Low-pass filter to smooth noisy series, or high-pass to isolate rapid changes
• ◆Periodicity detection — fft to expose cycles and seasonality that are invisible in the raw signal
• ◆Spectral processing — Transform with fft, edit the spectrum, and reconstruct with ifft

Example: transform, inspect, filter

// Transform a sampled signal into the frequency domain (native FFT)
{ "name": "data-grout@1/signal.fft@1",
  "arguments": { "data": "$sensor.readings", "sample_rate": 1000, "one_sided": true }
}

// Or get the power spectrum + dominant frequency directly
{ "name": "data-grout@1/signal.spectral@1",
  "arguments": { "data": "$sensor.readings", "sample_rate": 1000, "window": "hann" }
}

// Smooth the signal with a low-pass FIR filter (cutoff = 50Hz / 1000Hz)
{ "name": "data-grout@1/signal.filter@1",
  "arguments": { "data": "$sensor.readings", "type": "lowpass", "cutoff": 0.05 }
}

The records (frequency, magnitude, phase) and filtered values drop straight into prism.chart for spectrum and waveform plots.