Engineers searching “best filter for oscillatory sensor noise” often expect a low-pass shootout. GSRF is different: it is tuned to clamp amplitude toward a normal and kill mid-band ring— not to track slow waves like EMA.
Gain plateau (pure sine)
RMS AC gain vs period on locked pure-sine characterization (see chart on Evidence §2, CSV-backed SVG):
- EMA → gain approaches 1.0 on slow waves (tracks)
- GSRF Practical → gain plateaus near ~0.37 (clamps toward normal)
If your job is “follow the slow process variable,” EMA is the honest tool. If your job is “stop mid-band chatter around a known operating point,” the plateau is the product.
Sweet band & oscillation metric
| Result | Number | Notes |
|---|---|---|
| osc31 vs best EMA | ~+35% | phase6.2 · 3 seeds · full series |
| Sweet-band spectral kill | ~−70% | Same packs · not rolling-31-only |
| Frequency-sweep sweet spot | P ≈ 50–120 | Peak osc adv ~+51% at P=80 |
osc31 = std of (signal − rolling mean of length 31). Lower is calmer high-pass energy. Glossary: /glossary.
Where EMA still wins on osc31
Official frequency sweep: GSRF loses osc31 at very fast periods (~P=20) and very slow periods (~P=400–600) versus best EMA. Slow “oscillation” that is really drift you want tracked is an EMA job.
Sampling-rate translation
“Period 80” means eighty samples, not eighty seconds by magic. On 1-minute industrial traces, P=80 ≈ 80 minutes of ring timescale. Always map the sweet band to your Δt before buying a preset.
Verdict
Target mid-band ring near P≈50–120 samples. Expect gain clamp, not full slow-wave tracking. Validate with /try and your plant metric before production license.
Sources: characterization v4 sine_frequency_response · probes v5 FULL PSD · phase6.2 · /evidence#frequency · related: GSRF vs EMA mid-band