The background ladder: which stage removes what
Background controls act on different quantities and representations. Choose them according to the observation being analysed: an accidental count level, a declared reference count spectrum, a physical sample/holder signal, or a spectral display feature. These are not interchangeable estimates of one rate. See Background correction for the contribution and scale assumptions needed for count subtraction and for additive physical modelling advice.
A steady, uncorrelated count rate is one source of background in raw histograms. It can matter at continuous sources and in long-time pulsed-source data (Blundell, De Renzi, Lancaster, and Pratt, Muon Spectroscopy: An Introduction). Its effects after a nonlinear asymmetry ratio or a spectral transform depend on that operation. A physical holder polarisation instead follows its own relaxation or precession; exposure alone cannot determine its amplitude.
The ladder at a glance
Stage |
Operates on |
Removes |
When |
|---|---|---|---|
Grouping background mode (backgrounds) |
raw counts, pre-asymmetry / pre-FFT |
a flat count level or a declared reference spectrum (mutually exclusive modes, applied once) |
enable for supported reduced-data workflows |
Count-fit background nuisance (Count-domain fitting (α calibration & single histograms)) |
raw counts, in the fit |
a flat rate fitted jointly with the count model — independent of the grouping correction |
when fitting full counts, including their accidental background |
σ-clip baseline (Robust baseline offset) |
spectrum, post-FFT, display |
the spectral noise floor (a positive pedestal of redistributed counting noise) |
when measuring peak heights or areas above a true zero |
MaxEnt SpecBG (Fourier analysis) |
spectrum, display, ZF/LF |
the zero-frequency central peak of a ZF/LF field distribution |
ZF/LF field distributions with weak satellite structure buried under the central peak |
Stage 1 — the grouping background mode (pre-asymmetry, pre-FFT)
The grouping dialog’s Background selector offers Fixed values, Range average, and Tail fit for flat count levels, and Background run for a declared count spectrum. The modes are mutually exclusive; a reference spectrum can be time dependent. Subtraction acts on grouped counts before the asymmetry ratio. See Background correction for mode selection and uncertainty assumptions.
Supported asymmetry and grouped Fourier/MaxEnt preparation use the current count-background operation. Fourier input is rebuilt from the current grouping; its background status describes that preparation. Grouped/raw count-fit inputs and the Individual Groups display have a separate contract: they retain full count backgrounds and reject active reference subtraction. They do not consume the reference-subtracted group sums. See Count-domain fitting (α calibration & single histograms).
When to enable. Set a grouping background mode whenever the uncorrelated rate is non-negligible — routinely for continuous-source data (Range average), and for pulsed data analysed to long times (Tail fit), where even the small residual rate biases weak late-time relaxation.
Stage 2 — the count-fit background nuisance (in the fit, on raw counts)
Count-domain fitting (Count-domain fitting (α calibration & single histograms)) fits a flat background
term bg jointly with the count model \(N_0 e^{-t/\tau_\mu}[1 + s A
P(t)] + \mathrm{bg}\). This term is fitted against the raw counts.
Warning
The grouping background correction never reaches the count fit. The
count fit consumes full grouped counts, with deadtime correction only when
enabled and available. Poisson fitting requires eligible raw observations.
Active Background run subtraction is rejected even if the reference is
loaded; disable background correction or choose a non-reference mode. A
user who believes the grouping correction has already removed the flat rate,
and therefore fixes bg = 0, will bias \(N_0\) and α: the raw counts
still contain the full background. Let bg float (or seed it from the
grouping value) — do not assume Stage 1 has done the job for the count fit.
This is not a double subtraction; it is the opposite trap — a stage that looks downstream of the grouping correction but is not. The two are separate estimates of the same flat rate: Stage 1 stores it in the grouping and applies it to the reduced data; Stage 2 measures it from the raw counts the fit actually sees.
When to enable. Whenever you fit raw counts and want the background
determined from the same data and weighting as the rest of the count model —
the statistically clean route when a separate grouping estimate is unavailable
or untrusted. The fitted value is a legitimate measurement of the steady rate
and can be promoted into the grouping as a Fixed value (see
Count-domain fitting (α calibration & single histograms)).
Stage 3 — the σ-clip baseline (post-FFT, display)
A power or magnitude spectrum sits on a positive pedestal from the redistributed counting noise — distinct from the flat time-domain rate of Stages 1–2. The Fourier panel’s robust σ-clip baseline estimates that pedestal and subtracts it from the displayed spectrum, leaving sharp peaks intact (they are rejected as outliers during the estimate). It is a display-channel operation: it never alters the underlying transform, and its converged width doubles as the noise estimate behind the signal-to-noise readout. See Robust baseline offset in Frequency-domain conditioning for the σ-clip mechanism and its pitfalls.
When to enable. Whenever peak heights or areas must be measured above a true zero — comparing intensities across runs, integrating a line. It is orthogonal to Stages 1–2: subtracting the time-domain rate flattens the low-frequency ramp, while the σ-clip baseline removes the spectral noise floor that remains.
Stage 4 — MaxEnt SpecBG (post-reconstruction, display, ZF/LF)
In zero or longitudinal field the MaxEnt reconstruction returns a field distribution \(p(B)\) dominated by a strong zero-frequency central peak, which can bury weak satellite structure. The Zero-frequency background (SpecBG) control (ZF/LF mode only) subtracts a zero-centred pseudo-Voigt model of that central peak from the displayed spectrum. Like the σ-clip baseline it is display-only and never alters the reconstructed spectrum; unlike it, it targets one specific feature — the ZF/LF central peak — rather than a broadband pedestal. See Fourier analysis for the SpecBG controls.
When to enable. Only for ZF/LF MaxEnt field distributions where a central peak is hiding the structure you care about. It is unrelated to the transverse-field precession lines that the other stages serve.
Why they coexist
Each control has its own purpose and input contract. Count subtraction changes the observations used to form asymmetry; a count-fit nuisance models the full accidental rate in the fitted counts. Spectral display controls change the presentation of the transformed or reconstructed data. Applying several controls does not establish that their physical assumptions hold; inspect the owning workflow before combining them. A physical sample/holder contribution belongs in the fit model when it represents the measured polarisation.
See also
Background correction — the four mutually exclusive grouping modes and how to choose among them.
Count-domain fitting (α calibration & single histograms) — the count model and its
bgnuisance term.Frequency-domain conditioning — the σ-clip baseline and other post-FFT conditioning.
Fourier analysis — the FFT background inheritance and MaxEnt SpecBG.
References
S. J. Blundell, R. De Renzi, T. Lancaster, and F. L. Pratt, Muon Spectroscopy: An Introduction (Oxford University Press, Oxford, 2022) — the steady background count from uncorrelated detector hits at continuous sources, and its suppression at pulsed sources.