Concepts¶
Backends and grouping¶
Many detectors operate per backend (column group by default). This keeps
instrument-specific noise characteristics separate. You can select the backend
column with backend_col in the pipeline or --backend-col in the CLI.
Some detectors are global by design (across backends), including solar-event, eclipse-event, gaussian-bump, and glitch detection.
Residuals and noise¶
Pulsar timing uses residuals between observed and modeled TOAs. In many
analyses, residuals are treated as approximately Gaussian with known
uncertainties, so weights scale as 1/sigma^2. The general framework is
standard in timing packages such as tempo2 [Hobbs2006] [Edwards2006].
Bad measurement detection¶
The OU-based bad measurement detector models short-timescale residual structure with an Ornstein-Uhlenbeck (OU) process, a mean-reverting Gaussian process defined by the stochastic differential equation:
Outliers are identified via innovation statistics and multiple-testing control. The Benjamini-Hochberg procedure controls the false discovery rate (FDR) by comparing ordered p-values to a linearly increasing threshold [BH1995].
Transient events¶
Transient exponential recoveries are modeled as
where H is the Heaviside step function, t0 is the event time, A is
amplitude, and tau is the recovery timescale. This is a common template for
recovery-like disturbances in timing data [LKH2005].
Global astrophysical events¶
PQC also supports event templates that are fit globally:
Exponential dips over MJD (optional frequency dependence
1/f^alpha)Solar elongation events (per-year fit when supported, fallback to global fit)
Binary eclipses as a function of orbital phase (global fit)
Gaussian-bumps over MJD using model comparison
Glitches using step+ramp or peak+ramp model comparison
These event members are treated as non-outliers in bad_point and are still
tracked in event_member for downstream filtering/plotting.
Step and DM-step events¶
Step events model abrupt offsets at time t0:
DM-step events scale with observing frequency f according to the
dispersion delay law:
The 1/f^2 scaling reflects cold-plasma dispersion and is standard in
pulsar timing [LKH2005].
Membership semantics¶
Step and DM-step events use two masks:
*_applicable: points in the time window (t >= t0and within the membership window).*_informative: applicable points that also satisfy the per-point SNR threshold (|model_effect| / sigma > eta).
Event identity is carried in step_id and dm_step_id and is based on
applicability. Plotting and summary defaults use informative membership.
Covariate-conditioned preprocessing¶
Detectors can operate on residuals that are detrended and/or variance-rescaled as a function of covariates. This leaves the original residuals intact while adding processed columns for selected detectors.
References¶
Hobbs, G. B., Edwards, R. T., & Manchester, R. N. (2006). “tempo2, a new pulsar-timing package - I. An overview.” MNRAS, 369(2), 655-672. citeturn0search1
Edwards, R. T., Hobbs, G. B., & Manchester, R. N. (2006). “tempo2, a new pulsar timing package - II. The timing model and precision estimates.” MNRAS, 372(4), 1549-1574. citeturn0search0
Benjamini, Y., & Hochberg, Y. (1995). “Controlling the false discovery rate: a practical and powerful approach to multiple testing.” Journal of the Royal Statistical Society Series B, 57(1), 289-300. citeturn1search0
Uhlenbeck, G. E., & Ornstein, L. S. (1930). “On the theory of the Brownian motion.” Physical Review, 36, 823-841. citeturn2search0