Preparing and checking data

Before retrieving, establish what the file records: the signal, its spectral coordinate and units, the delay axis and its sign, and any independently measured spectrum. Then remove what the model does not describe (fringes, background, out-of-band noise) and put the trace on the retrieval grid. This chapter covers measured files, numerically simulated scans, and the marginal checks that test a trace for consistency before any retrieval is run.

Which page do I need?

I want to…	Read
load a measured trace and clean, window and regrid it	Cleaning and regridding measured data
remove interferometric fringes or an asymmetric baseline	Cleaning and regridding measured data
load a simulated scan and score a retrieval against its known pulse	Loading numerically simulated traces
test a trace against an independently measured spectrum	Marginal consistency checks
understand why the marginals constrain the trace	FROG trace marginals

• Cleaning and regridding measured data
• Loading numerically simulated traces
• Marginal consistency checks
• FROG trace marginals as consistency checks

Key concepts

Preprocessing produces a TraceData: a normalised trace on a uniform frequency and delay grid, with the carrier and the delay origin recorded. Conversion from a wavelength axis applies the \(\lambda^2\) density Jacobian. The frequency marginal of a trace is predicted by the pulse spectrum (exactly for SHG, as a transform-limited estimate for PG/TG/SD), so a mismatch flags a calibration or spectral-response problem before it biases the retrieval. A third-order efficiency exponent corrects a wavelength-dependent detection efficiency.

Main entry points

• list_datasets(), unit_to_si(): pick datasets from an HDF5 or NPZ file.

• load_and_clean(): the whole cleaning pipeline in one call.

• defringe_carrier(), arpls_baseline(): the individual cleaning steps.

• read_simulated_scan(): read a simulated scan with its truth fields.

• predicted_marginal(), auto_third_order_exponent(): marginal checks.

The experimental workflow runs the sequence end to end, and the data preparation examples show fringe and baseline removal. Next, model the measurement.