Noise

The noise command estimates noise levels in XAS data by analyzing the difference between even and odd data points. Unlike other manipulation commands, this is non-destructive — it creates a new column without altering existing data.

Basic Usage 

noise [options]

All options are optional; defaults typically work well for most XAS data.

Command Options

Option Description
--xaxiscol <axis> Axis column used for even-odd pairing (default: E).
--yaxiscol <data> Data column to analyze for noise (default: a).

Noise Estimation Method

The command uses the even-odd difference method:

  1. Pairs consecutive data points: (point 0, point 1), (point 2, point 3), …
  2. Computes a regression-based difference between even and odd subsequences
  3. Estimates noise standard deviation from these differences

This method is robust because:

  • True signal varies smoothly and cancels out in even-odd differences
  • High-frequency noise (point-to-point fluctuations) is preserved
  • It works well even with correlated noise

Output

The command creates a new data column named s<yaxiscol>:

  • If --yaxiscol a (default): creates column sa
  • If --yaxiscol chi: creates column schi
  • If --yaxiscol mu: creates column smu

This column contains the estimated standard deviation (noise level) at each point.

Examples 

# Estimate noise in default absorption column 'a'
noise

# Estimate noise in chi column
noise --yaxiscol chi

# Use k-axis for even-odd pairing
noise --xaxiscol k --yaxiscol chi

# Estimate noise in normalized mu
noise --yaxiscol mu

Use Cases

  • Data quality assessment: Identify noisy regions in scans
  • Weighting for fitting: Use noise estimates to weight data points in EXAFS fitting
  • Experiment comparison: Compare noise levels across different measurement conditions
  • Outlier detection: Identify regions where signal-to-noise ratio is poor
  • Visualization: Plot noise alongside data to assess quality

Tips and Best Practices

  1. Run early in workflow: Estimate noise before normalization or background removal to assess raw data quality

  2. Multiple estimates: Run noise on different columns (a, chi, mu) to understand how processing affects uncertainty

  3. Visual inspection: Always plot noise alongside data:

    noise
plot E:a --figure 1
plot E:sa --figure 1  # Plot absorption with noise overlay

See also: