Chromatography Theory Chapter 3 – Dealing with Noise by Smoothing
- Chromperfect
- Sep 24
- 2 min read
This chapter explains how detector noise impacts peak detection and quantification, and how chromatography smoothing the chromatogram (before integration) reduces those effects. It also clarifies what noise is (vs. drift), how to measure it, and when smoothing is appropriate in Chromperfect (v8 shown; theory is universal).
What is “noise” (vs. drift)?
Noise: rapid fluctuations on a timescale shorter than peak width.
Drift: slow baseline changes on a timescale longer than peak width.Smoothing reduces noise only. Removing drift requires subtracting a baseline chromatogram.
Measuring noise & S/N
Visual “peak-to-peak” noise overestimates randomness; prefer RMS noise measured on a flat baseline region.
Signal-to-Noise (S/N) = peak height ÷ RMS noise. As S/N falls, noise distorts baselines and results.
Practical thresholds: LOD (detectable peak) and LOQ (accurately integrated peak) can be expressed in amount units per analyte.
Chromperfect can measure noise during acquisition and in analysis; it can also auto-measure via a timed event (advanced).
How noise affects integration
On noisy traces the algorithm contracts/lowers baselines to avoid penetration by noise spikes.
Result: peak height and area are biased high (example: +8% height, +17% area for a narrow peak).
Noise near the peak top affects height; noise anywhere on the peak affects area. Baseline lowering is the major contributor.
Chromatography Smoothing, What is smoothing?
Replace each data point with a weighted average of neighbors (a convolution with a window). Needs adequate sampling rate.
Algorithms supported (Chromperfect):
Rectangular (moving average): simplest; reduces height, broadens peaks; area ~constant.
Triangular: central point weighted most; like two rectangular passes at half width.
Hamming: sinusoidal taper; less broadening than rectangular/triangular.
Savitzky–Golay (SG): polynomial fit; minimizes peak broadening; preserves areas and widths best; good for high-frequency noise; also provides derivatives (1st derivative zero at peak apex; 2nd derivative zeros at inflection points).
Median (non-linear): replaces with window median; files off outliers; slight loss of height/area; width unchanged.
Key control: Smoothing time (window width).
Too small → residual noise.
Too large → peak distortion, resolution loss.
Aim for a window spanning several noise cycles, still narrow vs. peak width.
Effects of smoothing (illustrative)
Example with heavy rectangular smoothing (1.5 s) shows ~82% noise reduction.
Areas for narrow peaks move closer to true values (e.g., excess area halved for the narrowest peak).
Very narrow peaks may lose height and widen; broader peaks are less affected.
Choosing an algorithm
For moderate noise and wider peaks, algorithms perform similarly.
For high noise with larger windows, rectangular may over-broaden; prefer Triangular, Hamming, or especially Savitzky–Golay.
SG is excellent for high-frequency noise and derivative work; less effective for low-frequency pump noise.
When to use smoothing
Baseline subtraction workflows: smooth the baseline chromatogram (no peaks → large window safe) to avoid adding noise to the analytical run.
Wide-range calibration: smoothing can stabilize low-amount points so the curve passes QC without discarding data or multiplying replicates.
Otherwise, don’t smooth unless needed.
How to apply in Chromperfect (v8)
Experiment in the Raw File Editor: try algorithms and window sizes; linear smooths accumulate over multiple passes; median does not (re-applying with same/smaller window has no effect).
Once set, store choices in the method so all processed raw files are smoothed consistently.
Comments