PeakLab v1 Documentation Contents            AIST Software Home            AIST Software Support

BEADS Baseline

BEADS: Baseline Estimation and Denoising using Sparsity

BEADS represents a modern, non-convex optimization approach to signal processing. Unlike Whittaker or morphological filters, BEADS performs simultaneous baseline correction and denoising by decomposing the signal into three distinct physical components.

The Sparsity Model

The algorithm assumes the observed signal ($y$) is the sum of:

  B (Baseline): A low-pass component representing slow, undulating drift.

  P (Peaks): A "sparse" component representing analytical peaks (non-zero only in small regions).

  N (Noise): High-frequency stationary white noise.

BEADS Parameter Overview

  Fc (Cutoff Frequency): This is the most critical parameter. It defines the "speed limit" for the baseline. A lower $Fc$ results in a stiffer baseline that ignores peaks, while a higher $Fc$ allows the baseline to follow faster drift.

  lambda_0 (Peak Sparsity): This parameter enforces the "sparsity" of the analytical peaks. It ensures that the peak signal returns to exactly zero in the baseline regions.

  lambda_1 (Baseline Slope): Penalizes the first derivative of the baseline. It prevents the baseline from having sudden "steps" or steep jumps.

  lambda_2 (Baseline Curvature): Penalizes the second derivative. It ensures the baseline transitions are "graceful" and smooth, similar to the $\lambda$ effect in Whittaker smoothing.

  Asym (Asymmetry): Since chromatographic peaks are unipolar (positive-only), this parameter weights the penalty to prevent the baseline from "dipping" below the noise floor during peak regions.

BEADS Parameter Interpretation

BEADS is a "Sparsity-based" approach. It doesn't just smooth; it attempts to decompose the signal into components. Because it uses a non-convex solver, the five parameters are highly interactive.

Usage Considerations

BEADS is ideal for high-noise environments where you wish to remove the baseline and "clean" the signal in a single mathematical pass without distorting peak shapes. BEADS has been reported to be highly effective for Raman spectral data and for complex LC-MS total ion chromatograms.

Optimization of Parameters

The BEADS algorithm's five parameters can be very challenging to tune and we strongly recommend that you use PeakLab's baseline genetic optimization to train BEADS to mimic your own human designed baseline. The BEADS algorithm requires careful tuning of the filter cutoff frequency and the sparsity regularization parameters.


This is an example of a BEADS algorithm for a natural product GC separation where there is a raised baseline. The BEADS optimization tracks the lower limits of the data in creating a rolling ball/convex hull/rubber band kind of baseline. In natural product separations, the "raised baseline" is often not an instrument artifact but a "forest" of unresolved trace peaks. BEADS is ideal here because its sparsity model treats the forest as noise while the asymmetry identifies the true chemical zero.
 

Comparison with Whittaker Fitted Baselines
 

Core References

Ning, X., Selesnick, I. W., & Duval, L. (2014). Chromatogram baseline estimation and denoising using sparsity (BEADS). Chemometrics and Intelligent Laboratory Systems, 139.