# Extract signal

The next step in quantification requires you to extract edge intensities from spectrum, while you disregard the underlying background intensity. To separate the edge intensity, you will need to fit, extrapolate, and then subtract a background model. It is important to consider the following items when you perform this critical extraction step:

• Which background model should I use?

• Where should I fit the background model to the data?

• What are the optimal width and position for the signal integration window?

## Background modeling

To extract the edge intensity, you must determine a model for the background of your spectrum. First identify a pre-edge fitting region that allows you to determine parameters of the fit. Then extrapolate this fit to estimate the background intensity below the edge signal. However, an accurate background subtraction may become difficult below 100 eV due to the large number of scattering processes in this region (e.g., plasmons tails, plural scattering).

Typically, the model is determined using linear least-squares methods using a single pre-edge region $$\Gamma$$. where

• $$\Gamma$$ = background fit window

• $$\Delta$$ = signal integration window

• $$I_{b}$$ = background intensity

• $$I_{k}$$ = signal intensity

### Power law

A power law is the most common background model.

$$J\left ( E \right )=AE^{-r}$$

where

• $$A$$ = scaling constant

• $$r$$ = slope exponent (usually 2 – 6)

This model has the physical basis when interpreted as the long energy tails of the preceding energy loss events. ### Background model placement

When you choose the optimal background placement, it is important to consider these parameters:

• The high energy side $$E_{be}$$ should be as close to but still preceding the edge (e.g., 5 eV) to avoid chemical shifts and broadening detector tails

• To limit statistical error, fit region $$\Gamma$$ should be as wide as possible

• To limit systematic error, you need to limit the fit region to 10 – 30% $$E_{k}$$ ### Rules of thumb to follow

• Background window end should be 5 eV from edge onset

• Background window width should be at most 30% edge energy

• May need to limit window size to avoid preceding edges where necessary

### Troubleshoot background extrapolation errors

Once background placement is made, it is important to review common errors.

• Unphysical

• Symptom – Obvious error where the background model crosses the spectrum and may cause the signal to become negative

• Solution – Increase the window size and/or offset it from the edge onset; you may need to limit the extrapolation distance of your analysis • Systematic errors

• Symptom – Small changes in the background window width or position have large effects on the background model

• Solution – Ensure small variations in window position do not change background fit significantly; increase size of window The background in this case changes rapidly as the window position is moved. A larger window and avoiding the edge onset region are recommended.
• Overlapping edges

• Symptom – Background extrapolation is ineffective for instances where the pre-edge region is obscured by the preceding edge

• Solution – Reduce the window size or placement as well as limit the signal window size and offset; a multiple linear least-squares (MLLS) fitting or model based approach may be necessary ## Width and position of signal integration window

When you choose the optimal signal integration window placement, it is important to consider:

• Statistical error – The region $$\Delta$$ should be as wide as possible and start at the steepest intensity increase of the spectrum

• Hydrogenic edges (e.g., K-, some L-edges) – Place window at threshold

• Delayed edges (e.g., L-, M-, N-, O-edges) – Offset by a few tens of eV

• Systematic error – Limit fit region to about 10% $$E_{k}$$, but it should cover all of the significant energy loss near edge structure (ELNES) changes White lines – Best to avoid inclusion for quantitative evaluation as their intensity can vary with chemical state, and are not well modeled in cross-section calculations ## Signal extraction with GMS 3 software

With Gatan Microscopy Suite® (GMS) 3 software, the process of signal extraction is highly automated. However, the guidelines and concepts above still must be considered. GMS 3 quantification utilizes a model based approach where the spectral background and the edge intensity are treated as single model. If there are overlapping edges present, they are also added to the model to allow separation of the overlap. Follow the below steps for EELS signal extraction in GMS 3 software.

1. Identify the edge features within the spectrum
2. Show fit regions on spectrum using the Show signal setup button in the Elemental Analysis window.
3. The edge model will be shown on the spectrum. Default values are typically adequate, but the regions of interest can be dynamically changed by the user if needed.
4. The EELS edge setup button allows specific set up for each edge:
1. Exclude ELNES – Removes the near edge structure from the analysis
2. Include plural scattering – Enables linking to low-loss spectrum and is required for absolute quantification
3. Most settings dynamically update if the fit region is adjusted on the spectrum Exclude ELNES allows the model to begin after a predetermined ELNES region. This leads to more consistent fitting to the calculated cross-section and should be used when fitting strong edges. For semi-quantitative mapping, the ELNES region should be included to get the best signal-to-noise (SNR) in the maps.
1. Edges that are close in energy are automatically tagged for overlap analysis. You can change this default using the Overlaps check box in the edge setup dialog.
2. The fit to the edge model less the models for the background and any proceeding edges yields the extracted signal intensity for that edge Overlapping edges are fit together as a single model. The total signal for the overlapped edge includes contribution from the preceding edge background.

## References

Joy, D.C.; Maher, D. M. The quantization of electron energy-loss spectra. J. Microsc. 124:37 – 48; 1981.

Egerton, R. F. A revised expression for signal/noise ratio in EELS. Ultramicroscopy. 9:387 – 390; 1982.

Leapman, R. D.; Swyt, C. R. Separation of overlapping core edges in electron energy loss spectra by multiple-least-squares fitting. Ultramicroscopy. 26:393 – 404; 1988.

Kothleitner, G.; Hofer F. Optimisation of the signal to noise ratio in EFTEM elemental maps with regard to different ionisation edge types. Micron. 29349 – 357; 1998.

Verbeeck, J., Van Aert, S. Model based quantification of EELS spectra. Ultramicroscopy. 101(2 – 4):207 – 24; 2004.

Riegler, K.; Kothleitner, G. EELS detection limits revisited: Ruby – a case study. Ultramicroscopy. 110(8); 2010.

Thomas, P.; Twesten, R. A Simple, Model Based Approach for Robust Quantification of EELS Spectra and Spectrum-Images. Microscopy and Microanalysis. 18(S2):968 – 969; 2012.