# Optimize spectrum

When you look at a spectrum for the first time, it may appear noisy. This is typically the result of collecting too few electrons and the spectrum has excessive shot noise. A common reason is the exposure time may be small due to settings from the previous spectrum (e.g., the core-loss was very bright). Increase the exposure time until the detector starts to saturate. Auto exposure mode is recommended.

To record a high quality spectrum, follow the steps below.

1. From the EELS palette, choose the correct camera setting. HDR is best for most single spectrum acquisitions where speed is not a concern. The signal-to-noise ratio (SNR) will give the lowest readout noise however.

2. Within the EELS Acquire Setup dialog, on the Options tab, check the Auto-align Summed Spectra box to sum together many short exposures to improve the energy resolution. This will have limited effect at low energy resolution dispersion, but can sharpen the ZLP and low-loss spectra at high dispersion. Apply HQ dark correction should always be enabled and will greatly reduce the noise in summed spectra and STEM spectrum images.

3. From the EELS pallet, choose an appropriate Sum Time. This will be the total live time the spectrum is integrated. If the Auto feature is enabled for capture, the single readout exposure with be determined upon pressing the capture button; otherwise, the view exposure will be used. The acquisition will continue until the live time reached. If sufficient SNR is reached before the live time is reached, the operator can press the Capture again button to finish the acquisition.

Once you increase in the integration time, there will be a notable improvement in the spectrum noise. The system is now ready to run an EELS quantification.

## Noise sources

### Shot noise

There are various sources of noise that commonly impact an experiment’s SNR. One source is shot noise, which results from the inherent statistical variation in the arrival rate of electrons on the detector.

​Shot noise = $$N{_{E}}^{\frac{1}{2}}$$ where $$N_{E}$$ is the expected number of electrons detected.

To reduce shot noise, your only recourse is to acquire more electrons, consider:

• Increasing the beam current

• Increasing exposure – Stop when the CCD saturates

• Summing frames

### Readout noise

This usually displays as a fixed noise value each time a pixel is readout. This noise is inherent to the process of converting CCD charge carriers into a voltage signal, as well as the subsequent processing and analog-to-digital conversion. For strong signals, this noise is insignificant, but sets a noise floor or fog level for weak signals. To mitigate this effect, increase the binning so fewer pixels are readout.

### Dark noise

This minor contributor to noise arises from statistical variation in the number of electrons thermally generated within the silicon structure of the CCD. Verify the detector is cooled to help avoid this source of noise.

### Poor gain correction

Occasionally you will encounter fixed patterns in the spectra that may be due to small errors in gain correction. When this type of redundancy occurs, consider these variations to your system setup.

Note: The noise from the gain correction increases as the number of frames summed in increased. It is recommend that your gain correction reference have at least as many frames as you are summing in your spectrum. This is not always practical, but at least 25 frames should be summed in the gain reference.