Live time

[Ben Buse]

It's great we've extracted live time maps from JEOL spectrometer with Yuji code

This got me thinking more generally - an eds physics question

How much does eds live time vary between silicate minerals?

Can eds live time plot as a function of mean atomic number?

Does variations in live time between materials scale from low beam currents to high beam currents?

Can we predict live time (dead time) for different materials?

[sem-geologist]

To answer your question: it depends a lot from EDS hardware. On EDS you have two dead time sources: 1. pre-amplifier feedback cap discharge 2. pulse processing  (when other pulse arrives to soon and its energy is impossible to measure precisely). The second type of dead time will statistically scale up with increasing count rate and wont produce noticeable artifacts on the raw hyper-map counts between two subsequent pixels. And so if only second type of dead time would be there you could kinda try making live time prediction vs count rate. However first type dead time is engaged depending from energy flux. That is – if there are more higher energy pulses then capacitor will get filled much faster and reset will be needed more often. And it is also true otherwise – if there is more low energy X-rays (and also continuum is dominated by low energy X-rays) then reset of capacitor will be required less often. The first type dead time is pulsed (that is it could affect every nth pixel, where "every nth" is not precise as it depends from count rate, and pixel dwell time, thus is unpredictable).

and just to demonstrate 1st type of dead time effect on high count rate microXRF based EDS hypercube:

source: for full discussion which includes the above picture you could look in old hyperspy issue thread there:
https://github.com/hyperspy/hyperspy/issues/2898
The above picture is mildly stripped, and stripes appear vertically, but I had seen datasets with higher count rates which had diagonal stripes (irregular).

In general, if you use EDS with: low current, reduced input counts (i.e. limiting aperture in front of EDS), low voltage application – then simple approximation of count rate to live time can solve your problems good enough. However if you use very high currents, full EDS input (no aperture), high voltage (much broader and stronger Bremstrahlung emission band) –  if there is none of live time/ dead time record per every pixel then quantification of hyperspectra is simply impossible. I.e. micro XRF scanning systems has even higher count rates, and without dead time correction raw hyper maps are severely artifacted (stripes) from type 1 dead time.

And that is what I like much about Roentec/Bruker EDS solution: they inject artificial 0keV pulses before pre-amplifier at known frequency before pulse processing and both types of dead times are then being taken into account - that creates nice 0keV peak which encodes all of physically important data for EDS data which makes Bruker EDS being useful in quantification, and not just a bunch of lines and colors for awe of the user – dead time, real time, live time, fano-factor, real measurement FWHM (at 0keV, but using Fiori equation it can be resolved to any energy) – all these information are directly encoded in 0keV peak.

I am still looking forward to ThermoFisher new generation of EDS, where lots of these problems with dead time and live-time differences between different materials could be solved I think more elegantly, as they use I think FPGA directly for all pulse-shapping (or maybe they don't do pulse shapping at all - that is a complete black box with very impressive high count rate handling) and counting, thus it could work linearly even at extreme count rates.

[Ben Buse]

To answer your question: it depends a lot from EDS hardware. On EDS you have two dead time sources: 1. pre-amplifier feedback cap discharge 2. pulse processing  (when other pulse arrives to soon and its energy is impossible to measure precisely). The second type of dead time will statistically scale up with increasing count rate and wont produce noticeable artifacts on the raw hyper-map counts between two subsequent pixels. And so if only second type of dead time would be there you could kinda try making live time prediction vs count rate. However first type dead time is engaged depending from energy flux. That is – if there are more higher energy pulses then capacitor will get filled much faster and reset will be needed more often. And it is also true otherwise – if there is more low energy X-rays (and also continuum is dominated by low energy X-rays) then reset of capacitor will be required less often.

Thank you for this explanation, what I thinking is for fast x-ray maps when the live time pulse of 10msec is to slow, for each pixel we have the spectra, and from the spectra we might be able to calculate the live time using a calibration dataset - the spectra contains information about number of xrays of different energies.

[sem-geologist]

Thank you for this explanation, what I thinking is for fast x-ray maps when the live time pulse of 10msec is to slow, for each pixel we have the spectra, and from the spectra we might be able to calculate the live time using a calibration dataset - the spectra contains information about number of xrays of different energies.

You mean Jeol EDS is capable only to check for life time in 100Hz? And not only that, it use that extremely poor value independently from count rate?