Multi-spectral Xrf Counting: Squeeze Twice as Much Information from Your Detector

Counting small peaks on large backgrounds in XRF applications is a process limited by the statistics of estimating both the counts and the background in the peak region. Improving energy resolution ΔE helps by increasing peak/background ratio and providing a more accurate estimate of the background but also reduces counting statistics due to increased pileup losses. Time variant filtering techniques have not been useful because their detector response functions vary with counting rate, precluding the accurate use of standards. Here we present a new time variant approach that takes a small set of shaping filters, applies the longest one to each detector output pulse, and places the result into a spectrum specific to that filter. By not commingling the filters' results, we produce multiple spectra whose detector response functions do not depend upon input count rate and can thus be used with standards-based analyses. Both theoretical and experimental measurements show that this approach allows the filters' information to be fully preserved, so that an optimized multi-spectral spectrometer can achieve the same signal to noise ratio in approximately half the counting time required by an optimized single filter spectrometer. INTRODUCTION A common requirement in XRF measurements is to determine the spectral line intensity IL of a weak peak sitting on a large background. Figure 1 indicates the situation schematically. The desired value IL is the sum of the counts in a region of interest (ROI) P2 that includes the peak minus an estimate B of the background obtained from regions B1 and B2, whose extent is typically limited by neighboring peaks that are separated from the peak of interest by values Es. In difficult cases, where Es is less than R, deconvolution methods will be required. R's value scales with the spectrometer’s energy resolution ΔE. For example, for R to include 95% of the peak area, its value would be 1.76 ΔE. In general the statistical accuracy of IL can be improved both by collecting more counts, making both the ROI area and background estimates more accurate, and by reducing ΔE, which both reduces the number of background counts in ROI P2 and increases the regions B1 and B2 over which B can be measured. Thus, while counting time to a required signal to noise ratio (SNR) can be reduced by increasing output counting rate (OCR) and reducing ΔE, these two variables cannot be adjusted independently for a given detector because increasing the OCR requires decreasing the spectrometer’s peaking time τp, which increases ΔE. OCR is related to the input count rate ICR by the dead time formula OCR = ICR exp(-τd ICR) , where τd = 2(τp + τg) in a digital spectrometer, τg being a trapezoidal filter’s flat top time. Figure 2 shows data from an 80 mm HPGe detector, where counts were collected for 10 sec at the point of maximum 0 100 200 300 400 0 200 400 600 80