Radiation detection with non-parametric decompounding of pulse pile-up

The non-parametric decompounding method addresses the challenge of pulse pile-up in radiation detection by modeling the problem as a compound Poisson process, enabling accurate energy distribution estimation in high-count-rate scenarios with reduced complexity.

CA3134143CActive Publication Date: 2026-11-10SOUTHERN INNOVATION INT
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CA3134143
Authority / Receiving Office
CA · CA
Patent Type
Patents
Current Assignee / Owner
Priority Date
2019-03-22
Filing Date
2020-03-23
Publication Date
2026-11-10
Estimated Expiration
2040-03-23

AI Technical Summary

Technical Problem

Existing radiation detection systems face challenges in accurately estimating the energy distribution of incident photons due to pulse pile-up, particularly at high count-rates, where overlapping pulses complicate the determination of individual pulse areas, leading to inefficiencies and limitations in spectroscopic analysis.

Method used

A non-parametric decompounding method is applied to model the pulse pile-up problem as a compound Poisson process, utilizing spectrum-sensitive statistics and characteristic functions to estimate the energy distribution of radiation, allowing for accurate energy estimation without relying on individual pulse identification.

Benefits of technology

This approach effectively handles pulse shape variation and high count-rates, providing accurate energy distribution estimates with reduced computational complexity, enhancing the efficiency and accuracy of radiation detection systems.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

A method of determining a spectrum of energies of individual quanta of radiation received in a radiation detector is disclosed. Spectrum sensitive statistics are computed from a time series of digital observations from the radiation detector, defining a mapping from a density of amplitudes of the pulses to the spectrum sensitive statistics. The spectrum is determined by estimating the density of amplitudes of the pulses by applying an inversion of the mapping to the spectrum sensitive statistics. The statistics may be based on a first set of nonoverlapping time intervals of constant length L at least as long as a duration of the pulses without regard to entirety of clusters of the pulses; and a second set of nonoverlapping time intervals of constant length L1 less than L also without regard to entirety of clusters of the pulses. A method of estimating count rate is also disclosed.
Need to check novelty before this filing date? Find Prior Art

Description

RADIATION DETECTION WITH NON-PARAMETRIC DECOMPOUNDING OF PULSE PILE-UP FIELD OF THE INVENTION A novel method is provided for estimating the energy distribution of quanta of radiation such as photons incident upon a detector in spectroscopic systems such as in X-ray or gamma-ray spectroscopy. The method is particularly useful for count-rate regimes where pulse pile-up is an issue. A key step in the derivation of the estimator of an embodiment of the invention is the novel reformulation of the problem as a decompounding problem of a compound Poisson process. The method can be applied to any form of radiation detector detecting quanta or other corpuscles of radiation, such as x-rays, gamma rays or other photons, neutrons, atoms, molecules, or seismic pulses. Applications of spec- troscopy from such detectors are well known. Such applications are described widely in the prior art, including in international patent applications PCT / AU2005 / 001423, PCT / AU2009 / 000393, PCT / AU2009 / 000394, PCTAU2009 / 000395, PCT / AU2009 / - 001648, PCT / AU2012 / 000678, PCT / AU2014 / 050420, PCT / AU2015 / 050752, PCT / AU- 2017 / 050514, and PCT / AU2017 / 050512. 1 BACKGROUND X-RAY and gamma-ray spectroscopy underpin a wide range of scientific, industrial and commercial processes. One goal of spectroscopy is to estimate the energy distribution of photons incident upon a detector. From a signal processing perspective, the challenge is to convert the stream of pulses output by a detector into a histogram of the area under each pulse. Pulses are generated according to a Poisson distribution whose rate corresponds to the intensity of X-rays or gamma-rays used to illuminate the sample. Increasing the intensity results in more pulses per second on average and hence a shorter time before an accurate histogram is obtained. In applications such as baggage scanning at airports, this translates directly into greater throughput. Pulse pile-up occurs when two or more pulses overlap in the time domain. As the count rate (the average number of pulses per second) increases so does the incidence of pulse pile-up. This increases the difficulty in determining the number of pulses present and the area under each pulse. In the limit, the problem is ill-conditioned: if two pulses start at essentially the same time, their superposition is indistinguishable from a single pulse. The response of an X-ray or gamma-ray detector to incident photons can be modelled as the superposition of convolutions of pulse shape functions <com:Math><semantics>Φj(t)<annotation encoding="application / x-tex">\Phi_j(t)< / annotation>< / semantics>< / com:Math> with Dirac-delta impulses, [Image disponible dans le document PDF, Image available in the PDF document] (1) The arrival times <com:Math><semantics>…,τ−1,τ0,τ1,…<annotation encoding="application / x-tex">\dots, \tau_{-1}, \tau_0, \tau_1, \dots< / annotation>< / semantics>< / com:Math> are unknown and form a Poisson process. Each photon arrival is modelled as a Dirac-delta at time <com:Math><semantics>τj<annotation encoding="application / x-tex">\tau_j< / annotation>< / semantics>< / com:Math>, and with amplitude <com:Math><semantics>aj<annotation encoding="application / x-tex">a_j< / annotation>< / semantics>< / com:Math> that is proportional to the photon energy and induces a detector pulse shape response <com:Math><semantics>Φi<annotation encoding="application / x-tex">\Phi_i< / annotation>< / semantics>< / com:Math>. The amplitudes <com:Math><semantics>ai<annotation encoding="application / x-tex">a_i< / annotation>< / semantics>< / com:Math> are realizations of identically distributed random variables <com:Math><semantics>Ai<annotation encoding="application / x-tex">A_i< / annotation>< / semantics>< / com:Math> whose common probability density function <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math> is unknown. The pulse shape function <com:Math><semantics>Φj<annotation encoding="application / x-tex">\Phi_j< / annotation>< / semantics>< / com:Math> is determined by the geometry of the detector and the photon interaction. In some systems the variation in pulse shape is minimal and may be ignored, while other systems (e.g., HPGe) individual pulse shapes may differ significantly from one another [1]. It is assumed all pulse shapes are causal i.e., <com:Math><semantics>Φj(t)=0<annotation encoding="application / x-tex">\Phi_j(t) = 0< / annotation>< / semantics>< / com:Math> for <com:Math><semantics>t<0<annotation encoding="application / x-tex">t < 0< / annotation>< / semantics>< / com:Math>, uni-modal, of finite energy, and decay exponentially toward zero as <com:Math><semantics>t→∞<annotation encoding="application / x-tex">t \to \infty< / annotation>< / semantics>< / com:Math>. The integral of the pulse shape functions are normalized to unity, i.e., <com:Math><semantics>∫−∞∞Φj(t)dt=1<annotation encoding="application / x-tex">\int_{-\infty}^{\infty} \Phi_j(t) dt = 1< / annotation>< / semantics>< / com:Math>, so that the area under the pulse is given by <com:Math><semantics>Aj<annotation encoding="application / x-tex">A_j< / annotation>< / semantics>< / com:Math>. The observed signal consists of the detector output corrupted by noise, i.e., [Image disponible dans le document PDF, Image available in the PDF document] (2) The mathematical goal of pulse pile-up correction is to estimate <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math> given a uniformly- sampled, finite-length version of <com:Math><semantics>s(t)<annotation encoding="application / x-tex">s(t)< / annotation>< / semantics>< / com:Math>. We assume throughout that the noise distribution <com:Math><semantics>w(t)<annotation encoding="application / x-tex">w(t)< / annotation>< / semantics>< / com:Math> is zero-mean Gaussian with known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. We also assume the photon arrival times form a homogeneous Poisson process with a known rate. Consisting of the detector response R corrupted by a noise process W, where <com:Math><semantics>Sk,Rk<annotation encoding="application / x-tex">S_k, R_k< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>Wk<annotation encoding="application / x-tex">W_k< / annotation>< / semantics>< / com:Math> represent the kth elements of each series and where <com:Math><semantics>0≤k<K<annotation encoding="application / x-tex">0 \le k < K< / annotation>< / semantics>< / com:Math>. Let S, R and W be the uniformly sampled time-series corresponding to these signals, [Image disponible dans le document PDF, Image available in the PDF document] (3) [Image disponible dans le document PDF, Image available in the PDF document] (4) [Image disponible dans le document PDF, Image available in the PDF document] (5) [Image disponible dans le document PDF, Image available in the PDF document] (6) [Image disponible dans le document PDF, Image available in the PDF document] Summary of Pulse Processing Methodologies: Numerous approaches have been proposed over the decades to address the issue of pulse pile-up. Approaches can be broadly catego- rized into two types: time-domain based and energy-domain based. A popular strategy is to attempt to detect when pile-up has occurred in the time domain, and either reject or compensate for the affected pulses. Early spectroscopic systems adopted a rejection- based approach along with matched filtering. The disadvantage of this approach is that an increasing proportion of pulses are rejected as the probability of pile-up grows. The system rapidly succumbs to paralysis, placing an upper limit on the count rate [1]. Strate- gies that compensate or correct for pile-up have grown in number with the increase of cheap computational power. These include template fitting [2], baseline subtraction [3], adaptive filtering [4, 5], sparse regression [6, 7] and more. These approaches all attempt to identify and compensate for pile-up in the time domain, and are generally best suited to systems with low pulse shape variation. The complexity of these approaches increases significantly with increasing variability between pulse shapes <com:Math><semantics>Φi<annotation encoding="application / x-tex">\Phi_i< / annotation>< / semantics>< / com:Math>. It can be shown that any method that attempts to characterise individual pulses will suffer from pile-up. The best that these approaches can do is to reduce the onset of pile-up. Energy-based ap- proaches attempt to address pile-up based on the statistics of an ensemble of pulses rather than individual pulses. They typically operate on histograms of estimated energy (the areas under clusters of pulses). The early work of Wielopolski and Gardner [8] and more recent extensions of their idea [9] operate primarily in the energy domain using ensemble-based strategies. Trigano et al. [10, 11] estimate the incident spectrum utiliz- ing marginal densities from the joint distribution of the statistical properties of variable length clusters of pulses where the beginning and end of each cluster is detected. This circumvents the need to identify individual pulses, and is robust to pulse shape variation. Ilhe et al.

[12] examine exponential shot-noise processes, restricting pulse shapes to a simple exponential to obtain tractable results. Further work

[13] has been done to allow a wider range of pulse shapes. In both cases, knowledge is required of the pulse shape, along with estimates of the characteristic function and its derivative. 2 SUMMARY OF THE INVENTION We chose an energy-based pile-up correction approach in order to i) avoid the limitations associated with the detection of individual pulses

[14] and to ii) be able to handle pulse shape variation without undue increase in computational complexity. Rather than uti- lizing the joint distribution [10, 11] or shot-process

[12] approaches, we recast the pile-up problem as a 'decompounding' problem of a compound Poisson process. A compound Poisson process is a discrete-time random process where each component consists of the sum of a random number of independent identically distributed random variables, where the number of random variables in each sum is Poisson distributed

[15] . 'Decompound- ing' of a compound Poisson process is the task of using the random sums to estimate the distribution from which the random variables have been drawn. Buchmann and Grübel

[16] formulated the decompounding of compound Poisson processes in the context of insurance claims and queuing theory. Decompounding of uniformly sampled compound Poisson processes has received some attention in recent times [16, 17, 18, 12, 19]. The context of these derivations frequently assume (reasonably) that each event is detectable (i.e., there is no ambiguity regarding the number of events), or that the density esti- mators are conditioned on at least one event occurring in each observation

[20] . These assumptions are of limited value when addressing the spectroscopic pile-up problem. The investigation of non-parametric decompounding without conditioning on event detection has received relatively little attention in the literature. Gugushvili

[18] proposes a non-parametric, kernel-based estimator for the decompounding problem in the presence of Gaussian noise. In embodiments of the invention, the inventors have conceived that once a method for selecting the kernel bandwidth is obtained, along with a method for transforming the observed detector output to fit the mathematical model, this estimator can be readily extended and applied to a reformulation of the spectroscopic pile-up problem. In accordance with a first broad aspect of the invention there is provided a method of determining a spectrum of energies of individual quanta of radiation received in a radiation detector, the method comprising the steps of: (1) obtaining a time series of digital observations from the radiation detector comprising pulses corresponding to the detection of the individual quanta; (2) computing spectrum sensitive statistics from the detector signal, the spectrum sensitive statistics defining a mapping from a density of amplitudes of the pulses to a spectrum sensitive statistics; (3) determining the spectrum by estimating the density of amplitudes of the pulses by applying an inversion of the mapping to the spectrum sensitive statistics. In embodiments, the spectrum sensitive statistics may be based on a sum of the digi- tal observations over a plurality of time intervals and the mapping may be defined using an approximate compound Poisson process, which may be augmented by a modelled noise. The mapping may be expressed as a relation between characteristic functions of the amplitudes, the spectrum sensitive statistics and the modelled noise. The charac- teristic functions of the spectrum sensitive statistics may be computed with the use of a histogram of the sum of the digital observations to which is applied an inverse Fourier transform. Computation of a characteristic function of the amplitudes may comprise the use of a low pass filter. In a first embodiment, the plurality of time intervals are nonoverlapping and have have a constant length L and each interval is selected to encompass zero or more approx- imately entire clusters of the pulses. This may be accomplished by requiring a maximum value of the detector signal at a beginning and end of each time interval. In this em- bodiment, the compound Poisson process may be defined as a sum of the amplitudes of the pulses in each time interval. The mapping may be expressed as defined in equations (40) and (41), which may be augmented by windowing functions. In a second embodiment, a plurality of intervals comprise a first set of nonoverlapping time intervals constant length L selected without regard to entirety of clusters of the pulses, and a 2nd set of nonoverlapping time intervals of constant length L1 less than L also selected without regard to entirety of clusters of the pulses. L is at least as long as a duration of the pulses and preferably L1 is less than the duration of the pulses. In this embodiment, the compound Poisson process may be defined as in Section 6. The mapping may be expressed as defined in Section 6. The second embodiment may utilise processes and calculations for each set of time intervals as defined for the set of time intervals in the first embodiment. In embodiments, a data-driven strategy is used that results in a near optimal choice for a kernel parameter, which minimises the integrated-square-of-errors (ISE) of the estimated probability density function of incident photon energies. According to a second broad aspect of the invention there is provided a method of estimating count rate of individual quanta of radiation received in a radiation detector, the method comprising the steps of: (1) obtaining a time series of digital observations from the radiation detector comprising pulses corresponding to the detection of the individual quanta; (2) computing spectrum sensitive statistics from the detector signal, the spectrum sensitive statistics using the intervals of constant length L and constant length L1 as described above in relation to the 1st broad aspect; (3) determining an estimate of a characteristic function of the compound Poisson process using formula (109); (4) estimating the count rate from the estimate of the characteristic function. Step (4) above may be achieved by using an optimisation routine or some other means to fit a curve, estimating a DC offset of a logarithm of the estimate of the characteristic function, or fitting a curve to the logarithm of the estimate of the characteristic function. The rest of this application is organized as follows. Sections 3, 4 and 5 relate to the 1st embodiment of the 1st aspect of the invention. Sections 3 provides preliminary background, defines notation, outlines the mathematical model and gives a derivation of the estimator of the 1st embodiment, including modifications. Section 4 shows the performance of the modified estimator of the 1st embodiment for both simulated and experimental data, and discusses the results. Section 5 provides a conclusion for the 1st embodiment. Section 6 describe the 2nd embodiment, with reference to the 1st embodiment where relevant. Section 7 describes the 2nd aspect of the invention, a novel method of estimating count rate. 3 DERIVATION OF ESTIMATOR OF THE FIRST EMBODIMENT The general approach we take to addressing pile-up is based on the following strategy; i) obtain statistics from s(t) that are sensitive to the distribution of incident photon energies, and estimate those statistics using the observed, finite-length sampled version of <com:Math><semantics>s(t)<annotation encoding="application / x-tex">s(t)< / annotation>< / semantics>< / com:Math>, ii) obtain a mapping from the density of incident photon energies to the statistical properties of the observed statistics, iii) estimate the density of the incident photon energies by inverting the mapping. Section 3.1 describes our choice of statistics. Section 3.2 argues that these statistics (approximately) have the same distribution as a compound Poisson process. Section 3.3 introduces a decompounding technique for recovering the spectrum from these statistics. It is based on the decompounding algorithm in

[18] but further developed to obtain near optimal performance in terms of the integrated square of error. Our approach to the pile-up problem follows the general theme of finding some statistics of s(t) that are sensitive to the underlying spectrum, estimating these statistics from a finite-time sampled version of s(t), then inverting the map that describes the statistical properties of these statistics given the underlying spectrum, thereby producing an estimate of the spectrum. 3.1 Choice of Statistic We wish to obtain estimates of the photon energies from the observed signal given in (2). In typical modern spectroscopic systems, the detector output <com:Math><semantics>s(t)<annotation encoding="application / x-tex">s(t)< / annotation>< / semantics>< / com:Math> is uniformly sampled by an ADC. Without loss of generality, we assume the raw observations available to the algorithm are <com:Math><semantics>{s(k):k∈ℤ≥0}<annotation encoding="application / x-tex">\{s(k): k \in \mathbb{Z}_{\geq 0}\}< / annotation>< / semantics>< / com:Math>. Since identification of individual pulses can be difficult, we look instead for intervals of fixed length <com:Math><semantics>L∈ℤ>0<annotation encoding="application / x-tex">L \in \mathbb{Z}_{>0}< / annotation>< / semantics>< / com:Math> containing zero or more clusters of pulses. Precisely, we define these intervals to be <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> where [Image disponible dans le document PDF, Image available in the PDF document] (7) [Image disponible dans le document PDF, Image available in the PDF document] [Image disponible dans le document PDF, Image available in the PDF document] (8) Here, <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math> is chosen as a trade-off between errors in the energy estimate and the probability of creating an interval. The value of <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math> should be sufficiently small to ensure the error in the estimate of total photon energy arriving within each interval is acceptably low, yet sufficiently large with respect to the noise variance to ensure a large number of intervals are obtained. Although the probability of partitioning the observed data into intervals approaches zero as the count-rate goes to infinity, this approach succumbs to paralysis at higher count-rates compared to pile-up rejection strategies based on individual pulses, since multiple photons are permitted to pile-up within in each interval. Section 4.2 describes the selection of L and <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math> for real data. Each interval contains an unknown, random number of pulses and may contain zero pulses. We estimate the total photon energy <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> using the sampled raw observations. Since the area under each pulse is proportional to the photon energy <com:Math><semantics>Ai<annotation encoding="application / x-tex">A_i< / annotation>< / semantics>< / com:Math> defined in (1), we let [Image disponible dans le document PDF, Image available in the PDF document] (9) The number of photon arrivals, the energy of each arriving photon and the detector output noise in each interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> are assumed to be random and independent of other intervals. For pulse shapes with exponential decay, a small amount of the photon energy arriving in an interval may be recorded in the next interval. The amount of leakage is proportional to <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math>, and is negligible for sufficiently small <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math>. Consequently, the estimates <com:Math><semantics>x1,x2,...<annotation encoding="application / x-tex">x_1, x_2, ...< / annotation>< / semantics>< / com:Math> may be treated as the realization of a weakly-dependent, stationary process where each estimate is identically distributed according to the random variable X. This relationship is illustrated in Fig. 1 for the noise free case using a typical pulse shape. 3.2 Approximation with Compound Poisson Process In this subsection we describe the distribution of X in terms of <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>. We will then invert this in section 3.3, to obtain an estimator for the density <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>. Using (9), (2), (1) and the fact that <com:Math><semantics>Φℓ(t)<annotation encoding="application / x-tex">\Phi_{\ell}(t)< / annotation>< / semantics>< / com:Math> is causal we have [Image disponible dans le document PDF, Image available in the PDF document] (10) As justified below, this simplifies to [Image disponible dans le document PDF, Image available in the PDF document] (11) [Image disponible dans le document PDF, Image available in the PDF document] (12) [Image disponible dans le document PDF, Image available in the PDF document] (13) Both <com:Math><semantics>yi<annotation encoding="application / x-tex">y_i< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>zi<annotation encoding="application / x-tex">z_i< / annotation>< / semantics>< / com:Math> are i.i.d. sequences of random variables. We denote their distributions by Y and Z. The distribution of Z is fully determined from the distribution of <com:Math><semantics>w(t)<annotation encoding="application / x-tex">w(t)< / annotation>< / semantics>< / com:Math>, which is assumed zero-mean Gaussian with known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. Moreover, Y is a compound Poisson process since the number of terms in the summation (number of photon arrivals in an interval of length L) has Poisson statistics. Equations <com:Math><semantics>(11)<annotation encoding="application / x-tex">(11)< / annotation>< / semantics>< / com:Math>–<com:Math><semantics>(13)<annotation encoding="application / x-tex">(13)< / annotation>< / semantics>< / com:Math> are justified as follows. The first term of <com:Math><semantics>(10)<annotation encoding="application / x-tex">(10)< / annotation>< / semantics>< / com:Math> represents leakage from earlier intervals and is approximately zero. This is easily shown for Gaussian noise by performing a Taylor expansion about <com:Math><semantics>ϵ=0<annotation encoding="application / x-tex">\epsilon = 0< / annotation>< / semantics>< / com:Math> [Image disponible dans le document PDF, Image available in the PDF document] (14) [Image disponible dans le document PDF, Image available in the PDF document] (15) Thus there is a finite but small probability that some energy belonging to a previous interval will be included in the current estimate. In practice, this contribution is com- parable to the noise for sufficiently small <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math>. The third term in (10) is zero since <com:Math><semantics>Φℓ(t)<annotation encoding="application / x-tex">\Phi_{\ell}(t)< / annotation>< / semantics>< / com:Math> is causal. The second term in (10) can be written as [Image disponible dans le document PDF, Image available in the PDF document] (16) where we assume the pulse shapes <com:Math><semantics>Φℓ(t)<annotation encoding="application / x-tex">\Phi_{\ell}(t)< / annotation>< / semantics>< / com:Math> are sufficiently smooth such that <com:Math><semantics>∑k=TiTj+L−1Φℓ(k−t)<annotation encoding="application / x-tex">\sum_{k=T_i}^{T_j+L-1} \Phi_{\ell}(k-t)< / annotation>< / semantics>< / com:Math> <com:Math><semantics>τℓ<annotation encoding="application / x-tex">\tau_{\ell}< / annotation>< / semantics>< / com:Math>) <com:Math><semantics>≈∫−∞∞Φℓ(t)dt=1<annotation encoding="application / x-tex">\approx \int_{-\infty}^{\infty} \Phi_{\ell}(t) dt = 1< / annotation>< / semantics>< / com:Math>. It approximates the total energy of all the photons arriving in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math>. Let <com:Math><semantics>νj<annotation encoding="application / x-tex">\nu_j< / annotation>< / semantics>< / com:Math> designate the number of photon arrivals in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math>. We assume <com:Math><semantics>νj<annotation encoding="application / x-tex">\nu_j< / annotation>< / semantics>< / com:Math> is a realization of a homogeneous Poisson process with rate parameter <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>, where <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> is expressed in terms of the expected number of photons per interval of length <com:Math><semantics>L<annotation encoding="application / x-tex">L< / annotation>< / semantics>< / com:Math>. Henceforth we shall assume that (11) holds exactly, and write [Image disponible dans le document PDF, Image available in the PDF document] (17) Finally, we write <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> as [Image disponible dans le document PDF, Image available in the PDF document] (18) where we assume Z has known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. In this subsection we model the statistic of section 3.1 using a compound Poisson process. This allows us to derive an estimator for the density <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math> in terms of observable quantities. The number of photons arriving in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> is a Poisson random variable which we designate <com:Math><semantics>νj<annotation encoding="application / x-tex">\nu_j< / annotation>< / semantics>< / com:Math>. The total energy in the interval Y can be modelled as a compound Poisson process i.e., [Image disponible dans le document PDF, Image available in the PDF document] (19) [Image disponible dans le document PDF, Image available in the PDF document] (20) where <com:Math><semantics>ℓj,1=min⁡{ℓ:Tj≤τℓ<Tj+L}<annotation encoding="application / x-tex">\ell_{j,1} = \min\{\ell: T_j \le \tau_\ell < T_j + L\}< / annotation>< / semantics>< / com:Math> is the index of the first photon arrival time in the interval, the arrival times are assumed ordered, and <com:Math><semantics>a(ℓi,1+k)<annotation encoding="application / x-tex">a_{(\ell_{i,1}+k)}< / annotation>< / semantics>< / com:Math> representing photon energy are independent realizations of the random variable A with density function <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>. The <com:Math><semantics>{νi}<annotation encoding="application / x-tex">\{\nu_i\}< / annotation>< / semantics>< / com:Math> form a homogeneous Poisson process with rate parameter <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>. The Poisson rate <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> is expressed in terms of the expected number of photons per interval of length <com:Math><semantics>L<annotation encoding="application / x-tex">L< / annotation>< / semantics>< / com:Math>. [Image disponible dans le document PDF, Image available in the PDF document] (21) The relationship between realizations of <com:Math><semantics>Y<annotation encoding="application / x-tex">Y< / annotation>< / semantics>< / com:Math> and the sampled detector response is illustrated in Fig. 1. The observed <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> can be approximated in terms of the <com:Math><semantics>yj<annotation encoding="application / x-tex">y_j< / annotation>< / semantics>< / com:Math> by substituting (2) into (9), [Image disponible dans le document PDF, Image available in the PDF document] (22) [Image disponible dans le document PDF, Image available in the PDF document] [Image disponible dans le document PDF, Image available in the PDF document] (23) [Image disponible dans le document PDF, Image available in the PDF document] (24) where <com:Math><semantics>yi<annotation encoding="application / x-tex">y_i< / annotation>< / semantics>< / com:Math> is the realization of the unobservable random variable Y that represents the photon energy in an interval of the discrete-time detector response, [Image disponible dans le document PDF, Image available in the PDF document] (25) where <com:Math><semantics>zi<annotation encoding="application / x-tex">z_i< / annotation>< / semantics>< / com:Math> is a realization of Z, an independent random variable representing errors in the sampling process and estimation of <com:Math><semantics>Tj<annotation encoding="application / x-tex">T_j< / annotation>< / semantics>< / com:Math>. We assume Z has known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. With these definitions of <com:Math><semantics>X<annotation encoding="application / x-tex">X< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>Y<annotation encoding="application / x-tex">Y< / annotation>< / semantics>< / com:Math>, the number of intervals which can be found in a finite length of detector output is a random variable <com:Math><semantics>N<annotation encoding="application / x-tex">N< / annotation>< / semantics>< / com:Math>. At high count-rates this approach succumbs to paralysis, as the probability of being able to partition the observed data into intervals approaches zero. The onset of paralysis occurs at higher count-rates compared to pile-up rejection based strategies, since multiple photons are permitted to pile-up within in each interval. Assume the time-series defined in <com:Math><semantics>(3)<annotation encoding="application / x-tex">(3)< / annotation>< / semantics>< / com:Math> – <com:Math><semantics>(6)<annotation encoding="application / x-tex">(6)< / annotation>< / semantics>< / com:Math> has been sampled uniformly. Without loss of generality, assume unit sample intervals beginning at <com:Math><semantics>t0=0<annotation encoding="application / x-tex">t_0 = 0< / annotation>< / semantics>< / com:Math> i.e., <com:Math><semantics>tk=k<annotation encoding="application / x-tex">t_k = k< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>0≤k<K<annotation encoding="application / x-tex">0 \le k < K< / annotation>< / semantics>< / com:Math>. Let R be a discrete-time random process representing the sampled detector response of (1). Let <com:Math><semantics>Y={Yj:0≤j<N}<annotation encoding="application / x-tex">Y = \{Y_j : 0 \le j < N\}< / annotation>< / semantics>< / com:Math> be a discrete-time random process whose components <com:Math><semantics>Yj<annotation encoding="application / x-tex">Y_j< / annotation>< / semantics>< / com:Math> represent the total photon energy arriving during a fixed time interval. A compound Poisson process can be used to model <com:Math><semantics>Y<annotation encoding="application / x-tex">Y< / annotation>< / semantics>< / com:Math>, i.e., [Image disponible dans le document PDF, Image available in the PDF document] (26) [Image disponible dans le document PDF, Image available in the PDF document] (27) where <com:Math><semantics>νj<annotation encoding="application / x-tex">\nu_j< / annotation>< / semantics>< / com:Math> is an independent Poisson random variable, and <com:Math><semantics>Ak<annotation encoding="application / x-tex">A_k< / annotation>< / semantics>< / com:Math> are independent identically distributed random variables with density function <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>. The <com:Math><semantics>{νj}<annotation encoding="application / x-tex">\{\nu_j\}< / annotation>< / semantics>< / com:Math> form a homogeneous Poisson process with rate parameter <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>. The process Y is not directly observable. Assume the pulse shape <com:Math><semantics>Φ(t)<annotation encoding="application / x-tex">\Phi(t)< / annotation>< / semantics>< / com:Math> has finite support. Let <com:Math><semantics>𝟙A(t)<annotation encoding="application / x-tex">\mathbb{1}_A(t)< / annotation>< / semantics>< / com:Math> be the indicator function for the set A. Let the pulse length <com:Math><semantics>ℓ<annotation encoding="application / x-tex">\ell< / annotation>< / semantics>< / com:Math> be given by <com:Math><semantics>ℓ=sup⁡({t:Φ(t)>0})−inf⁡({t:Φ(t)>0})<annotation encoding="application / x-tex">\ell = \sup(\{t : \Phi(t) > 0\}) - \inf(\{t : \Phi(t) > 0\})< / annotation>< / semantics>< / com:Math>.Let <com:Math><semantics>S=R+W={s(k):0≤k<K}<annotation encoding="application / x-tex">S = R + W = \{s(k) : 0 \le k < K\}< / annotation>< / semantics>< / com:Math> be a discrete-time random process representing the observed detector output given by <com:Math><semantics>(2)<annotation encoding="application / x-tex">(2)< / annotation>< / semantics>< / com:Math>. It consists of the detector response <com:Math><semantics>R<annotation encoding="application / x-tex">R< / annotation>< / semantics>< / com:Math> corrupted by a noise process <com:Math><semantics>W<annotation encoding="application / x-tex">W< / annotation>< / semantics>< / com:Math>. Without loss of generality, we assume unit sample intervals. From the observations <com:Math><semantics>S<annotation encoding="application / x-tex">S< / annotation>< / semantics>< / com:Math> we form the process <com:Math><semantics>X<annotation encoding="application / x-tex">X< / annotation>< / semantics>< / com:Math>, where [Image disponible dans le document PDF, Image available in the PDF document] (28) and where <com:Math><semantics>Zi<annotation encoding="application / x-tex">Z_i< / annotation>< / semantics>< / com:Math> is a random variable from an independent noise process of known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. A simple model for testing theory is obtained when we let the pulse shape <com:Math><semantics>Φ(t)=<annotation encoding="application / x-tex">\Phi(t) =< / annotation>< / semantics>< / com:Math> <com:Math><semantics>𝟙[0,1)(t)<annotation encoding="application / x-tex">\mathbb{1}_{[0,1)}(t)< / annotation>< / semantics>< / com:Math> in (1), in which case we let <com:Math><semantics>Xj=Sj<annotation encoding="application / x-tex">X_j = S_j< / annotation>< / semantics>< / com:Math>, and N is simply the sample length K. Obtaining <com:Math><semantics>Xi<annotation encoding="application / x-tex">X_i< / annotation>< / semantics>< / com:Math> from S is more complicated for real data. In that case we partition the process S into non-overlapping blocks of length L, where <com:Math><semantics>L>ℓ<annotation encoding="application / x-tex">L > \ell< / annotation>< / semantics>< / com:Math>. The Poisson rate <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> is expressed in photons per block. The start of each block <com:Math><semantics>Tj∈ℕ<annotation encoding="application / x-tex">T_j \in \mathbb{N}< / annotation>< / semantics>< / com:Math> is chosen such that the total energy of any pulse is fully contained within the block in which it arrives [Image disponible dans le document PDF, Image available in the PDF document] (29) Fig. 1 shows that <com:Math><semantics>Yj=∑k=TjTj+LRk<annotation encoding="application / x-tex">Y_j = \sum_{k=T_j}^{T_j+L} R_k< / annotation>< / semantics>< / com:Math>. We let <com:Math><semantics>Xj=∑k=T̂iT̂j+LSk<annotation encoding="application / x-tex">X_j = \sum_{k=\hat{T}_i}^{\hat{T}_j+L} S_k< / annotation>< / semantics>< / com:Math> [Image disponible dans le document PDF, Image available in the PDF document] (30) where <com:Math><semantics>T̂i<annotation encoding="application / x-tex">\hat{T}_i< / annotation>< / semantics>< / com:Math> is an estimate of <com:Math><semantics>Tj<annotation encoding="application / x-tex">T_j< / annotation>< / semantics>< / com:Math>. Section 4.2 describes the selection of L and <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math> for real data. With this definition of <com:Math><semantics>Xj<annotation encoding="application / x-tex">X_j< / annotation>< / semantics>< / com:Math>, the number of components in Y becomes a random variable for a given sample length K. At high count-rates this approach succumbs to paralysis, as the probability of being able to create a block approaches zero. The onset of paralysis occurs at higher count-rates compared to pile-up rejection based strategies, since multiple photons are permitted to pile-up within in each block. Let <com:Math><semantics>Y={Yj:0≤j<N}<annotation encoding="application / x-tex">Y = \{Y_j : 0 \le j < N\}< / annotation>< / semantics>< / com:Math> be a discrete-time random process whose components <com:Math><semantics>Yj<annotation encoding="application / x-tex">Y_j< / annotation>< / semantics>< / com:Math> are given by [Image disponible dans le document PDF, Image available in the PDF document] (31) [Image disponible dans le document PDF, Image available in the PDF document] (32) where <com:Math><semantics>L∈ℕ<annotation encoding="application / x-tex">L \in \mathbb{N}< / annotation>< / semantics>< / com:Math> is a constant chosen such that h and d is a small threshold value close to zero. The random variable <com:Math><semantics>Yi<annotation encoding="application / x-tex">Y_i< / annotation>< / semantics>< / com:Math> thus represents the total photon energy arriving during a fixed time interval of length <com:Math><semantics>L<annotation encoding="application / x-tex">L< / annotation>< / semantics>< / com:Math>. The value of <com:Math><semantics>d<annotation encoding="application / x-tex">d< / annotation>< / semantics>< / com:Math> ensures the signal associated with photon arrivals is very small at the start and end of each interval. This is illustrated in Fig. 1. A compound Poisson process can be used to model <com:Math><semantics>Y<annotation encoding="application / x-tex">Y< / annotation>< / semantics>< / com:Math>, i.e., [Image disponible dans le document PDF, Image available in the PDF document] (33) [Image disponible dans le document PDF, Image available in the PDF document] (34) [Image disponible dans le document PDF, Image available in the PDF document] (35) where <com:Math><semantics>νλ<annotation encoding="application / x-tex">\nu_{\lambda}< / annotation>< / semantics>< / com:Math> is a homogeneous Poisson process with rate parameter <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>, and <com:Math><semantics>Ak<annotation encoding="application / x-tex">A_k< / annotation>< / semantics>< / com:Math> are indepen- dent identically distributed random variables with density function <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>. Let <com:Math><semantics>S=R+W<annotation encoding="application / x-tex">S = R + W< / annotation>< / semantics>< / com:Math> be a discrete-time random process representing the sampled detector output given by (2). It consists of the detector response <com:Math><semantics>R<annotation encoding="application / x-tex">R< / annotation>< / semantics>< / com:Math> corrupted by a noise process <com:Math><semantics>W<annotation encoding="application / x-tex">W< / annotation>< / semantics>< / com:Math>. The process Y is not directly observable. Using <com:Math><semantics>(2)<annotation encoding="application / x-tex">(2)< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>(25)<annotation encoding="application / x-tex">(25)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>(32)<annotation encoding="application / x-tex">(32)< / annotation>< / semantics>< / com:Math>, we model observations by the process <com:Math><semantics>X={Xj:0≤j<N}<annotation encoding="application / x-tex">X = \{X_j : 0 \le j < N\}< / annotation>< / semantics>< / com:Math>, i.e., [Image disponible dans le document PDF, Image available in the PDF document] (36) [Image disponible dans le document PDF, Image available in the PDF document] (37) [Image disponible dans le document PDF, Image available in the PDF document] (38) [Image disponible dans le document PDF, Image available in the PDF document] (39) where Z is a noise process of known variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. All the random variables <com:Math><semantics>(νj,A1,…,Aνj,Zj)<annotation encoding="application / x-tex">(\nu_j, A_1, \ldots, A_{\nu_j}, Z_j)< / annotation>< / semantics>< / com:Math> involved in modelling a given observation <com:Math><semantics>Xj<annotation encoding="application / x-tex">X_j< / annotation>< / semantics>< / com:Math> are assumed independent. Let <com:Math><semantics>X1,X2,…,XN<annotation encoding="application / x-tex">X_1, X_2, \ldots, X_N< / annotation>< / semantics>< / com:Math> be N independent, identically distributed observations. Let <com:Math><semantics>X,Y,Z,A<annotation encoding="application / x-tex">X, Y, Z, A< / annotation>< / semantics>< / com:Math> be the collec- tions of <com:Math><semantics>Xj,Yj,Zj,Aj:0≤j<N<annotation encoding="application / x-tex">X_j, Y_j, Z_j, A_j : 0 \leq j < N< / annotation>< / semantics>< / com:Math>. Let the corresponding characteristic functions be <com:Math><semantics>ϕX,ϕY,ϕZ,ϕA.<annotation encoding="application / x-tex">\phi_X, \phi_Y, \phi_Z, \phi_A.< / annotation>< / semantics>< / com:Math> 3.3 Basic Form of Estimator We seek to invert the mapping from the distribution of photon energy <com:Math><semantics>A<annotation encoding="application / x-tex">A< / annotation>< / semantics>< / com:Math> to the distribu- tion of X. Our strategy is to first obtain the characteristic function of X in terms of <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math>, then invert the mapping assuming the count-rate and noise characteristics are known. Let <com:Math><semantics>ϕX,ϕY,ϕZ,ϕA<annotation encoding="application / x-tex">\phi_X, \phi_Y, \phi_Z, \phi_A< / annotation>< / semantics>< / com:Math> be the characteristic functions of <com:Math><semantics>X,Y,Z,A<annotation encoding="application / x-tex">X, Y, Z, A< / annotation>< / semantics>< / com:Math>. It is well known

[15] that for the compound Poisson process Y with rate <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>, [Image disponible dans le document PDF, Image available in the PDF document] (40) and since <com:Math><semantics>X=Y+Z<annotation encoding="application / x-tex">X = Y + Z< / annotation>< / semantics>< / com:Math> then [Image disponible dans le document PDF, Image available in the PDF document] (41) Given the observations <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> we can form an empirical estimate <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_X< / annotation>< / semantics>< / com:Math> of the characteristic function of X. Treating this as the true characteristic function, we can invert <com:Math><semantics>(40)<annotation encoding="application / x-tex">(40)< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>(41)<annotation encoding="application / x-tex">(41)< / annotation>< / semantics>< / com:Math> to obtain the characteristic function of A and then take the Fourier transform to find the amplitude spectrum <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math>. Specifically, using (40), (41) and exploiting the assumption that Z is Gaussian to ensure <com:Math><semantics>ϕZ(u)<annotation encoding="application / x-tex">\phi_Z(u)< / annotation>< / semantics>< / com:Math> will be non-zero <com:Math><semantics>∀u∈ℝ<annotation encoding="application / x-tex">\forall u \in \mathbb{R}< / annotation>< / semantics>< / com:Math>, we let <com:Math><semantics>γ:ℝ→ℂ<annotation encoding="application / x-tex">\gamma \colon \mathbb{R} \to \mathbb{C}< / annotation>< / semantics>< / com:Math> be the curve described by [Image disponible dans le document PDF, Image available in the PDF document] (42) [Image disponible dans le document PDF, Image available in the PDF document] (43) Temporarily assuming <com:Math><semantics>∀u,γ(u)≠0<annotation encoding="application / x-tex">\forall u, \gamma(u) \neq 0< / annotation>< / semantics>< / com:Math>, after taking the distinguished logarithm of (43) and rearranging we have [Image disponible dans le document PDF, Image available in the PDF document] (44) Ideally, <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math> is recovered by taking a Fourier transform [Image disponible dans le document PDF, Image available in the PDF document] (45) The basic form of our proposed estimator is given in (88) and is derived from (45) via a sequence of steps. First, <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> is estimated from the data (Step 1). Simply substituting this estimate for <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> in (42) does not produce an ISE optimal estimate of <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math>. The approximate ISE is obtained from an approximate estimate of the error distribution of <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> (Step 2). We then determine a sensible windowing function <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math> (in Step 3) and estimate <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> by [Image disponible dans le document PDF, Image available in the PDF document] (46) The windowing function <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math> is designed to minimise the approximate ISE between <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math> and our estimate of <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math> based on (44), (45) and (46), but with <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> in (44) replaced by (46). A similar idea is used for estimating <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> from (44): a weighting function <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> is found (in Step 4) such that replacing <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> in (45) by [Image disponible dans le document PDF, Image available in the PDF document] (47) produces a better estimate of <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math> than using the unweighted estimate <com:Math><semantics>1λdlog⁡(γ̂)<annotation encoding="application / x-tex">\frac{1}{\lambda} \operatorname{dlog}(\hat{\gamma})< / annotation>< / semantics>< / com:Math>. Finally, the weighting function <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> is modified (in Step 5) to account for the integral in (45) having to be replaced by a finite sum in practice. The following subsections expand on these five steps. 3.4 Estimating <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> An estimate of <com:Math><semantics>ϕX(u)<annotation encoding="application / x-tex">\phi_X(u)< / annotation>< / semantics>< / com:Math> is required to estimate <com:Math><semantics>γ(u)<annotation encoding="application / x-tex">\gamma(u)< / annotation>< / semantics>< / com:Math>. In this subsection we define a his- togram model and describe our estimation of <com:Math><semantics>ϕX(u)<annotation encoding="application / x-tex">\phi_X(u)< / annotation>< / semantics>< / com:Math> based on a histogram of the <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> values. Assume N intervals (and corresponding <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values) have been obtained from a finite length data sample. Although the empirical characteristic function [Image disponible dans le document PDF, Image available in the PDF document] (48) provides a consistent, asymptotically normal estimator of the characteristic function

[21] , it has the disadvantage of rapid growth in computational burden as the number of data points N and the required number of evaluation points <com:Math><semantics>u∈ℝ<annotation encoding="application / x-tex">u \in \mathbb{R}< / annotation>< / semantics>< / com:Math> increases. Instead, we use a histogram based estimator that has a lower computational burden. Assume that a histogram of the observed X values is represented by the <com:Math><semantics>2M×1<annotation encoding="application / x-tex">2M \times 1< / annotation>< / semantics>< / com:Math> vector <com:Math><semantics>𝒏<annotation encoding="application / x-tex">\boldsymbol{n}< / annotation>< / semantics>< / com:Math>, where the count in the mth bin is given by [Image disponible dans le document PDF, Image available in the PDF document] (49) All bins of the histogram have equal width. The bin-width is chosen in relation to the magnitude of the <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values. Since the effect of choosing a different bin width is simply equivalent to scaling the <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values, we assume the bin-width to be unity without loss of generality. The bins are apportioned equally between non-negative and negative data values. The number of histogram bins <com:Math><semantics>2M<annotation encoding="application / x-tex">2M< / annotation>< / semantics>< / com:Math> influences the estimator in various ways, as discussed in later subsections. For now, it is sufficient to assume that <com:Math><semantics>2M<annotation encoding="application / x-tex">2M< / annotation>< / semantics>< / com:Math> is large enough to ensure the histogram includes all <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values. We estimate <com:Math><semantics>ϕX(u)<annotation encoding="application / x-tex">\phi_X(u)< / annotation>< / semantics>< / com:Math> by forming a histogram of scaled <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values and take the inverse Discrete Fourier transform i.e., [Image disponible dans le document PDF, Image available in the PDF document] (50) This is a close approximation of the empirical characteristic function but where <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> terms have been rounded to the nearest histogram bin centre (and u contracted by a factor of <com:Math><semantics>2π<annotation encoding="application / x-tex">2\pi< / annotation>< / semantics>< / com:Math>). The term <com:Math><semantics>nm<annotation encoding="application / x-tex">n_m< / annotation>< / semantics>< / com:Math> simply counts the number of rounded terms with the same value. Clearly, this function can be efficiently evaluated at certain discrete points <com:Math><semantics>u∈<annotation encoding="application / x-tex">u \in< / annotation>< / semantics>< / com:Math> <com:Math><semantics>−M,…,M−1<annotation encoding="application / x-tex">-M, \ldots, M-1< / annotation>< / semantics>< / com:Math> using the fast Fourier Transform (FFT). 3.5 Error Distribution of <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_X< / annotation>< / semantics>< / com:Math> The design of the filters <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> in (46) and (47) rely on the statistics of the errors between <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_X< / annotation>< / semantics>< / com:Math> and the true characteristic function. In this subsection we define and describe the characteristics of these errors. We assume the density function <com:Math><semantics>fX<annotation encoding="application / x-tex">f_X< / annotation>< / semantics>< / com:Math> is sufficiently smooth (i.e., <com:Math><semantics>|dnfX(u) / dun|<Cn∈ℝ+∀n∈ℕ<annotation encoding="application / x-tex">|d^n f_X(u) / du^n| < C_n \in \mathbb{R}_+ \quad \forall n \in \mathbb{N}< / annotation>< / semantics>< / com:Math>) and that the width of the histogram bins are sufficiently small (relative to the standard deviation of the additive noise Z) such that the errors introduced by rounding <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values to the centre of each histogram bin are approximately uniformly distributed across each bin, have zero mean and are small relative to the peak spreading caused by <com:Math><semantics>Z<annotation encoding="application / x-tex">Z< / annotation>< / semantics>< / com:Math>. In other words, the source of error arising from the binning of <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> values is considered negligible. Due to both the statistical nature of Poisson counting and the expected count in each bin being non- integer <com:Math><semantics>(𝔼[nm]∈ℝ≥0)<annotation encoding="application / x-tex">(\mathbb{E}[n_m] \in \mathbb{R}_{\geq 0})< / annotation>< / semantics>< / com:Math>, discrepancies exist between the observed number of counts in any given histogram bin and the expected number of counts for that bin. We combine these two sources of error in our model and refer to it as 'histogram noise'. We emphasize that this noise is distinct from the additive noise <com:Math><semantics>Z<annotation encoding="application / x-tex">Z< / annotation>< / semantics>< / com:Math> modelled in (11), which causes peak spreading in the histogram. Let the probability that a realization of X falls in the m-th bin be [Image disponible dans le document PDF, Image available in the PDF document] (51) Let the normalized histogram error <com:Math><semantics>ϵm<annotation encoding="application / x-tex">\epsilon_m< / annotation>< / semantics>< / com:Math> in the m-th bin be the difference between the observed count <com:Math><semantics>nm<annotation encoding="application / x-tex">n_m< / annotation>< / semantics>< / com:Math> and the expected count <com:Math><semantics>𝔼[nm]=NpXm<annotation encoding="application / x-tex">\mathbb{E}[n_m] = N p_{X_m}< / annotation>< / semantics>< / com:Math> in the mth bin, relative to the total counts in the histogram <com:Math><semantics>N<annotation encoding="application / x-tex">N< / annotation>< / semantics>< / com:Math> i.e., [Image disponible dans le document PDF, Image available in the PDF document] (52) Using <com:Math><semantics>(50)<annotation encoding="application / x-tex">(50)< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>(51)<annotation encoding="application / x-tex">(51)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>(52)<annotation encoding="application / x-tex">(52)< / annotation>< / semantics>< / com:Math> we have [Image disponible dans le document PDF, Image available in the PDF document] (53) [Image disponible dans le document PDF, Image available in the PDF document] (54) [Image disponible dans le document PDF, Image available in the PDF document] (55) If the histogram is modelled as a Poisson vector, show that [Image disponible dans le document PDF, Image available in the PDF document] (56) [Image disponible dans le document PDF, Image available in the PDF document] (57) [Image disponible dans le document PDF, Image available in the PDF document] (58) Since the characteristics of the histogram noise can be expressed in terms of the total number of observed intervals <com:Math><semantics>N<annotation encoding="application / x-tex">N< / annotation>< / semantics>< / com:Math>, the impact of using observation data of finite length may be accounted for by incorporating this information into the design of <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math>. 3.6 Estimating <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> Having obtained <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_X< / annotation>< / semantics>< / com:Math>, the next task is to estimate <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math>. Rather than substitute <com:Math><semantics>ϕ̂X(u)<annotation encoding="application / x-tex">\hat{\phi}_X(u)< / annotation>< / semantics>< / com:Math> for <com:Math><semantics>ϕX(u)<annotation encoding="application / x-tex">\phi_X(u)< / annotation>< / semantics>< / com:Math> in (42), we instead use (46) as the estimator, which requires us to choose a windowing function <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math>. In this subsection we attempt to find a function <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math> that is close to optimal. When the distribution of errors in <com:Math><semantics>ϕ̂X(u)<annotation encoding="application / x-tex">\hat{\phi}_X(u)< / annotation>< / semantics>< / com:Math> are considered, the windowing function <com:Math><semantics>G(u)=Gopt(u)<annotation encoding="application / x-tex">G(u) = G_{\text{opt}}(u)< / annotation>< / semantics>< / com:Math> that results in the lowest ISE estimator of the form given in <com:Math><semantics>(46)<annotation encoding="application / x-tex">(46)< / annotation>< / semantics>< / com:Math> is [Image disponible dans le document PDF, Image available in the PDF document] (59) where <com:Math><semantics>ℜ{z}<annotation encoding="application / x-tex">\Re\{z\}< / annotation>< / semantics>< / com:Math> denotes the real component of <com:Math><semantics>z∈ℂ<annotation encoding="application / x-tex">z \in \mathbb{C}< / annotation>< / semantics>< / com:Math>. We cannot calculate <com:Math><semantics>Gopt(u)<annotation encoding="application / x-tex">G_{\text{opt}}(u)< / annotation>< / semantics>< / com:Math> since <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math> is unknown, so instead we attempt to find an approximation. We let [Image disponible dans le document PDF, Image available in the PDF document] (60) This is justified by considering the magnitude of the relative error between the functions <com:Math><semantics>gopt(u)<annotation encoding="application / x-tex">g_{\text{opt}}(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>g1(u)<annotation encoding="application / x-tex">g_1(u)< / annotation>< / semantics>< / com:Math> where [Image disponible dans le document PDF, Image available in the PDF document] (61) [Image disponible dans le document PDF, Image available in the PDF document] <com:Math><semantics>(62)<annotation encoding="application / x-tex">(62)< / annotation>< / semantics>< / com:Math> The magnitude of the relative error is given by [Image disponible dans le document PDF, Image available in the PDF document] (63) Since <com:Math><semantics>ℜ{ϕA}∈[−1,1]<annotation encoding="application / x-tex">\Re\{\phi_A\}\in[-1,1]< / annotation>< / semantics>< / com:Math>, we see the right hand side of (63) is maximum when <com:Math><semantics>ℜ{ϕA(u)}=<annotation encoding="application / x-tex">\Re\{\phi_A(u)\}=< / annotation>< / semantics>< / com:Math> <com:Math><semantics>−1<annotation encoding="application / x-tex">-1< / annotation>< / semantics>< / com:Math>. The relative error is thus bound by [Image disponible dans le document PDF, Image available in the PDF document] (64) which justifies the approximation when <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> is small, or when <com:Math><semantics>N|ϕZ|2(u)≫e4λ<annotation encoding="application / x-tex">N|\phi_Z|^2(u)\gg e^{4\lambda}< / annotation>< / semantics>< / com:Math>. Further- more, we note that the above bound is quite conservative. The distribution of photon energies in spectroscopic systems can typically be modelled as a sum of <com:Math><semantics>K<annotation encoding="application / x-tex">K< / annotation>< / semantics>< / com:Math> Gaussian peaks, where the kth peak has location <com:Math><semantics>μk<annotation encoding="application / x-tex">\mu_k< / annotation>< / semantics>< / com:Math> and scale <com:Math><semantics>σk<annotation encoding="application / x-tex">\sigma_k< / annotation>< / semantics>< / com:Math> i.e., [Image disponible dans le document PDF, Image available in the PDF document] <com:Math><semantics>(65)<annotation encoding="application / x-tex">(65)< / annotation>< / semantics>< / com:Math> [Image disponible dans le document PDF, Image available in the PDF document] (66) Consequently, the characteristic function will have the form [Image disponible dans le document PDF, Image available in the PDF document] (67) i.e., oscillations within an envelope that decays as <com:Math><semantics>e−cu2<annotation encoding="application / x-tex">e^{-cu^2}< / annotation>< / semantics>< / com:Math> for some <com:Math><semantics>c>0<annotation encoding="application / x-tex">c > 0< / annotation>< / semantics>< / com:Math>. The upper bound given by (64) is quite conservative since <com:Math><semantics>|ℜ{ϕA}|≪1<annotation encoding="application / x-tex">|\Re\{\phi_A\}| \ll 1< / annotation>< / semantics>< / com:Math> for most values of <com:Math><semantics>u<annotation encoding="application / x-tex">u< / annotation>< / semantics>< / com:Math>. The approximation error will be significantly smaller at most evaluation points across the spectrum. Having chosen <com:Math><semantics>G(u)<annotation encoding="application / x-tex">G(u)< / annotation>< / semantics>< / com:Math>, we can form an estimate of <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> using (46). The windowing function reduces the impact of histogram noise arising from the finite number of data samples. For large values of <com:Math><semantics>Ne−2λ|ϕZ(u)|2<annotation encoding="application / x-tex">Ne^{-2\lambda}|\phi_Z(u)|^2< / annotation>< / semantics>< / com:Math>, the impact of windowing is negligible and the estimator is essentially the same as using (42) directly. However, in the regions where [Image disponible dans le document PDF, Image available in the PDF document] (68) the windowing becomes significant, and acts to bound our estimate of <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> i.e., Using the fact that the noise Z is Gaussian (so <com:Math><semantics>ϕZ(u)∈ℝ<annotation encoding="application / x-tex">\phi_Z(u) \in \mathbb{R}< / annotation>< / semantics>< / com:Math> and hence <com:Math><semantics>|ϕZ|2=ϕZ2<annotation encoding="application / x-tex">|\phi_Z|^2 = \phi_Z^2< / annotation>< / semantics>< / com:Math>), and since <com:Math><semantics>e−2λ>0<annotation encoding="application / x-tex">e^{-2\lambda} > 0< / annotation>< / semantics>< / com:Math> we see that [Image disponible dans le document PDF, Image available in the PDF document] (69) [Image disponible dans le document PDF, Image available in the PDF document] (70) <com:Math><semantics><N<annotation encoding="application / x-tex"><\sqrt{N}< / annotation>< / semantics>< / com:Math>. (71) This ensures the argument to the distinguished logarithm in (47) remains finite even though <com:Math><semantics>lim⁡u→∞ϕZ(u)=0<annotation encoding="application / x-tex">\lim_{u\to\infty}\phi_Z(u)=0< / annotation>< / semantics>< / com:Math>. 3.7 Estimating <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> Once <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> has been obtained, we proceed to estimate <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> using (47). This requires another windowing function <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math>. In this subsection we find a function <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> for estimating <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> that is close to ISE optimal. We begin by defining a function <com:Math><semantics>ψ(u)<annotation encoding="application / x-tex">\psi(u)< / annotation>< / semantics>< / com:Math> for notational convenience [Image disponible dans le document PDF, Image available in the PDF document] (72) The ISE is minimized when <com:Math><semantics>H(u)=Hopt(u)<annotation encoding="application / x-tex">H(u) = H_{\text{opt}}(u)< / annotation>< / semantics>< / com:Math>, where the optimal filter <com:Math><semantics>Hopt(u)<annotation encoding="application / x-tex">H_{\text{opt}}(u)< / annotation>< / semantics>< / com:Math> is given by [Image disponible dans le document PDF, Image available in the PDF document] (73) (74) Again, we cannot calculate the optimal filter by using (73) – (74) since <com:Math><semantics>ϕX(u)<annotation encoding="application / x-tex">\phi_X(u)< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕϵ(u)<annotation encoding="application / x-tex">\phi_{\epsilon}(u)< / annotation>< / semantics>< / com:Math> are unknown. We instead make the following observations to obtain an ap- proximation of the ISE-optimal filter. 3.7.1 Initial Observations The optimal filter remains close to unity as long as the estimated <com:Math><semantics>ϕ̂A(u)<annotation encoding="application / x-tex">\hat{\phi}_A(u)< / annotation>< / semantics>< / com:Math> remains close to the true value of <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math>. This will invariably be the case for small values of u since [Image disponible dans le document PDF, Image available in the PDF document] (75) [Image disponible dans le document PDF, Image available in the PDF document] (76) Furthermore, equation (73) shows that if <com:Math><semantics>|ϕϵ(u)|≪|ϕX(u)|<annotation encoding="application / x-tex">|\phi_{\epsilon}(u)| \ll |\phi_X(u)|< / annotation>< / semantics>< / com:Math>, then <com:Math><semantics>ϕ̂X(u)=ϕX(u)+<annotation encoding="application / x-tex">\hat{\phi}_X(u) = \phi_X(u) +< / annotation>< / semantics>< / com:Math> <com:Math><semantics>ϕϵ(u)≈ϕX(u)<annotation encoding="application / x-tex">\phi_{\epsilon}(u) \approx \phi_X(u)< / annotation>< / semantics>< / com:Math> so <com:Math><semantics>Hopt(u)≈1<annotation encoding="application / x-tex">H_{\text{opt}}(u) \approx 1< / annotation>< / semantics>< / com:Math>. For larger values of u, when the magnitude of <com:Math><semantics>|ϕX(u)|<annotation encoding="application / x-tex">|\phi_X(u)|< / annotation>< / semantics>< / com:Math> be- comes comparable to or less than <com:Math><semantics>|ϕϵ(u)|<annotation encoding="application / x-tex">|\phi_{\epsilon}(u)|< / annotation>< / semantics>< / com:Math>, the estimator <com:Math><semantics>ϕ̂A(u)=1λdlog⁡((ϕX+ϕϵ) / ψ)(u)<annotation encoding="application / x-tex">\hat{\phi}_{A}(u) = \frac{1}{\lambda} \operatorname{dlog} ((\phi_{X} + \phi_{\epsilon}) / \psi)(u)< / annotation>< / semantics>< / com:Math> is dominated by noise and no longer provides useful estimates of <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math>. In the extreme case <com:Math><semantics>|ϕX(u)|≪|ϕϵ(u)|<annotation encoding="application / x-tex">|\phi_X(u)| \ll |\phi_{\epsilon}(u)|< / annotation>< / semantics>< / com:Math> so <com:Math><semantics>|ϕ̂X(u)|≈|ϕϵ(u)|<annotation encoding="application / x-tex">|\hat{\phi}_X(u)| \approx |\phi_{\epsilon}(u)|< / annotation>< / semantics>< / com:Math> and hence [Image disponible dans le document PDF, Image available in the PDF document] (77) The window <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> should exclude these regions from the estimate, as the bias introduced in doing so will be less than the variance of the unfiltered noise. Unfortunately the esti- mate of <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math> can be severely degraded well before this boundary condition is reached, so (77) is not particularly helpful. A more useful method for detecting when noise begins to dominate is as follows. 3.7.2 Filter Design Function Further manipulation of (67) shows that for typical spectroscopic systems, the magnitude of <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> will have the form [Image disponible dans le document PDF, Image available in the PDF document] (78) i.e.; a mean component that decays according to the peak widths <com:Math><semantics>σk<annotation encoding="application / x-tex">\sigma_k< / annotation>< / semantics>< / com:Math>, and a more rapidly decaying oscillatory component that varies according to the location of the spectral peaks <com:Math><semantics>μk<annotation encoding="application / x-tex">\mu_k< / annotation>< / semantics>< / com:Math>. In designing the window <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math>, we are interested in attenuating the regions in <com:Math><semantics>|ϕ̂A|<annotation encoding="application / x-tex">|\hat{\phi}_A|< / annotation>< / semantics>< / com:Math> where <com:Math><semantics>|ϕA|2≲|ϕϵ / ψ|2<annotation encoding="application / x-tex">|\phi_A|^2 \lesssim |\phi_\epsilon / \psi|^2< / annotation>< / semantics>< / com:Math>, i.e., where the signal power is less than the histogram noise that has been enhanced by the removal of <com:Math><semantics>ϕZ<annotation encoding="application / x-tex">\phi_Z< / annotation>< / semantics>< / com:Math> during the estimation of <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math>. To obtain an estimate of <com:Math><semantics>|ϕA|<annotation encoding="application / x-tex">|\phi_A|< / annotation>< / semantics>< / com:Math>, a low-pass, Gaussian shaped filter <com:Math><semantics>Hlpf(u)<annotation encoding="application / x-tex">H_{lpf}(u)< / annotation>< / semantics>< / com:Math> is convolved with <com:Math><semantics>|ϕ̂A|<annotation encoding="application / x-tex">|\hat{\phi}_A|< / annotation>< / semantics>< / com:Math> to attenuate all but the slowly varying, large scale features of <com:Math><semantics>|ϕ̂A|<annotation encoding="application / x-tex">|\hat{\phi}_A|< / annotation>< / semantics>< / com:Math>. We denote this <com:Math><semantics>|ϕAsmooth|(u)<annotation encoding="application / x-tex">|\phi_{\mathrm{Asmooth}}|(u)< / annotation>< / semantics>< / com:Math> [Image disponible dans le document PDF, Image available in the PDF document] (79) We see that <com:Math><semantics>|ϕϵ(u)|<annotation encoding="application / x-tex">|\phi_{\epsilon}(u)|< / annotation>< / semantics>< / com:Math> has a Rayleigh distribution with scale parameter <com:Math><semantics>σRay=12N<annotation encoding="application / x-tex">\sigma_{\text{Ray}} = \frac{1}{\sqrt{2N}}< / annotation>< / semantics>< / com:Math>. Consequently [Image disponible dans le document PDF, Image available in the PDF document] (80) It is well known that the cumulative distribution function of a Rayleigh distributed random variable <com:Math><semantics>XRay<annotation encoding="application / x-tex">X_{\text{Ray}}< / annotation>< / semantics>< / com:Math> is given by [Image disponible dans le document PDF, Image available in the PDF document] (81) [Image disponible dans le document PDF, Image available in the PDF document] (82) Hence, to assist with computing the window <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> we will make use of the function [Image disponible dans le document PDF, Image available in the PDF document] (83) [Image disponible dans le document PDF, Image available in the PDF document] (84) to control the shape of <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math>. The function <com:Math><semantics>αmin(u)<annotation encoding="application / x-tex">\alpha_{\min}(u)< / annotation>< / semantics>< / com:Math> provides an indication of how confident we can be that the estimate <com:Math><semantics>ϕ̂A(u)<annotation encoding="application / x-tex">\hat{\phi}_A(u)< / annotation>< / semantics>< / com:Math> contains more signal energy than noise energy. The approximation in (84) arises from the fact that <com:Math><semantics>|ϕ̂Asmooth|<annotation encoding="application / x-tex">|\hat{\phi}_{Asmooth}|< / annotation>< / semantics>< / com:Math> is also a random variable slightly affected by the noise <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math>. On occasion — particularly for larger values of <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math> — the histogram noise may result in sufficiently large values of <com:Math><semantics>αmin(u)<annotation encoding="application / x-tex">\alpha_{\min}(u)< / annotation>< / semantics>< / com:Math> to give a false sense of confidence, and potentially allow noisy results to corrupt the estimate of <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math>. To overcome this problem, the function was modified to be uni-modal in u [Image disponible dans le document PDF, Image available in the PDF document] (85) This modification was justified on the assumption that Gaussian noise causes <com:Math><semantics>ϕZ(u)<annotation encoding="application / x-tex">\phi_Z(u)< / annotation>< / semantics>< / com:Math> to be decreasing in <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>. Consequently we expect <com:Math><semantics>𝔼[|ϕϵ(u)| / ψ(u)]<annotation encoding="application / x-tex">\mathbb{E}[|\phi_{\epsilon}(u)| / \psi(u)]< / annotation>< / semantics>< / com:Math> to be increasing in <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>. If we ignore the local oscillations in <com:Math><semantics>ϕA(u)<annotation encoding="application / x-tex">\phi_A(u)< / annotation>< / semantics>< / com:Math> that are due to peak locations in <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math>, the envelope approximated by the smoothed <com:Math><semantics>|ϕAsmooth|(u)<annotation encoding="application / x-tex">|\phi_{Asmooth}|(u)< / annotation>< / semantics>< / com:Math> will be non-increasing in <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>. Equation (74) indicates the optimal window has the form <com:Math><semantics>λϕA(u) / (λϕA(u)+dlog(ϕ̂X / ϕX)(u)+<annotation encoding="application / x-tex">\lambda \phi_A(u) / (\lambda \phi_A(u) + \text{dlog}(\hat{\phi}_X / \phi_X)(u) +< / annotation>< / semantics>< / com:Math> <com:Math><semantics>dlog(G)(u)<annotation encoding="application / x-tex">dlog(G)(u)< / annotation>< / semantics>< / com:Math>, so the overall window shape will be decreasing in <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>. Hence, if the estimated characteristic function in the region of some <com:Math><semantics>u0<annotation encoding="application / x-tex">u_0< / annotation>< / semantics>< / com:Math> (where the signal to noise ratio is high) has determined that the window value should be <com:Math><semantics>H(u0)<1<annotation encoding="application / x-tex">H(u_0) < 1< / annotation>< / semantics>< / com:Math>, then it is reasonable to reject the suggestion that in the region <com:Math><semantics>u1>u0<annotation encoding="application / x-tex">u_1 > u_0< / annotation>< / semantics>< / com:Math> (where the signal to noise ratio will be worse) that <com:Math><semantics>H(u1)>H(u0)<annotation encoding="application / x-tex">H(u_1) > H(u_0)< / annotation>< / semantics>< / com:Math>. Using the knowledge that <com:Math><semantics>|Hopt(u)|<annotation encoding="application / x-tex">|H_{\text{opt}}(u)|< / annotation>< / semantics>< / com:Math> should be close to unity for small <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>, close to zero for large <com:Math><semantics>|u|<annotation encoding="application / x-tex">|u|< / annotation>< / semantics>< / com:Math>, and should 'roll off' as the signal-to-noise- ratio decreases — we consider two potential windowing functions as approximations of <com:Math><semantics>Hopt(u)<annotation encoding="application / x-tex">H_{\mathrm{opt}}(u)< / annotation>< / semantics>< / com:Math>. 3.7.3 Rectangle Window The indicator function provides a very simple windowing function [Image disponible dans le document PDF, Image available in the PDF document] (86) The threshold value <com:Math><semantics>α0<annotation encoding="application / x-tex">\alpha_0< / annotation>< / semantics>< / com:Math> determines the point at which cut-off occurs, and can be se- lected manually as desired (e.g., <com:Math><semantics>α0=0.95<annotation encoding="application / x-tex">\alpha_0 = 0.95< / annotation>< / semantics>< / com:Math>). Once the threshold is chosen, the estimator exhibits similar ISE performance regardless of peak locations in the incident spectra. Rather than requiring the user to select a window width depending on the incident spec- trum<com:Sup>1< / com:Sup>, the width of the window is automatically selected by the data via <com:Math><semantics>αmod(u)<annotation encoding="application / x-tex">\alpha_{\text{mod}}(u)< / annotation>< / semantics>< / com:Math>. While simplicity is the primary advantage of the rectangular window, the abrupt transition re- gion provides a poor model for the roll-off region of the optimal filter. The second filter shape attempts to improve on that. 3.7.4 Logistic Window A window based on the logistic function attempts to model smoother roll-off. It is given by [Image disponible dans le document PDF, Image available in the PDF document] (87) where <com:Math><semantics>α0<annotation encoding="application / x-tex">\alpha_0< / annotation>< / semantics>< / com:Math> again acts as a threshold of acceptance of the hypothesis that the signal energy is greater than the noise energy in the estimate <com:Math><semantics>ϕ̂A(u)<annotation encoding="application / x-tex">\hat{\phi}_A(u)< / annotation>< / semantics>< / com:Math>. The rate of filter roll off in the vicinity of the threshold region is controlled by <com:Math><semantics>β0>0<annotation encoding="application / x-tex">\beta_0 > 0< / annotation>< / semantics>< / com:Math>. This provides a smoother transition region than the rectangle window, reducing Gibbs oscillations in the final estimate of <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math>. Once again, although the parameters <com:Math><semantics>α0,β<annotation encoding="application / x-tex">\alpha_0, \beta< / annotation>< / semantics>< / com:Math> are chosen manually, they are much less dependent on <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> and can be used to provide close to optimal filtering for a wide variety of incident spectra. Typical values used were <com:Math><semantics>α0=0.95,β0=40.0<annotation encoding="application / x-tex">\alpha_0 = 0.95, \beta_0 = 40.0< / annotation>< / semantics>< / com:Math>. The performance of the rectangle and logistic window functions are compared in section 4. 3.8 Estimating <com:Math><semantics>fA<annotation encoding="application / x-tex">f_A< / annotation>< / semantics>< / com:Math> Having designed a window function <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> and thus an estimator <com:Math><semantics>ϕ̂A(u)<annotation encoding="application / x-tex">\hat{\phi}_A(u)< / annotation>< / semantics>< / com:Math>, the final task is to estimate <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math> by inverting the Fourier transform. This sub-section describes several <com:Sup>1< / com:Sup>Gugushvili

[18] proposed a non-parametric estimator for the general decompounding problem in which a rectangular windowing scheme was used. It requires manual selection of window width, which varies with <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math>. issues that arise with numerical implementation. Firstly, it is infeasible to evaluate <com:Math><semantics>ϕ̂X,γ̂(u)<annotation encoding="application / x-tex">\hat{\phi}_X, \hat{\gamma}(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> numerically on the whole real line. Instead we estimate it at discrete points over a finite interval. The finite interval is chosen sufficiently large such that a tolerably small error is incurred as a result of excluding signal values outside the interval. This is justified for <com:Math><semantics>fA(x)<annotation encoding="application / x-tex">f_A(x)< / annotation>< / semantics>< / com:Math> being a Gaussian mixture, since the magnitudes of <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> will decay as <com:Math><semantics>e−cu2<annotation encoding="application / x-tex">e^{-cu^2}< / annotation>< / semantics>< / com:Math> for some <com:Math><semantics>c>0<annotation encoding="application / x-tex">c>0< / annotation>< / semantics>< / com:Math>. The Fast Fourier Transform (FFT) is used to evaluate <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_X< / annotation>< / semantics>< / com:Math> at discrete points, and hence also determines the points where <com:Math><semantics>γ̂(u)<annotation encoding="application / x-tex">\hat{\gamma}(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> are evaluated. Likewise, the FFT is used to evaluate the final estimate of <com:Math><semantics>f̂A<annotation encoding="application / x-tex">\hat{f}_A< / annotation>< / semantics>< / com:Math> at discrete points. In order to use the FFT, the signals outside the interval should be sufficiently small to reduce impact of aliasing. The evaluation points also need to be sufficiently dense to avoid any 'phase wrap' ambiguity when evaluating <com:Math><semantics>dlog⁡(γ̂)(u)<annotation encoding="application / x-tex">d\log(\hat{\gamma})(u)< / annotation>< / semantics>< / com:Math>. Both these objectives can be achieved by increasing the number of bins <com:Math><semantics>2M<annotation encoding="application / x-tex">2M< / annotation>< / semantics>< / com:Math> in the histogram (zero-padding) until a sufficiently large number of bins is attained. As M increases, the sampling density of <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> increases, which allows phase wrapping to be detected and managed. A larger <com:Math><semantics>M<annotation encoding="application / x-tex">M< / annotation>< / semantics>< / com:Math> also allows aliasing (caused by the Gaussian shaped tails of <com:Math><semantics>|ϕX|<annotation encoding="application / x-tex">|\phi_X|< / annotation>< / semantics>< / com:Math>) to be negligible. Typically a value of M was chosen as the smallest power of two sufficiently large such that the non-zero values of the histogram were confined to the ‘lower half’ indexes i.e., <com:Math><semantics>M=min⁡{M:𝒏𝒎=0,|m|∈{M / 2,…,M},M=2N,N∈ℕ}.<annotation encoding="application / x-tex">M=\min\{M: \boldsymbol{n_m}=0, |m|\in\{M / 2,\dots,M\}, M=2^N, N\in\mathbb{N}\}.< / annotation>< / semantics>< / com:Math> Secondly, the distinguished logarithm in (47) is undefined if <com:Math><semantics>γ̂(u)=0<annotation encoding="application / x-tex">\hat{\gamma}(u) = 0< / annotation>< / semantics>< / com:Math>. In estimating <com:Math><semantics>γ(u)<annotation encoding="application / x-tex">\gamma(u)< / annotation>< / semantics>< / com:Math> from the data, there is a small but non-zero probability that the estimate will be zero. In this case, the distinguished logarithm in (47) is undefined and the technique fails. As |u| increases, <com:Math><semantics>|ϕX|(u)<annotation encoding="application / x-tex">|\phi_X|(u)< / annotation>< / semantics>< / com:Math> decreases and may approach <com:Math><semantics>|ϕϵ|(u)<annotation encoding="application / x-tex">|\phi_{\epsilon}|(u)< / annotation>< / semantics>< / com:Math>. When <com:Math><semantics>|ϕX|(u)<annotation encoding="application / x-tex">|\phi_X|(u)< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>|ϕϵ|(u)<annotation encoding="application / x-tex">|\phi_{\epsilon}|(u)< / annotation>< / semantics>< / com:Math> have similar magnitudes, the probability of <com:Math><semantics>|ϕX+ϕϵ|<annotation encoding="application / x-tex">|\phi_X + \phi_{\epsilon}|< / annotation>< / semantics>< / com:Math> (and hence <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math>) being close to zero can become significant. The filter <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> should roll off faster than <com:Math><semantics>|ϕX|(u)<annotation encoding="application / x-tex">|\phi_X|(u)< / annotation>< / semantics>< / com:Math> approaches <com:Math><semantics>|ϕϵ|(u)<annotation encoding="application / x-tex">|\phi_{\epsilon}|(u)< / annotation>< / semantics>< / com:Math> to reduce the impact this may have on the estimate. Ideally <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> should be zero in regions where noise may result in <com:Math><semantics>|γ̂|(u)<annotation encoding="application / x-tex">|\hat{\gamma}|(u)< / annotation>< / semantics>< / com:Math> being close to zero. Gugushvili has shown 18 that for a rectangular window, the probability of inversion failure approaches zero as the length of the data set increases <com:Math><semantics>N→∞<annotation encoding="application / x-tex">N \to \infty< / annotation>< / semantics>< / com:Math>. 3.9 Discrete Notation We digress momentarily to introduce additional notation. Throughout the rest of the paper, bold font will be used to indicate a <com:Math><semantics>2M×1<annotation encoding="application / x-tex">2M \times 1< / annotation>< / semantics>< / com:Math> vector corresponding to a discretely sampled version of the named function, e.g., <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> represents a <com:Math><semantics>2M×1<annotation encoding="application / x-tex">2M \times 1< / annotation>< / semantics>< / com:Math> vector whose values are given by the characteristic function <com:Math><semantics>ϕ̂A(u)<annotation encoding="application / x-tex">\hat{\phi}_A(u)< / annotation>< / semantics>< / com:Math> evaluated at the points <com:Math><semantics>u∈{0,1,…,M−1}<annotation encoding="application / x-tex">u \in \{0, 1, \dots, M-1\}< / annotation>< / semantics>< / com:Math> <com:Math><semantics>1,−M,…,−2,−1<annotation encoding="application / x-tex">1, -M, \ldots, -2, -1< / annotation>< / semantics>< / com:Math>. Square bracket notation [k] is used to index a particular element in the vector, e.g., <com:Math><semantics>ϕ̂A[M−1]<annotation encoding="application / x-tex">\hat{\phi}_A[M-1]< / annotation>< / semantics>< / com:Math> has the value of <com:Math><semantics>ϕ̂A(M−1)<annotation encoding="application / x-tex">\hat{\phi}_A(M-1)< / annotation>< / semantics>< / com:Math>. We also use negative indexes for accessing elements of a vector in a manner similar to the python programming language. Negative indexes should be interpreted relative to the length of the vector, i.e., <com:Math><semantics>ϕ̂A[−1]<annotation encoding="application / x-tex">\hat{\phi}_A[-1]< / annotation>< / semantics>< / com:Math> refers to the last element in the vector (which is equivalent to <com:Math><semantics>ϕ̂A[2M−1]<annotation encoding="application / x-tex">\hat{\phi}_A[2M-1]< / annotation>< / semantics>< / com:Math>). 3.10 Summary of Estimator The estimation procedure we use may be summarized in the following steps. 1. Partition sampled time series into intervals using (8). 2. Calculate <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> value for each interval according to (9). 3. Generate histogram <com:Math><semantics>n<annotation encoding="application / x-tex">n< / annotation>< / semantics>< / com:Math> from the <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> values. 4. Calculate <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_{X}< / annotation>< / semantics>< / com:Math> using the inverse FFT to efficiently evaluate (50) at various sample points. 5. Calculate <com:Math><semantics>ϕ𝒁<annotation encoding="application / x-tex">\phi_{\mathbf{Z}}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>𝑮<annotation encoding="application / x-tex">\mathbf{G}< / annotation>< / semantics>< / com:Math> at the appropriate points. 6. Calculate <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> via (46) using <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_{X}< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>G<annotation encoding="application / x-tex">G< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕZ<annotation encoding="application / x-tex">\phi_{Z}< / annotation>< / semantics>< / com:Math>. 7. Calculate <com:Math><semantics>|ϕAsmooth(u)|<annotation encoding="application / x-tex">|\phi_{\text{Asmooth}}(u)|< / annotation>< / semantics>< / com:Math>, a low-pass filtered version of <com:Math><semantics>|1λdlog(γ̂)(u)|<annotation encoding="application / x-tex">|\frac{1}{\lambda}\text{dlog}(\hat{\gamma})(u)|< / annotation>< / semantics>< / com:Math>. 8. Calculate <com:Math><semantics>αmod<annotation encoding="application / x-tex">\alpha_{\text{mod}}< / annotation>< / semantics>< / com:Math> via (83) and (85). 9. Calculate <com:Math><semantics>𝑯<annotation encoding="application / x-tex">\boldsymbol{H}< / annotation>< / semantics>< / com:Math> using <com:Math><semantics>𝜶mod<annotation encoding="application / x-tex">\boldsymbol{\alpha}_{\text{mod}}< / annotation>< / semantics>< / com:Math> and either (86) or (87). 10. Calculate <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_{A}< / annotation>< / semantics>< / com:Math> via (47) using <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>H<annotation encoding="application / x-tex">H< / annotation>< / semantics>< / com:Math>. If any element of <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> is zero and the corre- sponding element of <com:Math><semantics>𝑯<annotation encoding="application / x-tex">\boldsymbol{H}< / annotation>< / semantics>< / com:Math> is non-zero, the estimation has failed as the distinguished logarithm is undefined. 11. Calculate <com:Math><semantics>f̂A<annotation encoding="application / x-tex">\hat{f}_A< / annotation>< / semantics>< / com:Math> using the FFT of <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> according to [Image disponible dans le document PDF, Image available in the PDF document] (88) 3.11 Performance Measures The performance of the estimator is measured using the integrated square of the error (ISE). The ISE measures the global fit of the estimated density. [Image disponible dans le document PDF, Image available in the PDF document] (89) The discrete ISE measure is given by [Image disponible dans le document PDF, Image available in the PDF document] (90) where <com:Math><semantics>pA<annotation encoding="application / x-tex">p_A< / annotation>< / semantics>< / com:Math> is a <com:Math><semantics>2M×1<annotation encoding="application / x-tex">2M \times 1< / annotation>< / semantics>< / com:Math> vector whose elements contain the probability mass in the region of each histogram bin i.e., [Image disponible dans le document PDF, Image available in the PDF document] (91) The vector <com:Math><semantics>𝒑̂A<annotation encoding="application / x-tex">\hat{\boldsymbol{p}}_A< / annotation>< / semantics>< / com:Math> represents the corresponding estimated probability mass vector. 4 NUMERICAL RESULTS OF THE FIRST EMBODIMENT Experiments were performed using simulated and real data. 4.1 Simulations The ideal density used by Trigano et al.

[11] was used for these simulations. It consists of a mixture of six Gaussian and one gamma distribution to simulate Compton background. The mixture density is given by [Image disponible dans le document PDF, Image available in the PDF document] (92) where <com:Math><semantics>𝒩(μ,σ2)<annotation encoding="application / x-tex">\mathcal{N}(\mu, \sigma^2)< / annotation>< / semantics>< / com:Math> is the density of a normal distribution with mean <com:Math><semantics>μ<annotation encoding="application / x-tex">\mu< / annotation>< / semantics>< / com:Math> and variance <com:Math><semantics>σ2<annotation encoding="application / x-tex">\sigma^2< / annotation>< / semantics>< / com:Math>. The density of the gamma distribution is given by <com:Math><semantics>g(x)=(0.5+x / 200)e−(0.5+x / 200)<annotation encoding="application / x-tex">g(x) = (0.5 + x / 200)e^{-(0.5+x / 200)}< / annotation>< / semantics>< / com:Math>. The density was sampled at 8192 equally spaced integer points to produce the discrete vector <com:Math><semantics>pA<annotation encoding="application / x-tex">p_A< / annotation>< / semantics>< / com:Math> of probability mass. The FFT was taken to obtain <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math>, a sampled vector of <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> values. A particular count rate <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> was chosen for an experiment, corresponding to the expected number of events per observation interval. The expected pile-up density was obtained via (40). i.e., the discrete vector <com:Math><semantics>ϕA<annotation encoding="application / x-tex">\phi_A< / annotation>< / semantics>< / com:Math> was scaled by <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>, exponentiated, then scaled by <com:Math><semantics>e−λ<annotation encoding="application / x-tex">e^{-\lambda}< / annotation>< / semantics>< / com:Math> and finally an FFT was applied [Image disponible dans le document PDF, Image available in the PDF document] (93) Equation (93) was convolved with a Gaussian to simulate the effect of noise Z smearing out the observed spectrum [Image disponible dans le document PDF, Image available in the PDF document] (94) This represents the expected density of the observed spectrum, including pile-up and additive noise. Observation histograms were created using random variables that were distributed according to (94). Experiments were parameterized by the pair <com:Math><semantics>(N,λ)<annotation encoding="application / x-tex">(N, \lambda)< / annotation>< / semantics>< / com:Math> where <com:Math><semantics>N∈{104,105,106,107,108,109}<annotation encoding="application / x-tex">N \in \{10^4, 10^5, 10^6, 10^7, 10^8, 10^9\}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>λ∈{1.0,3.0,5.0}<annotation encoding="application / x-tex">\lambda \in \{1.0, 3.0, 5.0\}< / annotation>< / semantics>< / com:Math>. For each parameter pair <com:Math><semantics>(N,λ)<annotation encoding="application / x-tex">(N, \lambda)< / annotation>< / semantics>< / com:Math>, one thousand observed histograms were made. Estimates of the probability mass vector <com:Math><semantics>pA<annotation encoding="application / x-tex">p_A< / annotation>< / semantics>< / com:Math> were made using (88) with both (86) and (87) used for <com:Math><semantics>H(k)<annotation encoding="application / x-tex">H(k)< / annotation>< / semantics>< / com:Math>. A threshold value of <com:Math><semantics>α0=0.95<annotation encoding="application / x-tex">\alpha_0 = 0.95< / annotation>< / semantics>< / com:Math> was used for both window shapes, and <com:Math><semantics>β0=40.0<annotation encoding="application / x-tex">\beta_0 = 40.0< / annotation>< / semantics>< / com:Math> for the logistic shape. The discrete ISE measure of the error between each estimate <com:Math><semantics>p̂A<annotation encoding="application / x-tex">\hat{p}_A< / annotation>< / semantics>< / com:Math> and the true vector <com:Math><semantics>pA<annotation encoding="application / x-tex">p_A< / annotation>< / semantics>< / com:Math> were recorded. For comparison with asymptotic bandwidth results, estimates were made using a rectangular window whose bandwidth was selected according to the condition 1.3 specified by Gugushvili in

[18] i.e., <com:Math><semantics>hN=(ln⁡N)−β<annotation encoding="application / x-tex">h_N = (\ln N)^{-\beta}< / annotation>< / semantics>< / com:Math> where <com:Math><semantics>β<1 / 2<annotation encoding="application / x-tex">\beta < 1 / 2< / annotation>< / semantics>< / com:Math>. We emphasize that the <com:Math><semantics>β<annotation encoding="application / x-tex">\beta< / annotation>< / semantics>< / com:Math> of Gugushvili's filter is not to be confused with the <com:Math><semantics>β0<annotation encoding="application / x-tex">\beta_0< / annotation>< / semantics>< / com:Math> of (87). The asymptotic bandwidth criterion was implemented by using [Image disponible dans le document PDF, Image available in the PDF document] (95) [Image disponible dans le document PDF, Image available in the PDF document] (96) Three values for Gugushvilli's <com:Math><semantics>β<annotation encoding="application / x-tex">\beta< / annotation>< / semantics>< / com:Math> were trialled, namely <com:Math><semantics>β=1 / 2,1 / 3,1 / 4<annotation encoding="application / x-tex">\beta = 1 / 2, 1 / 3, 1 / 4< / annotation>< / semantics>< / com:Math>. Estimates were also made using a rectangular filter (95) with fixed bandwidths of various values <com:Math><semantics>α / M∈[0.2,0.4,0.6,0.8]<annotation encoding="application / x-tex">\alpha / M \in [0.2, 0.4, 0.6, 0.8]< / annotation>< / semantics>< / com:Math>. Finally time-series data was created according to (1) with an idealised rectangular pulse shape and 10<com:Sup>7< / com:Sup> pulses whose energies were distributed according to (92). The pulse length and count rate were chosen to give a Poisson rate <com:Math><semantics>λ=1.0<annotation encoding="application / x-tex">\lambda = 1.0< / annotation>< / semantics>< / com:Math>. The algorithm described by Trigano et al.

[11] was used to estimate the underlying amplitude density from a bi-dimensional histogram containing <com:Math><semantics>32×1024<annotation encoding="application / x-tex">32 \times 1024< / annotation>< / semantics>< / com:Math> (duration x energy) bins — this choice of bins reportedly giving the best accuracy and reasonable execution times. The performance and processing time of the core algorithm were recorded for comparison with our proposed algorithm. Fig. 2 plots a typical estimate <com:Math><semantics>p̂A<annotation encoding="application / x-tex">\hat{p}_A< / annotation>< / semantics>< / com:Math> made by the data-driven logistic shaped filter for an experiment with parameter pair <com:Math><semantics>(N=106,λ=3.0)<annotation encoding="application / x-tex">(N=10^6, \lambda=3.0)< / annotation>< / semantics>< / com:Math>. The true vector <com:Math><semantics>pA<annotation encoding="application / x-tex">p_A< / annotation>< / semantics>< / com:Math> (thin solid line), and the observed histogram <com:Math><semantics>𝒑̂𝑿<annotation encoding="application / x-tex">\hat{\boldsymbol{p}}_{\boldsymbol{X}}< / annotation>< / semantics>< / com:Math> (lower curve containing some noise) are also plotted. Pile-up peaks can be clearly seen in the observed histogram. Although the estimated density suffers from ringing (due to the Gibbs phenomenon), it otherwise estimates the true density and corrects the pile-up that was present in the observed histogram. Fig. 3 plots a typical estimate made at the same operating point as Fig. 2, but with an estimator having a rectangular filter where the bandwidth was selected using (96) and <com:Math><semantics>β=14<annotation encoding="application / x-tex">\beta = \frac{1}{4}< / annotation>< / semantics>< / com:Math>. This corresponds to the operating region in Fig. 6 where the performance of the fixed bandwidth filter <com:Math><semantics>(β=14)<annotation encoding="application / x-tex">(\beta = \frac{1}{4})< / annotation>< / semantics>< / com:Math> is approaching that of the data-driven filters. It is evident that while also correcting pile-up, the resulting estimate contains more noise. Fig. 4 shows the distribution densities of ISE measures as a function of sample count using a rectangular filter and various fixed bandwidths. Lines were plotted between distribution means (MISE) to assist visualization. The results for the data-driven rectangular filter (86) were also plotted, connected with a thicker curve. This clearly illustrates the weakness of fixed bandwidth filtering. For any fixed bandwidth, the ISE decreases as sample count increases, eventually asymptoting as the bias becomes the dominant source of error. At that point (which is noise and bandwidth dependent) the ISE remains largely constant despite increases in sample count. The fixed bandwidth excludes the use of some estimates <com:Math><semantics>ϕ̂A[k]<annotation encoding="application / x-tex">\hat{\phi}_{A}[k]< / annotation>< / semantics>< / com:Math> in the final calculation, even when they have high signal-to-noise- ratio (SNR). Fig. 4 also shows the results given by the rectangular filter with our proposed data-driven bandwidth selection. This curve lies close to the inflection point of each fixed bandwidth curve. This indicates the bandwidth selected for the data-driven rectangular filter is close to the optimal bandwidth value (for a rectangular filter) across the range of sample counts. Fig. <com:Math><semantics>5−Fig.<annotation encoding="application / x-tex">5 - \text{Fig.}< / annotation>< / semantics>< / com:Math> 7 show the distribution densities of the ISE measure as a function of the total number of estimates <com:Math><semantics>N<annotation encoding="application / x-tex">N< / annotation>< / semantics>< / com:Math> in each histogram at three count rates <com:Math><semantics>λ∈{1.0,3.0,5.0}<annotation encoding="application / x-tex">\lambda \in \{1.0, 3.0, 5.0\}< / annotation>< / semantics>< / com:Math>. The MISE curves for the logistic and rectangular filters are lower than those obtained using the bandwidth given by (96) for much of the region of application interest. There are various regions where the non-data-driven bandwidth <com:Math><semantics>(β=14)<annotation encoding="application / x-tex">(\beta = \frac{1}{4})< / annotation>< / semantics>< / com:Math> gives similar performance to the data-driven bandwidths, however this is not maintained across the whole range of sample counts. The logistic filter shape has slightly better performance than the rectangular filter shape, although the differences between the two filters appears relatively minor to the ISE measure. Table 1 compares the results between the proposed algorithm and the algorithm recently described in

[11] . The ISE for both methods were similar at the operating point under test (<com:Math><semantics>λ=1.0,N=107<annotation encoding="application / x-tex">\lambda = 1.0, N = 10^7< / annotation>< / semantics>< / com:Math>), however our proposed algorithm requires considerably less computation. Table 1: Comparison With Algorithm Described in

[11] [Image disponible dans le document PDF, Image available in the PDF document] 4.2 Real Data The estimator was applied to real data to assess its usefulness in practical applications. The threshold value <com:Math><semantics>ϵ<annotation encoding="application / x-tex">\epsilon< / annotation>< / semantics>< / com:Math> found in (8) was chosen to be approximately one half the standard deviation of the additive noise <com:Math><semantics>w(t)<annotation encoding="application / x-tex">w(t)< / annotation>< / semantics>< / com:Math>. This ensured a reasonably high probability of creating intervals, yet ensured errors in the estimation of interval energy were low. A value for the interval length L was chosen approximately four times the 'length' of a typical pulse - that is, four times the length of the interval <com:Math><semantics>{t:Φ(t)>ϵ}<annotation encoding="application / x-tex">\{t:\Phi(t)>\epsilon\}< / annotation>< / semantics>< / com:Math>. An energy histogram was obtained from a manganese sample, with a photon flux rate of nominally 10<com:Sup>5< / com:Sup> events per second. A slight negative skew was present in the shape of the main peaks of the observed histogram, suggesting a complicated noise source had influenced the system. This is barely visible in Fig. 8. The noise was modelled as a bimodal Gaussian mixture rather than a single Gaussian peak. A simple least-squares optimization routine was used to fit bimodal Gaussian parameters <com:Math><semantics>Z∼α1𝒩(μ1,σ12)+α2𝒩(μ2,σ22)<annotation encoding="application / x-tex">Z \sim \alpha_1 \mathcal{N}(\mu_1, \sigma_1^2) + \alpha_2 \mathcal{N}(\mu_2, \sigma_2^2)< / annotation>< / semantics>< / com:Math> to the noise peak located around bin index zero. A suitable value for <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> was chosen manually. The logistic filter with data-driven bandwidth was used to estimate the true density. Fig. 8 shows plots of the observed and estimated probability mass vectors. The main peaks (bins <com:Math><semantics>450∼600<annotation encoding="application / x-tex">450 \sim 600< / annotation>< / semantics>< / com:Math>) have been enhanced while the pile-up has been attenuated though not fully removed. The first order pile-up peaks have been reduced. The peak-to-pile-up ratio (ratio of the height of main peak to that of first pile-up peak) has increased from around 6 to around 120. These improvements are comparable to other state of the art systems (e.g.,

[11] ). There are several possible reasons the estimator fails to fully resolve pile-up. The accuracy of the estimator depends on correctly modelling the Gaussian noise peak. The bimodal Gaussian mixture modelled the noise peak such that the maximum error was less than 1% of the noise density peak. Given that the residual pile-up peaks in the estimated spectrum are below 1% of the main peak, the sensitivity of the estimator to errors in noise modelling may have contributed to this in some part. A second reason for the unresolved pile-up may be due to the uncertainty in the estimation of the observed spectrum. Several of the residual pile-up peaks are relatively close to the floor of the observed histogram. The residual peaks may simply be a noise induced artefact of the estimator. Finally, the mathematical model may be an overly simple approximation of the observed spectrum. The detection process includes numerous second-order effects that have not been included in the model (e.g., ballistic deficit, supply charge depletion, correlated noise, non-linearities, etc...). These minor effects may limit the accuracy of the pile-up correction estimator. 5 SUMMARY OF THE FIRST EMBODIMENT We have taken the estimator proposed by Gugushvili

[18] for decompounding under Gaussian noise, and adapted it for correcting pulse pile-up in X-ray spectroscopy. We have proposed a data-driven bandwidth selection mechanism that is easily implemented, and provides significant reduction in ISE / MISE across a broad range of sample counts of interest to spectroscopic applications (<com:Math><semantics>104∼109<annotation encoding="application / x-tex">10^4 \sim 10^9< / annotation>< / semantics>< / com:Math> counts). The data-driven rectangular bandwidth selection is close to optimal (for rectangular filters), and over the range of interest outperforms bandwidth selection based on asymptotic results or fixed bandwidth. Although initial results appear promising, further work is required to improve the performance for practical implementations. The estimation still contains 'ringing' arte- facts associated with the Gibbs phenomenon. Additional filter shape attempts to reduce this, there are other shapes that are closer to MSE optimal. 6 SECOND EMBODIMENT This section gives a summary of the spectrum estimator of the 2nd embodiment. The 2nd embodiment solves the problem of the 1st embodiment which requires entire clusters to be approximately encompassed in each interval. In the 2nd embodiment, the entire data series if desired can be used, and the overlap is compensated by introduction of 2 different interval lengths L and L1. We need to include a few additional terms not mentioned in the first embodiment. In particular <com:Math><semantics>ϕ̂X1<annotation encoding="application / x-tex">\hat{\phi}_{X1}< / annotation>< / semantics>< / com:Math>. The spectrum estimator is based on [Image disponible dans le document PDF, Image available in the PDF document] (97) The introduction of the filter <com:Math><semantics>H(u;α)<annotation encoding="application / x-tex">H(u;\alpha)< / annotation>< / semantics>< / com:Math> allows us to address several implementation issues that arise. The estimation procedure we use may be summarized in the following steps. 1. Partition sampled time-series into fixed length intervals <com:Math><semantics>[Tj,Tj+L),j∈ℕ<annotation encoding="application / x-tex">[T_j, T_j + L), j \in \mathbb{N}< / annotation>< / semantics>< / com:Math>. 2. Calculate <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> value for each interval according to <com:Math><semantics>xj=∑k∈[Ti,Ti+L)s(k)<annotation encoding="application / x-tex">x_j = \sum_{k \in [T_i, T_i + L)} s(k)< / annotation>< / semantics>< / com:Math>. 3. Generate histogram <com:Math><semantics>n<annotation encoding="application / x-tex">n< / annotation>< / semantics>< / com:Math> from the <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> values. 4. Calculate <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_{X}< / annotation>< / semantics>< / com:Math> using the inverse FFT of <com:Math><semantics>n<annotation encoding="application / x-tex">n< / annotation>< / semantics>< / com:Math>. 5. Partition the sampled time series into a different set of intervals with length <com:Math><semantics>L1<annotation encoding="application / x-tex">L_1< / annotation>< / semantics>< / com:Math>, and following similar calculations to obtain <com:Math><semantics>ϕ̂X1<annotation encoding="application / x-tex">\hat{\phi}_{X1}< / annotation>< / semantics>< / com:Math>. 6. Calculate <com:Math><semantics>ϕZ<annotation encoding="application / x-tex">\phi_{Z}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>G<annotation encoding="application / x-tex">G< / annotation>< / semantics>< / com:Math>. 7. Calculate <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> using <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_{X}< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>G<annotation encoding="application / x-tex">G< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>ϕZ<annotation encoding="application / x-tex">\phi_{Z}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>ϕ̂X1<annotation encoding="application / x-tex">\hat{\phi}_{X1}< / annotation>< / semantics>< / com:Math>, [Image disponible dans le document PDF, Image available in the PDF document] (98) 8. Calculate <com:Math><semantics>|ϕAsmooth(u)|<annotation encoding="application / x-tex">|\phi_{\text{Asmooth}}(u)|< / annotation>< / semantics>< / com:Math>, a low-pass filtered version of <com:Math><semantics>|1λdlog(γ̂)(u)|<annotation encoding="application / x-tex">|\frac{1}{\lambda}\text{dlog}(\hat{\gamma})(u)|< / annotation>< / semantics>< / com:Math>. 9. Calculate <com:Math><semantics>αmod<annotation encoding="application / x-tex">\alpha_{\text{mod}}< / annotation>< / semantics>< / com:Math>. 10. Calculate <com:Math><semantics>𝑯<annotation encoding="application / x-tex">\boldsymbol{H}< / annotation>< / semantics>< / com:Math> using <com:Math><semantics>𝜶mod<annotation encoding="application / x-tex">\boldsymbol{\alpha}_{\text{mod}}< / annotation>< / semantics>< / com:Math>. 11. Calculate <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> using <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> and <com:Math><semantics>H<annotation encoding="application / x-tex">H< / annotation>< / semantics>< / com:Math>. If any element of <com:Math><semantics>γ̂<annotation encoding="application / x-tex">\hat{\gamma}< / annotation>< / semantics>< / com:Math> is zero and the corresponding element of <com:Math><semantics>𝑯<annotation encoding="application / x-tex">\boldsymbol{H}< / annotation>< / semantics>< / com:Math> is non-zero, the estimation has failed as the distinguished logarithm is undefined. 12. Calculate <com:Math><semantics>f̂A<annotation encoding="application / x-tex">\hat{f}_A< / annotation>< / semantics>< / com:Math> using the FFT of <com:Math><semantics>ϕ̂A<annotation encoding="application / x-tex">\hat{\phi}_A< / annotation>< / semantics>< / com:Math> according to [Image disponible dans le document PDF, Image available in the PDF document] (99) 6.1 Algorithm Details Partition the detector output stream into a set of non-overlapping intervals of length <com:Math><semantics>L<annotation encoding="application / x-tex">L< / annotation>< / semantics>< / com:Math> i.e., <com:Math><semantics>[Tj,Tj+L),T0∈ℤ≥0,Tj+1≥Tj+L,j∈ℤ≥0<annotation encoding="application / x-tex">[T_j, T_j + L), T_0 \in \mathbb{Z}_{\geq 0}, T_{j+1} \geq T_j + L, j \in \mathbb{Z}_{\geq 0}< / annotation>< / semantics>< / com:Math>. Let <com:Math><semantics>xj<annotation encoding="application / x-tex">x_j< / annotation>< / semantics>< / com:Math> be the sum of the detector output samples in the <com:Math><semantics>j<annotation encoding="application / x-tex">j< / annotation>< / semantics>< / com:Math>th interval i.e., [Image disponible dans le document PDF, Image available in the PDF document] (100) Assuming L is greater than a pulse length, the jth interval may contain 'complete' pulses as well as pulses which have been truncated by the ends of the interval. It can be shown that <com:Math><semantics>xi<annotation encoding="application / x-tex">x_i< / annotation>< / semantics>< / com:Math> consists of a superposition of the energy of 'complete' pulses which we denote <com:Math><semantics>yj<annotation encoding="application / x-tex">y_j< / annotation>< / semantics>< / com:Math>, the energies of truncated pulses which we denote with <com:Math><semantics>y1j<annotation encoding="application / x-tex">y_{1j}< / annotation>< / semantics>< / com:Math> and noise <com:Math><semantics>zj<annotation encoding="application / x-tex">z_j< / annotation>< / semantics>< / com:Math> Let the detector output stream be partitioned into a second set of non-overlapping intervals <com:Math><semantics>[T1j,T1j+L1),T1,0∈ℤ≥0,T1,j+1≥T1,j+L,j∈ℤ≥0<annotation encoding="application / x-tex">[T_{1j}, T_{1j} + L_1), T_{1,0} \in \mathbb{Z}_{\geq 0}, T_{1,j+1} \geq T_{1,j} + L, j \in \mathbb{Z}_{\geq 0}< / annotation>< / semantics>< / com:Math> where <com:Math><semantics>L1<L<annotation encoding="application / x-tex">L_1 < L< / annotation>< / semantics>< / com:Math>. Let <com:Math><semantics>x1j<annotation encoding="application / x-tex">x_{1j}< / annotation>< / semantics>< / com:Math> be given by [Image disponible dans le document PDF, Image available in the PDF document] (101) If <com:Math><semantics>L1<annotation encoding="application / x-tex">L_1< / annotation>< / semantics>< / com:Math> is chosen to be slightly less than the pulse length, the <com:Math><semantics>x1j<annotation encoding="application / x-tex">x_{1j}< / annotation>< / semantics>< / com:Math> term will contain no 'com- plete' pulses, but consist of a superposition of only the energies of truncated pulses <com:Math><semantics>y1j<annotation encoding="application / x-tex">y_{1j}< / annotation>< / semantics>< / com:Math> and noise <com:Math><semantics>zj<annotation encoding="application / x-tex">z_j< / annotation>< / semantics>< / com:Math>. The number of truncated pulses in any interval has a Poisson distribution. We have [Image disponible dans le document PDF, Image available in the PDF document] (102) [Image disponible dans le document PDF, Image available in the PDF document] (103) We can decompose the total energy in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> into the energy con- tribution <com:Math><semantics>Y1<annotation encoding="application / x-tex">Y_1< / annotation>< / semantics>< / com:Math> from pulses that have been truncated and the energy contribution <com:Math><semantics>Y0<annotation encoding="application / x-tex">Y_0< / annotation>< / semantics>< / com:Math> from pulses that are fully contained in the interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math> i.e., [Image disponible dans le document PDF, Image available in the PDF document] (104) where <com:Math><semantics>Z0<annotation encoding="application / x-tex">Z_0< / annotation>< / semantics>< / com:Math> represents noise in the regions where pulses are fully contained in the interval (a length of <com:Math><semantics>L−L1<annotation encoding="application / x-tex">L - L_1< / annotation>< / semantics>< / com:Math>), and <com:Math><semantics>Z1<annotation encoding="application / x-tex">Z_1< / annotation>< / semantics>< / com:Math> represents noise in the regions where pulses are truncated (a length of <com:Math><semantics>L1<annotation encoding="application / x-tex">L_1< / annotation>< / semantics>< / com:Math>). Hence, [Image disponible dans le document PDF, Image available in the PDF document] (105) By combining (103) with (105) we have [Image disponible dans le document PDF, Image available in the PDF document] (106) Rearranging gives [Image disponible dans le document PDF, Image available in the PDF document] (107) [Image disponible dans le document PDF, Image available in the PDF document] (108) We can estimate <com:Math><semantics>ϕX1<annotation encoding="application / x-tex">\phi_{X_1}< / annotation>< / semantics>< / com:Math> in a similar manner that we estimated <com:Math><semantics>ϕX<annotation encoding="application / x-tex">\phi_X< / annotation>< / semantics>< / com:Math> or some other method, e.g., via the empirical characteristic function or by performing an FFT on the normalized histogram of <com:Math><semantics>x1j<annotation encoding="application / x-tex">x_{1j}< / annotation>< / semantics>< / com:Math> values. When performing the decompounding operation, the Poisson rate <com:Math><semantics>λ0<annotation encoding="application / x-tex">\lambda_0< / annotation>< / semantics>< / com:Math> for the reduced interval length <com:Math><semantics>L−L1<annotation encoding="application / x-tex">L-L_1< / annotation>< / semantics>< / com:Math> is used to account for the sub-interval over which the compound Poisson process <com:Math><semantics>Y0<annotation encoding="application / x-tex">Y_0< / annotation>< / semantics>< / com:Math> occurs. 6.2 Visualization of Internal Quantities To aid the reader's understanding, Figure 9 plots various quantities obtained during the estimation process. The upper blue curve (with a value around 0.3 at bin zero) plots <com:Math><semantics>|ϕ̂X|<annotation encoding="application / x-tex">|\hat{\phi}_X|< / annotation>< / semantics>< / com:Math>, the estimated characteristic function of the observed spectrum. A brown curve is used to show the true value of <com:Math><semantics>|ϕf|<annotation encoding="application / x-tex">|\phi_f|< / annotation>< / semantics>< / com:Math>, which is distinctly visible as the lower curve with periodic nulls in the region [6000, 10000]. The quantity <com:Math><semantics>|ϕϵ| / (λϕZe−λ)<annotation encoding="application / x-tex">|\phi_{\epsilon}| / (\lambda\phi_{Z}e^{-\lambda})< / annotation>< / semantics>< / com:Math> is shown in transparent red and appears as 'noise' whose average density peaks around bin #8000. The expected value of <com:Math><semantics>|ϕϵ| / (λϕZe−λ)<annotation encoding="application / x-tex">|\phi_{\epsilon}| / (\lambda\phi_Z e^{-\lambda})< / annotation>< / semantics>< / com:Math> is shown with a black dashed line. This is obtained using (75), the known value of <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math>, and assuming Gaussian noise with known <com:Math><semantics>σ<annotation encoding="application / x-tex">\sigma< / annotation>< / semantics>< / com:Math> to obtain <com:Math><semantics>ϕZ(k)<annotation encoding="application / x-tex">\phi_Z(k)< / annotation>< / semantics>< / com:Math>. The quantity <com:Math><semantics>|ϕ̂f|<annotation encoding="application / x-tex">|\hat{\phi}_f|< / annotation>< / semantics>< / com:Math> is shown with a transparent blue curve. This is barely visible as it coincides closely with <com:Math><semantics>|ϕf|<annotation encoding="application / x-tex">|\phi_f|< / annotation>< / semantics>< / com:Math> in the intervals <com:Math><semantics>[0,4000],[12000,16000],<annotation encoding="application / x-tex">[0, 4000], [12000, 16000],< / annotation>< / semantics>< / com:Math> and closely with <com:Math><semantics>|ϕϵ| / (λϕZe−λ)<annotation encoding="application / x-tex">|\phi_{\epsilon}| / (\lambda\phi_Z e^{-\lambda})< / annotation>< / semantics>< / com:Math> in the interval [5000, 11000]. Note that the colour of <com:Math><semantics>|ϕϵ| / (λϕZe−λ)<annotation encoding="application / x-tex">|\phi_{\epsilon}| / (\lambda\phi_Z e^{-\lambda})< / annotation>< / semantics>< / com:Math> appears to change from red to purple in the interval [5000, 11000] as both transparent plots overlap. A solid black line shows <com:Math><semantics>|ϕ̂fsmooth|<annotation encoding="application / x-tex">|\hat{\phi}_{fsmooth}|< / annotation>< / semantics>< / com:Math>, a low pass filtered version of <com:Math><semantics>|ϕ̂f|<annotation encoding="application / x-tex">|\hat{\phi}_f|< / annotation>< / semantics>< / com:Math>. The low pass filtering removes any local oscillations in <com:Math><semantics>|ϕ̂f(k)|<annotation encoding="application / x-tex">|\hat{\phi}_f(k)|< / annotation>< / semantics>< / com:Math> due to the peak localities, as described in paragraph on smoothing at the beginning of this section. The term <com:Math><semantics>|ϕ̂fsmooth(k)|<annotation encoding="application / x-tex">|\hat{\phi}_{\text{fsmooth}}(k)|< / annotation>< / semantics>< / com:Math> serves as an estimate of <com:Math><semantics>𝔼[|ϕf(k)|]<annotation encoding="application / x-tex">\mathbb{E}[|\phi_f(k)|]< / annotation>< / semantics>< / com:Math>. It can be seen that <com:Math><semantics>|ϕ̂f|<annotation encoding="application / x-tex">|\hat{\phi}_f|< / annotation>< / semantics>< / com:Math> provides a reasonably good estimate of <com:Math><semantics>|ϕf|<annotation encoding="application / x-tex">|\phi_f|< / annotation>< / semantics>< / com:Math> in the region where <com:Math><semantics>|ϕ̂fsmooth|≫𝔼[|ϕϵ| / (λϕZe−λ)]<annotation encoding="application / x-tex">|\hat{\phi}_{\text{fsmooth}}| \gg \mathbb{E}[|\phi_{\epsilon}| / (\lambda \phi_Z e^{-\lambda})]< / annotation>< / semantics>< / com:Math>. As these two quantities approach each other, the quality of the estimate deteriorates until it is eventually dominated by noise. The filter <com:Math><semantics>H(k)<annotation encoding="application / x-tex">H(k)< / annotation>< / semantics>< / com:Math> should include good estimates of <com:Math><semantics>|ϕf|<annotation encoding="application / x-tex">|\phi_f|< / annotation>< / semantics>< / com:Math> while excluding poor estimates. To find the regions where good estimates of <com:Math><semantics>|ϕf|<annotation encoding="application / x-tex">|\phi_f|< / annotation>< / semantics>< / com:Math> are obtained we address the question: Given <com:Math><semantics>𝔼[|ϕϵ| / (λϕZe−λ)]<annotation encoding="application / x-tex">\mathbb{E}[|\phi_{\epsilon}| / (\lambda\phi_{Z}e^{-\lambda})]< / annotation>< / semantics>< / com:Math>, what is the probability that the calculated values of <com:Math><semantics>ϕ̂f<annotation encoding="application / x-tex">\hat{\phi}_f< / annotation>< / semantics>< / com:Math> in a local region arise largely from noise? 7 COUNT RATE ESTIMATION The previous estimator assumed <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> was known. An estimate of <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> can be obtained without prior knowledge as follows. 1. Using <com:Math><semantics>ϕ̂X<annotation encoding="application / x-tex">\hat{\phi}_{X}< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>ϕ̂X1<annotation encoding="application / x-tex">\hat{\phi}_{X1}< / annotation>< / semantics>< / com:Math>, <com:Math><semantics>ϕZ<annotation encoding="application / x-tex">\phi_{Z}< / annotation>< / semantics>< / com:Math>, G from the previous section, calculate [Image disponible dans le document PDF, Image available in the PDF document] (109) 2. Using <com:Math><semantics>ϕ̂Y<annotation encoding="application / x-tex">\hat{\phi}_{Y}< / annotation>< / semantics>< / com:Math>, estimate the count rate. This can be done a number of ways. 3. One way is to use an optimization routine or some other means to fit a curve to <com:Math><semantics>ϕ̂Y<annotation encoding="application / x-tex">\hat{\phi}_{Y}< / annotation>< / semantics>< / com:Math>. The fitted parameters can be used to obtain an estimate of the count rate. 4. Another way involves estimating the DC offset of <com:Math><semantics>Ψ=dlog⁡(ϕ̂𝒀)<annotation encoding="application / x-tex">\Psi = \operatorname{dlog}(\hat{\phi}_{\boldsymbol{Y}})< / annotation>< / semantics>< / com:Math>. This can be done by averaging a suitable number of points of <com:Math><semantics>Ψ<annotation encoding="application / x-tex">\Psi< / annotation>< / semantics>< / com:Math>. The points obtained by filtering by <com:Math><semantics>H(u)<annotation encoding="application / x-tex">H(u)< / annotation>< / semantics>< / com:Math> in the previous section are usually suitable, although less points may also produce an adequate estimate. 5. Another way involves using an optimization engine or some other means to fit a curve to <com:Math><semantics>Ψ=dlog⁡(ϕ̂Y)<annotation encoding="application / x-tex">\Psi = d\log(\hat{\phi}_{Y})< / annotation>< / semantics>< / com:Math>. A suitable parameterized curve to fit <com:Math><semantics>dlog⁡(ϕ̂Y)<annotation encoding="application / x-tex">d\log(\hat{\phi}_{Y})< / annotation>< / semantics>< / com:Math> is given by [Image disponible dans le document PDF, Image available in the PDF document] (110) [Image disponible dans le document PDF, Image available in the PDF document] (111) [Image disponible dans le document PDF, Image available in the PDF document] (112) [Image disponible dans le document PDF, Image available in the PDF document] (113) and where <com:Math><semantics>K∈ℕ<annotation encoding="application / x-tex">K \in \mathbb{N}< / annotation>< / semantics>< / com:Math> is chosen to allow the curve fit to sufficient accuracy. The parameter <com:Math><semantics>λ<annotation encoding="application / x-tex">\lambda< / annotation>< / semantics>< / com:Math> provides an estimate of the count rate. The optimization engine is not required to give equal weighting to each point in <com:Math><semantics>Ψ<annotation encoding="application / x-tex">\Psi< / annotation>< / semantics>< / com:Math>. 8 DESCRIPTION OF FIGURES The following figures are to aid understanding of the process. Figure 1 shows one possible scheme used to partition the detector output. The illustration depicts the sampled detector response to three incident photons. To aid clarity of the figure, the effects of noise have been removed. The output response has been partitioned into several regions of equal length <com:Math><semantics>(L)<annotation encoding="application / x-tex">(L)< / annotation>< / semantics>< / com:Math>. The number of pulses arriving in each region is unknown to the processing system. One pulse has arrived in the first interval. Two pulses have arrived in the second interval. No pulses have arrived in the third interval. The total photon energy arriving within each interval is calculated as the statistic of interest, being the sum of all sample values in each interval. Intervals are not temporally aligned with pulse arrivals. Figure 2 illustrates the output of the estimation procedure. The true probability density of incident photon energy is plotted as a soild black line. The photons arrival rate is such that three photons on average arriving during any given interval <com:Math><semantics>[Tj,Tj+L)<annotation encoding="application / x-tex">[T_j, T_j + L)< / annotation>< / semantics>< / com:Math>. The standard deviation of additive noise in the detector output signal <com:Math><semantics>s(t)<annotation encoding="application / x-tex">s(t)< / annotation>< / semantics>< / com:Math> is equal to one histogram bin width. One million intervals were collected. A histogram was made of the total energy in each interval. This is plotted with the blue line. The effects of pile-up is clearly evident, particularly around bins 75, 150 and 225. The red trace plots the estimate of the true incident energy spectrum after the data has been processed by the system. Although some noise appears in the estimate, the effects of pile-up have been removed. The estimate is expected to correctly recover the true incident spectrum on average. This result was obtained using an internal filter whose bandwidth was determined automatically from the data. Figure 3 illustrates the same quantities as Figure 2 under the same operating conditions, however in this instance the bandwidth of the internal filter has been determined using asymptotic results from the literature. Although the estimated probability density of incident energies has been recovered, the variance is significantly greater compared to Figure 2. Figure 8 illustrates the operation of the system on real data. The blue trace plots the probability density of observed energy values, while the red trace plots the estimated true probability density of incident photon energies. There is no black trace as the true probability density is unknown. In this experiment, X-ray fluorescence of a manganese sample was used as a photon source. The photon arrival rate was around <com:Math><semantics>105<annotation encoding="application / x-tex">10^5< / annotation>< / semantics>< / com:Math> photons per second. The interval length was chosen such that the average time between photons corresponded to the length of two intervals. Sufficient data was collected and partitioned to form <com:Math><semantics>5.9x106<annotation encoding="application / x-tex">5.9x10^6< / annotation>< / semantics>< / com:Math> intervals. The standard deviation of the additive noise corresponds to 4.7 histogram bins. The estimation process has clearly reduced the pile-up peaks and enhanced the true peaks. Figure 9 illustrates various quantities obtained during the simulation of the system described in the 2nd embodiment. It is described in section 5.1 Visualization of Internal Quantities. Figures 9-12 relate to the 2nd embodiment. Figure 10 illustrates the observed and true probability density of input photon energies for the experiment from which Figures 9 - 13 were derived. The black trace plots the true probability density. The red trace plots the density expected to be observed when three photons on average arrive during a given interval length. The blue trace plots the actual observed density. Up to tenth-order pile-up can be seen in the observed density. Figure 10 included several plots arising from a typical spectroscopic system. The actual incident photon density ('Ideal Density') is plotted with a solid dark line. An observed histogram obtained by partitioning the time-series data is shown in dark blue. Distortion of the spectrum caused by pulse pile-up is evident Figure 13 plots various internal quantities using a logarithmic vertical axis. The dark blue curve that dips in the centre of the plot is <com:Math><semantics>|ϕ̂X|<annotation encoding="application / x-tex">|\hat{\phi}_{X}|< / annotation>< / semantics>< / com:Math>. The green quantity that crosses the plot horizontally is <com:Math><semantics>|ϕ̂Y|<annotation encoding="application / x-tex">|\hat{\phi}_{Y}|< / annotation>< / semantics>< / com:Math>. The upper cyan curve that dips in the centre of the plot is <com:Math><semantics>|ϕZ|.1<annotation encoding="application / x-tex">|\phi_Z|.1< / annotation>< / semantics>< / com:Math> Figure 11 illustrates the trajectory of the curve <com:Math><semantics>γ<annotation encoding="application / x-tex">\gamma< / annotation>< / semantics>< / com:Math> in the complex plane. Figure 12 illustrates internal quantities similar to Figure 9, however there are some additional signals. The horizontal red trace that is largely noise, and the corresponding black dashed line represent <com:Math><semantics>|ϕϵ|<annotation encoding="application / x-tex">|\phi_{\epsilon}|< / annotation>< / semantics>< / com:Math>, the magnitude of the characteristic function of the histogram noise. The transparent green plot that forms the 'noisy peak' in the center of the figure is the estimate <com:Math><semantics>|ϕÂ|<annotation encoding="application / x-tex">|\hat{\phi_A}|< / annotation>< / semantics>< / com:Math>. This quantity was plotted in blue in Figure 9, and was barely visible as it was obscured by <com:Math><semantics>|ϕX| / (λϕZe−λ)<annotation encoding="application / x-tex">|\phi_X| / (\lambda\phi_Z e^{-\lambda})< / annotation>< / semantics>< / com:Math>, which is not shown in Figure 12. The horizontal trace with an average value of -3 is a plot of <com:Math><semantics>|ϕY|<annotation encoding="application / x-tex">|\phi_Y|< / annotation>< / semantics>< / com:Math>. The cyan trace that begins with a value of zero at bin zero, and dips to a minimum around bin 8000 is <com:Math><semantics>|ϕZ|<annotation encoding="application / x-tex">|\phi_Z|< / annotation>< / semantics>< / com:Math>, the magnitude of the characteristic function of the additive Gaussian noise. Figure 13 relates to the 2nd broad aspect of calculating the count rate. It illustrates internal quantities used in the calculation of <com:Math><semantics>λ̂<annotation encoding="application / x-tex">\hat{\lambda}< / annotation>< / semantics>< / com:Math>. The cyan trace that begins with a value of zero at bin zero, and dips to a minimum around bin 8000 is <com:Math><semantics>|ϕZ|<annotation encoding="application / x-tex">|\phi_Z|< / annotation>< / semantics>< / com:Math>, the magnitude of the characteristic function of the additive Gaussian noise. The dark blue trace that dips to a minimum in the center of the Figure is <com:Math><semantics>|ϕ̂X|<annotation encoding="application / x-tex">|\hat{\phi}_X|< / annotation>< / semantics>< / com:Math>, the estimate of the characteristic function of the observed data. The yellow / green horizontal trace with an average value of -3 is the estimate of <com:Math><semantics>|ϕY|<annotation encoding="application / x-tex">|\phi_Y|< / annotation>< / semantics>< / com:Math>. References [1] G. F. Knoll, Radiation Detection and Measurement, 3rd Edition. New York: Wiley, 2000. [2] P. A. B. Scoullar and R. J. Evans, "Maximum likelihood estimation techniques for high rate, high throughput digital pulse processing," in 2008 IEEE Nuclear Science Symposium Conference Record, pp. 1668–1672, Oct. 2008. [3] M. Haselman, J. Pasko, S. Hauck, T. Lewellen, and R. Miyaoka, "FPGA-based pulse pile-up correction with energy and timing recovery," IEEE Transactions on Nuclear Science, vol. 59, pp. 1823–1830, Oct. 2012. [4] T. Petrovic, M. Vencelj, M. Lipoglavsek, R. Novak, and D. Savran, "Efficient re- duction of piled-up events in gamma-ray spectrometry at high count rates," IEEE Transactions on Nuclear Science, vol. 61, pp. 584–589, Feb. 2014. [5] B. A. VanDevender, M. P. Dion, J. E. Fast, D. C. Rodriguez, M. S. Taubman, C. D. Wilen, L. S. Wood, and M. E. Wright, "High-purity germanium spectroscopy at rates in excess of 10<com:Sup>6< / com:Sup> events / s," IEEE Transactions on Nuclear Science, vol. 61, pp. 2619–2627, Oct. 2014. [6] Y. Sepulcre, T. Trigano, and Y. Ritov, "Sparse regression algorithm for activity esti- mation in spectrometry," IEEE Transactions on Signal Processing, vol. 61, pp. 4347- 4359, Sept. 2013. [7] T. Trigano, I. Gildin, and Y. Sepulcre, "Pileup correction algorithm using an iterated sparse reconstruction method," IEEE Signal Processing Letters, vol. 22, pp. 1392- 1395, Sept. 2015. [8] L. Wielopolski and R. P. Gardner, "Prediction of the pulse-height spectral distortion caused by the peak pile-up effect," Nuclear Instruments and Methods, vol. 133, pp. 303–309, Mar. 1976. [9] N. P. Barradas and M. A. Reis, "Accurate calculation of pileup effects in PIXE spectra from first principles," X-Ray Spectrometry, vol. 35, pp. 232–237, July 2006.

[10] T. Trigano, A. Souloumiac, T. Montagu, F. Roueff, and E. Moulines, "Statistical pileup correction method for HPGe detectors," IEEE Transactions on Signal Pro- cessing, vol. 55, pp. 4871–4881, Oct. 2007.

[11] T. Trigano, E. Barat, T. Dautremer, and T. Montagu, "Fast digital filtering of spectrometric data for pile-up correction," IEEE Signal Processing Letters, vol. 22, pp. 973–977, July 2015.

[12] P. Ilhe, E. Moulines, F. Roueff, and A. Souloumiac, "Nonparametric estimation of mark's distribution of an exponential shot-noise process," Electronic Journal of Statistics, vol. 9, no. 2, pp. 3098–3123, 2015.

[13] P. Ilhe, F. Roueff, E. Moulines, and A. Souloumiac, "Nonparametric estimation of a shot-noise process," in 2016 IEEE Statistical Signal Processing Workshop (SSP), pp. 1–4, June 2016.

[14] C. McLean, M. Pauley, and J. H. Manton, "Limitations of decision based pile-up correction algorithms," in 2018 IEEE Statistical Signal Processing Workshop (SSP), pp. 693–697, June 2018.

[15] D. Snyder and M. Miller, Random Point Processes In Time And Space. New York: Springer-Verlag, 2, revised ed., Sept. 2011.

[16] B. Buchmann and R. Grübel, "Decompounding: An estimation problem for Poisson random sums," Annals of Statistics, pp. 1054–1074, 2003.

[17] B. van Es, S. Gugushvili, and P. Spreij, "Deconvolution for an atomic distribution," Electronic Journal of Statistics, vol. 2, pp. 265–297, 2008.

[18] S. Gugushvili, Non-Parametric Inference for Partially Observed Levy Processes. PhD, University of Amsterdam, Thomas Stieltjes Institute, 2008.

[19] S. Said, C. Lageman, N. Le Bihan, and J. H. Manton, "Decompounding on compact Lie groups," IEEE Transactions on Information Theory, vol. 56, pp. 2766–2777, June 2010.

[20] B. van Es, S. Gugushvili, and P. Spreij, "A kernel type nonparametric density esti- mator for decompounding," Bernoulli, vol. 13, pp. 672–694, Aug. 2007.

[21] J. Yu, "Empirical characteristic function estimation and its applications," Econo- metric Reviews, vol. 23, pp. 93–123, Dec. 2004.

Claims

1. A method of detecting a radiation by a radiation detector and determining a spectrum of energies of individual quanta of the radiation, the method comprising the steps of: (1) converting the detected radiation into a time series of digital observations comprising pulses corresponding to the detection of the individual quanta; (2) computing spectrum sensitive statistics from the detector signal, the spectrum sensitive statistics defining a mapping from a density of amplitudes of the pulses to the spectrum sensitive statistics using an approximate compound Poisson process; and (3) determining the spectrum by estimating the density of amplitudes of the pulses by applying an inversion of the mapping to the spectrum sensitive statistics; 2. The method of claim 1 further comprising basing the spectrum sensitive statistics on a sum of the digital observations over a plurality of time intervals.

3. The method of claim 2, further comprising augmenting the approximate compound Poisson process by a modelled noise.

4. The method of claim 3, further comprising expressing the mapping as a relation between characteristic functions of the amplitudes, the spectrum sensitive statistics and the modelled noise.

5. The method of claim 4, further comprising computing the characteristic functions of the spectrum sensitive statistics by applying an inverse Fourier transform to a histogram of the sum of the digital observations.

6. The method of claim 4 or claim 5, further comprising computing the characteristic functions of the amplitudes with a low pass filter.

7. The method of any one of claims 2 to 6, further comprising selecting each of the plurality of time intervals to encompass zero or more approximately entire clusters of the pulses, and defining the plurality of time intervals to be nonoverlapping and have a constant length L.

8. The method of claim 7, further comprising requiring a maximum value of the detector signal at a beginning and end of each time interval.

9. The method of claim 7 or 8, further comprising defining the approximate compound Poisson process as a sum of the amplitudes of the pulses in each time interval.

10. The method of any one of claims 2 to 6, further comprising selecting the plurality of intervals to include: a first set of nonoverlapping time intervals of constant length L without regard to entirety of clusters of the pulses; and a second set of nonoverlapping time intervals of constant length L1 less than L also without regard to entirety of clusters of the pulses; wherein L is at least as long as a duration of the pulses.

11. The method of claim 10, further comprising selecting L1 to be less than the duration of the pulses.

12. The method of any one of claims 1 to 11, further comprising using a data driven strategy selected to result in a near optimal choice for a kernel parameter which minimises an integrated square of errors of an estimated probability density function of the energies of the individual quanta of radiation.

13. A method of detecting a radiation by a radiation detector and estimating a count rate of individual quanta of the radiation, the method comprising the steps of: (1) converting the detected radiation into a time series of digital observations from the radiation detector comprising pulses corresponding to the detection of the individual quanta; (2) computing spectrum sensitive statistics from the time series, based on a sum of the digital observations over a plurality of time intervals, the spectrum sensitive statistics defining a mapping from a density of amplitudes of the pulses to the spectrum sensitive statistics using an approximate compound Poisson process, the plurality of time intervals including: a first set of nonoverlapping time intervals of constant length L selected without regard to entirety of clusters of the pulses; and a second set of nonoverlapping time intervals of constant length L1 less than L also selected without regard to entirety of clusters of the pulses; wherein L is at least as long as a duration of the pulses; (3) determining an estimate <com:Math><semantics>ϕ̂𝒚<annotation encoding="application / x-tex">\widehat{\phi}_{\mathbf{y}}< / annotation>< / semantics>< / com:Math> of a characteristic function of the approximate compound Poisson process using [Image disponible dans le document PDF, Image available in the PDF document] where <com:Math><semantics>G<annotation encoding="application / x-tex">G< / annotation>< / semantics>< / com:Math> is a windowing function, <com:Math><semantics>∅̂X<annotation encoding="application / x-tex">\widehat{\emptyset}_X< / annotation>< / semantics>< / com:Math> is an estimate of a characteristic function of a sum of the digital observations over each nonoverlapping time interval in the first set, <com:Math><semantics>∅Z<annotation encoding="application / x-tex">\emptyset_Z< / annotation>< / semantics>< / com:Math> is a characteristic function of a modelled noise process and <com:Math><semantics>∅̂X1<annotation encoding="application / x-tex">\widehat{\emptyset}_{X_1}< / annotation>< / semantics>< / com:Math> is an estimate of a characteristic function of a sum of the digital observations over each nonoverlapping time interval in the second set.; (4) estimating the count rate from the estimate of the characteristic function.

14. The method of claim 13, further comprising estimating the count rate by using an optimisation routine or other means to fit a curve, estimating a DC offset of a logarithm of the estimate of the characteristic function, or fitting a curve to the logarithm of the estimate of the characteristic function.