A comprehensive experimental characterization of the iPIX gamma imager

The results of more than 280 different experiments aimed at exploring the main features and performances of a newly developed gamma imager, called iPIX, are summarized in this paper. iPIX is designed to quickly localize radioactive sources while estimating the ambient dose equivalent rate at the measurement point. It integrates a 1 mm thick CdTe detector directly bump-bonded to a Timepix chip, a tungsten coded-aperture mask, and a mini RGB camera. It also represents a major technological breakthrough in terms of lightness, compactness, usability, response sensitivity, and angular resolution. As an example of its key strengths, an 241Am source with a dose rate of only few nSv/h can be localized in less than one minute.


Introduction
In-situ localization of radiation emitting objects or hotspots is of prime concern during decommissioning and clean-up activities of nuclear facilities at the end of their life-cycle. It is also of great importance in homeland security for detecting illicit traffic of nuclear materials as well as in crisis managements of radiological emergency situations for recovering orphan or lost radioactive sources on the one side, and for delimiting suspicious radiological contaminated areas after an accident event or malevolent act on the other side [1][2][3]. Gamma imaging technique, applied for the first time by Anger in 1952 as a human body tomographic scanner to detect early stage tumours [4], provides an optimal solution to track Xor γ-ray emitting objects from greater distances than conventional rate meters, thus significantly reducing the radiation dose received by operators in line with the well-known ALARA principle. However, the associated equipment usually used for medical applications is generally too bulky, heavy, and complex to be employed for in-field monitoring.
-1 -An earlier attempt to develop a portable gamma imager was carried out in the 1990s by CEA. The outcome of this effort was the CARTOGAM instrument [5], which has been widely commercialized by CANBERRA during the last decade. This instrument is a mature technology combining an inverted double-cone collimator (i.e., two right circular cones placed apex-to-apex) to be used as a single pinhole aperture, a CsI/Tl (thallium doped caesium iodide) scintillation crystal, a multi-channel image intensifier, and a CCD camera. With the exception of the pinhole aperture, the whole instrument is fully shielded against background radiation.
CARTOGAM applies the same approach of a basic photographic camera (i.e., a light-opaque box with a pinhole aperture through which the light coming from an object is projected inside to form an inverted image). Although it is still demonstrating good performances, namely at severe irradiation conditions, its heavy weight (∼18 kg) poses several constraints from a practical point of view. In addition, its detection efficiency is quite limited by the small size of its pinhole aperture.
To overcome these shortcomings, CANBERRA has recently developed a second-generation gamma imager, called iPIX (figure 1), in the framework of a partnership agreement with CEA [6][7][8]. iPIX integrates a 1 mm thick CdTe (cadmium telluride) detector directly bump-bonded to a Timepix chip [9], a tungsten coded-aperture mask (i.e., a front-end collimator with several pinholes) [10], and a mini RGB camera. Under this configuration, the real view of the scene under study is firstly taken by the mini RGB camera and is subsequently superimposed by the associated X-or γ-ray information (i.e., a sort of color map display) derived as a result of the mask pattern design (see section 2.2 for more details).
The coded-aperture technique has played a crucial role in astronomy to help resolving the origin of γ-ray bursts from distant galaxies [11] and can be used under a mask/anti-mask mode [12] to reduce the background noise interference without the need of heavy radiation shielding. This allows for an ultra-portable (2.35 kg) and compact (9 × 9 × 18.8 cm 3 ) device, which is remotely controlled (via a single Ethernet cable up to 80 meters long) by means of an easy-to-use, fully ergonomic, and highly intuitive Graphical User Interface (GUI). The latter when launched is systematically configured and, with only one click, a new acquisition can be started.
Other main features of iPIX are: • Easily deployable and plug-and-play system, • Automatic parameter settings, • Dynamic display of the measurement progress, • Industrial design for use in harsh environments, • Dust and water proof, • Easy to clean, especially for unwanted radioactive contamination, • Protection against unpredictable vibrations, chocks and falls, • More than five hours autonomy, and • Motorized pan-tilt support.
-2 - The present paper is mainly focused on the description of the iPIX gamma imager. It also analyzes and discusses the results of a comprehensive experimental study, carried out with ISO narrow X-ray beams or N-Series [13] and standard radioactive sources.

Timepix chip
The Timepix chip is a pixelated CMOS ASIC directly bump-bounded to a given semiconductor substrate in order to detect X-or γ-rays (hereafter referred to as "photons"). It consists of 256 × 256 square pixels with 55 µm side, providing a global detection area of 14 × 14 mm 2 . Each pixel contains ∼550 transistors with ∼13.5 µW static power consumption and it has its own electronic circuitry: charge-sensitive amplifier, discriminator with polarity control pin, 4-bit DAC offset for threshold adjustment, synchronization logic, 8-bit configuration register, and 14-bit counter.
The photon detection is based on a depletion region that is formed in the vicinity of the PN junction under a reverse-bias voltage. The number of collected charge carriers (electrons and holes) is expected to be proportional to the amount of the energy deposited in the depletion region by the passage of ionizing radiation. However, the active pixels give no signal when the pre-set threshold level (THL) is not crossed. The 14-bit counter of each pixel operates only when the Timepix shutter is open (see section 2.1.1) and has an overflow control that stops at 11810 counts [9]. All the Timepix parameters are easily configured via Pixelman, a cross-platform and modular software tool developed by the Czech Technical University in Prague [14]. Figure 2 compares the energy dependence of photon absorption probability, which is directly related to the photon detection efficiency, for the three substrates (Si, CdTe and GaAs) that can be currently chosen for the Timepix chip. According to this figure, CdTe has globally a better photon absorption probability due to its high atomic number and density (Z = 48.52 and ρ = 5.32 g cm −3 ). The maximum difference, up to a factor 20 around 50-70 keV, is shown between CdTe and Si. At this energy range, the photon absorption probability on CdTe is near 50%. However, the three materials provide quite comparable behaviour at photon energies above 400 keV showing low absorption probabilities (<1%).

Counting modes
Every pixel of the Timepix chip can be independently configured to run in one of the following four different operating modes: • Masked mode: individual pixels can be masked.
• Medipix (MPX) mode: the pixel counter is incremented by one each time its output signal crosses the pre-set threshold level.
• Time-over-Threshold (ToT) or energy mode: the counter is incremented continuously as long as the pixel output signal is above the pre-set threshold level. The number of clock cycles is counted until the pixel output signal crosses back below this threshold. As stated in section 2.1.3, the number of clock cycles can be directly related to the voltage amplitude of the pixel output signal.
• Time-of-Arrival (ToA) or time mode: the counter is incremented continuously from the time the pixel output signal crosses the pre-set threshold level until the closure of the shutter. This mode is beyond the iPIX scope and will not be addressed hereafter since it is essentially used in the Large Hadron Collider (LHC) at CERN for accurate tracking of fundamental particles [15].
A reference clock with up to 100 MHz frequency is used for the last two counting modes. However, as the iPIX default clock frequency is fixed to 9.6 MHz, each ToT unit corresponds to a time period of 1.042×10 −7 sec.
As pointed out before, all the above counting modes rely on the configured shutter-time, which is the active time during which the Timepix sensor is kept switched on. When this shutter-time ends, pixel values are retrieved from the buffer and cleared. The process repeats for the duration of the measurement (i.e., until the chosen number of frames is reached). A comparison between MPX and ToT modes is illustrated in figure 3.

Charge sharing between adjacent pixels
When an incident photon (γ-or X-ray) interacts with an electron within the depletion region, mainly via photoelectric effect or Compton scattering, this primary electron then ionizes nearby atoms releasing more electron-hole pairs. Consequently, these charge carriers drift in opposite directions towards the corresponding electrodes because of the electric field and may be collected at the same time by several adjacent pixels, resulting in a "cluster".
Therefore, each registered cluster can be directly associated to the original photon interaction. This means that each cluster is equivalent to a "count" for conventional radiation detectors. In addition, the whole charge created by any incident photon can be revealed by summation of the fractional charges collected by all the pixels forming its associated cluster. The extent of charge sharing effect generally depends on the incident photon energy, pixel size, substrate layer, applied bias voltage, and the interaction depth within the sensitive volume [16].
A "cluster reconstruction" method (see figure 4) has been applied to correct for the above charge sharing effect. This method consists of determining the span of each registered cluster and defining the associated pixel with the highest ToT value as the possible location on the sensor surface of the original photon interaction. All the registered clusters can then be reduced to just the origin interaction pixels, improving the precision of the measurement. Furthermore, when using the ToT mode, the reading value of all the nearby pixels forming a given cluster can be summed to return the total energy deposited by the primary incident photon.
The method of cluster reconstruction also allows for a direct integration of the whole clusters registered along successive frames taken during the same measurement. Therefore, the number of registered clusters divided by the detection "live-time" (i.e., the preset shutter-time multiplied by the total recorded frames) directly provides the conventional "count rate" or "cps".
In principle, as illustrated in figure 4, both ToT and MPX modes are able to generate similar count rates under identical irradiation conditions. The only difference in MPX mode is that, as all the pixels associated to a given cluster provide the same reading value (one hit or 1), an arbitrary criterion regarding the possible location of the original photon interaction must be applied. This is why the cluster reconstruction step in the MPX mode is represented in figure 4 with a hatched arrow.

Pixel output signal
The pixel output signal is approximated to a triangular pulse, with ∼100 ns rise time and a fall time that can reach several tens of microseconds for the photon energy of interest, ranging between 10 keV and 1.5 MeV. Due to this triangular shape, the time duration (i.e., the number of clock cycles divided by the clock frequency in ToT mode) of any pixel output signal above the pre-set threshold level remains proportional to its voltage amplitude, which in turn is directly related to the portion of the total energy deposited by the incident photon that has been transferred to the considered pixel.
To have an idea how the ToT value varies with the associated energy deposited by the incident photon, figure 5 illustrates the example of single-pixel clusters. The green curve in this figure puts into evidence that the resulting ToT-energy calibration curve can be generally described by the following non-linear function depending on four parameters (a, b, c and d): This function has been experimentally confirmed by Jakubek et al. [17] showing a linear behaviour down to a given critical energy (usually near 10 keV) because of the threshold level that must be applied to avoid unwanted leakage current contribution (see section 2.1.6). Below this critical energy, the calibration curve deviates from linearity.

Pulse pile-up under intense photon fluxes
Under an excessive photon flux, the detector becomes saturated resulting in the pulse pile-up phenomenon. As illustrated in figure 6, when successive pulses are very close in time the system will wrongly record them as a single count with combined amplitude.

Pulse summation due to extended shutter-times
When selecting an excessive shutter-time, even in case of a low incident photon flux, successive signals can be collected and accumulated before it expires (see figure 7). The resulting reading value under ToT mode is the sum of all of the individual signals that are detected on the particular pixel within this excessive shutter-time. For this purpose, the shutter-time must be conveniently chosen to avoid this effect. If successive signals are recorded on the same pixel under ToT mode, the corresponding reading would lead to a meaningless numerical value. This effect can be minimized by decreasing the shutter-time to a limit of not setting it shorter than the time it takes for the pulse to be fully collected by the pixel. In addition, a very short shutter-time has a great risk of not providing enough statistics, especially at low photon fluxes. The iPIX response dependence with regards to the pre-set shutter-time is further explored in section 3.1.

Threshold equalization
Threshold equalization is used to compensate the pixel-to-pixel threshold variations of the Timepix readout chip due to local transistor mismatches. This procedure is normally applied using the inherent electronic noise as a trigger (i.e., the measurement is carried out in a radiation-free environment). In that case, the system tries to automatically fine-tune the 4-bit DAC offset for every pixel to be as close as possible to an optimum threshold level common to all pixels so that unwanted leakage current contribution is conveniently discarded but not the input pulses of interest.
A new equalization method was specifically developed for iPIX to well fix the leaking current issue. This method considers the real sensor response, instead of the inherent electronic noise, under a homogeneous X-or γ-ray flux in the energy range between 50-100 keV.

Coded-aperture mask
Three coded-aperture masks, providing two different mask ranks (see figure 8) and three different thicknesses, are currently available for the iPIX platform. They are identified according to their rank and thickness: R7e4, R7e8 and R13e2. The label "e" is an abbreviation of the word épaisseur (a French translation of "thickness"). Their planar area is four times that of the CdTe detector and their patterns were designed according to the well-known uniformly redundant arrays [18].
The benefit of a high mask rank is a more precise angular resolution with counterpart a decrease of the detection efficiency due to its less open area. Whereas increasing the mask thickness provides more photon filtering thus enhancing the signal-to-noise ratio but, in turn, it reduces the iPIX offaxis response (see section 3.7 for more details). The two masks of rank 7 can be inverted by a 90 • rotation to perform anti-mask measurements. The application of the mask R13e2 is only limited for low photon energies. More specifically, this mask is used to detect the presence of plutonium in nuclear wastes via the 59 keV γ-ray of 241 Am.
-8 - By placing the coded-aperture mask in front of a photon beam, emitted from a given radioactive hotspot located within its field-of-view (FOV), a shadow of the "illuminated" portion of its pattern is cast onto the detector surface. It is thus necessary to apply a de-convolution process of the resulting data collected with the Timepix chip to accurately reconstruct the location of the radioactive hotspot on the scene under study. The interested reader is referred to [18] and [19] for more information about the mathematical rules behind the coded-aperture technique.

Irradiation set-up
The iPIX experimental characterisation has been carried out at by means of the wide-range irradiation equipment located at CANBERRA's Loches site as well as the KIT X-ray irradiation facility.
The CANBERRA irradiation equipment (figure 9) consists of: 1. A "big" irradiator with six interchangeable standard 137 Cs sources of nominal activities between 3 GBq and 30 TBq to provide a dose rate ranging from 20 µSv/h to 20 Sv/h at several source-to-detector distances.

2.
A "small" irradiator with two interchangeable standard 137 Cs (∼300 MBq) and 60 Co (∼70 MBq) sources to provide a dose rate at the chosen measurement point ranging from 2 µSv/h to 100 µSv/h.

3.
A "portable" irradiator based on a standard 241 Am source with ∼1.85 GBq nominal activity to provide a dose rate ranging from 1 to 50 µSv/h.
The KIT X-ray irradiation facility (figure 10) is equipped with secondary standards and is able to provide the ISO standard narrow beams or N-Series [13]. All the X-ray beams used in this study, together with their theoretical energy distributions, are listed in table 1 and table 2. Increasing input -9 -

JINST 11 P08012
electron currents were applied for each X-ray energy (i.e., at constant acceleration voltages in kV) in order to increase the associated flux at the considered measurement point.
For each irradiation run at the KIT X-ray irradiation facility, the following steps have been essentially used for the determination of the iPIX response to X-ray beams: • Firstly, a basic quantity such as air Kerma rate orK air expressed in mGy/h units was directly measured with a calibrated ionization chamber at the considered distance (∼2 meters).
• Then, the appropriate operational quantity, in terms of the ambient dose equivalent rate oṙ H * (10) expressed in mSv/h units, was derived by the application of the associated conversion coefficient derived from ICRP-74 [20]. The corresponding values for every X-ray beam at varied input currents are summarized in tables 1 and 2.
• Lastly, the iPIX device under test was placed at the same measurement distance to evaluate its response.
Almost the same steps were followed at CANBERRA irradiation facilities with the only difference of using the calibrated ionization chamber earlier before at five reference distances for the considered standard source to derive the associated fitting parameters (a and b) of the following empirical relationship:K where x is the source-to-detector distance in cm.

Data quality
In this section, a qualitative study of the measurement data collected at the KIT X-ray facility is performed. In this case, the iPIX count rates measured at each X-ray beam have been normalized to the corresponding dose rate (see tables 1 and 2) in order to facilitate comparison between the different shutter-times used. Indeed, when increasing the X-ray beam currents the whole area of the measured spectra must be linearly proportional to the associated dose rates. As expected, the masks R7e8 and R7e4 provide quite similar results whereas the ones obtained with R13e2 when the incident photon energy is above 200 keV were quite unsatisfactory (i.e., the radioactive hotspot is not anymore localized) due to its small thickness. The iPIX measurement data collected with the mask R7e8 are plotted in figures 11-21 to show the typical evolution of the cluster ToT distributions with respect to the incident X-ray beam energy and current as well as to the Timepix shutter-time.
According to these figures, when the shutter-time is appropriately chosen to avoid signal distortions eventually due to pulse pile-up and/or summation effects, as explained in sections 2.1.4 and 2.1.5, the associated count rates per unit of mSv/h must not change when only the X-ray beam current is varied. This is the case when 10 −4 sec shutter-time is chosen for all the considered ISO X-ray narrow beams and at practically the whole range of applied currents.
Conversely, caution should be taken when choosing a shutter-time of 10 −5 sec since it cannot be applied for photon energies above 150 keV. In fact, above this energy, the effect of incomplete -13 -      charge collection of the pixel output signals becomes apparent. Shutter-times higher than 10 −4 sec can only be applied for a limited range of X-ray beams and at very low currents with the exception of 1 sec that completely failed in providing physically meaningful results.
Furthermore, a clear distinction has always to be made between radioactive hotspot localization (possible even under pulse pile-up and/or summation effects) and dose rate evaluation that can be done only when dose rate linearity is fully guaranteed.
It must be noticed that interactions of low energy photons with the CdTe substrate occur primarily via the photoelectric process giving rise to a single full-energy peak, which as expected shifts to higher ToT values with increasing the beam acceleration voltage. When the incident photon energy is above 40 keV, an additional peak appears on the left at around 13 ToT (i.e., ∼1.4 µsec) as shown in figure 15. In the beginning, we believed that it is a consequence of the possible -16 -  escape from the detector sensitive volume of characteristic fluorescence X-rays (23-31 keV) locally emitted by Cd or Te atoms after photoelectric interactions. Nevertheless, this peak remains fixed and contrary to the full-energy peak it does not move to the right with increasing the incident photon energy. For this reason, we think that it could be a direct detection of extra characteristic fluorescence X-rays emitted by other elements outside the CdTe substrate. The lack of knowledge about the actual composition of all surrounding materials does not allow us to make any kind of prediction regarding this additional peak.

ToT-energy calibration
Only clearly differentiated and sharp full-energy peaks were considered for the ToT-energy calibration. This was the case for the incident photon mean energies ranging from 10 to 90 keV and 10 −4 sec shutter-time (see figures [11][12][13][14][15][16][17]. Each one of these peaks was fitted with a Gaussian function to derive the associated ToT value of its geometric centre. The results obtained are plotted in figure 22, where all the ToT values were divided by the default clock frequency to express them in time units. A satisfying linearity is observed between incident photon energies and their respective ToT values.   Despite the small thickness of the CdTe substrate, the corresponding 662 keV peak of 137 Cs as well as its Compton edge (∼478 keV) are also visible on logarithmic scale (figure 23). This ultimately ratifies the validity of the ToT-energy linearity also at high energies. The energy resolution for the 662 keV peak was found to be approximately 9%.
-19 - Similar 137 Cs spectra were obtained for other dose rates but considering highly significant statistics as well as appropriate shutter-times to avoid signal distortions due to pulse pile-up and/or summation effects (as explained in sections 2.1.4 and 2.1.5).
Taking into account the above ToT-energy calibration relationship, the expected ToT for the 1500 keV energy is ∼85 µsec so that, as a general rule, the iPIX shutter-time should never decrease below 10 −4 µsec to be sure that all pixel output signals generated in the photon energy range of interest (10 keV-1.5 MeV) will be fully recorded.

Dose rate linearity
The variation of the iPIX count rates with respect to the reference dose rates was studied at the CANBERRA big irradiator considering the two masks R7e4 (see figure 24) and R7e8. Again, these masks provide quite similar results. In fact, depending on the chosen shutter-time, almost all the dose rate response curves are linear along a given range until reaching a maximum, after which the radioactive hotspot is no longer localized. These curves finally drop off towards a full saturation of the Timepix detector when a single cluster containing all of its pixels is registered. It can be also seen in figure 24 that the iPIX dose rate response may be considered as quite linear over a wide range reaching nearly 8 decades (10 −6 -10 1 Sv/h) if the Timepix shutter-time is correctly chosen to avoid signal distortions due to pulse pile-up and/or summation effects.

Dose calibration factor vs. incident photon energy
The iPIX dose calibration factor is energy dependent and is defined as the number of the registered net counts per second (cps) divided by the photon ambient dose equivalent rate (in mSv/h units) at the measurement point. The corresponding values obtained for all the CANBERRA radioactive -20 -  sources as well as for the KIT X-ray narrow beams are displayed on figure 25. This figure shows that the two masks R7e4 and R7e8 provide quite similar response whereas that of R13e2 is almost two times lower as a result of its high rank as stated in section 2.2. On the other hand, the iPIX response behaviour with respect to the incident photon energy is fully consistent with figure 2. Indeed, the dose calibration factor for an 241 Am radioactive source is ∼30 and ∼50 times higher than for 137 Cs and 60 Co, respectively.

Cluster sizes vs. incident photon energy
In this section, we examine the dependence of the cluster size with respect to incident photon energy. In each case, the shutter-time was appropriately chosen to avoid signal distortions on the Timepix detector due to pulse pile-up and/or summation effects. Once again, as all the coded-aperture masks provide quite similar results, only those obtained with the mask R7e8 are shown in figures 26 and 27. According to figure 26 a discrimination criterion, especially between low (e.g. 241 Am) and high photon energies (e.g. 137 Cs and 60 Co), can be applied but only for a qualitative analysis to estimate an energy range of the observed hotspot(s). Conversely, the same criterion is not practical to quantify the individual contribution of a given radionuclide to the overall dose rate as the cluster size distributions generated by different photon energies are significantly overlapping. In other words, there is not a clearly distinguishable signature in terms of cluster size distribution allowing both identification and quantification of radionuclides.
Another aspect to remark is that an upper size threshold of 20 pixels can be applied to discard unwanted cluster events such as the ones that may be generated by cosmic rays.

Sensitivity and influence of background radiation
The iPIX sensitivity has been studied with the low-activity point sources listed on table 3 but placing the CANBERRA small irradiator (figure 9) behind it in order to simulate, if needed, different background contributions. All the results obtained directly using the iPIX automatic -22 -   It can be observed in these figures that iPIX, once positioned, was fully able to localize by itself -without any further actions from the user -all the low-activity sources listed in table 3. Two criteria must be usually satisfied to decide if a radioactive source is adequately localized: 1) apparition of a hotspot in a clean picture without any kind of interferences or misleading artifacts, and 2) the corresponding peak is quite differentiated in a 3D profile plot, which shows the registered counts associated to each Timepix pixel after the de-convolution process (see section 2.2). Another aspect to highlight is that the parallax correction is no longer possible at source-to-detector distances less than 50 cm (see the mismatch between the actual location of the considered radioactive sources and that of the associated colored hotspots as determined by iPIX in figures 28 and 29).
1Under which iPIX is able to autonomously select, among other parameters, the appropriate shutter-time and the minimum number of frames needed for every measurement environment.

JINST 11 P08012
The minimum times needed, considering data collection only, to localize the above low-activity point sources are summarized in table 4. In all cases, the mask R7e8 mask always gives the fastest localization time. Moreover, iPIX was able to localize the 412 kBq 241 Am point source, providing a dose rate at the measurement point of only 2 nSv/h, with a minimum of 70 frames of 1 sec shutter-time. Additional measurements were also carried out as shown in table 5 to prove that iPIX has better performances than the CARTOGAM system. Further experiments with a similar configuration were performed, but placing this time the portable 241 Am source (see section 2.3) in front of iPIX. The source-to-detector distance was 330 cm in order to provide an ambient dose equivalent rate of ∼1 µSv/h at the measurement point. The following steps were sequentially applied: • CANBERRA small irradiator entirely shielded.
• Selecting the 60 Co source of the CANBERRA small irradiator to provide a background dose rate of ∼215 µSv/h at the measurement point.
• Selecting the 137 Cs source of the CANBERRA small irradiator to provide a background dose rate of ∼310 µSv/h at the measurement point.
The shutter-time was automatically fixed by iPIX to 1 sec and up to 200 frames were taken. The results obtained are presented in figures 31-33. Except for the mask R13e2, the two enhanced background radiation levels have caused a minimal impact to localize the portable 241 Am radioactive source.

Off-axis response
The evaluation of the iPIX off-axis response was carried out by means of the CANBERRA big irradiator. The iPIX gamma camera was firstly oriented so that the corresponding hotspot was visualized exactly at the center of its field-of-view. Thereafter, only its pan (horizontal) angle was varied sequently between −25 • to +25 • in steps of 5 • while keeping its tilt (vertical) position fixed -27 -

R7e8
R7e4 R13e2 to 0 • , and vice versa. The measurements were performed using the masks R7e4 and R7e8. The −25 • to +25 • variation was considered to include the whole iPIX field-of-view. The observed loss in counts for each mask was very similar with respect to both pan and tilt rotation angles. Figure 34 shows the average count rates, normalized to normal incidence (Pan = Tilt = 0 • ), measured with these two masks at varied pan or tilt rotation angle. The response deviation from an ideal behaviour (i.e., following the cosine function represented by a red dashed curve in -28 -

R7e8
R7e4 R13e2 Figure 33. Results obtained for the 1.85 GBq 137 Am source located at a distance of 330 cm to the iPIX with an enhanced background dose rate of ∼215 µSv/h generated by the 60 Co source of the CANBERRA small irradiator.
figure 34) is within 3%-12% for the mask R7e4 and never exceeds the limit of 25% for the mask R7e8. These deviations, which can be considered as negligible from the perspective of radiation dosimetry, are mainly due to the collimation effect of incident photons at the same coded-aperture mask. Admittedly, as illustrated in figure 35, the number of transmitted photons is decreased under non-normal incidences and thicker masks affecting moreover the shape of the projected open holes (they become oval instead of round) at the detector surface. Nevertheless, as proved with a preliminary proof-of-concept prototype of iPIX [8], the selfattenuation effect is even more accentuated at low-energy photons ( 241 Am like sources) and the corresponding decrease in the iPIX response may surpass 50% for both the mask R13e2 and R7e4 (or even 80% for the mask R7e8) when the non-normal incidence is above 10 • .
Concerning hotspot visualization, figure 36 displays an example of the superimposed pictures obtained at varied pan angles. According to this figure, a mirroring effect (i.e., false hotspot localization on the opposite side) may occur at the limit of the iPIX field-of-view (±25 • ). As illustrated in figure 37, this misleading artifact is caused by the projection on the detector surface of identical portions of the mask pattern. Therefore, it is suggested to systemically repeat any measurement resulting in a peripheral radioactive hotspot until this last is being targeted more directly. This will also have the benefit of increasing measurement statistics by mitigating the mask self-collimation effect.

Angular resolution
An important requirement in gamma imaging is the ability to track several radioactive sources at the same time. The angular resolution, θ, is the most convenient parameter to define the distinction limit between two small and adjacent radioactive sources. Theoretically, it depends on the diameter, d, of the open holes on the coded-aperture mask used and the distance, D, between the outer surface -29 -  of this mask and the Timepix detector, as follows: As illustrated in figure 38, the angular resolution can also be experimentally estimated from the vertical or horizontal axial profile of the hotspot peak, taking into account the associated FWHM (Full With at Half Maximum). Since the iPIX field-of-view (FOV) has a square shape, its side corresponds to the maximum number of available pixels (i.e., 256) on the considered axis and we can write then: The theoretical and experimental angular resolutions for the three iPIX masks are summarized in table 6. According to this table, the experimental values obtained with the three masks agree to within 68% (± 1σ) confidence level of their associated uncertainties with the theoretical ones. They also confirm that the angular resolution does not depend on the incident photon energy.
-31 -   To illustrate the optimal angular resolution of iPIX when compared to the existing gamma imagers [5,21], additional measurements were also carried out using three identical low-activity 241 Am point sources (340 kBq). These sources were placed on a horizontal line at 40 cm distance (i.e., providing 3 × 12 nSv/h dose rates at the measurement point). As summarized in figure 39, the following experimental configurations were considered: a. All the sources were close to each other but the outside sources have remained separated from the central one by ∼4 cm, which is the diameter of their circular support. This represents around ±6 • separation angle.
b. The outside sources were separated by ±5 cm (i.e., ±7 • ) from the central source.
c. The outside sources were separated by ±10 cm (i.e., ±14 • ) from the central source.

JINST 11 P08012
Satisfying results were obtained with the masks R7e4 and R7e8 (see figure 39 and figure 40, respectively). However, it is not possible to determine the exact location of each one of the three sources when they are very close to each other (i.e., ±6 • separation angle). In addition, the strong decrease of count rates at non-normal incidences corroborate the mask self-collimation effect, which has already been discussed in the previous section. Due to this effect, the outside sources are less visible at ±14 • separation angle.

Conclusions
We have carried out a comprehensive experimental characterization of the iPIX gamma imager to explore its main features and performances. In view of all the results obtained with ISO narrow X-ray beams at varied acceleration voltages and currents, we have observed that the Timepix shuttertime is the most critical parameter for an effective localization of radioactive hotspots and must be conveniently chosen to avoid signal distortions on the Timepix detector due to pulse pile-up and/or summation effects. Compared to existing gamma imagers, iPIX has an enhanced angular resolution (between 2.5 • and 6.0 • , depending on the coded-aperture mask used) and a high response sensitivity at low energies, tolerating at the same time high levels of background radiation (around 200-310 µSv/h).
Regarding its localization capability, an 241 Am source providing only 2 nSv/h dose rate at the measurement point was clearly spotted within 70 sec of data collection time. In the same way, iPIX was able to localize low-activity 137 Cs and 60 Co sources (providing 100 nSv/h and 1 µSv/h dose rates at the measurement point, respectively) with less than 2.5 min of data collection time.
This study has demonstrated that the mask R7e8 can be applied to almost the whole energy range of interest (10 keV-1.5 MeV). Whereas mask R13e2 is limited to only low photon energies -33 - (e.g. 59 keV γ-rays of 241 Am, whose detection usually indicates the presence of plutonium in nuclear wastes). As an intermediate application, mask R7e4 is able to localize radioactive sources in the medium photon energy range of 137 Cs. Finally, the iPIX dose rate response curve was equally checked with several standard 137 Cs sources of different activities showing a quite linear behaviour over a wide range covering nearly 8 decades (10 −6 -10 1 Sv/h).