2.1.

Detector and Analog Signal Processing in Orbit

The X-ray microcalorimeter detector consists of $6<span\; class="naked\_sign">\times </span><span\; class="naked\_aural">ばつ</span>6=36\mathrm{pixels}$ (pixel 0 to 35) constituting an array with a side length of 5 mm. The array is situated in the center of an Si frame that fans out the electrical connections to the pixels. Portions of the front and back of the frame are coated with Au to assist in the transport of heat, including, on the back, the portions of the frame between pixels.⁸ One of the read-out channels is connected to a pixel (pixel 12) that is displaced from the array and outside of the aperture and is under constant illumination by a dedicated $\mathrm{Fe}_{55}$ calibration source. One corner pixel of the main array is not read out to accommodate this calibration pixel. Each pixel comprises a HgTe X-ray absorber and a suspended Si thermistor. The energy deposited by photoelectric absorption of X-rays is dissipated into heat, which is conducted in a 50-mK heat sink with a thermal time scale of a few ms.¹⁷

Underneath the substrate of the X-ray microcalorimeter, an Si detector is placed with a side length of 10 mm and a thickness of 0.5 mm that is used as an anti-coincidence detector (anti-co).⁸ Cosmic-ray particles (mostly minimum-ionizing protons) interact with the detectors mainly via ionization losses along their track lengths and are detected both by the anti-co (as charge) and the microcalorimeter (as heat). The anti-co spectrum peaks at 200 keV, whereas the spectrum of calorimeter events coincident with anti-co events peaks at 6 keV, with these peaks corresponding to the energy deposited by a relativistic proton at normal incidence to the anti-co (0.5-mm-thick Si)¹⁸ and a calorimeter absorber (0.0085-mm-thick HgTe), respectively. The pulse characteristics of cosmic-ray and X-ray events in the calorimeter are identical, which is why the anti-co is essential for the rejection of particle events. The anti-co detector is read out by two readout circuits for redundancy.

Each of the 36 microcalorimeter pixels and two anti-co channels is read out individually using a pair of signal and return lines without multiplexing. The output impedance of each detector circuit is reduced by a junction field-effect transistor (JFET) source-follower operated at 130 K¹⁷ before being transferred from inside to outside of the cryostat via a feed-through. The analog signal is received by the analog electronics called X-ray box (XBOX)¹⁷ placed outside of the cryostat. The XBOX provides the bias voltages to the microcalorimeter and anti-co detectors and processes their output by shaping, amplifying, filtering, and sampling at a rate of $80\mu \mathrm{s}$ (12.5 kHz) with a bipolar 14-bit depth analog-to-digital converter (ADC). The signal-chain hardware up to this part is provided by NASA.

2.2.

Digital Signal Processing in Orbit

The rate of the entire time-series data ($12.5\mathrm{kHz}<span\; class="naked\_sign">\times </span><span\; class="naked\_aural">ばつ</span>16\mathrm{bit}<span\; class="naked\_sign">\times </span><span\; class="naked\_aural">ばつ</span>38$ channels) is too large to fit into the telemetry bandpass to ground stations. Therefore, a significant data reduction needs to be made in the orbit. Events are detected, and their time-series data (we call "pulse" hereafter) are processed. A handful of characteristic values are derived for each event, which are downlinked at five $\sim 10\text{-}\min$ contacts per day.¹⁹ When resources are available in the data recorder and the telemetry bandpass, the pulse data of randomly selected events are downlinked for diagnostic purposes, which are called pulse records. This data reduction is performed using the onboard digital electronics called the pulse shape processor (PSP) provided by JAXA.²⁰

The digitized time-series data are sent from the XBOX to the PSP, which consists of two identical units (PSP-A and PSP-B). Each unit hosts one field programmable gate array (FPGA) board and two central processing unit (CPU) boards. One FPGA board processes 18 microcalorimeters and 1 anti-co channel. One CPU board processes nine microcalorimeters and 1 or 0 anti-co channels by default. The two anti-co channels are readouts of the same set of events. We used data from the A side in the presented analysis. All cross-channel processing, such as vetoes, is done on the ground.

The FPGA board is responsible for calculating the time derivative of the incoming microcalorimeter data (ADC samples) from the XBOX using a box car filter. It triggers event candidates based on a threshold applied to the time derivative. The threshold was changed once in orbit (at 19:50 on October 31, 2023) from 75 to 120 as a precaution before the first attempt to open the GV and has been left at that value. The FPGA board also detects events in the anti-co detector data by applying a threshold in ADC samples and derives characteristic values of individual pulses: (1) the arrival time when the threshold is exceeded, (2) the maximum ADC sample value [pulse height amplitude (PHA)], and (3) the number of consecutive intervals between samples exceeding the threshold (DURATION). We recognize events with a minimum PHA of 71 (equivalent to an energy deposition of 30.5 keV at PSP A side) and a minimum DURATION of 2 to be cosmic-ray events.

The CPU board is responsible for detecting secondary pulses overlapping others and deriving characteristic values of each pulse in the microcalorimeter data. One of the values is the event grade, which is based on how an event is isolated in time from the others on the same channel. Events with no preceding pulse in a fixed time are called p(rimary) and the others s(econdary). Events without others within $\pm 69.92\mathrm{ms}$ are H(igh) grade, those without others within $\pm 17.52\mathrm{ms}$ are M(id) grade, and the rest are L(ow) grade. For the H and M grade events, the time series data of a pulse are correlated with templates for better characterization of the energy and arrival time. The template is made for each pixel from the average pulse shape of numerous events from 5.9 keV X-ray photons and noise spectra obtained in the absence of pulses. After the correlation at several shifted times (TICK_SHIFT), characteristic values such as the arrival time and pulse height (PHA; the amplitude of correlation) are derived by interpolation. Here, the TICK_SHIFT is a value that represents the extent to which the template pulse is shifted in time relative to an incoming pulse from the initial value. It is calculated by the PSP and takes integer values ranging from $-8$ to $+7$ in the unit of a sample ($80\mu \mathrm{s}$).^{20 ,21} For L grade events, template fitting is not done, and their pulses are characterized similarly to anti-co events based on thresholding. Because of this difference, a systematic offset of arrival time arises among different grades. This is calibrated as a function of incoming X-ray energy for each channel²² and corrected on the ground to $\lesssim 5\mu \mathrm{s}$.

2.3.

Data Processing on the Ground

The ground processing takes several stages from levels 0 to 3 (L0 to L3). The L0 products are raw packet telemetry files in the flexible image transport system (FITS) format. The L1 products are generated by the pre-pipeline developed by JAXA,²³ in which raw telemetry values are converted to engineering values in the First FITS Files. The relative arrival time correction is made at this stage. The L2 products are generated by the pipeline (PL) developed by NASA,²⁴ in which the engineering values in the L1 products are converted into physical values based on the calibration database. The L3 products are for quick-look assessment for scientific purposes such as X-ray images, spectra, and light curves.

Event screening is performed in the PL using the xapipeline task for the satellite-wide processing and the rslpipeline task for the Resolve-specific processing. The following processing items are performed.

3.1.

Screening Based on Pulse Shape

The PSP derives several values and flags to characterize individual pulses (Fig. 1). DERIV_MAX is the maximum value of the time derivative. RISE_TIME is defined as the time from the maximum time derivative to the zero-crossing time interpolated for sub-sample resolution. Along with TICK_SHIFT, PHA, and arrival times (Sec. 2.2), they are characteristic values of pulses. The maximum and minimum values of TICK_SHIFT are used to indicate that the template fitting was not successful. Events with such values are screened. Other than this, the event properties are not used for the initial screening.

Flags are also given to indicate anomalies in the pulse shape (Fig. 2). The QUICK_DOUBLE flag indicates an increase of the time derivative during the decay phase before the trigger threshold is crossed again, which corresponds to the rising edge of the original pulse. SLOW_PULSE and SLOPE_DIFFER flags are set for abnormal rise or slow time scales of primary and secondary pulses, respectively.²⁰ Events with any of these flags are removed, except for events flagged with SLOPE_DIFFER that have $\mathrm{PI}>22000$. Many normal X-ray events at high energy ($E>11\mathrm{keV}$) receive the SLOPE_DIFFER flag, and these events should not be removed. We examined pulse records of many events with QUICK_DOUBLE, SLOW_PULSE, and SLOPE_DIFFER flags and found that they indeed exhibit abnormal pulse shapes.

The FLAG_CLIPPED flag is set for events with the possibility of hitting the maximum of the ADC. Such a flag could be given during the onboard processing when the entire pulse shape is available, but this function was not implemented. In Resolve, the rslplsclip task was developed to flag events of possible ADC saturation based on the pulse height.²⁵ We examined the algorithm using the ground data set taken from February 14, 2022, at 14:05 to February 16, 2022, at 08:40 using a rotating target source, in which targets of an X-ray generator rotate to provide multiple lines over a wide range of energy in the integrated spectra in all event grades.¹² Figure 3 shows the result, in which most FLAG_CLIPPED events appear at energies above 20 keV. Note that the maximum energy before clipping depends on pixel and detector heatsink temperature. Such events are found to degrade the energy resolution at 22 keV. The reason for this change (for high and mid-grade events) is that the shape of the gain scale is significantly affected by the pulse clipping and is not accounted for using the standard gain correction techniques. This is not, however, included in the initial screening. Many clipped events are background, from cosmic rays with long tracks in the absorbers, but some may be signals; hence, they should be included with appropriate screening and redistribution matrix functions in the future.

3.2.

Screening Based on Relative Timing of Multiple Events

3.3.

Screening Based on Time Intervals

Fig. 8

5.1.

Screening Criteria

Although the initial screening in the current PL processing is sufficient to satisfy the requirement (Sec. 4), additional screening, such as applied to the ASTRO-H SXS data, but optimized for Resolve, is expected to improve the signal-to-noise ratio and extend the Resolve capability for specific sources. We exploit the fact that the characteristic values, not flags, of events (i.e., DERIV_MAX, RISE_TIME, and TICK_SHIFT) are tightly correlated with each other for X-ray events but not for background. For example, the higher the energy deposited, the greater the drop in the resistance of the thermistor, and the resulting change in the electrical component of the signal response time leads to a correlation between DERIV_MAX and RISE_TIME. Events deviating from this correlation should be considered as background events. Here, we used DERIV_MAX as opposed to another proxy of energy used in the previous work⁷ as the correlation is more linear between DERIV_MAX and RISE_TIME. Other criteria such as frame (Sec. 3.2.4) and cross-talk (Secs. 3.2.2 and 3.2.3) events are expected to improve screening further, but they are not presented in this work as they are currently under study.

Figure 9 shows the two-dimensional distribution of Hp and Mp events using the in-orbit data of LMC X-3 observed in late November and early December 2023. The gray dots indicate events after the initial screening. The correlations among the X-ray events are evident as the main single sequence. The distributions deviating from the main sequence are mostly made with background events, which is confirmed by inspecting the pulse shapes when the pulse record data are available. An exception is the distribution deviating upward from the main sequence at the largest DERIV_MAX end, which is made by X-ray events that are clipped. These events may be signals and hence should not be removed. Only for the purpose of clarifying the main sequence, we applied the frame-event screening (Sec. 3.2.4), which efficiently eliminates many background events though with false coincidence. Based on the frame-screened distribution of events (black dots), we defined the screening criteria shown in blue lines. Note that the presented screening applies to the data taken with the event threshold of 120. TICK_SHIFT depends on the trigger threshold; thus, different criteria for TICK_SHIFT were developed for the data taken with the event threshold of 75.

5.2.

Evaluation

We now evaluate the performance of the additional screening. The result of screening depends on the count rate of the celestial object; the background may be reduced further but at a sacrifice of data due to false coincidence. There may be some pixel dependence. For the assessment, therefore, we used the data sets of different origins (Table 3), which include the ground data with the GVO and the in-orbit data of Abell 2319 and LMC X-3 observations and NXB data.

Abell 2319 is a galaxy cluster and is a diffuse source with a uniform distribution of events over the microcalorimeter array.³⁴ LMC X-3 is a blackhole binary in the Large Magellanic Cloud³⁵ and a point source making a concentrated event distribution at the array center. We conducted spectral fitting of LMC X-3, the brightest of the two, and found that the X-ray signals below 1.7 keV are negligible; thus, we use this energy as the boundary. The GVO data set was taken when the GV was open on the ground and provides data below $\sim 1.7\mathrm{keV}$, a region probed only by background events and spectral redistribution when the GV is closed.

We apply the additional screening and see how the background and signal events are affected. For signal events, we focus on the Hp and Mp grade events used for spectroscopy. Signal and background events dominate different energy bands for these data sets. Signal events dominate the whole band (0.1 to 20 keV) for the GVO data and in the science band (1.7 to 12 keV) for the two observations in the closed GV configuration. Background events dominate the whole band for the NXB data and in the soft band $<1.7\mathrm{keV}$ for the two observations.

Table 4 shows the ratio of the event rate after screening to that before in the energy band where signal or noise events dominate. The reduced rate in the signal-dominant band is due either to the false coincidence or contamination of background events in the band. More events are removed by screening for the lowest count rate data set of Abell 2319 than for the other two sets. This is likely because background events are non-negligible in the signal-dominant band of the lowest count rate data set. Based on these data, we conclude that the percentage of real source data removed by the additional screening is less than 6%. For Abell 2319 and the NXB, the additional screening reduced the rate by $\sim 95\%$ below 1.7 keV. The reduction is lower for the highest count rate data set of LMC X-3. This is likely due to spectral redistribution from the signal dominant band at $>1.7\mathrm{keV}$ to the background dominant band at $<1.7\mathrm{keV}$.²⁹ These redistributed events, resulting from phenomena such as the removal of energy by escaping photons or electrons after absorption of an X-ray, represent signal information and hence should not be removed as background but should be modeled through the redistribution matrix.

Table 4

Ratio of the event rate after the additional screening to that before separately for the signal or noise-dominant energy bands.

The resultant NXB spectrum after the additional screening is shown in Fig. 8. The count rate is $1.0<span\; class="naked\_sign">\times </span><span\; class="naked\_aural">ばつ</span>{10}^{-3}\mathrm{counts}{\mathrm{s}}^{-1}{\mathrm{keV}}^{-1}$ in the 0.3 to 12 keV band, which is smaller than the requirement by a factor of 2. The additional screening is particularly useful below $\lesssim 2\mathrm{keV}$, which will be accessible after the GVOs are in orbit.

References