3.1.

Compton Camera

Taking advantage of 15 years of development,^{11–13,31,36–42} the SGD Compton camera is optimized to detect gamma rays up to 600 keV. A Compton camera consists of a stack of 32 layers of 0.6-mm-thick Si pixel sensors⁴³ followed by 8 layers of 0.75-mm-thick CdTe pixel sensors,^44,45 surrounded by 2 layers of 0.75-mm-thick CdTe pixel sensors on the four sides as shown in Fig. 2(b). Each Si sensor has a $16\times 16$ array of $3.2\times 3.2{\mathrm{mm}}^2$ pixels, and each CdTe sensor has an $8\times 8$ array of $3.2\times 3.2{\mathrm{mm}}^2$ pixels. Each layer of CdTe sensors in the stack is composed of $2\times 2$ CdTe sensors to match the area of Si sensors, and each layer of CdTe sensor on the side is composed of $2\times 3$ CdTe sensors. The pixel size is optimized to minimize the number of pixels for lower power consumption while avoiding the pixel size being the dominant contribution to the angular resolution of Compton kinematics. The thicknesses of Si and CdTe sensors are determined from constraints on the bias voltages required to operate the sensors at the best condition. The location of the CdTe sensors on the side is slightly displaced in the horizontal direction to allow placement of readout application-specific integrated circuits (ASICs) at the corner of the sensor. This arrangement allows a placement of the CdTe sensors on the side very close to the stacked Si and CdTe sensors. The average distance between stacked sensor layers is 1.8 mm, realizing a very compact scattering volume. The length and placement of the CdTe sensor on the side and the distance from the stacked sensors are determined to maximize the coverage of gamma rays scattered in the Si sensors with the minimum area of the CdTe sensors on the side.

Signals from $8\times 8\text{pixels}$ of Si and CdTe sensors are read out by a 64-channel ASIC, VATA-SGD,⁴⁶ which can produce trigger signals, digitize pulse heights, and can be configured digitally. The signal processing chain of the Compton camera consists of 208 front-end cards (FECs) followed by four ASIC driver boards (ADBs) and an ASIC control board (ACB) as shown in Fig. 4(a). FECs are located at the four corners of the sensor modules in the sensor stack, and the backside of the CdTe sensor on the side, and have only ASICs (one ASIC for the stack module and six ASICs for the side module) and passive components, such as capacitors and miniature connectors. Five daisy chains of eight ASICs in the sensor stack are connected to one ADB, which is located on each side of the Compton camera. Two daisy chains of six ASICs in the side sensors are connected to the same ADB. Each ADB transmits digital signals between seven daisy chains of ASICs and the ACB. Only power lines and digital lines are required between ADBs and FECs, and all fast digital signals are differential to minimize the electromagnetic interference. ASICs are controlled by a field programmable gate array (FPGA) on the ACB attached at the bottom of the Compton camera. These components are packed in a $12\times 12\times 12{\mathrm{cm}}^3$ aluminum enclosure. Such a high-density stacking of Si and CdTe sensors is critical to obtain a high efficiency for Compton reconstruction of 10% to 15% around 100 keV. A detailed design of the SGD Compton camera is described in Ref. 41.

To suppress the leakage current of the Si, CdTe, and APD sensors and to avoid the gain variation of the APD, the temperature of the bottom structure attached to the Compton cameras and the BGO support structure needs to be kept at $-20°\mathrm{C}\pm 5°\mathrm{C}$ and below $-15°\mathrm{C}\pm 5°\mathrm{C}$, respectively, with an orbital variation of $<3°\mathrm{C}$. The Compton camera is designed to conduct the heat of 4 W from ASICs and to keep the temperature of all sensors within 5°C of the bottom structure interface. A detailed description of thermal design of the Compton camera is given in Ref. 47.

A custom ASIC, VATA-SGD, is developed for the SGD based on the VIKING architecture,^46,48 which has been known for good noise performance and employed in various satellites, such as Swift,⁴⁹ PAMELA,⁵⁰ and AGILE.⁵¹ In the SGD Compton camera, very little space is left for the electronics; therefore, the ASIC must integrate trigger and digitization capabilities in a single die. The VATA-SGD consists of 64 channels of CSAs with low noise [$180e^{-}$ (RMS) at 6 pF load] and low-power consumption ($\lesssim 0.3\mathrm{mW}/\text{channel}$). Following each CSA, a trigger circuit is formed by a fast shaper with a peaking time of $0.6\mu \mathrm{s}$ and a discriminator. The CSA output is also connected to a slow shaper with a peaking time of $\sim 3\mu \mathrm{s}$ followed by a hold circuit so that the peak voltage can be held using the above trigger signal. The hold circuit outputs of all 64 channels are simultaneously digitized by the Wilkinson-type analog-to-digital converter where the time duration of the voltage ramp to cross the sampled voltage is measured by a counter. The conversion time is $<100\mu \mathrm{s}$ using the external clock of 10 MHz for 10-bit digitization. The digitized data are transferred via a differential serial output. To reduce the data transfer time, the ASIC can send only the data above a certain digital value which can be adjusted for each channel. The ASIC produces all necessary analog bias currents and voltages by internal digital-to-analog converters except for the main bias current which sets the scale of all bias currents; this is provided by an external resistor on the FEC. Owing to all the functionality integrated in the ASIC, we only need an FPGA, several digital drivers and receivers, and passive components (resistors and capacitors) to operate 208 ASICs in a Compton camera. The data input and output circuits on the ASIC are designed to allow daisy chaining of multiple ASICs to reduce the number of signal paths. In the SGD Compton camera, six or eight ASICs are daisy chained.

3.2.

BGO Active Shield and Support Structure

The thick active shield made of 25 BGO scintillators coupled with APDs is employed to reduce in-orbit background of the SGD.^52–55 The BGO crystal is dense ($7.1{\mathrm{g}/\mathrm{cm}}^3$), has a high stopping power (i.e., short radiation length of 1.1 cm), high optical transparency, and ability to form a large crystal although its light output is lower than NaI or CsI.

In addition to providing anticoincidence signals for cosmic rays and gamma rays from outside of the FOV, the BGO shield is also used to reduce the number of SAA protons penetrating to the cameras since those protons are the main cause of the activation of sensor materials. The BGO shape is designed so that any trajectory that intersects with the Compton camera must go through at least 3 cm (2.7 radiation length) of BGO before it reaches the camera. We employ a modular mechanical structure for the BGO shield where each BGO crystal is supported by a CFRP base plate to make it easier to handle each BGO module. The BGO enclosure consists of a CFRP base plate that is glued to the BGO crystal via ${\mathrm{BaSO}}_4$-based reflector painted on the BGO and CFRP covers. The ${\mathrm{BaSO}}_4$-based reflector is chosen for the adhesive strength with modest reflectivity. The remaining sides of the BGO crystal are covered by both enhanced specular reflector and Gore-Tex sheet for better reflection properties.

The APD is chosen for the photon detector of the BGO shield mainly due to its compact size compared with photomultipliers and also due to compatibility with a modular structure of the shield. Although the larger APD yields better photon collection efficiency (nearly proportional to the root square of the area), the capacitance and the leakage current of the APD also increases proportionally to the area. Based on the experiments with $3\times 3$, $5\times 5$, $10\times 10$, and $20\times 20{\mathrm{mm}}^2$ APDs, we concluded that 10-mm size gives the best signal-to-noise ratio for our application. We employ HPK S8664-55 with slightly modified structure for less leakage current; those were used by the Compact Muon Solenoid, an experiment for the Large Hadron Collider at the European Organization for Nuclear Research. The APD is encapsulated by silicone resin rather than epoxy resin, to avoid cracks that appeared in epoxy resin due to thermal cycles.

Since the gain of the APD is temperature dependent ($-3\%/°\mathrm{C}$), the temperature of the APD needs to be controlled within 3°C to keep the gain variation within 10%. If the temperature variation cannot be controlled within 3°C, the APD bias voltage needs to be adjusted to compensate for the gain change.

Signals from the APDs are routed to CSAs in shielded boxes located in the close vicinity of the APDs on the SGD-S enclosure. Since the APD capacitance is relatively large, $\sim 270\mathrm{pF}$, a low-noise CSA is developed to obtain $\sim 1300$ electrons (FWHM).

3.3.

Fine Collimator

The BGO active shield has an opening of $9.2\times 9.2{\mathrm{deg}}^2$ (FWHM), which is rather large, and thus would result in the CXB higher than non-x-ray backgrounds and substantial source confusions within the FOV below $\sim 150\mathrm{keV}$. FCs are installed in the opening of the BGO active shield, to reduce the FOV to 0.55 deg (FWHM). The collimator consists of $16\times 16$ array of square cells with a pitch of 3.2 mm and a length of 324 mm. Each cell is divided by 0.1-mm-thick phosphor bronze (PCuSn), yielding an aperture opening of $\sim 94\%$. These choices of the material and the geometry yield negligible leakage for 100 keV and 10% leakage at 150 keV and 0.55 deg from the FOV center as shown in Fig. 3. To achieve the effective area $>20{\mathrm{cm}}^2$ at 100 keV, the loss of the effective area due to the FC aperture opening and misalignment must be $<20\%$.⁵⁶ The effective FC aperture opening, including the effect of the imperfect shape of the FCs, is expected to be better than 90%. Misalignment of the FCs may be caused by misalignment among six collimators, thermal distortion of the SGD-S structure, and misalignment of the SGD-S with respect to the satellite optical axis. The total misalignment should be below 3.3 arc min to keep the total loss below 20%.

Fig. 3

Analytical calculation of the transparency of the fine collimator as a function of the angle from the FOV center.

3.4.

Electronics

The SGD electronics system consists of two processing chains, one for Compton cameras and the other for APDs as shown in an SGD electronics block diagram in Fig. 4(b). The Compton camera electronics system includes three Compton cameras, two camera HV power supplies, two DC/DC converters, a camera power management unit (CPMU), and a mission I/O (MIO) board. The CPMU controls power supplies (both HV power supplies and DC/DC converters) and HV levels, and monitors power supply voltages and currents, and temperatures around cameras. HV power supplies are controlled via slow serial data link and have low-pass filters to avoid a sudden change of the HV output. The MIO board sends control signals to the camera and receives the data from the camera via low-voltage differential signaling (LVDS). The MIO board also formats the camera data, controls the CPMU, and communicates with the SGD-DE via SpaceWire network interface.

Fig. 4

Block diagram of (a) the Compton camera electronics systems and (b) the SGD electronics system.

The APD electronics system includes 25 APDs, 25 APD-CSAs, four APD HV power supplies, two APD processing and management units (APMUs), and an MIO board. The APMU receives APD signals from APD-CSAs and continuously digitizes them with flash ADCs. The FPGA on the APMU immediately applies a digital filter on the digitized APD signal and issues anticoincidence signals with three threshold levels: the lowest possible threshold to detect any particle that deposits energy in the BGO, a threshold to detect minimum ionizing particles, and a threshold to detect iron nuclei. We have two digital filters for the lowest possible threshold to provide a fast anticoincidence to abort the data acquisition in progress and a slow anticoincidence for offline analyses with a better energy resolution. Other APMU functions include creating pulse height histograms, detection of transient events in BGOs, control of power supplies and their voltages, and monitoring of power supply voltages and temperature around APDs. The MIO board sends control signals to and receives digital data from the APMUs via LVDS.

3.5.

Data Acquisition System

The data acquisition of the Compton camera in the SGD is initiated by the primitive trigger issued from any ASIC. The primitive triggers from all ASICs in a Compton camera are processed by the ACB in the Compton camera. The ACB checks the coincidence pattern of those primitive triggers for any sign of charged-particle tracks that tend to produce triggers in many consecutive layers. The coincidence time is $0.4\mu \mathrm{s}$. If ACB does not find any sign of the charged-particle tracks, it sends the trigger signal to the MIO board. The MIO board sends back trigger acknowledgment to the ACB if no other Compton camera is in the data acquisition phase. Since the data acquisition activities in one Compton camera affect the trigger performance of the other Compton cameras in the same SGD-S, we do not accept new triggers during the data acquisition. Upon reception of the trigger acknowledgment from the MIO board, the ACB sends hold signal to all ASICs in the Compton camera with appropriate delay ($\sim 3\mu \mathrm{s}$) so that the slow shaper signal of the ASIC can be sampled near the peak of the pulse shape and activate the digitization mode. While we wait for the ASIC to stabilize after the digitization mode is activated, the MIO board checks the output of the fast anticoincidence signals from the BGO system. If any anticoincidence signal is issued within $6\mu \mathrm{s}$ of the activation of the digitization mode, the data acquisition is aborted before the actual digitization is started to minimize the dead time. Unless the abort signal is issued by the MIO board within $6\mu \mathrm{s}$ from the trigger acknowledgment, the ACB activates the digitization sequence of all ASICs in the Compton camera. This BGO anticoincidence functionality was not activated during the verification phase of the satellite operation. In addition, slow anticoincidence signals from the BGO system, with lower energy threshold for better performance, were also recorded for offline analyses. A detailed description of the Compton camera data acquisition is given in Ref. 41.

The BGO data acquisition operates independently of the Compton camera. The FPGA on the APMU continuously process the BGO signal and sends anticoincidence signals to the MIO board whenever it detects signals above thresholds. It also produces histograms of the peak energies at a certain interval. Those histograms are used for gain calibrations of each BGO module. Gamma-ray bursts may be detected if it finds a sudden increase of the anticoincidence rates. A detailed description of the BGO signal processing is given in Ref. 55.

The SGD data mainly consist of the event and HK data. The event data include the absolute time, the time after previous trigger, raw ASIC data, and various flags, including anticoincidence flags from BGOs. The HK data include time, temperatures, power supply status, and measured voltages of the power supplies.

6.1.

BGO Shield Performance

All of the 50 BGO shield modules in the SGD1-S and SGD2-S worked stably in orbit. Figure 7(a) shows spectra for a BGO module with different observing orbits divided by SAA passages. The 511-keV annihilation line and other gamma-ray lines from radioactivation of detector materials apparent in this figure can be used to derive the energy scale and the threshold energy of the BGO signals. The stability of the energy scale is also confirmed to be within 10% by monitoring the peak position of the 511-keV line. Figure 7(b) compares the distributions of the lowest threshold energies with the slow filter of all BGO shield modules obtained on ground (black) and in orbit (red). Note that the on-ground performance is slightly different from that in Ref. 55 since it derived the results from different on-ground tests. The in-orbit data show slightly better thresholds presumably due to the lower operating temperature, which indicate the anticoincidence performance of the BGO shield is not worse than that verified on ground.

Fig. 7

(a) Spectra for a BGO module with different observing orbits divided by SAA passages. (b) Comparisons of the distributions for the BGO threshold energies obtained on ground (black dashed) and in orbit (red solid). (c) BGO count rates as a function of time for the average rate (black curve), and the BGO modules with highest rate (red curve) and lowest rate (blue curve). The data were taken from 18:00 UTC on March 23 till 18:00 on March 24.

Figure 7(c) shows the BGO count rates for the lowest threshold with the slow filter as a function of time in one day for the average rate (black curve), and the BGO modules with highest rate (red curve) and lowest rate (blue curve). Background variations due to the varying cutoff rigidity (COR) of cosmic rays and the SAA passages can be seen. The BGO shields operated properly even during SAA passages and the sharp increases of the count rates were as expected. Outside of the SAA passages, gradual modulations of the count rate are found to be well anticorrelated with the COR values. The average count rates vary by module and ranges from 0.5 to 1.8 kHz per module, which is consistent with our expectation of 0.5 to 2.1 kHz based on the Suzaku HXD BGO data.

The SGD BGO system is very massive and good at detecting gamma rays and measuring spectra from transient events, such as gamma-ray bursts and terrestrial gamma-ray flashes. There was no autodetection of transient events during the SGD operation period since its search parameters had not been adjusted on orbit. However, we found a candidate for a gamma-ray burst (GRB) by manually searching for transient events. This GRB was found to be GRB 160324A. The top panel of Fig. 8(a) shows the count rate of some SGD BGOs as a function of the time since 15:58:34 UT (this is the trigger timing by INTEGRAL SPI-ACS), where clear increase of the count rate is observed. The middle and bottom panels of Fig. 8(a) also show the count rates of INTEGRAL SPI-ACS and Konus-Wind. This event could have been autodetected if the proper search parameters had been in place. Figure 8(b) shows the energy spectrum observed by the SGD BGOs in the 30-s period indicated in Fig. 8(a). The background spectra are estimated by a linear interpolation of the 30-s period before and after the burst and subtracted from the observed spectra. A fit to a power-law function yields a power-law index of $-{3.03}_{-0.31}^{+0.28}$ and a fluence of $1.38\times {10}^{-5}\mathrm{erg}/\mathrm{cm}$ in the 100- to 1000-keV band. Here, the effective area is obtained from the MC simulation that was verified by the calibration tests with various source directions on ground. These results for spectral parameters are within the expected ranges found in typical gamma-ray bursts.

Fig. 8

(a) Count rate as a function of the time for some SGD BGOs (top), INTEGRAL SPI-ACS (middle), and Konus-Wind (bottom). Time for Konus-Wind was adjusted by 0.6915 s with respect to that for SPI-ACS to account for the arrival time difference due to the different satellite locations. A clear increase of the count rate is observed in the SGD BGOs. The hatched region in the top panel indicates the time period used for following spectral analysis. (b) Energy spectrum observed by three SGD BGOs with the most count increases. Three different colors indicate the spectra of three BGOs.

6.2.

Fine Collimator Alignments

The alignment of each FC was adjusted after they were installed. As a result, the viewing axis of the FC was aligned within 1.5 arc min of the nominal viewing axis of the SGD-S (an average misalignment was 0.8 arc min). This is much smaller than the FC alignment requirement of 3.3 arc min to accommodate remaining misalignment factors, such as thermal distortion of the SGD-S structure and misalignment of the SGD-S with respect to the satellite optical axis.

We checked the count rate of each Compton camera during the Crab observation since misalignment among the FCs would appear as a dispersion of the count rates among Compton cameras. The thickness dispersion of the active and inactive layers of Si sensors would affect the count rates; however, those effects are expected to be much smaller than the FC misalignment effect. Figure 9(a) shows the spectra for all pixels in the top four layers of Si sensors without anticoincidence signals in the active shield where most hits are due to photoabsorption of the low-energy gamma rays. (Hereafter, we always require the absence of any anticoincidence signals in the active shield for the analysis of the Compton camera data.) Different colors show spectra for different Compton cameras. Since we have not adjusted the thresholds of each channel, the spectra below 30 keV are not consistent. The count rates in the energy range, 35 to 60 keV, where the effects of threshold dispersions and background subtraction are negligible, are listed in Table 1. The count rates agree with each other within 3% among the six Compton cameras. This corresponds to $<1.0\mathrm{arc}\min$ of misalignments among six FCs, which is consistent with the average misalignment of 0.8 arc min.

Fig. 9

(a) Spectra for all pixels in the top four layers of Si sensors where most hits are due to photoabsorption of the low-energy gamma rays. Different colors show spectra for different Compton cameras. (b) Observed spectrum for all pixels in the top four layers of the Si sensors of the Compton camera 1 in the SGD1 (red), the background spectrum taken one day before the Crab observation (green), the observed spectrum with the background subtraction (blue), and the simulation spectrum (black). The simulation assumes a power-law spectrum with a power-law index of $-2.11$.⁵⁷ The lower panel shows the residual [(observed counts) – (background) − (simulation)]/(simulation), as a function of the energy.

Table 1

Count rates of top four layer of Si sensors for six Compton cameras during the Crab observation.

To confirm the alignment with the optical axis of the satellite, we compared the observed rate for the Crab Nebula with that from the Monte Carlo simulation assuming the nominal Crab flux. Figure 9(b) shows the observed spectrum of the Compton camera 1 in the SGD1 (red), the background spectrum taken one day before the Crab observation (green), the observed spectrum with the background subtraction (blue), and the simulation spectrum (black). In the simulation of the Crab Nebula observation, we assumed a power-law spectrum, $N·E^{-\mathrm{\Gamma }}$, with $\mathrm{\Gamma }=2.11$.⁵⁷ By scaling the integrated rate of the simulated spectrum in the 25- to 60-keV range to match the observed rate, we obtain $N=8.97\pm 0.21$ (statistical error only), which corresponds to a flux of $(8.28\pm 0.20)\times {10}^{-9}\mathrm{erg}/{\mathrm{cm}}^2/{\mathrm{s}}^2$ in the 25- to 60-keV range. Here, the energy range is extended to 25 keV since the simulation accounts for the threshold effects. A recent measurement of the absolute flux for the Crab Nebula by NuSTAR (observations were done in October 2015 and April 2016) gives $\mathrm{\Gamma }=2.106\pm 0.006$, $N=9.71\pm 0.16$,⁵⁷ which corresponds to a flux of $(9.25\pm 0.05)\times {10}^{-9}\mathrm{erg}/{\mathrm{cm}}^2/\mathrm{s}$ in the 25- to 60-keV range (a correlation between $N$ and $\mathrm{\Gamma }$ is taken into account). The Crab flux observed by the SGD is $\sim 10\%$ lower than that by NuSTAR, which is not inconsistent with the expected losses of 7% for the effective area due to known misalignment and distortion of the FCs, considering that we still have unknown misalignment effects, such as the thermal distortion of the SGD-S structure and misalignment of the SGD-S with respect to the satellite optical axis. (The satellite operation was terminated before we had opportunities to measure those remaining misalignment effects.)

6.3.

Noise Performance of Compton Camera

Noise performance and energy resolution of the Si and CdTe sensors in the Compton camera are critical to reject backgrounds from nuclear emission lines and to reject continuum backgrounds by Compton kinematics with high efficiencies. The energy resolution of the sensors was verified by the width of nuclear emission lines observed by the Compton cameras before and after the launch.

Figure 10 shows single-hit spectra of CdTe sensors in a Compton camera and a spectrum for Compton events⁴² obtained during the thermal vacuum test of the satellite and corresponding spectra in orbit. Compton events are double-hit events, in which the energy values of two hits are consistent with a Compton scattering followed by a photoabsorption. (Consistency between the inferred direction of the incident photon and the FOV is not required at this point.) Nuclear emission lines from naturally occurring radioisotopes are clearly detected in the spectra taken on ground. In-orbit, nuclear emission lines from activated materials are observed. For example, the lines around 192, 247, 336, 392, and 396 keV are considered to originate from activated cadmium. The 396-keV line is more prominent in Compton events since it is actually a sum of 151- and 245-keV lines from ${}^{111}\mathrm{Cd}$. The Compton kinematics constraints reduce this line; however, some of them still remain. The energy resolutions obtained from the fit differ between the CdTe sensors in the stack and in the side. The spectra for the CdTe sensors in the side suffer more backgrounds from the surrounding material. Because of this, the peak fit may include more tail components for the CdTe sensors in the side.

Fig. 10

(a) Single-hit spectra of CdTe sensors in a Compton camera taken on ground. The spectrum of all channels in 48 CdTe sensors on the side and that of all channels in 32 CdTe sensors in the stack are shown in black and red, respectively. (b) Single-hit spectra of CdTe sensors on the side (black) and in the stack (red) taken in orbit. (c) Spectrum for Compton events in a Compton camera taken on ground. (d) Spectrum for Compton events in a Compton camera taken in orbit.

Figure 11 summarizes the energy resolution of single hits as a function of the line energies for Si and CdTe sensors. Black and red points indicate the energy resolutions for Si and CdTe sensors on ground, respectively. Blue points show the energy resolutions for CdTe sensors in orbit. Two points for each energy show the energy resolution in the stack (better resolution) and in the side. The energy resolution in orbit is slightly better than the performance observed on ground due to lower operating temperature in orbit, and no degradation was observed from the performance obtained in the component level tests.^23,24,41

Fig. 11

Energy resolution (FWHM) of single hits as a function of the line energies for Si and CdTe sensors. Black (circle) and red (square) points indicate the energy resolutions for Si and CdTe sensors on ground, respectively. Black and red lines show fits to a linear function for black and red points, respectively. Blue (triangle) points show the energy resolutions for CdTe sensors in orbit.

6.4.

Compton Reconstruction Performance

Rejection of backgrounds coming from outside of the FOV is the most crucial capability for the SGD. We verified the background rejection capability using the data observed before the maneuver to the Crab Nebula.

Figure 12(a) shows the light curve of Compton events in a Compton camera of the SGD1. The energy range is from 50 to 600 keV, and no event selection based on the Compton-scattering angle is applied. The gaps in the light curve correspond to SAA passages (where the data acquisition is disabled). The background count rate increases soon after the SAA passages and decreases after the SAA. The background variation is produced by radiations from activated material due to SAA protons.

Fig. 12

(a) Light curve of Compton events in 50 to 600 keV from a Compton camera of the SGD1. (b) The background spectra obtained on the above Compton camera. The red histogram shows the background spectrum from orbits after SAA passages, corresponding to the red periods in (a), and the green histogram shows the spectrum from orbits without SAA passages, corresponding to the green periods in (a). The black histogram shows the on-ground background spectrum obtained in the thermal vacuum test of the Hitomi satellite.

Figure 12(b) shows the background spectra of Compton events in the same Compton camera after the event selection on the Compton-scattering angle. The background spectrum from orbits right after the SAA passages [corresponding to the red hatched period in Fig. 12(a)] is shown in red, and the spectrum from orbits without SAA passages [corresponding to the green hatched period in Fig. 12(a)] is shown in green. For comparisons, the background spectrum on ground is shown in black. The difference between the red and green spectra is due to radiations from short-lived isotopes activated by the SAA, including the 511-keV line and other nuclear emission lines. The difference between the green and black spectra is due to radiation from long-lived isotopes activated by the SAA, and those would accumulate over time until an equilibrium is reached of isotope production and decay.

Figure 13(a) compares spectra for the background in orbit before (black) and after (blue) the event selection based on the constraint on the scattering angle between the FOV and scattered photon direction as described in Ref. 42. With this selection, the background rate is reduced by a factor of 3 in 100 to 200 keV. This rejection factor appears small because the majority of non-gamma-ray backgrounds and partial events where photoabsorption takes place outside of the Compton camera are rejected due to nonphysical scattering angle calculated from energies, and also gamma-ray backgrounds from the outside of the FOV may still satisfy the constraint on the scattering angle between the FOV and scattered photon direction by chance. This event selection is not optimized for the observed backgrounds and further optimization is on going for the scientific analysis of the Crab Nebula. Figure 13(b) shows the resulting spectrum of four operational Compton cameras for the Crab observation after the background subtraction. The background spectrum was estimated using the observation data 1 day earlier. The total observation time is 5167 seconds after a dead-time correction. Figure 13(b) also shows the spectrum estimated by the Monte Carlo simulation using the same event reconstruction and selection as the observation. In this simulation, we assume the same spectral function for the Crab Nebula as Fig. 9(b). The observed data and the simulation agree with each other above 100 keV, and the observed data are slightly below the simulation in 60- to 100-keV range.

Fig. 13

(a) Spectra for the background in orbit before (black) and after (blue) the event selection on the Compton kinematics. (b) Spectrum of events which pass Compton reconstruction criteria for the data taken during the Crab observation. The background is subtracted. The events from four Compton cameras (excluding two Compton cameras, which were turned off during the Crab observation) are summed. The spectrum of the Monte Carlo simulation is shown in red.

The analysis of the gamma-ray polarization from the Crab Nebula is ongoing, and the result will be published in a separate paper.