Article Index

The Soft X-Ray Imager (SXI)

The THESEUS Soft X-ray Imager (SXI) comprises 4 DU. Each DU is a wide field lobster eye telescope using the optical principle first described by Angel (1979) with an optical bench as shown in Figure 3.1. The optics aperture is formed by an array of 8x8 square pore Micro Channel Plates (MCPs) supplied by PHOTONIS France SAS, (Avenue Roger Roncier, 19100 Brive La Gaillarde, France). The MCPs are 40x40 mm2 and are mounted on a spherical frame with radius of curvature 600 mm (2 times the focal length of 300 mm). The open aperture provided by each plate is 38x38 mm2 so that a 1 mm wide strip around the edges is available for mounting/gluing. Table 1 summarizes SXI characteristics. The mechanical envelope of a SXI module has a square cross-section 320x320 mm2 at the optics end tapering to 200x200 mm2 at detector. The depth of the detector housing is 200 mm giving an overall module length of 500 mm. The optics assembly has a mass of 2.5 kg, the detector assembly 13 kg and the structure 2.5 kg giving an overall module mass of 18 kg.


  • 1

Energy band (keV)


Telescope type:

Lobster eye

Optics aperture (mm2)


Optics configuration

8x8 square pore MCPs

MCP size (mm2)


Focal length (mm)


Focal plane shape


Focal plane detectors

CCD array

Size of each CCD (mm2)


Pixel size (µm)


Pixel Number


Number of CCDs


Field of View (square deg)

~1 sr

Angular accuracy (best, worst) (arcsec)

(<10, 105)

Power [W] 27.8
Mass [kg] 40
Figure 1. Optical elements of a SXI module. Table 1. The SXI characteristics.


The left-hand side of Figure 2 shows the optics frame of the breadboard model for the SVOM MXT lobster eye telescope which comprises 21 square MCPs mounted over a 5x5 grid (the corners are unoccupied for this instrument). The front surface is spherical with radius of curvature 2000 mm giving a focal length of 1000 mm. The design proposed for SXI uses the same plate size and exactly the same mounting principle but a shorter focal length, (300 mm), so the radius of curvature of the front surface must be 600 mm. The right-hand panels of Figure 2 shows a schematic of a single plate and a micrograph that reveals the square pore glass structure. The focal plane of each SXI module is a spherical surface of radius of curvature 600 mm situated a distance 300 mm (the focal length) from the optics aperture. The detectors for each module comprise a 2x2 array of large format CCDs baselined to be supplied by e2v (e2v technologies (UK) Ltd, 106 Waterhouse Lane, Chelmsford, Essex, CM1 2QU, England), where each CCD has an active area of 81.2x81.2 mm2 (1/6 to be used as a frame store). The detectors are tilted to approximate to the spherical focal surface.


  • 2
Figure 2. Left: The SVOM MXT lobster eye optic aperture frame. Top right: A schematic of a single square pore MCP. Bottom right: A micrograph of a square pore MCP showing the pore structure. This plate has a pore size d=20 microns and a wall thickness w=6 microns.


SXI Calibrations

SXI optic: X-ray beam line testing (University of Leicester or Panter MPE) to measure the focal length, the effective area and the point spread function as a function of off-axis angle and energy. SXI detector: Vacuum test facility (University of Leicester or OU) to measure the gain and energy resolution as a function of energy. SXI end-to-end: X-ray beam line facility (University of Leicester or Panter, MPE) to confirm the focal length and measure the instrument effective area and PSF as a function of photon energy and off-axis angle. SXI in orbit calibration: use cosmic sources to confirm in-flight alignment, plate scale, point spread function, effective area, vignetting and energy resolution. Regular monitoring of cosmic sources to monitor calibration. Optional internal X-ray source above CCDs to measure CTI.


SXI Performance, Sensitivity and Data Rate

The imaging area of the CCDs sets the field of view of each module. Each CCD has dimensions 81.2x81.2 mm2 which, with a focal length of 300 mm, and allowing for the frame store gives an active area of 15.5x12.92 square degrees. The field of view of each DU is provided by 4 CCDs giving 801.3 square degrees. Thus, a compliment of 4 SXI modules has a total field of view of 3200 square degrees (0.9 steradians).


  • 3
Figure 3 The point spread function of the SXI.


The point spread function is shown in Figure 3. The inner dotted square shows the off-axis angle at which the cross arms go to zero as determined by the L/d ratio of the pores. For optimum performance at 1 keV we require L/d=50. The outer dotted square indicates the shadowing of the cross-arms introduced by the gap between the individual MCPs in the aperture. The central true-focus spot is illustrated by the projection plot to the left. The FWHM is 4.5 arc minutes and all the true-focus flux is contained by a circular beam of diameter 10 arc minutes. The collecting area, within a 10 arc minutes beam centered on the central focus, as a function of energy is shown in Figure 4. The optics provides the area plotted in black. The red curve includes the quantum efficiency of the CCD and the transmission of the optical blocking filter comprising a 60 nm Aluminium film deposited over the front of the MCPs and 260 nm of Aluminium plus 500 nm of parylene on the surface of the detectors. Because the angular width of the optics MCP-array is 2.3 degrees larger than the CCD-array the field of view is unvignetted at 1 keV and above so the collecting area shown in Figure 5 is constant across the field of view.


  • 4
  • 5
Figure 4. Collecting area as a function of energy. The black is the optics only. The red curve includes the quantum efficiency of the CCD and the transmission of the optical blocking filter.     Figure 5. The position accuracy of the SXI as a function of source and background count. R90 is the error radius that contains 90% of the derived positions


Using the Rosat All-sky Survey data we can estimate the count rate expected from the diffuse sky (Galactic and Cosmic) and point sources. The average is 1.12x10-5 cts s-1 per square arc minute, although at the Galactic poles it will be a factor of 2 less than this and in the Galactic plane a factor of 3 or so higher. The expected particle background rate in the CCDs is 0.0013 cts s-1 mm-1 so the mean background rate (sky+particles) is 0.0016 cts s-1 in a circular beam of diameter 10 arc minutes. The sensitivity to transient sources using this background rate and a false detection probability of 1.0x10-10 is shown as a function of integration time in the science requirements web-page. For longer integration times the source count required rises, e.g. to 30 counts for a 10000 second integration. The count rate expected from a typical GRB is 2.0 cts s-1. We find that 94% of the Swift BAT bursts (before Sept. 16 2010) would be detected by the SXI. The X-ray light curve of the afterglow would be detected to >1000 seconds after the trigger for a large fraction of the bursts.

Figure 5 shows how the position accuracy R90 (the radius of a circle which contains 90% of the derived positions) improves as the source count S increases. If we include the background count then R90=C.S/(S+gB)1/2 where C=255 arc seconds. The constant g=0.86 depends on the relative size of the focused spot and the beam used. At the sensitivity limit we get R90=105 arc seconds in 50 seconds integration and R90=74.5 arc seconds in 2000 seconds integration. Of course when source are detected well above the threshold the R90 will be much smaller. If we get >1000 source counts then R90<10 arc seconds and the position accuracy will be limited by the systematic errors in the aspect solution. The average count rate per module is ~40 cts s-1 dominated by the diffuse X-ray sky component. This will rise to ~120 cts s-1 in the Galactic plane. Bright transients will produce 10 cts s-1. Therefore the total maximum count rate expected from 5 modules is ~650 cts s-1 and the average is ~250 cts s-1.


  • 6
Figure 6. The two stage trigger algorithm. Top left-hand panel: the detected event distribution ΔT=4 seconds and a source count rate of 40 cts s-1 over the full PSF. The cross-arms are rotated wrt the detector axes to demonstrate how this can be handled. The Top right-hand panel: the detected event distribution in the patch of sky aligned to the cross-arm axes of the PSF (shown as the red rectangle in the top left-hand panel). The red cross-patch indicates the area used for the second stage of the algorithm. Bottom panels: the histograms along columns and rows in the patch.


Trigger Algorithm

The ideal algorithm would be some form of matched filtering using the full PSF distribution but because of the extent of the PSF this would be far too computationally heavy. At the other extreme a simple scheme would be to search for significant peaks using the cell size commensurate with the central peak in the PSF. This would be faster but utilizes only ~25% of the total flux detected. The scheme described below is a two stage process which exploits the cross-arm geometry but avoids computationally expensive 2-D cross-correlation. For the 1st stage the focal plane is divided into square patches with angular side length ~4d/L=1/12 radians aligned with the cross-arm axes. The dotted central square shown in Figure 3 indicates the size and orientation of such a patch. The optimum size of such patches depends on the HEW of the lobster-eye optic and the background count rate. The patches could correspond to detector elements or tiles in the focal plane, e.g. CCD arrays. The peak profile is the line spread profile of the central spot and cross-arms of the PSF. The remainder of the histogram distribution arises from events in the cross-arm parallel to the histogram direction, the diffuse component of the PSF and any diffuse background events not associated with the source. Because we are looking for transient sources the fixed pattern of the steady sources in the field of view at the time would have to be subtracted from the histogram distributions. As the pointing changes this fixed pattern background would have to be updated. A transient source is detected if a significant peak is seen in both histograms.

The sensitivity of detection and accuracy of the derived position of the source within the patch depends on the bin size of the histograms, the HEW of the central peak of the PSF and the background. For the most sensitive detection the bin size should be approximately equal to the HEW but this will limit the accuracy of the position. If the bin size of the histograms is chosen to be significantly smaller than the width of the HEW then the histograms can be smoothed by cross-correlation with the expected line width profile of the peak-cross-arm combination. Using the smaller bin size the histograms can also be used to estimate the position centroid of the source within the patch. The significance used for this first stage should be low, e.g. 2.5 sigma. This will provide candidate positions for the second stage.

For each of the candidate positions identified in the first stage a cross-arm patch is set up to cover just the detector area which is expected to contain a fraction of the full cross-arms and the central peak in the PSF. The cross-patch dimensions are changed depending on the integration time ΔT. For short integration times the total background count will be small and the cross-patch size is set large to capture a large fraction of the counts from the cross-arms and central peak.

The above considers a single value of ΔT. We envisage that a series of searches would be run in parallel each using a different integration time so that the sensitivity limit as a function of ΔT is covered. The basic scheme is illustrated in Figure 6. The total source count assumed for this illustration was 40 spread over the full PSF as plotted in Figure 3. The bin size used for the histograms was 1 arc min, and the HEW of the central peak of the PSF is approximately 4 arc mins. We have tested the algorithm over a range of integration times and background conditions. It achieves the sensitivity limit plotted in the science requirements web-page. When a significant transient peak is identified the position must be converted to sky coordinates using the current aspect solution (from Payload star trackers). Positions of all transients found must be cross-correlated with known source catalogues, e.g. Rosat All-Sky Survey, Flare stars, Swift BAT catalogue etc. Any position which does not match a known position must be passed to the Space Craft as a potential trigger position.

The processing required to implement the above is as follows:

1) Extract frames from the CCD at ΔT=2. seconds – this is the fastest rate set by the frame time. 2) Apply event reconstruction algorithm to the frames to give an event list with positions in CCD pixels and a pulse height. 3) Convert the pixel positions into a local module coordinate frame which is aligned to the cross-arms of the PSF. 4) Accumulate counts in the 1-D histograms. 5) Subtract the fixed source/background pattern from the histograms. 6) Scan the histograms for significant peaks and extract candidate positions for further analysis. 7) Set up the cross-arm mask at candidate positions to look for significant peaks. Calculate an accurate position in the local module coordinate frame for the peak. 8) Convert this position into global sky coordinates (quaternion). 9) Check positions against on-board catalogues to weed out known sources. 10) Communicate unidentified transients to the Space Craft. 4)-7) must be repeated for different Δt values e.g. 2, 20, 200, 2000 seconds.