Chandra's Ultimate Angular Resolution: Studies of the HRC-I Point Spread Function

Michael Juda & Margarita Karovska
(Harvard-Smithsonian CfA)


The Chandra High Resolution Camera (HRC) should provide an ideal imaging match to the High-Resolution Mirror Assembly (HRMA). The laboratory-measured intrinsic resolution of the HRC is ~20 microns FWHM. HRC event positions are determined via a centroiding method rather than by using discrete pixels. This event position reconstruction method and any non-ideal performance of the detector electronics can introduce distortions in event locations that, when combined with spacecraft dither, produce artifacts in source images. We compare ray-traces of the HRMA response to "on-axis" observations of AR Lac and Capella as they move through their dither patterns to images produced from filtered event lists to characterize the effective intrinsic PSF of the HRC-I. A two-dimensional Gaussian, which is often used to represent the detector response, is NOT a good representation of the intrinsic PSF of the HRC-I; the actual PSF has a sharper peak and additional structure.


Chandra's High-Resolution Mirror Assembly (HRMA) is the highest resolution X-ray optic produced for astronomical observations. The High Resolution Camera (HRC) was designed to make full use of the HRMA resolution. Each HRC detector uses a pair of micro-channel plates (MCPs) to convert the incoming X-ray to charge. The front MCP channel diameter and pitch determines ultimate possible resolution of the HRC (HRC-I diameter/pitch = 10/12.5 um, HRC-S = 12.5/15 um). Subsequent processing by 2nd MCP and read-out are likely to add blur. The charge-cloud from back of the MCPs is "imaged" on crossed-grid of wires. X-ray event positions are calculated using the centroid of the charge on the wire grid. The charge-cloud centroid is determined per axis with a "three-tap" algorithm. The pixel size for positions is arbitrary; the default value for standard processing is ~6.43 um or 0.1318 arcsec, which over-samples the PSF. Laboratory measurements of the spatial resolution on a "flight-like" system had FWHM of 20-25 um.

MCP principle of
Figure 1: HRC principle of operation: Incoming X-ray interacts with wall of channel releasing (at most) a few electrons into the channel. Applied HV accelerates electrons, which impact the channel wall releasing more electrons, amplifying the signal. The charge cloud exiting the MCP is collected on a grid of wires. The charge on a wire is resistively divided between charge amplifiers. The event location is determined by a centroid of the amplifier signals.

Performance Details

There are a few issues with the implementation of the charge-cloud readout (both expected and unexpected) that compromise the determination of the X-ray location. "Gaps" are generated in images formed using the simple centroid since an increasing portion of charge is lost when approaching mid-point between amplifiers. De-gap corrections are applied to shift the event positions to close the gap. Electronic ringing in amplifier strings occurs for a subset of events. If not corrected, the ringing produces jet-like artifacts. The affected events can be identified and a partial signal correction made (1 of 3 signals per axis). The de-gap correction attempts to fix the impact of any residual distortions from the ringing. Non-matching gains/offsets in amplifier strings exist that were not observed at component level. Again, de-gap corrections attempt to fix the induced distortions.The impact of these issues may differ on each axis and lead to differing resolutions along the two detector axes. Standard processing applies best know corrections for event-position reconstruction and standard filtering rejects events with obviously bad positions. Due to the amplifier-ringing issue some of the remaining calculated event positions may be shifted slightly away from their actual location. In the analysis presented here we have eliminated the events affected by the ringing by rejecting all events with an amplifier scale (AMP_SF) setting of 3.

Gaussian Fits

A sherpa thread[1] describes how to use an image from a ChaRT[2] produced ray-trace as the kernel in a two dimensional fit of image data. The ChaRT ray-trace does not include any detector effects. Gaussian approximations to detector effects can be added using MARX[3] or the CIAO tool psf_project_ray. The sigma of the Gaussian should be derived from calibration. Using the ray-trace result without including a contribution from the instrument as a kernel in fits to on-axis unresolved sources provides an estimate for added blur from the detector (and aspect) as well as an assessment of whether a Gaussian is an appropriate model for the detector contribution. Performing such a fit on an observation of AR Lac taken in the first months of the mission yields a Gaussian FWHM of 4.69 HRC pixels (0.618 arcsec) for a sigma of 0.262 arcsec. Unfortunately, the residuals of the fit suggest that this Gaussian blur to the ray-trace is a poor representation to the actual instrument response. This might be expected given the assumptions made in this simple treatment. We have no reason to expect the HRC intrinsic PSF to be a Gaussian.


In order to study the intrinsic HRC-I point-spread response we first generated ray-trace simulations for a set of AR Lac calibration observations using SAOTrace[4] (the engine behind ChaRT). We modeled the effect of spacecraft dither by calculating set of observed, source-rate-weighted, pointing offsets (~2000 offsets per observation). A ray-trace was generated for each pointing offset. The output rays from each ray-trace were "spatially quantized" onto a triangular grid that reflects the MCP channel structure, with all rays that impact a given channel reassigned to the coordinates of the channel center. Rays were then given random offsets within the channel area to mimic the loss of event location information produced by the charge cascade down the channel and the processing by the second MCP. The known pointing offset for each ray-trace was used to "de-dither" the set of ray-traces to a common center. The CIAO tool psf_project_ray was used to convert the combined ray-traces to an event list. The ray-trace events were used as a kernel in 2D-Gaussian fits of the observed source image. We worked with images rotated to the HRC-I coordinate frame so that the results do not have to be corrected for the differing spacecraft roll between observations. The table below gives the resulting FWHM for observations of AR Lac taken each year to monitor the HRC performance. The fit ellipticity parameter for most of the fits was ~0.2 but with no preferred direction. The FWHM values of these fits are slightly smaller than those using ChaRT ray-traces due to the randomization within the MCP channel that was included the ray-traces.

ObsIDDateFWHM (pixels)sigma (arcsec)
1385 1999-10-054.59 +/- 0.050.257 +/- 0.003
996 1999-12-093.80 +/- 0.180.213 +/- 0.010
1484 2000-12-123.91 +/- 0.150.219 +/- 0.008
2608 2002-01-273.94 +/- 0.150.221 +/- 0.008
4294 2003-02-223.90 +/- 0.140.218 +/- 0.008
5060 2004-09-133.74 +/- 0.130.210 +/- 0.007
5979 2005-09-273.47 +/- 0.080.194 +/- 0.004
6519 2006-09-203.55 +/- 0.080.199 +/- 0.004
8298 2007-09-173.58 +/- 0.080.200 +/- 0.004
9640 2008-09-073.42 +/- 0.340.191 +/- 0.019
105782009-09-243.23 +/- 0.730.181 +/- 0.041

Figure 2 shows the fit results for the first of these observations. The observed source event distribution is more peaked than the best-fit Gaussian; this is characteristic of all the fits. A comparison of the residuals among all the fits showed the same systematic pattern.

2-D Gaussian fit results
Figure 2: Results of a 2D-Gaussian fit, using a ray-trace of the optics as a kernel, to a HRC-I on-axis observation of a point source (ObsID 1385). The source image and ray-trace were binned in the HRC-I detector coordinate frame at 0.2 times the nominal pixel size. The vertical white bar in the source image is 1 arcsec in length. Non-zero ellipticity was allowed in the fit. The color scale for the smoothed residuals is red/blue for data greater/less than the model. The systematic pattern in the smoothed residuals is characteristic of all fits that we have performed.


A systematic pattern in the fit residuals suggests that there is underlying structure in the HRC-I PSF relative to the simple Gaussian. In order to understand these residuals we have deconvolved each of the AR Lac observations with the ray-trace simulations. The resulting images should provide a guide to what we can expect in the intrinsic structure of the HRC-I PSF. Figure 3 shows the deconvolved images for the AR Lac observations. This time-sequence of images seems to imply that there has been evolution in the structure. By the year 2002 observation a hook-like structure has appeared extending ~0.8 arcsec to the negative-V side of the main peak. The amount of signal in this hook region is ~6%. The main peak appears more extended along the U-axis but also exhibits a triangular shape. The triangular shape may be related to the MCP channel structure.

AR Lac deconvolutions
Figure 3: Deconvolution of observed source image with ray-trace image for the AR Lac calibration observations. The vertical white bar is 1 arcsec in length. By the 2002 observation a hook-shaped extension to the -V direction has appeared. This feature appears to be stable over the last several years.

The AR Lac observations were all performed in the same region of the detector so the obvious question is whether this structure is related to this position. Figure 4 shows similar images for a series of observations of Capella in which the HRC-I position was translated behind the telescope aim-point. The translation was large enough that distinctly different regions of the detector and charge amplifiers were used in imaging the source. Structure similar to that in the later AR Lac observations is present in each of these observations.

Capella deconvoltions
Figure 4: Similar to figure 3 but for Capella calibration observations that translated the HRC-I behind the telescope aim-point. The offset position is indicated above each panel. The offsets are large enough that the dither samples non-overlapping regions of the detector. The same hook-like feature as seen in the AR Lac data is observed.

The Bottom Line

The intrinsic PSF of the HRC-I is not well-represented by a Gaussian; there is significant underlying structure that is not represented by a simple function. While we are working to understand the process by which the HRC hardware or the processing of its signals could produce this structure, we cannot yet rule out that it is due to un-modeled performance of the HRMA. An examination of suitable ACIS observations may resolve this question. Any observer who sees (or wishes to find) structure in their HRC observations on arcsec scales should proceed with caution. Applying a filter to exclude AMP_SF = 3 events can help provide additional cleaning of the data. Two (or more) observations at differing spacecraft rolls should help separate image structure due to detector artifacts from structure in their source.


[1] "Accounting for PSF Effects in 2D Image Fitting"
[2] "ChaRT: The Chandra Ray Tracer" C. Carter, et al. ADASS XII ASP Conference Series, Vol. 295, 2003, p.477
[3] MARX
[4] SAOTrace

Last modified: Tue May 18 14:35:55 EDT 2010

Dr. Michael Juda
Harvard-Smithsonian Center for Astrophysics
60 Garden Street, Mail Stop 70
Cambridge, MA 02138, USA
Ph.: (617) 495-7062
Fax: (617) 495-7356