Skip to main content

X-ray ghost imaging using individual synchrotron pulses


Ghost imaging is an indirect method in which most of the X-rays used for the experiment never actually interact with the sample. The first demonstration of X-ray ghost imaging has been carried out at beamline ID19. The method uses two copies of the same speckled beam and the image is retrieved indirectly by correlating the two measured signals. This experiment could lead to the development of low-dose medical X-ray diagnostics and diffraction-imaging at free electron lasers.

  • Share

The time structure of the ESRF synchrotron beam has been used to demonstrate the first ever instance of X-ray ghost imaging. Ghost imaging is an indirect imaging method that has received considerable attention in recent years [1].  In its simplest form, ghost imaging requires two identical copies of a structured (speckled) beam, as illustrated in Figure 1. One copy is directed on the sample and only the total intensity (not the image) transmitted or scattered by the specimen is recorded by a bucket (point) detector. The second, empty, beam is imaged by a pixel array detector, to form the reference image. In this way the sample image is never recorded directly. Remarkably though, the sample image can be reconstructed by repeating the experiment many times, each time with a different, but correlated, speckle pattern. The image is retrieved by correlating the bucket signals and the reference images. This can be done with an electronic correlator or, as in the case of this experiment, working offline on the whole data sequence.

Schematic diagram of a ghost-imaging experiment.

Figure 1. Schematic diagram of a ghost-imaging experiment. A speckled X-ray beam is split into two copies. One interacts with a sample and only the total intensity transmitted by the specimen is recorded.  The second beam is imaged. Neither of the detectors measures an image of the sample. The 'ghost' image is retrieved by correlating the detector outputs. In our experiment, we actually used a single detector, and perform an offline correlation of the data.

Combining the special '4-bunch' filling mode of the ESRF with the intense X-ray pulses produced by the undulators at beamline ID19, we produced the first proof of principle demonstration of ghost imaging in the hard X-ray spectral range. In 4-bunch mode, the temporal separation between the bunches is approximately 704 ns, corresponding to a frequency of about 1.42 MHz. In our experiment, the X-ray pulses were monochromatised to an energy of 20 keV by a pair of silicon crystals. The beam was then split using a thin silicon crystal in Laue diffraction, to produce the two identical copies required by the ghost imaging protocol. The sample (an X-ray opaque Cu wire) was inserted in one of the beams and both beams were detected using the same camera frame.
We used an ultrafast camera to image the individual X-ray pulses, for an average of only a few consecutive pulses. In this way the characteristic shot noise of the synchrotron emission process is dominant and the beam contains natural speckles. In this proof-of-concept experiment, the bucket signal was calculated by summing the intensity measured in the sample beam and correlated with the reference beam image instead of using a separate bucket and reference detector.

The power spectrum of the bucket signal is shown in Figure 2a. The sharp peak at 0.71 MHz corresponds to half the storage ring frequency. Low frequency components of the spectrum arise from mechanical vibrations of the optics (monochromator and crystal beam splitter), which are problematic. Vibrations cause small changes in the crystal orientation which produce intensity variations in both the diffracted and transmitted beam.  Such variations are actually anti-correlated and therefore obscure the genuine speckle correlation coming from the electron bunch structure.

ower spectrum of the bucket signals and ghost image.

Figure 2. (a) Power spectrum of the bucket signals.  Two peaks are visible in the high frequency part (marked by arrows). They correspond to half the ESRF storage ring frequency (0.71 MHz) and to an alias of the same frequency (1.15 MHz). The actual storage ring frequency (1.42 MHz) was not resolved due to the insufficient portal resolution of the camera. (b) Ghost image of a copper wire measured by windowing the 0.71 MHz frequency component. (c) By windowing the 'wrong' frequency component (in this case 0.8 MHz) the ghost image is not retrieved.

To eliminate the effect of mechanical vibrations, we performed Fourier filtering of the spectrum of both reference and bucket signals.  The true speckle correlation was isolated by windowing the 0.71 MHz peak (equivalent to taking the average of two pulses) in both reference and bucket. 

Figure 2b shows the recovered ghost image of the wire, using the filtered spectrum. Figure 2c shows the ghost image obtained when the window selects a nearby spectral region instead of the 0.71 MHz peak. In this case the speckle correlation is not present and the sample image is not recovered.

To make sure that the effect we were measuring was genuine, we repeated the experiment with different conditions, for instance by exchanging the role of bucket and reference beam, or by using the correlation peak at 1.15 MHz, which corresponds to an alias of the true storage ring frequency. In all cases we were able to reconstruct the sample’s ghost image after Fourier filtering.

Our experiment confirms that shot noise produced in the synchrotron emission of isolated electron bunches can be successfully used for correlation experiments. This result might inspire new ideas on single molecule diffraction imaging at FELs. By reducing the dose of the bucket beam one could devise a “diffraction without destruction” protocol using ghost imaging [2].

X-ray ghost imaging has other potential applications that include spectroscopy such as X-ray fluorescence where one cannot generally use an imaging detector. Also, it could be highly beneficial for low dose radiology as most of the photons in a ghost imaging experiment never actually interact with the sample.


Principal publication and authors
Experimental X-ray ghost imaging, D. Pelliccia (a,b,e), A.O. Rack (c), M. Scheel (d), V. Cantelli (e,c) and D.M. Paganin (e), Phys. Rev. Lett. 117, 113902 (2016); doi: 10.1103/PhysRevLett.117.113902.
(a) RMIT University, Melbourne (Australia)
(b) Australian Synchrotron, Melbourne (Australia)
(c) ESRF
(d) Synchrotron Soleil, Gif-sur-Yvette (France)
(e) Helmholtz-Zentrum Dresden-Rossendorf, Dresden (Germany)
(e) Monash University, Melbourne (Australia)


[1] B.I. Erkmen and J.H. Shapiro, Adv. Opt. Phot. 2, 405-450 (2010).
[2] Z. Li, N. Medvedev, H. Chapman, and Y. Shih, arXiv:1511.05068 (2015).

Top image: Ghost image of a copper wire.