February_EDFA_Digital

edfas.org 13 ELECTRONIC DEV ICE FA I LURE ANALYSIS | VOLUME 24 NO . 1 the error is unacceptably large. However, given that the damage induced in the sample is a function of the speed of the scan and the overlap of the beam positions, reduc- ing the total number of beam positions in the image can have a tremendous effect on the overall sample stability during the experiment. It is also important to note that there are many possible algorithms for the reconstruc- tion of these sub-sampled images, and more are being developed daily. All the existing algorithms use the same construct to inpaint the image and so after the hardware solution is in place, the methodology can improve with every algorithm upgrade. EXAMPLE RECONSTRUCTIONS There is not enough space in this brief publication to go through thedetails of the reconstructionalgorithm. The Beta Process Factor Analysis (BPFA) approach to inpaint sub-sampled images [8,10,11] was utilizedanda fewexamples Fig. 2 (a) A series of dictionary elements (Fourier components in this case) can be combined with defined scalar weightings to reproduce the true 1D signal. (b) Reducing the number of points in the true signal (c-f) increases the speed and decreases the dose in the sampling of the experiment, but still generates a good fit to the data. At some point (f) the level of sub- sampling leads to an unacceptable error in the reconstruction. This level of sub-sampling is material and instrument resolution dependent. [8] (a) (b) (c) (d) (e) (f)