Search

Modern photon science performed at high repetition rate free-electron laser (FEL) facilities and beyond relies on 2D pixel detectors operating at increasing frequencies (towards 100 kHz at LCLS-II) and producing rapidly increasing amounts of data (towards TB/s). This data must be rapidly stored for offline analysis and summarized in real time. While at LCLS all raw data has been stored, at LCLS-II this would lead to a prohibitive cost; instead, enabling real time processing of pixel detector raw data allows reducing the size and cost of online processing, offline processing and storage by orders of magnitude while preserving full photon information, by taking advantage of the compressibility of sparse data typical for LCLS-II applications. We investigated if recent developments in machine learning are useful in data processing for high speed pixel detectors and found that typical deep learning models and autoencoder architectures failed to yield useful noise reduction while preserving full photon information, presumably because of the very different statistics and feature sets between computer vision and radiation imaging. However, we redesigned in Tensorflow mathematically equivalent versions of the state-of-the-art, "classical" algorithms used at LCLS. The novel Tensorflow models resulted in elegant, compact and hardware agnostic code, gaining 1 to 2 orders of magnitude faster processing on an inexpensive consumer GPU, reducing by 3 orders of magnitude the projected cost of online analysis at LCLS-II. Computer vision a decade ago was dominated by hand-crafted filters; their structure inspired the deep learning revolution resulting in modern deep convolutional networks; similarly, our novel Tensorflow filters provide inspiration for designing future deep learning architectures for ultrafast and efficient processing and classification of pixel detector images at FEL facilities., Comment: 9 pages, 9 figures

Modern photon counting pixel detectors enabled a revolution in applications at synchrotron light sources and beyond in the last decade. One of the limitations of current detectors is reduced counting linearity or even paralysis at high counting rates, due to dead-time which results in photon pile-up. Existing dead-time and pile-up models fail to reproduce the complexity of dead-time effects on photon counting, resulting in empirical calibrations for particular detectors at best, imprecise linearization methods, or no linearization. This problem will increase in the future as many synchrotron light sources plan significant brilliance upgrades and free-electron lasers plan moving to a quasi-continuous operation mode. We present here the first models that use the actual behavior of the analog pre-amplifiers in spectroscopic photon counting pixel detectors with constant current discharge (e.g., Medipix family of detectors) to deduce more accurate analytical models and optimal linearization methods. In particular, for detectors with at least two counters per pixel, we completely eliminate the need of calibration, or previous knowledge of the detector and beam parameters (dead-time, integration time, large sets of synchrotron filling patterns). This is summarized in several models with increasing complexity and accuracy. Finally, we present a general empirical approach applicable to any particular cases where the analytical approach is not sufficiently precise., Comment: 7 figures