By Noah Schnitzer, Suk Hyun Sung @ Hovden Lab

Code accompanying our paper: Schnitzer, N., Sung, S. H., & Hovden, R. (2020). Optimal STEM Convergence Angle Selection Using a Convolutional Neural Network and the Strehl Ratio. Microscopy and Microanalysis, 26(5), 921-928.

A MATLAB library for quick simulation of the STEM probe and electron Ronchigram, and training of a CNN to assess probe quality from the Ronchigram.

See also: http://ronchigram.com/ (source) for a similar JS/Wasm project.

MATLAB scripts and functions are organized into folders but are mutually dependent -- all subfolders must be added to the path to work properly. assessment and simulation functions are well tested for reasonable inputs, dataset_building scripts are not and should be used cautiously/as examples.

To get started, check out misc/example.m

• assessment: functions to calculate heuristics for aberration functions
  • indiv_p4_calculator.m: Calculate the individual aberration phase shift convergence angle for a single aberration
  • par_strehl_calculator.m: Calculate the Strehl ratio converence angle (parallel) for aberrations
  • pi4_calculator.m: Calculate the total aberration phase shift convergence angle for a single aberration
  • probe_sizer.m: Wraps resolution_test.m to calculate 50% probe current diameter for given aberration functions and convergence angles
  • resolution_test.m: Poorly named probe size assessment, should be used with effprobe option
  • strehl_calculator.m: Calculate the Strehl ratio converence angle for aberrations
• CNN: scripts to train and test CNN on simulated Ronchigrams
  • lim_net.m: Train CNN used in paper
  • transfer_learning.m: Transfer learn on Alexnet
• dataset_building: scripts and functions to build data sets. Note many parameters are embedded in scripts, have only been tested for limitied inputs.
  • aberration_series.m: Calculates heuristics as specific aberrations are varied. Used to e.g. find defocus to balance other aberrations.
  • dataset_generator.m: Generates a dataset with a ~ uniform distibution of convergence angles.
  • defocus_distribution.m: More slowly generates a dataset with ~ uniform distribution of convergence angles and defocus set to compensate other aberrations.
  • distribution_generator.m: Subroutine to generate aberrations and dynamically scale to try to get uniform CA distribtion.
• misc: miscellaneous scripts and functions, e.g. utilities and examples
  • ab_set.mat: Test set of in-focus randomly generated aberrations with a broad range of optimal convergence angles. Read by ronchi_game.m, used for testing network and human performance in paper.
  • colordef.m: Colors from paper
  • example.m: Brief walkthrough for Ronchigram simulation and heuristic calculation
  • get_aberration.m: Calculates the phase shift for a specific aberration
  • normalize_data.m: Normalizes data to min 0 max 1
  • px_to_ang.m : Calculates the scale factor Å/px for a given accelerating voltage and simulation dimension
  • radial_average.m: Calculates a radial average
  • ronchi_game_labeling.m: Minimal version of ronchi_game GUI Application for human aperture selection. This version has no feedback and minimal UI for quick labeling of testing data. Requires a file ab_set.mat in the working directory with a variable abs with containing any number of aberrations. User selections will be saved in game_res_1.mat file in the working directory.
• ronchi_game: GUI Application for human aperture selection, with optional feedback.
• simulation: functions for STEM probe and Ronchigram simulation
  • aberration_generator.m: Generates aberrations with random magnitude and angle out to 5th order. Relative scale of magnitudes is based off observations from AC-STEM, Kirkland 2011 Ultramicrosc.
  • aperture_mask.m: Generates a binary disc of given radius
  • calculate_aberration_function.m: Calculates phase shift for an aberration function
  • calculate_probe.m: Calculates STEM probe given ab phase shift, convergence angle
  • shifted_ronchigram.m: Calculates a Ronchigram for an aberration function, shift in Ronch center relative to obj aperture, and convergence angle
  • shifted_ronchigram_o.m: Legacy shifted_ronchigram to match parameters (e.g. grating resize factor) of trained networks.