slmsuite.hardware.slms.screenmirrored.ScreenMirrored

class ScreenMirrored(display_number, bitdepth=8, verbose=True, **kwargs)

Bases: SLM

Wraps a pyglet window for displaying data to an SLM.

Caution

This class currently supports SLMs with 8-bit precision or less. In the future, this class will support 16-bit SLMs using RG color.

Important

Many SLM manufacturers provide an SDK for interfacing with their devices. Using a python wrapper for these SDKs is recommended, instead of or in supplement to this class, as there often is functionality additional to a mirrored screen (e.g. USB for changing settings) along with device-specific optimizations.

Note

There are a variety of python packages that support blitting images onto a fullscreen display.

• Simple DirectMedia Layer (SDL) wrappers:

  • pygame (link), which also supports OpenGL. Only supports one screen.

  • sdl2 (readthedocs) through the PySDL2 package. Requires additional libraries.

• Open Graphics Library (OpenGL) wrappers:

  • moderngl (readthedocs), an OpenGL wrapper focusing on a pythonic interface for core OpenGL functions.

  • OpenGL (link) through the PyOpenGL/PyOpenGL_accelerate package, a very light OpenGL wrapper.

  • pyglet (readthedocs), a light OpenGL wrapper.

• GUI Library wrappers:

  • gi (readthedocs), through the PyGObject package wrapping GTK and other GUI libraries.

  • pyqt6 (link), through the PyQt6 package wrapping the version 6 Qt GUI library.

  • tkinter (link), included in standard python, wrapping the Tcl/Tk GUI library.

  • wx (link), through the wxPython package wrapping the wxWidgets GUI library. slmpy (GitHub) uses wx.

slmsuite uses pyglet as the default display package. pyglet is generally more capable than the mentioned SDL wrappers while immediately supporting features such as detecting connected displays which low-level packages like OpenGL and moderngl do not have. pyglet allows us to interact more directly with the display hardware without the additional overhead that is found in GUI libraries. Most importantly, pyglet is well documented.

However, it might be worthwhile in the future to look back into SDL options, as SDL surfaces are closer to the pixels than OpenGL textures, so greater speed might be achievable (even without loading data to the GPU as a texture). Another potential improvement could come from writing cupy datastructures to OpenGL textures directly, without using the CPU as an intermediary. There is some precedent for transferring data from CUDA (on which cupy is based) to OpenGL, though cupy does not currently directly support this.

Important

ScreenMirrored uses a double-buffered and vertically synchronized (vsync) OpenGL context. This is to prevent “tearing” resulting from data being modified during a display write: rather, all monitor writes are synchronized such that clean frames are always displayed. This feature is similar to the isImageLock flag in slmpy, but is implemented a bit closer to the hardware.

window

Fullscreen window used to send information to the SLM.

Type

pyglet.window.Window

tex_shape_ratio

Ratio between the SLM shape and the (power-2-padded) texture stored in OpenGL.

Type

(int, int)

buffer

Memory used to load data to the OpenGL memory. Of type np.uint8.

Type

numpy.ndarray

cbuffer

Array of length prod(shape) * 4 (4 bytes per RGBA). Maps to the same memory as buffer. Used to load data to the texture.

Type

pyglet.gl.GLubyte

texture

Identifier for the texture loaded into OpenGL memory.

Type

pyglet.gl.GLuint

Methods

close	Closes frame.
info	Get information about the available displays, their indexes, and their sizes.
load_vendor_phase_correction	Loads vendor-provided phase correction from file, setting phase_correction.
point_spread_function_knm	Fourier transforms the wavefront calibration's measured amplitude to find the expected diffraction-limited perfomance of the system in "knm" space.
set_measured_amplitude_analytic	Sets measured_amplitude used for hologram generation in the absence of a proper wavefront calibration.
spot_radius_kxy	Approximates the expected radius of farfield spots in the "kxy" basis based on the near-field amplitude distribution measured_amplitude.
write	Checks, cleans, and adds to data, then sends the data to the SLM and potentially waits for settle.

__init__(display_number, bitdepth=8, verbose=True, **kwargs)

Initializes a pyglet window for displaying data to an SLM.

Caution

slmsuite makes use of user-supplied knowledge of SLM pixel size: SLM.dx_um, SLM.dy_um. Be sure these values are correct.

Caution

An SLM designed at 1064 nm can be used for an application at 780 nm by passing wav_um=.780 and wav_design_um=1.064, thus causing the SLM to use only a fraction (780/1064) of the full dynamic range. Be sure these values are correct. Note that there are some performance losses from using this modality (see write()).

Caution

There is some subtlety to complex display setups with Linux. Working outside the default display is currently not implemented.

Parameters

• display_number (int) – Monitor number for frame to be instantiated upon.

• verbose (bool) – Whether or not to print extra information.

• **kwargs – See SLM.__init__() for permissible options.

References

14

close()

Closes frame. See SLM.

static info(verbose=True)

Get information about the available displays, their indexes, and their sizes.

Parameters

verbose (bool) – Whether or not to print display information.

Returns

The number and geometry of each display.

Return type

list of (int, (int, int, int, int)) tuples

load_vendor_phase_correction(file_path)

Loads vendor-provided phase correction from file, setting phase_correction. By default, this is interpreted as an image file and is padded or unpadded to the shape of the SLM. Subclasses should implement vendor-specific routines for loading and interpreting the file (e.g. Santec loads a .csv).

Parameters

file_path (str) – File path for the vendor-provided phase correction.

Returns

phase_correction, the vendor-provided phase correction.

Return type

numpy.ndarray

point_spread_function_knm(padded_shape=None)

Fourier transforms the wavefront calibration’s measured amplitude to find the expected diffraction-limited perfomance of the system in "knm" space.

Parameters

padded_shape ((int, int) OR None) – The point spread function changes in resolution depending on the padding. Use this variable to provide this padding. If None, do not pad.

Returns

The point spread function of shape padded_shape.

Return type

numpy.ndarray

set_measured_amplitude_analytic(radius, units='norm')

Sets measured_amplitude used for hologram generation in the absence of a proper wavefront calibration. FourierSLM includes capabilities for wavefront calibration via wavefront_calibrate(). This process also measures the amplitude of the source on the SLM and stores this in measured_amplitude. measured_amplitude is used for better refinement of holograms during numerical optimization. If one does not have a camera to use for wavefront_calibrate(), this method allows the user to set an approximation of the source amplitude based on an assumed \(1/e\) amplitude (\(1/e^2\) power) Gaussian beam radius.

Parameters

• radius (float) – Radius in normalized units to assume for the source Gaussian beam.

• units (str in {"norm", "nm", "um", "mm", "m"}) – Units for the given radius.

Returns

measured_amplitude.

Return type

numpy.ndarray

spot_radius_kxy()

Approximates the expected radius of farfield spots in the "kxy" basis based on the near-field amplitude distribution measured_amplitude.

Returns

Average radius of the farfield spot.

Return type

float

write(phase, phase_correct=True, settle=False)

Checks, cleans, and adds to data, then sends the data to the SLM and potentially waits for settle. This method calls the SLM-specific private method _write_hw() which transfers the data to the SLM.

Warning

Subclasses implementing vendor-specific software should not overwrite this method. Subclasses should overwrite _write_hw() instead.

Caution

The sign on phase is flipped before converting to integer data. This is to convert between the ‘increasing value ==> increasing voltage (= decreasing phase delay)’ convention in most SLMs and slmsuite’s ‘increasing value ==> increasing phase delay’ convention. As a result, zero phase will appear entirely white (255 for an 8-bit SLM), and increasing phase will darken the displayed pattern. If integer data is passed, this data is displayed directly and the sign is not flipped.

Important

The user does not need to wrap (e.g. numpy.mod(data, 2*numpy.pi)) the passed phase data, unless they are pre-caching data for speed (see below). write() uses optimized routines to wrap the phase (see the private method _phase2gray()). Which routine is used depends on phase_scaling:

• phase_scaling is one.

  Fast bitwise integer modulo is used. Much faster than the other routines which depend on numpy.mod().

• phase_scaling is less than one.

  In this case, the SLM has more phase tuning range than necessary. If the data is within the SLM range [0, 2*pi/phase_scaling], then the data is passed directly. Otherwise, the data is wrapped by \(2\pi\) using the very slow numpy.mod(). Try to avoid this in applications where speed is important.

• phase_scaling is more than one.

  In this case, the SLM has less phase tuning range than necessary. Processed the same way as the phase_scaling is less than one case, with the important exception that phases (after wrapping) between 2*pi/phase_scaling and 2*pi are set to zero. For instance, a sawtooth blaze would be truncated at the tips.

Caution

After scale conversion, data is floor() ed to integers with np.copyto, rather than rounded to the nearest integer (np.around() equivalent). While this is irrelevant for the average user, it may be significant in some cases. If this behavior is undesired consider either: write() integer data directly or modifying the behavior of the private method _phase2gray() in a pull request. We have not been able to find an example of np.copyto producing undesired behavior, but will change this if such behavior is found.

Parameters

• phase (numpy.ndarray or None) –

  Phase data to display in units of \(2\pi\), unless the passed data is of integer type and the data is applied directly.

  • If None is passed to write(), data is zeroed.

  • If the array has a larger shape than the SLM shape, then the data is cropped to size in a centered manner (unpad).

  • If integer data is passed with the same type as display (np.uint8 for <=8-bit SLMs, np.uint16 otherwise), then this data is directly passed to the SLM, without going through the “phase delay to grayscale” conversion defined in the private method _phase2gray(). In this situation, phase_correct is ignored. This is error-checked such that bits with greater significance than the bitdepth of the SLM are zero (e.g. the final 6 bits of 16 bit data for a 10-bit SLM). Integer data with type different from display leads to a TypeError.

  Usually, an exact stored copy of the data passed by the user under phase is stored in the attribute phase. However, in cases where phase_scaling not one, this copy is modified to include how the data was wrapped. If the data was cropped, then the cropped data is stored, etc. If integer data was passed, the equivalent floating point phase is computed and stored in the attribute phase.

• phase_correct (bool) – Whether or not to add phase_correction to phase.

• settle (bool) – Whether to sleep for settle_time_s.

Returns

display, the integer data sent to the SLM.

Return type

numpy.ndarray

Raises

TypeError – If integer data is incompatible with the bitdepth or if the passed phase is otherwise incompatible (not a 2D array or smaller than the SLM shape, etc).