This package implements a CLI script for generating pseudorandom noise from M-sequences. JSON metadata is generated for each run containing the generator configurations to reproduce the run.
Tested on python 3.8, 3.10, 3.11
$ python3 -m pip install .
# test, dev dependencies
$ python3 -m pip install .[test, dev]
usage: mnoisy-gen [-h] -i IMAGE_SIZE -n NUM_FRAMES [-t FRAME_TIME_SEC] [-l SEQUENCE_LENGTH] [-s INITIAL_SEED] [-o OUTPUT_FILE]
$ mnoisy-gen -i 24 -n 3
OR
$ python -m mnoisy -i 15 -n 30 -l 10 -s 1234
The output filename is extracted from the metadata and prefixed with rebuild-
usage: mnoisy-rebuild [-h] -f FILENAME [-i INDEX_OF_FRAME]
$ mnoisy-rebuild -f data.json -i 26
-
Currently, the update function up to M-sequences of length 12 are implemented. That means image sizes are limited to 143x143 px, otherwise periodicity is introduced to the signal.
-
The performance of matplotlib animation is sluggish. Running the script and updating the canvas in realtime performs poorly on my system; however the GIF that is saved to file plays back at the expected frames per second.
-
Based on my understanding, an M-Sequence generates N = m*2-1 signal bits. To create an NxN image, multiple (N) sequences should be generated with different seeds then stacked to form the image.
- The NoiseGrid generator uses the generic NoiseGenerator interface, so different subclasses may be easily implemented and injected. New generators should also be added to the factory registry.
-
The metadata schema should be probably split into several sub schemas, corresponding to each component.
- Using pydantic data models would simplify this validation and schema definition.
-
The m-sequence update function registry and also the NoiseGeneratorFactory are pretty rudimentary (just a dict). Could use a more sophisticated implementation to register them using decorators.
-
More thorough tests: I added the --debug flag to dump full frame *.npy for generating test fixtures.
pip install .[test]
python -m pytest --cov=src/ tests/