TorchSig v2.1.0

What's Changed

• Major restructure of DatasetMetadata (removed) and SignalBuilder (renamed to BaseSignalGenerator) to allow for more flexible signal types, using HierarchicalMetadataObject class to handle metadata across dataset and signal generators
• Removed models and image_datasets
  • torchsig-models repository to be released soon
• Combined StaticDataset and CustomDataset
• Code cleanup and optimizations from v2.0.0
  • cleaned up unused libraries, upgrade to numpy > 2
  • 411 bug readthedocs errors by @ereoh in #412
  • removing zarr imports by @MattCarrickPL in #414
• v2.1.0 Release by @TorchDSP in #448

Full Changelog: v2.0.0...v2.1.0

TorchSig v2.0.0

What's Changed

This release has major structural changes, see the TorchSig v1.1.0 -> v2.0.0 Migration Guide for more details.

Additionally, we have released a TorchSig-GUI (in beta).

• Datasets
  • Fixed: TorchSigIterableDataset num_samples
  • New: Custom Datasets to use user data within TorchSig
  • Update: Unified Narrowband and Wideband
• Transforms
  • Fixed: IQImbalance
  • New:
    • Clock Jitter, Drift
    • Rust resampler
  • Update: Level 2 impairments, TargetTransform replaced with MetadataTransform
• Signals
  • Update: Signal, DatasetSignal, DatasetDict restructured into Signal
• Utils
  • New: Zarr replace by HDF5
  • Update: Restructured DatasetCreator and File Handler class
  • WorkingSeedingDataLoader guarantees reporoducibility for DataLoaders
• Miscellaneous
  • Added MIT license
  • Updated Dockerfule
• General bug fixes
• forcing specific numcodecs version by @MattCarrickPL in #385
• 2.0.0 dev by @ereoh in #410

New Contributors

• @romanziske help with SigMF file reader in #301

Full Changelog: v1.1.0...v2.0.0

TorchSig v1.1.0

What's Changed

• fix: #381
• fix: include YAML config files in package installation by @romanziske in #301
• re-enabling seed param, fixes notebook crash on dataset generation by @MattCarrickPL in #308
• 1.1.0 dev by @ereoh in #375
• Signal generation bug fixes and optimizations
  • #369
  • #351
  • #316
  • #315
  • #314
  • #313
  • #312
  • #311
  • #299
  • #293
• Remaining transforms implemented in #350
• gr-spectrumdetect updated in #326, #294
• Dataset writing optimizations (zarr) in #364, #328, #287
• Dataset YAML example in #307
• Dataset reproducibility fix in #345, #286
• PNG/JPG capabilit in #368, #284
• General bug fixes
  • #363
  • #345
  • #320
  • #319
  • #310
  • #300
  • #296
  • #355

New Contributors

• @romanziske made their first contribution in #301

Full Changelog: v1.0.0...v1.1.0

TorchSig v1.0.0

What's Changed

• v1.0.0-beta release by @ereoh in #273
• fixing f-string print problem in notebook by @MattCarrickPL in #276
• Fix synthetic spectrogram notebook, add min_num_signals to generate scripts by @ereoh in #278
• sync main with 1.0.0-dev by @MattCarrickPL in #295

Full Changelog: v0.6.1...v1.0.0

TorchSig v1.0.0-beta

Since this is almost an entirely new version of TorchSig, it is not backwards compatible with previous versions.

What's Changed (a lot!)

tldr; Look at our example notebooks, and start with getting_started.ipynb

• Datasets
  • Dataset configuration with new class DatasetMetadata, which holds configuration information about dataset
    • supports YAML configuration
    • Official Narrowband and Wideband datasets are define in datasets/default_configs with loaders
  • Two types of dataset classes: NewDataset and StaticDataset
    • NewDataset support infinite or finite signal generation, but does not store samples in memory
      • NewNarrowband, NewWideband
    • StaticDataset supports loading a dataset from disk
      • StaticNarrowband, StaticWideband
  • Datamodules for custom Narrowband/Wideband dataset and Official Narrowband/Wideband datasets
• Signals
  • New signal classes: DatasetSignal, DatasetDict.
  • SignalMetadata redefined as class instead of TypedDict.
  • SignalBuilder for signal generation.
  • Signal lists variable names updated.
• Transforms
  • Transforms redefined as changing signal data and signal metadata
    • Grouped into three categories: SignalTransform, DatasetTransform, and Impairments.
    • Many transforms refactored, optimized, or removed.
  • Target transforms redefined as label/output creators, do not change signal data at all
    • Can ask for any configuration of labels
• Utils
  • Random class Seedable to manage randomness across all TorchSig objects
  • Various printing and verification code
  • Dataset writing and reading using Zarr and YAML
    • New class: DatasetCreator
• Examples
  • Examples no longer need to be run in any specific order.
  • Moved Jupyter annotation tool example to tools/examples.
• Tests
  • More comprehensive test suit with Pytest.
• Google-style docstrings have been added for most of the code.
  • Updated documentation.

What Has Not Changed

• gr-spectrumdetect
• image_datasets

Notes

• v0.6.1 pretrained models are out of date.
• OFDM signal not fully featured
  • 0.6.1 OFDM has a simulated protocol on it: Resource blocks, pilots, etc. that does not exist in 1.0 yet.
  • 0.6.1 OFDM has some windowing in the time domain acting as pulse shaping, that does not exist in 1.0 yet
  • Otherwise the OFDM is accurate and representative, it is idealized and textbook-like, just as our other signals are
• We are depreciating our support of models. Specifically, we plan on only maintaining our pretrained models.
  • Our code should be compatible with any PyTorch compatible model, or should be easy to make compatible.
  • Still keep example notebooks to show how to train and use a model with our datasets.
• gr-spectrumdetect had not been updated with most recent code.

Known Bugs

• NewDatasets are not perfectly reproducible.
  • To guarantee reproducibility, we recommend writing the dataset to disk and loading back in the ensure correctness of samples.

Open-Source Collaboration

• We would really like to involve the open-source community, and are moving development to Github.
• Please feel free to make contributions towards bugs, new features, current issues, or anything on our Roadmap on the wiki.
• Github workflows for Pylint and Pytest to merge with main:
  • Pylint score must be above 9/10
  • Pytests should all pass without error

TorchSig v0.6.1

What's Changed

• fix ReadTheDocs pipeline by @ereoh in #251
• Fix for #254 -- accept dict dtype in is_signal by @IsaiahHarvi in #255
• v0.6.1 release by @ereoh @MattCarrickPL in #267
• Releases
  • example timm model loading, example notebook 11
  • example sigmf loading, example notebook 12 (#257)
  • new pretrained models for narrowband and wideband with example notebook 13
• Bugs with example notebooks 6-10, 13
• remove Numpy and OpenCV constraints

New Contributors

• @IsaiahHarvi contributed in #255
• @romanziske contributed in #261
• @agm-eratosth contributed in #248
• @sciafri contributed in #247
• @jogomojo contributed in #246

Full Changelog: v0.6.0...v0.6.1

TorchSig v0.6.0

• New signals added (53 → 61 total signals)
  • New signals:
    • FM (frequency modulation)
    • AM-DSB (amplitude modulation, double side band, with carrier)
    • AM-DSB-SC (amplitude modulation, double side band, suppressed carrier)
    • AM-LSB (amplitude modulation, lower side band)
    • AM-USB (amplitude modulation, upper side band)
    • Chirp Spread Spectrum
    • Linear Frequency Modulated Data (LFM-Data, both up and down chirps)
    • Linear Frequency Modulated Radar (LFM-Radar, up-chirps only)
• Refactored 53 out of dataset names and codebase
  • Sig53 → Narrowband
  • WidebandSig53 → Wideband
• Datasets, transforms, example notebooks, ect. updated with new classes listed in torchsig/datasets/signal_classes.py
• Removed iq_samples_per_symbol from metadata
• Optuna integration
  • examples/08_example_optuna_yolo.ipynb
• Synthetic spectrogram image dataset improvements
  • examples/09_example_synthetic_spectrogram_dataset.ipynb: Generating and using synthetic spectrograms.
  • examples/10_example_yolo_annotation_tool.ipynb: New feature to draw SOI bounding boxes in Jupyter notebooks.
  • Spectrogram wideband texturing
• Bug Fixes
  • Narrowband dataset crashing due to Transforms
  • RandomResample
  • Frequency Hopper
  • Normalize
  • Transition bandwidth calculations
• Documentation
  • gr-spectrumdetect README
    • detect.pt also attached to release if needed.
  • README has minimum/recommended specifications
• Python >= 3.9

TorchSig v0.5.3.1

• Fix Github LFS bandwidth limits with gr-spectrumdetect/examples/detect.pt file

TorchSig v0.5.3

Closed the following issues:

1. Fixed: incorrect bounding boxes, too large in frequency, both G/FSK and others.
2. Fixed: Reports of problem that DescToList tuple transform has a problem in type conversion, being a list vs list-of-lists. Says this causes a crash somewhere in one of the example notebooks.
3. Fixed: "rearranged the order in which meta['start'] and meta['stop'] values were being updated in relation to meta['num_samples'] it was causing a new_rate **2 change to the new_start and new_stop variables"
4. Fixed: impaired dataset generation bug - RandomTimeShift resulted in signals being shifted "out of bounds"
5. Validate all notebooks are working properly
6. Fixed: frequency hopper functionality.
7. Fixed: OFDM modulator creating incorrect bounding box on mc/deterministic-modulators.
8. Fixed: OFDM bandwidth wrap around +-fs/2 boundary
9. Upgrade: replace sp.resample_poly() within synthetic.py with a higher quality resampler
10. Upgrade: PFB resamplers need to use fred harris approximation for number of branches
11. Released: gr-spectrumdetect: The out-of-tree (OOT) module gr-spectrumdetect incorporates a YOLOv8x TorchSig ML model to detect signals in real time.
12. Upgrade/Release: 05_example_wideband_yolo_to_disk.ipynb: This notebook showcases using the WBSig53 dataset to train a YOLOv8 model.
13. Release: 06_example_wideband_yolo.ipynb: This notebook showcases using the WBSig53 dataset to train a YOLOv8 model.
14. Release: 07_example_classify_yolo.ipnyb: This notebook showcases using the Sig53 dataset to train a YOLOv8 classification model.

@pvallance, @MattCarrickPL

v0.5.2.1

1. Merge issue with previous release: Fixed pickle issue causing OOM issues. No longer stores tensors to db
2. Removed unused print statements add collate_fn in generate_sig53.py to avoid torch Dataloader from converting the labels to tensors.

@pvalance