Analysis and recovery of IBM MFM floppy image data
This is where I experiment with methods to more reliably decode IBM MFM floppies based on flux transition data.
There are two source subdirectories: csrc/, the C++ version, and python/, the Python version. I'm currently working on the C++ version; the Python version is older.
I may be picking up this project again, time permitting, as I've thought of new ideas that could handle errors more robustly. If you have got any marginal or failing floppy images that you'd be willing to share (without PII or other private data), let me know by issue/PR or by email to my commit Author mail address. Such images could help me tune the algorithms so that they'll have better success at recovering data.
To compile the C++ version, use cmake. A typical invocation would be as follows:
$ cd ~/Sources/
$ git clone https://github.com/kristomu/flux-analyze.git
$ cd flux-analyze
$ mkdir build
$ cd build
$ cmake -S ../csrc -B .
$ make
Then use by invoking flux-analyze by specifying one or more FluxEngine .flux
filenames on the command line. If more than one flux file is specified, they'll
be treated as different images of the same floppy.
To use the Python version, run mfm_decoder interactively in ipython, i.e.
ipython3 -i mfm_decoder.py, then invoke demonstrate("your-file.au"). The
Python version uses FluxEngine .au dumps as input. The README.md file in the
Python subdirectory contains more information.
The tracks/ directory contains some sample tracks in FluxEngine and .au format. The .au files are all compressed with bz2. (Uncompress before using.)
- low_level_format_with_noise: every sector is 0x00 or 0xF6. Some weird effects on the flux transitions in certain places. Useful for testing recovery ideas later, because the data pattern on the corrupted sectors is known.
- MS_Plus_OK_track - OK track from a Microsoft Plus! floppy. This one can be decoded fine with FluxEngine.
- MS_Plus_warped_track - Warped track from a Microsoft Plus! floppy. FluxEngine can't cleanly decode every sector, but both flux-analyze versions can.