Tools and Reproducibility

Both MATLAB and Python are widely used in wireless research. The choice depends on your group's conventions and the task at hand.

Recommendation: Use whichever tool your collaborators use. For new projects with deep learning components, Python is the de facto standard. For quick link-level BER simulations, MATLAB remains highly productive.

MATLAB ecosystem:

• Communications Toolbox: modulators, channel models, LDPC/turbo codecs
• Phased Array System Toolbox: antenna patterns, beamforming
• 5G Toolbox: NR waveform generation, PDSCH/PUSCH processing
• CVX: disciplined convex programming

Python ecosystem:

• NumPy / SciPy: core numerical computing
• Matplotlib: publication-quality figures
• sionna (NVIDIA): GPU-accelerated link-level simulator with differentiable components
• commpy: basic modulation, coding, channel models
• CVXPY: convex optimization modeling
• PyTorch / TensorFlow: deep learning for communications
• QuaDRiGa (MATLAB/Octave with Python wrapper): 3GPP-compliant channel generation

System-level simulators:

• Vienna 5G SLS (MATLAB): multi-cell, multi-user system simulation
• ns-3 (C++/Python): network-level protocol simulation
• MATLAB 5G Toolbox system-level examples

When reporting results, always specify: tool name, version number, and any non-default settings.

Figures are the most scrutinized part of a paper. Follow these guidelines:

Axes and labels:

• Always label both axes with units
• Use "$E_b/N_0$ (dB)" not just "SNR"
• Log scale for BER ($y$-axis), linear for rate
• Start BER plots at $10^0$ or $10^{-1}$, not at an arbitrary value

Lines and markers:

• Use distinct line styles (solid, dashed, dotted) and markers
• Ensure the figure is readable in grayscale (not just color)
• Analytical curves: solid lines without markers
• Simulation points: markers (possibly with connecting lines)
• When analytical and simulation agree, show both to validate

Legends and annotations:

• Place legend outside the plot or in a clear region
• Label curves directly when possible (less clutter than a legend)
• Annotate key features (crossing points, asymptotic slopes)

Technical details:

• Vector formats (PDF, EPS) for line plots, not raster (PNG, JPEG)
• Font size in figures should match the paper body text
• Use consistent notation between equations and figure labels

Document class: IEEE papers use IEEEtran; journals and conferences have specific templates.

Math conventions:

• Vectors in bold lowercase: \mathbf{h}
• Matrices in bold uppercase: \mathbf{H}
• Sets in calligraphic: \mathcal{S}
• Use \text{SNR} not SNR in equations (upright font for acronyms)

Common packages:

• amsmath, amssymb: essential math environments
• bm: bold math symbols (\bm{\theta})
• algorithm2e or algorithmicx: pseudocode
• booktabs: professional tables (\toprule, \midrule)
• siunitx: consistent unit formatting
• tikz/pgfplots: programmatic figures

Bibliography: Use BibTeX with the IEEE bibliography style. Check that all references have complete metadata (year, volume, pages). ArXiv preprints should include the arXiv ID.

Every wireless researcher should use git for simulation code. Key practices:

• Commit early, commit often: Each experiment should be a tagged commit so you can trace any figure back to exact code.
• Branch for experiments: Create a branch for each new idea; merge to main only when validated.
• Tag submissions: git tag submission-v1 when you submit a paper. This makes camera-ready revisions traceable.
• Never hard-code parameters: Use config files that are committed alongside the code.
• Include plotting scripts: The exact script that generates each figure should be in the repository.
• .gitignore data files: Large data and binary files should not be in git; use git-LFS or external storage.

The combination of a git hash in your paper's simulation section and a publicly available repository is the gold standard for reproducibility.