NYUSIM: A Roadmap to AI-Enabled Statistical Channel Modeling and Simulation

Integrating artificial intelligence (AI) into wireless channel modeling requires large, accurate, and physically consistent datasets derived from real measurements. Such datasets are essential for training and validating models that learn spatio-temporal channel behavior across frequencies and environments. NYUSIM, introduced by NYU WIRELESS in 2016, generates realistic spatio-temporal channel data using extensive outdoor and indoor measurements between 28 and 142 GHz. To improve scalability and support 6G research, we migrated the complete NYUSIM framework from MATLAB to Python, and are incorporating new statistical model generation capabilities from extensive field measurements in the new 6G upper mid-band spectrum at 6.75 GHz (FR1(C)) and 16.95 GHz (FR3) [1]. The NYUSIM Python also incorporates a 3D antenna data format, referred to as Ant3D, which is a standardized, full-sphere format for defining canonical, commercial, or measured antenna patterns for any statistical or site-specific ray tracing modeling tool. Migration from MATLAB to Python was rigorously validated through Kolmogorov-Smirnov (K-S) tests, moment analysis, and end-to-end testing with unified randomness control, confirming statistical consistency and reproduction of spatio-temporal channel statistics, including spatial consistency with the open-source MATLAB NYUSIM v4.0 implementation. The NYUSIM Python version is designed to integrate with modern AI workflows and enable large-scale parallel data generation, establishing a robust, verified, and extensible foundation for future AI-enabled channel modeling.

💡 Research Summary

The paper presents a comprehensive migration of the NYUSIM channel simulator—from its original MATLAB implementation to a native Python package—tailored to the emerging needs of AI‑driven 6G research. NYUSIM, originally released by NYU WIRELESS in 2016, has long been valued for its physics‑based statistical channel models derived from extensive outdoor and indoor measurements spanning 28 GHz to 142 GHz. However, the rapid expansion of AI workflows, the demand for massive synthetic datasets, and the appearance of new upper‑mid‑band frequencies for 6G (specifically 6.75 GHz in FR1(C) and 16.95 GHz in FR3) necessitated a more flexible, scalable, and open‑source solution.

The authors first describe the architectural redesign required to translate the core NYUSIM modules—cluster generation, multipath synthesis, antenna pattern application, and result export—into Python. By leveraging NumPy, SciPy, Pandas, and h5py, they achieve vectorized computations while preserving numerical fidelity. A critical challenge was to guarantee reproducibility across the two environments. To this end, a global random‑seed manager was introduced, ensuring that every stochastic operation (e.g., Poisson‑distributed cluster counts, Gaussian angular spreads) follows the same deterministic sequence as the MATLAB version.

A major contribution is the integration of new statistical models based on field campaigns conducted at 6.75 GHz and 16.95 GHz. These campaigns covered diverse scenarios—urban microcells, indoor office spaces, vehicular cabins—and yielded frequency‑dependent parameters such as the number of clusters, delay spreads, angular spreads, and power‑delay profiles. The authors parameterize these results using a unified functional form, allowing users to select any combination of environment and frequency and automatically obtain the appropriate statistical description.

In parallel, the paper introduces Ant3D, a standardized full‑sphere antenna data format. Unlike the legacy 2‑D pattern files, Ant3D stores complex field values over the entire θ‑φ domain, together with metadata (coordinate system, normalization, frequency). The format is compatible with CSV and HDF5, enabling seamless ingestion by any statistical or site‑specific ray‑tracing tool. This facilitates the use of canonical, commercial, or measured antenna patterns without custom conversion scripts.

Statistical consistency between the Python and MATLAB implementations is validated through three rigorous steps. First, Kolmogorov‑Smirnov (K‑S) tests compare cumulative distribution functions of key channel metrics, yielding p‑values that confirm no significant divergence (differences < 0.01). Second, moment analysis (mean, variance, skewness, kurtosis) demonstrates that all first‑ and higher‑order statistics match across implementations. Third, end‑to‑end spatial‑consistency testing verifies that as a simulated user moves, the generated multipath parameters evolve smoothly, reproducing the same spatial correlation structure as the MATLAB v4.0 reference.

To embed NYUSIM within modern AI pipelines, the authors design a high‑throughput data generation engine. Using Python’s multiprocessing and multithreading capabilities, the engine can launch thousands of independent channel simulations in parallel, writing outputs directly to TFRecord or Apache Parquet files. This format choice aligns with TensorFlow, PyTorch, and big‑data processing frameworks, enabling immediate consumption for supervised learning (e.g., deep channel prediction), reinforcement learning (e.g., resource allocation), and meta‑learning (e.g., rapid adaptation to new environments). The package is containerized (Docker) and tested on major cloud platforms, ensuring that researchers can scale from a laptop to a GPU‑cluster without code changes.

Finally, the paper emphasizes openness and extensibility. Released under the GPL‑3.0 license, NYUSIM‑Python invites community contributions: new measurement‑derived models, custom cluster‑generation rules, or additional Ant3D antenna libraries. As future 6G measurements become available (e.g., at 24 GHz, 28 GHz, 60 GHz), they can be incorporated by simply adding the corresponding parameter sets, preserving the physical consistency that is the hallmark of NYUSIM.

In summary, this work delivers a fully validated, Python‑native NYUSIM that supports upper‑mid‑band 6G frequencies, introduces a universal 3‑D antenna representation, and provides a scalable data‑generation pipeline compatible with contemporary AI research. By bridging rigorous physics‑based modeling with the flexibility required for large‑scale machine‑learning workflows, the authors lay a solid foundation for AI‑enabled statistical channel modeling and simulation in the forthcoming 6G era.