Joint Modeling of Received Power, Mean Delay, and Delay Spread for Wideband Radio Channels

Ayush Bharti, Ramoni Adeogun, Xuesong Cai, Wei Fan, Francois-Xavier Briol, Laurent Clavier, Troels PedersenIEEE Transactions on Antennas and Propagation 2021 · OpenReview

TL;DR

This paper introduces a multivariate log-normal distribution to jointly model key characteristics of wideband radio channels: received power, mean delay, and delay spread. By treating these as correlated variables, their model captures the underlying channel behavior more accurately than independent modeling, leading to improved simulations and analysis for wireless communication systems.

What Problem Does This Paper Address?

Characterizing wideband radio channels is crucial for designing effective wireless communication systems. A common approach is to summarize the complex channel behavior using "temporal moments," specifically received power, mean delay, and root mean square (RMS) delay spread. While these moments are known to be correlated in reality, much of the existing research has modeled them independently. This simplification, however, can lead to inaccurate simulations and biased analyses because it ignores the inherent dependencies between these channel parameters. The paper aims to address this gap by proposing and validating a joint statistical model that accounts for these correlations.

Methods

The core of the paper's contribution is the proposal of a multivariate log-normal distribution to model the raw temporal moments (mo, m1, m2) of the received signal. The raw temporal moments are derived from the instantaneous power of the received signal. The standardized temporal moments (received power, mean delay, and RMS delay spread) are then derived from these raw moments. The multivariate log-normal distribution is chosen because it's a natural extension of the log-normal distribution, which has been shown to fit individual temporal moments well, and it allows for analytical tractability.

The authors estimate the parameters of this multivariate log-normal distribution using maximum likelihood estimation on the raw temporal moments. They then use this fitted model to simulate new channel realizations and compare them against the measured data. Model comparison is performed using the Akaike Information Criterion (AIC), a standard metric for evaluating statistical models that balances goodness-of-fit with model complexity.

Datasets

The proposed model was validated using five different radio channel measurement datasets:

  • Lund Data: Indoor, 58-62 GHz band.
  • Lille Data: Indoor, 59-61 GHz band.
  • AAU-Industry: Indoor, 3-8 GHz band.
  • AAU-Hall: Indoor, 28-30 GHz band.
  • AAU-Outdoor: Outdoor, 28-30 GHz band.

These datasets cover various scenarios (indoor/outdoor, different room sizes, industrial environments) and frequency ranges, providing a comprehensive evaluation of the model's robustness.

Key Insights and Ideas

  • Correlation is Key: The most significant insight is that received power, mean delay, and RMS delay spread are not independent. Ignoring this correlation leads to a loss of crucial information about the channel's behavior, as demonstrated by the poorer fits of independent models.
  • Multivariate Log-Normal as a Sweet Spot: The multivariate log-normal distribution offers a good balance between accurately capturing the dependency structure of the temporal moments and maintaining analytical tractability and ease of estimation.
  • Raw Moments are Easier to Model Jointly: While the standardized moments are of direct interest, modeling the raw temporal moments jointly with a multivariate log-normal distribution proved more straightforward and yielded better results than directly modeling the standardized moments.
  • AIC as a Model Selector: The AIC metric effectively highlighted the superiority of joint models over independent ones, even when the independent models provided good fits to the marginal distributions.

Key Results

  • The proposed multivariate log-normal model provided a good fit to the raw temporal moments across all five diverse measurement datasets.
  • In three out of five datasets, the multivariate log-normal model achieved a better joint fit (lower AIC) compared to a multivariate Gaussian model.
  • Crucially, the joint models (both multivariate log-normal and Gaussian) significantly outperformed independent models in terms of AIC, demonstrating the importance of accounting for correlations.
  • The model successfully captured the correlation structure between the standardized temporal moments (received power, mean delay, and RMS delay spread), which independent models failed to do. This is visually evident in the scatter plots.

Limitations and Future Work

  • Marginal Fit vs. Joint Fit: While the AIC is a good overall measure, the paper notes that a good marginal fit (e.g., for individual moments) does not guarantee a good joint fit. The multivariate log-normal model, while good, sometimes had slight deviations in the right tail of the data, which the authors suggest might be acceptable depending on the application.
  • Model Complexity: The authors acknowledge that more complex models, such as copulas, could potentially offer even better fits but would come at the cost of increased complexity and reduced interpretability.
  • Future Work: The paper suggests that further research could explore the dependency of the covariance matrix on physical factors like transmitter-receiver distance and antenna directivity. Understanding these effects is crucial for developing more sophisticated and predictive channel models. The authors also mention that the effects of antenna directivity on the covariance matrix are not yet well-understood and warrant further investigation.

Summary updated

Abstract

We propose a multivariate log-normal distribution to jointly model received power, mean delay, and root mean square (rms) delay spread of wideband radio channels, referred to as the standardized temporal moments. The model is validated using experimental data collected from five different measurement campaigns (four indoor scenarios and one outdoor scenario). We observe that the received power, the mean delay, and the rms delay spread are correlated random variables, and therefore, should be simulated jointly. Joint models are able to capture the structure of the underlying process, unlike the independent models considered in the literature. The proposed model of the multivariate log-normal distribution is found to be a good fit for a large number of wideband data sets.
BibTeX
@article{bharti2021joint,
  author = {Ayush Bharti and Ramoni Adeogun and Xuesong Cai and Wei Fan and Francois-Xavier Briol and Laurent Clavier and Troels Pedersen},
  title = {{Joint Modeling of Received Power, Mean Delay, and Delay Spread for Wideband Radio Channels}},
  journal = {IEEE Transactions on Antennas and Propagation},
  year = {2021},
  url = {https://openreview.net/forum?id=VnPwfacoVs},
}
Figure