[Paper Review] A Sum-Product Model as a Physical Basis for Shadow Fading

Jari Salo|ArXiv.org|Feb 18, 2007

Millimeter-Wave Propagation and Modeling23 references3 citations

TL;DR

This paper proposes a sum-product signal model as a unified physical basis for shadow fading in mobile radio channels, demonstrating through simulations that it naturally produces log-normal distributed local mean power regardless of underlying interaction distributions. The model improves upon conventional product and sum models by being physically plausible, applicable across global and local distance scales, and achieving superior goodness-of-fit to log-normal statistics with fewer layers (3–5) than required by the product model.

ABSTRACT

Shadow fading (slow fading) effects play a central role in mobile communication system design and analysis. Experimental evidence indicates that shadow fading exhibits log-normal power distribution almost universally, and yet it is still not well understood what causes this. In this paper, we propose a versatile sum-product signal model as a physical basis for shadow fading. Simulation results imply that the proposed model results in log-normally distributed local mean power regardless of the distributions of the interactions in the radio channel, and hence it is capable of explaining the log-normality in a wide variety of propagation scenarios. The sum-product model also includes as its special cases the conventional product model as well as the recently proposed sum model, and improves upon these by: a) being applicable in both global and local distance scales; b) being more plausible from physical point of view; c) providing better goodness-of-fit to log-normal distribution than either of these models.

Motivation & Objective

• To address the long-standing lack of a physically plausible explanation for the universal log-normal distribution of shadow fading in mobile radio channels.
• To overcome limitations of the conventional product model, which requires unrealistically high numbers of interactions (30–100) and assumes identical attenuation for all multipaths.
• To extend the applicability of the sum model, which explains local shadow fading but fails at global scales, by integrating it into a broader framework.
• To unify existing models under a single signal model that explains log-normality across diverse propagation environments.

Proposed method

• Formulates a sum-product signal model where signal propagation is represented as a cascade of random coupling matrices (multiplicative layers) followed by a summation over plane waves.
• Models the received signal power as a quadratic form involving random matrices and vectors representing channel interactions.
• Uses Monte Carlo simulations to evaluate the distribution of locally averaged power across varying numbers of rays (N), interaction distributions (beta, R, L), and coupling matrix sizes.
• Applies the Kolmogorov-Smirnov test to assess goodness-of-fit between simulated power distributions and the log-normal distribution.
• Compares the sum-product model’s performance against the conventional product model and the sum model in terms of convergence speed and distribution fit.
• Analyzes the impact of distribution parameters and matrix size on convergence to log-normality.

Experimental results

Research questions

• RQ1 Can a unified sum-product signal model explain the log-normal distribution of shadow fading across both local and global propagation scales?
• RQ2 How does the sum-product model’s convergence to log-normality compare to that of the conventional product and sum models in terms of required number of interactions?
• RQ3 Does the sum-product model remain robust to variations in the statistical distribution of channel interactions (e.g., beta, Rayleigh, Laplace)?
• RQ4 What physical advantages does the sum-product model offer over the product model, particularly in terms of plausibility and multipath representation?
• RQ5 Can the sum-product model explain the observed independence of shadow fading standard deviation from distance, contrary to the product model’s prediction?

Key findings

• The sum-product model achieves a good fit to the log-normal distribution with only 3–5 layers of channel interactions, significantly fewer than the 20–100 required by the conventional product model.
• The model exhibits convergence to log-normality regardless of the underlying distribution of channel interactions (beta, Rayleigh, Laplace), indicating broad applicability.
• The standard deviation of the simulated power distribution decreases with increasing N, indicating a balancing effect between averaging and nonlinearity in the model.
• The sum model emerges as a special case of the sum-product model when intermediate propagation processes are constant and only plane wave powers vary at the receiver.
• The sum-product model provides a better goodness-of-fit to log-normal statistics than either the product or sum model alone, especially in global-scale scenarios.
• The model resolves inconsistencies in the product model, such as distance-dependent standard deviation and identical attenuation assumptions across all paths, by allowing independent multipath attenuation and realistic path loss scaling.