An MGF-based Unified Framework to Determine the Joint Statistics of Partial Sums of Ordered Random Variables

Sung Sik Nam, Mohamed-Slim Alouini, Hong-Chuan Yang

Introduction

In modern wireless communications and signal processing, researchers frequently encounter "order statistics." These arise whenever a system selects a subset of available signals based on their strength—for example, choosing the three strongest signals from a ten-antenna array to maximize performance. While the underlying signals may be independent and identically distributed (i.i.d.), the act of sorting them introduces complex mathematical dependencies: the $n$-th strongest signal is, by definition, always greater than or equal to the $(n+1)$-th signal.

Calculating the statistics of the sums of these ordered variables (such as the total power of the top $K_s$ signals) is a notoriously difficult problem. Traditional methods often rely on "successive conditioning," a technique that requires bespoke, case-specific mathematical manipulations that do not generalize well to different configurations. This paper introduces a unified, systematic framework to derive the joint probability density functions (PDFs) for partial sums of ordered random variables using a compact Moment Generating Function (MGF) approach.

The Analytical Framework

The authors propose a two-step analytical "recipe" that simplifies the derivation process:

  1. Derivation of the Joint MGF: The framework first obtains analytical expressions for the joint MGF of the partial sums. To keep these expressions manageable, the authors utilize a technique involving the interchange of multiple integrals. By ensuring that integration limits are "tight"—defined by the most immediate neighboring variables—complex multiple integrals are reduced to compact, solvable forms.
  2. Inverse Laplace Transform: Once the compact MGF is established, an inverse Laplace transform is applied to arrive at the joint PDF.

Handling Discontinuous Sums

A significant challenge in order statistics occurs with "discontinuous" sums, where a group of ordered variables of interest is separated by other variables. The authors solve this by temporarily increasing the dimensionality of the problem. They break the groups into smaller, continuous chunks, derive the higher-dimensional PDF, and then perform finite integration to "collapse" the result back to the desired lower-dimensional form.

The "Common Functions" Building Blocks

To standardize the derivations, the paper introduces three "Common Functions" that serve as the mathematical building blocks for the framework:

  • $c(\gamma, \lambda)$: A mixture of a Cumulative Distribution Function (CDF) and an MGF.
  • $e(\gamma, \lambda)$: A mixture of an Exceedance Distribution Function (EDF) and an MGF.
  • $\mu(\gamma_a, \gamma_b, \lambda)$: An "Interval MGF" for variables falling strictly between two specific bounds.

By using these functions, the framework provides a consistent methodology that can be applied to any combination of ordered random variables, moving away from the "tricks" required by previous literature.

Key Results and Applications

The primary contribution of this research is a set of generalized theorems (Theorems 4.1 through 5.3) that provide the joint PDFs for various partial sum configurations.

Rayleigh Fading and Exponential RVs

For practical engineering applications, the authors provide specific closed-form formulas for i.i.d. exponential random variables. This is particularly valuable because exponential variables are the standard mathematical model for Rayleigh fading in wireless channels. These ready-to-use formulas allow engineers to bypass the complex derivation process for standard communication scenarios.

System Performance Analysis

The paper demonstrates the utility of the framework through two primary application examples:

  • Multiuser Scheduling: Analyzing the Signal-to-Interference-plus-Noise Ratio (SINR) where users are selected based on channel quality.
  • Adaptive Diversity Combining: Evaluating the performance of schemes like Minimum Selection Generalized Selection Combining (MS-GSC), where the number of combined signals changes dynamically.

Limitations and Future Directions

While the framework is a significant step toward a unified theory of order statistics, some challenges remain:

  • Numerical Evaluation: For non-exponential distributions, the final inverse Laplace transform may still require numerical methods, such as Bromwich contour integrals, or specialized software like MATLAB or Mathematica.
  • The I.I.D. Assumption: The current framework assumes that the underlying random variables are independent and identically distributed. Extending these results to non-identically distributed or correlated variables remains an open area for future research.
  • Computational Effort: Cases involving "split groups" still require a final stage of finite integration, which adds a layer of computational complexity, though it remains simpler than traditional successive conditioning methods.

Summary updated

Abstract

Order statistics find applications in various areas of communications and signal processing. In this paper, we introduce an unified analytical framework to determine the joint statistics of partial sums of ordered random variables (RVs). With the proposed approach, we can systematically derive the joint statistics of any partial sums of ordered statistics, in terms of the moment generating function (MGF) and the probability density function (PDF). Our MGF-based approach applies not only when all the K ordered RVs are involved but also when only the Ks (Ks < K) best RVs are considered. In addition, we present the closed-form expressions for the exponential RV special case. These results apply to the performance analysis of various wireless communication systems over fading channels.
BibTeX
@article{nam2010mgf,
  author = {Sung Sik Nam and Mohamed-Slim Alouini and Hong-Chuan Yang},
  title = {{An MGF-based Unified Framework to Determine the Joint Statistics of Partial Sums of Ordered Random Variables}},
  journal = {IEEE Trans. Information Theory},
  year = {2010},
  url = {https://openreview.net/forum?id=YbQ9p8Piso1},
}
Figure