Authors: T.S. Jayram and Erdal Arıkan
Abstract
This technical note provides a unified treatment of the fundamental inequalities used in the proofs of channel polarization theorems for binary-input discrete memoryless channels (B-DMC). By analyzing the properties of a specific mapping function, the authors establish tight bounds relating a channel's symmetric capacity to its reliability, simplifying the mathematical foundations of polar coding theory.
Introduction
Polar coding, introduced by Erdal Arıkan, represents a landmark achievement in information theory as the first constructive method to achieve the capacity of symmetric binary-input discrete memoryless channels. The efficacy of polar codes relies on the phenomenon of "channel polarization," where a sequence of channel transformations results in a set of bit-channels that are either nearly noiseless or completely noisy.
To rigorously prove that this polarization occurs, researchers must track the evolution of two primary channel metrics:
- Symmetric Capacity $I(W)$: A measure of the maximum rate at which information can be reliably transmitted.
- Bhattacharyya Parameter $Z(W)$: An upper bound on the probability of a decoding error, representing the channel's reliability.
While various inequalities relating these two parameters were previously known, they were often derived through disparate and indirect methods. This paper addresses this fragmentation by providing a single, elegant mathematical framework to derive these essential bounds.
The Unified Framework: The $\phi$ Function
The authors center their analysis on a specific "bridge" function, $\phi(u)$, which translates between the entropy-based world of capacity and the reliability-based world of the Bhattacharyya parameter. The function is defined as:
$$\phi(u) = \mathcal{H}\left(\frac{1-\sqrt{1-u^2}}{2}\right)$$
where $\mathcal{H}$ is the binary entropy function. The methodological core of the paper involves treating the channel as a random variable—specifically using the Blackwell measure. This perspective allows $I(W)$ and $Z(W)$ to be expressed as expectations of functions of this variable.
By employing calculus-based analysis (including Taylor series expansions of inverse hyperbolic functions), the authors prove that $\phi(u)$ is strictly convex and that $\phi(\sqrt{w})$ is strictly concave. Applying Jensen’s Inequality to these properties allows for the derivation of global bounds that apply to any binary-input channel.
Key Results and Insights
The primary contribution of the paper is Theorem 4, which establishes the fundamental "guardrails" for channel polarization:
$$Z(W) \geq 1 - I(W) \geq \phi(Z(W))$$
These bounds are shown to be "tight," meaning the equality holds for the two most significant channel models in coding theory:
- Binary Erasure Channel (BEC): Represents the extreme where $I(W) + Z(W) = 1$.
- Binary Symmetric Channel (BSC): Represents the extreme where $I(W) + \phi(Z(W)) = 1$.
Furthermore, the authors demonstrate a cross-disciplinary link, showing that these information-theoretic inequalities are equivalent to bounds between the Hellinger distance and the Jensen-Shannon divergence, which are widely used in statistics and machine learning to compare probability distributions.
Significance and Limitations
By unifying these inequalities, the authors provide the mathematical assurance that polar codes will converge to the correct limits as block lengths increase. This note simplifies the pedagogical and theoretical approach to understanding why polarization is an intrinsic property of binary channels.
However, as a theoretical note, the scope is intentionally narrow:
- Input Alphabet: The analysis is strictly limited to binary-input channels. Extending these unified inequalities to non-binary or continuous-input channels remains an open area for research.
- Practical Implementation: The paper focuses on the existence of bounds rather than the algorithmic complexity of calculating these parameters in real-world 5G hardware implementations.
- Optimization: While these inequalities are vital for proofs, they do not directly address the practical construction of polar codes for specific finite block lengths.
In conclusion, Jayram and Arıkan provide a concise and powerful mathematical toolset that reinforces the theoretical stability of polar coding, grounding the complex phenomenon of channel polarization in the fundamental properties of convex functions.
Summary updated