[Paper Review] Binary Linear Codes with Optimal Scaling and Quasi-Linear Complexity.

Arman Fazeli, S. Hamed Hassani+2 more|arXiv (Cornell University)|Nov 3, 2017

Error Correcting Code Techniques36 references9 citations

TL;DR

This paper introduces binary polar codes with large kernels that achieve optimal scaling exponent μ = 2 on the binary erasure channel (BEC), matching the theoretical limit of random linear codes. By constructing and analyzing ℓ×ℓ binary kernels, the authors prove that polar codes can attain both optimal scaling and quasi-linear complexity O(n log n) for encoding and decoding, significantly improving upon conventional polar codes with μ = 3.63.

ABSTRACT

We prove that, for the binary erasure channel (BEC), the polar-coding paradigm gives rise to codes that not only approach the Shannon limit but do so under the best possible scaling of their block length as a~function of the gap to capacity. This result exhibits the first known family of binary codes that attain both optimal scaling and quasi-linear complexity of encoding and decoding. Our proof is based on the construction and analysis of binary polar codes with large kernels. When communicating reliably at rates within $\varepsilon > 0$ of capacity, the code length $n$ often scales as $O(1/\varepsilon^{\mu})$, where the constant $\mu$ is called the scaling exponent. It is known that the optimal scaling exponent is $\mu=2$, and it is achieved by random linear codes. The scaling exponent of conventional polar codes (based on the $2 imes 2$ kernel) on the BEC is $\mu=3.63$. This falls far short of the optimal scaling guaranteed by random codes. Our main contribution is a rigorous proof of the following result: for the BEC, there exist $\ell imes\ell$ binary kernels, such that polar codes constructed from these kernels achieve scaling exponent $\mu(\ell)$ that tends to the optimal value of $2$ as $\ell$ grows. We furthermore characterize precisely how large $\ell$ needs to be as a function of the gap between $\mu(\ell)$ and $2$. The resulting binary codes maintain the recursive structure of conventional polar codes, and thereby achieve construction complexity $O(n)$ and encoding/decoding complexity $O(n\log n)$.

Motivation & Objective

• To close the gap between conventional polar codes and the theoretical capacity limit of binary linear codes on the BEC.
• To achieve optimal scaling exponent μ = 2, which is the best possible scaling for binary codes.
• To maintain low encoding and decoding complexity while approaching the Shannon limit.
• To characterize the required kernel size ℓ to achieve a desired scaling exponent μ(ℓ) close to 2.
• To prove that polar codes with large kernels preserve the recursive structure enabling O(n) construction and O(n log n) complexity.

Proposed method

• The authors construct binary polar codes using ℓ×ℓ kernels instead of the standard 2×2 kernel.
• They analyze the scaling exponent μ(ℓ) of these codes on the BEC using recursive channel polarization techniques.
• The method involves proving that μ(ℓ) → 2 as ℓ increases, using rigorous analysis of kernel transformation properties.
• The recursive structure of the codes is preserved, enabling efficient O(n) construction and O(n log n) encoding/decoding.
• The authors derive explicit bounds on how large ℓ must be to achieve a given μ(ℓ) within ε of the optimal μ = 2.
• The analysis relies on the behavior of kernel-induced channel transformations and their convergence to the optimal polarization pattern.

Experimental results

Research questions

• RQ1 Can binary polar codes with large kernels achieve the optimal scaling exponent μ = 2 on the BEC, matching random linear codes?
• RQ2 What is the required kernel size ℓ to achieve a scaling exponent μ(ℓ) arbitrarily close to 2?
• RQ3 Do large-kernel polar codes maintain the quasi-linear complexity of conventional polar codes?
• RQ4 How does the scaling exponent μ(ℓ) of large-kernel polar codes depend on the kernel size ℓ?
• RQ5 Can the recursive structure of polar codes be preserved while improving scaling performance?

Key findings

• The scaling exponent μ(ℓ) of polar codes constructed with ℓ×ℓ kernels approaches the optimal value of 2 as ℓ increases.
• For any ε > 0, there exists a finite ℓ such that μ(ℓ) ≤ 2 + ε, proving convergence to the optimal scaling.
• The code length n scales as O(1/ε^μ) with μ(ℓ) → 2, achieving the best possible scaling for binary codes.
• The construction maintains O(n) complexity for code generation and O(n log n) complexity for encoding and decoding.
• The paper provides explicit bounds on ℓ as a function of the desired proximity to μ = 2.
• This is the first known family of binary linear codes that simultaneously achieve optimal scaling and quasi-linear complexity on the BEC.