Abstract
(Partial) Unit Memory ((P)UM) codes provide a powerful possibility to construct convolutional codes based on block codes in order to achieve a high decoding performance. In this contribution, a construction based on Gabidulin codes is considered. This construction requires a modified rank metric, the so-called sum rank metric. For the sum rank metric, the free rank distance, the extended row rank distance and its slope are defined analogous to the extended row distance in Hamming metric. Upper bounds for the free rank distance and the slope of (P)UM codes in the sum rank metric are derived and an explicit construction of (P)UM codes based on Gabidulin codes is given, achieving the upper bound for the free rank distance.
Introduction
In traditional network coding, data is often sent in discrete blocks, and error control is handled using rank-metric block codes like Gabidulin codes. However, in "multi-shot" network coding scenarios—where a network is used multiple times to transmit a sequence of blocks—treating each block independently is inefficient. Convolutional codes are better suited for sequences because they create dependencies between successive blocks, allowing the decoder to use temporal context to correct errors.
This paper addresses the lack of structured, high-performance convolutional codes for rank-based environments. The authors propose a new construction for convolutional codes based on Gabidulin codes, specifically designed for the sum rank metric.
Methodology
The authors focus on (Partial) Unit Memory codes, which are convolutional codes with a memory of one ($m=1$). This simplicity makes them algebraically manageable while still providing significant coding gain. The construction follows several key steps:
- Defining the Metric: They utilize the sum rank metric, which calculates the total rank of error matrices across multiple time slots.
- Adapting Distance Measures: They define the "free rank distance" and "slope" (the rate at which error-correcting capability increases with sequence length) specifically for the sum rank metric.
- Algebraic Construction: Instead of using generator matrices, they build the codes using parity-check matrices composed of submatrices from Gabidulin codes. To ensure the resulting matrices have the necessary full-rank properties, they employ normal bases from finite field theory.
Key Results and Insights
The paper establishes that upper bounds for distances in the Hamming metric also serve as upper bounds for the sum rank metric. This allows the authors to leverage classical (P)UM code research to derive the following:
- Upper Bounds: For a UM code, the free rank distance $d_{free}$ is bounded by $2n - k + 1$. For a PUM code, it is $n - k + \nu + 1$.
- Optimal Construction: The authors provide an explicit construction for (P)UM codes with dual memory $m_H = 1$. They prove this construction is optimal in terms of free rank distance, hitting the theoretical upper bound.
- Slope Performance: The proposed construction achieves a slope of $\alpha = (n - k + 1)/2$, which is exactly half of the theoretical maximum. While not "slope-optimal," it represents a significant step forward for rank-metric convolutional codes.
Conclusion
This work bridges the gap between (P)UM codes and Gabidulin codes, providing a robust framework for multi-shot network coding. While the paper focuses on construction and distance properties, future work is required to develop low-complexity decoding algorithms, potentially adapting the Dettmar-Sorger algorithm used for Hamming-metric PUM codes, and to explore generator-based constructions.
Summary updated