Power Allocation for the Base Matrix of Spatially Coupled Sparse Regression Codes

Nian Guo, Shansuo Liang, Wei Han

Abstract

We investigate power allocation for the base matrix of a spatially coupled sparse regression code (SC-SPARC) for reliable communications over an additive white Gaussian noise (AWGN) channel. A conventional SC-SPARC allocates power uniformly to the non-zero entries of its base matrix. Yet, to achieve the channel capacity with uniform power allocation, the coupling width and the coupling length of the base matrix must satisfy regularity conditions and tend to infinity as the rate approaches the capacity. For a base matrix with a pair of finite and arbitrarily chosen coupling width and coupling length, we propose a novel power allocation policy, termed V-power allocation (VPA). VPA puts more power to the outer columns of the base matrix to jumpstart the decoding process and less power to the inner columns, resembling the shape of the letter V. We show that VPA outperforms uniform power allocation since it ensures successful decoding for a wider range of signal-to-noise ratios given a code rate in the limit of large blocklength. In the finite blocklength regime, we show by simulations that power allocations imitating the shape of the letter V improve the error performance of a SC-SPARC.

Introduction

Sparse Regression Codes (SPARCs) have emerged as a powerful class of error-correction codes designed for the Additive White Gaussian Noise (AWGN) channel. When combined with spatial coupling—a technique that links multiple code blocks together in a chain-like structure—they can theoretically reach the Shannon capacity, the absolute limit of communication efficiency. This phenomenon, known as threshold saturation, allows the performance of a low-complexity belief propagation decoder to match that of an optimal maximum-likelihood decoder.

However, a significant gap exists between theoretical ideals and practical engineering. Standard theory assumes an infinitely long coupling chain and infinitely many connections (coupling width). In real-world applications, these parameters must be finite. When using standard Uniform Power Allocation (UPA), where every section of the message receives equal power, the decoding process often fails to initiate in these finite-length scenarios. This paper addresses this bottleneck by proposing a novel power allocation strategy for the base matrix of SC-SPARCs.

The Problem with Uniform Power

Decoding in spatially coupled codes relies on a "sliding window" or "wave" effect. The process typically begins at the boundaries (the edges of the chain) and moves inward. These boundaries act as a "seed"; once the edges are successfully decoded, they provide side information that lowers the decoding threshold for adjacent sections.

Under UPA, the power assigned to these boundary sections is often insufficient to trigger the initial decoding step unless the coupling width and length are extremely large. If the boundaries do not have enough "oomph" to decode first, the entire message remains stuck, leading to a high Block Error Rate (BLER) even at relatively high signal-to-noise ratios (SNR).

V-Power Allocation (VPA)

To overcome the limitations of UPA, we propose V-power allocation (VPA). Instead of spreading power evenly across the base matrix, VPA concentrates more power at the outer columns and less power in the middle, creating a power profile that resembles the letter V.

The VPA Algorithm

The design of VPA is driven by an iterative process using State Evolution (SE), a mathematical tool that predicts the behavior of the decoder in the limit of large section lengths. The algorithm determines the minimum power required for each section of the code to ensure that the decoding wave successfully propagates from the edges to the center. By strengthening the "seed" at the boundaries, VPA ensures that the decoding process is "jumpstarted" even when the coupling parameters are finite and arbitrarily chosen.

Key Insights

  1. Symmetry and Efficiency: The optimal power allocation for diagonal-band base matrices is found to be symmetric around the center. By optimizing the power distribution, we define a Power-Rate Function (PRF) that quantifies the minimum power needed to achieve a specific data rate.
  2. Theoretical Superiority: We mathematically prove that the PRF of VPA is always lower than or equal to that of UPA ($P_V(R) \leq P_U(R)$). This means VPA can guarantee successful decoding at lower power levels for the same data rate.
  3. Implementation Advantage: Unlike previous iterative power allocation methods that vary power section-by-section within a specific message, VPA modifies the base matrix itself. This makes it easier to implement during the hardware design phase, as the power profile is built into the code's blueprint.

Experimental Results

We evaluated the performance of VPA using both numerical state evolution and Monte Carlo simulations on synthetic AWGN channels. To manage computational complexity, Hadamard design matrices were utilized in the simulations.

  • Error Performance: In the finite-blocklength regime (with parameters such as section length $M=512$ and coupling length $L_C=15$), VPA consistently achieved a lower BLER than UPA at SNRs between 9.5dB and 12.0dB.
  • Convergence Speed: Simulations demonstrated that VPA-coded messages converged to a successful decode in fewer iterations compared to UPA at identical signal strengths, highlighting the efficiency of the V-shaped "seed" effect.

Conclusion and Future Work

V-power allocation provides a robust solution for making SC-SPARCs practical for real-world communication systems. By breaking the "uniformity myth," we show that non-uniform power distribution is essential for finite-length spatially coupled codes.

While VPA shows significant gains, future research is needed to address the Peak-to-Average Power Ratio (PAPR) in physical radio hardware, as varying power levels can impact amplifier efficiency. Additionally, while the theory holds for large section lengths, further optimization for very small blocklengths remains an open and promising area of study.

Summary updated

Abstract

We investigate power allocation for the base matrix of a spatially coupled sparse regression code (SC-SPARC) for reliable communications over an additive white Gaussian noise channel. A conventional SC-SPARC allocates power uniformly to the non-zero entries of its base matrix. Yet, to achieve the channel capacity with uniform power allocation, the coupling width and the coupling length of the base matrix must satisfy regularity conditions and tend to infinity as the rate approaches the capacity. For a base matrix with a pair of finite and arbitrarily chosen coupling width and coupling length, we propose a novel power allocation policy, termed V-power allocation. V-power allocation puts more power to the outer columns of the base matrix to jumpstart the decoding process and less power to the inner columns, resembling the shape of the letter V. We show that V-power allocation outperforms uniform power allocation since it ensures successful decoding for a wider range of signal-to-noise ratios given a code rate in the limit of large blocklength. In the finite blocklength regime, we show by simulations that power allocations imitating the shape of the letter V improve the error performance of a SC-SPARC.
BibTeX
@inproceedings{guo2023power,
  author = {Nian Guo and Shansuo Liang and Wei Han},
  title = {{Power Allocation for the Base Matrix of Spatially Coupled Sparse Regression Codes}},
  booktitle = {ISTC},
  year = {2023},
  doi = {10.1109/istc57237.2023.10273571},
}
Figure