A Novel Negative ℓ1 Penalty Approach for Multiuser One-Bit Massive MIMO Downlink with PSK Signaling

Zheyu Wu, Bo Jiang, Ya‐Feng Liu, Yu‐Hong Dai

TL;DR

This paper introduces a novel "negative ℓ₁ penalty" approach to tackle the challenging one-bit precoding problem in massive MIMO systems with PSK modulation. By reformulating the problem into a min-max optimization and employing an efficient alternating optimization algorithm, the proposed method achieves superior bit-error-rate performance with significantly lower computational cost compared to existing state-of-the-art techniques.

What Problem Does This Paper Address?

Modern wireless communication systems, like 5G and beyond, heavily rely on massive MIMO (Multiple-Input Multiple-Output) technology to boost spectrum and energy efficiency. This involves using base stations with a very large number of antennas. However, equipping each antenna with high-resolution digital-to-analog converters (DACs) is prohibitively expensive and power-hungry. A cost-effective solution is to use one-bit DACs, which can only output signals with two amplitude levels (e.g., +1 or -1).

The core challenge lies in designing the precoding vectors at the base station when using these one-bit DACs. This "one-bit precoding" problem is notoriously difficult due to the discrete nature of the one-bit constraint. Existing methods often struggle with either performance limitations (e.g., high symbol error rates) or computational complexity, especially in massive MIMO scenarios. This paper focuses on a specific formulation of this problem that leverages "constructive interference" (CI) and Phase Shift Keying (PSK) modulation, aiming to find precoders that minimize the bit-error-rate (BER) while being computationally feasible.

Methods

The authors propose a new approach based on a "negative ℓ₁ penalty" model. Here's a breakdown:

  1. Negative ℓ₁ Penalty Model: They introduce a novel penalty function that incorporates the one-bit constraint directly into the objective function using a negative ℓ₁-norm term. This is a key departure from previous methods that often rely on linear programming (LP) relaxations. The authors prove that this penalty model is an "exact reformulation," meaning its optimal solutions are also optimal for the original, discrete problem, provided the penalty parameter is sufficiently large.
  2. Min-Max Reformulation: The penalty model is then transformed into an equivalent min-max problem. This reformulation is crucial for enabling the development of an efficient optimization algorithm.
  3. Alternating Optimization (AO) Algorithm: To solve the min-max problem, they propose an alternating optimization algorithm. This algorithm iteratively updates two sets of variables: the precoding vector x and an auxiliary variable y related to the penalty term. Each update step is designed to be computationally efficient, involving matrix-vector multiplications and projections. The algorithm is proven to converge to a stationary point.
  4. Homotopy/Continuation Technique: To further improve practical performance, they employ a homotopy or continuation technique. This involves starting with a small penalty parameter and gradually increasing it, tracing the solution path. This helps in finding a good initial point for larger penalty values and ultimately leads to a one-bit solution.

Datasets

Not discussed in the paper. The evaluation is based on simulated massive MIMO channels and random channel matrices.

Key Insights and Ideas

  • The Power of Negative ℓ₁ Penalty: The core insight is that a negative ℓ₁ penalty can effectively enforce discrete constraints in optimization problems like one-bit precoding, offering an alternative to LP relaxations that can be computationally intensive.
  • Exactness of the Penalty: Proving the "exactness" of the penalty model is critical. It guarantees that solving the continuous penalty problem yields the correct discrete solution for the original problem, simplifying the overall approach.
  • Min-Max for Tractability: Reformulating the problem into a min-max structure, combined with the specific structure of the penalty, allows for the development of an efficient alternating optimization algorithm.
  • Efficiency in Massive MIMO: The proposed AO algorithm's low per-iteration complexity makes it particularly well-suited for the high-dimensional problems encountered in massive MIMO systems, where traditional methods can become intractable.

Key Results

  • BER Performance: The proposed "CI-1-Bit NL1P" approach consistently outperforms state-of-the-art CI-based algorithms (like MSM and OPSU) in terms of bit-error-rate (BER) across different system configurations (e.g., 8-PSK and 16-PSK modulation). The paper reports SNR gains of up to 6dB and 2.5dB in specific scenarios.
  • Computational Efficiency: The NL1P approach demonstrates significantly lower CPU time compared to other CI-based methods, especially as the number of users increases. This is attributed to the AO algorithm's efficient per-iteration computations, avoiding the complex interior-point methods used in LP relaxation-based algorithms.
  • Scalability: The computational cost of the proposed method grows much slower with system scale compared to methods like P-BB and OPSU, making it more practical for large-scale deployments.

Limitations and Future Work

  • Parameter Tuning: While the paper provides theoretical guidance for parameter selection in the AO algorithm, practical tuning might still be required for optimal performance in diverse scenarios.
  • Generalization to Other Modulations: The current work focuses on PSK modulation. Extending the negative ℓ₁ penalty approach to other modulation schemes (e.g., QAM) could be a valuable future direction.
  • Hardware Implementation: While the computational efficiency is highlighted, further research into the practical hardware implementation of the proposed precoding scheme would be beneficial.

Summary updated

Abstract

This paper considers the one-bit precoding problem for the multiuser downlink massive multiple-input multiple-output (MIMO) system with phase shift keying (PSK) modulation and focuses on the celebrated constructive interference (CI)-based problem formulation. The existence of the discrete one-bit constraint makes the problem generally hard to solve. In this paper, we propose an efficient negative ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> penalty approach for finding a high-quality solution of the considered problem. Specifically, we first propose a novel negative ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> penalty model, which penalizes the one-bit constraint into the objective with a negative ℓ <inf xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</inf> -norm term, and show the equivalence between (global and local) solutions of the original problem and the penalty problem when the penalty parameter is sufficiently large. We further transform the penalty model into an equivalent min-max problem and propose an efficient alternating optimization (AO) algorithm for solving it. The AO algorithm enjoys low periteration complexity and is guaranteed to converge to the stationary point of the min-max problem. Numerical results show that, compared against the state-of-the-art CI-based algorithms, the proposed algorithm generally achieves better bit-error-rate (BER) performance with lower computational cost.
BibTeX
@inproceedings{wu2022novel,
  author = {Zheyu Wu and Bo Jiang and Ya‐Feng Liu and Yu‐Hong Dai},
  title = {{A Novel Negative ℓ<sub>1</sub> Penalty Approach for Multiuser One-Bit Massive MIMO Downlink with PSK Signaling}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2022},
  doi = {10.1109/icassp43922.2022.9747030},
}
Figure