Divide and Conquer: One-bit MIMO-OFDM Detection by Inexact Expectation Maximization

Mingjie Shao, Wing-Kin Ma

TL;DR

This paper tackles the challenge of detecting signals in massive MIMO-OFDM systems using low-power, one-bit analog-to-digital converters (ADCs). They propose an Expectation-Maximization (EM) based "divide and conquer" approach that breaks down the complex detection problem into smaller, manageable sub-problems, leading to efficient and accurate signal recovery.

What Problem Does This Paper Address?

Massive MIMO (Multiple-Input Multiple-Output) systems, with their large number of antennas, promise significant improvements in wireless communication. However, equipping all these antennas with high-resolution ADCs is prohibitively expensive and power-hungry. One-bit ADCs offer a compelling solution by drastically reducing cost and power consumption. The catch? This extreme quantization introduces severe signal distortion, making traditional signal detection methods ineffective. Specifically, in MIMO-OFDM (Orthogonal Frequency-Division Multiplexing) systems, the one-bit quantization destroys the subcarrier orthogonality, which is crucial for efficient detection in unquantized systems. This transforms a problem that could be decoupled across subcarriers into a single, massive, and computationally intractable problem. The paper aims to overcome this by developing an efficient detection method for one-bit quantized MIMO-OFDM signals.

Methods

The core of the paper's contribution lies in adapting the Expectation-Maximization (EM) algorithm for one-bit MIMO-OFDM detection. The authors frame the problem as a Maximum Likelihood (ML) detection task. They leverage the EM algorithm's ability to iteratively refine estimates. Crucially, the M-step of their EM algorithm allows them to decouple the detection problem across subcarriers, mirroring the successful "divide and conquer" strategy used in unquantized MIMO-OFDM.

They introduce "inexact" variants of the EM algorithm to improve computational efficiency. These inexact methods involve using approximate solutions within the M-step, rather than solving the sub-problems exactly. Specifically, they explore:

  • Exact BCD: EM Algorithm: This is the classical EM approach where the M-step involves solving box-constrained quadratic programs for each subcarrier.
  • Inexact BCD: One-Step PG in M-Step: This variant uses a single step of a Projected Gradient (PG) method to approximate the solution in the M-step.
  • Inexact BCD: Inexact Update in M-Step: This approach employs a limited number of iterations of the FISTA (Fast Iterative Shrinkage-Thresholding Algorithm) method, an accelerated projected gradient algorithm, to solve the M-step sub-problems.

The authors interpret these methods through the lens of Block Coordinate Descent (BCD) and variational inference, providing a deeper understanding of their operation.

Datasets

Not discussed in the paper. The paper focuses on algorithmic development and evaluates performance through numerical simulations using synthetic data generated according to a specified channel model.

Key Insights and Ideas

  • EM as a "Divide and Conquer" Tool: The paper highlights that the EM algorithm, particularly its M-step, can effectively restore the subcarrier-wise decoupling lost due to one-bit quantization. This is the central "divide and conquer" insight.
  • Inexact EM for Efficiency: The key innovation is the development of inexact EM variants. By not requiring exact solutions in the M-step, significant computational savings can be achieved, making the algorithm practical for massive MIMO scenarios.
  • Trade-off between M-step Accuracy and Iterations: The paper implicitly discusses a trade-off: more accurate M-steps might reduce the total number of EM iterations needed for convergence, but each iteration is more computationally expensive. Conversely, less accurate M-steps are faster per iteration but might require more EM iterations.
  • Potential for Broader Applicability: The authors suggest that the core EM framework and the "divide and conquer" strategy could be extended to other MIMO detection methods beyond the ML approach studied here.

Key Results

The proposed EM algorithms, particularly the inexact variants, demonstrate strong performance in terms of Bit Error Rate (BER).

  • Superior to One-Bit ZF: The proposed EM algorithms consistently outperform the one-bit Zero-Forcing (ZF) detector.
  • Better than Unconstrained EM: The proposed EM algorithms achieve a performance gain of approximately 5 dB over the unconstrained regularized EM algorithm from prior work [14] in the mid-to-high SNR region.
  • Robustness in Large Systems: Even for very large system dimensions (e.g., N=512, K=36, W=1024), the EM algorithms maintain satisfactory BER performance, while other methods like 1BOX struggle.
  • Convergence: The paper shows that the negative log-likelihood monotonically decreases with EM iterations for all proposed algorithms, indicating convergence. The inexact FISTA-type EM converges comparably to the exact EM, within 20-40 iterations.

Limitations and Future Work

  • Computational Complexity: While inexact EM improves efficiency, the paper acknowledges that evaluating DFT/IDFT operations in each EM iteration can still incur significant computational cost, especially for large numbers of subcarriers.
  • Convergence Analysis of Inexact EM: The paper notes that while there are strong indications of convergence for the inexact EM schemes (especially the FISTA variant), a formal convergence analysis is challenging due to the nature of the optimization variables (distributions, not simple vectors) and is left as future work.
  • Extension to Other Detection Methods: The authors explicitly state that future work will explore extending the "divide and conquer" EM strategy to other MIMO detection methods that offer powerful performance in classical MIMO detection.
  • Other Quantization Schemes: The paper focuses on one-bit quantization and QAM constellations. Future work could explore other quantization levels or modulation schemes.

Summary updated

Abstract

Adopting one-bit analog-to-digital convertors (ADCs) for massive multiple-input multiple-output (MIMO) implementations has great potential in reducing the hardware cost and power consumption. However, distortions caused by quantization raise great challenges. In MIMO orthogonal frequency-division modulation (OFDM) detection, coarse quantization renders the orthogonal separation among subcarriers inapplicable, forcing us to deal with a problem that has a very large problem size. In this paper we study the expectation-maximization (EM) approach for one-bit MIMO-OFDM detection. The idea is to iteratively decouple the MIMO-OFDM detection problem among subcarriers. Using the perspective of block coordinate descent, we describe inexact variants of the classical EM method for providing more flexible and computationally efficient designs. Simulation results are provided to illustrate the potential of the divide-and-conquer strategy enabled by EM.
BibTeX
@inproceedings{shao2021divide,
  author = {Mingjie Shao and Wing-Kin Ma},
  title = {{Divide and Conquer: One-bit MIMO-OFDM Detection by Inexact Expectation Maximization}},
  booktitle = {IEEE International Conference on Acoustics, Speech and Signal Processing},
  year = {2021},
  doi = {10.1109/icassp39728.2021.9414559},
}
Figure