[Paper Review] Probabilistic Receiver Architecture Combining BP, MF, and EP for Multi-Signal Detection

Daniel J. Jakubisin, R. Michael Buehrer+1 more|arXiv (Cornell University)|Apr 17, 2016

Error Correcting Code Techniques44 references3 citations

TL;DR

This paper proposes a novel probabilistic receiver architecture that integrates belief propagation (BP), mean field (MF) approximation, and expectation propagation (EP) for multi-signal detection in MIMO and multi-user systems. By structuring the factor graph into distinct subgraphs—Gaussian BP for multi-signal detection, MF for channel estimation and observations, and discrete BP for demodulation—the framework achieves superior detection performance with theoretical justification and low complexity through EP-based message passing between discrete and continuous domains.

ABSTRACT

Receiver algorithms which combine belief propagation (BP) with the mean field (MF) approximation are well-suited for inference of both continuous and discrete random variables. In wireless scenarios involving detection of multiple signals, the standard construction of the combined BP-MF framework includes the equalization or multi-user detection functions within the MF subgraph. In this paper, we show that the MF approximation is not particularly effective for multi-signal detection. We develop a new factor graph construction for application of the BP-MF framework to problems involving the detection of multiple signals. We then develop a low-complexity variant to the proposed construction in which Gaussian BP is applied to the equalization factors. In this case, the factor graph of the joint probability distribution is divided into three subgraphs: (i) a MF subgraph comprised of the observation factors and channel estimation, (ii) a Gaussian BP subgraph which is applied to multi-signal detection, and (iii) a discrete BP subgraph which is applied to demodulation and decoding. Expectation propagation is used to approximate discrete distributions with a Gaussian distribution and links the discrete BP and Gaussian BP subgraphs. The result is a probabilistic receiver architecture with strong theoretical justification which can be applied to multi-signal detection.

Motivation & Objective

• To address the poor performance of standard BP-MF frameworks in multi-signal detection due to ineffective MF approximation for signal separation tasks.
• To develop a factor graph construction that separates detection, channel estimation, and demodulation into specialized subgraphs for improved inference.
• To reduce computational complexity by applying Gaussian BP to equalization factors while preserving probabilistic message passing.
• To integrate expectation propagation (EP) to bridge discrete demodulation and continuous detection subgraphs, enabling accurate message exchange.
• To provide a theoretically grounded, low-complexity receiver architecture applicable to MIMO and multi-user detection with strong performance.

Proposed method

• The joint probability distribution is factorized into three subgraphs: a Gaussian BP subgraph for multi-signal detection, an MF subgraph for observations and channel estimation, and a discrete BP subgraph for demodulation and decoding.
• Expectation propagation (EP) is used to approximate discrete distributions with Gaussian distributions, enabling message passing between the discrete BP and Gaussian BP subgraphs.
• Messages from the MF subgraph to observation factors are computed using posterior means and variances derived from channel coefficient estimates and prior information.
• The Gaussian BP subgraph computes posterior beliefs for transmitted symbols using approximate sufficient statistics derived from channel state information and received signals.
• The MF subgraph computes messages to channel coefficients based on likelihoods from observations and prior distributions, using conjugate priors and moment matching.
• The resulting architecture enables iterative message passing across subgraphs, with convergence guaranteed by the region-based free energy framework.

Experimental results

Research questions

• RQ1 Can a BP-MF-EP hybrid framework outperform standard BP-MF architectures in multi-signal detection tasks?
• RQ2 Is the MF approximation effective for multi-signal detection, particularly in interference-limited scenarios?
• RQ3 How can the factor graph be restructured to decouple detection, channel estimation, and demodulation for improved performance?
• RQ4 What is the impact of applying Gaussian BP to equalization factors in terms of complexity and accuracy?
• RQ5 Can expectation propagation effectively bridge discrete and continuous message passing in a hybrid receiver architecture?

Key findings

• The MF approximation is shown to be ineffective for multi-signal detection due to its inability to handle hard constraints in signal separation models.
• The proposed factor graph decomposition significantly improves detection performance compared to standard BP-MF frameworks.
• The use of Gaussian BP for equalization factors reduces computational complexity while maintaining high accuracy in multi-signal detection.
• Expectation propagation enables accurate and efficient message exchange between discrete demodulation and continuous detection subgraphs.
• Theoretical justification is provided via region-based free energy minimization, ensuring convergence to stationary points of the approximation.
• Numerical results demonstrate improved bit error rate (BER) performance in MIMO and multi-user scenarios, particularly in low SNR regimes.