MIMO Detection via Gaussian Mixture Expectation Propagation: A Bayesian Machine Learning Approach for High-Order High-Dimensional MIMO Systems

1 st Shachar Shayovitz, 2 nd Doron Ezri, 3 rd Yoav Levinbook · 2024

Read MIMO Detection via Gaussian Mixture Expectation Propagation: A Bayesian Machine Learning Approach for High-Order High-Dimensional MIMO Systems by 1 st Shachar Shayovitz, 2 nd Doron Ezri, 3 rd Yoav Levinbook (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-09068.txt.

Introduction

Shachar Shayovitz, Doron Ezri, and Yoav Levinbook worked on signal decoding in MIMO systems. Their paper, from twenty twenty-four, deals with systems that send many data streams at once in the same band. That trick boosts speed but the streams clash and noise builds up. Decoding them well is the hard part. This matters most for the uplink, from user to tower.

The team built a new decoding method that grows out of a known one. It models the messages passed in decoding with a mix of bell curves. That mix fits the true signal shape better when the old method fails. Readers get a way to decode dense, high order MIMO signals with more care.

Catalog record

CREATOR1 st Shachar Shayovitz, 2 nd Doron Ezri, 3 rd Yoav Levinbook
YEAR2024
COLLECTIONPAPERS
FORMATTXT
FILE SIZE31 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2412-09068

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

MIMO Detection via Gaussian Mixture Expectation Propagation: A Bayesian Machine Learning Approach for High-Order High-Dimensional MIMO Systems 1 st Shachar Shayovitz, 2 nd Doron Ezri, 3 rd Yoav Levinbook Abstract MIMO systems can simultaneously transmit multiple data streams within the same frequency band, thus exploiting the spatial dimension to enhance performance. MIMO detection poses considerable challenges due to the interference and noise introduced by the concurrent transmission of multiple streams. Efficient Uplink (UL) MIMO detection algorithms are crucial for decoding these signals accurately and ensuring robust communication. In this paper a MIMO detection algorithm is proposed which improves over the Expectation Propagation (EP) algorithm. The proposed algorithm is based on a Gaussian Mixture Model (GMM) approximation for Belief Propagation (BP) and EP messages. The GMM messages better approximate the data prior when EP fails to do so and thus improve detection. This algorithm outperforms state of the art detection algorithms while maintaining low computational complexity. I Introduction In recent years, Multiple-Input Multiple-Output (MIMO) technology has emerged as a cornerstone in modern wireless communication systems, offering significant improvements in spectral efficiency, reliability, and capacity [ 1 , 2 ] . By leveraging multiple antennas at both the transmitter and receiver, MIMO systems can simultaneously transmit multiple data streams within the same frequency band, thus exploiting the spatial dimension to enhance performance. The detection of signals in MIMO systems, however, poses considerable challenges due to the interference and noise introduced by the concurrent transmission of multiple streams. This task is increasingly challenging as the dimensions of the system (number of transmit and receive antennas) and the constellation order (the number of symbols in the modulation scheme) grow. Traditional detection methods, such as Maximum Likelihood Detection (MLD), provide optimal performance but are computationally prohibitive for large-scale MIMO systems. The complexity of MLD grows exponentially with the number of antennas and the constellation order. For instance, in a system with N_{t} transmit antennas and M -QAM modulation, the ML detector needs to evaluate M^{N_{t}} possible transmitted symbol combinations. This exponential growth makes MLD infeasible for large MIMO systems, especially in real-time applications. As a result, suboptimal yet computationally efficient algorithms, including Linear Detectors (e.g., Zero Forcing, Minimum Mean Squared Error), Successive Interference Cancellation, and various heuristic and probabilistic approaches, have been extensively studied [ 3 , 4 , 5 , 6 , 7 , 8 , 9 , 10 , 11 ] . Message-passing algorithms have gained prominence in MIMO detection due to their ability to efficiently handle the inherent complexity of these systems. These algorithms, inspired by Belief Propagation (BP) [ 12 ] in graphical models, iteratively exchange messages between nodes in a Factor Graph (FG) representing the joint posterior probability distribution of the transmitted symbols given received observations. This iterative approach facilitates the computation of marginal posterior distribution over the transmitted symbols, enabling effective detection and decoding. Among the various message-passing techniques, the Sum-Product Algorithm (SPA) and its variants, such as the Approximate Message Passing (AMP) and the Generalized Belief Propagation (GBP), have been extensively studied. These algorithms [ 13 , 14 , 15 ] leverage the structure of the MIMO channel model to simplify the detection process, achieving a balance between computational complexity and detection accuracy. Expectation Propagation (EP) [ 16 ] , a Bayesian Machine Learning technique which is a more recent development in the field, has also shown promise in MIMO detection. EP extends traditional message-passing algorithms by approximating the posterior distribution with a series of moment matching steps, which can improve the accuracy of the approximations. This technique has been successfully applied to various problems in signal processing and machine learning, demonstrating its potential in enhancing the performance of MIMO detection systems. In summary, the combination of exponential complexity, high-dimensional search space, amplified noise and interference effects, and the need for scalable algorithms make MIMO detection in large dimensions with high constellation orders a formidable challenge. This motivates the ongoing research into more efficient and robust detection algorithms that can operate effectively in these challenging scenarios. In this paper, a MIMO detection algorithm based on a GMM approximation for EP is proposed. It is observed that the true prior for the data symbols is a discrete uniform distribution and thus the Gaussian prior used in Linear Minimum Mean Square Error (LMMSE) and EP is not accurate. We propose to approximate certain EP messages as GMMs and improve the resulting posterior accuracy. II System Model In this section, the mathematical model for the received UL signal is defined. Let n be the number of data symbols from some constellation (for simplicity we assume the same constellation for all symbols), where \mathcal{A} denotes the set of symbols in the constellation and E_{s} is the mean symbol energy. The transmitted symbol vector \mathbf{u}\in\mathcal{A}^{n} is an n\times 1 i.i.d. vector. The symbols are transmitted over a flat-fading complex MIMO channel defined by \mathbf{H}\in\mathbb{C}^{m\times n} , where each coefficient is drawn according to a proper complex zero-mean unit-variance Gaussian distribution and m is the number of receiving antennas. The channel output \mathbf{y}\in\mathbb{C}^{m} is given by: \mathbf{y}=\mathbf{H}\mathbf{u}+\mathbf{n} (1) where \mathbf{n} is an additive white circular-symmetric complex Gaussian noise vector with independent zero-mean components and \sigma^{2}_{w} variance. Given the model in ( 1 ), the posterior probability for the transmitted symbol vector \mathbf{u} : p\left(\mathbf{u}|\mathbf{y}\right)\propto p\left(\mathbf{y}|\mathbf{u}\right)p\left(\mathbf{u}\right)\propto\mathcal{N}(\mathbf{y};\mathbf{H}\mathbf{u},\,\sigma_{w}^{2}\mathbf{I})\Pi_{i=1}^{n}\mathds{1}_{u_{i}\in\mathcal{A}} (2) where \mathds{1}_{u_{i}\in\mathcal{A}} is the indicator function that takes value one if u_{i}\in\mathcal{A} and zero otherwise. Note that p(\mathbf{u})\propto\Pi_{i=1}^{n}\mathds{1}_{u_{i}\in\mathcal{A}} is uniform across all points in \mathcal{A}^{n} . Similarly to the formulation in [ 13 ] , the MIMO model is transformed from complex valued to real valued by an equivalent double-sized real-valued representation. This representation is obtained by considering the real and imaginary parts separately. The complex notations are the same as the real notations to the keep the notations uncluttered. The minimum symbol error optimal detection rule is the Maximum A-Posteriori (MAP) detector which requires the knowledge of the posterior distribution p\left(u_{i}|\mathbf{y}\right) . This posterior can be calculated using marginalization on the joint posterior proposed in ( 2 ): p\left(u_{i}|\mathbf{y}\right)\propto\int_{\mathbf{u}^{-i}}\mathcal{N}(\mathbf{y};\mathbf{H}\mathbf{u},\sigma_{w}^{2}\mathbf{I})\Pi_{j=1}^{N}\mathds{1}_{u_{j}\in\mathcal{A}}\,d\mathbf{u}^{-i} (3) where \mathbf{u}^{-i}=[u_{1},u_{2},...,u_{i-1},u_{i+1},...,u_{m}] is the vector of all symbols except u_{i} . The multi-dimensional integral in ( 3 ) is intractable and approximations are needed in order to find a solution. This is exactly where the BP algorithm comes in and provides an iterative message passing algorithm which computes an approximation of this integral for all u_{i} . Since this work is focused on the MAP detector for the marginal posterior, we will use the Sum Product Algorithm [ 17 ] which is a specific variant of the BP algorithm for computing marginal posteriors. In order to define the SPA messages, we first introduce the reader to Factor Graphs. A Factor Graph (FG) is a bipartite graph representing the factorization of a function. In probability theory and its applications, FGs are used to represent factorization of a probability distribution function, enabling efficient computations, such as the computation of marginal distributions through the sum–product algorithm. The FG for the posterior in ( 2 ) is shown in Fig. 1 for m=5 . The SPA messages are defined in [ 17 ] for a general FG. p\left(\mathbf{y}|\mathbf{u}\right) u_{3} u_{2} u_{1} u_{4} u_{5} p(u_{1}) p(u_{2}) p(u_{3}) p(u_{4}) p(u_{5}) q^{(l)}_{5}(u_{5}) q^{(l)}_{-5}(u_{5}) Fig.

Read the complete text