Expectation Propagation Detector for Extra-Large Scale Massive MIMO

Abstract

The deployment of extremely large-scale antenna array at the base station, which is referred to as extra-large-scale massive multiple-input-multiple-output (MIMO) system, is a brand new communication paradigm allowing for significant performance gain at the cost of excessively large amount of fronthaul data and ultra-high computational complexity. These practical challenges inspire the subarray-based processing architecture. Moreover, the spatial no-stationarity occurs and different portions of array observe different channel properties. Therefore, the aforementioned aspects should be considered into detector design for extra-large-scale massive MIMO systems. In this paper, the iterative detector is developed leveraging the expectation propagation (EP) principle based on the factor graph describing the subarray-based architecture. Its convergence is intuitively verified from the evolution analysis. We also demonstrate that different sizes and numbers of subarrays yield similar performance. In addition, we consider various implementation related aspects. We show that the non-stationary property can be exploited to reduce complexity. To avoid excessively high-complexity matrix inversion, we propose a hierarchical implementation architecture to enable parallel computation of simple-form matrix inversion.Moreover, we modify the proposed EP-based detector which requires only one feedforward from each subarray to the central processing unit to minimize the transfer latency. Simulation results demonstrate that the proposed detector outperforms its counterparts and verify the validity of our analyses.

Cell-Free Massive MIMO : Scalability, Signal Processing and Power Control

“…Changing perspective, different portions of the ELAA observe either the same channels with varying strength or totally different channels towards each user. In this regard, subarray-based architectures have been proposed in [112,115] to reduce the complexity of the centralized baseband processing by exploiting the spatial non-stationary properties of the channels.…”

A Linear Bayesian Learning Receiver Scheme for Massive MIMO Systems

“…In this section, we briefly review the best Bayesian based receiver in the literature, Expectation Propagation (EP) receivers [11]- [14], [19], [20], which is a combination of Bayesian and MMSE concepts. The main idea of the EP is to iteratively approximate the distribution of a random transmitted symbol vector, x by using a Gaussian probability distribution function (PDF) approximation based on the received signals y, p (t) (x|y), and a pair of tuning parameters, (λ (t) , γ (t) ) which are obtained from the exponential family distributions [17].…”

“…Complexity LBL O(N KT ) AMP [15] O(N KT ) MMSE O(N 2 K + N K) EP in [11] O((N 2 K + N K)T ) EP in [14] O((N K 2 + N K)T ) ML [8] O(M K ) receiver; the representative of best linear receivers, MMSE [9] scheme; the representative of Bayesian receivers EP [11], [14] and AMP [15] schemes; and the exhaustive search based ML scheme [8] are tabulated in Table I. The table indicates that the computational complexity of the proposed receiver increases linearly with the number of antennas, N and users, K by avoiding matrix inversion operations.…”

A Message Passing Based Receiver for Extra-Large Scale MIMO

“…Both channel estimation and data detection problems with one-bit quantization are solved with MP techniques in [6]. Recently, authors in [7] used expectation propagation (EP) to solve the symbol detection problem in XL-MIMO systems. They have exploited the subarray structure to model their EP scheme.…”