Decentralized Massive MIMO Processing Exploring Daisy-Chain Architecture and Recursive Algorithms

Abstract

Algorithms for Massive MIMO uplink detection and downlink precoding typically rely on a centralized approach, by which baseband data from all antenna modules are routed to a central node in order to be processed. In the case of Massive MIMO, where hundreds or thousands of antennas are expected in the base-station, said routing becomes a bottleneck since interconnection throughput is limited. This paper presents a fully decentralized architecture and an algorithm for Massive MIMO uplink detection and downlink precoding based on the Coordinate Descent (CD) method, which does not require a central node for these tasks. Through a recursive approach and very low complexity operations, the proposed algorithm provides a good trade-off between performance, interconnection throughput and latency. Further, our proposed solution achieves significantly lower interconnection data-rate than other architectures, enabling future scalability. This paper is build upon previous results presented at the 2018 SiPS Conference [1] and in IEEE Wireless Communication Letters [2]. The authors are with the

Decentralized Linear MMSE Equalizer Under Colored Noise for Massive MIMO Systems

“…where AWGN with a diagonal covariance matrix [6]- [8], [10]- [18]. This allows for the covariance matrix to be naturally decomposed into multiple diagonal submatrices that perfectly fit the distributed implementation of LMMSE equalization in DBP architectures.…”

“…However, with an increasing number of BS antennas, conventional centralized LMMSE detectors encounter two major bottlenecks: 1) Excessive communication bandwidth: The rapid growth of the number of BS antennas brings an exceedingly high amount of raw baseband data, including channel state information (CSI), received signal, and noise samples, that must be transferred between the radio-frequency (RF) chains and the centralized computing fabric, as shown in Fig. 1 with the red arrow lines [6]- [8]. This issue is particularly evident in a 256-antenna BS with an 80MHz bandwidth and 12-bit digital-to-analog converters, where the raw baseband data throughput can reach 1Tbps, significantly exceeding the existing capacity of BS internal interface standards [9].…”

“…To address the limitations of traditional CBP architectures, recent studies have explored a promising alternative called decentralized baseband processing (DBP) [6]- [8], [10]- [18]. As shown in Fig.…”

“…In light of this consideration, the authors in [10] presented a decentralized implementation that directly obtained MRC, ZF, and LMMSE equalization matrices in a feedforward DBP architecture. Meanwhile, the work [8] proposed a decentralized algorithm based on gradient descent to obtain the ZF equalization matrix for a daisy chain architecture.…”

“…It should be noted that prior works [6]- [8], [10]- [18] all assumed that the BS receiver noise is an ideal additive white Gaussian noise (AWGN), which has a diagonal noise covariance matrix. This allows for the covariance matrix to be naturally decomposed into multiple diagonal submatrices that perfectly fit the distributed implementation of LMMSE equalization in DBP architectures.…”