At-scale deployment of a disaggregated near-real-time radio access network (RAN) Intelligent Controller (near-RT RIC) remains challenging in user-centric RANs, given concurrent objectives of low control-latency and low deployment cost. The placement problem of disaggregated near-RT RIC is formulated as a mixed-integer nonlinear programming (MINLP) that jointly minimizes control-latency and deployment cost across cloud and edge computing nodes. To address the computational complexity and adapt to network dynamics, a multi-armed bandit (MAB) approach, specifically, cost-subsidy explore-then-commit (CSETC) policy, is leveraged to decouple the joint latency and cost objective into metric-specific policies for online placement under feasibility constraints. In simulations with 512 user equipment (UEs), the proposed CS-ETC-RIC algorithm achieves up to 67% and 55% lower total control-latency than the co-located near-RT RIC and the HG-RIC benchmarks, respectively, while activating fewer edge computing nodes and thereby lowering deployment cost. These findings indicate a scalable and effective mechanism for user-centric RAN management within the O-RAN ALLIANCE framework.

We address the degradation of wireless quality caused by phase difference variations among access points (APs) in cell-free massive multiple-input multiple-output (CFmMIMO) systems employing coherent joint transmission (CJT) under time division duplexing (TDD). Using 5G-compliant equipment, we analyze real measurement data and show that inter-AP phase differences can fluctuate on the millisecond timescale. To mitigate this issue, we propose a predictive calibration method that employs time-series regression to estimate and compensate short-term phase variations during precoding. Simulation results based on the measured data demonstrate that the proposed method improves spectral efficiency (SE) by up to 0.63 bit/s/Hz.

This paper presents a full-stack testbed for evaluating inter-site central processing unit (CPU) cooperation in large-scale cell-free massive multiple-input multiple-output (CF-mMIMO) systems. In such systems, multiple CPUs are distributed across different sites to address computational and fronthaul capacity limitations. However, this distributed architecture introduces inter-site interference that degrades radio quality at site edges. To mitigate this issue, two inter-site CPU cooperation schemes, tight CPU cooperation (TCC) and loose CPU cooperation (LCC), have been proposed. Despite their theoretical promise, their practical feasibility in an end-to-end system and their effectiveness under real outdoor conditions remain unclear. To address these gaps, we have developed a full-stack 5G testbed to experimentally validate the feasibility and performance of TCC and LCC. Measurements conducted over a 5400 m2outdoor area demonstrate that both methods significantly improve uplink throughput by suppressing inter-site interference. We further analyze the relationship between throughput, signal power, and interference power on a per-user basis. Finally, we discuss key challenges for scaling CF-mMIMO deployments.

Toward 6G in around 2030, cell-free massive MIMO (CF-mMIMO) is expected to provide high radio quality everywhere. To deploy CF-mMIMO over a wide area, a distributed site architecture has been proposed. Central processing units (CPUs), which process radio signals, are deployed to multiple sites, which are physical bases for aggregating radio signals. In a distributed site architecture, the radio quality of user equipments (UEs) located at the border of the site degrades because access points (APs) cannot be used in signal processing between sites. We consider the application of a multi-homing technique to the fronthaul (FH) between the AP and sites. By multi-homing, an AP connects to multiple sites via the fronthaul in order to cooperate with APs in different sites. Since the radio signals of the multi-homing AP can be received by the CPUs of multiple sites, radio quality is improved by the cooperation of APs in different sites. However, multi-homing has an issue: the FH cost increases due to the installation of additional FH links. In order to balance the suppression of FH cost increase and improvement in radio quality, we propose an AP connection method for minimizing the number of total FH links. In the proposed AP connection method, APs in different sites are connected directly. The number of FH links can be reduced because connected APs can share one FH link. The results of computer simulation and FH cost estimation show that the proposed multi-homing methods improve the 5%-tile throughput 1.7-fold compared to the distributed site architecture while suppressing the FH cost increase compared to existing connection methods.

We analyze the impact on the wireless quality of time division duplex Cell-Free massive multiple-input multiple-output (TDD CF-mMIMO) caused by phase variation in reference signals that modulate RF signals due to clock distribution over a radio access network (RAN). Phase variation in reference signals causes temporal phase variation due to temporal difference in transmission and reception in TDD. In addition, since APs are distributed among a RAN, phase variation also occurs due to clock distribution through fronthaul depending on the placement of reference clocks (RCs), which is a challenge for practical wide-area deployment. We propose RC distribution for practical user-centric RAN, considering use of IEEE 1588 or radio over fiber (RoF) technology. We analyze phase variation of reference signals and evaluate SINR of TDD CF-mMIMO by numerical simulations, revealing that clock distribution from distributed RCs with RoF secures high SINR.

In mobile networks, network performance can be severely degraded by a signaling storm, a phenomenon characterized by control plane (C-plane) congestion due to excessive signaling messages. This paper identifies congestion of inter-site links and overload of a specific NF instance as factors leading to a signaling storm. To address these problems, we propose a combined placement and routing algorithm approach. Firstly, our joint placement and routing algorithms perform geographically distributed deployment of C-plane NF instances according to the fluctuating mobility patterns of each user equipment (UE). Secondly, our routing algorithm identifies an appropriate C-plane NF instance to manage each UE, considering both the utilization rate and the number of contexts possessed by each NF instance. Extensive simulation evaluations demonstrate that our proposed methods significantly reduce inter-site signaling messages compared to existing placement scenarios. Furthermore, the dispersion of C-plane NF instances’ utilization rate is reduced, enhancing network performance and efficiency.

The next-generation mobile networks will be devel- oped based on virtualized and distributed radio access network (RAN), necessitating cost-effective and flexible deployment solutions to meet latency requirements without incurring excessive costs. The O-RAN alliance has defined the architecture for these future networks, where the RAN controller is split into the Non-Real-Time RAN Intelligent Controller (Non-RT RIC) and the Near-Real-Time RAN Intelligent Controller (Near-RT RIC). These controllers, virtualized and deployable on general- purpose processors at edge nodes, can be distributed across the entire geographical area covered by a mobile network. This is particularly vital for user-centric RAN with cell-free massive MIMO (CF-mMIMO). The inherent high control traffic overload, due to not only continuous access point (AP) cluster updates but also from extensive controls for the user-centric RAN, runs the risk of overloading the transport network. Therefore, con- troller deployment optimization becomes essential. Additionally, in scenarios involving critical applications, latency is a key factor to consider, adding to the optimization challenges. This paper addresses the Near-RT RIC deployment problem, considering latency requirements and transport network limitations. Mixed integer linear programming (MILP) is utilized to solve the deployment issue. The solution's effectiveness is validated against different network topologies and design parameters.

This paper proposes a distributed deep reinforcement learning (DRL) method with multiple learners for AP clustering in large-scale Cell-Free massive MIMO (CF-mMIMO). In the deployment of large-scale CF-mMIMO with many user equipments (UEs) and access points (APs), it is necessary to perform AP clustering according to the demand and movements of each UE in a lightweight manner and with high inference accuracy. However, existing DRL-based methods have struggled to learn diverse and site-specific radio environments and provide high inference accuracy with small amounts of data and small neural network (NN) models for lightweight. To address this problem, the proposed method classifies learners for each radio environment with the Reference Signal Received Power (RSRP) between surrounding APs of the UE to perform learning and inference with these multiple learners. Furthermore, the proposed method dynamically adjusts the association between UE and learners based on the fluctuation in RSRP due to the UE's movements, thereby ensuring sufficient agility for user mobility. This dynamic association of UEs and learners for each radio environment enables efficient learning and improved inference accuracy by focusing on UEs in similar radio environments, even with small amounts of data and small NN models. Simulation evaluations based on actual urban structures demonstrated that the proposed method realizes AP clustering with higher inference accuracy than existing methods, even with small amounts of learning data and small NN models.

We consider scalable RAN management for large-scale deployment in a distributed central processing units (CPUs) environment with cell-free massive MIMO (CF-mMIMO). In distributed CPUs among multiple sites, there is a problem of low radio quality for users at the site edge areas, which are the boundaries between sites. This is due to inter-site interference between UEs processed by different CPUs and a reduction in received signal power owing to the inability to associate with the surrounding APs connecting to other sites. To address this problem, our approach is to deploy the vCPUs of users at the site edges to higher-level sites based on the physical hierarchical structure of the RAN. This hierarchical deployment allows the formation of broad AP clusters associated with group APs across the sites and improves radio quality. However, deploying vCPUs at higher-level sites causes the radio signal to flow into the backhaul (BH), significantly increasing the transmission load depending on the size of the AP cluster. Thus, we formulate an optimization problem to improve user throughput everywhere under the constraints of the RAN physical resources by managing the deployment of vCPUs and AP clustering. This optimization problem is non-linear and non-convex and requires inverse matrix calculations, resulting in computational complexity. Therefore, we propose a lightweight list-processing algorithm with reference signals of APs around users that does not use inverse matrices calculation and metaheuristic search. Simulation results show that the proposed method improves user throughput and provides lightweight calculation with a large number of UEs compared to existing methods.

The authors have been studying a user-centric radio access network (RAN) towards the realization of homogeneous wireless quality "anywhere anytime" by introducing Cell-free massive MIMO (CF-mMIMO) technology. Assuming a double-star network, which is frequently seen in optical access networks, central processing units (CPUs) which process CF-mMIMO signals will be deployed at central and edge sites. Due to this configuration, user-centric RAN has a unique characteristic not shared by a conventional RAN; i.e., the wireless quality varies with CPU placement, because the number of available access points (APs), which affects the wireless quality, differs between the central and the edge sites. However, transport links between the central sites and edge sites will be subject to a heavy load when all APs are connected to the central site, because radio signals are transferred over them. That is, the placement of CPUs is an important problem to be solved to realize homogeneous wireless quality taking the load on the transport links into account. In this paper, we propose a CPU placement configuration where CPUs are deployed both at the central site and edge sites. Numerical simulation shows that the proposed placement configuration can flexibly adapt to the achieved wireless quality and throughput by changing vCPU placement and radio band width allocation to the central and edge sites.

We propose an interference suppression method that reduces the amount of data transmitted between sites and the computational load of radio signal processing as well as maintaining radio quality in wide-area deployment for cell-free massive MIMO (CF-mMIMO). In a large-scale CF-mMIMO, the central processing units (CPUs) that process CF-mMIMO signals are assumed to be deployed at multiple sites. There is the problem of inter-site interference between UEs connected to CPUs at different sites. Conventional methods which deal with this problem require the exchange of radio signals between CPUs, which significantly increases the transmission load between the sites. In this paper, we consider the application of coordinated beamforming (CoBF) techniques to CF-mMIMO to reduce the inter-site transmission load. For the CoBF across the sites, the channel estimation processing function for inter-site interference is operated at each site by exchanging the pilot assignment information instead of relying on radio signals. Inter-site interference is suppressed by signal processing using the estimated channel. However, since this approach requires extensive channel estimation across sites, the computational load is high. Thus, we propose a method to reduce the computational load by adjusting the range of channel estimation for each user according to the required radio quality and interference power. Simulation results show that the proposed method can significantly reduce the transmission and computational load for the same area throughput compared to existing methods. Furthermore, we show the evaluation results of experiments using the CF-mMIMO testbed and the feasibility of the proposed method in suppressing inter-site interference.

For the efficient use of limited frequency resources, the spectrum sharing allocation between mobile networks and different existing radio systems is promising. In this scheme, we focus on an allocation method that involves sharing spectrum bands with a mobile network operator (MNO). Recently, to improve the spectrum efficiency, dynamic spectrum allocation methods according to the MNO demand were proposed. However, in demand-based dynamic allocation, channel discontinuity in the frequency and time directions in the same MNO occurs when mapping the channel to MNOs. This channel discontinuity reduces the utility of shared frequencies. To address this problem, we propose a novel spectrum allocation method that maximizes frequency and time continuity for demand-based dynamic spectrum sharing. Our performance evaluation shows that compared with the existing methods, the proposed algorithm can improve the spectrum efficiency by 6%, channel continuity by 23%, and fairness by 5%.