Dynamic traffic forecasting and fuzzy-based optimized admission control in federated 5G-open RAN networks

The adaptive nature of fifth-generation (5G) wireless network architecture offers a significant opportunity to enhance system capacity and provide more efficient radio resource utilization. Adaptivity is achieved partly by recent advances in network function virtualization, network slicing, and the coexistence of multiple radio access technologies (RAT) [1]. One of the principal incentives behind redesigning cellular networks is to serve a plethora of devices with different requirements. There are numerous techniques for enhancing system capacity (e.g., the use of ultra-dense small cell distribution, millimeter waves (mmWave), new radio (NR), and intelligent cognitive radio) [2‐7]. However, there is little emphasis in the literature on the interpolation of current standards with existing ones to guarantee a seamless multi-operator orchestration. Such a seamless ecosystem is required to provide enhanced quality of experience (QoE) and efficient radio resource utilization [8, 9].

The coexistence of various heterogeneous wireless networks provides better performance by accumulating system capacity, supporting higher data rates, and reducing latency and packet loss. From the implementation perspective, operators often use existing network infrastructure to serve traditional voice and Web browsing applications as they offer satisfactory service performance. Also, network operators expect that whenever users are outside 5G coverage, available legacy RATs are needed to provide a seamless service to end users [10, 11]. Such coexistence is becoming the dominant feature of the current and next generation of cellular networking. In this regard, the novel concept of the open radio access network (O-RAN) could be considered a complementary option to the new 5G RATs [12, 13].

Network selection using the tenant-based approach is a common technique to provide heterogeneous RAT services [1, 14‐17]. Selecting the best RAT from an available set adds latency, leading to network congestion, particularly when the number of devices requesting access to the network is extensive. Furthermore, inefficient admission control saturates the network, impractical for providing real-time services with stringent QoE demand [6, 18]. Furthermore, lack of optimal network resource utilization and fairness impact the network’s QoS. In such scenarios, the choice of technique for selecting the best access network from multiple heterogeneous RATs by the user remains an open problem. We propose a forecasting and admission control (FAC) framework to support the presented federated O-RAN. This framework builds on dynamic traffic demand and augmented by particle filtering, followed by a network selection method using NSGA-II fuzzy-logic optimization to address the challenges mentioned above.

This work focuses on an efficient tenant-aware network configuration in the federated O-RAN, an extended version of the well-known O-RAN architecture, to achieve this autonomously. A federation controller is added to the O-RAN architecture and acts as a switch to select the optimal network based on various networks and user statistics, including network bandwidth, required data rate, latency sensitivity of the requested service, packet loss, priority, and signal strength. The contributions in this paper are (1) a novel federation framework for tenant-aware network configuration which features a dynamic demand-estimation scheme embedded with fuzzy-logic-based optimization for the optimal network selection, and (2) two algorithms in which we establish a multivariate service allocation priority factor and admission queue and build a service profile used for service monitoring. (3) We analytically compare to several state-of-the-art methods for admission control and resource allocation schemes and show how FAC is more efficient and provides higher QoE.

The remainder of the paper is organized as follows. Related work in correspondence with the proposed work is described in detail and also summarized in Table 1 in Sect. 2. The system model considered in the scope of the proposed work, its design, and statistics are explained in detail along with a summary of key symbols used throughout this paper are listed in Table 2 in Sect. 3. Section 4 presents the proposed forecasting and admission control scheme and its respective algorithms. Section 5 introduces the evaluation schemes. We showed our results with a detailed comparison in Table 3 in Sect. 6. The conclusion is presented in Sect. 7.

5G promises on-demand service deployment, support of service heterogeneity, and coordination of multiple access network technologies. Over the last few years, multiple research teams initiated work on the transformation of RANs by integrating various technologies to provide more agile services to end-users. For example, 3GPP TR 38.804 Release 14 describes dual connectivity between 4G LTE and 5G NR. Work by 3GPP in TR 37.900 Release 15 explains multi-RAT deployment architecture. It also identifies relevant scenarios and their radio frequency requirements for implementing the Multi-Standard Radio (MSR) Base Station [9, 15‐17, 19]. More recently, open RAN (O-RAN) (aka V-RAN), a novel architecture with interoperability of various networks at its core principle, has been proposed. O-RAN is an emerging technology that incorporates virtualization and intelligence in networks. One such example is the OpenRAN project by Telecom Infra, a software-driven architecture that evolved from the concept of Cloud RAN (C-RAN). It provides a solution based on software-defined networks and openness of general-purpose hardware [20]. Another individual effort is the xRAN forum, an open-source alternative to the traditional RAN architecture, which provides a solution by separating data and control planes of network devices and opening interfaces with intelligence between various RAN’s building blocks [21].

In 2018, the C-RAN Alliance and the xRAN forum merged into an open radio access network (O-RAN) to support the evolution of 5G networks and beyond. More than 160 well-known contributors from large and small companies, vendors, academic institutions, and network operators are participating in the standardization of this technology (e.g., Nokia, Hewlett Packard Enterprise, Intel, Vodafone [12]). This multi-vendor, interoperable technology eliminates dependencies and opens the protocols and interfaces between various components to incorporate intelligence into RAN to support different deployment scenarios [13, 22, 23]. This paper continues the trend by introducing a federation layer within an O-RAN architecture to enable dynamic traffic forecasting, efficient admission control, and service monitoring.

In addition to the coexistence of various access technologies, future demand prediction (via efficient forecasting techniques) is among the operators’ main challenges. Network operators strive to make resource management and orchestration (MANO) processes highly automated to cope with the volatile demand. To realize this, Sciancalepore et al. proposed a Holt-Winters theory-based mobile traffic forecasting and slice scheduling approach for admission control in 5G networks [26]. Two low-complexity algorithms were developed into a geometric knapsack problem to ensure optimal slice admission and better QoE. The authors further proposed enhancements to their work for traffic and user mobility analysis on guaranteed and best-effort traffic. The signaling-based network slicing broker is used for capacity forecasting of the cellular network [27]. Raikwar et al. proposed the concept of using the Holt-Winters method for predicting vehicular traffic demand in long- and short-term traffic windows for an efficient transportation management system [28]. Tseliou et al. proposed a multi-tenant slicing capacity framework for on-demand resource allocation in LTE networks. The authors integrated the capacity broker into the 3GPP architecture for extracting variations in traffic patterns. The technique improves traffic forecasting based on the Monte Carlo method [1]. For short-term forecasting, a few other network models using the Monte Carlo approach can be found in [29‐31]. Narmanlioglu et al. investigated a Bayesian technique for predicting active users in the LTE network to ensure efficient resource allocation [32]. Miao et al. proposed a multi-spatiotemporal framework for forecasting cellular user traffic [33]. Another forecasting framework based on the Bayesian model and Markov chain Monte Carlo (MCMC) techniques is proposed in [34] to predict the spatiotemporal information of traffic distribution in a cellular network. Holt-Winters and Bayesian are basic exponential smoothing techniques. These techniques are simple, yet work well over short time series in a linear system using prior assumptions about the user. Monte Carlo-based forecasting techniques make predictions according to data from previous instances [42]. However, these approaches are not suitable for forecasting demand in dense networks with limited or no prior information about user demand and network capacity. In this paper, we implement a hybrid Monte Carlo-based particle filtering technique to predict traffic demand. The hybrid Monte Carlo-based particle filtering technology has proved its superiority over other approaches, due to its non-dependency on previous data samples in nonlinear and non-Gaussian systems, and its multimodal processing capability makes it suitable for a wide range of communication applications.

Fuzzy-logic optimization proved to be among the most effective approaches in coping with the lack of information and associated uncertainties surrounding users. A significant amount of research is available on fuzzy-logic optimization. Bouali et al. proposed a novel, context-aware, user-driven framework for network selection. The authors implemented fuzzy multiple-attribute decision-making (MADM) approach to select the best RAT of the network’s defined policies [18]. Goudarzi et al. proposed a hybrid model that uses a multi-point algorithm in heterogeneous networks for the most suitable RAT selection. This scheme implements biogeography-based optimization (BBO) on RAT selection probabilities obtained from a Markov decision process (MDP) [36]. Kaloxylos et al. applied a fuzzy-logic approach for efficient RAT selection mechanism between (H)eNBs and Wi-Fi APs, addressing static and low-mobility users [37]. The authors of [35] proposed a fuzzy call admission control scheme for wireless multimedia networks. Khan et al. [38] proposed a hybrid fuzzy-logic-based genetic algorithm (H-FLGA) for resource allocation in 5G-driven VANETs. Zeng et al. [39] proposed a fuzzy-logic-based multi-criterion scheme to investigate a coordinated NOMA system for resource allocation in 5G. Their proposed resource allocation algorithms serve channel-gain-based subchannel allocation (SCG-SA), low-complexity, fuzzy-logic user-ranking-order-based joint resource allocation (FLURO-JRA). In [40], Silva et al. proposed a self-tuning approach based on fuzzy-logic for resource allocation during handover in small, dense cells. This approach compared the received signal level with a proposed fuzzy-logic-based threshold derived from the user’s velocity, signal power, and channel quality. This way, the number of handovers was reduced, and a lower failed handover ratio was achieved. Shrimali et al. [41] proposed a weighted-sum-based, multi-objective optimization framework for resource allocation in cloud infrastructure. This framework applied the fuzzy-logic approach to generate coefficients of the defined objectives. These coefficients used by the genetic algorithm to generate Pareto optimal solutions. The existing research is considering few application requirements or network statistics for best network selection among two heterogeneous RATs and resource allocation. However, today’s network is more complex and dynamic due to multiple and heterogeneous application requirements and network statistics, and especially in the case of the coexistence of various heterogeneous RATs such as in O-RAN. In such scenarios, Information about uncertainty on application requirements is essential for fuzzy-logic operations. This is because higher uncertainty leads to inefficient network selection, overload, and agreed QoE degradation [43]. The proposed framework applies the Non-Dominated Sorting Genetic Algorithm II, coupled with the fuzzy-logic approach to provide an optimal network selection based on user forecasted demand, network capacity and fitness policies of the fuzzy-logic model to reduce the effects of uncertainty (see Fig. 3).

The management of resource allocation to support a massive amount of heterogeneous traffic flow in proportion to network capacity is still an open issue [44]. In this section, a system model, known as federated O-RAN (FORAN), is presented, which is similar to the O-RAN architecture, as given in [12]. A federation controller is added to the O-RAN referenced architecture, which acts as a switch to select the optimal network based on various network and user demand statistics as discussed in the detail in the following subsections.

In this section, we propose a forecasting and admission control framework. This framework applies the aforementioned sampling-based forecasting technique to obtain optimal network selection by the key entities of the federation controller. The federation controller consists of three key entities: the Demand and Capacity Analyzer (DCA), the Network selection and configuration function (NSCF), and QoS/QoE and Traffic Flow management (QTFM). The systematic diagram of the proposed Forecasting and Admission Control (FAC) framework processed by these entities is illustrated in Fig. 3 and discussed in detail in the following subsections.

We define the evaluation parameters in this section. In the context of evaluating the proposed framework, the chosen performance metrics set out to be the assessment of resources and network utilization, resource allocation fairness among the tenants as well as their associated user satisfaction level, as discussed in the following subsections. The performance evaluation framework is aligned with the existing work from the literature for comparison.

A simulation model developed in MATLAB to evaluate the performance of the proposed framework. A virtual network is constructed with a set of varied system parameters to support four distinguished services. The number of tenants associated with the virtual network is considered to be \(U=[5, 330]\) for a heterogeneous service provisioning. The average number of users associated with the tenant u is \(E[u_i]=100\), as considered in [26] and [27]. Significantly, \(\beta =[10^{-2}, 10^{-7}]\), \(\iota = [10, 200]\) ms, \(\varphi =[1, 5]\), \(\tau _h= [10, 100]\) MHz, and \({\mathcal {R}}\_Op=500\) are the considered ranges of tenant-service-specific packet loss, latency sensitivity, priority, desired resource demand, and available operator resources for each service belong to \({\mathcal {S}}\), respectively. The overall demand is normalized to 0 or 1 for simplicity.

Figures 5, 6, 7, 8, and 9 show the performance of the proposed dynamic traffic forecasting and fuzzy-based admission control in the context of user satisfaction level, resource allocation fairness, and utilization gain. The results are compared with existing schemes in the literature and summarized in Table 3. We compare to mobile traffic forecasting (MTF) [26], reinforcement learning (RL-NSB) [27], online auction (ORAN) and greedy algorithm [48], and bankruptcy game (BG) [49], based resource allocation and admission control schemes.

In this paper, we proposed a dynamic traffic forecasting and admission control framework for a federated open radio access network (O-RAN). In this framework, a three-stage approach, namely demand and capacity analyzer, network selection and configuration, and QoS/QoE and traffic flow management have been proposed. This framework predicts the future traffic demand for the optimal network selection among multiple heterogeneous access networks and resource management to ensure better tenant QoE and O-RAN network utilization. The considered demand characteristics are bandwidth, latency sensitivity, quality of service demand, service priority type, and packet loss ratio. In this work, a fuzzy-logic-based network selection scheme with a multi-variate admission priority feature is introduced for optimal admission control and service allocation to the tenants. Moreover, a QoS/QoE-based service monitoring approach is also presented to update the demand via a forecasting modifier to allocate resources approximately closer to the tenant’s actual demand, which improves overall network QoS and tenant QoE. The proposed framework outperforms the existing legacy approaches in terms of enhanced tenant QoE and fairness of resource allocation, efficient network utilization, and better user satisfaction levels for various heterogeneous services of future wireless networks. For future work, we aim to enhance the framework with signaling optimization on homogeneous user-application-specific characteristics for service provisioning of complex networks of the future.