Abstract
An important concern that 5G networks face is supporting a wide range of services and use cases with heterogeneous requirements. Radio access network (RAN) slices, understood as isolated virtual networks that share a common infrastructure, are a possible answer to this very demanding scenario and enable virtual operators to provide differentiated services over independent logical entities. This article addresses the feasibility of forming 5G slices, answering the question of whether the available capacity (resources) is sufficient to satisfy slice requirements. As spectral efficiency is one of the key metrics in 5G networks, we introduce the minislot-based slicing allocation (MISA) model, a novel 5G slice resource allocation approach that combines the utilization of both complete slots (or physical resource blocks) and mini-slots with the adequate physical layer design and service requirement constraints. We advocate for a probabilistic characterization that allows to estimate feasibility and characterize the behavior of the constraints, while an exhaustive search is very computationally demanding and the methods to check feasibility provide no information on the constraints. In such a characterization, the concept of phase transition allows for the identification of a clear frontier between the feasible and infeasible regions. Our method relies on an adaptation of the Wang-Landau algorithm to determine the existence of, at least, one solution to the problem. The conducted simulations show a significant improvement in spectral efficiency and feasibility of the MISA approach compared to the slot-based formulation, the identification of the phase transition, and valuable results to characterize the satisfiability of the constraints.
Introduction
5G technology represents a paradigm shift from previous generations, moving beyond simple mobile broadband to a multi-purpose infrastructure. It must simultaneously support high-bandwidth video (eMBB), ultra-reliable low-latency communications (URLLC), and massive machine-type communications (mMTC). To manage these diverse requirements on shared physical hardware, the industry has turned to "network slicing"—creating isolated logical networks tailored to specific service needs.
However, a critical bottleneck in 5G deployment is the feasibility of these slices. Before an operator can optimize a network, they must first determine if a specific set of user demands can be met within the finite pool of available radio resources. Traditional allocation methods often struggle with this, leading to either wasted spectrum or system failure under heavy load.
The Challenge of Resource Feasibility
In a typical 5G New Radio (NR) environment, resources are organized into a grid of Physical Resource Blocks (PRBs). Each PRB consists of a set of subcarriers in the frequency domain and symbols in the time domain.
The primary challenge is the combinatorial explosion inherent in checking feasibility. For a system with even a modest number of users and base stations, checking every possible resource combination can involve upwards of $2^{90}$ permutations. This computational nightmare makes exhaustive searches impossible for real-time or even near-real-time network management.
The MISA Model: Beyond Traditional Slots
To address the limitations of traditional "slot-based" allocation—where a user is granted a full 14-symbol slot—this paper introduces the Minislot-based Slicing Allocation (MISA) model.
Unlike standard models, MISA allows for the allocation of "mini-slots," which can be as small as 2 symbols. This granularity is particularly beneficial for URLLC services that require low latency but transmit relatively small amounts of data. By using mini-slots, the network can pack these services more tightly, preserving larger blocks of spectrum for high-bandwidth eMBB users.
Probabilistic Characterization via DESS
To overcome the "curse of dimensionality" in feasibility checking, the authors developed the Density of States for Slices (DESS) algorithm. This approach adapts the Wang-Landau algorithm, a technique from statistical physics originally used to study complex systems like protein folding.
DESS is an "importance sampling" method. Instead of checking every possible configuration, it estimates the "density of states" to build a probability distribution of how many ways the network constraints can be satisfied. This allows the system to detect "rare events"—the few valid allocation configurations hidden in a sea of invalid ones—much faster than standard Monte Carlo simulations.
Key Results and the "Phase Transition"
The study's simulations revealed several critical insights into 5G network behavior:
- The Phase Transition: Feasibility in 5G networks behaves like a physical phase transition. There is a sharp "cliff" or frontier; if an operator increases demand just slightly beyond this point, the probability of finding a feasible allocation drops from nearly 100% to 0% almost instantly.
- Efficiency Gains: The MISA approach demonstrated at least a 2x improvement in spectral efficiency over traditional slot-based models (SOSO) in common configurations. While SOSO often becomes 100% infeasible when the number of users exceeds the number of resource blocks, MISA remains robust.
- Computational Scalability: The DESS algorithm scales with a complexity of approximately $O(V^{4.11})$. While computationally intensive, it represents a massive leap over the exponential complexity of exhaustive searches, making it a viable tool for pre-calculating network limits.
- Constraint Sensitivity: Interestingly, the analysis showed that mini-slot orthogonality constraints (ensuring users do not interfere with one another) are often more restrictive and harder to satisfy than the actual data-rate requirements of the services.
Conclusion
The MISA model and the DESS algorithm provide a powerful framework for understanding and managing 5G resource allocation. By moving to a minislot-based approach and utilizing probabilistic characterization, operators can significantly boost spectral efficiency and better predict the limits of their network slices. Future work will look to extend this model to multi-cell environments and incorporate dynamic, time-varying traffic patterns to better reflect real-world mobility.
Summary updated