Data-driven Bandwidth Adaptation for Radio Access Network Slices

TL;DR

The paper tackles SLA-driven QoS for multiple RAN slices by addressing packet-delay guarantees with a data-driven Bandwidth Demand Estimator (BDE). It introduces a two-function BS architecture (BDE and Network Slice Multiplexer) and casts the BDE bandwidth decision as an infinite-horizon DP, solved via model-based reinforcement learning with periodic transition estimation and VI, yielding an $\epsilon$-soft policy. A key contribution is per-slice learning and a practical testbed evaluation on a 3GPP LTE Amarisoft system, showing substantial bandwidth savings while meeting per-slice delay targets $Q_i(t)\le Q_c$ and SLA probabilities $P_i$. The results demonstrate scalability and real-world viability for dynamic PRB allocation, leveraging domain insights like cost monotonicity to accelerate learning. Overall, the work provides a scalable, data-driven approach to satisfy SLAs in multi-slice RANs with non-trivial delay QoS requirements.

Abstract

The need to satisfy the QoS requirements of multiple network slices deployed at the same base station poses a major challenge to network operators. The problem becomes even harder when the desired QoS involves packet delays. In that case, network utility maximization is not directly applicable since the utilities of the slices are unknown. As a result, most related works learn online the utilities of all slices and how to split the resources among them. Unfortunately, this approach does not scale well for many slices. Instead, it is needed to perform learning separately for each slice. To this end, we develop a bandwidth demand estimator; a network function that periodically receives as input the traffic of the slice and outputs the amount of bandwidth that its MAC scheduler needs to deliver the desired QoS. We develop the bandwidth demand estimator for QoS involving packet delay metrics based on a model-based reinforcement learning algorithm. We implement the algorithm on a cellular testbed and conduct experiments with time-varying traffic loads. Results show that the algorithm delivers the desired QoS but with significantly less bandwidth than non-adaptive approaches and other baseline online learning algorithms.

Figures (27)

• Figure 1: First, the BDE observes the state $X_i(t)$ of NS $i$ and estimates the PRBs $W_i(t)$ needed to deliver the desired QoS at slot $t$. Second, the NSM receives all bandwidth demands $\{W_i(t)\}_{i \leq n}$ and decides which ones to satisfy given the limited bandwidth at the BS and the SLAs of all NSs.
• Figure 2: At the start of slot $t$, the BDE observes the state $X(t)$ of the NS. Then, it estimates the number of PRBs $W(t)$ required to deliver the desired QoS throughout slot $t$ and allocates it to the MAC scheduler of the NS. At the end of slot $t$, it computes the cost $c(t)$ based on $W(t)$, the QoS feedback $Q(t)$ and the desired bound $Q_c$.
• Figure 3: The AMARI Callbox can emulate multiple LTE BSs and LTE CNs. Any commercial 3GPP compliant device, e.g., a smartphone, can connect to it via LTE once the Subscriber Identity Module (SIM) card is registered in the emulated LTE CN. The AMARI UE Simbox allows the emulation of tens of UEs without the need of new hardware devices. Both PCs are equipped with SDRs and antennas that allow them to communicate via LTE. The depicted setup uses 4 antennas on each PC which enables 2x2 MIMO.
• Figure 4: The Amarisoft PCs communicate over-the-air via LTE. Each of them has two Ethernet ports. The first port connects to the Internet. The second port connects directly to an Ubuntu PC. The Ubuntu PC runs Algorithm \ref{['ovalgo']} and configures the Amarisoft PCs. It issues ping and API commands to the Amarisoft PCs, and downloads and parses their traffic logs online.
• Figure 5: A small excerpt of a traffic log. It shows the traffic in the UE Simbox across the LTE stack for two subframes, i.e., $2$ ms. The payload of the IP packets can be read as well. The BDE parses online such logs in their .txt format. The logs are similar to Wireshark logs enriched with LTE metadata.