Artificial neural network assisted polling scheme for central coordinated low-latency WLANs

Abstract

Focusing on uplink transmission, a low-latency medium access control scheme is proposed and investigated for central coordinated wireless local area networks (WLANs). By categorizing user-stations using a trained artificial neural network and avoiding polling empty user-stations during the polling period, effective polling of user-stations can be realized and network latency can be reduced. The artificial neural network is trained by samples collected from a request-based polling scheme, and the boundary for categorization decisioning is selected according to the calculated outputs of training samples. Compared to the conventional controlled channel access scheme, the proposed scheme achieves up to 32.9% reduction in network latency, and significantly improves the network capacity in delivering services requiring 200 μs average latency. Moreover, to improve service quality for user-stations delivering mission-critical services, a latency guarantee mechanism is incorporated. The incorporated scheme records the last access time for each of these user-stations and requires these user-stations to report their buffer statuses when they are in danger of exceeding their latency targets. Simulation results show that the incorporated scheme can ensure a delay below 500 μs for 99.9% of packets of these user-stations, enabling a WLAN to realize ultrahigh quality low-latency performance.

Introduction

LOW latency is always an important property that most telecommunication networks seek. With low-latency property, real-time data can be delivered over telecommunication networks and various delay-sensitive applications can be realized to enhance the interaction between human and environment, which prospers the development of modern society [1,2].

In recent years, delay-sensitive applications such as tele-diagnose/surgery, visual/augment reality, and industrial automation have attracted extensive interest as they are envisioned revolutionary and are expected to boost the life quality [3]. However, they require a stringent 1 ms end-to-end latency for the delivery of tactile/control/steering data, which exposes a big challenge to conventional telecommunication networks. To ensure 1 ms end-to-end latency, a network latency of 500 μs is desired as data need to travel both uplink and downlink, and a network latency toward 100 μs is preferred when considering the time budget for data processing [4]. Therefore, aiming at realizing such low latency, a lot of works have been reported to reduce network latency, especially in uplink direction.

As far as optical access networks (OANs) are concerned, a proposal using Bayesian estimator and maximum-likelihood sequence estimator during bandwidth allocation is proposed to achieve low latency [5]. An artificial neural network (ANN) based traffic burst status prediction scheme is proposed for OANs to optimize bandwidth allocation decisioning and latency performance [6]. Reinforced learning is utilized in OANs for maximum bandwidth optimization and latency reduction [7]. Users pairing, dedicated wavelength, service class based scheduling discipline, and double per priority queue method are investigated in OANs to ensure both low latency and service quality [8], [9], [10]. Also, future OAN roadmap has been envisioned to facilitate low-latency performance [11].

In the field of radio access networks (RANs), network slicing based resource allocation, multipath transmission, and erasure coding are adopted for low-latency and high-reliability fronthaul [12], [13], [14]. Maximum burst and data reception timing considered dynamic bandwidth allocation scheme is proposed for energy-efficient and low-latency RANs [15]. Transmission power and resource block allocation of RANs are optimized by greedy heuristic algorithm and near-optimal solution to achieve low latency [16]. Gated service medium access control (MAC) scheme is proposed for RANs to allocate each user sufficient bandwidth for low-latency transmission [17].

Although latency reduction for tactile/control/steering data delivery has been widely studied in OANs and RANs, there are still very limited investigations carried out in the field of wireless local area networks (WLANs). For WLANs, supporting real-time traffic is challenging [18]. So far, Saldana et al. investigated how frame size affects latency and throughput and showed that small frame size can help realizing 500 μs latency [19]. Feng et al. studied the feasibility of realizing 100 μs latency with IEEE 802.11 hybrid coordination function [20]. Wei et al. set a strict schedule for channel access and realized both 200 μs latency and high service quality [21]. Nevertheless, these proposals can only support very limited number of user-stations (uSTAs) and can only afford very low load. In our previous work, we proposed a request-based polling scheme named RPA for WLANs and achieved 200 μs latency with promising service quality under a 8-uSTA configuration [22]. However, it is still sensitive to uSTA quantity and therefore requires further improvement. Clearly, although WLAN is one of the most common networks in our daily life, it is still not ready to deliver future low-latency services.

In this paper, focusing on uplink direction, we propose an ANN assisted polling scheme named NNAP to facilitate future low-latency services delivery over the WLAN that multiple uSTAs are centrally coordinated by an access point (AP). In this scheme, uSTA polling list is decisioned based on the categorization result that obtained from the outputs of a trained ANN. Therefore, unnecessary polling of uSTAs can be avoided, and network latency can be reduced effectively. Service quality enhancement is investigated as well by incorporating a buffer status reporting based check mechanism. The main contributions of this paper are: (a) providing a novel NNAP scheme for central coordinated low-latency WLANs; (b) providing a boundary selecting method for ANN output categorization in the proposed NNAP scheme; (c) providing a latency guarantee mechanism for service quality enhancement of the proposed NNAP scheme; and (d) investigating the latency and service quality performances of the proposed NNAP scheme.

The rest of the paper is organized as follows. Section II briefly reviews the conventional MAC functions of WLAN and our previously proposed RPA scheme. Section III introduces and analyzes the proposed NNAP scheme. Section IV presents simulation results and discussions. A concise conclusion is conducted in Section V.