锐韬WiFi笔记2

Introduction

Commonly known as WiFi, the IEEE 802.11 standard was released back in the late 90s, with the aim to provide a low complex, and cost efficient, wireless connectivity solution. Currently in its 6th generation, the proliferation of WiFi has been driven by the constant revision of the standard, since periodic amendments have made possible to face the increasing requirements of newer use-cases. Wireless data services will continue to grow, with upcoming applications, such as virtual/augmented reality, video/game streaming and cloud based services, requesting vasts amounts of data with the most demanding throughput, latency, and reliability requirements. To address such expectations, the 802.11be Task Group (TGbe) was created in May 2019 to address the development of new specifications to fuel the upcoming WiFi 7.

Referred to as IEEE 802.11be Extremely High Throughput (EHT) [draft11be], this amendment aims to increase the WiFi throughput, while reducing the end-to-end latency and improving the reliability of communications [lopez2019ieee]. For such purpose, the Multi-link Operation (MLO111Throughout this paper, we will refer to the multi-band/multi-channel operation feature as the MLO, following the notation of the TGbe.) is considered a main candidate feature, as it promotes the use of multiple wireless interfaces to allow concurrent data transmission and reception in access points (APs) and stations (STAs) with dual- or tri-band capabilities.

Indeed, the interest in the use of the MLO framework is rapidly increasing. Latency in real-time applications has been already studied in [naik2021can, lacalle2021analysis, carrascosa2021experimental], showing that MLO is able to significantly reduce worst-case latency. Besides, authors in [lacalle2021analysis] extended their analysis to evaluate the reliability over multiple links, showing a high delivery rate when having multiple uncorrelated links. Also, an end time alignment mechanism to allow the use of parallel downlink transmissions through different links to stations without simultaneous transmit and receive (STR) capability is presented in [naribole2020simul]. Such approach is intended to maximize the spectrum efficiency. Analogously, an opportunistic backoff mechanism is proposed in [naribole2020simultaneous] to allow non-STR stations to resume their backoff timers, if an ongoing transmission is identified to not cause a collision. Authors in [yang2019ap] suggest that the use MLO per se may not be sufficient enough without coordination between APs, proposing a coordination framework to achieve high throughput requirements in high density areas.

The integration of a framework capable to operate at the same time over multiple wireless interfaces brings up new challenges and research opportunities. In this context, we find that MLO compliant devices will have the ability to transmit and receive packets with different quality-of-service (QoS) requirements over multiple links. Such functionality, which was not allowed in past amendments, is called traffic identifier (TID) to link mapping, and opens up to conceive new traffic management mechanisms. For instance, we may find all TIDs to be assigned to all links, allowing a full adaptive load balancing strategy, as traffic may be moved partially or fully between multiple links. On the contrary, other approaches may rely on having a dedicated link assigned to an specific QoS traffic, which implies a more rigid and less flexible load balancing solution, but ensuring that only traffic with the same QoS requirements share the same set of resources.

Fig. 1: Multi-link architecture and transmission modes representation. Each PHY color represents a different band/channel for each one of the different interfaces.

Multi-link operation

Architecture

The first architectural change is found in the redefinition of classical APs or STAs into the so-called multi-link capable devices (MLDs). Either AP MLDs or STA MLDs, refer to single devices with multiple wireless interfaces222Instead of interfaces, the TGbe defines them as affiliated AP/STAs. However, for sake of simplicity, and comprehensive purposes, we will keep referring to them as interfaces.. The most relevant aspect about that remains on the fact that MLDs will provide a unique MAC instance to the upper layers, without losing the independent parameters of each interface. To achieve that, TGbe proposed to divide the MAC sub-layer functionalities in two different levels [PatilEHT2019_2]. Figure 1 depicts the MLD architecture, representing both MAC sub-layer levels.

First, there is the upper MAC (U-MAC), which is a common part of the MAC sub-layer for all the interfaces. In the U-MAC, we find that link agnostic operations take place. We refer, for instance, to sequence number assignation, and MAC service data units (MSDUs) aggregation/de-aggregation. In this context, it is important to point out that the sequence number assignation must be performed at the U-MAC, since packets belonging to the same traffic flow can be fragmented and transmitted over different links. Such approach, then, eases the packet reordering at the receiver side. Additionally, common management functions for all links, such as setup, association and authentication take placed in this layer.

Below the U-MAC, we find the low MAC (L-MAC). This lower level, which is independent for each interface, is in charge of link specific functionalities like the channel access. In this context, we find that having individual L-MAC instances allow interfaces to keep their own channel parameters if needed. Inherently, this implementation also grants each interface to keep track of their own enhanced distributed channel access (EDCA) queues (one for each access category) to hold the traffic until its transmission. Other functionalities in the L-MAC layer are the management and control frame generation, as well as the MAC header creation and validation, when transmitting and receiving respectively [LevyEHT2020].

The motivation behind this two-tier architecture is to permit MLO-capable devices to move traffic from one link to another, being totally transparent to upper-layers. Hence, load balancing techniques may be useful to minimize the spectrum usage inefficiency of current standardized multi-band approaches, in which per client transmissions are only performed either in one band or another, by leveraging the use of all the available resources. However, such architecture entails a more complex design, requiring not only to design new methods to perform traffic to link allocation, if not also to rethink low-level aspects regar