A Tutorial on Wi-Fi 8: The Journey to Ultra High Reliability

Abstract

Wi-Fi is evolving to support new areas such as extended reality, Industrial Internet of Things, and environments with numerous devices. Unlike previous generations that aimed to maximize peak and user-perceived throughput, Wi-Fi 8 (also known as IEEE 802.11bn Ultra High Reliability) intends to make Wi-Fi operation more reliable, i.e., to improve performance across the full range of scenarios, particularly in worst-case conditions. Specifically, 802.11bn is striving to increase throughput at various signal to interference and noise ratio levels, diminish the 95th percentile of latency, decrease packet losses, and reduce power consumption. To attain these ambitions, Wi-Fi 8 is going to introduce a series of groundbreaking mechanisms, including distributed-tone resource units, unequal modulations, prioritized and non-primary channel access, advanced roaming features, multi-AP coordination, and so on. We present the current status of Wi-Fi 8 development with an emphasis on the targets and enabling technologies. This tutorial also discusses open research directions that require the development of new intelligent methods and algorithms.

1 Introduction

Over the past three decades, Wi-Fi has become an essential part of our daily lives. Wireless Local Area Networks (WLANs) continue to develop and grow rapidly. By 2030, Wi-Fi is expected to generate $18 billion in revenue, reflecting a more than fourfold increase from 2021 [1]. Wi-Fi’s ubiquity spans numerous sectors, residential and enterprise, enhancing remote work capabilities, customer experiences, and operational efficiency. Industries such as retail, education, transportation, healthcare, and industrial automation increasingly adopt Wi-Fi and rely on its capabilities [2, Doc. 23/0079r10].

However, despite the widespread adoption, current Wi-Fi often falls short in certain conditions, e.g., in interference-overcrowded environments, which lead to service disruptions, or during client movement, causing long adaptations or re-connection. As a result, enhancing reliability and stability remains critical for future development. The upcoming eighth generation of Wi-Fi (Wi-Fi 8) shall ensure that Quality of Service (QoS) requirements are satisfied across the full range of scenarios [3].

Looking ahead, Wi-Fi aims to cover emerging use cases with much more stringent requirements, such as industrial automation and eXtended Reality (XR), which exceed the capabilities of current WLANs; see [4] and [2, Doc. 23/0079r10]. To satisfy these demands, the next 8th generation of Wi-Fi will incorporate breakthrough mechanisms that help achieve high throughput, extremely low latency, near-perfect reliability, and reduced power consumption; see [3, 4] and [2, Doc. 23/0480r3].

The development of Wi-Fi 8 centers around the new IEEE 802.11bn standard amendment, initiated by the Ultra High Reliability (UHR) Study Group (SG) and continued by the UHR Task Group (TG), briefly named TGbn. Following the timeline of previous Wi-Fi generations, Wi-Fi 8 is going to enter the market by 2027–2028 and face the scenarios of the 2030s and beyond (IMT-2030 [5]). This means that Wi-Fi 8 will compete for the market with the 5G-Advanced and 6G [6] under the 3GPP Release 20 Specification, so TGbn aims to make 802.11bn competitive technology against the cellular systems.

Since its inception, Wi-Fi 8 has garnered much attention in several tutorials and research papers. The tutorials [3, 4, 7] give a brief overview of the UHR activities. However, since then, the route of UHR has taken several drastic turns, rendering the information in these white papers and tutorials outdated. For example, they lack several features (e.g., distributed-tone resource units, dynamic bandwidth management, and unavailability reporting) and mention other unrelated features (e.g., full duplex transmission and constellation shaping). The research papers address individual 802.11bn mechanisms and new solutions for potential integration into Wi-Fi 8. Specifically, they cover:

• coordination of multiple Access Points (Multi-AP coordination) [8,9,10];

• time-sensitive networking (TSN) functionalities [11];

• seamless roaming concepts [12, 13];

• power save mechanisms [14];

• novel channel access schemes for Wi-Fi 8 [15].

These papers mostly focus on specific narrow improvement areas, lacking a panoramic and multifaceted analysis of new features.

In this tutorial, we spotlight the 802.11bn amendment for Wi-Fi 8, offering a detailed analysis of its features in the context of IMT-2030 target scenarios and corresponding key performance indicators (KPIs). We present candidate solutions for Wi-Fi 8 and discuss them. Our primary contribution is an exploration of enabling technologies and open research directions of Wi-Fi 8.

The rest of the paper is organized as follows. Section 2 presents the evolution of Wi-Fi, highlighting its journey and current open issues. Next, in Section 3, we describe the development timeline of 802.11bn, next-generation scenarios, and their key requirements, along with the formal objectives. Section 4 provides an at a glance overview of potential Wi-Fi 8 mechanisms. Then, we thoroughly examine crucial candidate features and open research directions in Sections 5–7, dedicated to the Physical Layer (PHY), the Medium Access Control Layer (MAC), and coordination of multiple Wi-Fi networks, respectively. Next, Section 8 briefly discusses machine learning within 802.11. Finally, Section 9 concludes the tutorial.

Table 1 presents the acronyms used throughout the paper, excluding local ones, i.e., which are defined and used within a single paragraph.

Table 1. List of acronyms.

2 Evolution of IEEE 802.11

Since its inception in 1997, Wi-Fi, governed by IEEE 802.11, has evolved from a modest technology with a maximum data rate of 2 Mbps in the 2.4 GHz band into an indispensable component of modern multi-gigabit connectivity, enabling a multitude of applications and devices. Below we describe the development of its main milestones, relevant for Wi-Fi 8, and corresponding features.

2.1 IEEE 802.11a and IEEE 802.11g

The first major leap comes with the 802.11a amendment, which builds a PHY basement for almost all the following amendments. It describes operation in the 5 GHz band at data rates up to 54 Mbps. Its most notable feature is orthogonal frequency-division multiplexing (OFDM), which, unlike previous single-carrier schemes, combats high-frequency attenuation, narrowband interference, and frequency-selective fading caused by multipath. 802.11a frame body consists of 4 μs OFDM symbols including 0.8 μs guard interval, each using 52 of 64 subcarriers (also called tones): 48 QAM-modulated data tones and 4 pilot tones for phase correction, while the remaining ones serve as edge and guard tones.

802.11g transfers OFDM to the 2.4 GHz band. Notably, it is the first amendment to face challenges related to compatibility with legacy devices and their heritage modulations defined in the 802.11 standard, such as complementary code keying and phase shift keying with direct-sequence spread spectrum (DSSS).

2.2 IEEE 802.11e

The 802.11e amendment significantly enhances the original standard by introducing QoS capabilities [16]. It focuses on supporting differentiated services and thus extends the legacy Distributed Coordination Function (DCF) with two mechanisms: Enhanced Distributed Channel Access (EDCA) and Hybrid Coordination Function (HCF) Controlled Channel Access (HCCA). EDCA, the more widely implemented approach, provides the differentiated services via the contention-based channel access and traffic prioritization. Specifically, it classifies frames into four distinct access categories (ACs) for voice, video, best effort, and background traffic. EDCA then leverages differentiated channel access parameters for these ACs, whereas DCF uses the same parameters for all frames. For each AC, EDCA specifies:

• an arbitration interframe space (AIFS) that replaces a common distributed interframe space (DIFS);

• minimum and maximum contention window sizes for a random backoff;

• a transmission opportunity (TXOP) limit that defines a maximum time interval, during which a device can perform frame exchanges.

This prioritization scheme provides better service for time-sensitive applications while maintaining backward compatibility with DCF. Although HCCA by design furnishes polling-based channel access with guaranteed service intervals through Traffic Specifications (TSPECs), its overhead and complexity have limited its practical adoption, leaving EDCA as the primary QoS mechanism.

802.11e also brings in the Block Acknowledgment (Block Ack) mechanism that allows multiple data frames to be acknowledged with a single Block Ack frame. Instead of requiring an individual ACK for every single frame, the recipient can collect multiple frames and then send a bitmap back to the sender indicating which frames were received correctly. This significantly reduces protocol overhead and enhances throughput for bulk data transfers.

2.3 IEEE 802.11n

802.11n, subsequently named Wi-Fi 4, marks a tenfold increase in nominal data rates up to $600$ Mbps in both $2.4$ GHz and $5$ GHz [17]. This improvement is mainly achieved by two innovations in PHY, namely the Multiple Input Multiple Output (MIMO) technology, which allows communicating in up to four spatial streams (SSs), and twice as wide $40$ MHz channels. To provide compatibility with devices operating in $20$ MHz channels, 802.11n introduces the channel bonding rules. Specifically, the primary $20$ MHz channel is defined in each network. Then, having won a contention in the primary channel, a device can expand the operating bandwidth by concatenating the secondary $20$ MHz. 802.11n also adds a higher code rate, brings in a shorter $0.4$ μs guard interval, and increases the number of (data $+$ pilot) tones used: up to $52 + 4$ in a $20$ MHz channel and up to $(54 + 3) \times 2$ in a $40$ MHz channel. Each of these features slightly increases nominal data rates. In addition to the legacy binary convolutional codes, 802.11n introduces more efficient low-density parity-check (LDPC) codes, which are optional in Wi-Fi 4 yet have become mandatory starting from Wi-Fi 6. To further combat overhead, 802.11n MAC introduces two types of aggregation of packets into a single frame, thus eliminating the need for multiple preambles and interframe spaces: an aggregated MAC Service Data Unit (A-MSDU) and an aggregated MAC Protocol Data Unit (A-MPDU). A-MSDU uses a solitary MAC header and checksum in the entire frame, while A-MPDU encapsulates each aggregated packet with a MAC header and checksum, implying more overhead but allowing independent packet decoding. Finally, with A-MPDU, 802.11n leverages Block Ack frames introduced in 802.11e, which acknowledge the correctly received packets in A-MPDU with further retransmissions of the corrupted ones.

2.4 IEEE 802.11ac

802.11ac (Wi-Fi 5) continues the throughput increase trend, pushing nominal data rates to $7$ Gbps [18]. First, it allows for even wider channel bonding up to $160$ MHz, i.e., a device can sequentially add the idle secondary $20$ MHz, $40$ MHz, and $80$ MHz channels to its primary. However, if the primary channel is busy, while the rest of the spectrum is idle, it cannot be used, which limits 802.11ac performance in dense networks. As the $2.4$ GHz band does not have spectrum even for a $80$ MHz channel, 802.11ac operates only in $5$ GHz. Also, 802.11ac allows using non-contiguous $80+80$ MHz channels.

Next, the maximal number of SSs is increased to eight. Moreover, since client devices usually do not have so many antennas, 802.11ac introduces downlink (DL) multi-user (MU) MIMO. With MU-MIMO, an access point (AP) can assign SSs to up to four STAs in parallel. 802.11ac also increments the QAM constellation order and begins to leverage $256$-QAM, yielding a $33\%$ data rate boost. Notably, to accommodate the increased data rates and to further reduce overhead, the maximum frame length is scaled up to $4.6$ MB from $65$ KB in 802.11n.

2.5 IEEE 802.11ah

The 802.11ah amendment, known as Wi-Fi HaLow, aims to address the burgeoning Internet of Things (IoT) market [19,20,21]. It operates in the sub-$1$ GHz bands and uses narrow channels ranging from $1$ MHz to $16$ MHz. Although not commercially successful yet, Wi-Fi HaLow introduces valuable concepts. One of its novelties is the Target Wake Time (TWT) mechanism [21], which is crucial for battery-supplied sensors and is upgraded in the following standard amendments. With TWT, a station (STA) negotiates with an AP a set of Service Periods (SPs), during which they may communicate, consequently, the STA can turn off its radio outside the TWT SPs to save energy.

2.6 IEEE 802.11ax

Since 802.11ax (Wi-Fi 6), the focus has shifted from increasing nominal data rates to improving user-perceived throughput and quality of experience; see [22, 23] and [2, Doc. 14/0165r1]. To enhance real-world performance, 802.11ax introduces several features related to spectral efficiency. The major one is the orthogonal frequency-division multiple access (OFDMA), which enables flexible frequency allocation. It allows an AP to split the frequency channel into smaller pieces, called Resource Units (RUs), and allocate them to different STAs simultaneously so that each STA gets at most one RU (of possibly various widths). Table 2 shows the maximum number of RUs for each channel bandwidth available in Wi-Fi 6. To empower OFDMA, 802.11ax changes the OFDM numerology making symbols four times longer and the number of tones four times greater. A positive side effect of this update is that the relative guard interval overhead drops to about $6\%$ for the new $12.8$ μs symbol and the $0.8$ μs guard interval. Also, 802.11ax introduces $1024$-QAM modulation, which is $25\%$ faster than $256$-QAM used in 802.11ac, and broadens Wi-Fi capabilities to the $6$ GHz band.

Table 2. Maximum number of identical RUs for each channel bandwidth in Wi-Fi 6.

To enable uplink (UL) OFDMA and UL MU-MIMO, another novel 802.11ax feature, Wi-Fi 6 brings in Trigger frames (TFs) that initiate and synchronize the UL MU transmission sequence, as well as carry its parameters: Modulation and Coding Scheme (MCS), power, duration, etc. Moreover, TFs can be used to implement centralized trigger-based (TB) channel access. Specifically, the AP can deprioritize or even disable random channel access for all associated STAs by advertising the MU EDCA parameter set and start transmissions only with TFs [24]. Note that TB channel access finally seems to be a success over previous 802.11 attempts (such as 802.11s reservations, 802.11ah Restricted Access Window (RAW), etc.) to implement long-awaited reservation-based channel access. To improve power efficiency for STAs, 802.11ax also expands the TWT mechanism from 802.11ah to be used in conjunction with OFDMA transmissions and TFs.

Importantly, dense deployments have been in the focus of 802.11ax developers [2, Doc. 14/0165r1] sparking many mechanism ideas and debates within the 802.11 Working Group. As a result, 802.11ax moved from the restriction of possibly colliding transmissions in neighboring networks (Basic Service Sets, BSSs) to the Spatial Reuse (SR) paradigm, i.e., simultaneous transmissions of multiple overlapping BSSs (OBSSs). The key features of the SR-based operation are as follows.

First, 802.11ax introduces BSS Coloring defining unique identifiers (“colors”) that help a device easily distinguish traffic from its BSS from that from OBSSs. Second, 802.11ax uses two Network Allocation Vectors (NAVs) to maintain separate channel reservation timers for intra-BSS and inter-BSS traffic, optimizing medium access efficiency. Third, 802.11ax implements quiet time periods to silence the STAs while allowing for dedicated peer-to-peer (P2P) communication. Notably, quiet time periods are not the same as quieting and quiet intervals from 802.11h. The latter are used to silence each STA in the BSS so that the AP can listen for radar signals and then perform a channel switching decision.

Another important feature of SR is the OBSS Packet Detect (OBSS PD) mechanism, which allows a device to transmit concurrently with OBSS transmissions as long as its own transmission does not damage the ongoing one. Specifically, 802.11ax introduces an adjustable OBSS PD threshold to detect signals from OBSSs. If the signal strength is below this threshold, the device considers the medium idle and can transmit with a limited power. Namely, 802.11ax allows increasing the OBSS PD threshold if the transmit power is decreased by the same value in dB, which ensures that the interference is the same as if devices were located far from each other. Moreover, 802.11ax allows several BSSs to establish a Spatial Reuse Group (SRG) operating with an additional SRG OBSS PD threshold. Simultaneously having two thresholds, SRG and non-SRG, the device can follow different SR strategies for transmissions within its SRG and outside. If SRG STAs can tolerate some interference, the SRG OBSS PD threshold can be increased to enable more aggressive channel access. On the contrary, if the traffic requires a more reliable service, lowering the SRG OBSS PD threshold decreases the level of interference in SRG and helps avoid possible collisions.

To overcome 802.11ac’s channel bonding constraint—where if the secondary $20$ MHz is occupied, a device can use only the primary $20$ MHz channel but not the other segments of a wide channel—802.11ax introduces preamble puncturing. This feature allows selective exclusion (“puncturing”) of some busy $20$ MHz subchannels within $80$ MHz or wider channels during MU OFDMA transmissions, while still using available spectrum segments. By dynamically bypassing only the occupied subchannels (e.g., skipping a $20$ MHz subchannel in an $80$ MHz channel while using the remaining $60$ MHz), 802.11ax significantly improves throughput in interference-crowded deployments where narrowband OBSS traffic would otherwise block entire wide channels.

Note that many successors to 802.11ax SR features have found their niche in 802.11bn, albeit in a more centralized, and thus potentially more effective, implementation.

Finally for 802.11ax, let us highlight that Wi-Fi 6’s ambition is fourfold improvement in user-perceived throughput [2, Doc. 14/0165r1] thanks to the mechanisms described, while at PHY the maximum nominal data rate is only $37\%$ higher than that in 802.11ac (reaching up $9.6$ Gbps).

2.7 IEEE 802.11be

Wi-Fi 7 (802.11be) [25, 26] aims to support real-time applications requiring high throughput and low latency. 802.11be increases the maximum bandwidth to $320$ MHz, available only in the $6$ GHz band, effectively doubling nominal throughput. It also introduces $4096$-QAM, which is $20\%$ faster than $1024$-QAM. All these features combined give a data rate of up to $2.88$ Gbps per SS, or at max $23$ Gbps with eight SSs [27].

To provide the flexibility of using even more spectrum simultaneously and to eliminate the limitations of legacy channel bonding, 802.11be introduces Multi-link Operation (MLO). This key MAC feature allows multi-link devices (MLDs) to exchange data on multiple frequency channels simultaneously. Many papers show that MLO improves throughput, reliability, latency, and jitter by enabling load balancing and frame duplication across multiple links [28, 29]. The 802.11be amendment also supports additional MLD versions, which represent different states of a tradeoff between hardware complexity and performance. Enhanced multi-link single-radio (EMLSR) non-AP MLD can listen on multiple links but exchange frames in one link at a time. Enhanced multi-link multi-radio (EMLMR) non-AP MLD operates simultaneously on multiple links, but can remap its auxiliary antennas to achieve more SSs and use faster MCSs at the link, where an AP transmitted a specific TF.

Additionally, 802.11be brings in the concept of multiple resource units (MRUs) [27], which addresses inefficiencies in the RU allocation scheme of 802.11ax. Specifically, in Wi-Fi 6, an AP cannot assign RUs to STAs with an arbitrary width ratio, e.g., the AP cannot divide an $80$ MHz channel with a $3:1$ ratio for two STAs because a $60$ MHz RU is not defined. Also, even half-dividing of channels in 802.11ax leaves a piece of the frequency band unused, e.g., a $20$ MHz channel splits into two $106$-tone RUs yet with a single remaining $26$-tone RU (${\sim}\,2$ MHz) that may be unexploited, as shown in Table 2. Thus, Wi-Fi 7 defines some new RUs, called MRUs, which consolidate several adjacent regular RUs. For example, a $484+242$-tone MRU combines two regular RUs into a $60$ MHz RU. As a result, MRUs solve (most of) the described problems, make OFDMA more flexible, and enable preamble puncturing for single-user (SU) transmissions. However, the combinations of 802.11ax RUs in MRUs are strictly predefined, as well as still only one MRU can be assigned to an STA.

802.11be also introduces Triggered TXOP Sharing (TXS), which allows an AP to allocate a portion of its TXOP to a specific STA for data transmission either back to the AP or directly to another STA. The TXS capability differs from the reverse direction protocol, pioneered in 802.11n, which does not allow multiple STAs to share the allocated time and cannot allocate only a part of the remaining TXOP to be used. Thus, the TXS is beneficial for P2P communications because the AP can protect the associated STAs’ transmissions from interference within the BSS.

Another vital feature is Restricted Target Wake Time (R-TWT) [30], an extension of the TWT mechanism from 802.11ah and 802.11ax. With R-TWT, when an AP and a group of STAs negotiate an R-TWT SP schedule, all STAs in that BSS are prohibited from crossing the R-TWT SP start times with transmissions, which explains an introduced “restrictiveness.” In this way, R-TWT purports to clear the beginning of the R-TWT SP from transmissions, thus allowing devices to contend for the channel without delay. Consequently, an STA with time-sensitive traffic is highly likely to access the medium and complete its urgent delivery within the R-TWT SP.

Several promising features, such as Multi-AP coordination, were initially planned for Stage 2 of 802.11be but were postponed due to implementation complexity and development time constraints. These features hold significant potential for performance gains but require much effort to implement them.

2.8 Open Issues and Technical Limitations

Despite significant advances in the evolution of Wi-Fi, its performance still falls short in certain areas. The problems arise due to overcrowding of the spectrum and increasing demands of users in everyday scenarios, as well as innovative use cases that require sophisticated QoS, e.g., ultra reliability in the industrial domain.

Wireless technologies suffer from unstable connections and unreliable data delivery, which presents a significant hurdle to achieving seamless Wi-Fi operation [3]. Initially designed for the unlicensed spectrum, Wi-Fi was built with an acceptance of non-zero packet loss ratio (PLR) and collisions. However, as the number of devices increases, the amount of interference grows exponentially, exacerbating reliability issues. Notably, Cisco refers to interference as a silent killer of Wi-Fi performance [31]. Because of the swelling amount of interference, legacy mechanisms that were once satisfactory no longer meet today’s demands of throughput, latency, and reliability in dense environments.

Moreover, people’s demands are also constantly growing: we are becoming more and more finicky. Nowadays even minor disruptions, such as spikes in video or audio streaming or momentary drops in video resolution below “$720$p”, significantly affect user perception. Besides, current Wi-Fi capabilities fall short of serving the emerging applications in scenarios, such as XR and real-time video conferencing or gaming, which demand latency of a few dozens of ms and user-perceived throughput up to a few Gbps. As Wi-Fi expands into emerging industrial machine-to-machine scenarios, the challenges of reliability and latency become even more pronounced.

Furthermore, some targets of previous Wi-Fi generations have not been met. For example, to achieve deterministic delivery and bounded low latency, Wi-Fi 7 introduced R-TWT. However, its operation has some concerns: R-TWT restrictions have effect only within the BSS. STAs from OBSSs may trespass the start of R-TWT SPs, which can result in collisions, packet loss, and increased delivery time [2, Doc. 23/0226r1]. Moreover, recent experiments show that existing equipment will likely disrupt the R-TWT operation [32]. Another direction of 802.11be was the integration of the TSN paradigm into Wi-Fi. Many already existing Wi-Fi features can be harmonized with the TSN logic, e.g., scheduled channel access with TFs, traffic classification even from 802.11e to enable prioritized packet treatment, and Fine Time Measurement protocol to enable tight synchronization [2, Doc. 21/0628r0]. However, Wi-Fi still lacks a frame preemption technique to allow time-critical packets to interrupt a long frame transmission and achieve immediate channel access [33], which, however, raises a question of a tradeoff between performance of background and time-critical traffic. In addition, current Wi-Fi versions are almost incapable of providing zero-loss seamless inter-BSS mobility, which is an essential part of TSN scenarios [33].

Given these limitations, Wi-Fi is expected to refine its mechanisms and launch new tools to improve its overall performance. Thus, the long journey of Wi-Fi shall reach new horizons of reliable operation.

3 Timeline, Scenarios, and Targets of Wi-Fi 8

The next generation of Wi-Fi continues to improve network performance in many directions, expanding into new scenarios and targets. We first present the development timeline for Wi-Fi 8. To cover the whole variety of improvement directions, we address the novel Wi-Fi use cases and market requirements. Finally, we focus on KPIs that the next Wi-Fi generation will aspire to improve.

3.1 Wi-Fi 8 Timeline

Looking back at the development timeline of previous amendments, 802.11ax and 802.11be, the complete path for 802.11bn is expected to span $5$–$6$ years [2, Docs. 22/0046r1, 22/0059r0], as shown in Fig. 1.

Fig. 1.

Standard amendments’ development timeline.

The UHR SG, responsible for the scope of 802.11bn, commenced its operation in September 2022 with the goal of formulating requirements for future scenarios and devising mechanisms to satisfy them. Then, TGbn (aka the UHR TG) was created in July 2023, with its first meeting held in November 2023. This group focuses on technologies to enhance WLAN reliability, increase throughput, reduce latency, and lower power consumption [2, Doc. 22/0708r0]. Draft 0.1 was completed in February 2025, D0.3 in June 2025 [34], and Draft 1.0 is expected by September 2025. Drafts 2.0 and 3.0 are awaited by May 2026 and January 2027, respectively. After 802.11 approval, 802.11bn could be published (“Pub.” in Fig. 1) in 2028. Note that the devices that are capable of Wi-Fi 8 features are likely to enter the open market early, e.g., in 2027, ready to tackle scenarios of the 2030s, including IMT-2030 topics of interest; see [2, Docs. 22/0961r0, 21/1809r0] and [5].

3.2 Wi-Fi 8 Scenarios

Let us distinguish three groups of scenarios that Wi-Fi 8 is going to cover. Table 3 lists their corresponding numerical KPIs.

Table 3. Wi-Fi 8 scenarios and the corresponding KPIs.

Extended reality. One of the most eagerly anticipated and rapidly developing areas is eXtended Reality (XR), encompassing augmented reality, virtual reality, and mixed reality. XR-related applications—such as high-quality video streaming to wireless headsets, immersive gaming and training, metaverse experiences, and work with digital twins—demand high bitrates; see [2, Docs. 22/0030r1, 22/0059r0, 22/0697r0]. Complex metaverse environments are often rendered remotely in a cloud and need to be delivered with low latency. Typical XR requirements [2, Doc. 22/0952r0] include:

• throughput per device up to a few Gbps;

• latency below $20$ ms;

• PLR below $10^{-5}$.

802.11be has already targeted XR scenarios, leveraging features such as MLO and R-TWT. Namely, Wi-Fi 7 manages to satisfy strict QoS requirements of XR for a few devices with good channel conditions [35]. However, challenges remain when low latency and high bitrate use cases encounter environments with numerous real-time connections and mediocre channel quality, i.e., when multiple transmissions occur simultaneously in multiple networks. Thus, Wi-Fi still needs more comprehensive and robust latency-aware solutions.

Industrial IoT. The Industrial Internet of Things (IIoT) scenarios include robotics, industrial automation, smart agriculture, and monitoring systems. Cheap industrial devices, like sensors and actuators, individually do not require much throughput, but together they generate many packets [24]. IIoT applications require extremely reliable connections, comprehensive coverage [2, Doc. 22/1493r1], and stable handover for moving devices [2, Doc. 22/1919r5]. IIoT systems generally demand [33]:

• millisecond latency or less with a minimal jitter;

• PLR in the range of $10^{-7}$ to $10^{-3}$;

• strict synchronization and deterministic message delivery.

Moreover, initial attempts to integrate TSN with Wi-Fi, which is essential for industrial applications, have faced challenges, as described in Section 2.8. The upcoming 802.11bn aims to address these latency and reliability issues to satisfy IIoT requirements.

High-density deployments. The ubiquity of Wi-Fi in homes, offices, and smart IoT environments creates high-density deployments with numerous OBSSs [2, Doc. 22/0030r1], severely impacting network performance [2, Doc. 22/1809r0]. Typical scenarios include streaming high-resolution videos or playing real-time games in the coexistence of third-party load. The high-level ambition here is to provide the same quality of operation for multiple devices as experienced by a single device. The corresponding requirements are:

• per-user data rates from a few Mbps to sub-Gbps [2, Doc. 22/1919r5];

• latency of a few dozen milliseconds for real time [2, Doc. 22/0030r1];

• PLR below $10^{-3}$.

Efficient collision-free channel access, better management of direct and infrastructure links [2, Doc. 22/0046r1], comprehensive coverage, and reduced power consumption are crucial for these scenarios. Current Wi-Fi capabilities are insufficient, necessitating the aforementioned advancements in the next generation.

3.3 Wi-Fi 8 KPIs

Let us summarize the KPIs that shall be achieved by the upcoming 802.11bn technology.

Throughput. Most of the scenarios require throughput improvements. However, the developers of 802.11bn emphasize that boosting the maximum nominal throughput does not convert into real-life experience enhancement, especially for edge users. Many devices have a low or medium signal to interference and noise ratio (SINR) or operate in dense deployments, so the new amendment will target practical throughput in such non-optimal conditions.

Some studies predict user throughput demands of about $2$ Gbps by 2030s [2, Doc. 22/0694r0], and certainly, the goal is to provide this throughput benchmark under all range of conditions.

Latency. Worst-case latency and jitter have first become a focus in Wi-Fi 7 and continue to be a valuable concern in Wi-Fi 8, especially in the context of XR and IIoT [2, Docs. 22/1919r5, 22/0779r0]. Although Wi-Fi 7 was anticipated to solve latency issues, many relevant features were deferred due to complexity and lack of consensus, while the techniques included in the 802.11be standard work only in some cases.

In particular, R-TWT, which was purposefully created to achieve limited latency, suffers from being “non-mandatory”. The R-TWT rules may be violated both by Wi-Fi 7 and legacy devices. MLO is useful to reduce latency, but it cannot bound latency. Hence, proposals for latency reduction, especially in OBSS scenarios, are still needed in 802.11bn.

Another source of latency-related issues is roaming (aka handover), i.e., transition of an STA from one AP to another. Maintaining uninterrupted connections and seamless communication is crucial for emerging use cases. However, long handover procedures for mobile devices increase packet delivery delays beyond the limits [2, Doc. 22/1874r2], making it impossible to satisfy strict QoS requirements. Therefore, the support for enhanced mobility and roaming has garnered special attention for 802.11bn.

Reliability. Wi-Fi 8, as its name “Ultra High Reliability” suggests, puts reliability at the forefront of all scenarios. Let us offer a dual interpretation of the reliability term.

First, we define reliability as the loss rate, usually PLR, that Wi-Fi 8 aims to reduce in order to accommodate emerging applications. For instance, IIoT scenarios involve cooperative robot control, where loss of commands—which may occur due to interference, channel shadowing and fading, roaming, etc.—can lead to equipment malfunction. Lowering PLR in these contexts directly means a reliability enhancement.

Second, we associate reliability with the ability of Wi-Fi to maintain the required performance under worst-case conditions. The intent is to ensure that KPIs—such as throughput and latency—are satisfied reliably across diverse scenarios, including adverse conditions: low SINR range and OBSS presence [2, Doc. 23/0028r6].

Power consumption. Another overarching goal of Wi-Fi 8 is to reduce the power consumption, which is supported by a great demand for cheap and easy-to-manage battery-powered devices. This goal shall be addressed by new power saving mechanisms, as well as by further improvement of the existing ones.

Also, residential and enterprise users are concerned about the global environment and energy consumption [2, Doc. 23/1254r1]. Hence, they seek to improve the power efficiency of their household appliances. Hence, AP devices are also proposed to include power save (PS) modes, which are missing in Wi-Fi 7 [2, Doc. 23/0244r2].

3.4 Wi-Fi 8 Targets

To provide a concise representation of the Wi-Fi 8 targets, we refer to the Project Authorization Request (PAR) and the Criteria for Standards Development (CSD) for 802.11bn documents [2, Docs. 23/0480r3, 23/0079r10], a description of the amendment scope, its purpose, and needs. For the benchmarking, PAR states that current 802.11be implementations achieve multi-Gbps throughput, sub-$10$ ms latency, and less than $0.1\%$ packet loss.

802.11bn shall add Ultra High Reliability capabilities to WLANs for both isolated BSS and OBSS scenarios. Specifically, in comparison with 802.11be, 802.11bn targets to:

• increase throughput by $25\%$ at different SINR levels, referred to as the rate-vs-range performance improvement;

• reduce the $95$th percentile of latency by $25\%$;

• decrease MAC protocol data unit (MPDU) loss by $25\%$, especially during transition between BSSs.

Additionally, Wi-Fi 8 shall enhance PS for APs (including mobile APs) and tweak P2P operation compared to 802.11be.

4 Wi-Fi 8 at a Glance

To achieve the ambitious goals of Wi-Fi 8, 802.11bn is expected to provide a wide range of groundbreaking mechanisms. This section provides a concise overview of Wi-Fi 8 features, grouped into three categories: PHY, MAC, and Wi-Fi networks coordination, while the detailed description of these features is given in Sections 5, 6, and 7, respectively. Note that, to spotlight the full landscape of Wi-Fi 8, we also discuss relevant features that were not included in the draft but were actively discussed during the development of 802.11bn. Table 4 maps Wi-Fi 8 features (including candidate ones) from these groups to the KPIs introduced in Section 3.3, as well as explicitly marks which features have made it into the 802.11bn draft.

Table 4. Wi-Fi 8 candidate features mapped to KPIs.

4.1 PHY Candidate Features

Unlike previous Wi-Fi generations, 802.11bn introduces no quantitative PHY improvements toward higher nominal data rates. Although adding more antennas—and thus more SSs—can improve throughput, the affiliated cost and complexity are limiting factors. Moreover, increasing the number of SSs beyond eight significantly raises sounding overhead. 802.11bn prioritizes performance in the low-to-mid SINR range and, therefore, has no incentive to use modulations above the existing $4096$-QAM. Additionally, extending the channel width beyond $320$ MHz in the sub-$7$ GHz bands is hardly feasible in real-world scenarios.

The UHR SG initially elaborated on expanding UHR operation into the millimeter-wave (mmWave) band to get more spectrum. This initiative has been continued by the separate Integrated mmWave TG in the corresponding 802.11bq amendment. Following the example of 802.11bq, the Enhanced Light Communication SG is exploring the use of the light spectrum. According to some sources (e.g., [7]), mmWave could become part of Wi-Fi 8 in a broad sense, which is why we briefly cover these alternative spectrum horizons in this tutorial.

Transmission reliability improvements, the central focus of 802.11bn, and its associated rate-vs-range tradeoff are addressed through multiple features. One notable innovation is distributed-tone RUs (DRUs). DRUs avoid regulatory power spectral density (PSD) limitations by distributing tones and thus achieve higher transmission power across a wider bandwidth, thereby improving SINR for multiple STAs in UL transmissions. To help receivers detect and suppress interference, 802.11bn also allows adding interference mitigation (IM) pilots, further improving transmission reliability. Among the features of Wi-Fi 8 is unequal modulation (UEQM), which enhances reliability by applying different modulations across SSs. Originally introduced in 802.11n but later removed in 802.11ac due to implementation complexity, UEQM is now regaining attention in light of stricter reliability demands. It naturally couples with the newly introduced MCSs, which are derived from existing modulations and coding rates to further improve rate-vs-range. Wi-Fi 8 also enhances LDPC efficiency by doubling the maximum codeword length. Finally, 802.11bn defines the Enhanced Long Range (ELR) frame format, which promises to improve performance of edge users thanks to the quadruplication of data and a specialized preamble.

Even though the number of SSs remains unchanged, developers of 802.11bn try to improve the accuracy of beamforming (BF) and MIMO precoding and optimize the sounding procedure. Various techniques for channel state information (CSI) compression, such as channel impulse response feedback, variable angle quantization, and implicit feedback, have been proposed to reduce BF feedback overhead. The AP can recommend STAs precoders for TB UL transmission to increase SINR at the AP. Additionally, smooth BF techniques are suggested to mitigate phase response discontinuities at various CSI compression stages, thereby improving reliability in high-noise environments.

4.2 MAC Candidate Features

As for MAC, TGbn discusses several mechanisms to implement low-latency channel access. The first one is prioritized EDCA (P-EDCA) that allows an STA with an urgent frame to access the channel with reduced contention, thereby cutting the delay distribution tail. To accomplish this, the STA sends a defer signal, which temporarily excludes all STAs with non-urgent traffic from contention. The second one is non-primary channel access (NPCA) that enables gaining a TXOP in a secondary channel if the primary one is busy. The third channel access modification discussed for UHR is preemption, a method to stop an ongoing transmission and start a new urgent one so that the latency constraint can be satisfied. Then, to enhance low-latency P2P communication in the base channel, 802.11bn introduces group-based TXOP allocations, allowing multiple P2P STAs to communicate within a shared portion of a TXOP. Also, to reduce interference, Wi-Fi 8 enables an AP to coordinate P2P operation in the off-channel, i.e., in the channel outside the AP’s bandwidth. Finally, a set of changes improves the QoS signaling by increasing the maximum queue size in Buffer Status Reports (BSRs), flexibly repurposing AC indices for different traffic classes, enhancing queue management, and introducing tools to stipulate strict latency requirements.

To improve spectrum utilization in Wi-Fi networks, 802.11bn introduces several dynamic bandwidth management techniques. First, dynamic subband operation (DSO) allows an AP with a wider bandwidth than its associated STAs to assign them frequency resources beyond their operational bandwidth, excluding the primary channel. Second, dynamic bandwidth expansion (DBE) enables an AP to temporarily increase the BSS bandwidth when conditions permit.

Unavailability reporting is a new mechanism in UHR networks that allows STAs to notify peers about time intervals when they are unavailable due to in-device coexistence (IDC) with other technologies. Peers can leverage this information to avoid transmitting frames during these times, or even to switch off their radios to save power. To accommodate both periodic and aperiodic traffic patterns, 802.11bn introduces periodic unavailability operation (PUO) and dynamic unavailability operation (DUO) modes, respectively. Moreover, PUO can also be used to implement a scheduled PS mechanism for the AP.

Wi-Fi power consumption is a critical concern due to the prevalence of battery-powered devices, coupled with emerging eco-design regulations. Even AP emissions reduction is also gaining much attention nowadays. Each generation of Wi-Fi focuses on increasing throughput, but the mechanisms that enable it, e.g., the use of wide frequency channels, multiple SSs, and high-rate MCSs, significantly increase power consumption. This effect is only exacerbated as the number of devices grows. 802.11bn introduces the dynamic PS mode, both for APs and client STAs, which will allow them to switch to the low capability mode and thus save energy, while remaining available for receiving some signals and switching back to the full capability mode after receiving a special frame. TGbn also improves the MLO framework by defining cross-link PS, which allows MLDs to manage power states across all links via a single active link. Finally, 802.11bn brings in scheduled AP PS, which operates on the basis of PUO. It enables APs to save power during the planned idle periods, while still accommodating legacy STAs and maintaining essential communication capabilities.

4.3 Wi-Fi Networks Coordination Candidate Features

Nowadays, high latency and low reliability are often experienced when devices move between Wi-Fi networks. Thus, to improve performance, the seamless roaming tools are necessary. The key 802.11bn offering here is the Seamless Mobility Domain (SMD) feature that defines a single entity covering multiple AP MLDs, which may not be colocated within a physical device. Within an SMD, the context, i.e., the states of handshakes, sequence numbers, security keys, and capabilities, can be transferred between multiple AP MLDs, i.e., among Wi-Fi networks. Such coordination reduces the unavailability time and decreases the loss ratio when a client MLD device roams from one Wi-Fi network to another. An SMD also enables a step-by-step per-link transition of the client MLD between AP MLDs, which promises to enable seamless connectivity.

Besides roaming enhancements, Wi-Fi 8 defines numerous Multi-AP coordination schemes, which may become the major innovations of Wi-Fi 8. These schemes have been discussed in 802.11be but were postponed due to specification complexity, so 802.11bn has taken up the baton along with the baggage of studies done. In a nutshell, Multi-AP leverages the classical Wi-Fi mechanisms—R-TWT, SR, BF, and others—by strengthening them thanks to enabling their cooperative work across multiple BSSs. Correspondingly, 802.11bn introduces Co-RTWT, Co-SR, Co-BF, and other AP coordination schemes, each varying with targets, efficiency, complexity, and overhead. To integrate Multi-AP into 802.11bn, the developers reassess potential gains in different scenarios and rework the frame format, signaling frames, and many coordination procedures.

5 PHY Candidate Features

5.1 PHY’s Quantitative Growth: A Closed Chapter?

One of the conventional approaches to enhance Wi-Fi performance is quantitatively improving PHY techniques, e.g., multiplying the channel width following the Shannon theorem. However, for Wi-Fi 8, the focus dramatically shifts away from PHY’s quantitative growth. Wi-Fi is approaching the limits of spectral efficiency, which makes it challenging to increase network throughput and density by old approaches: simply incorporating higher-order modulation, additional antennas, and wider channels [2, Doc. 22/0729r1]. Let us delve into each of these topics.

Higher-order modulations. The common way to boost data rates is to increase the peak order of modulation. Wi-Fi 7 has raised the top modulation to $4096$-QAM ($4$K-QAM). However, the further increase of the modulation order to $8$K-QAM or $16$K-QAM is unlikely because of the diminishing relative data rate growth (adding $16$K-QAM gives only $16.7\%$ nominal throughput gain vs. $4$K-QAM). Besides, higher-order modulations are difficult to implement and apply outside idealistic channel conditions, while the purpose of 802.11bn is to improve performance in realistic scenarios with imperfect channels and limited channel knowledge. As a result, Wi-Fi 8 is unlikely to increase the modulation order.

More antennas. With MIMO, 802.11ac uses up to eight SSs. In an attempt to achieve a quantitative gain from spatial multiplexing, 802.11be was initially expected to scale the number of SSs up to $16$ in DL across all the scheduled STAs. However, it has been postponed to future generations due to larger sizes of the devices and, most importantly, the huge sounding and feedback overhead, which can consume more than half of a channel time [25]. 802.11bn is trying to solve this overhead problem, but it is still likely to leave more antennas for the future: many of today’s devices tend to have four antennas at most. In other words, the market has not yet fully embraced the benefits of eight antennas, mainly due to cost and form factors. The authors of [2, Doc. 22/1580r1] suggest postponing more SSs for the next-next-generation, i.e., Wi-Fi 9.

Wider channels. Another straightforward way to increase throughput is through wider channels. Wi-Fi 7 has already scaled up the channel bandwidth up to $320$ MHz. However, the usage of so wide channels in the sub-$7$ GHz band reduces the frequency reuse factor [2, Doc. 22/0729r1] because only few such channels are available. Moreover, according to the presentation [2, Doc. 22/1580r1], the relative usage of $160$ MHz channels and wider is still below $1\%$. Therefore, widening channels in the same spectrum will not benefit realistic scenarios, so it is likely to be ignored by 802.11bn vendors even if included in the specification.

At the same time, Wi-Fi requires more spectrum. For example, the paper [36] shows that the existing number of channels in the sub-$7$ GHz bands may not be enough to serve a dense in-school scenario, while increasing the number of channels (and the allocated spectrum, e.g., in mmWave bands) boosts capacity and helps satisfy latency requirements. However, this calls for the allocation of higher-frequency bands.

5.2 Horizons of Alternative Spectrum

A reasonable response to the lack of spectrum in the sub-$7$ GHz bands is to master the horizons of alternative spectrum and to build a joint operation across the all available bands.

Millimeter-wave spectrum. Expanding into the millimeter-wave (mmWave) spectrum is a potential avenue for further throughput improvement. The mmWave band, ranging from $45$ GHz to $72$ GHz, offers ample spectrum globally, which presents substantial opportunities for enhancing system capacity and QoS provision [2, Doc. 22/1580r1].

Despite the appeal of increasing bandwidth, mmWave operation raises serious concerns. First and foremost, it increases complexity and cost of devices. Second, mmWave bands have short ranges, resulting in lower reliability over distances. Finally, operation in a band of higher frequency will cause a plethora of impairments, e.g., carrier frequency offset (CFO), phase noise, power amplifier non-linearity, etc. [2, Docs. 22/1395r0, 22/1865r1, 23/0066r2]. The viability of mmWave heavily depends on the quality of the solutions to these problems.

Fig. 2.

Comparison of realistic rates in the sub-$7$ GHz and mmWave bands (source: [2, Doc. 23/0165r2]).

The submission [2, Doc. 23/0165r2] provides a fair evaluation of realistic rates for $60$ GHz clients in comparison to existing solutions in the sub-$7$ GHz band. The authors highlight that mmWave operation necessitates the use of slower MCSs and a lower number of SSs than those in the classical bands. According to the simulation results [2, Doc. 23/0165r2], the $60$ GHz band does not significantly improve data rates despite wider bandwidth, as shown in Fig. 2. Therefore, the authors conclude that investing effort in mmWave may not be a one-size-fits-all solution as its benefits in realistic scenarios are questionable.

That is why this topic is considered by the separate Integrated mmWave (IMMW) TG, which started the development of the 802.11bq amendment in 2025. This group focuses on non-standalone operation in the unlicensed mmWave bands using SU OFDM-based transmissions [2, Docs. 24/0116r7, 24/0549r6]. Aside from Wi-Fi’s previous attempts to conquer mmWave (i.e., 802.11ad and 802.11ay [37]), first, the goal of 802.11bq is to reuse the classical PHY (from the sub-$7$ GHz band) as much as possible to simplify devices and thus reduce their costs. Second, in contrast to 802.11ad and 802.11ay that define standalone mmWave operation, the 802.11bq amendment requires devices to support at least one of the legacy bands, i.e., devices with the IMMW functionality shall work cooperatively with canonical Wi-Fi toward extended MLO, see Fig. 3. The intent behind this is to provide basic connectivity with control and management frames over sub-$7$ GHz links, while using an unreliable but high-throughput mmWave link only to offload data.

Fig. 3.

Extended MLO design with the mmWave and light bands (inspired by [2, Doc. 24/1594r0]).

Although mmWave is not directly included in 802.11bn, the UHR and IMMW TGs are going to finish by the same time. Thus, according to some studies [7], the devices that are capable of Wi-Fi 8+ in the broad sense (the IEEE 802.11 standard of the 2030s) are likely to provide extended MLO with mmWave links.

Light spectrum. Inspired by the idea of integrating mmWave, the community has also raised ideas for synthesis with the infrared and visible light spectrum. The recently published 802.11bb [38] is harmonized with 802.11ax, but not with 802.11be and 802.11bn, so it lacks many modern features, such as MLO, and requires an update. This idea is developed in the Enhanced Light Communication (ELC) SG, started in September 2024, with the continuation under the auspices of 802.11br TG [2, Docs. 24/0185r2, 24/1600r5]. These groups have yet to contemplate the feasibility of optical bands and, similar to mmWave, to consider joint operation with the classical sub-$7$ GHz bands as shown in Fig. 3.

5.3 Transmission Reliability Improvements

Although 802.11bn does not improve nominal throughput, it defines several features for boosting throughput at different ranges, as well as providing improved support for poor channel conditions, e.g., in dense scenarios.

Distributed-tone RUs. Unfavorable channel conditions—the target for 802.11bn—are typically represented by edge clients that suffer from low receive power. Increasing transmit power could solve this problem. However, due to regulatory rules, devices cannot exceed the PSD limit defined per a narrow piece of bandwidth, e.g., a Low Power Indoor (LPI) AP cannot emit more than $10$ dBm per $1$ MHz in the $6$ GHz band following ETSI regulations [2, Doc. 23/0037r0]. The developers of 802.11bn overcome this limitation by introducing distributed-tone RUs (DRUs); see [34] and [2, Doc. 23/0037r0]. Their idea is to redistribute the tones of small RUs over a wide bandwidth (see Fig. 4). This allows devices to increase their total transmit power without violating the PSD limits. Such an approach promises up to $11$ dB power boost when small RUs are distributed over an $80$ MHz channel [2, Doc. 23/0037r0]. It naturally motivates the use of DRUs in UL OFDMA, as multiple devices can use different DRUs to simultaneously increase their overall transmit PSD and spectral efficiency.

Fig. 4.

Regular RU and distributed-tone RU (DRU) tone plans for a part of a 20 MHz channel [34].

Figure 4 compares the tone plans for regular RUs and DRUs on the example of a right part of a $20$ MHz channel. Specifically, while a regular RU spans a continuous subset of subcarriers, each DRU spreads its tones across the entire available distribution bandwidth (DBW). Succeeding the hierarchical structure of regular RUs, DRU tone plans use $26$-tone DRUs as basic building blocks, so larger DRUs combine the tones of smaller DRUs, as shown in Fig. 4. 802.11bn is going to support DBW of $\{20, 40, 60, 80\}$ MHz. Note that a DBW of $60$ MHz indicates that DRUs can work with preamble puncturing (see Section 2.6), i.e., a DRU can spread over $60$ MHz in an $80$ MHz channel, forsaking the remaining $20$ MHz subchannel. The community also proposed to introduce DBW up to $160$ MHz as an optional feature [2, Docs. 23/1448r0, 23/1988r2, 24/0500r0], but it did not manage to appear in the current draft [34]. Thus, each DRU would span only up to an $80$ MHz channel segment. This peculiarity enables hybrid PHY Protocol Data Units (PPDUs) that combine regular RUs and DRUs in different $80$ MHz subchannels of wide channels [2, Docs. 24/0400r0, 24/0014r0]. In turn, it allows serving both DRU-capable and DRU-incapable STAs simultaneously.

Since the spread of data tones increases in DRUs, the pilot tone assignment problem arises. To remind, the pilots have a predetermined phase, and their main goal is to assist in fine CFO correction. TGbn is also interested in inter-carrier interference cancellation, as well as easier channel smoothing [34]. As a result of these discussions, the DRU pilots are designed to have enough separation yet be far from central and channel edge (guard) subcarriers [34]. Notably, the pilots of larger DRUs merge subsets of pilots from smaller DRUs, resulting in a well-spaced pilot set.

Interference mitigation pilots. A random, unexpected interference in unlicensed bands is one of the main reasons for limited reliability in WLANs [2, Doc. 24/1264r0]. Traditional reliability-aware methods, e.g., using more robust MCSs, retransmitting failed packets, or switching channels, usually fall short in the presence of such interference.

To address this issue, TGbn introduces interference mitigation (IM) pilots. These IM pilots are additional subcarriers (or repurposed data tones), which are located in every OFDM symbol and distributed across the tones. They help resist interference of any bandwidth, which may begin anywhere within a PPDU and which existing CFO pilots and midambles usually cannot handle. Specifically, thanks to these IM pilots, a receiver can identify and mitigate interference through advanced receive BF algorithms. For example, with multiple receive antennas, it can be done with the Minimum Variance Distortionless Response (MVDR) algorithm [2, Doc. 23/1490r1].

The simulation results from [2, Doc. 24/1264r0] show that when the new pilots occupy $16$–$20\%$ of subcarriers IM significantly improves the SINR performance by over $10$ dB, particularly in the high SINR range [2, Doc. 23/1490r1]. Thus, it enables operation with fast MCSs even during strong interference, considerably increasing goodput. Besides, the contribution [2, Doc. 25/0808r0] suggests using three additional IM pilots per $26$-tone RU (and doubling this value for each RU of a sequentially larger size). It consumes about $9$–$12\%$ of the resources for IM pilots but is assumed to be sufficient in terms of the SINR performance [2, Doc. 25/0808r0]. IM pilots also allow for achieving low and stable error rates with high throughput, thus helping link adaptation algorithms to converge faster and stay robust. However, IM is less applicable in DL MU-MIMO scenarios, where STAs typically have fewer antennas than needed to handle interference from SSs to other users [2, Doc. 24/1785r2].

The IM technique can also enhance SR as a signal receiver becomes more resilient to OBSS interference. This aligns with the goal of SR to allow concurrent transmissions while limiting interference. So, IM has a strong advantage to be used in addition to SR and Coordinated SR (Co-SR) operation.

Unequal modulations and new MCSs. In MIMO systems, SSs often experience different channel gains. However, the current 802.11 specification spreads encoded bits regularly over all SSs and mandates the use of the same MCS for them, which reduces spectral efficiency. To address this issue, 802.11bn introduces the Unequal Modulation (UEQM) feature, which allows different constellation orders across SSs with the same code rate [34]. The key advantage of UEQM is its ability to adapt the modulation order to SINR of each SS. In particular, according to the currently proposed scheme [34], SSs can use a constellation up to two steps lower than that of the SS with the highest channel gain. To maintain manageable complexity, the possible UEQM patterns are limited to those listed in Table 5, where “$\mathcal{S}$” denotes the modulation $2^{2\mathcal{S}}$-QAM of the first SS. Such UEQM schemes are applicable to up to four SSs and only with LDPC and non-MU transmissions with the lowest modulations higher than BPSK. These constraints greatly simplify the UEQM design, e.g., enabling the indication of UEQM patterns only by a pair of bits, while still keeping UEQM effective: the contributions [2, Docs. 24/0498r4, 24/0474r3] demonstrate more than $2$ dB SINR performance gain.

Table 5. Unequal modulation (UEQM) patterns [34].

Notably, UEQM exists in 802.11n, then, it turned out to be rarely used in real devices due to high complexity and was eventually removed in 802.11ac [18]. However, with the current focus on reliability, the idea is being revived.

UEQM uses the same code rate across multiple SSs to apply a joint LDPC coding, but this highlights an inconsistency in or even contradicts with the current MCS grid [2, Doc. 24/0498r4]. Specifically, several combinations of QAM constellations and code rates are not defined. For example, as shown in Fig. 5, the $2/3$ code rate is only paired with $64$-QAM, limiting the applicability of UEQM. To address this, the MCS grid is extended with several missing combinations highlighted in Fig. 5. These new MCSs promise to improve throughput in intermediate SINR regions by up to $30\%$ and enhance the rate-vs-range performance both with and without UEQM [2, Docs. 24/0498r4, 24/0469r0]. Note that these additions leverage existing QAM and coding functionalities in 802.11, so the implementation complexity is minimal.

Fig. 5.

Wi-Fi MCS grid with 802.11be MCSs (except those with the dual carrier modulation) and the new UHR MCSs.

Continuing the idea of UEQM over different SSs, TGbn developers have discussed a similar feature in the frequency domain [2, Doc. 24/1132r0], but it has not been included in the current draft. The idea is to use different MCSs at different subcarriers or RUs within an MRU, which helps combat frequency-selective interference, akin to adaptive bit loading in light communication technologies [38].

Finally, several contributions [2, Docs. 24/1177r0, 24/0812r2] pitch multi-layer transmissions, which achieve different reliability levels within the same modulation constellation. This feature is not included in the current 802.11bn draft [34], but promises significant benefits in the context of UHR. The core idea is to categorize bits into two (or more) groups (e.g., codewords, PPDUs, or bit flows) that require different levels of robustness and then map each group to separate constellation positions with different reliability. More critical bits are assigned to the most reliable positions (e.g., the first and the fifth bits in $256$-QAM), while less critical bits occupy the remaining positions (e.g., the remaining six bits in $256$-QAM). These bit groups can then be decoded separately: the first one as a lower-order modulation (e.g., QPSK) and the second one as a higher-order QAM excluding the most protected bits. While the second group incurs a minor SINR penalty of about $1$ dB, the reliable one benefits from up to a $10$ dB gain [2, Doc. 24/1177r0]. Moreover, this multi-layer scheme enables a single frame to be sent to two devices simultaneously, with each device receiving dedicated bits with a distinct reliability level. If the preambles are backward compatible, legacy devices can still decode the more robust portion. Note that this approach of serving multiple users within the same time-frequency resource is a form of Non-orthogonal Multiple Access (NOMA); see [2, Doc. 18/1957r3] and [39].

LDPC improvements. Since 802.11n, when Wi-Fi LDPC was introduced, data rates have increased significantly, but the codeword lengths have remained the same, resulting in suboptimal coding performance. Thus, 802.11bn is going to improve the current LDPC coding pipeline by doubling the maximum codeword length up to $2 \times 1944 = 3888$ bits [2, Doc. 23/1985r6]. To incorporate longer codewords, the contribution [2, Doc. 23/1985r6] proposes continuing to leverage the protograph lifting approach, already existing within 802.11. Thus, the parity check matrices for $3888$-bit codewords are built upon the new, larger matrix prototypes (aka exponent matrices) that are filled with subblocks of size $81 \times 81$, as it has been done for $1944$-bit codewords since 802.11n. Protograph lifting represents a good compromise between performance and complexity. Also, it facilitates decoding parallelism. The results shown in [2, Doc. 23/1985r6] promise a stable gain of a few SINR decibels for fast MCSs.

Let us also note that 802.11bn enhances the efficiency of LDPC codes for short frames. During LDPC encoding and rate matching procedures in Wi-Fi 6 and Wi-Fi 7, the number of coded bits is rounded up to fill in a quarter of an OFDM symbol, which is a remnant of the fourfold increase in OFDM symbol duration in Wi-Fi 6. However, since only full OFDM symbols are transmitted, it is a shortcoming (or even a “bug”) because this process forsakes bits corresponding to the remainder of the last OFDM symbol that is filled as padding [2, Doc. 24/1054r0]. Such an underutilization is particularly noticeable for frames of a few OFDM symbols long. The 802.11bn standard addresses this inefficiency and changes the formulas to round up the coded bits to full OFDM symbols. Thanks to it the futile padding bits are transformed into useful bits that are used during the LDPC rate matching, e.g., increasing the number of bits that are repeated per codeword in the over-puncturing case. This modification promises to improve reliability (i.e., bit or packet error rate) of short frames, which, for reference, are prevalent in IIoT applications.

Enhanced long range PPDU. Wi-Fi 8 defines three types of PPDU formats shown in Fig. 6, each tailored to different transmission modes and goals. First, 802.11bn inherits the MU PPDU format from previous Wi-Fi generations. Despite its name implying transmission to multiple users, UHR MU PPDUs are used for both SU and MU transmissions. Second, 802.11bn continues to support the TB PPDU format, used for UL OFDMA and UL MU MIMO transmissions performed in response to a TF from an AP. Third, 802.11bn introduces the new UHR Enhanced Long Range (ELR) PPDU format, geared toward improving the rate-vs-range performance for edge users.

Fig. 6.

UHR PPDU formats [34].

The primary motivation for ELR is the asymmetry in link budget between UL and DL in current Wi-Fi systems: APs typically transmit with a significantly (up to $10$ dB [2, Doc. 24/0875r1]) higher power compared to battery-powered client STAs. The ELR PPDU format offers a $6$ dB gain, while the remaining UL-DL power gap can be mitigated through antenna and spatial diversity gains [2, Doc. 24/0921r0]. Consequently, ELR partially overcomes the link budget imbalance and improves spectral efficiency for edge STAs, thereby enhancing transmission reliability.

To achieve this gain, in the UHR ELR PPDU Data field, encoded data bits are quadruplicated across four $52$-tone RUs in a $20$ MHz channel as shown in Fig. 7, with phase rotation applied on certain data tones to reduce the peak-to-average power ratio [2, Doc. 14/1486r1]. Such quadruplication provides exactly $6$ dB improvement. ELR PPDUs support only BPSK and QPSK at the $1/2$ code rate, yielding PHY data rates of $1.67$ Mbps and $3.33$ Mbps, respectively, in the only supported $20$ MHz channel bandwidth.

Fig. 7.

Data encoding in UHR ELR PPDUs [34].

The ELR PPDU preamble, depicted in Fig. 6, is also designed to provide reliability enhancements. For backward compatibility, it begins with a legacy (pre-UHR) preamble structure, as all 802.11bn frames do. Next, there is a two-symbol ELR-MARK, which identifies the PPDU as ELR and carries the BSS color information for early OBSS frame filtering [2, Doc. 24/1571r2]. The preamble also contains UHR training fields and the ELR-SIG field that reinforces robustness by repeating critical information from earlier fields, such as MCS and the BSS color [2, Doc. 24/1485r2]. The ELR-SIG field consists of two separately encoded OFDM symbols and employs repetition across four $52$-tone RUs, mirroring the approach for the ELR Data field shown in Fig. 7. Finally, all training fields in the ELR preamble incorporate an additional $3$ dB power boost (compared to UHR non-ELR PPDUs), which can achieve $6$ dB link budget gain with support of CFO pre-correction [2, Doc. 24/1573r1].

5.4 Beamforming and MIMO

Beamforming (BF) and MIMO, the major features of Wi-Fi technology, are suitable for increasing throughput at various SINR levels, improving reliability, and reducing delays. However, their effectiveness heavily depends on channel sounding and precoder formation procedures. Hence, the following features have been discussed in the UHR SG and TG, but have not been included in the current 802.11bn draft [34].

Compression of CSI feedback. Most BF techniques rely on CSI, which includes feedback matrices per subcarrier group. Transmitting these matrices consumes a significant amount of channel time, especially with many subcarriers, antennas, and devices. Even the Artificial Intelligence and Machine Learning (AIML) Topic Interest Group highlights the reduction of CSI feedback overhead as a crucial issue [2, Docs. 22/0950r2, 22/1563r2]. Numerous feedback compression techniques have been proposed, including reduced channel impulse response, differential Givens rotations, variable angle quantization, implicit feedback techniques, and many others [40,41,42]. However, only angle decomposition of unitary matrices and subcarrier decimation, or grouping have been integrated into the current 802.11 standard. A search for a “perfect” compression algorithm is still ongoing. Such an algorithm should have only a reasonable increase in complexity and provide significant gain in various channel conditions, preferably validated in real-world scenarios.

Beamforming in trigger-based UL. Another potentially fruitful direction is to introduce BF for TB PPDUs in UL. Currently, 802.11 APs cannot instruct STAs to apply BF in TB UL. At the time of 802.11ax specification, the feature, which burdens the STA-side with sounding and precoding mechanisms, was thought to be too complex. However, using BF for long UL TB PPDUs per STA promises to provide about $2$–$4$ dB gain over the entire SINR range [2, Doc. 20/1672r2], which positions UL BF as a viable candidate for reliable Wi-Fi.

As mentioned, BF in TB UL needs a procedure for STAs to perform sounding, either independently of the AP or when triggered by an AP [2, Doc. 23/0263r0]. A traditional approach to sounding uses Null Data Packets (NDPs) sent by the STAs and followed by the AP’s response with CSI [2, Docs. 24/0243r2, 24/0869r0]. In contrast, a more innovative approach suggests implicit sounding, which does not require feedback from the AP, thus reducing overhead [2, Doc. 22/1392r0]. In this approach, the AP sends NDP frames and STAs measure the reciprocal channel directly. As the AP has higher transmit power than STAs, it likely provides better measurement accuracy. Moreover, the training fields of NDP can be integrated into TFs, saving extra time.

Smooth beamforming. Channel estimation often employs frequency-domain smoothing to mitigate noise and decrease error rates. However, this process requires the effective channel’s frequency response after BF to be continuous; otherwise, the filtering will distort the channel estimation and degrade reliability. Such discontinuity can be caused by various factors described below and smooth BF is designed to combat this effect [2, Doc. 22/1392r0].

First, in the standard compressed CSI feedback, the feedback matrix undergoes a column-wise phase shift. This approach discards information about the similarity of adjacent subcarriers and reduces the feedback size [2, Doc. 22/1842r0]. Without this information, the transmitter cannot compensate for sudden phase changes, which may occur at faded subcarriers. The missing phase information can either be added to the feedback [2, Doc. 23/1906r1] or restored at the receiver using complex, implementation-specific solvers [43].

Second, discontinuity also occurs when two singular values from BF matrices at adjacent subcarriers are similar [2, Doc. 05/1635r1]. During singular value decomposition (SVD) computation, eigen-channels are ranked by their singular values. Thus, when the singular values at adjacent subcarriers are close, the corresponding steering matrices can swap. If this is reflected in CSI, the transmitter will mix up the SSs in a certain frequency range, and smoothing at the receiver will impair reception. Alternatively, the transmitter would need a complicated framework to track such swaps. To address these issues, the SVD computation shall be harmonized across subcarriers [2, Docs. 22/1413r1, 22/1869r1]. Additionally, equal singular values can result in numerical precision problems in matrix operations when building the CSI feedback.

Third, channel discontinuity is exacerbated by subcarrier grouping (i.e., tone decimation) in CSI feedback. Transmitters perform interpolation, and if the matrices in the CSI feedback are discontinuous across frequencies, the interpolation becomes invalid. Such discontinuity after interpolation may be tackled independently at either the receiver or transmitter side.

Fourth, discontinuity may arise from angle quantization [2, Doc. 22/1842r0]. Beamformers can resolve this by applying a smoothing filter to the beam-steering matrix across subcarriers or by using the fine quantization option.

5.5 Open Issues

While Wi-Fi 8 shifts away from traditional quantitative PHY improvements and attempts to adopt new PHY approaches, the latter, in turn, result in numerous open issues. These issues require careful consideration to ensure robust performance gains while maintaining backward compatibility and cost-effectiveness.

To begin with, mmWave technologies, while enabling high data rates, face fundamental physical and practical challenges that hinder their reliable deployment in dynamic wireless environments. First, mmWave knowingly suffers from high susceptibility to blockage. When a line-of-sight or dominant path is blocked, the connection most probably fails. The devices need a reliable method to determine the current link quality and quickly switch links from mmWave to sub-$7$ GHz and vice versa to transmit packets. Second, to maintain sufficiently high received power, BF should be widely used. Beams need to be both narrow and precise, which is especially difficult when the environment is highly dynamic and/or client STAs rotate. Third, enabling such precise BF induces substantial beam training overhead. Existing beam sweeping and training algorithms, such as those standardized in 802.11ad and 802.11ay, incur high latency and degrade spectral efficiency. This motivates the need for smarter beam search and training algorithms. Notably, AIML has offered powerful tools for beam search and management, but it remains a question whether these techniques generalize well to real-world time-varying and noisy environments without large amounts of carefully labeled training data. Finally, supporting efficient BF and channel estimation in practice must explicitly account for hardware limitations and impairments inherent at mmWave. Imperfections such as high phase noise and nonlinear power amplifier distortion under the high peak-to-average ratio conditions impact the precision and robustness of BF. Additionally, high-resolution wideband analog-to-digital converters consume considerable power, which constrains scalability. This makes energy-efficient solution more appealing for mmWave, e.g., hybrid precoding architectures.

Next, DRUs is an encouraging idea but further examination is needed to test the viability of the mechanism. The proposed structure may experience inter-carrier interference problems, so the process of synchronization is a point of interest. Besides, the MAC perspective of DRU operation has not been analyzed. The AP has to determine which RU type for an STA to use in a TB UL frame, but the STA has no standardized ways to tell the AP if switching to DRUs is helpful or not [2, Doc. 24/1058r0].

Although 802.11bn is going to support DRUs only for TB PPDUs in UL [34], let us elaborate on how they might be useful in DL as well. Interestingly, different DRUs may naturally exhibit very similar channel conditions because their tones are the same up to a shift of no more than $2.2$ MHz [2, Doc. 23/1516r0]. In other words, each DRU smooths frequency selectivity within itself, averaging different distributed tones, so the SINR values promise to be nearly identical in different DRUs. As a result, the rate adaptation could be effectively performed independently of the RU index or even its size, which could simplify resource planning, yet a detailed analysis is in demand.

IM pilots have several associated problems as well. First, IM mode selection is to be integrated into link adaptation algorithms, so new advanced search techniques are required [2, Doc. 24/1264r0]. Second, the introduced IM pilots, besides interference mitigation, can be leveraged for other purposes, e.g., enhanced CFO compensation and online channel adaptation. The additional purposes may enable reliable reception in diverse conditions.

Although UEQM is a promising approach, its current gain estimates rely on assumptions of perfect channel estimation and perfect link-to-rate matching, which may overstate its practical benefits. Furthermore, as a MIMO technique, UEQM requires careful consideration of the impact of non-ideal channel knowledge [2, Doc. 24/0890r1]. Moreover, the algorithms for rate selection (how to choose MCSs for different SSs and RUs) and the specific signaling are in demand [2, Docs. 24/0474r3, 24/1216r1].

The operation with ELR faces several challenges. First, the extended duration of the ELR preamble raises concerns about its efficiency, especially while multiple antennas can enhance the reception rate without the preamble extension [2, Doc. 24/0921r0]. Second, backward compatibility poses issues [2, Doc. 24/1508r0]. A STA using request-to-send (RTS) and clear-to-send (CTS) frames to protect its ELR transmission faces potential problems: non-ELR RTSs may go unnoticed by the AP while ELR RTSs may be ignored by legacy STAs, with the mirrored issues for ELR CTSs. Using multiple frames is a potential solution (e.g., accompany the exchange of ELR RTS/CTS with ordinary, non-ELR, CTS-to-self and/or CTS frames [2, Doc. 24/1508r0]), but a more elegant one is desirable.

Several enhancements are proposed for BF and MIMO. For example, UL MIMO precoding can be amended with coordination by the AP [2, Doc. 23/0027r1]: it would collect channel sounding data and instruct each STA on the precoder to use considering all user channel cross-correlations. This would prevent STAs from performing BF independently, which, due to lack of information, imposes mutual interference [2, Doc. 24/1265r0]. Such coordination will potentially improve the total throughput vs. SINR performance by up to $8$ dB despite negotiation overhead and channel aging [2, Doc. 23/0725r0]. These results are certainly promising and can even contribute to the Coordinated BF (Co-BF) Multi-AP scheme (see Section 7.8) in the future. The STAs could use precoders that mitigate MU interference at their AP, as well as additionally cancel interference to OBSSs. However, the non-convex nature of the problem makes it computationally intensive. Creating an algorithm that deals with interference from other devices and imperfect CSI is another challenge. Besides, smooth BF shows a relatively low gain of about $2$ dB of SINR at the expense of large specification effort and high complexity. Smoothing has been an open issue for a long time but still has not been solved in 802.11. Incorporating these ideas into the standard would require further analysis to find a way to make them simple or beneficial enough.

In the end, let us recall here Hybrid Automatic Repeat reQuest (HARQ), which stably improves reliability in cellular systems but has yet to be adopted in Wi-Fi. Research during the development of 802.11be confirmed HARQ’s value, demonstrating about $3$–$4$ dB SINR gain in Wi-Fi MIMO scenarios [44]. Unfortunately, many concerns have been raised about the complexity of PHY reorganization to solve problems such as MPDU and codeword misalignment [2, Doc. 22/1880r1]. Currently, these concerns have not been resolved, so to leverage HARQ in Wi-Fi the novel simple PHY design for further generations is in demand.

6 MAC Candidate Features

6.1 Low-latency Channel Access

Although Wi-Fi already supports extremely high nominal data rates and low transmission delays, the provision of limited latency requires further improvement. A clear latency bottleneck is associated with the random and distributed nature of Wi-Fi channel access, i.e., EDCA [2, Doc. 22/1519r0], which implements CSMA/CA with a random backoff. While EDCA has been proven to guarantee fairness and backward compatibility, it falls short in providing low and deterministic delays. Let us note that other channel access schemes, e.g., scheduled TB channel access introduced in 802.11ax, are still built upon EDCA. In this regard, we explore the direction of low-latency channel access enhancements.

Prioritized EDCA. Since 802.11bn aims to provide guaranteed latency, it seeks, for the first time in generations, to develop an improved version of totally distributed channel access with prioritization. This modification shall leave EDCA distributed to remain operational in scenarios with many BSSs, as well as shall not hinder the service of legacy devices, which do not understand the new rules.

Fig. 8.

An example of prioritized EDCA (P-EDCA) operation (inspired by [2, Doc. 23/1065r0]).

For that, the 802.11bn draft defines the prioritized EDCA (P-EDCA) mechanism [34], which targets to reduce the access distribution delay tail and operates as shown in Fig. 8. P-EDCA divides STAs into two groups: with and without prioritized access. When the prioritized STAs need to access the channel, they send a Defer Signal CTS (DS-CTS) frame, following the previous TXOP and after some interframe space and short contention. This interframe space may equal to AIFS of the voice AC (“AC_VO”), and the contention window may be set to a small value or even zero as shown in Fig. 8. These values correspond to the default parameters recommended by 802.11bn [34] and ensure that P-EDCA channel access is shorter than a time when non-prioritized STAs can gain access to the medium. Having received the DS-CTS frame, other STAs either suspend their contention operation thanks to the NAV set or defer channel access for a long extended interframe space (EIFS) if they cannot decode it. As a result, with properly configured channel access parameters the prioritized STAs appear to be isolated from the other STAs in the channel. After the DS-CTS frame, only the STAs that have sent it participate in ordinary contention, but using the new parameters from the P-EDCA parameter set advertised by an AP. Specifically, the prioritized STAs wait for P-EDCA AIFS, then randomly initialize the backoff counter and count it down to initiate a frame exchange.

To mitigate the adverse effects of P-EDCA on non-prioritized STAs, 802.11bn establishes additional rules. First, only UHR STAs with pending data corresponding to the highest-priority AC (i.e., voice) can participate in P-EDCA contention. Once it begins, STAs suspend contentions for the other ACs. Second, an STA can follow P-EDCA only after reaching a small number of consecutive failed retransmissions, e.g., two. Also, it can use P-EDCA only a few times in a row, e.g., only once.

The simulation results from [2, Docs. 23/1065r0, 23/2126r3, 24/0467r1] show that P-EDCA by $3$–$4$ times reduces the tail of latency in a vast number of scenarios while preserving legacy STAs performance.

For better latency performance in UL and environments with multiple networks, the defer signal may be unique for each BSS [2, Doc. 23/2126r3]. Also, to reduce the chance of collision of DS-CTS frames, several researchers [2, Doc. 24/1193r0] suggest that an STA can decide whether it is eligible to send the DS-CTS frame by the additional randomized procedure that determines priority even within the most prioritized traffic identifier (TID).

Non-primary channel access. In Wi-Fi networks, any wideband transmission shall cover the primary $20$ MHz subchannel. For example, an STA operating in a $160$ MHz channel cannot transmit when its primary $20$ MHz is occupied, even if the remaining $140$ MHz is idle. To address the described bandwidth waste, the 802.11bn draft defines non-primary channel access (NPCA) that enables an STA to access the secondary channel while the primary channel is known to be busy [15, 34].

Fig. 9.

Non-primary channel access (NPCA) scheme (inspired by [2, Doc. 24/0803r1]).

Figure 9 shows a general scheme of NPCA. When an STA listens on a primary channel and finds it busy by an OBSS transmission, the STA can switch to the other auxiliary channel, termed the NPCA primary channel, predetermined by the AP. Then, the STA contends in the new channel, i.e., it invokes a new backoff procedure, following the standard rules. After obtaining a TXOP, to ensure that the receiver also switched to a different channel, the STA initiates a frame exchange with an initial control frame (ICF) followed by an initial control response frame (ICR), e.g., with the RTS/CTS handshake [2, Doc. 24/0829r0]. Finally, when the primary channel becomes idle, the STA switches back to it [2, Docs. 24/0070r2, 24/0496r1].

The simulation results show that NPCA promises significant throughput gains by leveraging underutilized channels [2, Doc. 23/0962r0], [15], as well as improves latency of pending traffic.

Notably, to limit the complexity of NPCA, the 802.11bn developers have agreed to the following constraints [34]. First, NPCA shall not assume the capability to detect or decode a frame and obtain busyness information (e.g., NAVs) of the secondary channel concurrently with the primary channel. Second, the BSS shall only have a single NPCA primary channel as an alternative to the primary one.

Preemption. Another feature dedicated to low-latency channel access is preemption. In general, preemption is a method to stop the ongoing transmission and start serving low-latency traffic to satisfy the delay requirements [2, Doc. 24/0389r0]. Notably, it is one of the missing components preventing Wi-Fi from fully supporting the TSN functionality [33] related with prioritized packet treatment and immediate access for time-critical packets. Although Wi-Fi 8 is unlikely to incorporate preemption directly, the UHR SG has devoted significant effort to discussing it. Therefore, let us explore its key principles.

Suppose a device is transmitting a long PPDU carrying non-urgent traffic when a low-latency packet arrives. In this case, the device interrupts the ongoing PPDU transmission and sends a new PPDU with latency-sensitive data, thereby performing preemption [2, Docs. 21/0670r0, 23/0018r1, 23/0092r0]. Afterward, the device resumes the deferred non-urgent PPDU transmission, starting with a new, though potentially shortened, preamble.

Preemption schemes become more complex when the source of urgent data is not the current sender. Suppose a TXOP holder transmits non-urgent traffic, while a low-latency packet is buffered on another device. To let the latter perform preemption within the TXOP [2, Docs. 23/0018r1, 23/0092r0, 23/1886r3], the TXOP holder should suspend transmissions of non-urgent traffic in favor of delivering low-latency data. Such a suspension is easily implemented if devices have an independent radio interface to signal about the presence of urgent data [45], while devices with a single radio interface require a more sophisticated transmission scheme. To facilitate this scheme, a dedicated low-latency session shall be pre-established [2, Doc. 24/0389r0]. During this session, the TXOP holder segments long PPDUs into smaller PPDUs [2, Doc. 23/1229r1], creating time gaps between them and signaling preemption allowance in these sliced PPDUs. With this preemption allowance, the device with low-latency packets may preferentially transmit preemption request frames [2, Docs. 23/0378r0, 24/0389r0] during the time gaps to notify the TXOP holder of pending urgent data. Upon receiving the preemption request, the TXOP holder orchestrates the delivery of low-latency data [2, Doc. 24/0168r0]. If no preemption request is received, the TXOP holder resumes transmission of non-urgent traffic, perhaps using a shortened preamble, as shown in Fig. 10.

Fig. 10.

An example of preemption within the TXOP (inspired by [2, Doc. 24/0168r0]).

Preemption dramatically increases the complexity of the transmission procedure; consequently, it will most likely not become a part of 802.11bn. However, it may be gradually added in future amendments to the Wi-Fi standard. Thus, we can expect that in its early versions preemption will be limited to the TXOP holder and the TXOP responder [2, Doc. 24/0389r0], when the TXOP holder is the AP. Allowing other STAs to transmit their preemption requests during the time gaps may lead to contention between the possible transmitters. Limiting the contention while collecting the low-latency indication and BSRs for preemption can be implemented in various ways [2, Docs. 23/1909r1, 24/0131r0]. For example, an AP, being the TXOP holder, can send the special TF after its PPDUs. In this TF, some dedicated random access RUs may be allocated and used for sending the low-latency indication, BSRs, or even the low-latency data itself simultaneously with regular PPDU transmission [2, Doc. 24/0442r3]. The TXOP responder can be additionally prioritized by allowing contention-free channel access right after the end of PPDU from the TXOP holder [2, Doc. 24/0416r1], while the TXOP holder signals in its frames that it allows the other STAs with low-latency traffic to contend for channel access after a longer time interval. Moreover, the TXOP responder AP can add a preemption request to its Block Ack frame to signal that it needs to send urgent data. To make sure that the TXOP holder’s traffic is served fairly, preemption can be enabled only if the TXOP holder signalizes, e.g., in the PHY header of its frames, that it solicits preemption requests and allows the low-latency packet transmission [2, Doc. 24/0131r0].

To sum up, the preemption mechanisms offer a huge improvement in the latency direction [46]. For example, the simulation results [2, Doc. 22/1880r1] show that preemption reduces latency by $60$–$80\%$ for different ACs, including the $90$th percentile, i.e., showing gains in worst-case situations. However, implementing preemption introduces significant complexity and leaves many open challenges, which is, probably, the reason why the specific preemption mechanisms have been left out of the 802.11bn draft [34]. At the same time, TGbn takes the first steps toward it by introducing several low-latency indication mechanisms (see Section 6.1 and [2, Docs. 24/1195r1, 23/0045r1, 24/0264r1]), while preemption emerges as a promising candidate for leveraging them in the future standard amendments.

P2P improvements. Another goal of 802.11bn is to improve the latency of P2P communications between the client STAs [34], e.g., when a computer streams video to an XR headset. This communication can be performed in the base channel, i.e., the channel used by the associated AP, and in the off-channel, i.e., the channel outside the AP’s bandwidth.

The base channel P2P operation can be improved by extending the TXS mechanism introduced in 802.11be. In 802.11be, an AP can share its TXOP with one STA, which can use it to send its data to another STA [27]. Note that with this TXOP sharing, the AP allows only a specific STA to send its PPDUs. To enable bidirectional data exchange between two STAs, the AP shall send a TF to share its TXOP with one STA, then with another one, etc. However, in some scenarios, two or more client STAs shall be able to quickly exchange data. To avoid the repeated TXOP reallocations and the induced overhead, the 802.11bn draft introduces the group-based TXOP allocation, which is expected to shorten delivery delays [2, Docs. 23/1929r0, 24/0392r1]. Such an allocation would enable bidirectional communications and would allow P2P devices to use prioritized channel access parameters during the allocated TXOP [2, Doc. 24/0403r2].

To facilitate the off-channel P2P communications, 802.11bn enables an AP to advertise in its BSS a set of channels recommended for P2P communications by including the related information in its Beacons and Probe Response frames [2, Doc. 24/0393r3]. The AP can also promote time intervals for P2P transmissions, which can coincide with the AP’s PS schedule [2, Doc. 22/1528r1] (see Section 6.4). However, in high-density environments, the P2P link is still likely to face interference from OBSSs because different BSSs may have non-aligned P2P schedules [2, Doc. 23/0294r1]. To solve this problem, TGbn has included in the 802.11bn draft the Coordinated Channel Recommendation mechanism (Co-CR in the vein of Multi-AP names, see Table 6), which will allow the APs from different BSSs to ensure that they advertise P2P channels without interference. Therefore, the P2P STAs will be able to communicate in the scheduled channels and time intervals directly across the OBSSs [2, Doc. 23/1424r0]. Such cooperation is a laconic example of how Multi-AP leverages Wi-Fi performance. Note that Section 7 describes more Multi-AP schemes considered for 802.11bn.

Table 6. Multi-AP coordination schemes.

QoS signaling improvements. Developers of 802.11bn have also highlighted the limitations of the current QoS signaling. Specifically, an AP may receive from STAs BSRs, i.e., queue sizes, related to particular TIDs. The current 802.11be-capable device is capable of reporting up to ${\sim}\,2$ MB that, however, does not match the 802.11be peak data rate and the maximum frame size of 802.11be, which is ${\sim}\,15$ MB; see [2, Docs. 24/0963r3] and [27]. This shortfall forces too frequent polling procedures, which increases overhead and limits Wi-Fi performance both in terms of throughput and latency. To address it, the 802.11bn draft introduces the Enhanced BSR Control field, in which the maximum per-TID queue size up to ${\sim}\,35$ MB can be informed [34].

Meanwhile, multiple traffic flows with different latency requirements and thus priorities may be mapped to a single Wi-Fi queue associated with one of the four EDCA ACs. As a result, a head-of-line problem may occur for an urgent frame when other low-latency but less urgent frames are in the queue [2, Doc. 24/0463r2]. This problem can be addressed by Traffic Specification (TSPEC) and Traffic Classification (TCLASS) functionalities from 802.11e, as well as by the Stream Classification Service (SCS) with the QoS Characteristic Element, recently updated in 802.11be. These mechanisms allow specifying delay bound, minimum data rate, etc. on a per-TID basis, thus making flexible scheduling decisions. However, currently, only two TIDs at max can be mapped to the AC with the highest priority. 802.11bn is going to fix this issue and allow remapping unused TIDs of low-priority ACs so more than two TIDs correspond to high-priority ACs [34]. These TIDs may also be used dynamically.

Next, 802.11bn makes the first steps toward supporting the Low Latency, Low Loss, and Scalable Throughput (L4S) architecture. In short, L4S uses the Explicit Congestion Notification (ECN) in the IP header to improve queue management, thus minimizing queuing delay and packet drops due to congestion. UHR APs will optionally support the signaling of L4S ECN, so when congestion is experienced at MAC due to a build-up in the transmit queue, they can notify the upper layers and enforce the necessary adaptations.

The crux of another issue lies in the lack of indication of unpredictable urgent traffic in Wi-Fi [2, Doc. 24/1195r1], so the 802.11bn draft defines a mechanism that will help the TXOP responder (i.e., the STA that responds to a frame from the TXOP holder) indicate low-latency traffic needs [34]. The general idea is to explicitly specify the presence of data and its delay budget from the TXOP responder [2, Doc. 24/1195r1], which will compel actions from the TXOP holder if the delay budget is less than the remaining TXOP time. Besides the 802.11bn draft itself, the authors of [2, Docs. 23/0045r1, 24/0264r1] propose an urgency-based BSR specifically designed for latency-sensitive traffic, which provides information on whether the amount of data pending for transmission has exceeded or is about to exceed its delay bound. In a similar vein, the authors of [2, Doc. 23/0740r0] suggest including a timing component in the report to identify the most urgent packets as a part of dynamic QoS feedback. These features are expected to help reduce delivery delays by enabling tools of low-latency channel access, including those of future generations, e.g., preemption, as discussed in Section 6.1.

6.2 Dynamic Bandwidth Management

Dynamic subband operation. The 802.11bn developers highlight the mismatch in operating bandwidths between APs and client STAs [2, Docs. 23/2141r3, 24/1863r1]. State-of-the-art APs are expected to support up to $320$ MHz channels, while client STAs are often limited to much smaller bandwidths. Since an STA’s operating bandwidth shall include the primary channel, such a discrepancy between an AP and its associated STAs causes large portions of the BSS bandwidth to go unused.

To address this, the 802.11bn draft brings in the tool of dynamic subband operation (DSO) [34]. With pre-negotiated DSO capabilities, an UHR AP can dynamically (i.e., on a per-TXOP basis) allocate frequency resources to STAs outside of their current operating bandwidth. This part of bandwidth, named the DSO subband, excludes the BSS primary channel but is present in the AP’s operating bandwidth. The AP initiates a DSO frame exchange in the primary channel with a special DSO ICF, sufficiently padded to provide enough time for an STA to switch. When the STA receives the DSO ICF with an assigned RU in the DSO subband, this STA transitions to the indicated DSO subband and then transmits the ICR on the allocated RU. Then, a regular frame exchange follows, using RUs of the DSO subband. Finally, the STA shall switch back to the subband with the primary channel. Performance evaluations show that DSO can increase throughput by $1.5$–$3$ times in high-density scenarios [2, Doc. 24/1863r1].

Note that DSO is conceptually similar to NPCA (see Section 6.1), but DSO channel transitions are fully orchestrated by the AP and explicitly start in the primary channel.

Dynamic bandwidth expansion. Another challenge addressed by TGbn is a general underutilization of wide bandwidths such as $160$ MHz and $320$ MHz channels. Enterprise and similar environments cannot routinely use wide channels and often remain conservatively limited to $40$ MHz [2, Docs. 24/0088r1, 24/0815r1]. At the same time, these scenarios could leverage wider channels under controlled circumstances, e.g., with coordination between neighboring APs (see Section 7). However, legacy channel-change mechanisms introduce long switching delays and connection interruptions [2, Doc. 24/0088r1].

To tackle this issue, the 802.11bn draft defines a mechanism for dynamic bandwidth expansion (DBE) that allows an AP to dynamically expand (and modify or reset) its operating bandwidth in BSS for UHR STAs [34]. Note that such a DBE bandwidth change is announced in advance (e.g., several beacon intervals) and is performed via signaling in the primary channel. Therefore, it minimizes bandwidth-switching delays for UHR STAs and does not disrupt operation of legacy STAs.

6.3 Unavailability Reporting

To keep sizes and cost of mobile devices low, manufacturers often make Wi-Fi STAs share radio components with other technologies, e.g., Bluetooth, ultra-wideband, or with other Wi-Fi STAs within the same MLD, which raises in-device coexistence (IDC) issues [2, Doc. 23/2078r5]. To address this topic, 802.11bn introduces unavailability reporting mechanisms that allow an STA to inform its peers about upcoming unavailability times [2, Doc. 23/1964r1]. Notably, the traditional PS signaling is usually insufficient to notify the peers about the next unavailability interval before it comes [2, Doc. 23/2078r5]. Thanks to unavailability reporting, STAs help the AP avoid scheduling and transmitting frames during these intervals [2, Doc. 24/0856r0], ensuring better resource utilization, minimizing interference, and saving energy.

Unavailability reporting has some similarities with the opportunistic PS mechanism introduced in 802.11ax, which allows an AP to announce that it is not planning to transmit anything to an associated STA during a specified time period and thus the STA may become unavailable until the end of this period. However, opportunistic PS is controlled by the AP, while unavailability reporting can be used by any STA.

To organize cooperation between STAs and their AP during predictable and periodic unavailability intervals (e.g., when Bluetooth frames with audio data are transmitted), 802.11bn defines the periodic unavailability operation (PUO) mode [34]. The PUO reporting is likely to be realized via a modified version of the TWT mechanism with new fields for signaling [34]. Upon receiving the PUO report, the AP will adjust its scheduling accordingly.

Similarly, for notification of unpredictable or urgent unavailability events, 802.11bn introduces the dynamic unavailability operation (DUO) mode [34]. Within this DUO mode, STAs indicate unavailability start times and durations at the beginning of a TXOP or during it in modified frames, such as BSR Polls or multi-STA Block Ack frames [2, Docs. 23/1964r1, 24/1558r2].

Partial unavailability has been considered to ensure the frame exchange remains possible, albeit with reduced capabilities, such as lower MCS, fewer SSs, narrower bandwidth, or limited subchannels [2, Docs. 24/0675r1, 24/1109r1]. A STA could also specify whether it is unavailable for transmission, reception, or both, and update or remove its previously reported unavailability schedules [2, Docs. 24/1109r1, 24/1504r0]. 802.11bn may as well include the cross-link unavailability signaling for MLDs [2, Doc. 24/0420r2].

The 802.11bn draft extends the unavailability reporting to APs, which is crucial for mobile APs (e.g., smartphones) and infrastructure APs with multiple connectivity technologies [2, Doc. 24/1108r2]. For instance, a smartphone streaming music via Bluetooth while tethering Wi-Fi can use the unavailability reporting functionality to avoid collisions and ensure efficient use of resources and energy. The PUO signaling specifically enables scheduled AP PS [2, Doc. 24/2040r9] because it allows the AP to inform the associated STAs when it will be unavailable and to schedule switching off its interfaces during the indicated time intervals.

6.4 Power Save

Power consumption has always been an important issue for Wi-Fi, mostly because the majority of devices are mobile and battery-powered. At the same time, the “eco-design” regulations emerge and force very low power consumption both of STAs and APs [2, Doc. 22/1790r0]. Thus, 802.11bn aims to provide power save (PS) mechanisms both for AP and STAs.

Dynamic power save. The Wi-Fi module consumes $3$–$10\%$ of smartphone power [2, Doc. 22/1841r0], out of which more than a half is used for the radio listening [2, Doc. 22/1414r1]. 802.11 introduced a series of PS mechanisms like the basic PS mode, the automatic PS delivery, or the TWT mechanism [47], which imply a complete shutdown of the radio interface. However, such approaches increase the maximal delay up to the shutdown duration, and thus cannot be used in scenarios when the delays are critical and the traffic is unpredictable, which are a focus of 802.11bn.

To avoid long unavailability periods and at the same time to reduce power consumption, the 802.11bn draft defines the dynamic PS mode that allows an STA to switch between:

• low capability mode, when the STA can receive only simple PPDUs, e.g., in a $20$ MHz channel with one SS and slow MCS; and

• high capability mode, when the STA can collect PPDUs at its maximum capability [2, Docs. 23/1875r1, 23/1965r2, 23/2003r1].

The transition from the low capability to the high capability mode is initiated by a TF, which may provide more accurate information about the expected transmission, e.g., the number of SSs, bandwidth, and MCSs. Such a PS mode is similar to the spatial multiplexing PS [2, Doc. 22/1423r2] proposed for 802.11be, but the latter requires an STA to listen to the channel with its whole operating bandwidth [2, Docs. 23/2003r1, 23/1875r1].

The STA needs time to switch from the low to the high capability mode. To make sure that the STA is ready to receive PPDUs right after the end of a TF, 802.11bn proposes to include in the TF the Intermediate Frame Check Sequence (IFCS) [2, Doc. 24/1246r0] followed by data or signaling for other client STAs and padding fields that extend the frame for the necessary duration [2, Docs. 23/1873r1, 24/0450r3, 24/1261r0]. The STA may start switching to the high capability mode after successfully decoding IFCS, while the other STAs wait until the end of the TF to then receive their portions of data. If the TF is intended for several STAs that have different switching time requirements, there can be multiple IFCSs within it [2, Doc. 24/1256r0].

Cross-link power save. Importantly, TGbn shall enhance the MLO framework by enabling cross-link PS, addressing inefficiencies in the current per-link power management [14, 34]. In 802.11be, when a client MLD wishes to enter PS mode across all links, it shall perform separate frame exchanges over each enabled link, which is both energy-inefficient and time-consuming [2, Doc. 24/0602r0]. To overcome this issue, the 802.11bn draft introduces the cross-link PS signaling, which allows a pair of MLDs to coordinate PS modes across multiple links via a single active link. For instance, an AP MLD and a client MLD can maintain connectivity over one link while keeping other radios off, activating them only upon receiving a cross-link wake-up frame. This approach provides greater flexibility in managing the tradeoff between power consumption and performance for MLDs.

AP power save. The previous 802.11 standards did not pay much attention to AP PS, however, nowadays, with the dense Wi-Fi network deployment and increasing capabilities of Wi-Fi devices, AP power consumption becomes an issue [14]. Currently, AP power consumption can be reduced by narrowing the AP channel bandwidth, disabling extra links of an AP MLD, decreasing the number of SSs, and using less-power-demanding MCSs [2, Docs. 23/0010r0, 23/2040r1]. However, such configuration changes have a transition duration of several beacon intervals, so using them results in long periods of reduced BSS performance in terms of throughput and latency.

Given the mentioned limitation, the 802.11bn draft [2, Docs. 24/0833r0, 23/2040r1] defines the new dynamic AP PS and scheduled AP PS mechanisms. The former is the same as the non-AP STA dynamic PS mode [2, Doc. 24/0659r1] described above. With the latter, similarly to the 802.11ah scheduled AP PS [19], an AP announces in beacons that during a time window, it will switch to specific states. The states may include [2, Doc. 24/0097r0]:

• the doze state, when the AP is unavailable;

• the listen state, when the AP can receive frames but defers responses to them;

• the reduced capability state, e.g., when the AP uses a single antenna, low bandwidth, or low-modulation MCSs;

• the awake state, when the AP can receive and transmit data without limitations.

Non-AP STAs that receive the schedule may turn off their radios when the AP is in the doze or listen state. On the AP side, it can be implemented using the Broadcast TWT mechanism with the specific signaling [34]. Specifically, scheduled AP PS enables the AP to schedule switching between the doze and awake states using the PUO mechanism (see Section 6.3). It can also be accompanied by the cross-link signaling [2, Doc. 24/1166r0], thus enforcing switching to the awake state on one link by signaling on the other link.

6.5 Open Issues

The described MAC features bring in not only new opportunities but also a range of open challenges that call for further research. To fully harness these new mechanisms and ensure their practical adoption in real-world devices, the community shall carefully investigate the following issues.

Starting with the features of low-latency channel access, P-EDCA raises several concerns. Foremost, its efficiency in complicated environments, including hidden nodes and several OBSSs, requires further study. Specifically, P-EDCA requires all STAs in the network to receive the defer signals (the DS-CTS frames). If some devices do not detect the defer signal, they may interfere with a low-latency transmission, making P-EDCA inefficient. Additional research is needed on cases where one BSS uses P-EDCA while another does not. If these BSSs have similar channel access parameters, the delay distribution of the BSS without P-EDCA likely degrades. Moreover, if the BSS that does not use P-EDCA tries to provide low-latency delivery using other mechanisms, e.g., tuning the EDCA parameters, the resulting performance of both mechanisms is unclear. Determining the appropriate parameters, such as the number of failed transmission attempts to start P-EDCA, is crucial for network performance. Finally, when an STA transmits with P-EDCA, the performance of the other STAs decreases because the former is not deprioritized after P-EDCA transmissions. This could be a significant problem if one or several STAs overuse P-EDCA.

Although NPCA clearly offers advantages, its performance needs to be evaluated under non-ideal channel conditions to fully assess its effectiveness. To communicate via NPCA, both devices shall switch to the NPCA primary channel, requiring a consistent view of channel occupancy. However, in Wi-Fi networks, especially in OBSS environments, devices may perceive third-party transmissions differently (known as channel assessment mismatch or the hidden node problem), which limits the utility of NPCA in such scenarios [2, Docs. 24/0829r0, 24/0670r2]. Additionally, NPCA operation may conflict with existing MAC logic, e.g., the standard allows an STA to terminate its TXOP earlier in the primary channel by sending the CF-End frame. Frequent channel switching can also significantly increase power consumption, making energy efficiency a key concern [2, Doc. 24/1260r2]. Moreover, to ensure synchronized switching, NPCA exchanges begin with a mandatory handshake, introducing overhead. Channel switching delays further contribute to channel time consumption [2, Docs. 23/0797r1, 23/0962r0]. Besides NPCA, the same open issue with channel switching delays and overhead applies to DSO and DBE. Also, advanced sounding and scheduling algorithms are in demand to perform effective dynamic bandwidth allocations and channel switching. Together, these factors may degrade the performance of the aforementioned innovations, making it essential to quantify their impact through further research.

Preemption also faces numerous open challenges, e.g., its coordination, response frame transmission schemes, enabling preemption in OBSS scenarios, protecting a TXOP from hidden nodes, etc. [2, Docs. 23/1242r1, 24/0636r1, 24/1074r0]. Harmonization of a preemption protocol and specific management frame formats requires much effort, which is why, after long discussions, preemption has been postponed for future generations. Moreover, the preemption latency gains should be clarified in a vast set of scenarios.

Some scenarios of Wi-Fi usage, e.g., XR, will certainly benefit from the improved P2P operation with the group-based TXOP allocation and off-channel P2P communication. However, these improvements raise several new issues. First, the AP needs to gather information about the STA traffic and directions of communication in order to allocate a TXOP for a group of STAs and to keep this information up-to-date. Using this mechanism with many groups may require a complicated scheduler that will provide a necessary amount of channel resources to different kinds of traffic and will manage them in a fair way. The scheduling problem is further complicated in the case of multi-BSS operation because APs from different BSSs need to coordinate their P2P channels.

Importantly, frequent unavailability at STAs receiving low-latency traffic creates scheduling challenges. The AP shall manage frame exchanges to ensure that STAs have sufficient time to report unavailability [2, Doc. 24/1848r1]. However, discrepancies between notified and actual unavailability durations complicate scheduling. Additionally, STAs that lose medium synchronization during long unavailability events call for recovery mechanisms, such as the AP-assisted recovery [2, Doc. 24/1848r1]. These issues highlight the need for robust algorithms to serve low-latency traffic in UHR networks while accounting for dynamic unavailability.

The new PS mechanisms raise some open issues. The first one is the handling of legacy devices and the combined use of the new and legacy PS mechanisms to increase efficiency [2, Doc. 24/0813r0]. The second one is the interdependence between AP PS and the functionality of the associated STAs. Specifically, the new AP PS mechanisms shall not disrupt the operation of legacy STAs that do not recognize the novel signaling. To address this concern, APs may be required to stay awake as long as at least one STA is active, as done in 802.11ah. This leaves the question of power saving open for further evaluation. The third one is restricting the network availability when the AP is in the doze or listen state, which may hinder the functionality of roaming and satisfying stringent QoS requirements.

7 Wi-Fi Networks Coordination Candidate Features

A key direction of the development of 802.11bn involves operating in deployments with numerous Wi-Fi networks. The most feasible way to enable stable Wi-Fi operation in such environments is to let multiple Wi-Fi networks (i.e., APs) coordinate with each other, which is directly what 802.11bn is going to do. Specifically, the developers focus on seamless roaming for devices moving between APs (see Section 7.1) and several Multi-AP coordination mechanisms that allow APs to optimally share resources between them (see Sections 7.2–7.9).

7.1 Seamless Roaming

Existing Wi-Fi struggles to provide low latency and high reliability when a device roams from one network to another. First, the fast fading exacerbates channel operations. Second, the switching procedure itself usually takes a long time [2, Doc. 24/0679r4]. This makes roaming one of the most anticipated directions to improve in Wi-Fi 8.

For more than a decade, several roaming improving amendments, such as 802.11k, 802.11v, 802.11w, and 802.11r, have been incorporated into Wi-Fi. Currently, even with all these amendments, when an STA moves between multiple APs, there are no strict timing guarantees [13]. The STA also needs to perform additional security negotiation and re-establishment of the context [2, Doc. 24/1851r2], i.e., the states of the Block Ack agreement, sequence numbers, TWT schedules, etc. Moreover, in some cases, such as STA’s PS mode, additional frame exchanges are required that can consume up to $70$ ms [2, Docs. 22/1874r2, 24/0679r4]. Such connection interruptions are much more than can be tolerated by emerging applications.

Fig. 11.

An example of the roaming procedure under seamless mobility domain (inspired by [2, Doc. 23/1908r2]).

The pipe dream here is to reduce roaming times to near zero, which raises the intention to develop new tools. The core offering of UHR is the introduction of the seamless mobility domain (SMD) covering multiple, possibly non-colocated, AP MLDs [34]. A client MLD (a non-AP MLD) initially authenticates with the SMD entity through an AP, much like in modern Wi-Fi networks. Then the client MLD can seamlessly roam from the current AP MLD to the target one, where both AP MLDs belong to the same SMD unit, as schematically shown in Fig. 11. The “seamlessness” of the roaming procedure is provided by the fact that the context can be transferred (at least partially) between AP MLDs within an SMD, as well as the unity of an SMD prevents from the need of full re-association and re-authentication. For example, based on the request frame from the client MLD or on the internal decision, the current AP MLD can transfer the context related to the client MLD to the target AP MLD via backhaul. Such coordination between AP MLDs (in fact, between networks) significantly reduces or even completely eliminates the time required to start data exchange between the client MLD and the target AP MLD after roaming. The current AP MLD may also forward frames to the target AP MLD or backward if some frames have not been delivered before roaming.

Interestingly, the concept of an SMD covering multiple AP MLDs allows for a step-by-step per-link transition from the current AP MLD to the target one [2, Doc. 23/1908r2]. In other words, the client MLD switches links from one AP MLD to another one at a time, which creates the intermediate roaming stage when the client MLD is concurrently connected by one link to the current AP MLD and by another link to the target AP MLD, as shown in Fig. 11. Thanks to the described concept, the client MLD has an uninterrupted connection with the SMD entity, so reliability and latency of data exchange are not affected by roaming.

7.2 Multi-AP Concept

A pivotal feature of Wi-Fi 8 is the Multi-AP functionality. Planned back in 802.11be, it found its way into 802.11bn. This feature is anticipated to enhance throughput, latency, as well as reliability of Wi-Fi networks.

With Multi-AP, a network consists of a sharing AP, shared APs, and client STAs. The sharing AP acts as an all-encompassing controller of shared APs that can transmit data to their STAs or receive it from their STAs in a coordinated way. Also, shared APs can be organized into a virtual BSS and can be connected to the sharing AP by a wired or a wireless link (or both) [2, Doc. 22/1394r1]. Prior to any coordinated transmission, the sharing AP learns the shared APs’ status (capabilities, channel, pathloss, buffer), selects the transmitting shared APs based on the provided information, and chooses the most suitable transmission mode based on a performance metric (throughput, latency, reliability). Commonly, a TXOP-based operation is considered, where the sharing AP gets channel access, and then coordinated transmissions occur. Specifically, the sharing AP sends a TF to synchronize all shared APs and dictate or negotiate the parameters of the following coordinated data transmission. Another option is the SP-based operation, where parameters and time periods of the coordinated transmissions are pre-negotiated. This is a long-term approach that results in less overhead, but induces less flexibility and reacts slower to sudden events. Note that the 802.11bn draft obligates the sharing and shared APs (and thus their BSSs) use the same primary $20$ MHz channel.

Different Multi-AP coordination schemes have been considered during discussions in TGbn, listed in Table 6 in the increasing order of potential gain and complexity. We discuss them with the most promising proposals, though TGbn has included only five items in the specification draft D0.3 [34].

7.3 Coordinated Channel Recommendation

As described in Section 6.1, to enable off-channel P2P communications, AP may advertise recommended channels and time intervals for P2P transmissions [2, Doc. 24/0393r3]. To align P2P schedules between OBSSs, the 802.11bn draft proposes the Coordinated Channel Recommendation (Co-CR) mechanism. With Co-CR, different APs can agree to promote the same channels for off-channel P2P communications [2, Doc. 23/1424r0], enabling STAs from different BSSs to communicate with each other. It should be noted that this mechanism is optional, so an AP may reject a request from another AP to coordinate its off-channel.

7.4 Coordinated R-TWT

Similar to Co-CR, Coordinated R-TWT (Co-RTWT) allows an AP to advertise R-TWT SPs and thereby extend their protection beyond its own BSS. Namely, the AP announces its R-TWT SPs to other BSSs, where the corresponding information is promoted to STAs, so they also comply with the restrictions. Hence, the problem of R-TWT being violated by OBSS interference diminishes. This is especially fruitful for enhanced latency and reliability. Besides, Co-RTWT might be bidirectional: instead of dictating their R-TWT SPs, the APs can coordinate them together so that they do not overlap [2, Docs. 23/2022r1, 24/0388r0].

7.5 Coordinated TDMA

As mentioned in Section 6.1, 802.11be defines the TXOP sharing mechanism that allows an AP to share its TXOP with an associated STA. TGbn extends this concept to the Multi-AP case, allowing an AP to share its TXOP with some other APs, which results in Coordinated Time Division Multiple Access (Co-TDMA). Co-TDMA is a straightforward scheme where APs schedule their transmissions to alternate (i.e., to avoid overlapping) time resources [10]. The protocol operates in three phases, as shown in Fig. 12: polling, TXOP allocation, and TXOP return. During the polling phase, the sharing AP sends an ICF to request information (e.g., Multi-AP capabilities, requested medium time, bandwidth, and traffic details [2, Docs. 23/0249r1, 24/0842r0]) from other APs, which respond if they intend to participate. In the TXOP allocation phase, the sharing AP transmits an MU-RTS TF, which informs the shared AP about the allocated TXOP portion. Then, the shared AP responds with a CTS frame, initiating its coordinated transmission. If a shared AP exhausts its buffer, it returns the unused TXOP to avoid medium underutilization. Potentially, the same TXOP can be shared with multiple APs sequentially [2, Doc. 23/0249r1]. Notably, in Co-TDMA, even APs with different primary channels can communicate [2, Doc. 23/0041r0] if their primary channels overlap within their operating bandwidths and they are not punctured [2, Doc. 23/0249r1].

Fig. 12.

An example of a Co-TDMA procedure between three APs [34].

The NAV protection is critical in Co-TDMA to ensure that only the intended BSS accesses the channel during shared TXOP periods. Usually, Wi-Fi devices use the NAV to reserve time for TXOPs: when an STA receives a frame with a NAV duration, it treats the medium as busy for that period. Consider a scenario where an AP in BSS1 acquires a TXOP and shares part of it with BSS2. The NAV set by the AP in BSS1 could inadvertently block transmissions in BSS2 during the shared period. To address this issue, two TXOP concepts are introduced [2, Doc. 23/1910r1]: the Nominal TXOP for coordinated access scheduling and the Basic TXOP for backward compatibility. However, hidden STAs missing NAV-setting frames can still cause TXOP handover failures in Co-TDMA. Solutions include adjusting the basic NAV signaling [2, Doc. 24/0227r1], using additional frames that end TXOP [2, Doc. 24/0382r0], or employing MU EDCA and RTS/CTS handshakes to reduce UL collisions [2, Docs. 24/0423r0, 24/1701r2]. These mechanisms ensure BSS2 STAs exclusively access the medium during the shared phase while BSS1 STAs regain access afterward.

7.6 Coordinated OFDMA

The coordinated OFDMA (Co-OFDMA) scheme improves upon the logic of Co-TDMA by operating with time-frequency resources (instead of time resources in Co-TDMA). Namely, the sharing AP gains access to a wide channel and allocates a portion of it to another (shared) AP for a particular time. Co-OFDMA works similarly to the classical OFDMA, but now across multiple neighboring APs (and thus BSSs). Unlike Co-TDMA and other Multi-AP schemes in the standard draft [34], the APs using Co-OFDMA shall have distinct primary channels.

Co-OFDMA has been discussed since the development of 802.11be and is widely studied in literature [48,49,50]. In [48], a simple Co-OFDMA-based transmission scheme is proposed, where a sharing AP sends a TF to a shared AP, inviting it in a coordinated transmission in the current TXOP. When scheduling, the primary channels of BSSs are always assigned to the respective APs, possibly along with some other subchannels. Within the allocated resources, the APs can send data in DL or solicit data from their STAs. All transmissions across the coordinating BSSs are intended to be synchronous. The authors of [48] develop a mathematical model for the simplest scenario with two $40$ MHz BSSs with different primary channels. Their analysis shows that Co-OFDMA increases the system throughput by up to $50\%$ (and higher with other enhancements) and guarantees UL-DL fairness. The paper [49] elaborates on CSI collection and scheduling in Co-OFDMA networks. It demonstrates significant reliability improvements, especially under low time-varying wireless channels. The authors note the importance of accurate CSI: measurements done with a period less than channel coherence time provide a close-to-optimal error rate, while longer periods lead to rapid performance deterioration. Thus, efficient Co-OFDMA scheduling in fast-varying channels is quite challenging.

The developers of 802.11bn have also explored Co-OFDMA in residential deployments [2, Doc. 22/1567r0]. They show that, although Co-OFDMA may reduce collisions and may allow using faster MCSs, the overall throughput gain is modest ($<10\%$). Given the minor throughput benefits, latency improvements are also questionable and should be clarified. Furthermore, researchers often assume that the sharing AP can allocate bandwidth in units of $20$ MHz. However, the 802.11 preamble typically carries identical information over $80$ MHz frequency blocks. So, while finer allocation enables enhanced scheduling flexibility and higher gains, it requires a new PPDU format for Co-OFDMA with narrower frequency granularity, which is a significant roadblock [2, Doc. 23/0768r0]. Addressing the effort required and the availability of Co-TDMA, which offers similar gains with lower specification complexity, Co-OFDMA has not seen further development in 802.11bn.

7.7 Coordinated Spatial Reuse

Coordinated SR (Co-SR) enhances the SR mechanism from 802.11ax, which reduces interference in OBSSs. Co-SR brings significant performance improvements by cooperatively controlling transmission power.

For Co-SR, two coordination types have been considered [2, Doc. 24/0640r0]. In a simpler one-way coordination, a sharing AP selects a transmission power sufficient to reach the target SINR at an STA and instructs the shared AP on its power level for a TXOP. Note that 802.11bn will support Co-SR only for two participating APs, i.e., a single shared AP. In a more effective two-way coordination, the sharing AP determines a transmission power for itself and the shared AP based on received signal strength and, optionally, target SINR, maximizing overall performance. For that, APs gather measurement data from their STAs, which monitor power of Beacon frames from neighboring BSSs, and share this information among themselves.

Interestingly, the performance difference between one-way and two-way coordination is minor [2, Doc. 24/0640r0], even if the sharing AP can solve the optimization problem effectively. The simulation results show that Co-SR achieves stable DL throughput gains compared to 802.11ax time-division duplex or EDCA [51] with no degradation [2, Doc. 24/0839r1] and nearly doubles it in some instances [2, Doc. 22/1822r0]. Moreover, Co-SR further outperforms no-SR and classic-SR cases [2, Doc. 22/1970r0]. On top of that, Co-SR can enjoy a variety of improvements. For example, the paper [52] proposes a hierarchical multi-armed bandit algorithm that identifies a list of AP-STA pairs, which can be serviced simultaneously using Co-SR.

Co-SR can work well in combination with other schemes. For instance, the paper [8] presents a framework integrating Co-SR and Co-TDMA. The authors propose methods for forming shared AP groups and compare scheduling algorithms in terms of throughput and delay. Additionally, Co-BF can extend Co-SR when an STA from another BSS is nearby and the AP has more antennas than SSs [2, Doc. 24/0635r0].

7.8 Coordinated Beamforming

Coordinated BF (Co-BF) is a more advanced Multi-AP coordination scheme than Co-SR. It allows APs to perform BF to increase power at receivers and reduce interference at the considered set of STAs in OBSSs by steering away from (nulling to) them. To accomplish this, each AP requires CSI knowledge for the channels between itself and those OBSS STAs in neighboring BSSs. Each AP then spends each available degree of freedom either to carry one SS of useful data to a desired user or to create one null to suppress interference at an OBSS device. For example, an AP that transmits using two SSs to a two-antenna STA while nulling to a single-antenna STA from OBSS required a total of three degrees of freedom (and at least three antennas at the AP). To reduce complexity, Co-BF in 802.11bn will be limited to two participating APs. Moreover, each AP may serve multiple STAs, but can allocate each no more than two SSs.

The Co-BF frame transmissions incur several synchronization issues. While non-synchronized transmissions offer scheduling flexibility [2, Doc. 24/0142r1], they increase error rates due to two key issues. First, legacy preamble portions that are not beamformed can have different durations and can interfere; beamforming them is feasible but unconventional. Second, nulling mitigates interference inside the OFDM symbols but not at their boundaries [2, Doc. 24/0635r0]. To prevent such an effect of extra interference at these boundaries of OFDM symbols, devices in neighboring BSSs can align OFDM symbols so that the interference falls within OFDM guard intervals and gets discarded. To avoid these issues, the Co-BF frame transmissions in 802.11bn shall be fully synchronized across different OBSSs. Specifically, the Co-BF frame exchange begins with the announcement of parameters that determine the legacy preamble portions, the number and durations of OFDM symbols, the number of STAs and their SSs, etc.

Notably, Co-BF leverages two sounding methods: sequential sounding and joint sounding [2, Doc. 24/1582r2]. Sequential sounding is a simple extension of MU sounding in 802.11be: a sharing AP sends NDPs and collects feedback from its STAs, and then shared APs do the same. This method accommodates lower-capability STAs and suits better for full-rank nulling, i.e., when APs spend all available degrees of freedom for beamforming and nulling. Although good for full-rank nulling, sequential sounding is less effective for partial-rank nulling, when APs use less degrees of freedom than available. In joint sounding, multiple APs transmit one common NDP frame with a longer orthogonal sequence used for sounding. This method enables partial-rank nulling and additionally provides about $15$–$50\%$ throughput gains [2, Doc. 23/0776r1], but requires STAs to support higher capabilities, e.g., eight SSs. The overhead difference between these sounding types is minimal [2, Doc. 24/1582r2], but joint sounding introduces additional complexity at the STA side because it is required to support more feedback transmissions and advanced interference cancellation algorithms. Co-BF adapts both sounding types and can operate in diverse network conditions and STA capabilities.

Co-BF is useful when OBSS STAs have delay-sensitive traffic. The simulation results show that it can achieve nearly an order of magnitude reduction in terms of worst-case packet latency over 802.11ax schemes; see [3] and [2, Doc. 24/0880r0]. Co-BF can also double the sum throughput compared to the single AP and Co-SR schemes in various scenarios [2, Doc. 19/0772r1]. However, to be effective, Co-BF needs to work in scenarios with high OBSS interference, where STAs need high SINR and APs have a sufficient degree of freedom for both BF and nulling [2, Doc. 24/0880r0].

7.9 Joint Transmission

The Joint Transmission (J-TX) scheme is the most advanced Multi-AP coordination scheme. All APs act as a single virtual multi-antenna AP, sharing not only control information but also user data [2, Doc. 22/2188r0]. Also, J-TX allows performing mutual BF for all devices simultaneously, theoretically providing the highest throughput among the schemes. However, it requires a multi-gigabit backhaul link and tight time and frequency synchronization [53]. Modern time and frequency synchronization techniques are generally sufficient [2, Doc. 22/1821r1], but additional verification is needed. Alternatively, STAs can assist the synchronization process [2, Doc. 24/0488r1]. They typically have timely knowledge of signal quality and can report the witnessed delay/phase/frequency deviations between the APs.

The simulation results from [2, Doc. 22/1821r1] indicate that J-TX achieves higher and more stable SINR than Co-BF. The bigger number of SSs gives a higher gain in J-TX: with one transmit SS, Co-BF and J-TX have similar throughput, but with two transmit SSs, J-TX outperforms Co-BF by $50\%$. Both Co-BF and J-TX significantly surpass uncoordinated schemes, although Co-BF has a slightly lower queuing delay because it does not require data sharing among APs. Further simulations show that J-TX also enhances robustness and maintains higher rates over longer ranges [2, Doc. 22/2188r0].

Still, being complicated, J-TX is currently not planned for 802.11bn. Yet, if Co-BF shows real benefit in deployment, the passion for a better tool will emerge. So, J-TX has the potential to enter Wi-Fi in the future and remains a relevant research topic.

7.10 Open Issues

While Wi-Fi 8 introduces advanced coordination mechanisms for roaming and multi-AP networks, several challenges remain unresolved. Open issues include roaming latency, Co-TDMA fairness and efficiency, fine time and frequency synchronization in J-TX and others.

Despite many benefits of seamless roaming, the community has raised concerns about this direction, making it a highly debated topic. First and foremost, to enable the SMD functionality, 802.11bn needs to determine the architecture and standardize the interface between the non-colocated AP MLDs, including inter-vendor operation [2, Docs. 23/0231r0, 23/0705r0, 23/1131r0]. Additionally, there are sparking discussions on how to identify links of non-colocated AP MLDs [2, Docs. 23/0170r1, 22/1910r3, 23/1090r0]. The discovery procedure for non-colocated APs, i.e., to which APs an STA can connect, is also required [2, Doc. 23/0705r0]. Moreover, the community highlights the inherent problems of unavoidable buffering, duplication, and out-of-order delivery, and thus communication delays within and between the affiliated APs, which may cause frames to be late by up to $20$ ms. The implementation of reorder buffers poses scalability challenges, as it remains unclear how and where the received reordering buffer should be stored [2, Doc. 23/0322r0]. Finally, the security issue of sharing security keys within SMD demands attention [2, Doc. 24/0679r4]: the frames exchange that is needed to re-establish a connection should be minimized while ensuring that this connection is secure.

Beyond seamless roaming, Multi-AP coordination presents additional open challenges that require further research. Researchers should clarify the prospective gains of Multi-AP across various scenarios. The schemes differ in potential efficiency, but also in complexity, network demands, and overhead size for negotiating AP transmissions, schedules, sounding, and data sharing. It is vital to estimate the gains from Multi-AP schemes while considering the overhead they impose.

Co-TDMA faces a fairness vs. efficiency dilemma: while granting a piece of TXOP to a shared AP improves latency, it may unfairly suppress neighboring non-participating STAs. Allowing non-interfering STAs to contend during the allocated time (supported only by UHR devices) or applying multiple EDCA tables (similar to MU EDCA) can address this issue [2, Doc. 25/0086r0], but it requires further investigation. Another Co-TDMA challenge is processing delays: since the scheduling complexity increases shared APs may require additional time. Solutions include adding padding to account for delays (reducing efficiency) or notifying shared APs of schedule allocations upfront via dedicated signaling [2, Doc. 23/0739r1]. Furthermore, the absence of accurate BSRs or airtime information can hinder medium efficiency. Enhancements such as allocating multiple STAs or improving BSR mechanisms are needed [2, Doc. 23/0261r0].

We have mentioned that Co-OFDMA, unlike other schemes, requires the coordinating APs to have different primary channels, which contradicts with the current 802.11bn draft. If the APs use the different primary channels, their coordination is complicated. Meanwhile, they still share the same wide bandwidth and may occasionally block each other’s primary channels. The APs may puncture the corresponding subchannels, which reduces contention and accelerates traffic delivery. However, consistent puncturing may lead to underutilization. The paper [54] proposes an algorithm to increase puncturing effectiveness in scenarios with multiple OBSSs with different primary channels. The ideas from the paper could be reused for Multi-AP coordination. Moreover, such puncturing algorithm could be merged with time scheduling by leveraging R-TWT SPs of neighboring BSSs and so provide higher performance and fairness. However, this direction has not yet been specifically explored.

Several issues remain in Co-SR. Developing a fast and accurate receive power measurement and exchange mechanism is crucial, as inaccuracies and overheads can degrade the Co-SR performance [2, Docs. 22/1822r0, 22/1970r0]. Additionally, algorithms for selecting optimal transmission parameters (e.g., power, MCS, bandwidth) are needed [2, Doc. 24/0529r1]. Finally, UL Co-SR requires further refinement, particularly for the simultaneous transmission of TFs [2, Doc. 22/1822r0].

In the context of Co-BF, the APs with limited spatial degrees of freedom will need to optimize their use. The APs shall use an algorithm that balances between the number of spatial data streams and interference nulling quality. A significant challenge here is CSI acquisition overhead: the sounding duration scales with the total number of SSs, creating a bottleneck. To mitigate this issue, implicit channel sounding can be employed, sacrificing some CSI accuracy for a substantial reduction in overhead. This tradeoff and the corresponding Co-BF performance shall be further evaluated.

For J-TX to be realized in Wi-Fi, APs shall be finely synchronized in time and frequency. More studies are required to prove that existing techniques satisfy J-TX requirements and to determine their limits (e.g., the maximum number of APs and STAs, bandwidth, and transmission duration).

Realizing wireless backhaul would be especially valuable for J-TX. It would greatly simplify deployment and enhance the appeal of Multi-AP schemes, yet achieving high-speed wireless backhaul between APs in practical scenarios is extremely challenging: its throughput shall at least surpass that of the AP-STA link, or better exceed it several times over [2, Doc. 19/1588r0]. Most proposals imply that AP-to-AP communication happens just before coordinated transmission, but introducing special coordination time instances could improve efficiency [2, Doc. 24/0478r0], though the gain is still unknown.

Shared AP selection is a challenging task in any Multi-AP scheme. Given all the information provided by the potentially-shared APs, the sharing AP needs to elect APs to share its limited resources, e.g., considering the low-latency traffic requirements [2, Doc. 24/0941r0]. Besides, optimal STA assignment in Multi-AP transmissions is an open problem as well [55].

For those seeking information on the Multi-AP topic beyond this tutorial, let us recommend the dedicated survey [56].

Finally, let us talk about relaying, which represents another promising yet challenging direction for Wi-Fi 8, particularly when integrated with Multi-AP coordination. While traditional relay approaches focus on extending coverage through sequential transmissions, their synergy with Multi-AP schemes could unlock new efficiencies in dense deployments. Contributions have demonstrated notable rate-vs-range improvements [2, Docs. 24/0650r1, 24/0887r0], but TGbn remains cautious due to specification and implementation complexity. Besides, a streamlined, latency-aware relay protocol is still missing, with open questions around acknowledgment procedures, sounding optimizations for relay links, and dynamic relay selection in Multi-AP environments [2, Docs. 24/0105r0, 24/1565r1, 24/1578r0]. Resolving these issues in a relatively simple and cost-effective could position relaying as a complementary tool for enhancing end-to-end QoS.

8 Artificial Intelligence & Machine Learning

To the end of this tutorial, let us briefly discuss the domain of Artificial Intelligence and Machine Learning (AIML), which is rapidly becoming a driver of innovation in wireless networking. Traditional machine learning algorithms, such as rate control, have already been integrated into Wi-Fi operating logic. Numerous studies demonstrate that AIML algorithms can improve resource utilization, minimize energy consumption, increase reliability, and enhance adaptability to changing environments [52, 57]. As a result, the recent surge in AIML advancements opens up new possibilities for Wi-Fi 8.

The AIML Topic Interest Group, operated concurrently with the UHR SG, created a technical report [2, Doc. 22/0987r25] that describes “use cases” for AIML applicability in 802.11 systems [2, Doc. 23/0924r2]. The first use case—“Enabling Efficient AIML Model Sharing”—focuses on issues of exchanging the training data and AIML models among STAs and APs, which is a general infrastructure step to enable AIML-based operations. Other use cases leverage AIML to enhance WLAN performance. These include:

• Compression of CSI feedback: AIML algorithms can refine CSI information from fewer and smaller feedback reports, enriching system throughput;

• Distributed channel access: AIML can optimize channel access parameters, enhancing throughput and latency;

• Roaming: AIML techniques can predict STA moving patterns and scanning thresholds, accelerating the roaming process and improving reliability;

• Multi-AP coordination: AIML can enhance Multi-AP performance through better coordination decisions, e.g., grouping and power allocations in Co-SR and Co-BF.

Although AIML is broadly recognized for its ability to improve WLAN performance, its usage within 802.11 systems, at least by now, does not directly require mandatory protocol changes. Consequently, while we anticipate the widespread adoption of AIML algorithms in Wi-Fi 8 devices, AIML is beyond the scope of the 802.11bn standard and thus this tutorial. We gratefully refer the reader to the studies listed in [58] and [2, Doc. 22/0987r25] for detailed information about KPIs, numerical gains, and open questions of each use case. Also note that the AIML Standing Committee, the successor to the AIML Topic Interest Group, continues to investigate use cases for AIML applicability in 802.11 systems and inquire into the technical feasibility of AIML-enabling features.

9 Conclusions

IEEE 802.11bn (Wi-Fi 8) represents the next phase of evolution in Wi-Fi technology. In contrast to the commonplace challenges of increasing peak network throughput, 802.11bn Ultra High Reliability is intended to focus on connectivity reliability, i.e., improving performance under non-ideal and realistic conditions, as well as in emerging scenarios. The development of Wi-Fi 8 is still ongoing. Nevertheless, the community already has a vision of what 802.11bn shall achieve and what mechanisms can help accomplish it.

In this tutorial, we have provided a background to Wi-Fi 8, highlighting the major milestones in the Wi-Fi evolution timeline. Then, we have discussed the target scenarios, including innovative XR, IIoT, and high-density deployments. The paper has mapped the goals to several KPIs that 802.11bn will enhance: throughput, latency, reliability, and power consumption. Furthermore, we have presented current views and perspectives on Wi-Fi 8 candidate features, divided into three groups: PHY, MAC, and Wi-Fi networks coordination. We have discussed the most vivid mechanisms, emphasizing their core ideas, potential gains, and open issues.

We hope that this tutorial will provide an overview of 802.11bn features and related challenges. It will help the community to focus on the most important issues and develop effective solutions for Wi-Fi 8, paving the way for ultra-reliable wireless networks.