802.11ax introduces OFDMA technology to provide parallel transmission with multiple users simultaneously. OFDMA divides the transmission bandwidth into multiple non-overlapping orthogonal subcarriers, and the user can select one or more subcarriers for parallel transmission. Since resource management and allocation have always been a key issue in wireless networks, many researchers have studied the OFDMA technology of WIFI6.

Specifically, in [13], the authors propose a distributed RU selection method for deep reinforcement learning (C-DRL) based on convolutional neural networks (CNNs), in which each STA is trained locally online based only on energy detection and confirmation packets. In [14], to improve the efficiency of uplink random access (OURA), a multi-busy tone arbitration mechanism (MBTA) was developed to reduce conflicts between sites competing for RU. In [15], the author considers the joint strategy design of resource unit allocation and power control of WIFI6 uplink, and splits the problem into minimized power problems and RU allocation problems under delay constraints to achieve the goal of reducing average latency. In [16], the authors combine WIFI6’s OFDMA technology and TWT technology to propose a scheduling channel access scheme to ensure that the STA is close to but smaller than RU at each TWT wake-up time, avoiding conflicts and improving throughput.

In [1], the authors designed a heuristic algorithm for multiple, dense user wireless scenarios to meet throughput while resisting AP failures. In [17], the authors mainly study the self-healing ability of wireless cellular networks, and compensate adjacent cells through antenna reconstruction and power compensation, thereby filling the coverage gap and improving users’ QoS.

However, the above work rarely combines the technology of WIFI6 with the fault recovery scenario. A fault determination and service set division model for AP failures in WIFI6 scenarios using the technical features of OFDMA is proposed. Then a WiFi6 dynamic channel optimization method for fault tolerance is proposed. To sum up, this paper studies the fault tolerance of wireless networks to ensure that the network after an AP of failure can still complete the system data transmission, and ultimately ensure the continuity of the network.

Table 1 lists the differences between this paper and the existing works.

WIFI6 introduces OFDMA technology to provide multi-user parallel transmission. OFDMA divides the transmission bandwidth into multiple non-overlapping orthogonal subcarriers, and users can select one or more subcarriers for parallel transmission. As shown in Fig. 1, the traditional WIFI protocol adopts the transmission mode of Orthogonal Frequency Division Multiplexing (OFDM), which allows only one user to transmit at any time; WIFI6 achieves multi-user parallel transmission by introducing OFDMA technology at the PHY layer, OFDMA has the following three advantages over OFDM technology: First, a more detailed channel allocation, WIFI6 can choose the best RU according to the channel quality for data transmission; The second is to provide better QoS, in OFDMA mode, since a sender only occupies part of the entire channel’s resources, it can send data from multiple users at the same time, thus reducing the latency of STA access; The third is more user concurrency and higher user bandwidth, as can be seen from the Fig. 1, OFDMA can satisfy multi-user transmission in the same time slot, providing more user concurrency.

WiFi6 enhances the AP’s function, which supports two uplink transmission modes, random access and scheduled access [18]. Downlink transmission is simpler compared to uplink transmission, and this paper focuses on the more complex OFDMA uplink access problem:

WiFi6 supports two access modes for OFDMA uplink transmission, uplink OFDMA random access (UORA) and uplink OFDMA non-random access (UONRA). The former STA obtains RU resources by competing for access, while the latter requires STAs and AP to interact with each other in advance to complete the data transmission.

Fig. 2 shows the transmission process of UONRA. Through the information interaction process between the buffer status reports (BSR) frame and the Clear to Send (CTS) frame, the AP obtains the STA cache report information, and then the AP starts the uplink data transmission by sending a trigger frame (TF) frame, which contains the scheduling information (transmission time, power control information and RU allocation information). Then, the STAs receiving the TF frame upload the data in the physical protocol data unit (PPDU). Finally, the AP feeds back the MU-acknowledgement character (MU-ACK) to confirm the acceptance [19,20].

In view of the requirement of high reliable and high-quality communication in the wireless network of power, it is assumed that all STAs are covered by AP signals in a global strong coverage network, and each STA is guaranteed to be within the communication range of at least two APs. The neighbor AP listening method is used to determine whether the AP is faulty. In 802.11 communication protocol, data is encapsulated into a frame. A frame refers to a block of data in communication. According to 802.11 protocol, there are three types of frames, namely control frames, management frames and data frames.

The common control frames include Request to Send (RTS) frames, CTS frames and Acknowledge Character (ACK) frames, which are mainly used for handshake communication and forwarding acknowledgement at the end of the competitive and non-competitive periods. As the name implies, data frames are used to carry data.

Management frames are mainly used to control and negotiate the relationship between APs and STAs, such as association, authentication, synchronization, etc. This section focuses on management frames. Among them, the beacon frame is a very important frame, mainly used to announce the presence of an AP network. By sending beacons periodically, the AP can let the mobile users know the existence of the network and thus adjust the necessary parameters to join the network. In the basic network, the AP must be responsible for sending beacon frames, which range over the basic service area. In this paper, the principle of beacon frame is used. Each AP not only periodically sends beacon frame to the STAs under its jurisdiction, but also receives beacon frames from its neighboring APs.

AP will send beacon frames periodically to announce the existence of the 802.11 network. Neighboring APs always listens for the beacon frame. If the node fails and the neighboring node fails to receive the beacon frame within a certain period, the node loses the ability to communicate.

Fig. 3 is a beacon frame sending mode, the STAs passively listen to the beacon frame around the AP periodic broadcast (about 100 ms), the frame contains all the service parameters of the AP. Not only all STAs can receive the beacon frame, but also neighboring APs can listen to beacon frames to determine whether the network is functioning properly.

If an AP fails to receive beacon frames from neighboring APs on time, the AP may be restarted due to abnormal power failure or common configuration delivery. The restart time of a WIFI6 device is about 3–5 min, but the processing delay of most real-time applications is in the millisecond type. If STAs wait for the device to restart and then transfer task, it will eventually cause most tasks to fail to meet the latency standard. This paper takes 5 min as the cycle. After dynamic channel allocation is started, beacon frames from the neighboring AP are received again every 5 min. If they are received, the restart is considered successful. After an AP is restarted, the terminal connects to the original AP after completing the current services to ensure full utilization of resources.

At the same time, if the STA cannot complete the data transmission after repeated retransmission, a new AP is selected for reconnection. Selecting connection for APs is described in the next section.

Based on the above analysis, this paper determines whether the AP is in normal operation by receiving beacon frames sent periodically by the neighboring APs. This section describes that when an AP fails, the terminals covered by that failed AP will join the neighboring APs to continue receiving services under the condition that the service delay is satisfied. Fig. 4 shows an overlapping basic service set (OBSS), where each STA transmits data within the coverage of its own AP.

When an AP in the network fails, the STAs will select a neighboring AP to connect, as shown in Fig. 5. These users are set to STAif. However, there are many factors that determine whether AP and STA can communicate, such as the power level of the sender, the distance between the sender, the SINR, etc. The transmitting power of all APs is adjusted to be the same and the maximum. However, there may be multiple selection APs left by users. How to overcome the influence of multiple factors and select the optimal neighboring AP to access is the main problem to be solved in this section. This algorithm can be seen in Table 3. (1)maxmize ∑j=1MSINRj (2)subjecttotj≤Thrj ∀j∈STAIf (3)∑i∈Ixij=1,∀j∈STAIf (4)Cj=Blog2(1+SINRj),∀j∈J

Thrj represents the latency threshold for STA j to complete the task, tj is represented by the formula tj=djCj.dj represent the service size of STA j, Cj represents the information transmission rate of STA j, and J represents the collection of all STA devices under the network. B is the channel bandwidth. Eq. (3) indicates that each STA can only maintain a connection to at most one AP at the same time.

After the affected STAs re-connect AP, this paper weighs the balance between the information interaction overhead and the data transmission rate, and designs a dynamic access selection model (DASM) for different data frames, specifically the long frame scheduling access (UONRA), and the short frame random competitive access (UORA), which improves the efficiency of data transmission and the utilization of resources. Fig. 2 is the transmission process of UONRA [21].

As can be seen from the Fig. 6, the UONRA mechanism requires two rounds of information interaction to be completed before data transmission, which is not friendly to the transmission of short data frames. Meanwhile, it has been demonstrated in the literature that UORA is more suitable for short frame transmission [14].

However, it is impossible to distinguish whether a data frame is long or short with a definite value. In addition, whether from the perspective of resource awesome or scheduling allocation, it is desirable to reduce blank fields to improve resource utilization. 802.11ax stipulates that the AP can appropriately adjust the duration of each Transmission Opportunity (TXOP) [22]. Therefore, the transmission of multi-frame data can be performed within one TXOP. For synchronous transmission, this paper sets the length of each frame as the maximum length of the group of data frames, and padding is performed on other terminals. Minimize the length difference of each group of data frames to achieve the goal of improving resource utilization.

The AP allocates the ratio of short and long frames to the number of requested users to obtain the boundary of short and long frames. Subsequently, the STA is informed of the boundary based on the information interaction phase and selects the appropriate transmission mode.

The AP differentiates RU resources by following the principle of minimizing the internal variance between UORA and UONRA scheduling data while satisfying the resource reservation variance varS. Define the constraint αsch that indicates the number of resources used for scheduling accesses as a percentage of the total number of scheduling. αsch is a value which between 0 and 1. Based on the above analysis, the optimization objectives and constraints of DASM are summarized as follows. (5)minvarS, s∈{SUONRA,SUORA} (6)Subjectto KSUONRA≥αschKwhere varS=1|S|[∑s=1S(Thrs−Thr¯S)], SUONRA and SUORA represent the user sets for UONRA and UORA respectively. K is the total RU resources. Thrs is the maximum completion delay threshold of the STA. Thr¯S=∑s=1SThrs|S|, |S| is the number of all users.

After the DASM, random access for short frames is not discussed in this paper, while for scheduling access, SAO-MF introduces a scheduling utility function and considers both delay and priority factors.

The SINR determines the MCS level of the RU assigned to the STA. In SAO-MF, the SINR is defined by formula (7): (7)SINRjk=Powjk|hjk|2N0k+∑z≠jPowzk|hzk|2where hjk represents the channel gain of user j on RU k. Powjk represents the power of user j on RU k. N0k represents the white noise on RU k.

The STA feeds CSI information to the AP through BSR frames, In SAO-MF, the statistical modeling of the channel is referenced to [23] as follows. (8)hjk=GtGrLpAsAfwhere Lp is path loss, Gt represent antenna gain of transmitter, Gr represent antenna gain of receiver, As and Af are two random variables, which represent shadow effect and fast fading effect respectively.

Each STA feeds Thrs back to the AP through BSR interaction, and SAO-MF specifies that each frame of data transmission length should be kept within the delay threshold, otherwise that data will wait for the next frame to be transmitted.

SAO-MF specifies five levels of task types, and STA feeds this information to the AP during the BSR interaction. Referring to [24], SAO-MF sets the following five services to classify the priority, where a lager number indicates a lower priority. As shown in Table 4.

Finally, the SAO-MF mechanism to maximize the scheduling utility function is formulated as follows. (9)max∑k∈K∑j∈SUONRA⁡xjkUjk (10)subjectto xjk∈{0,1},∀j∈SUONRA,∀k∈K (11)∑j∈SUONRAxjk≤1,∀k∈K (12)αdat∑k∈K∑j∈SUONRATjk≤dj,∀j∈SUONRAwhere xjk indicates whether the kth RU is allocated to user j, xjk=1 indicates yes, otherwise 0. αdat represents the percentage of the RUs allocated to STA. For example, αdat=1 means that all data is sent. The purpose of formula (9) is to maximize the total utility function of the system.

In SAO-MF, the utility function is equal to SINR transmitted by each STA. (13)Ujk=SINRjkQoSPjwhere Ujk represents the utility function of user j on RU k. SINRjk represents the SINR transmitted by user j on RU k. QoSPj is the urgency of services. QoS is between 0 and 1.

Defining the emergency factor Pj=αdel(Thrj∑j∈SUONRAThrj)+αurg(Prij∑j∈SUONRAPrij) ensures that high-priority users get a larger utility function. We can see that the smaller the value of Pj, the higher the priority, the larger the value of Ujk. In addition, if QoS=1, service priority and emergency are not considered.