An adaptive channel access method

Abstract

The application discloses a kind of self-adapting channel access methods, specifically including the following steps: step 1, listening stage: the channel energy of unlicensed frequency band and service label are monitored, current service type is identified and service priority weight is distributed;Step 2, decision stage: calculate the preemption factor by dynamic energy detection threshold, judge whether to trigger channel preemption based on preemption factor;Step 3, access stage: after preemption succeeds, avoid conflict with other protocol signals by pre-distributing time slot code, and send data;Step 4, conflict prediction and backoff adjustment stage: predict the conflict probability based on LSTM model, carry out adaptive backoff window adjustment, and handle abnormal conditions according to abnormal handling mechanism.The application can make high-priority service preoccupy channel more quickly, while avoiding low-priority service to carry out invalid competition when channel quality is poor, improve spectrum utilization, while reducing conflict rate and delay.

Description

Technical Field

[0001] This invention pertains to the field of wireless communication technology, and specifically relates to an adaptive channel access method. Background Technology

[0002] In unlicensed frequency bands (such as the 5.8GHz ISM band), the LBT (Before Access) mechanism is widely used to avoid collisions caused by multiple devices transmitting signals simultaneously. The LBT mechanism determines whether to access the channel by listening to whether the channel is idle. If the channel is idle, the device is allowed to transmit signals; otherwise, it enters a backoff state. Traditional LBT mechanisms use a fixed Energy Detection Threshold (EDT) and a backoff window. The workflow is as follows: before transmitting, the device listens to the channel energy; if it is below the fixed energy detection threshold (EDT), it is considered idle and transmits data; otherwise, it enters a backoff state.

[0003] With the explosive growth of differentiated service demands such as Ultra Reliable Low Latency Communication (URLLC), Enhanced Mobile Broadband (eMBB), and Massive Machine Type Communication (mMTC), the traditional LBT mechanism faces the following shortcomings:

[0004] 1. Mismatch between static thresholds and dynamic services: Fixed EDT cannot distinguish service priorities. For example, URLLC services need to quickly preempt the channel, but a high EDT may cause excessively long waiting times; while mMTC services are prone to redundant contention and channel congestion due to an excessively low EDT.

[0005] 2. Lack of environmental awareness in backoff mechanisms: Existing exponential backoff algorithms (such as CW adjustment in 802.11ax) rely only on the number of historical collisions and do not combine real-time channel load prediction, resulting in backoff windows that are too large (wasting resources) or too small (exacerbating collisions) under high load.

[0006] 3. Increased cross-protocol interference: When Wi-Fi coexists with heterogeneous protocols such as LTE-U / LAA, the fixed CCA (Free Channel Assessment) mechanism cannot identify protocol characteristics, leading to cross-system signal collisions (such as Wi-Fi preamble being misjudged as noise). Summary of the Invention

[0007] The purpose of this invention is to provide an adaptive channel access method. This invention enables high-priority services to preempt the channel more quickly, while avoiding invalid contention by low-priority services when the channel quality is poor. This improves spectrum utilization while reducing collision rate and latency.

[0008] The technical solution of this invention: an adaptive channel access method, specifically including the following steps:

[0009] Step 1, Listening Phase: Monitor the channel energy and service labels of the unlicensed frequency band, monitor channel state information based on the monitored channel energy, identify the current service type and assign service priority weights;

[0010] Step 2, Decision-making stage: Based on channel state information and service priority weight, calculate the preemption factor through dynamic energy detection threshold, and determine whether to trigger channel preemption based on the preemption factor;

[0011] Step 3, Access Phase: After successful preemption, data is sent by pre-allocating time slot codes to avoid conflicts with other protocol signals;

[0012] Step 4, Conflict Prediction and Backoff Adjustment Stage: Based on the LSTM model, the conflict probability is predicted, the adaptive backoff window is adjusted, and abnormal situations are handled according to the anomaly handling mechanism.

[0013] In the aforementioned adaptive channel access method, in step 1, the service label includes physical layer identifiers for URLLC, eMBB, and mMTC; in the allocation of service priority weights, the service weight for URLLC is 1, the service weight for eMBB is 0.7, and the service weight for mMTC is 0.3.

[0014] In the aforementioned adaptive channel access method, in step 2, the dynamic energy detection threshold (EDT) current The calculation formula is as follows:

[0015] EDT current =EDT base -α·P priority +β·CSI variation ;

[0016] Among them, EDT base It is the basic energy detection threshold, α is the business priority weighting factor, and P priority It is the service priority weight, β is the channel state change rate factor, and CSI variation It is the variance of the rate of change of channel state information calculated based on a sliding window.

[0017] In the aforementioned adaptive channel access method, the variance of the channel state information change rate (CSI) is... variation The calculation formula is as follows:

[0018] CSI variation =Var(CSI) window );

[0019] CSI window =[CSI t-N+1 CSI t-N+2 CSI t ];

[0020] Among them, CSI window It is a sliding window of size N, containing channel state information values ​​from the most recent N time moments; Var() represents the calculation of the variance function.

[0021] In the aforementioned adaptive channel access method, the preemption factor Q is calculated using the following formula:

[0022]

[0023] Among them, P priority It is the business priority weight, EDT current This is the current dynamic energy detection threshold.

[0024] In the aforementioned adaptive channel access method, step 4, the LSTM collision prediction model is implemented through the following steps:

[0025] Step 4.1: Collect historical collision counts, channel occupancy rates, and interference pulse characteristics as input features;

[0026] Step 4.2: Construct time series data using a sliding window, and perform normalization and outlier filtering;

[0027] Step 4.3: Deploy a lightweight LSTM model for real-time inference to predict conflict probabilities and determine risk levels.

[0028] In the aforementioned adaptive channel access method, the number of historical collisions N collision Let I be the number of channel collisions in the most recent N periods, and let I be the interference pulse characteristic. pulse The proportion of detected Wi-Fi preamble features is represented by the channel occupancy rate, which indicates the proportion of busy channel conditions. The calculation formula is as follows:

[0029]

[0030] In the formula, T busy It is the channel busy time, T total It is the total channel time.

[0031] In the aforementioned adaptive channel access method, the LSTM model predicts the collision probability P. collision The calculation formula is:

[0032] P collision =LSTM(X) history X current );

[0033] Among them, X history X represents historical window data. current Indicates the characteristics of the current moment;

[0034] The historical window data X history The calculation formula is as follows:

[0035] X history =[X t-4 ,X t-3 ,...,X t ];

[0036] Among them, X t Indicates the characteristics at time t;

[0037] The formula for calculating the current time feature is as follows:

[0038]

[0039] Where, N collision Indicates the number of historical collisions, CBR represents the channel occupancy rate, and I pulse Indicates an interference pulse;

[0040] Conflict probability P collision A value greater than 0.5 indicates a high risk of conflict.

[0041] Conflict probability P collision If the value is less than or equal to 0.5, the risk level is determined to be low conflict risk.

[0042] In the aforementioned adaptive channel access method, the adaptive backoff window adjustment includes contention window calculation, service priority coupling, and dynamic adjustment of the γ factor.

[0043] In the aforementioned adaptive channel access method, the contention window CW adaptive The calculation formula is as follows:

[0044] CW adaptive =CW min ×[1+floor(γ×P collision )];

[0045] Among them, CW min This represents the minimum contention window size, γ represents the dynamic conflict threshold adjustment factor, and floor() represents the floor function;

[0046] In the aforementioned adaptive channel access method, the service priority coupling process involves URLLC services using zero backoff window preemption of the protection time slot, and mMTC services using contention window (CW). adaptive Superimposed binary exponent retreat.

[0047] In the aforementioned adaptive channel access method, the calculation formula for the dynamic collision threshold adjustment factor γ is as follows:

[0048]

[0049] In the aforementioned adaptive channel access method, the anomaly handling mechanism includes collision oscillation suppression and model failure backoff;

[0050] The conflict oscillation suppression refers to three consecutive P... collision When the value is greater than 0.9, force a CW reset. adaptive =CW min ×5;

[0051] The model failure rollback refers to switching to the baseline backoff algorithm when the LSTM model inference times out and / or becomes abnormal.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] The dynamic energy detection threshold mechanism of this invention makes the timing of channel access for services of different priorities more reasonable, reducing conflicts caused by unreasonable competition. Through the coupling mechanism of dynamic energy detection threshold and service priority, this invention allows high-priority services to quickly preempt the channel when channel conditions permit, avoiding wasted waiting time due to fixed thresholds; low-priority services increase their energy detection threshold when the channel deteriorates, reducing ineffective competition for busy channels, thus enabling a more rational allocation of spectrum resources to different services and improving overall spectrum utilization. For high-priority services such as URLLC, this invention lowers their energy detection threshold, allowing them to preempt the channel first, significantly reducing transmission latency for high-priority services. This invention, through LSTM collision prediction and linear backoff window adjustment mechanisms, accurately predicts the collision probability and adaptively adjusts the backoff window, allowing equipment to compete for the channel more scientifically based on actual conditions, effectively reducing the possibility of collisions. Simultaneously, the LSTM collision prediction mechanism of this invention allows high-priority services to skip random competition and directly enter the transmission stage when the probability of collision is high, further ensuring low latency requirements. In scenarios where 3GPP LBT and Wi-Fi coexist, this invention plays a crucial role in cross-layer collaborative LTT protocol reconfiguration. This invention can automatically identify Wi-Fi preambles and dynamically adjust the LBT listening period to avoid conflicts with Wi-Fi signals. Furthermore, by adaptively adjusting the conflict probability threshold γ, this invention optimizes the access strategies for devices using different protocols, ensuring harmonious coexistence between different protocols and improving communication stability in heterogeneous network environments. Attached Figure Description

[0054] Figure 1 This is a flowchart of the method of the present invention;

[0055] Figure 2 This is a block diagram of the method of the present invention;

[0056] Figure 3 This is a schematic diagram illustrating a multi-service coexistence application scenario of the present invention;

[0057] Figure 4 This is a flowchart of the decision-making stage of the present invention;

[0058] Figure 5 This is a flowchart of the dynamic energy detection threshold update process of the present invention;

[0059] Figure 6 This is a flowchart of the conflict prediction and backoff window length adjustment based on LSTM of the present invention. Detailed Implementation

[0060] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0061] Example 1: An adaptive channel access method, such as Figure 1 As shown, by reconstructing the LBT into a three-layer LTT (Listen-Decision-Access) architecture, a "decision layer" is inserted between the traditional "listen-transmit" stages to achieve joint optimization of service priority, channel state, and cross-protocol characteristics. By adding a "decision" layer after the "listen" stage of the traditional LBT, combined with dynamic service priority and channel state prediction, a flexible energy detection threshold (EDT) adjustment mechanism is adopted, and channel access strategies are optimized through intelligent prediction of collision probability and adaptive backoff windows. The modules of this invention are as follows: Figure 2 As shown, the adaptive channel access method includes a dynamic energy detection threshold calculation module, a channel energy detection module, and a channel preemption module. The dynamic energy detection threshold calculation module includes a cross-layer listening module, a channel state monitoring module, and a service priority judgment module. The cross-layer listening module acquires the current channel energy level, service tag, and Wi-Fi preamble characteristics. The channel state monitoring module monitors the channel state information based on the acquired current channel energy level and evaluates channel quality. The service priority judgment module determines priority based on the service tag and assigns priority weights. The dynamic energy detection threshold calculation module dynamically calculates the energy detection threshold based on the results from the cross-layer listening module, channel state monitoring module, and service priority judgment module. The channel energy detection module detects the energy level in the channel and judges the channel state based on the dynamic energy detection threshold. The channel preemption module makes a preemption decision based on the results from the channel energy detection module. The specific steps of this invention are as follows:

[0062] Step 1, Listening Phase: Monitor the channel energy and service labels of the unlicensed frequency band, monitor channel state information based on the monitored channel energy, identify the current service type and assign service priority weights;

[0063] In this step, the service tags include physical layer identifiers for URLLC, eMBB, and mMTC; such as Figure 3 As shown, various services (URLLC, eMBB, mMTC, etc.) share the unlicensed frequency band for data transmission. Due to different service requirements, each service uses different channel access parameters on the unlicensed frequency band.

[0064] In this embodiment, the weight of URLLC service is 1, the weight of eMBB service is 0.7, and the weight of mMTC service is 0.3 in the allocation of service priority weights.

[0065] During the listening phase, channel and service information are collected. The physical layer and MAC layer work in parallel and collaboratively across layers, and the collected information is transmitted to the decision-making phase to provide a basis for subsequent decision-making. The physical layer detects channel energy and service labels (such as URLLC identifiers), while the MAC layer synchronously receives Wi-Fi preamble features (such as OFDM symbol structures).

[0066] Specifically, Physical Layer (PHY) sensing includes channel energy detection and service label identification. Channel energy detection involves the physical layer continuously monitoring the energy level in the channel, typically measured using metrics such as Received Signal Strength Indicator (RSSI) or Received Signal Strength (RSS). A higher RSSI / RSS value indicates a stronger signal in the channel, potentially suggesting other devices are transmitting data. Service label identification involves the physical layer attempting to identify whether a specific service's signal is present in the current channel. This is achieved by parsing specific fields in the received signal. For example, if the device supports URLLC services, the physical layer will attempt to detect the presence of URLLC service identifiers (e.g., specific physical layer frame headers or control information) in the signal.

[0067] Specifically, the MAC layer is responsible for receiving and parsing the preamble of the Wi-Fi signal. The preamble is the beginning of the Wi-Fi signal and contains important control information, such as the modulation scheme and coding rate. By parsing the preamble, the MAC layer can determine whether a Wi-Fi signal exists in the current channel and obtain the characteristic information of the Wi-Fi signal (e.g., OFDM symbol structure).

[0068] Step 2, Decision-making stage: Based on channel state information and service priority weights, a preemption factor is calculated using a dynamic energy detection threshold. Based on this preemption factor, it is determined whether to trigger channel preemption. Figure 4 As shown;

[0069] In this step, the decision-making phase, based on the information collected in the listening phase, comprehensively considers service priority and channel status to decide whether to preempt the channel.

[0070] Figure 5 The flowchart for updating the current dynamic energy detection threshold is as follows:

[0071] Step 1: Initialize basic configuration parameters: Set initial values ​​for the parameters according to the dynamic energy detection threshold calculation formula. baseThe basic energy detection threshold can be configured according to the actual application scenario. This value is usually set based on regulatory restrictions, hardware performance, and empirical values; in this embodiment, it is set to -90dBm. α is the service priority weighting factor, used to adjust the impact of priority on EDT; in this embodiment, it is set to 10. β is the channel state change rate factor, used to adjust the impact of channel state on EDT; in this embodiment, it is set to 5. N is the size of the configured sliding window, used to calculate the variance of the CSI change rate; in this embodiment, it is set to 20, and then the CSI is initialized. window Empty.

[0072] Step 2: Periodic or event-triggered updates. Periodic updates refer to executing the update process once every certain period of time (e.g., 10ms). Event-triggered updates refer to triggering the update process when the following events are detected: a new service arrives with a different priority than the current service; or the channel state changes significantly (e.g., CSI changes exceed a certain threshold).

[0073] Step 3: Obtain the current business priority weight (P) priority Then, it identifies the current service type. Based on the header information of the data packet or through other signaling mechanisms, it identifies the service type that needs to be sent (e.g., URLLC, eMBB, mMTC); based on the service type, it obtains the corresponding priority weight P. priority .

[0074] Step 4: Obtain Current Channel State Information (CSI). Measure the current Channel State Information (CSI) through the physical layer. CSI can be Channel Quality Indicator (CQI), Signal-to-Interference-plus-Noise Ratio (SINR), or Received Signal Strength Indicator (RSSI), etc. Add the current CSI value (CSIt) to the sliding window CSI. window middle.

[0075] Step 5: Determine CSI window Has the window size reached the preset window size N? If the window is full, remove CSI. window The oldest CSI value is used to add the current CSI value (CSIt) to the CSI. window At the end, keep the window size at N. If the window is not full, directly add the current CSI value (CSIt) to the CSI value. window The end. CSI window It is expressed as follows:

[0076] CSI window =[CSI t-N+1 CSI t-N+2 CSI t ].

[0077] Step 6: Determine if the window size meets the conditions for calculating variance: Ensure that the CSI window contains at least two CSI values ​​in order to calculate the variance. The specific formula is as follows:

[0078] CSI variation =Var(CSI) window );

[0079] Among them, CSI window It is a sliding window of size N, containing channel state information values ​​from the most recent N time moments; Var() represents the calculation of the variance function.

[0080] Step 7: Calculate the Dynamic Energy Detection Threshold (EDT) current The calculation formula is as follows:

[0081] EDT current =EDT base -α·P priority +β·CSI variation ;

[0082] Among them, EDT base This is the basic energy detection threshold, which can be configured according to the actual application scenario; α is the business priority weighting factor, used to adjust the impact of priority on EDT; P priority This refers to the service priority weights (i.e., URLLC = 1, eMBB = 0.7, mMTC = 0.3), β is the channel state change rate factor, used to adjust the impact of channel state on EDT, and CSI. variation It is the variance of the rate of change of channel state information calculated based on a sliding window, used for real-time detection of channel degradation.

[0083] Step 8: Update the Energy Detection Threshold (EDT) current The calculated EDT current It is applied to the physical layer as the energy detection threshold for subsequent channel eavesdropping, ending the current periodic or event-triggered energy detection threshold update process and waiting for the next update.

[0084] The preemption factor Q is used to measure the priority of preempting the channel, and the calculation formula is as follows:

[0085]

[0086] Among them, P priority This refers to the business priority weight, which indicates the priority of the current business. Higher priority businesses have higher weight values; EDT current It is the current dynamic energy detection threshold, which indicates the current channel busyness. A higher EDT indicates that the channel is relatively busy.

[0087] The preemption factor Q is compared with the preset threshold Q.threshold Compare the values. If Q is greater than the preset threshold Q... threshold If this occurs, the preemption logic is triggered, indicating that the current service has a high priority and the channel is relatively idle, so it is possible to attempt to preempt the channel.

[0088] Step 3, Access Phase: After successful preemption, data is sent by pre-allocating time slot codes to avoid conflicts with other protocol signals;

[0089] In this step, the access phase involves preempting the channel and transmitting data while avoiding conflicts with Wi-Fi signals. A pre-allocated timeslot code (e.g., an OFDM symbol mask) is used to preempt the channel and avoid collisions with Wi-Fi signals; after preempting the channel, the device can begin transmitting data. A timeslot code is a special signal sequence that can be used to identify different devices or services. By sending a specific timeslot code, a device can inform other devices that it is using the channel, thereby avoiding collisions.

[0090] Step 4, Conflict Prediction and Backoff Adjustment Stage: Based on the LSTM model, the conflict probability is predicted, the adaptive backoff window is adjusted, and abnormal situations are handled according to the anomaly handling mechanism.

[0091] In this step, such as Figure 6 As shown, the specific process for predicting conflict probability and adjusting the adaptive backoff window is as follows:

[0092] Step 1: Initialize the parameters;

[0093] ①LSTM Input Dimension: Define the input feature dimension (number of historical collisions N) collision Channel occupancy rate (CBR) and interference pulse characteristics (I) pulse );

[0094] ② Competition window parameter: CW min γ is the minimum contention window size (e.g., 16 time slots); γ is the dynamic conflict threshold adjustment factor (initial value: γ = 2);

[0095] ③ Data collection period: Set the statistical period (e.g., collect data once every 100ms);

[0096] ④LSTM model configuration: Number of hidden layer neurons: 32 (can be adjusted as needed); Timesteps: 5 (5 historical data periods, can be adjusted as needed); Activation function: tanh / ReLU; Output layer: Sigmoid (conflict probability 0~1).

[0097] Step 2: Data acquisition and preprocessing;

[0098] Number of historical conflicts N collision Channel occupancy rate (CBR) and interference pulse characteristics (I)pulse Data collection is being conducted; historical conflict count N collision This represents the number of channel collisions within the most recent N periods; interference pulse characteristic I pulse The percentage of detected Wi-Fi preamble features (such as OFDM symbols); Channel occupancy rate (CBR) represents the percentage of busy channel states, calculated using the following formula:

[0099]

[0100] Among them, T busy It is the channel busy time, T total It is the total channel time.

[0101] Data preprocessing includes sliding window construction, normalization, and outlier filtering; sliding window construction refers to maintaining a time series window to store historical data from 5 periods; normalization involves processing N... collision CBR and I pulse Perform Min-Max normalization; outlier filtering means removing noisy data caused by transient interference (e.g., CBR>95% is considered outlier).

[0102] Step 3: Train and deploy the LSTM model;

[0103] ①Offline training phase

[0104] Dataset generation: Collect historical conflict event data (input X, label Y = actual conflict result 0 / 1)

[0105] Model training: Loss = Binary Cross-Entropy, Optimizer: Adam;

[0106] Model validation: Prediction accuracy is evaluated using ROC curves;

[0107] ② Online deployment

[0108] Real-time inference engine: Deploying lightweight LSTM models (such as TensorFlow Lite) with dynamic parameter updates:

[0109] The model weights are updated at a fixed interval, such as every 24 hours.

[0110] Fixed intervals, such as incremental updates every hour (online learning of the latest conflict patterns);

[0111] Step 4: Perform real-time conflict probability prediction;

[0112] Input feature construction: Extracting features at the current time step: Piecing together historical window data: X history =[X t-4 ,Xt-3 ,...,X t ];

[0113] Among them, X t This represents the characteristics at time t.

[0114] Conflict probability is calculated using an LSTM model for inference.

[0115] P collision =LSTM(X) history X current );

[0116] Then, the probability threshold is determined:

[0117] If P collision A value greater than 0.5 indicates a high risk of conflict.

[0118] If P collision ≤0.5 is considered a low risk of conflict.

[0119] Step 5: Adaptive backoff window adjustment;

[0120] Competition window calculation, dynamically adjusted based on predicted conflict probability:

[0121] CW adaptive =CW min ×[1+floor(γ×P collision )];

[0122] Among them, CW min This represents the minimum contention window size, γ represents the dynamic conflict threshold adjustment factor, and floor() represents the floor function.

[0123] Implement business priority coupling. For high-priority services (URLLC): if P collision If the value is greater than 0.5, skip the random backoff and directly preempt the protection time slot, while adopting an asymmetric competition strategy (such as fixed backoff window CW=0). For low-priority services (mMTC): force the application of CW. adaptive And this is compounded by index retreat.

[0124] The γ factor is dynamically adjusted and updated adaptively based on network load.

[0125]

[0126] γ∈[1,5];

[0127] Step 6: Handle abnormal situations according to the exception handling mechanism;

[0128] Conflict oscillation suppression: If P is predicted three times consecutively collision >0.9, force reset CWadaptive =CW min ×5;

[0129] Model failure rollback: When LSTM inference times out or malfunctions, switch to the baseline backoff algorithm (such as 802.11axEDCA).

[0130] The dynamic energy detection threshold mechanism of this invention makes the timing of channel access for services of different priorities more reasonable, reducing conflicts caused by unreasonable competition. Through the coupling mechanism of dynamic energy detection threshold and service priority, this invention allows high-priority services to quickly preempt the channel when channel conditions permit, avoiding wasted waiting time due to fixed thresholds; low-priority services increase their energy detection threshold when the channel deteriorates, reducing ineffective competition for busy channels, thus enabling a more rational allocation of spectrum resources to different services and improving overall spectrum utilization. For high-priority services such as URLLC, this invention lowers their energy detection threshold, allowing them to preempt the channel first, significantly reducing transmission latency for high-priority services. This invention, through LSTM collision prediction and linear backoff window adjustment mechanisms, accurately predicts the collision probability and adaptively adjusts the backoff window, allowing equipment to compete for the channel more scientifically based on actual conditions, effectively reducing the possibility of collisions. Simultaneously, the LSTM collision prediction mechanism of this invention allows high-priority services to skip random competition and directly enter the transmission stage when the probability of collision is high, further ensuring low latency requirements. In scenarios where 3GPP LBT and Wi-Fi coexist, this invention plays a crucial role in cross-layer collaborative LTT protocol reconfiguration. This invention can automatically identify Wi-Fi preambles and dynamically adjust the LBT listening period to avoid conflicts with Wi-Fi signals. Furthermore, by adaptively adjusting the conflict probability threshold γ, this invention optimizes the access strategies for devices using different protocols, ensuring harmonious coexistence between different protocols and improving communication stability in heterogeneous network environments.

[0131] In summary, this invention enables high-priority services to preempt the channel more quickly, while avoiding ineffective competition by low-priority services when the channel quality is poor. This improves spectrum utilization while reducing collision rate and latency.

Claims

1. An adaptive channel access method, characterized in that, Specifically, the following steps are included: Step 1, Listening Phase: Monitor the channel energy and service labels of the unlicensed frequency band, monitor channel state information based on the monitored channel energy, identify the current service type and assign service priority weights; Step 2, Decision-making stage: Based on channel state information and service priority weight, calculate the preemption factor through dynamic energy detection threshold, and determine whether to trigger channel preemption based on the preemption factor; Step 3, Access Phase: After successful preemption, data is sent by pre-allocating time slot codes to avoid conflicts with other protocol signals; Step 4, Conflict Prediction and Backoff Adjustment Stage: Based on the LSTM model, the conflict probability is predicted, the adaptive backoff window is adjusted, and abnormal situations are handled according to the anomaly handling mechanism. In step 2, the dynamic energy detection threshold The calculation formula is as follows: ； in, It is the basic energy detection threshold. It is a business priority weighting factor. It is a business priority weight. It is the channel state change rate factor. It is the variance of the rate of change of channel state information calculated based on a sliding window; The preemption factor The calculation formula is as follows: ； In step 4, the LSTM model predicts the conflict probability through the following steps: Step 4.1: Collect historical collision counts, channel occupancy rates, and interference pulse characteristics as input features; Step 4.2: Construct time series data using a sliding window, and perform normalization and outlier filtering; Step 4.3: Deploy a lightweight LSTM model for real-time inference to predict conflict probabilities and determine risk levels; The LSTM model predicts the probability of conflict. The calculation formula is: ； in, Represents historical window data. Indicates the characteristics of the current moment; The historical window data The calculation formula is as follows: ； in, express Time characteristics; The formula for calculating the current time feature is as follows: ； in, Indicates the number of historical conflicts. Indicates channel occupancy rate. Indicates an interference pulse; Conflict probability A value greater than 0.5 indicates a high risk of conflict. Conflict probability If the value is less than or equal to 0.5, the risk level is determined to be low conflict risk.

2. The adaptive channel access method according to claim 1, characterized in that: In step 1, the service tags include physical layer identifiers for URLLC, eMBB, and mMTC; in the allocation of service priority weights, the weight of URLLC service is 1, the weight of eMBB service is 0.7, and the weight of mMTC service is 0.

3.

3. The adaptive channel access method according to claim 1, characterized in that: The variance of the channel state information change rate The calculation formula is as follows: ； ； in, It is the size of A sliding window containing the most recent Channel state information values ​​at each moment; This indicates the calculation of the variance function.

4. The adaptive channel access method according to claim 1, characterized in that: The number of historical conflicts For the most recent Number of channel collisions within a period, characteristics of interference pulses The proportion of detected Wi-Fi preamble features is represented by the channel occupancy rate, which indicates the proportion of busy channel conditions. The calculation formula is as follows: ； In the formula, It is a busy time for the channel. It is the total channel time.

5. The adaptive channel access method according to claim 1, characterized in that: The adaptive backoff window adjustment includes contention window calculation, business priority coupling, and dynamic conflict threshold adjustment factors. Dynamic adjustment.

6. The adaptive channel access method according to claim 5, characterized in that: The competition window The calculation formula is as follows: ； in, Indicates the minimum contention window size. This represents the dynamic conflict threshold adjustment factor. This represents the floor function.

7. The adaptive channel access method according to claim 5, characterized in that: The process of prioritizing services involves URLLC services using a zero-backoff window preemption to secure time slots, while mMTC services use a contention window. Superimposed binary exponent retreat.

8. The adaptive channel access method according to claim 5, characterized in that: The dynamic conflict threshold adjustment factor The calculation formula is as follows: 。 9. The adaptive channel access method according to claim 1, characterized in that: The anomaly handling mechanism includes conflict oscillation suppression and model failure rollback; The conflict oscillation suppression refers to three consecutive times. Force reset when the value is greater than 0.

9. ; The model failure rollback refers to switching to the baseline backoff algorithm when the LSTM model inference times out and / or becomes abnormal.

Images