Machine type communication equipment channel access method based on position

Abstract

The invention discloses a position-based channel access method for machine type communication equipment, which comprises the following steps of: collecting position and communication parameter historical data, namely original data or characteristic data, from roadside equipment through a base station, and learning channel selection with a period of time or periodic characteristics; therefore, a mapping function of position information and approximate optimal channel access is obtained, and the sum rate of an uplink access system is maximized; based on a channel access strategy of a position, intelligent channel prediction and authorization-free rapid access are realized by adopting a centralized training and distributed execution framework driven by a deep neural network; for the active equipment without authorization transmission, the access success probability is improved, and the communication overhead and power consumption are reduced; when the equipment connection density range is fixed, the convergence speed of unsupervised learning and the calculation overhead of a learning model are improved, efficient deployment on resource-constrained equipment is further achieved, and the method has the advantages that the overhead of control signaling interaction is reduced, the calculation energy consumption of an equipment end is reduced, and the road monitoring continuity is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of communication technologies, and in particular, to a method for a location-based machine-type communication device to access a channel. Background Art

[0002] Currently, in the large-scale machine-type communication scenario of a smart city, efficiently and timely monitoring the environment of a road section with dense traffic flow is the basis for ensuring road traffic safety. Therefore, in order to achieve real-time data collection, complex environment perception, key data transmission, and vehicle-road collaborative control, a large number of Internet of Things devices are often arranged on both sides of the road, including road surface state sensors, vision sensors, vehicle alarm receivers, communication and collaborative perception modules, etc.

[0003] However, due to the characteristics of high-density connection, low-power communication, and sporadic transmission of a large number of machine-type communication devices accessing, access failure or delay problems may occur frequently. First, there are network congestion and resource competition factors. For example, a large number of devices usually adopt contention-based random access. When they simultaneously initiate access requests to a capacity-limited base station, unreasonable network resource planning, insufficient available channels, and a sharp increase in the probability of preamble collision will all lead to access failure. Second, there are wireless environment and signal interference factors. Affected by dense buildings, complex terrain, electromagnetic noise, and co-channel interference, severely attenuated device signals may lead to access failure because they cannot meet the minimum signal-to-noise ratio requirement. Third, there are device capability and power consumption limitation factors. For example, in the commonly used energy-saving mode, some low-end sensors need to be awakened after a long sleep cycle to respond to the network, thus there is a possibility of missing the access window. In addition, there is also the possibility of security authentication timeout failure due to the inability to support complex algorithms.

[0004] The present invention is applicable to the scenario where all machine-type communication devices in a smart city have sensor and cellular network communication functions. In the real urban communication scenario, limited by objective conditions such as network coverage blind spots, complex multipath propagation environment, and significant penetration loss, existing active terminal devices face three technical bottlenecks in channel access optimization: insufficient response efficiency of high-delay-sensitive access algorithms; difficulty in jointly optimizing signaling interaction overhead and device energy consumption; limited convergence speed of the optimal access decision in a dynamic environment. This directly results in the difficulty of traditional access schemes in jointly maximizing network capacity and spectral efficiency, becoming a key factor restricting the communication performance in high-density urban areas.

[0005] In the prior art, existing license-free random access technologies mainly focus on improving signal modulation and demodulation, coding and decoding, synchronization, etc. at the physical layer, as well as operations such as resource scheduling, conflict resolution, and access control at the data link layer (MAC layer). For example, in

CN 116684986 A

CN 116321515 A

US2024196429 A1

CN 118612885 A

CN 115021780 A

CN 115499938 A

[0006] Currently, most of the prior art adopts fixed or semi-static resource allocation strategies, such as fixed time-frequency resource block partitioning, which are difficult to meet the high-density access requirements of a large number of users. Although some solutions introduce non-orthogonal multiple access or interference cancellation technologies to improve the multiplexing efficiency, their computational complexity and implementation cost are relatively high, and advanced algorithms such as zero-forcing precoding have high hardware requirements. At the same time, multi-user interference and the diversity of power levels increase the complexity of channel separation and interference cancellation. In addition, the existing architecture relies on the base station to obtain the closed-loop feedback of key information such as instantaneous channel state or user geolocation, but there are challenges in the real-time acquisition and update of channel information in a large-scale user scenario. Especially in a complex wireless environment, the time-varying characteristics of the channel are similar to the time scale of feedback processing delay, which easily leads to the partial invalidation of the acquired channel information when used for resource allocation. For example, the effective time window is shortened to the sub-millisecond level and the feedback delay significantly occupies the channel coherence time.

[0007] Furthermore, existing solutions usually use the pilot sequences or preambles sent by terminal devices as the main input. However, the randomness and collision probability of these data are relatively high, and preamble collisions are likely to occur, especially when the number of users is large. Among them, some solutions require pre-training a deep learning model based on supervised learning, which relies on historical data for prediction. However, the accuracy and timeliness of the original data and its labeled data used for model training may be insufficient. For example, preamble collisions may deteriorate the signal-to-noise ratio of training samples, thereby affecting the prediction effect.

[0008] In summary, there are at least one of the following technical problems:

[0009] It is difficult to adapt to the high-density access requirements of a large number of users by adopting a fixed or semi-static resource allocation strategy.

[0010] Although some solutions introduce non-orthogonal multiple access or interference cancellation techniques, their computational complexity and implementation cost are relatively high, and they have high requirements for hardware.

[0011] The multi-user interference and the diversity of power levels increase the complexity of channel separation and interference cancellation.

[0012] The base station needs to obtain channel state information or user location information to optimize resource allocation. However, in a large-scale user scenario, the real-time acquisition and update of channel information face challenges. Especially in a complex wireless environment, the accuracy and timeliness of channel estimation may be insufficient.

[0013] Existing solutions usually use the pilot sequences or preambles sent by terminal devices as the main input. However, the randomness and collision probability of these data are relatively high, and preamble collisions are likely to occur, especially when the number of users is large.

[0014] Some solutions require pre-training a deep learning model based on supervised learning and rely on historical data for prediction. However, the accuracy and timeliness of the original data and its labeled data used for model training may be insufficient, thereby affecting the prediction effect. SUMMARY OF THE INVENTION

[0015] The main objective of the present invention is to provide a method for channel access of machine-type communication devices based on location, so as to solve the problem that in the prior art, it is difficult to adapt to the high-density access requirements of a large number of users by using fixed or semi-static resource allocation strategies. Although some solutions introduce non-orthogonal multiple access or interference cancellation technologies, their computational complexity and implementation costs are relatively high, and they have relatively high requirements for hardware. The diversity of multi-user interference and power levels increases the complexity of channel separation and interference cancellation. It is necessary for the base station to obtain channel state information or user location information to optimize resource allocation, but in a large-scale user scenario, the real-time acquisition and update of channel information face challenges. Especially in a complex wireless environment, the accuracy and timeliness of channel estimation may be insufficient. Existing solutions usually rely on pilot sequences or preambles sent by terminal devices as the main input, but the randomness and collision probability of these data are relatively high, especially when the number of users is large, it is easy to cause preamble collision problems. Some solutions require pre-training a deep learning model based on supervised learning, which relies on historical data for prediction. However, the accuracy and real-time nature of the original data and its labeled data used for model training may be insufficient, thus affecting the prediction effect.

[0016] To achieve the above objective, according to one aspect of the present invention, there is provided a method for channel access of machine-type communication devices based on location, including:

[0017] Collecting historical data of location and communication parameters from roadside devices by the base station, that is, original data or feature data, and learning the channel selection for a period of time or with periodic characteristics, so as to obtain a mapping function between location information and approximately optimal channel access, and maximizing the sum rate of the uplink access system;

[0018] Based on the location-based channel access strategy, a centralized training and decentralized execution framework driven by a deep neural network is adopted to achieve intelligent channel prediction and unauthorized fast access;

[0019] For active devices with unauthorized transmission, improving the access success probability, reducing communication overhead and power consumption;

[0020] When the connection density range of fixed devices is reached, the convergence speed of unsupervised learning can be increased and the computational overhead of the learning model can be reduced, thereby realizing the efficient deployment of the model on resource-constrained devices and significantly improving its generalization ability.

[0021] Preferably, the centralized training and decentralized execution framework includes:

[0022] Step 1: Determine the RRC state of the NR system in which the device is located;

[0023] Step 2: Receive a data packet containing model parameters transmitted from the base station;

[0024] Step 3: Record or update the location information of the current device;

[0025] Step 4: Trigger the access process to establish a connection or optimize the process to resume the connection;

[0026] Step 5: Select a channel according to the model output to perform unscheduled uplink transmission;

[0027] Step 6: If the data arrival, the availability of pre-configured resources, and the uplink synchronization status meet the requirements, proceed to Step 3. If the data arrival, the pre-configured resources, and the uplink synchronization status do not meet the requirements, proceed to Step 1.

[0028] Preferably, in Step 1, based on historical signaling interaction, system message parsing, context management, and timer control, the device senses and determines the current RRC state, as well as the availability of uplink synchronization status and pre-configured resources. If the device is in the idle state and the synchronization is invalid, the device needs to first trigger contention-based random access to obtain a valid timing advance (TA) value. If the device is in the inactive state, uplink synchronization can be maintained through periodic TA updates.

[0029] Preferably, Step 2 includes: The device sends an access request. When a device in the idle state initially establishes a connection, it randomly selects a specific physical random access channel (PRACH) resource group.

[0030] Preferably, Step 2 includes: The base station sends a recovery confirmation message. The nearby base station directly responds to contention resolution and uplink resource allocation. By using the collected raw-level data and channel knowledge map, the base station trains a neural network model in the access training module with the goal of maximizing the system sum rate until the convergence condition is met.

[0031] Preferably, Step 2 includes: The base station transmits model parameters. Combining with the device status selection, the base station dynamically selects a transmission mechanism and performs model distribution. That is, after completing the training of the model with raw-level data, the base station transmits a model parameter data packet containing neural network weights to a large-scale device through the physical downlink shared channel (PDSCH), supplemented by physical downlink control channel (PDCCH) scheduling and reference signal (RS) optimization. For devices with a low update frequency, lightweight model parameters are broadcast through the physical broadcast channel (PBCH) to update the execution network.

[0032] Preferably, the algorithm framework for centralized training includes:

[0033] First, use orthogonal frequency division multiplexing (OFDM) technology to divide the NR channel bandwidth into multiple mutually orthogonal sub-channels. Each sub-channel includes a sub-carrier, and at most one active device is assigned within the same time slot;

[0034] Secondly, using the historical data of the wireless propagation environment and the uplink transmission system in the urban scenario, preprocess the raw-level data to obtain an encapsulated dataset, then call the encapsulated dataset training configuration to find the optimal value of the model, and finally save the determined model parameters.

[0035] Thirdly, transform the integer programming problem of the channel problem into a sum-rate maximization problem, that is, maximize the sum of the achievable rates of the system under the given OFDM bandwidth budget, sub-channel allocation number constraint, and minimum required data rate.

[0036] Finally, after the centralized training is completed, save the sharded model parameters for decentralized execution by the devices.

[0037] Preferably, step 3 includes:

[0038] Read the location. After the active device with no adjusted deployment location reads the location information from the memory, it performs license-free access based on the signal quality using the two-step method or the four-step method.

[0039] Update the location. For the active device with displacement alarm caused by event-driven or dynamic calibration, the device first triggers repositioning and obtains the updated location information.

[0040] Preferably, step 4 includes: when the active device initiates access from the idle state, it needs to complete the initial access through the complete random access process; when the device is in the non-active state, it can directly send a resume request to the base station through the two-step RRC ResumeRequest.

[0041] Preferably, step 5 includes:

[0042] Collect historical location data and data rate measurements;

[0043] Learn the active mode, channel quality, and interference level;

[0044] Learn the mapping function between the device location and channel access;

[0045] Perform generalization learning;

[0046] Select a channel according to the probability distribution of the output channel selection;

[0047] The uplink transmission data is used for environmental monitoring and vehicle-road cooperation.

[0048] Applying the technical solution of the present invention has the following technical effects:

[0049] Through centralized training, the base station can learn the historical data of channel states in complex environments and the spatial characteristics of device distributions, establish a global channel selection prediction model, thereby reducing the overhead of control signaling interaction. For devices that need to be relocated, only the updated coordinates need to be input again to obtain the model output, significantly reducing the computing energy consumption of device terminals.

[0050] In the decentralized execution phase, active devices can quickly select the optimal channel for license-free access only based on the locally sensed location information and the policies output by the model, which is beneficial to ensuring the continuity of road monitoring. It is adapted to large-scale deployed road traffic scenarios, while reducing deployment costs and privacy risks. In addition, license-free access reduces the continuous listening energy consumption of devices during the contention phase.

[0051] Through the in-depth mining of location data by the neural network model, accident prevention, emergency response, and traffic efficiency improvement can be achieved without violating privacy, building a safe and sustainable traffic ecosystem for smart cities. In particular, it has core advantages for road safety applications, such as scalability and low-cost deployment, real-time traffic situation awareness and accident warning, dynamic traffic flow optimization and emergency response optimization, etc. Brief Description of the Drawings

[0052] The specification drawings forming a part of this application are used to provide a further understanding of the present invention. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0053] Figure 1 Shows a schematic flowchart of a location-based machine-type communication device channel access method according to the present invention;

[0054] Figure 2 Shows Figure 1 A channel access scenario diagram of a high-density urban environment of the location-based machine-type communication device channel access method in;

[0055] Figure 3 Shows Figure 1 A schematic diagram of a license-free access wireless communication system of the location-based machine-type communication device channel access method in;

[0056] Figure 4 Shows Figure 1 A user flowchart of a centralized training and decentralized execution framework of the location-based machine-type communication device channel access method in;

[0057] Figure 5 Shows Figure 1 A channel access flowchart of connection establishment of the location-based machine-type communication device channel access method in;

[0058] Figure 6 shows Figure 1 the flowchart of location-based random access of the location-based machine-type communication device channel access method in

[0059] Figure 7 shows Figure 1 the task flowchart of high-density scenario access of the location-based machine-type communication device channel access method in Detailed implementation manners

[0060] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present invention will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0061] As Figures 1 to 7 shown, the embodiment of the present invention provides a location-based machine-type communication device channel access method, including: collecting historical data of positions and communication parameters from roadside devices by a base station, that is, original data or feature data, and learning the channel selection for a period of time or with periodic characteristics, so as to obtain a mapping function between location information and approximate optimal channel access, and maximize the sum rate of the uplink access system; based on the location-based channel access strategy, adopting a centralized training and decentralized execution framework driven by a deep neural network to achieve intelligent channel prediction and unauthorized fast access; for active devices with unauthorized transmission, improving the access success probability, reducing communication overhead and power consumption; when the connection density range of fixed devices is fixed, improving the convergence speed of unsupervised learning and the computational overhead of the learning model, and then achieving efficient deployment on resource-constrained devices and enhancing the generalization of the model.

[0062] The present invention adopts a centralized training and decentralized execution framework driven by a deep neural network to achieve intelligent channel prediction and unauthorized fast access, and promotes the implementation of large-scale Internet of Things. The present invention uses the idea of unsupervised learning, collects historical data of positions and communication parameters from roadside devices by a base station, that is, original data or feature data, and learns the channel selection for a period of time or with periodic characteristics, so as to obtain a mapping function between location information and approximate optimal channel access, and maximize the sum rate of the uplink access system. For active devices with unauthorized transmission, the access success probability can be improved, and the communication overhead and power consumption can be reduced. When the connection density range of fixed devices is fixed, the labeling cost of training data and the computational overhead of the learning model can be reduced, and then efficient deployment on resource-constrained devices can be achieved, and the generalization of the model can be enhanced.

[0063] The present invention is oriented towards large-scale Internet of Things applications such as smart cities. The wireless communication system is used to provide communication services targeting sensing and data collection for machine-type communication devices requesting access simultaneously, that is, to monitor the road conditions in real time, collect data, and issue safety warnings. In a real urban street scenario, the channel access of a single active device is interfered by the signals of other devices. According to an intelligent sub-channel selection module, the data transceiver process in the multi-carrier fifth-generation mobile communication system (5G) new radio (NR) wireless communication system is realized by local decision-making. The present invention proposes a workflow of a centralized training and decentralized execution framework, including all steps of broadcasting model parameters from the base station, initialization / reconnection, performing sub-channel access, and data transmission.

[0064] The present invention relies on large-scale machine-type communication devices (such as environmental monitoring sensors, microwave radars, lightweight cameras, and infrastructure status monitoring sensors, etc.) and new radio access technologies (such as the 5G NR radio access network of Release 18 of 3GPP TS 38.104, etc.). In the frequency bands of the newly specified FR2 (FR2-1 corresponds to the 24250 MHz–52600 MHz millimeter-wave band and FR2-2 corresponds to the 52600 MHz–71000 MHz millimeter-wave band), the corresponding working frequency bands are n257-n263, the frequency spacing of adjacent sub-carriers is 60 kHz and 120 kHz, and each resource block contains 12 sub-carriers. Since 5G NR uses a smaller sub-carrier spacing, in an urban scenario, small data, low-power, and high-density uplink transmissions can better utilize the balance characteristics it provides among coverage, power consumption, and anti-interference ability.

[0065] Among them, affected by application requirements, environmental factors, positioning technologies, etc., static sensors or roadside infrastructure obtain current location information in the ways of on-demand sporadic updates (e.g., once a month when the signal source changes or the location migrates) and monitoring cycle updates (e.g., daily calibration at low frequency and hourly reporting at high frequency), generating raw-level data, including Cartesian coordinate data within the network coverage. For devices with long-term unchanged deployment locations, the pre-configured coordinates are read from the device storage; for devices that need to be dynamically adjusted, the positioning updates are indirectly inferred by technologies such as network assistance, environmental feature matching, and location mapping table maintenance. Within a period of time, the measurement data records accessed by devices in the service area can be collected by the base station, including the connection density range with scene characteristics and channel data at specific locations. Among them, the device connection density supported in the urban scene is called feature-level data. Compared with the raw-level data, the feature-level data can more accurately reflect the user interference degree and multipath channel characteristics, which is beneficial to guiding the hyperparameter optimization of long-time-domain prediction. After preprocessing the above two levels of data, an unlabeled dataset is generated for model training of the access training module. Therefore, while optimizing the access strategy, the labeling cost is reduced and the data utilization rate is improved. The access execution module can download the trained neural network parameters and then input the coordinates of the device into the model to select the base station for access without waiting for dynamic scheduling authorization.

[0066] In the training stage, the base station can collect the raw-level data of active users based on historical random access processes. Considering the impact of changes in the spatial distribution of devices on the prediction accuracy, a periodic evaluation is set to retrain the constructed model. Optionally, the base station can obtain the feature-level data according to the preambles reported by the devices or the pre-configured resource groups.

[0067] The process is as follows:

[0068] 1. Determine the RRC state of the device in the NR system: Based on historical signaling interactions, system message parsing, context management, and timer control, the device senses and determines what RRC state it is currently in (e.g., idle state or inactive state), as well as the uplink synchronization state and the availability of pre-configured resources. If the device is in the idle state and the synchronization is invalid, the device needs to first trigger contention-based random access to obtain a valid timing advance (TA) value; if the device is in the inactive state, the uplink synchronization can be maintained through periodic TA updates.

[0069] 2. Receive a data packet containing model parameters transmitted from the base station. 2a. Device sends an access request: When an idle-state device initially establishes a connection, it randomly selects a specific Physical Random Access Channel (PRACH) resource group, for example, and sends an RRC connection request or a small data packet to any serving base station. The channel resources for actual data transmission, such as the time-frequency location of the Physical Uplink Shared Channel (PUSCH), will be dynamically allocated by the base station in the random access response. 2b. Base station replies with an acknowledgment message: The nearby base station directly responds with contention resolution and uplink resource allocation, such as replying with acknowledgment messages like contention resolution identifiers, TA amount adjustment, etc. Through the collected raw-level data and the channel knowledge map, the base station trains a neural network model in the access training module with the goal of maximizing the system sum rate until the convergence condition is met. 2c. Base station transmits model parameters:

[0070] Combined with the device state selection, the base station dynamically selects a transmission mechanism and conducts model distribution. That is, after completing the training of the model with raw-level data, the base station transmits a data packet containing model parameters such as neural network weights to the massive devices through the Downlink Shared Channel (PDSCH), supplemented by Physical Downlink Control Channel (PDCCH) scheduling and Reference Signal (RS) optimization techniques. In addition, for devices with a lower update frequency, lightweight model parameters can also be broadcast through the Physical Broadcast Channel (PBCH) to update the execution network.

[0071] The algorithm framework for centralized training is designed as follows: First, the Orthogonal Frequency Division Multiplexing (OFDM) technology is used to divide the NR channel bandwidth into multiple mutually orthogonal sub-channels, where each sub-channel includes a sub-carrier, and at most one active device is allocated within the same time slot. For example, configure the sub-carriers according to the channel bandwidth supported by FR2 and the corresponding maximum transmission bandwidth to confirm the number of sub-channels supported by the network. For example, if the frequency interval between adjacent sub-carriers is 60 kHz and the number of resource blocks is 66, then the number of sub-carriers is 794; if the frequency interval between adjacent sub-carriers is 120 kHz and the number of resource blocks is 32, then the number of sub-carriers is 384. Second, use the historical data of the wireless propagation environment and the uplink transmission system in the urban scenario to preprocess the raw-level data to obtain an encapsulated data set, then call the encapsulated data set training configuration to find the optimal value of the model, and finally save the determined model parameters. Third, transform the integer programming problem of the channel problem into a sum rate maximization problem, that is, maximize the sum of the achievable rates of the system under the given OFDM bandwidth budget, sub-channel allocation number constraint, and minimum achievable data rate requirement. It can be solved by means such as heuristic search and distance-based slot ALOHA, etc., but the cost is high and it is not applicable to considering the complex wireless environment in reality.

[0072] Among them, three methods are designed to optimize the model training: a) Conduct model training in the base station access training module, that is, through parameter sharing, structure decoupling, and dynamic batching, share the same set of weights for the location information inputs of simultaneously active users, and then independently apply operations such as fully connected layers for end-to-end training. b) Minimize the custom loss function transformed from the optimization objective during training, that is, the weighted sum of the channel capacity and the norm matrix metric terms. c) Optionally, if the system knows the feature-level data as prior information, it can assist in setting the weight coefficients of the loss function to improve the prediction accuracy of the model for performance metrics. For example, resource optimization can be achieved by reasonably allocating the weight terms representing the channel capacity and user collisions.

[0073] Finally, after the centralized training is completed, the sharded model parameters are saved for decentralized execution by the devices. 1. Record or update the location information of the current device. Combining the device spatial distribution characteristics reflected by the historical data in steps (2a) and (2b), the device obtains its own location information, triggers the random access step according to the current state, and then selects a channel and transmits data according to the model output of the updated parameters in step (2c). 3a. Read location. For active devices with no adjusted deployment location, after reading the location information from the memory, use the two-step method or the four-step method for grant-free access based on the signal quality. If the reference signal received power (RSRP) measured by the device is higher than the msgA-RSRP-Threshold, preferentially select the two-step RA; otherwise, use the four-step RA. For example, use the two-step method in high signal-to-noise ratio regions, while in weak coverage or edge regions, the device needs to fallback to the four-step method to ensure access reliability. In particular, to reduce the interaction with the base station, the non-active state of RRC_CONNECTED is introduced in NR to enable small data packet transmission (SDT) and the two-step method to reuse Msg3 for data transmission. Although the optimized process skips the RRC connection reconstruction, the uplink resources still need to be obtained through the random access process. 3b. Update location. For active devices with displacement alarms caused by event-driven or dynamic calibration, the device first triggers a relocalization and obtains updated location information. For example, in network-assisted positioning technologies, the device achieves centimeter-level positioning by measuring the time difference of arrival (TDOA) or angle of arrival (AoA) of multiple 5G NR-supported dedicated positioning reference signals (PRS); in location inference based on the random access process, the base station indirectly infers the device location based on the preambles or preconfigured resource groups reported by the device, where different preamble sequences correspond to different geographical regions. Similarly, limited by resource preconfiguration and signal quality, the device uses the four-step method or the two-step method to perform random access, and both need to complete channel access through a contention mechanism.

[0074] 4. Trigger the access process to establish a connection or optimize the process to resume the connection. The triggering conditions for the two-step random access are mainly initial access (from RRC_IDLE to RRC_CONNECTED) and the state transition of RRC_INACTIVE. When an active device initiates access from the idle state, it needs to complete the initial access through a complete random access process; when the device is in the non-active state, it can directly send a resume request to the base station through the RRC Resume Request of the two-step method. For example, in Release 18, the two-step random access reduces the latency and signaling overhead by combining signaling steps and optimizing resource allocation. 4a. MsgA transmission (PRACH preamble + PUSCH payload transmission), that is, the active device combines the preamble (used to identify the UE and assist the base station to complete uplink synchronization) and data (including the unique identifier and RRC connection request), and sends them to the base station through time-division multiplexing (TDM) or frequency-division multiplexing (FDM). 4b. MsgB reception (random access response (RAR) + contention resolution), that is, the base station issues TA, temporary C-RNTI allocation, contention resolution identifier, and backoff indication through PDCCH / PDSCH. For the reception mechanism, the device listens for MsgB within the configured time window (msgB-ResponseWindow-r16) and decodes it using the dedicated RNTI. If the contention resolution is successful, the device completes the access; if it fails or does not receive MsgB, it triggers retransmission or reverts to the four-step process.

[0075] 5. Select a channel according to the model output to perform unscheduled uplink transmission. The channel access step needs to be triggered in both states in step (4), but the non-active state greatly simplifies the process through context retention and protocol optimization, while the idle state needs to execute a complete access process. Then, according to the model that updates the parameters based on the data packet, the active device selects a subchannel to perform unscheduled uplink transmission.

[0076] 4a. Collect historical location data and data rate measurements: Select the location information (raw-level data) of each device that has successfully established a connection. If the user distribution and active mode change, this data is not used in the dataset.

[0077] 4b. (Optional) Scenario adaptation: If there is data such as device location and environmental parameters at the raw level that need to be updated or corrected, regenerate the dataset for training to update the model parameters.

[0078] 4c. Learn the active mode, channel quality, and interference level: Use the training dataset generated from the raw-level data as input, and through unsupervised learning and deep neural networks, initially obtain the mapping function between the active user location and the optimal subchannel selection probability distribution.

[0079] 4d. (Optional) Parameter Optimization: According to time-varying scenario information, such as characteristic data like device density, set the corresponding input dimensions and weight parameters of the custom loss function, and then perform model training.

[0080] 4e. Learn the mapping function between device location and channel access: Use the shared parameter model optimized with feature-level data for training at the base station, while for decentralized execution, deploy the trained independent models on multiple devices for parallel execution.

[0081] 4f. (Optional) Generalized Learning: If the dimensions of the features and data vary greatly, further adjust the model construction as needed, such as parameters like the number of layers and neurons.

[0082] 4g. Select a channel according to the probability distribution of the output channel selection. The data transmitted uplink is ultimately used for tasks such as environmental monitoring and vehicle-road collaboration.

[0083] 6. Determine whether data arrives, pre-configured resources are available, and the uplink synchronization status meets the requirements. If the access conditions are triggered, such as uplink data needs to be sent, the device is in an idle state or non-active state, the data packet is small, and there are available pre-configured grant-free resources, then the device uses the selected sub-channel for uplink transmission; otherwise, temporarily release the connection and wait for an indication.

[0084] From the above description, it can be seen that the above embodiments of the present invention achieve the following technical effects:

[0085] Through centralized training, the base station can learn the historical data of channel states in complex environments and the spatial characteristics of device distributions, establish a global channel selection prediction model, thereby reducing the overhead of control signaling interaction. For devices that need repositioning, only need to input the updated coordinates again to obtain the model output, significantly reducing the computing energy consumption at the device side.

[0086] In the decentralized execution stage, active devices only quickly select the optimal channel for grant-free access based on the locally sensed location information and the policy output by the model, which is beneficial to ensuring the continuity of road monitoring. It adapts to large-scale deployed road traffic scenarios, while reducing deployment costs and privacy risks. In addition, grant-free access reduces the continuous listening energy consumption of devices during the competition stage.

[0087] Through in-depth mining of location data by the neural network model, accident prevention, emergency response, and traffic efficiency improvement can be achieved without violating privacy, building a safe and sustainable traffic ecosystem for smart cities. In particular, it has core advantages for road safety applications, such as scalability and low-cost deployment, real-time traffic situation perception and accident warning, dynamic traffic flow optimization and emergency response optimization, etc.

[0088] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for a machine-type communication device to access a channel based on location, characterized in that Including: Collecting historical data of location and communication parameters from roadside devices by the base station, namely raw data or feature data, and learning the channel selection for a period of time or with periodic characteristics, so as to obtain the mapping function between location information and approximate optimal channel access, and maximize the sum rate of the uplink access system; Based on the location-based channel access strategy, a centralized training and decentralized execution framework driven by a deep neural network is adopted to achieve intelligent channel prediction and unauthorized fast access; For active devices with unauthorized transmission, improve the access success probability, reduce communication overhead and power consumption; When the connection density range of fixed devices is fixed, improve the convergence speed of unsupervised learning and the computational overhead of the learning model, and then achieve efficient deployment on resource-constrained devices and improve the generalization of the model.

2. The method for channel access of the location-based machine-type communication device according to claim 1, wherein The centralized training and decentralized execution framework includes: Step 1: Determine the RRC state of the NR system where the device is located; Step 2: Receive the data packet containing model parameters transmitted from the base station; Step 3: Record or update the location information of the current device; Step 4: Trigger the access process to establish a connection or optimize the process to resume the connection; Step 5: Select a channel according to the model output to perform unscheduled uplink transmission; Step 6: If the data arrival, pre-configured resource availability, and uplink synchronization status meet the requirements, go to Step 3. If the data arrival, pre-configured resources, and uplink synchronization status do not meet the requirements, go to Step 1.

3. The method for channel access of the location-based machine-type communication device according to claim 2, wherein, In Step 1, according to historical signaling interaction, system message parsing, context management, and timer control, the device senses and determines the current RRC state, as well as the availability of uplink synchronization status and pre-configured resources. If the device is in the idle state and the synchronization is invalid, the device needs to first trigger contention-based random access to obtain an effective timing advance (TA) value; If the device is in the inactive state, uplink synchronization can be maintained through periodic TA updates.

4. The method for channel access of the location-based machine-type communication device according to claim 2, wherein Step 2 includes: The device sends an access request. When an idle-state device initially establishes a connection, it randomly selects a specific physical random access channel (PRACH) resource group.

5. The method for a location-based machine-type communication device to access a channel according to claim 2, wherein Step 2 includes: The base station sends a recovery confirmation message. The nearby base station directly responds to contention resolution and uplink resource allocation. Based on the collected raw-level data and channel knowledge map, the base station trains the neural network model in the access training module with the goal of maximizing the system sum rate until the convergence condition is met.

6. The method for a location-based machine-type communication device to access a channel according to claim 2, characterized in that Step 2 includes: The base station transmits model parameters. Combining with the device status selection, the base station dynamically selects the transmission mechanism and distributes the model. That is, after completing the training of the model with raw-level data, the base station transmits the data packet of model parameters containing neural network weights to a large number of devices through the downlink shared channel (PDSCH), supplemented by physical downlink control channel (PDCCH) scheduling and reference signal (RS) optimization. For devices with a low update frequency, lightweight model parameters are broadcast through the physical broadcast channel (PBCH) to update the execution network.

7. The method for a location-based machine-type communication device to access a channel according to claim 2, wherein The algorithm framework of the centralized training includes: First, the NR channel bandwidth is divided into multiple orthogonal sub-channels by using Orthogonal Frequency Division Multiplexing (OFDM) technology, where each sub-channel includes a sub-carrier, and at most one active device is allocated within the same time slot; Second, using the historical data of the wireless propagation environment and the uplink transmission system in the urban scenario, preprocess the original-level data to obtain an encapsulated data set, then call the encapsulated data set to train the configuration to find the optimal value of the model, and finally save the determined model parameters; Third, transform the integer programming problem of the channel problem into a sum rate maximization problem, that is, maximize the sum of the achievable rates of the system under the given OFDM bandwidth budget, sub-channel allocation number constraint, and minimum achievable data rate requirement; Finally, after the centralized training is completed, save the sharded model parameters for the decentralized execution of the devices.

8. The method for a location-based machine-type communication device to access a channel according to claim 2, wherein The said step 3 includes: Read location. After the active device with no adjustment in the deployment location reads the location information from the memory, it performs license-free access based on the signal quality using the two-step method or the four-step method; Update location. For the active device with displacement alarm caused by event-driven or dynamic calibration, the device first triggers repositioning and obtains the updated location information.

9. The method for a location-based machine-type communication device to access a channel according to claim 2, wherein The said step 4: includes that when the active device initiates access from the idle state, it needs to complete the initial access through the complete random access process; when the device is in the non-active state, it can directly send a resume request to the base station through the two-step RRC Resume Request.

10. The method for accessing a channel of a location-based machine-type communication device according to claim 2, wherein, The said step 5 includes: Collect historical location data and data rate measurements; Learn the active mode, channel quality, and interference level; Learn the mapping function between the device location and channel access; Conduct generalization learning; Select a channel according to the probability distribution of the output channel selection; The uplink transmission data is used for environmental monitoring and vehicle-road collaboration.

Images