An IoT data transmission method based on edge computing
By extracting semantic features and calculating comprehensive transmission priority on the IoT terminal device side, and combining dynamic time slot preemption and multi-agent deep reinforcement learning of the edge computing gateway, the problem of data semantic value not being reflected and the difficulty in dealing with time-varying channel characteristics in the existing technology is solved, realizing deterministic low-latency transmission of high-value data and system resource optimization.
Patent Information
- Application Number
- CN202611340522.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-09-01
- Publication Date
- 2026-11-17
AI Technical Summary
Existing IoT data transmission solutions fail to reflect the semantic value of the data content itself, and fixed-period scheduling mechanisms are unable to cope with the drastic time-varying characteristics of wireless channels, making it difficult for high-value, high-urgency data to obtain deterministic low-latency transmission guarantees at the edge computing node side.
By extracting semantic features and calculating comprehensive transmission priority on the IoT terminal device side, and combining dynamic time slot preemption and multi-agent deep reinforcement learning of the edge computing gateway, the semantic resource joint scheduling of data packets is realized, and the transmission priority and channel resource allocation are dynamically adjusted.
It enables dynamic adjustment of transmission priority based on the semantic value of data, reduces the occupation of invalid and redundant data, alleviates uplink congestion, ensures that high-value data arrives within a preset time delay, and improves the efficiency of system resource utilization.
Smart Images

Figure CN123123475A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of edge computing Internet of Things (IoT) data transmission technology, and more specifically to an IoT data transmission method based on edge computing. Background Technology
[0002] With the rapid development of applications such as Industrial Internet of Things (IIoT), autonomous driving, and smart cities, massive numbers of IoT terminal devices need to transmit the collected data to edge computing nodes in real time for low-latency processing. Currently, IoT data transmission is usually based on traditional protocol stacks, and the media access control layer generally uses carrier sense multiple access (CMA) collision avoidance mechanism to allocate channel resources. To meet the requirements of deterministic latency, the industry has also proposed Time-Sensitive Networking (TSN) and Deterministic Networking (DSN) technologies. The core idea is to allocate fixed-period transmission time slots to data streams of different priorities through a pre-planned fixed gating schedule and network-wide clock synchronization.
[0003] The closest prior art to this invention is an edge computing congestion control method based on priority queues. In this method, IoT terminal devices mark data packets according to pre-configured static priority labels. The data packets are sent to the edge computing gateway through a contention access method. The edge computing gateway maintains multiple fixed-length first-in-first-out queues corresponding to the priority labels and takes data packets from each queue in a strict priority manner for forwarding. If a priority queue overflows, the data packet with the longest queuing time in that queue is discarded.
[0004] The aforementioned existing technologies use static priority tags configured only once, without considering the dynamic changes in the value of the data content itself. This results in high-value information possibly failing to reach edge nodes in a timely manner because it belongs to a low-priority device. Furthermore, the method does not perform semantic differentiation and compression of the data packet content, and the fixed-period scheduling mechanism is difficult to cope with the drastic time-varying characteristics of wireless channels. When the queue overflows, the method decides to discard objects based solely on the queuing time rather than the semantic value of the data, making it difficult to ensure the deterministic transmission delay of critical data and the quality of system decision-making. Summary of the Invention
[0005] The technical problem this invention aims to solve is that in wireless edge IoT network environments with massive concurrent access, time-varying channels, and unpredictable packet sizes and arrival times, existing transmission schemes rely on statically configured priority tags and packet byte sizes for transmission queuing and resource allocation. This fails to reflect the semantic value of the data content itself, and the fixed-period scheduling mechanism is unable to cope with the drastic time-varying characteristics of wireless channels, making it difficult for high-value, high-urgency data to obtain deterministic low-latency transmission guarantees at the edge computing node side.
[0006] To address the aforementioned technical problems, this invention provides an IoT data transmission method based on edge computing, comprising: an IoT terminal device collecting raw sensing data and inputting the raw sensing data into a pre-trained semantic feature extraction model to obtain a semantic feature vector;
[0007] Calculate the semantic entropy based on the probability distribution of the semantic feature vectors;
[0008] The urgency of the task is calculated based on the remaining transmission delay budget of the task to be transmitted and the proximity of the corresponding sensor readings to the preset safety threshold.
[0009] The overall transmission priority of the data packet is calculated based on the semantic entropy, the task urgency, and the energy factor obtained after normalizing the remaining battery power of the IoT terminal device.
[0010] The integrated transmission priority and semantic compression rate are filled into the extended header of the media access control frame and sent to the edge computing gateway along with the data packet;
[0011] The edge computing gateway parses the extended header and determines whether to trigger dynamic time slot preemption based on queue congestion. In congested conditions, it determines available time slots based on the semantic fidelity loss of low-priority data streams and reallocates these slots to data streams with higher priority. Furthermore, the edge computing gateway can collect channel state information and remaining battery power from each terminal device. By introducing an attention-based multi-agent deep reinforcement learning method, it models the joint scheduling problem of channel and semantic resources as a restricted Markov decision process and solves for the optimal scheduling strategy, dynamically distributing semantic compression rates and transmission licenses.
[0012] The technical effects that this invention can achieve include:
[0013] 1. Replacing traditional packet size and static priority labels with data semantic value as the basis for transmission queuing and resource allocation avoids the problem of mismatch between priority and data value;
[0014] 2. By extracting and compressing semantic features, the occupation of wireless bandwidth by invalid and redundant data is reduced, thus alleviating uplink congestion;
[0015] 3. By using a dynamic time slot preemption mechanism based on semantic fidelity loss assessment, the system can respond to severe channel fluctuations in real time, ensuring that high-value data arrives within the preset delay limit;
[0016] 4. By using multi-agent deep reinforcement learning based on attention mechanism to jointly optimize channel resources and terminal semantic compression strategy, the overall resource utilization efficiency of the system is improved. Attached Figure Description
[0017] Figure 1 This is a flowchart of the data processing of the IoT terminal device of the present invention;
[0018] Figure 2 This is a flowchart illustrating the dynamic time slot preemption and reshaping process of the edge computing gateway in this invention.
[0019] Figure 3 This is a flowchart of the channel and semantic joint optimization scheduling process of the present invention. Detailed Implementation
[0020] The following is in conjunction with the appendix Figure 1-3 The specific embodiments of the present invention will be further described below. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.
[0021] I. System Overall Architecture
[0022] The system provided in this embodiment of the invention mainly consists of a semantic extraction and compression module on the IoT terminal device side, a wireless access channel, a semantic-aware media access control and scheduling module and a distributed artificial intelligence joint optimization and scheduling module on the edge computing gateway side.
[0023] Specifically, the system includes several IoT terminal devices. Each terminal device contains a data acquisition unit, a lightweight semantic feature extraction unit, and a semantic entropy and task urgency calculation unit. The terminal devices are connected to the edge computing gateway through a wireless access channel. The wireless access channel is subject to channel fading and feeds back channel status information to the terminal devices. The edge computing gateway contains an extended frame parsing module from front to back.
[0024] The system comprises a semantic-aware media access control scheduling queue module, a dynamic time slot preemption and reshaping control unit, and a channel and semantic joint optimization scheduling unit based on multi-agent deep reinforcement learning. The semantic-aware media access control scheduling queue module forms a dynamic queue by sorting data packets according to their comprehensive transmission priority, unlike the traditional fixed-length first-in-first-out queue. The output of the edge computing gateway is connected to the back-end edge processing server and can also be connected to the upper-layer cloud center through a non-essential path. The channel and semantic joint optimization scheduling unit based on multi-agent deep reinforcement learning issues semantic compression rate and transmission permission control signals to each terminal device. Together with the data acquisition, feature extraction, and priority calculation of the terminal devices, it forms a closed-loop processing link in which perception, decision-making, and execution cooperate.
[0025] II. Semantic Feature Extraction and Semantic Entropy Calculation
[0026] Before transmitting data, IoT terminal devices first extract features from the raw sensed data using a built-in lightweight autoencoder model to obtain low-dimensional semantic feature vectors. These semantic feature vectors are then discretized and quantized, dividing them into... The algorithm identifies discrete symbol intervals and calculates the probability distribution of elements within each interval. This probability distribution is obtained by normalizing the hidden layer activation distribution output by the autoencoder. Based on this, the semantic entropy of the data packet is calculated according to the following formula:
[0027] ;
[0028] in, Indicates the first The semantic entropy of a data packet is used to quantify the value of the information carried by that data packet. The higher the semantic entropy value, the richer the effective information carried by the data packet. This represents the total number of discrete symbols that are divided after the semantic feature vector is quantized. Indicates the first In the semantic feature vector of the data packet, the first... The probability of a discrete symbol appearing.
[0029] III. Calculation of Task Urgency and Overall Transmission Priority
[0030] While calculating semantic entropy, IoT terminal devices also read the remaining transmission latency budget of the task to which the data packet belongs, as well as the proximity of the sensor reading corresponding to the data packet to a preset safety threshold, and calculate the task urgency according to the following formula:
[0031] ;
[0032] in, Indicates the first The urgency of the task to which each data packet belongs. This represents the remaining tolerable transmission delay budget for the data packet; the smaller the value, the more urgent the task is in terms of time. This indicates the proximity of the sensor reading corresponding to the data packet to a preset safety threshold. The value ranges from 0 to 1, with a larger value indicating that the sensor is closer to the danger threshold. and These are pre-defined weighting coefficients used to balance the contribution of time delay urgency and numerical proximity danger to the urgency of the task.
[0033] The terminal device further reads its current remaining battery power and performs normalization processing to obtain the energy factor. This energy factor is used to prevent low-power devices from running out of power too quickly due to frequent participation in high-priority contention. The terminal device calculates the overall transmission priority of the data packet based on the following formula, taking into account semantic entropy, task urgency, and energy factor:
[0034] ;
[0035] in, Indicates the first The final overall transmission priority of each data packet. , , Weight coefficients are pre-set or dynamically issued by the edge computing gateway, and satisfy the following conditions: The terminal device will calculate the overall transmission priority. The semantic compression rate it currently uses is filled into the semantic priority identifier field of the Media Access Control Frame Extension Header and sent to the edge computing gateway along with the data packet.
[0036] The complete process described above is as follows: The terminal device collects raw sensing data, inputs the raw data into a lightweight autoencoder to extract semantic feature vectors, performs discrete quantization on the semantic feature vectors and counts the probability of occurrence of each symbol, calculates semantic entropy according to the semantic entropy formula, reads the remaining delay budget and sensor values to calculate the task urgency, reads the remaining battery power to calculate the energy factor, calculates the comprehensive transmission priority according to the comprehensive priority formula, writes the priority and semantic compression rate into the extended header of the media access control frame and sends it to the edge computing gateway, forming a complete processing closed loop from data acquisition to transmission completion.
[0037] IV. Engineering Implementation of the Media Access Control Frame Extended Header
[0038] To ensure compatibility between the present invention and existing communication protocols, no modification to the physical layer baseband technology is required. Only a custom extended header field is added to the existing Media Access Control (MAC) protocol frame to carry semantic priority identification information. Specifically, the extended frame structure, from left to right, consists of the original MAC header field, the newly added semantic priority identification extended header field, the original MAC payload field, and the original frame check field. The formats of the original header field, payload field, and frame check field remain unchanged. The semantic priority identification extended header field is further subdivided into a comprehensive transmission priority subfield, a semantic compression rate subfield, and a remaining delay budget subfield. Each subfield is allocated according to a preset bit length. (This is part of the extended frame parsing module of the edge computing gateway.)
[0039] After receiving the data packet, the extended header is parsed to obtain the comprehensive transmission priority, semantic compression rate and remaining delay budget information, and queue rearrangement and time slot allocation are performed accordingly.
[0040] V. Dynamic Time Slot Preemption and Reshaping Algorithm for Edge Deterministic Networks
[0041] Extended frame parsing module for edge computing gateways;
[0042] After receiving data packets from each terminal device, the semantic priority identifier field in the extended header is parsed, and the parsed data packets are sent to the semantically aware media access control scheduling queue module according to the overall transmission priority. The dynamic time slot preemption and reshaping control unit continuously calculates the current queue congestion level according to the following formula:
[0043] ;
[0044] in, Indicates the current queue congestion level. This indicates the amount of data currently queued in the transmission queue. This indicates the maximum amount of data the queue can hold. Not exceeding the preset congestion threshold In this case, the control unit performs routine queuing and forwarding of data packets in the queue according to the overall transmission priority from high to low. Exceeding the preset congestion threshold In such cases, the current channel is determined to be in a congested state, triggering a dynamic time slot preemption mechanism.
[0045] The criterion for preempting time slots is the semantic fidelity loss assessment result. That is, before preempting a time slot occupied by a low-priority data stream, the degree of semantic fidelity degradation after the data stream is delayed is assessed, specifically calculated according to the following formula:
[0046] ;
[0047] in, Indicates the first A low-priority data stream was delayed. The amount of loss in post-semantic fidelity. This indicates the initial semantic fidelity of the data stream under no-latency conditions. The sensitivity coefficient for the semantic fidelity of this type of data stream decays over time, and is pre-calibrated by the timeliness type of the data stream.
[0048] The dynamic time slot preemption and reshaping control unit traverses all low-priority data streams that have been allocated time slots but have not yet completed transmission, calculates their semantic fidelity loss one by one, and sorts these data streams in ascending order of semantic fidelity loss. The communication time slots occupied by one or more low-priority data streams at the top of the sort are determined as preemptible time slots. The control unit then reallocates the determined preemptible time slots to the data streams with the highest overall transmission priority in the queuing queue and updates the time slot allocation table. The preempted low-priority data streams are put back to the end of the queue to wait for the next round of scheduling. The above process is continuously executed in a loop. The complete closed-loop scheduling process is as follows: receive data packets and parse the extended header, update the queue congestion level, determine whether the congestion threshold is exceeded, if not, forward the data according to the original scheduling order and return to the receiving step, if the threshold is exceeded, traverse the low-priority data streams to calculate the semantic fidelity loss, determine the preemptible time slots and reallocate them, update the time slot allocation table and forward the data streams, and return to the receiving step.
[0049] VI. Joint Optimization Scheduling of Channel and Semantics Based on Edge Distributed Artificial Intelligence
[0050] The edge computing gateway's channel and semantic joint optimization scheduling unit, based on multi-agent deep reinforcement learning, collects channel state information and remaining battery power of each IoT terminal device at a preset period to construct the current state vector of each terminal device. The state vector includes channel state information and remaining battery power.
[0051] This scheduling unit models the joint scheduling problem of channel and semantic resources as a restricted Markov decision process, and its optimization objective is expressed as:
[0052] ;
[0053] The above objectives are achieved by ensuring that the end-to-end transmission delay does not exceed a preset delay limit. Furthermore, the energy consumption of each terminal device does not exceed the preset energy consumption budget. Solve under the constraints, where, This represents the scheduling strategy to be solved. This represents the discount factor, used to balance the weights of current and future returns. Indicates the first The immediate reward obtained by the system at each decision moment is calculated according to the following formula:
[0054] ;
[0055] in, This indicates the total number of IoT terminal devices currently participating in the scheduling. Indicates the first The weighting coefficient of each terminal device's data stream is determined by its historical average comprehensive transmission priority. Indicates the first The decision moment The actual semantic fidelity achieved by each data stream. This represents the penalty coefficient for delay default. For indicator functions, when the first The actual latency of each data stream Exceeding the latency limit The value is 1 if the condition is met, and 0 otherwise.
[0056] To dynamically measure the importance of each terminal device's state information to the global scheduling decision, the scheduling unit introduces an attention mechanism module into the multi-agent network structure. The state vectors of each terminal device are input into this module, and the attention weight corresponding to each terminal device's state is calculated according to the following formula:
[0057] ;
[0058] in, Indicates the first The attention weight of each terminal device's state in the global decision-making process. Indicates the first Current status characteristics of each terminal device The original attention score is calculated by the learnable scoring network. The scheduling unit performs weighted fusion of each state vector according to the attention weight to obtain a global state representation that reflects the overall situation of the entire network. This global state representation is then input into the pre-trained policy network and value network. The policy network outputs the semantic compression rate and transmission permission action that each terminal device should adopt at the moment. The value network outputs the evaluation of the value of the current state.
[0059] The action output is applied to the actual IoT data transmission environment through the environmental interaction module, generating new feedback on latency, energy consumption, and semantic fidelity. This feedback is used by the reward calculation module to generate an immediate reward, and the experience sample consisting of the current state, action, reward, and the state at the next moment is stored in the experience replay buffer. The scheduling unit samples a batch of experience samples from the experience replay buffer and uses a near-end policy optimization method to update the parameters of the policy network and the value network respectively, so that the expected value of the cumulative discount reward is improved under the conditions of latency and energy consumption constraints, and then enters the next scheduling cycle.
[0060] VII. Alternative Implementation Methods
[0061] In the semantic feature extraction stage on the terminal side, as an alternative implementation method, instead of using an autoencoder model, a signal processing method based on wavelet transform or principal component analysis can be used to extract the frequency domain or principal component features of the data, and the information entropy of the feature energy distribution can be used to approximate the aforementioned semantic entropy, which is suitable for scenarios where the terminal has limited computing power and it is difficult to run neural network models.
[0062] In the joint scheduling of channel and semantics, as an alternative implementation method, instead of using multi-agent deep reinforcement learning, a drift plus penalty algorithm based on Lyapunov optimization can be used. In each scheduling cycle, the drift amount is the weighted sum of the queue backlog length and the semantic fidelity loss. By solving the time-slot optimization problem, the allocation scheme of semantic compression rate and transmission license is obtained. This alternative method does not require offline training and can be used in scenarios with high real-time requirements and a lack of historical training data.
[0063] Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make possible changes and modifications without departing from the spirit and scope of the present invention. Therefore, any modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall fall within the protection scope defined by the claims of the present invention.
Claims
1. An edge computing-based Internet of Things data transmission method, characterized in that, include: IoT terminal devices collect raw sensing data and input the raw sensing data into a pre-trained semantic feature extraction model to obtain semantic feature vectors; The semantic entropy is calculated based on the probability distribution of the semantic feature vector, and the semantic entropy is used to characterize the information value carried by the original perceptual data; The urgency of the task is calculated based on the remaining transmission delay budget of the task to be transmitted and the proximity of the corresponding sensor readings to the preset safety threshold. The overall transmission priority of the data packet is calculated based on the semantic entropy, the task urgency, and the energy factor obtained after normalizing the remaining battery power of the IoT terminal device. The integrated transmission priority and the semantic compression rate currently used by the IoT terminal device are filled into the semantic priority identifier field of the media access control frame extension header and sent to the edge computing gateway along with the data packet; The edge computing gateway receives the data packet, parses the semantic priority identifier field, and obtains the queue congestion level based on the ratio of the amount of data queued in the current transmission queue to the upper limit of the queue capacity. When the queue congestion exceeds a preset congestion threshold, for each of the multiple low-priority data streams that are currently occupying communication time slots and have not yet completed transmission, the semantic fidelity loss of each low-priority data stream after delayed transmission is calculated. The multiple low-priority data streams are sorted in ascending order of semantic fidelity loss. The communication time slots occupied by the top-ranked low-priority data streams are determined as preemptible time slots. These preemptible time slots are then reallocated to the data streams in the queuing queue that have the highest overall transmission priority, thereby achieving dynamic preemption and reshaping of time slots.
2. The method of claim 1, wherein, The semantic entropy is calculated according to the probability distribution of the semantic feature vector, including: performing discrete quantization processing on the semantic feature vector to obtain a plurality of discrete symbols, and counting the probability distribution of the occurrence of each discrete symbol, and calculating the semantic entropy according to the following formula: ; wherein, denotes the semantic entropy of the th data packet, denotes the total number of discrete symbols, denotes the probability of the th discrete symbol occurring in the semantic feature vector of the th data packet.
3. The method of claim 1, wherein, The urgency of the task is calculated according to the following formula: ; wherein, represents the task urgency of the task to which the represents the remaining transmission latency budget, represents the proximity, with a value range of 0 to 1, and is a pre-set weight coefficient. 4. The method of claim 1, wherein, The overall transmission priority is calculated according to the following formula: ; wherein, denotes the integrated transmission priority of the , , is a weight coefficient dynamically issued by the edge computing gateway, and satisfies . 5. The method of claim 1, wherein, The semantic fidelity loss is calculated according to the following formula: ; wherein, represents the semantic fidelity loss of the first low priority data stream after being delayed represents the initial semantic fidelity of the first low priority data stream in the case of no delay, is a sensitive coefficient of semantic fidelity decay over time. 6. The method of claim 1, wherein, Also includes: The edge computing gateway collects the channel status information and remaining battery power of each IoT terminal device and constructs a state vector corresponding to each IoT terminal device. Each state vector is input into the attention mechanism module, the attention weight corresponding to each state vector is calculated, and the state vectors are weighted and fused according to the attention weights to obtain the global state representation. The global state representation is input into a pre-trained policy network to obtain the semantic compression rate and transmission permission corresponding to each IoT terminal device, and then distributed to the corresponding IoT terminal device.
7. The method of claim 6, wherein, The policy network is pre-trained using a reinforcement learning method based on a Restricted Markov Decision Process (RPM). The RPM is constrained by the end-to-end transmission latency not exceeding a preset latency limit and the terminal energy consumption not exceeding a preset energy consumption budget. The optimization objective is to increase the expected cumulative discount reward. The cumulative discount reward is calculated based on the semantic fidelity and latency violation penalty of each IoT terminal device.
8. The method of claim 1, wherein, The Media Access Control Frame Extended Header is positioned between the original Media Access Control Frame Header field and the original Media Access Control Frame Payload field, and includes a Comprehensive Transmission Priority subfield, a Semantic Compression Rate subfield, and a Remaining Delay Budget subfield.
9. An edge computing based Internet of Things data transmission system, characterized in that, This includes IoT terminal devices and edge computing gateways; The IoT terminal device includes a data acquisition unit, a semantic feature extraction unit, and a semantic priority calculation unit. The data acquisition unit is used to acquire raw sensing data. The semantic feature extraction unit is used to extract features from the raw sensing data to obtain semantic feature vectors and calculate semantic entropy based on the probability distribution of the semantic feature vectors. The semantic priority calculation unit is used to calculate a comprehensive transmission priority based on the semantic entropy, task urgency, and energy factor, and fill the comprehensive transmission priority and semantic compression rate into the extended header of the media access control frame before sending it. The edge computing gateway includes an extended frame parsing module, a semantic-aware scheduling queue module, and a dynamic time slot preemption and reshaping control unit. The extended frame parsing module is used to parse the extended header of the media access control frame to obtain the comprehensive transmission priority and the semantic compression rate. The semantic-aware scheduling queue module is used to queue data packets according to the comprehensive transmission priority. The dynamic time slot preemption and reshaping control unit is used to determine the preemptible time slots based on the semantic fidelity loss of low-priority data streams when the queue congestion exceeds a preset congestion threshold, and to reallocate the preemptible time slots to the data streams with the highest comprehensive transmission priority in the queuing queue.
10. The system of claim 9, wherein, The edge computing gateway also includes a channel and semantic joint optimization scheduling unit based on multi-agent deep reinforcement learning. The channel and semantic joint optimization scheduling unit is used to collect the channel state information and remaining battery power of each IoT terminal device, calculate the attention weights corresponding to each IoT terminal device through the attention mechanism module and perform weighted fusion to obtain a global state representation, and generate and issue the semantic compression rate and transmission license corresponding to each IoT terminal device based on the global state representation.
