Communication method and electronic equipment

Abstract

The invention discloses a communication method and electronic equipment, and the method comprises the steps: analyzing the service quality demands of data streams according to the application scene and equipment characteristics of the Internet of Things; transmission characteristic parameters of available channels are obtained, a channel characteristic matrix is formed, and the data stream service quality attribute vector is used as input to be matched with the channel characteristic matrix; quantifying the matching degree of the data stream and the available channel, and generating an adaptation degree matrix; obtaining a mapping relation from the data stream to the available channel; according to the attribute of the environment where the Internet of Things equipment is located, collecting signal parameters of available channels, adjusting a channel allocation strategy in combination with the mapping relation, and selecting a target channel for data transmission; indexes of available channels are collected, statistical indexes are analyzed, and the performance of channel shunting is quantified; and optimizing the weight and the greedy algorithm of each attribute in the weighted Euclidean distance adaptation degree function, and performing communication according to the optimized weighted Euclidean distance adaptation degree function and the greedy algorithm.

Description

Technical Field

[0001] This application belongs to the field of communications, specifically relating to a communication method and an electronic device. Background Technology

[0002] In IoT multi-channel communication scenarios, different types of data streams have varying requirements for performance indicators such as latency, reliability, and power consumption during transmission. Real-time audio and video streams require low latency and high reliability, with relatively lower power consumption requirements; low-power sensor data streams are extremely sensitive to power consumption, with relatively lower requirements for latency and reliability; high-priority control commands have extremely high requirements for latency and reliability, but are not sensitive to power consumption. Therefore, in order to transmit these differentiated data streams on the same channel, the sudden large bandwidth demands of real-time audio and video streams can lead to channel congestion, causing increased latency and packet loss rates for other data streams. Summary of the Invention

[0003] The purpose of this application is to provide a communication method and electronic device that can solve the problems of channel congestion, which leads to increased latency and packet loss rate in the transmission of other data streams.

[0004] To solve the above-mentioned technical problems, this application is implemented as follows: In a first aspect, embodiments of this application provide a communication method, the method comprising: analyzing the quality of service (QoS) requirements of different data streams based on IoT application scenarios and device characteristics; acquiring transmission characteristic parameters of multiple available channels to form a channel characteristic matrix; using a data stream QoS attribute vector as input and calculating the matching degree with the channel characteristic matrix; quantifying the matching degree between the data stream and the available channels by constructing a weighted Euclidean distance fit function to generate a fit matrix; solving the fit matrix using a greedy algorithm to obtain the mapping relationship between the data stream and the available channels; collecting signal parameters of each available channel based on the attributes of the environment in which the IoT device is located, adjusting the channel allocation strategy in conjunction with the mapping relationship, and selecting a target channel for data transmission; collecting data transmission statistical indicators of each available channel, analyzing the statistical indicators using a weighted moving average algorithm to quantify the performance of channel splitting; optimizing the weights of each attribute in the weighted Euclidean distance fit function and the greedy algorithm based on the performance of channel splitting, and performing communication based on the optimized weighted Euclidean distance fit function and the greedy algorithm.

[0005] In a second aspect, embodiments of this application provide an electronic device including a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the method described in the first aspect.

[0006] Thirdly, embodiments of this application provide a readable storage medium on which a program or instructions are stored, which, when executed by a processor, implement the steps of the method described in the first aspect.

[0007] Fourthly, embodiments of this application provide a chip, the chip including a processor and a communication interface, the communication interface being coupled to the processor, the processor being used to run programs or instructions to implement the method as described in the first aspect.

[0008] In the embodiments of this application, the problems of channel congestion leading to increased latency and packet loss rate in other data stream transmissions can be solved. Attached Figure Description

[0009] Figure 1 This is a flowchart illustrating a communication method provided in an embodiment of this application; Figure 2 This is a schematic flowchart of a communication device provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of another electronic device provided in an embodiment of this application. Detailed Implementation

[0010] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such use of data can be interchanged where appropriate so that embodiments of this application can be implemented in orders other than those illustrated or described herein. Furthermore, in the specification and claims, "and / or" indicates at least one of the connected objects, and the character " / " generally indicates that the preceding and following objects are in an "or" relationship.

[0012] The embodiments of this application will be described in detail below with reference to the accompanying drawings and specific examples and application scenarios.

[0013] Figure 1 This diagram illustrates a flowchart of the query method provided in an embodiment of this application. Figure 1 As shown, the method includes: S101: Analyze the service quality requirements of different data streams based on IoT application scenarios and device characteristics.

[0014] This step may include: analyzing the latency sensitivity, reliability requirements, security requirements, and power consumption budget service quality attributes of different data streams based on IoT application scenarios and device characteristics, and obtaining the service quality attribute vector for each data stream; based on the service quality attribute vector of the data stream, combined with specific business requirements and network environment, determining the transmission protocol and data processing method applicable to each data stream to meet the service quality requirements.

[0015] Based on IoT application scenarios and device characteristics, this study analyzes the latency sensitivity, reliability requirements, security requirements, and power budget service quality attributes of different data streams, obtaining service quality attribute vectors for each data stream. Based on these service quality attribute vectors, and considering specific business needs and network environment, suitable transmission protocols and data processing methods are determined to meet the service quality requirements of each data stream. For data streams with high latency sensitivity, the MQTT low-latency transmission protocol is adopted, and dedicated transmission resources, including dedicated queues or virtual links, are reserved in the network through software-defined networking. Edge computing is employed during data processing, pushing data processing tasks down to the network edge closer to the data source to reduce data transmission latency. Furthermore, data processing algorithms are optimized, including incremental learning and online learning methods, to further reduce processing latency. For data streams with high reliability requirements, the MQTT-SN reliable transmission protocol is adopted, improving transmission reliability through multi-path transmission and data redundancy. Error correction mechanisms, such as forward error correction, are used at the receiving end to recover damaged data. Multipath transmission can centrally coordinate multiple transmission paths through SDN and dynamically schedule traffic based on path status to ensure reliability. FEC introduces redundant data at the sending end and uses it to recover from errors at the receiving end, reducing retransmission overhead. For data streams with high security requirements, the secure and reliable AES encryption algorithm is used to encrypt the data, and the TLS security protocol is employed during transmission to prevent data theft or tampering. A two-way authentication mechanism is also used to authenticate devices and users to prevent unauthorized access. For devices with limited power budgets, the low-power CoAP transmission protocol and data processing algorithm are prioritized. The device's operating state is dynamically adjusted through device operating mode scheduling and idle time prediction power management mechanisms to reduce energy consumption while meeting service quality requirements. In the data processing, semantic analysis, including knowledge graphs and ontology reasoning, is first used to extract semantic information from multi-source heterogeneous data, and the data is then cleaned and labeled accordingly. Next, based on specific business logic, data fusion algorithms such as Kalman filtering and Bayesian inference are used to integrate information from multiple data sources to form a consistent, high-quality dataset. Finally, according to the needs of upper-layer applications, the fused data is sliced, compressed, and preprocessed, and then assembled and distributed through message queues and publish-subscribe mechanisms to provide timely and accurate services to relevant business applications.

[0016] For example, when analyzing the quality of service (QoS) attributes of IoT data streams, the Analytic Hierarchy Process (AHP) can be used to perform pairwise comparisons and weight calculations for each attribute. For instance, in a smart city application, expert scoring determines the importance ratio of latency sensitivity, reliability, security, and power budget to be 5:4:4:3. AHP then calculates the weight vector for each attribute as (0.3125, 0.25, 0.25, 0.1875), which can be used to quantitatively assess the QoS requirements of different data streams. Regarding transmission protocol selection, a mapping model between network state and protocol performance can be established based on Markov Decision Processes (MDPs) to dynamically select the optimal protocol for different data streams. For latency-sensitive video surveillance data, when network latency increases from 50ms to 100ms, the video frame rate needs to be reduced from 25FPS to 18FPS to meet real-time requirements. In this case, switching to the MQTT protocol can be used; when the network returns to normal, switching back to the HTTP protocol improves transmission efficiency. For algorithm optimization of latency-sensitive data, incremental learning algorithms, such as incremental SVM and online stochastic gradient descent, can be considered. Incremental training is performed after receiving a certain amount of new data (e.g., 1000 data points), and the average latency of 10 consecutive incremental training iterations is controlled within 200ms, while traditional batch training takes more than 2 seconds. In the design of multipath transmission schemes, Yen's Algorithm can be used to dynamically select the top K optimal paths (e.g., K=3), and combined with the leaky bucket algorithm to achieve load balancing between paths, ensuring that the utilization difference between paths does not exceed 10%. At the receiving end, FEC mechanisms such as Reed-Solomon codes can be used to reduce the bit error rate from 10⁻⁴ to below 10⁻⁶. For data with high security requirements, high-strength encryption algorithms such as AES-256 and ECC-384 can be used, combined with secure hash functions such as SHA-3 for integrity verification, controlling the proportion of tampered data to below 0.01%. In terms of power consumption management, reinforcement learning algorithms such as Q-Learning can be used to schedule the device's operating mode (e.g., active, sleep, low power) based on the device's power consumption, workload, network environment, and other state characteristics. This minimizes energy consumption while meeting service quality requirements, and simulations have shown that it can save more than 15% energy compared to traditional methods. For semantic extraction and fusion of IoT data, word embedding models such as Word2Vec can be used to pre-train IoT corpora to obtain more targeted semantic representations. Then, a knowledge graph for IoT can be constructed, and implicit semantic information can be obtained through ontology reasoning to automatically label the data. During fusion processing, multi-sensor information fusion algorithms such as covariance intersection (CI) and information distance weighted average (IWAD) can be used to associate data from different sources and modalities, eliminating redundancy and inconsistency.Finally, based on the requirements of real-time and concurrency in data association, high-performance message queues such as Kafka and ZeroMQ are flexibly used for streaming processing and distribution, so that the end-to-end latency is controlled within the 10ms range.

[0017] S102: Obtain the transmission characteristic parameters of multiple available channels to form a channel characteristic matrix. Take the data flow service quality attribute vector as input and calculate the matching degree with the channel characteristic matrix.

[0018] This step may include: obtaining the instantaneous signal-to-noise ratio and coherent bandwidth physical layer parameters of currently available channels through channel detection; estimating the maximum transmission rate, average latency, and packet loss rate of each channel; obtaining the power consumption characteristics of the channels through power detection to form a channel characteristic matrix reflecting the real-time transmission characteristics of each channel; quantifying the service quality attribute vector of the data stream into specific indicators of transmission rate, latency, reliability, and energy efficiency, and assigning different weights to them; and performing weighted Euclidean distance calculation between the channel characteristic matrix and the specific indicators of transmission rate, latency, reliability, and energy efficiency, and assigning different weights to them, to obtain the matching degree between each channel and the service quality requirements of the data stream.

[0019] Channel sounding is used to obtain the instantaneous signal-to-noise ratio and coherent bandwidth physical layer parameters of currently available channels. Based on Shannon's formula and queuing theory models, the maximum transmission rate, average delay, and packet loss rate of each channel are estimated. Simultaneously, power detection is used to obtain the power consumption characteristics of the channels, ultimately forming a channel characteristic matrix reflecting the real-time transmission characteristics of each channel. The service quality attribute vector of the data stream is quantified into specific indicators of transmission rate, delay, reliability, and energy efficiency, and assigned different weights. Weighted Euclidean distance is then calculated between these indicators and the channel characteristic matrix to obtain the matching degree between each channel and the data stream's service quality requirements. Based on the matching degree calculation results, fuzzy comprehensive evaluation and TOPSIS multi-attribute decision-making methods are used to comprehensively evaluate and rank each channel, obtaining the optimal channel with the highest matching degree to the data stream's service quality requirements. When selecting channels, not only the matching degree of a single channel is considered, but the feasibility of multi-channel parallel transmission or multi-hop relay transmission is also evaluated. The Kuhn-Munkres algorithm is used to construct the optimal channel combination, forming virtual channels that adapt to the data stream's service quality requirements. Stream segmentation and data replication mechanisms are used to achieve multi-channel collaborative transmission. For latency-sensitive data streams, channels or channel combinations that meet their latency limits are selected. For reliability-sensitive data streams, channels with packet loss rates below their tolerance thresholds are chosen, and SR-ARQ or HARQ reliable transmission is employed when necessary, using concurrent multiple access and spatiotemporal diversity to achieve diversity gain. After determining the optimal transmission channel, the transmission strategy is dynamically optimized based on channel characteristics. This includes adjusting transmit power using the Goldsmith-Varaiya power allocation algorithm, matching instantaneous channel states with adaptive channel coding and modulation schemes, or dynamically adjusting retransmission strategies and parameters based on channel prediction. Simultaneously, Turbo codes, LDPC codes for channel error correction, Alamouti spatiotemporal diversity codes for MIMO, and OFDM and NOMA multiple access are comprehensively applied to ensure transmission performance under adverse effects such as channel fading and interference. When multiple data streams are transmitted concurrently, proportional fairness and max-min fairness service-aware resource allocation algorithms are used to improve channel utilization and total system throughput while meeting the quality of service requirements of critical data streams, thus providing services to the data streams.

[0020] For example, when acquiring channel transmission characteristics, the channel impulse response can be estimated by sending pilot sequences, and the channel gain curve can be fitted using the least squares method. When the signal-to-noise ratio is 20dB, the estimation error is less than 1dB. Simultaneously, the channel service process is probabilistically modeled using LargeDeviations theory to obtain the probability distribution function of queuing delay. When the packet arrival rate is 0.5Mbps and the average service rate is 1Mbps, the error does not exceed 5%. Then, the Friis free-space propagation model is used to calculate the relationship between distance and channel attenuation, and energy consumption under different transmit powers is quantitatively analyzed. When the power range is 100mW to 1W, the relative error of energy consumption estimation is controlled within 10%. When calculating the data stream service quality matching degree, the various indicators are first normalized. Positive indicators such as transmission rate, reliability, and energy efficiency are linearly transformed, while negative indicators such as delay are inversely quantified. Then, the weight vector is determined using the analytic hierarchy process (AHP), and the comprehensive weighted matching degree is calculated. In channel evaluation and selection, a fuzzy comprehensive evaluation model based on Mamdani inference can be constructed. Membership functions for channel characteristics and data flow requirements are defined, a fuzzy relation matrix is ​​established, and a comprehensive score for each channel is obtained through fuzzy synthesis operations. The channel with the highest score is selected as the optimal channel. The Kuhn-Munkres algorithm is used to solve the bipartite graph matching problem between data flow and channel, finding the optimal match that maximizes system throughput. With 5 data flows and 3 channels, the algorithm complexity is O(n^3), and results can be obtained within 10ms. Regarding dynamic optimization of transmission strategies, Kalman filtering is used to predict the SINR variation trend of the channel. Based on the predicted value, the Goldsmith-Varaiya algorithm is used to dynamically allocate transmit power, maximizing the transmission rate while ensuring the target bit error rate. This achieves a rate gain of over 20% compared to fixed power allocation. In terms of coding and modulation optimization, modulation schemes of different orders, such as BPSK and QPSK, are selected based on the channel SINR, and channel codes with different error correction capabilities, such as convolutional codes and Turbo codes, are matched. Adaptive switching achieves a trade-off between coding gain and rate. At a SINR of 5dB, Turbo codes offer a 3dB higher coding gain than convolutional codes. In MIMO transmission optimization, Alamouti codes are used for transmit diversity, and at the receiver, maximum ratio combining or minimum mean square error criteria are used for diversity combining. In a 2-transmit, 2-receive, Gaussian channel, the diversity order of Alamouti codes is 2, and the coding gain can reach 10dB. In OFDM transmission, the maximization of system capacity criterion is used for dynamic allocation of resources such as subcarriers and power. The Lagrange multiplier method transforms the optimization objective into a dual problem, achieving a spectral efficiency improvement of over 30% compared to static OFDMA.In NOMA transmission, different power allocation factors are used for different services. During SIC reception, high-priority services are prioritized. With 2 users and a spectral efficiency below 1 bit / Hz, throughput more than 30% higher than OMA can be achieved. In concurrent multi-data-stream transmission, the MaxWeight scheduling algorithm can allocate channel resources based on the principle of proportional fairness. Specifically, it solves a linear programming problem, iteratively updating weights on a time-slot scale to maximize the system service rate region. The algorithm complexity is O(n), suitable for real-time scheduling on the order of 1ms.

[0021] S103: By constructing a weighted Euclidean distance fit function, the matching degree between the data stream and the available channels is quantified, and a fit matrix is ​​generated.

[0022] This step may include: normalizing the service quality attribute vector and channel characteristic matrix of the data stream, mapping each dimension index to the [0, 1] interval; using the analytic hierarchy process to determine the weight of each index in the service quality attributes, and obtaining the weight vectors of latency, bit error rate, and power consumption indexes by calculating and verifying consistency through pairwise comparisons, forming a weighted Euclidean distance metric; and obtaining the fit value between the data stream and each channel through matrix operations to generate the fit matrix.

[0023] By constructing a weighted Euclidean distance fit function, the matching degree between the data stream and the channel is quantified, and a fit matrix is ​​generated. If the data stream is real-time audio / video latency-sensitive, it is preferentially mapped to a channel with lower latency. If the data stream is sensor data low-power type, it is preferentially mapped to a channel with lower power consumption. If the data stream is control command reliability-sensitive, it is preferentially mapped to a channel with strong anti-interference capability and low bit error rate. When constructing the weighted Euclidean distance fit function, the service quality attribute vector and channel characteristic matrix of the data stream are first normalized, mapping each dimension index to the [0, 1] interval to eliminate the influence of dimensions. Then, the analytic hierarchy process (AHP) is used to determine the weights of each index in the service quality attributes. Through pairwise comparisons and consistency checks, the weight vectors of latency, bit error rate, and power consumption are obtained, forming a weighted Euclidean distance metric. Finally, matrix operations are used to obtain the fit value between the data stream and each channel, generating a fit matrix. After generating the fit matrix, different mapping strategies are used to allocate the data stream to the most suitable channel based on its service attributes. For latency-sensitive data streams, the channel with the smallest latency dimension fit (below a given threshold) in the fit matrix is ​​found, and the data stream is mapped to this channel. If multiple channels meet the conditions, the channel with the smallest capacity dimension fit is selected to balance the load of different channels. For low-power data streams, a similar approach is used, prioritizing power consumption adaptability. When multiple channels have power consumption adaptability below a threshold, the channel with the highest signal-to-noise ratio (SNR) adaptability is selected for transmission to achieve higher energy efficiency. For reliability-sensitive data streams, channels with bit error rate (BER) adaptability below a given threshold are considered first. If multiple channels meet the criteria, their adaptability in SNR and anti-interference capabilities is further compared, and the channel with the lowest overall adaptability is selected. In the actual dynamic channel allocation process, pilot sequences are periodically sent and channel impulse response changes are analyzed to monitor channel state changes in real time. Once a significant change in channel state is detected, the channel characteristic matrix is ​​updated and the adaptability matrix is ​​recalculated, promptly adjusting the channel mapping for affected data streams to ensure continuous quality of service. For composite data streams with multiple service attributes, the service quality attribute vector is first obtained through user preference queries and attribute sorting interactions. Then, the weight of each attribute is determined using the analytic hierarchy process (AHP). Based on this, the adaptability of each dimension is weighted and fused by setting dimensional weight vectors to generate a comprehensive adaptability between the composite data stream and the channel. The channel with the lowest comprehensive adaptability is selected for transmission. When no channel can independently meet the service quality requirements of the data stream, the data stream is split into multiple sub-streams and mapped in parallel to multiple channels with relatively high adaptability for simultaneous transmission. To achieve synchronization and reliable reassembly between sub-streams, interleaving coding is used to divide the original data slices into different sub-streams, and the receiving end uses the corresponding deinterleaving mechanism for reassembly. At the same time, a timing mark is introduced for each sub-stream packet, which the receiving end uses to check for packet loss or out-of-order delivery, and sends a retransmission request to the sending end if necessary. After the reliability-sensitive data stream is mapped to the channel, an adaptive fault-tolerant mechanism is further adopted to dynamically adjust the error correction coding and retransmission strategy according to the channel state.By tracking the statistical characteristics of channel signal-to-noise ratio and packet loss rate quality indicators, a performance threshold is set. When the channel quality is detected to be below the lower threshold, the redundancy of error correction coding is increased, and the number of retransmissions is appropriately increased. Conversely, when the channel quality improves and exceeds the upper threshold, the redundancy is reduced and the retransmission cycle is shortened, so as to adaptively balance reliability and transmission overhead.

[0024] For example, when performing hierarchical analysis on service quality attribute indicators, user or domain expert judgments on the importance of different indicators can be collected through questionnaires, expert interviews, etc., to construct pairwise comparison matrices. For instance, for data transmission in a smart home scenario, 10 users can be invited to score latency, reliability, and power consumption on a scale of 1-9. The average score is then used to obtain the judgment matrix. The eigenvector corresponding to the largest eigenvalue is calculated using the root-finding method and normalized to obtain a weight vector of (0.62, 0.28, 0.10). The consistency test shows a CI value less than 0.1. When generating the fit matrix, for a smart home data stream, its normalized service quality attribute vector is (0.9, 0.8, 0.2), and the normalized characteristic vector of a channel is (0.3, 0.1, 0.5). The weighted Euclidean distances between the data stream and the channel in the three dimensions of latency, reliability, and power consumption are 0.37, 0.20, and 0.03, respectively, and the fit value is 0.60. When the threshold value is set to 0.75, it can be mapped to the channel for transmission. During dynamic channel monitoring, a Gaussian white noise sequence can be sent as a pilot every 10ms. At the receiver, the least squares algorithm is used to estimate the channel impulse response. If the mean square error of the previous estimation result exceeds 0.3, it is determined that the channel state has changed significantly. For the calculation of the overall fit of composite data streams, the weights of each attribute can be determined first using the analytic hierarchy process (AHP), such as (0.5, 0.3, 0.2). Then, the overall fit value D=(d1^0.5×d2^0.3×d3^0.2)^(1 / (0.5+0.3+0.2)) is obtained by weighted geometric mean, where di is the fit of the i-th attribute. In multi-channel aggregation transmission, the data stream can be divided into N sub-streams using cyclic interleaving, with each sub-stream assigned a sequence number. The receiver reassembles the data streams according to the sequence numbers. Simultaneously, sub-stream packets are encapsulated in the format of (sequence number, data, CRC). The receiving end uses CRC check to determine if there are any erroneous packets and sets the timeout for the reassembly buffer to 1.5 times the channel RTT. Once an erroneous packet or timeout is detected, a NACK request for retransmission is sent to the sending end. In adaptive fault-tolerant control, the average signal-to-noise ratio (SNR) and packet loss rate of the channel's most recent 100 transmissions can be tracked, with upper and lower thresholds for SNR set at 15dB and 10dB respectively, and upper and lower thresholds for packet loss rate set at 10^-3 and 10^-5 respectively. Once performance indicators are detected to exceed the threshold range, the error correction coding rate is adjusted accordingly (e.g., switching between four code rates: 1 / 2, 2 / 3, 3 / 4, and 5 / 6) and the number of retransmissions (adjusted within the range of 1 to 4), and the current optimal configuration is recorded as the basis for subsequent adaptive adjustments.

[0025] S104: Use a greedy algorithm to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channels.

[0026] Based on the service quality attribute vector of the data stream and the characteristic matrix of the channel, a greedy algorithm is used to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channel.

[0027] A greedy algorithm is used to solve the fitness matrix, obtaining the optimal mapping relationship between data streams and channels, achieving balanced distribution of data streams across multiple channels. Based on the mapping relationship, different types of data streams are assigned to the most suitable channels for transmission. Based on the QoS attribute vector of the data stream and the characteristic matrix of the channel, a greedy algorithm is used to solve the fitness matrix to obtain the optimal mapping relationship between the data stream and the channel. The basic idea of ​​the greedy algorithm is to select the locally optimal solution in each step to achieve global optimization. First, the fitness matrix is ​​arranged in ascending order by row or column. Then, starting from the first row, the channel with the smallest fitness is selected sequentially until all data streams are allocated to the channel or all channels are occupied. During the solution process, a channel load balancing factor α is introduced. Channels with an allocated traffic ratio exceeding a certain threshold of the mean are penalized by a penalty coefficient, and their fitness value is multiplied by (1+α) to reduce the probability of this channel being selected later, thus achieving load balancing of the data stream on the channel. If the QoS attributes of the data stream are difficult to quantify with a single index, the analytic hierarchy process (AHP) can be used to determine the weight of each attribute to form a comprehensive fitness value. Specifically, a judgment matrix is ​​constructed by pairwise comparison and consistency check is performed. The eigenvector corresponding to the largest eigenvalue is calculated by the root-finding method and normalized to obtain the weights of latency, reliability, and power consumption attributes. Then, a weighted average is used to obtain the comprehensive fitness value. If different data streams have different QoS attribute priorities, attribute priority sorting can be introduced. First, a greedy matching is performed based on the highest priority attribute. Then, among the channels satisfying the highest priority attribute, selection is made based on the next highest priority attribute, and so on. Based on the data stream-to-channel mapping relationship obtained by the greedy algorithm, data streams with different service attributes are allocated to the most suitable channels for transmission. For delay-sensitive data streams, the channel with the lowest end-to-end latency is prioritized, and priority queuing and reserved time slots are used to ensure its priority transmission. For reliability-sensitive data streams, the channel with the lowest bit error rate is prioritized, and Turbo codes, LDPC codes, strong error correction coding, and ARQ and HARQ retransmission mechanisms are used. For low-power data streams, the channel with the lowest energy consumption is prioritized, and power consumption is reduced through power control and discontinuous reception mechanisms. During channel allocation, n*m channel multiplexing units can be constructed using AND, OR, and NOT gates, where n is the number of channels and m is the number of data streams multiplexed per channel. Each unit corresponds to a combination of a channel and a data stream. Dynamic selection is performed based on the combined fit value Dij. Simultaneously, OR gate logic is used between rows and columns to ensure that each data stream is assigned to only one channel, while each channel can reuse multiple data streams. If the number of allocated data streams k in the i-th channel is greater than the average, a penalty term is added to the selection signal of subsequent units in this row, converting Dij to Dij×(1+αk) to prevent excessive data stream aggregation. Similarly, the larger the QoS attribute deviation of the j-th data stream, the more penalty is applied to subsequent units in this column, converting Dij to Dij×(1+β|Sj'-Sj|), where Sj and Sj' are the target QoS attribute value of data stream j and the QoS attribute value provided by the currently matched channel, respectively, and β is the deviation penalty factor. Through this negative feedback mechanism, data streams are evenly distributed among channels, and the QoS performance deviations of each data stream are balanced.When idle channels exist and multiple data streams have similar QoS requirements, they can be bundled and multiplexed onto a single channel to improve channel utilization. The number of bundled data streams should not exceed the channel's maximum multiplexing capacity, and their QoS attribute weighted deviation should not exceed a certain threshold. If individual data streams cannot be satisfied by the current channel, a secondary matching channel is selected from the idle channel set. A virtual channel is constructed through dynamic channel borrowing, channel bandwidth bundling, and cognitive radio. Based on the cognitively perceived channel state, the transmission parameters of the virtual channel, including modulation order, coding rate, and transmit power, are dynamically optimized. To adapt to dynamic scenarios, a two-level dynamic optimization mechanism is designed. The first level, based on the data stream mapping matrix, fine-tunes the mapping matrix by combining the real-time load status and interference level of each channel, or triggers the second level of optimization. The second level uses deep reinforcement learning for optimization, quantifying QoS performance and fairness with a reward function R=(1-ω)U+ωS, where U is the data stream QoS attribute satisfaction rate, S is the channel utilization and load balancing degree, and ω is the balancing factor. A knowledge graph is constructed by integrating the multidimensional attributes of data streams and channels with their correlations. High-order semantic features are extracted using graph convolution and graph attention networks, serving as the state space of the DRL. The channel allocation matrix serves as the action space, and the REINFORCE algorithm of a GAN generator is used for policy gradient learning. An RNN discriminator evaluates the expected reward of actions and guides the generator optimization. Knowledge transfer and federated learning are used to share learned policy knowledge across different scenarios, improving the model's generalization ability. Simultaneously, dedicated channels are built around reliable, low-latency critical services, equipped with primary and backup channel disaster recovery schemes to maximize transmission quality.

[0028] For example, when using a greedy algorithm for data stream channel matching, the fit matrix can be preprocessed using a normalization method to normalize the weights of each dimension to the [0,1] interval. Taking an industrial control scenario as an example, the weight vectors of the three attributes of latency, reliability, and power consumption are determined to be (0.7, 0.2, 0.1) through AHP analysis. The three-dimensional attribute vector of a certain data stream is (50ms, 10^-5, 1mW), which, after mapping to the normalized space, becomes (0.05, 0.2, 0.1). The weighted Euclidean distance between this vector and the channel attribute vector (0.02, 0.1, 0.3) is 0.63, which is less than the threshold of 0.8, allowing for priority matching. When balancing the load of the matched channels, a penalty coefficient α is set to 0.5. If the allocated traffic accounts for more than 30% of the average, the fit value is multiplied by the penalty coefficient 1.5. When performing channel multiplexing, assuming there are 4 channels and 8 data streams, 4^8 = 32 gating units can be constructed. Each unit contains 4 NAND gates and 2 OR gates. If channel 2 has 3 multiplexed data streams, exceeding the average of 2, a penalty factor is introduced for subsequent units in that row, converting D24 from 0.3 to 0.3(1+0.51) = 0.45. If the delay deviation of data stream 5 is 20%, a penalty factor is introduced for that column, converting D15 from 0.2 to 0.2(1+0.60.2) = 0.224. When building the DRL optimization framework, the A3C algorithm can be used, designing a 2-layer GCN to extract knowledge graph features and a 3-layer GRU to implement policy generation. Reward = 0.8U + 0.2*S, where U is the QoS satisfaction rate and S is the channel utilization rate. After training for 5000 rounds, the test set accuracy can reach over 95%, which is 10% higher than the greedy algorithm. For critical services, three dedicated channels can be statically configured, each mapping its data source and data frame. Fault recovery is achieved through primary / backup switching, with a switching time of less than [time value missing].

[0029] S105: Based on the attributes of the environment in which the IoT device is located, collect the signal parameters of each available channel, combine them with the mapping relationship, adjust the channel allocation strategy, and select the target channel for data transmission.

[0030] This step may include: collecting data transmission statistics such as throughput, average latency, and packet loss rate for each channel by embedding lightweight monitoring probes into the network protocol stack of IoT devices; aggregating and calculating the collected channel-level and flow-level statistical data to form a global channel status view; smoothing each statistical indicator using a weighted moving average algorithm, with the smoothing coefficient adaptively adjusted according to the historical fluctuation characteristics of the indicators; constructing a multi-objective evaluation model based on service distribution characteristics and service quality requirements, and using the analytic hierarchy process (AHP) to determine the weights of each indicator and quantify the performance of channel traffic splitting.

[0031] Based on the signal strength and interference factors in the environment where IoT devices are located, the signal strength and interference parameters of each channel are collected in real time. Combined with the data stream and channel mapping relationship, the channel allocation strategy is dynamically adjusted to select the optimal channel for data transmission. IoT devices, through their integrated communication modules including WiFi, Bluetooth, and LoRa, sense the signal strength, signal-to-noise ratio, and bit error rate (BER) of various available channels in real time. Channel probing agents deployed on IoT devices actively send and analyze probing frame sequences, periodically assessing channel status and aggregating the sensing data from numerous dispersed devices to an edge gateway or cloud platform. Simultaneously, considering factors such as electromagnetic interference levels, terrain features, and building obstructions in the device's environment, the communication quality of each channel is comprehensively evaluated. Through data mining analysis, a correlation model between channel quality and the device's environment is established, forming a channel characteristic knowledge base tailored to IoT application scenarios. When an IoT device generates a new data stream, based on its service quality requirements (e.g., latency sensitivity, reliability priority), the most suitable candidate channel is selected from the data stream channel mapping matrix, and its priority is initially determined based on the channel's historical quality score in the knowledge base. For data streams whose transmission paths cross multiple network segments, a hierarchical channel selection mechanism is adopted. First, a transmission path is selected in the backbone network based on end-to-end latency and reliability factors. Then, the optimal access channel for each hop is determined by combining the real-time quality scores of each channel in the access network. Simultaneously, real-time channel parameters for each link segment are collected along the entire transmission path. An end-to-end comprehensive channel quality score is dynamically calculated using a weighted scoring method, serving as the basis for adjusting channel priorities. During actual data transmission, the communication quality of the selected channel is continuously monitored. When key quality parameters, such as the bit error rate, exceed a certain threshold, or the quality difference with other channels exceeds a certain threshold, the handover evaluation process for alternative channels is initiated. The handover evaluation needs to balance multiple factors, including communication quality, available bandwidth, and handover overhead, and a fuzzy comprehensive evaluation method can be used for multi-attribute decision-making. For transmission tasks with large data volumes and latency sensitivity, the MPTCP multipath transmission protocol can be used to slice the data and map it to multiple alternative channels for concurrent transmission. Combined with packet numbering and timestamp mechanisms, data integrity and timing are ensured. To further improve the adaptability and accuracy of channel allocation, an adaptive channel scheduling mechanism based on deep reinforcement learning is introduced. The channel quality parameters and data stream service quality attributes are used as the state space, while channel selection, data stream mapping, and transmission parameter configuration are used as the action space. Reward functions are designed to improve service quality, fairness, and resource utilization. The optimal channel allocation strategy is solved using a DQN value function approximation algorithm. During training, experience replay and a target network mechanism are used to improve learning efficiency and stability. Simultaneously, a federated learning framework is used for collaborative training among distributed IoT terminals, sharing environmental awareness and policy optimization knowledge without leaking local data, forming a channel allocation strategy model with a global perspective. To balance channel resource utilization efficiency and allocation fairness, a virtual resource pool is used to abstract and uniformly manage physical channels. Based on the network slicing concept, logical channel resource pools oriented towards different service levels are constructed.For critical business data streams, a dedicated resource pool is reserved, coupled with a priority scheduling and preemptive allocation mechanism to flexibly meet their service quality requirements. For ordinary data streams, a balanced scheduling and on-demand allocation method is adopted within the shared resource pool, improving resource utilization while adhering to the principle of proportional fairness. The scheduling strategy can also be adaptively optimized according to changes in channel load, appropriately reducing the resource occupancy ratio of ordinary data streams when the risk of channel congestion increases. Furthermore, the service quality of IoT data transmission can be further improved by combining the network's built-in communication optimization mechanisms. For latency-sensitive data, channel coding methods with high coding gain and low complexity, such as LDPC codes, are prioritized to reduce decoding latency. For low-power devices, coding mechanisms that balance coding efficiency and energy consumption, such as LT codes, are selected to extend battery life. For scenarios with high reliability requirements, Raptor code fountain code mechanisms are used to enhance the robustness of data transmission. Simultaneously, the collaboration between network slicing and edge computing can be utilized to offload data compression and security encryption tasks to the edge side, reducing the processing burden on the end side and shortening end-to-end transmission latency. Through the aforementioned multi-level, multi-dimensional, end-to-end channel allocation and transmission optimization, the service quality of IoT data transmission is maximized, providing highly reliable and low-latency communication guarantees for key IoT applications such as smart cities, intelligent manufacturing, and connected vehicles. During the optimization process, by balancing data acquisition frequency, compression ratio, and redundancy parameters, end-to-end communication overhead is controlled, keeping overall energy consumption within a set budget. Simultaneously, a closed-loop feedback loop is formed between service quality perception and channel allocation. Service quality assessment results promptly guide the dynamic evolution of allocation strategies, and the effectiveness of channel allocation continuously enhances the accuracy of service quality perception. This continuous optimization throughout the entire data transmission lifecycle constantly pushes the service quality limits of IoT applications.

[0032] For example, during channel state awareness, IoT devices scan every 10ms using their WiFi modules. By analyzing metrics such as RSSI and SNR in beacon frames and combining them with application-layer statistics such as bit error rate and retransmission rate, a channel quality feature vector containing seven key attributes is constructed. A single classifier such as One-Class SVM is then used to identify "abnormal" channels. When an interference source is detected, its modulation method and center frequency are analyzed using a spectrum energy detection algorithm to estimate its increased noise floor power. A corresponding strategy is then matched from the knowledge base, such as switching to a LoRa channel with stronger anti-interference capabilities. When matching channels based on service quality requirements, for latency-sensitive data streams, the top three candidate channels with the best latency attribute values ​​are selected, and redundant transmission and data compression are used to ensure latency requirements are met. For reliability-sensitive data streams, a combination of random linear network coding and HARQ is used, with parallel transmission on the two selected channels with the lowest bit error rate. The coding redundancy is dynamically adjusted, with the redundancy packet ratio not exceeding 20%. During multipath transmission, MPTCP's sub-stream mechanism is used to split the original data stream into three sub-streams, each mapped to one of the top three channels in terms of latency, reliability, and power consumption. Each data packet is appended with a 12-bit sequence number. The receiver reassembles the packets based on these sequence numbers. If three consecutive packets with the same sequence number are received, sub-stream migration is triggered. In deep reinforcement learning, channel switching is used as the agent, the quality of service attribute value as the state, and the score for each switch as the reward. The goal is to maximize the long-term cumulative reward. A double DQN algorithm is employed, with two Q-networks using a 3-layer MLP structure. The input is a combination of state and action features, corresponding to the comprehensive score of the available channels. With an initial learning rate of 0.01 and an experience replay capacity of 10,000, after 2000 training rounds, the average cumulative reward on the test set is 17% higher than that of the fixed channel allocation scheme. When allocating logical channel resources, for critical business data streams, 20% of the physical channel resource pool is allocated as a dedicated pool, and the two physical channels with the best latency and reliability attributes are selected to form a redundant transmission pair. Ordinary data streams share the remaining 80% of the resource pool, employing WFQ dynamic scheduling with weights determined by service priority. Weight coefficients are recalculated each scheduling cycle. Regarding collaborative compression and transmission, for video surveillance data streams with large data volumes and low latency, ROI adaptive coding based on random forests is used, achieving an average compression ratio of 30%. The compressed video slices are then encoded and stored on multiple edge nodes using an NC distributed storage system, with parallel decoding and transmission upon request, reducing end-to-end latency.

[0033] S106: Collect data transmission statistics for each available channel, and analyze the statistics using a weighted moving average algorithm to quantify the performance of channel splitting.

[0034] Real-time data transmission statistics such as throughput, latency, and packet loss rate for each channel are collected. A weighted moving average algorithm is used to analyze these statistics and quantitatively evaluate the performance of channel offloading. By embedding lightweight monitoring probes into the network protocol stack of IoT devices, real-time data transmission statistics such as throughput, average latency, and packet loss rate for each channel are collected. Deployed at the MAC layer, the probes efficiently capture channel-level traffic statistics through a combination of bypass sampling and conditional triggering. Simultaneously, traffic mirroring and data analysis modules are deployed at key network element nodes in core switches and edge routers to aggregate and calculate the collected channel-level and flow-level statistical data, forming a global channel status view. A weighted moving average algorithm is used to smooth various statistical indicators, with the smoothing coefficient adaptively adjusted based on the historical fluctuation characteristics of the indicators; for example, EWMA is used to dynamically track the changing trend of latency jitter coefficient. By calculating key performance indicators such as channel utilization, 95th percentile latency, and fault blocking probability, combined with service distribution characteristics and service quality requirements, a multi-objective evaluation model is constructed. The analytic hierarchy process (AHP) is used to determine the weights of each indicator, quantitatively evaluating the effectiveness of each channel traffic splitting scheme. The Gini coefficient and resource occupancy deviation fairness metrics are introduced to evaluate the balance of channel allocation for different priority services and their matching degree with service quality. Based on the evaluation results, when there is a severe load imbalance between channels, dynamic load balancing is triggered by threshold detection. The traffic splitting threshold is adaptively correlated with the channel utilization level, and a simulated annealing algorithm is used to solve for the optimal migration scheme of the data flow. When the end-to-end latency of critical services violates SLA constraints, the channel scheduling strategy and queue management mechanism are dynamically adjusted by subscribing to quality of service events, prioritizing traffic that violates constraints and reserving dedicated resources. When the latency jitter and packet loss rate across the entire network exceed the set performance alarm threshold, a grey prediction model is used to assess the medium- and long-term growth trend of traffic demand, initiating channel expansion and upgrade processes in advance to eliminate potential bottlenecks. To further improve channel utilization efficiency and operational flexibility, a grouping mechanism based on traffic patterns and QoS characteristics is introduced. The K-means clustering algorithm is used to cluster the packet length distribution, burst characteristics, and QoS level matrix attributes of the service data flow, extracting traffic model parameters. For different clustering results, a logical channel resource pool is constructed to adapt to them. Dynamic time slot multiplexing and code division multiple access mechanisms are used within the pool to achieve flexible sharing of channel resources. When traffic patterns change significantly, new cluster centers are identified using anomaly detection algorithms, and the size and mapping method of the channel resource pool are adaptively adjusted. In weighted moving average analysis, an adaptive weight estimation strategy is adopted to address the non-stationary characteristics of network state and service patterns. The ARIMA model is used to characterize the periodic variation features of traffic, and the smoothing coefficient is adjusted in an event-triggered manner based on the prediction results. Wavelet analysis is used to extract multi-scale burst features of traffic, and the weights of each scale are obtained through energy spectral density mapping, adaptively matching the smoothing coefficient with the scale. Bayesian inference is used to learn the state transition probability of traffic online, and the smoothing coefficient is dynamically adjusted based on the posterior state probability.Simultaneously, a service quality feedback loop is established, integrating multi-dimensional experience scores based on latency sensitivity, reliability requirements, and rate satisfaction of the data flow to form a comprehensive channel experience quality evaluation matrix. Through fuzzy logic reasoning, the quantitative scores of each dimension are mapped to the channel adjustment strategy domain. In the fuzzy controller, membership levels are divided based on the strength of latency sensitivity and reliability requirements, a fuzzy rule base is constructed, approximate reasoning is performed using the Mamdani inference mechanism, and defuzzification is completed using the centroid method, generating fuzzy control quantities for channel resource adjustment and traffic splitting optimization. At the IoT edge, a deterministic service enhancement mechanism is introduced for critical business data flows. Network slicing is used to divide hard-isolated virtual edge networks, and TSN and DetNet are used to provide deterministic bandwidth guarantees and latency jitter control within the virtual networks. For business interaction modes, a service-oriented abstract channel model is constructed, reserving end-to-end physical resources to guarantee deterministic SLAs. Within the virtual edge network, for the multi-connection concurrent transmission requirements of business data flows, load balancing and seamless flow switching for multipath transmission are achieved through traffic-aware dynamic channel binding. Meanwhile, DPI-based intelligent traffic identification is adopted at network entry and exit points to mark key business data flows with priority and map them to high-priority dedicated transmission queues in the forwarding plane. Combined with TSN time-aware traffic scheduling, the queuing latency of key business data is minimized.

[0035] For example, a lightweight monitoring probe is embedded in the MAC layer of an IoT device. Leveraging the multi-core parallel capabilities of the network processor, channel data packets are bypassed and mirrored to the probe buffer at a sampling rate of 1:10000. Full traffic capture is triggered via 5-tuple matching, with each capture lasting no more than 10ms. Simultaneously, the probe performs real-time stream summarization on the captured data, extracting key statistical indicators such as throughput, average latency, and packet loss rate for each channel, and reports this data to the MEC edge cloud in real time via the gRPC protocol. Multi-source probe data is aggregated and analyzed at the MEC node. The EWMA algorithm is used to track the trend of latency jitter coefficient changes. The initial smoothing coefficient α is set to 0.2, and the α value is adaptively adjusted every 100 samples with an adjustment step size of 0.05. An alarm event is triggered when the channel utilization exceeds 0.7, the 95th percentile latency exceeds 50ms, or the fault blocking probability exceeds 10^-5. Combining the experience rate, reliability score, and service quality weight matrix of the service data stream, a fuzzy comprehensive evaluation method is used to score and rank the service carrying capacity of each channel. The weights are then determined using the analytic hierarchy process (AHP) to generate channel overload probability prediction results. For overloaded channels, a load balancing process is initiated, and a simulated annealing algorithm is used to search the traffic migration strategy space. The degradation of latency jitter for critical services is controlled to not exceed 20%, latency jitter for non-critical services to not exceed 100%, and the impact of migration on the packet loss rate of critical services is not to exceed 1%. When performing QoS clustering and grouping of IoT data streams, the RSVP reservation parameters of service requests are extracted, and the length distribution and burst characteristics of data packets are collected. The data streams are divided into 35 QoS levels using the K-means algorithm. An independent channel resource pool is allocated for each level, and dynamic time slot multiplexing (TDMA) is used within the pool. The maximum time slot occupancy ratio for critical services does not exceed 80% of the total pool resources, while non-critical services use the Aloha protocol to statistically reuse remaining resources. When a new QoS cluster center is detected, a resource pool boundary adjustment is triggered, with an adjustment step size of 5% of the total resources. In traffic anomaly detection, wavelet analysis is used to extract multi-scale energy spectral density features of traffic. The Daubechies-4 wavelet basis is selected, with a decomposition scale of 5, corresponding to a time scale span of 100ms to 1.6s. The adaptive smoothing coefficient obtained through wavelet scale energy weight mapping is adjusted from 0.2 to 0.8. In the fuzzy controller design for quality of service feedback mapping, latency sensitivity and reliability requirements are used as inputs, divided into three membership levels: high, medium, and low. A Gaussian membership function is used, with corresponding means of 0.8, 0.5, and 0.2, and standard deviations of 0.1 for all three levels. The output space represents the channel adjustment strategy, divided into two fuzzy subsets: reducing low-priority traffic and increasing the priority of critical flows. Mamdani inference employs a product-summation-centroid defuzzification method, and the inference rule base consists of 25 IF-THEN rules.When providing deterministic service enhancements within the TSN virtual edge network, a service function chain is abstracted and a service matrix is ​​constructed based on the interaction patterns of key services. The matrix elements represent the interaction latency requirements between services. A graph coloring algorithm is used to partition the service matrix, with each partition mapped to a virtual channel. Transmission slot resources are reserved based on the most stringent latency requirements within each partition. For concurrent transmission across channels, a traffic-aware dynamic channel binding strategy is designed. Using the flow quintuple as the binding key, the source address hashing method is employed to allocate flows to multiple available channels. By actively probing the RTT differences of the links and combining congestion prediction feedback, seamless handover of flows is triggered, with a single handover latency not exceeding [a certain value].

[0036] S107: Based on the performance of the channel splitting, optimize the weights of each attribute in the weighted Euclidean distance adaptation function and the greedy algorithm, and perform communication based on the optimized weighted Euclidean distance adaptation function and the greedy algorithm.

[0037] Based on the channel splitting performance evaluation results, the weights of each attribute in the weighted Euclidean distance adaptation function are dynamically optimized, and the greedy algorithm strategy is optimized to continuously improve the overall performance of multi-channel communication. The optimized adaptation function and channel allocation algorithm are then applied to the subsequent data transmission process.

[0038] Based on the channel splitting performance evaluation results obtained in the previous stage, an adaptive dynamic optimization method is adopted to continuously improve the channel attribute weight configuration in the weighted Euclidean distance adaptation function. First, the correlation between each attribute and service quality is measured using information gain and Gini coefficient statistical indicators to preliminarily determine the importance ranking of attribute weights. On this basis, an attribute-service quality multivariate regression model is constructed, and a subset of key attributes is extracted using the L1 regularized sparsity penalty method to quantify the weight coefficients of each key attribute. Then, a hierarchical weight dynamic optimization strategy is designed for the service quality-sensitive characteristics of service data. For latency-sensitive services, the transmission latency weight is increased; for reliability-sensitive services, the packet loss rate weight is increased. The weight adjustment range is quantitatively mapped to the sensitivity of service data, and an adjustment upper limit is set to avoid weight imbalance. Through service-adaptive hierarchical optimization, the attribute weight vector of the adaptation function is dynamically matched with the service feature vector, improving the accuracy of optimization. For the greedy algorithm in channel allocation, a reinforcement learning-based strategy optimization framework is introduced. Offline data collection is performed on typical communication scenarios in multi-source heterogeneous environments, including channel parameters, service characteristics, allocation schemes, and quality of service feedback. This data is then cleaned and feature-engineered to construct a high-quality training sample set for the algorithm. Expert experience and combinatorial optimization algorithms are used to pre-label the sample data, forming near-optimal allocation decisions as prior knowledge for reinforcement learning. Based on this, an asynchronous advantage algorithm, A3C, integrating policy gradients and Q-learning, is designed to train multiple agents in parallel. A nonlinear combination of channel attributes and service characteristics is used as the state space, and selectable channel combinations are used as the discrete action space. Quality of service improvement and resource utilization are used as reward functions. GPU-accelerated distributed training continuously improves the long-term returns of the greedy algorithm. During training, an active learning mechanism is designed to dynamically adjust the exploration-exploitation tradeoff based on the channel comprehensive scoring matrix and the Thompson sampling strategy, focusing learning on key channels and services that significantly improve global returns. Through iterative optimization, an allocation strategy matrix that adapts to the environment is finally formed. Before applying the optimized algorithm in practice, a high-fidelity simulation environment for multi-channel communication is constructed using DigitalTwin. An end-to-end communication simulation platform was built using physical modeling of communication links, simulation of service data flows, and virtualization of hardware devices. A pre-trained digital twin model was then used to synchronize key attributes of the physical channel in real time. The optimized algorithm was then deployed to the simulation platform, and large-scale verification tests were conducted using Monte Carlo simulations to evaluate the statistical confidence level of the optimization effect. For scenarios where performance did not meet expectations, bottlenecks were diagnosed through visualization analysis, and the fitness function and allocation strategy were iteratively optimized until the performance requirements were met. Simultaneously, the resource overhead and latency loss caused by algorithm optimization were evaluated, balancing the optimization costs and benefits, and adaptively adjusting the algorithm's complexity and update frequency. Quantitative indicators and probability distributions of the optimization effect were obtained through simulation verification, providing a basis for the engineering deployment of the algorithm.After the optimized algorithm is verified through simulation, it is dynamically loaded into the network forwarding plane by the SDN controller. Leveraging the programmability of the data plane, the attribute weight vector of the adaptation function and the allocation strategy matrix are flexibly mapped to flow table rules, adjusting channel mapping relationships and data forwarding paths in real time, allowing the optimization results to take effect dynamically online. In response to changes in the data plane state, the controller periodically polls the statistical information of each forwarding node, extracts flow table hit rate and link utilization KPIs, and monitors key events during algorithm operation, achieving dynamic synchronization and real-time closed-loop between optimization logic and network status. When abnormal fluctuations in service quality occur, the root cause is diagnosed and assessed whether it is caused by algorithm iteration. If it is caused by environmental changes, algorithm retraining is triggered; if it is caused by limitations of the optimization scheme, other candidate algorithm optimization schemes in the knowledge base are invoked, and the optimal scheme is selected through A / B testing to ensure service continuity. To address the cold start problem of algorithm optimization in new environments, considering the differences in various business scenarios, a metric-based scenario similarity matching approach is adopted. Similar scenarios are retrieved from a historical scenario-algorithm knowledge base, and corresponding algorithm parameters and logical structures are extracted and ported to the new environment as the starting point for optimization, significantly reducing sampling complexity. Then, parameter fine-tuning and model distillation transfer learning are used to quickly adapt to the new environment under small sample conditions while maintaining generalization performance. Simultaneously, meta-learning maps the hand-designed neural network structure to an optimizable hyperparameter space, using heuristic search to quickly find the optimal network architecture that better fits the target scenario, thereby guiding the dynamic evolution of the algorithm model. Through the synergy of scenario similarity matching and meta-learning, continuous evolution of the optimization algorithm in changing environments is achieved, constantly pushing the Pareto frontier of service quality across multiple scenarios.

[0039] For example, in the dynamic optimization of channel attribute weights, for environmental monitoring services in a smart city scenario, information gain analysis using the information gain algorithm reveals that the information gain values ​​for transmission delay and packet loss rate are 0.45 and 0.3, respectively, significantly higher than other attributes such as power consumption (0.1). Therefore, the first two attributes are given higher weights in the adaptation function design. Further, a multivariate linear regression model is fitted to the attribute-quality of service model, and L1Lasso regularization is applied for sparsity learning to identify two key attributes: delay and packet loss rate, with regression coefficients of 0.6 and 0.4, respectively. Since the delay sensitivity of this service is in the millisecond range, the delay weight is further increased to 0.8 to enhance the dominant role of delay in the adaptation function. In optimizing channel allocation strategies using reinforcement learning, a simulation environment with 100 tracks and 500 channels was constructed for an AGV scheduling scenario in an industrial park. 2000 offline data points were sampled as the training set. The Gumbelsoftmax technique was applied to parameterize the channel selection process, and distributed asynchronous training was conducted using the A3C algorithm with 10 agents running in parallel. After 5000 iterations, the reward converged to 95 points. Key performance indicators such as latency and reliability were improved by 15% and 10% respectively compared to the default greedy algorithm, and the channel selection of the Actor network tended to stabilize. During training, the data was prioritized for replay based on the comprehensive channel score, with the sampling probability proportional to the score. Monte Carlo Tree Search was used to guide the action selection of exploration, focusing the exploration on the top 20% of channels. The average score of the new strategy improved by 10% compared to priority sampling. In the Digitaltwin simulation verification phase, for a real-time data transmission scenario involving multiple sensors in a smart park, a heterogeneous communication scenario covering the park was first constructed using a 5G / WiFi6 / WLAN system-level simulation platform. Virtualization was then used to map key attributes of the physical channels on demand, obtaining high-fidelity digital twin channel model parameters. The optimized channel allocation algorithm was then deployed to the simulation platform, with a data sampling frequency of 10ms and a concurrency of 100. 10,000 Monte Carlo simulations were run to obtain the distribution characteristics of key performance indicators such as end-to-end latency, throughput, and reliability. Statistical results show that the optimized algorithm has a 99.5% probability of latency jitter less than 10ms and a reliability of nine nines, meeting the performance expectations for smart park applications. The time and space overhead of one inference iteration was also measured, with an average time of 5ms and memory usage of 10MB. The inference frequency can be flexibly adjusted to balance optimization effects and resource costs.When applying the optimization algorithm in a production environment, a P4 programmable switch is used as the SDN forwarding plane. Combined with controllers such as ONOS, the channel allocation matrix is ​​dynamically distributed to flow table entries. The matching field is the data flow quintuple, and the corresponding action is the output port. The distribution period is configured to be 100ms. The controller subscribes to the performance telemetry data of the forwarding plane via gRPC. Prometheus extracts key indicators such as latency and packet loss rate. Alertmanager detects abnormal performance degradation. If it is caused by algorithm iteration, the algorithm version is rolled back to the previous stable version via Git. If it is caused by sudden traffic, resampling and training are performed. If the reward improvement does not exceed 1% after 5 iterations of optimization, a new stable version is released. Regarding the algorithm's multi-scenario adaptation, contrastive loss in metric learning is used to sample pairwise service scenarios from the knowledge base. This forces the L2 distance between scenario vectors (including channel and service attributes) of the same service in the embedding space to be less than a threshold, while the distance between different scenarios is greater than a threshold, thereby extracting cross-scenario discriminative features. In the new AGV scheduling scenario, scene vectors are extracted through embedding, and nearest neighbor scenes are retrieved from the knowledge base. With scene similarity greater than 80%, the 90th percentile latency of the direct transfer optimization model in this scenario is reduced from 150ms to 100ms, and transfer convergence is achieved with only 500 sampled data points. Furthermore, AutoML is used to search for the Actor network structure, defining three search dimensions with four candidate values ​​per dimension. After 50 iterations of optimization, the optimal combination is selected: 8 fully connected layers, 256-dimensional ReLU, and ResNet shortcut, which outperforms manually designed network structures by 5 percentage points, accelerating the algorithm's generalization and evolution.

[0040] Based on the above features and functions, this method solves the problem of effectively allocating different types of data streams across different channels according to their respective quality of service requirements, and dynamically adjusting channel allocation strategies to adapt to environmental changes and network conditions, thereby optimizing data transmission performance. Specifically, it achieves efficient matching between data streams and channel characteristics, optimizing the quality and efficiency of data transmission; real-time monitoring and adjustment of channel allocation strategies enable the invention to flexibly respond to environmental changes and network condition fluctuations, improving the reliability and stability of communication; and by quantifying the matching degree between data streams and channels and generating a fit matrix, it further enhances the overall performance of multi-channel communication. In summary, this invention provides a multi-channel communication method for the Internet of Things (IoT), significantly improving data transmission efficiency and quality, reducing power consumption, and enhancing system reliability and security.

[0041] The technical problems to be solved by this invention patent include: 1. Sudden surges in bandwidth demand can lead to channel congestion. This invention patent provides a multi-channel communication method for the Internet of Things (IoT), which rationally distributes traffic among different channels and dynamically adjusts channel allocation strategies to adapt to environmental changes and network conditions, thereby optimizing data transmission performance. Specifically, it achieves efficient matching between data streams and channel characteristics, optimizing the quality and efficiency of data transmission.

[0042] 2. Improved communication reliability and stability. This invention provides a multi-channel communication method for the Internet of Things (IoT), which monitors and adjusts channel allocation strategies in real time. This allows the invention to flexibly respond to environmental changes and network status fluctuations, thereby improving communication reliability and stability.

[0043] 3. Reduced power consumption. This invention patent provides a multi-channel communication method in the Internet of Things (IoT), which generates an adaptation matrix by quantifying the matching degree between data streams and channels, thereby further improving the overall performance of multi-channel communication.

[0044] Figure 2 The diagram shows the structure of the positioning device provided in the embodiment of this application. The device 200 includes: an analysis module 210, a calculation module 220, a processing module 230, a determination module 240, an adjustment module 250, a quantization module 260, and an optimization module 270.

[0045] Analysis module 210 analyzes the service quality requirements of different data streams based on IoT application scenarios and device characteristics; The calculation module 220 obtains the transmission characteristic parameters of multiple available channels, forms a channel characteristic matrix, and uses the data stream service quality attribute vector as input to calculate the matching degree with the channel characteristic matrix. The processing module 230 quantifies the matching degree between the data stream and the available channels by constructing a weighted Euclidean distance adaptation function, and generates an adaptation matrix; The determination module 240 uses a greedy algorithm to solve the fitness matrix and obtain the mapping relationship between the data stream and the available channels; The adjustment module 250 collects the signal parameters of each available channel according to the attributes of the environment in which the IoT device is located, and adjusts the channel allocation strategy in combination with the mapping relationship to select the target channel for data transmission. The quantization module 260 collects data transmission statistics of each available channel, and uses a weighted moving average algorithm to analyze the statistics and quantify the performance of channel splitting. The optimization module 270 optimizes the weights of each attribute in the weighted Euclidean distance adaptation function and the greedy algorithm based on the performance of the channel splitting, and performs communication based on the optimized weighted Euclidean distance adaptation function and the greedy algorithm.

[0046] Optionally, the step of analyzing the quality of service requirements of different data streams based on IoT application scenarios and device characteristics includes: Optionally, based on the IoT application scenarios and device characteristics, analyze the latency sensitivity, reliability requirements, security requirements, and power consumption budget service quality attributes of different data streams to obtain the service quality attribute vector of each data stream; Based on the service quality attribute vector of the data stream, and in combination with specific business requirements and network environment, the transmission protocol and data processing method applicable to each data stream are determined to meet the service quality requirements.

[0047] Optionally, the step of obtaining transmission characteristic parameters of multiple available channels to form a channel characteristic matrix, and using the data stream service quality attribute vector as input to calculate the matching degree with the channel characteristic matrix includes: The instantaneous signal-to-noise ratio and coherence bandwidth physical layer parameters of the currently available channels are obtained through channel sensing. Estimate the maximum transmission rate, average delay, and packet loss rate for each channel; The power consumption characteristics of the channel are obtained by power detection, and a channel characteristic matrix reflecting the real-time transmission characteristics of each channel is formed. The service quality attribute vector of the data stream is quantified into specific indicators such as transmission rate, latency, reliability, and energy efficiency, and different weights are assigned to them. The channel characteristic matrix is ​​compared with the specific indicators of transmission rate, latency, reliability, and energy efficiency, and weighted Euclidean distance is calculated to obtain the matching degree between each channel and the service quality requirements of the data stream.

[0048] Optionally, the step of quantifying the matching degree between the data stream and the available channels by constructing a weighted Euclidean distance fitness function and generating a fitness matrix includes: The service quality attribute vector and channel characteristic matrix of the data stream are normalized, and the indicators of each dimension are mapped to the interval [0, 1]. The weights of each indicator in the service quality attributes are determined by the analytic hierarchy process (AHP). Through pairwise comparison judgment matrix calculation and consistency verification, the weight vectors of latency, bit error rate, and power consumption indicators are obtained, forming a weighted Euclidean distance metric. The fitness matrix is ​​generated by obtaining the fitness value between the data stream and each channel through matrix operations.

[0049] Optionally, the step of using a greedy algorithm to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channels includes: Based on the service quality attribute vector of the data stream and the characteristic matrix of the channel, a greedy algorithm is used to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channel.

[0050] Optionally, data transmission statistics for each available channel are collected, and a weighted moving average algorithm is used to analyze the statistics to quantify the performance of channel offloading, including: By embedding lightweight monitoring probes into the network protocol stack of IoT devices, data transmission statistics such as throughput, average latency, and packet loss rate of each channel are collected. The collected channel-level and flow-level statistical data are aggregated and calculated to form a global channel state view; A weighted moving average algorithm is used to smooth various statistical indicators, and the smoothing coefficient is adaptively adjusted according to the historical fluctuation characteristics of the indicators. Based on the characteristics of service distribution and service quality requirements, a multi-objective evaluation model is constructed, and the weight of each indicator is determined by the analytic hierarchy process (AHP) to quantify the performance of channel offloading.

[0051] The apparatus provided in this application embodiment can execute the methods described in the preceding method embodiments and achieve the functions and beneficial effects of the methods described in the preceding method embodiments, which will not be repeated here.

[0052] Figure 3 The diagram illustrates the hardware structure of an electronic device implementing the embodiments of this application. Referring to the diagram, at the hardware level, the electronic device includes a processor and optionally, an internal bus, a network interface, and a memory. The memory may include RAM, such as high-speed random-access memory (RAM), or non-volatile memory, such as at least one disk storage device. Of course, the electronic device may also include other hardware required for other services.

[0053] The processor, network interface, and memory can be interconnected via an internal bus, which can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be categorized as an address bus, data bus, control bus, etc. For ease of illustration, only a single bidirectional arrow is used in this diagram, but this does not imply that there is only one bus or one type of bus.

[0054] Memory is used to store programs. Specifically, programs may include program code, which includes computer operation instructions. Memory may include main memory and non-volatile memory, and provides instructions and data to the processor.

[0055] The processor reads the corresponding computer program from non-volatile memory into main memory and then executes it, forming a device at the logical level that locates the target user. The processor executes the program stored in memory and specifically performs the following: Figure 1-2 The methods disclosed in the embodiments shown achieve the functions and beneficial effects of the methods described in the preceding method embodiments, and will not be repeated here.

[0056] The above is as stated in this application. Figure 1-2 The methods disclosed in the illustrated embodiments can be applied to or implemented by a processor. The processor may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above methods can be completed by integrated logic circuits in the processor's hardware or by instructions in software form. The processor can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc.; it can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly embodied in the execution of a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software module can reside in a mature storage medium in the field, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, or registers. This storage medium is located in memory, and the processor reads information from the memory and, in conjunction with its hardware, completes the steps of the above method.

[0057] The electronic device can also execute the methods described in the preceding method embodiments and achieve the functions and beneficial effects of the methods described in the preceding method embodiments, which will not be repeated here.

[0058] Of course, in addition to software implementation, the electronic device of this application does not exclude other implementation methods, such as logic devices or a combination of hardware and software, etc. In other words, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.

[0059] This application also proposes a computer-readable storage medium that stores one or more programs, which, when executed by an electronic device including multiple applications, cause the electronic device to perform... Figure 1-2 The methods disclosed in the embodiments shown achieve the functions and beneficial effects of the methods described in the preceding method embodiments, and will not be repeated here.

[0060] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the above query method embodiments and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0061] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0062] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0063] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0064] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A communication method, the method comprising: Based on IoT application scenarios and device characteristics, analyze the service quality requirements of different data streams; The transmission characteristic parameters of multiple available channels are obtained to form a channel characteristic matrix. The data flow service quality attribute vector is used as input to calculate the matching degree with the channel characteristic matrix. By constructing a weighted Euclidean distance fit function, the matching degree between the data stream and the available channels is quantified, and a fit matrix is ​​generated; A greedy algorithm is used to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channels; Based on the attributes of the environment in which the IoT device is located, the signal parameters of each available channel are collected, and combined with the mapping relationship, the channel allocation strategy is adjusted to select the target channel for data transmission. Data transmission statistics of each available channel are collected, and a weighted moving average algorithm is used to analyze the statistics to quantify the performance of channel diversion. Based on the performance of the channel splitting, the weights of each attribute in the weighted Euclidean distance fitness function and the greedy algorithm are optimized. Communication is then performed based on the optimized weighted Euclidean distance fitness function and the greedy algorithm.

2. The method according to claim 1, characterized in that, The analysis of service quality requirements for different data streams based on IoT application scenarios and device characteristics includes: Based on IoT application scenarios and device characteristics, the latency sensitivity, reliability requirements, security requirements, and power consumption budget service quality attributes of different data streams are analyzed to obtain the service quality attribute vector of each data stream. Based on the service quality attribute vector of the data stream, and in combination with specific business requirements and network environment, the transmission protocol and data processing method applicable to each data stream are determined to meet the service quality requirements.

3. The method according to claim 1, characterized in that, The process of obtaining transmission characteristic parameters of multiple available channels to form a channel characteristic matrix, and using the data stream service quality attribute vector as input to calculate the matching degree with the channel characteristic matrix includes: The instantaneous signal-to-noise ratio and coherence bandwidth physical layer parameters of the currently available channels are obtained through channel sensing. Estimate the maximum transmission rate, average delay, and packet loss rate for each channel; The power consumption characteristics of the channel are obtained by power detection, and a channel characteristic matrix reflecting the real-time transmission characteristics of each channel is formed. The service quality attribute vector of the data stream is quantified into specific indicators such as transmission rate, latency, reliability, and energy efficiency, and different weights are assigned to them. The channel characteristic matrix is ​​compared with the specific indicators of transmission rate, latency, reliability, and energy efficiency, and weighted Euclidean distance is calculated to obtain the matching degree between each channel and the service quality requirements of the data stream.

4. The method according to claim 1, characterized in that, The step of constructing a weighted Euclidean distance fit function to quantify the matching degree between the data stream and the available channels, and generating a fit matrix, includes: The service quality attribute vector and channel characteristic matrix of the data stream are normalized, and the indicators of each dimension are mapped to the interval [0, 1]. The weights of each indicator in the service quality attributes are determined by the analytic hierarchy process (AHP). Through pairwise comparison judgment matrix calculation and consistency verification, the weight vectors of latency, bit error rate, and power consumption indicators are obtained, forming a weighted Euclidean distance metric. The fitness matrix is ​​generated by obtaining the fitness value between the data stream and each channel through matrix operations.

5. The method according to claim 1, characterized in that, The step of using a greedy algorithm to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channels includes: Based on the service quality attribute vector of the data stream and the characteristic matrix of the channel, a greedy algorithm is used to solve the fitness matrix to obtain the mapping relationship between the data stream and the available channel.

6. The method according to claim 1, characterized in that, Data transmission statistics for each available channel are collected, and a weighted moving average algorithm is used to analyze these statistics to quantify the performance of channel offloading, including: By embedding lightweight monitoring probes into the network protocol stack of IoT devices, data transmission statistics such as throughput, average latency, and packet loss rate of each channel are collected. The collected channel-level and flow-level statistical data are aggregated and calculated to form a global channel state view; A weighted moving average algorithm is used to smooth various statistical indicators, and the smoothing coefficient is adaptively adjusted according to the historical fluctuation characteristics of the indicators. Based on the characteristics of service distribution and service quality requirements, a multi-objective evaluation model is constructed, and the weight of each indicator is determined by the analytic hierarchy process (AHP) to quantify the performance of channel offloading.

7. A positioning device, comprising: The analysis module analyzes the service quality requirements of different data streams based on IoT application scenarios and device characteristics. The calculation module obtains the transmission characteristic parameters of multiple available channels, forms a channel characteristic matrix, and calculates the matching degree between the data flow service quality attribute vector as input and the channel characteristic matrix. The processing module quantifies the matching degree between the data stream and the available channels by constructing a weighted Euclidean distance adaptation function, and generates an adaptation matrix. The module determines the fit matrix using a greedy algorithm to obtain the mapping relationship between the data stream and the available channels; The adjustment module collects signal parameters of each available channel based on the attributes of the environment in which the IoT device is located, and adjusts the channel allocation strategy in conjunction with the mapping relationship to select the target channel for data transmission. The quantization module collects data transmission statistics for each available channel, and analyzes the statistics using a weighted moving average algorithm to quantify the performance of channel splitting. The optimization module optimizes the weights of each attribute in the weighted Euclidean distance adaptation function and the greedy algorithm based on the performance of the channel splitting, and performs communication based on the optimized weighted Euclidean distance adaptation function and the greedy algorithm.

8. An electronic device, comprising: processor; as well as A memory configured to store computer-executable instructions, which, when executed, use the processor to perform the steps of the session method according to any one of claims 1-6.

9. A computer-readable medium storing one or more programs, which, when executed by an electronic device including a plurality of applications, cause the electronic device to perform the steps of the session method according to any one of claims 1-6.

10. A computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions that, when executed by a computer, perform the steps of the session method according to any one of claims 1-6.

Images