A power internet of things dual-mode communication method based on deep learning channel estimation

Abstract

This invention discloses a dual-mode communication method for the power Internet of Things (IoT) based on deep learning for channel estimation, relating to the field of power IoT communication technology. The method includes: designing a unified MAC layer protocol and setting a bit error rate (BER) requirement factor field in the data frame structure; obtaining power line and low-power wireless communication link parameters based on this, and determining the optimal communication path by introducing perturbation parameters; processing the initial channel matrix using deep learning to generate feature channel weights and perform residual connections, outputting channel state information; dynamically adjusting the BER requirement factor using a gradient adjustment method and calculating the sub-stream allocation ratio; and splitting the service data into sub-streams proportionally and transmitting them through corresponding links. This invention achieves intelligent selection of power IoT communication links and adaptive allocation of service data.

Description

Technical Field

[0001] This invention relates to the field of power Internet of Things (IoT) communication technology, specifically to a dual-mode communication method for power IoT based on deep learning-based channel estimation. Background Technology

[0002] With the integration of smart grids and the energy internet, the demand for wide coverage and high reliability transmission in the information sensing layer of the power internet of things (IoT) is becoming increasingly urgent. The link layer, also known as the MAC layer (Medium Access Control), is crucial. Currently, the main communication methods for the power IoT are low-voltage power line communication and low-power wireless communication. However, both have significant limitations. PLC communication relies on power line transmission and is affected by factors such as power grid noise and impedance changes, resulting in low reliability in complex power grid environments. Traditional channel estimation methods have large errors in low signal-to-noise ratio environments, making it difficult to meet the requirements of high-reliability communication. Existing deep learning channel estimation schemes mostly use fully connected networks or complex deep convolutional networks, resulting in a huge number of parameters, making it difficult to run in real time on power IoT terminals with limited computing resources. Low-power wireless communication is easily affected by building obstruction and electromagnetic interference, making it difficult to meet the full coverage requirements of complex scenarios such as buildings. Furthermore, existing fusion solutions only focus on the simple superposition of the physical or network layers, failing to achieve true cross-layer optimization.

[0003] Therefore, the present invention provides a dual-mode communication method for the power Internet of Things based on deep learning for channel estimation, which improves the above-mentioned technical problems. Summary of the Invention

[0004] The purpose of this invention is to provide a dual-mode communication method for the power Internet of Things based on deep learning-based channel estimation, aiming to solve at least one of the technical problems existing in the prior art.

[0005] The technical solution of this invention is: a dual-mode communication method for the power Internet of Things based on deep learning for channel estimation, comprising the following steps: Design a unified MAC layer protocol, set a bit error rate requirement factor field in the unified data frame structure, and output a unified data frame structure containing the bit error rate requirement factor field. Link parameters of power line communication links and low-power wireless communication links are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to service type, and disturbance parameters are introduced to adjust the path search process, outputting the optimal communication path. In the optimal communication path, the power line communication link and the low-power wireless communication link receive pilot signals to obtain the initial channel matrix. The initial channel matrix is ​​then subjected to global average pooling through a convolutional layer to obtain a compressed feature vector. After dimensionality reduction and dimensionality increase processing of the compressed feature vector through a convolutional layer, feature channel weights are generated. The feature channel weights are then multiplied element-wise with the initial channel matrix and residually concatenated with the initial channel matrix to output channel state information. The bit error rate requirement factor is obtained from the bit error rate requirement factor field in the unified data frame structure. The bit error rate requirement factor is dynamically adjusted using a gradient adjustment method based on the channel state information. The sub-stream allocation ratio is calculated based on the adjusted bit error rate requirement factor and the channel state information. The business data is split into sub-streams according to the sub-stream allocation ratio and transmitted separately through the power line communication link and the low-power wireless communication link in the optimal communication path.

[0006] The design of the unified MAC layer protocol includes setting a bit error rate requirement factor field in the unified data frame structure, and outputting a unified data frame structure containing the bit error rate requirement factor field, including: The beacon period is divided into contention-based and non-contention-based time slots. During the contention-based time slots, a carrier sense multiple access mechanism is used to control node access, while during the non-contention-based time slots, a time division multiple access mechanism is used to allocate transmission time slots. Construct an initial data frame structure that includes a preamble field, a network type field, and a bit error rate requirement factor field; In the initial data frame structure, the preamble field is set to the known sequence required for pilot signal reception and channel matrix acquisition, the network type field is set to the identification code that identifies the power line communication link and the low-power wireless communication link, and the bit error rate requirement factor field is set to the bit error rate parameter required for substream allocation ratio calculation. The output contains a unified data frame structure that includes a bit error rate requirement factor field.

[0007] The link parameters of the power line communication link and the low-power wireless communication link are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to the service type, and disturbance parameters are introduced to adjust the path search process. The output of the optimal communication path includes: Power line communication links and low-power wireless communication links are identified from the network type field of a unified data frame structure, and the delay, bit error rate, communication rate and routing hop count of each link are obtained as link parameters. Based on the business type, select the corresponding weight coefficient combination from the pre-configured weight coefficient library, and construct the path cost function based on the link parameters and the weight coefficient combination; Obtain the current number of network nodes and calculate the perturbation parameter. When the number of consecutive convergences of the path search reaches a preset convergence threshold, introduce the perturbation parameter to perform Brownian motion perturbation on the path cost function. The optimal next-hop node is searched based on the perturbed path cost function, and the optimal communication path is output. The optimal communication path is used for subsequent pilot signal reception and channel matrix acquisition.

[0008] The perturbation parameter is related to the network size, and its calculation formula is as follows: ; Where η is the perturbation parameter, and N is the number of network nodes, with N>1.

[0009] The formula for the path cost function is: ; in, Let be the path cost function, D be the delay, E be the bit error rate, B be the communication rate, and H be the number of hops. For the weighting coefficients corresponding to the delay, The weighting coefficients corresponding to the bit error rate. Weighting coefficients corresponding to communication rates This represents the weight coefficient corresponding to the number of route hops.

[0010] The optimal communication path, consisting of a power line communication link and a low-power wireless communication link, receives pilot signals to obtain an initial channel matrix. A global average pooling method is then applied to the initial channel matrix via a convolutional layer to obtain a compressed feature vector. This compressed feature vector is then subjected to dimensionality reduction and expansion processing via another convolutional layer to generate feature channel weights. The feature channel weights are then element-wise multiplied with the initial channel matrix, and a residual concatenation is performed between the feature channel weights and the initial channel matrix. The output channel state information includes: Extract the power line communication link and the low-power wireless communication link from the optimal communication path as the target channel link, and control the target channel link to receive pilot signals; The received pilot sequence is extracted from the preamble field of the unified data frame structure, the locally known pilot sequence is obtained, and the ratio of the received pilot sequence to the locally known pilot sequence is calculated using the least squares method to obtain the initial channel matrix. The initial channel matrix is ​​input into a deep residual network, and global average pooling is performed on the initial channel matrix to obtain a compressed feature vector. The compressed feature vector is then sequentially input into a dimensionality reduction convolutional layer and an dimensionality increase convolutional layer for dimensionality reduction and dimensionality increase processing to obtain an dimensionality increase feature vector. The upgraded feature vector is normalized and activated to generate feature channel weights. The feature channel weights are multiplied element-wise with the initial channel matrix to obtain a weighted channel matrix. The weighted channel matrix is ​​then residually concatenated with the initial matrix to output the channel state information. The channel state information includes the bit error rate of the power line communication link, the bit error rate of the low-power wireless communication link, the communication rate of the power line communication link, and the communication rate of the low-power wireless communication link.

[0011] The steps include obtaining the bit error rate (BER) requirement factor based on the BER requirement factor field in the unified data frame structure, dynamically adjusting the BER requirement factor using a gradient adjustment method based on channel state information, and calculating the substream allocation ratio based on the adjusted BER requirement factor and channel state information. The initial bit error rate requirement factor is extracted from the bit error rate requirement factor field of the unified data frame structure, and the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link are extracted from the channel state information. The error gradient value is calculated based on the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link. The initial bit error rate requirement factor is updated according to the preset adjustment step size and the error gradient value to obtain the adjusted bit error rate requirement factor. The power line communication link rate and the low-power wireless communication link rate are extracted from the channel state information. The proportion of power line communication sub-streams and the proportion of low-power wireless communication sub-streams are calculated based on the adjusted bit error rate requirement factor, the power line communication link rate, and the low-power wireless communication link rate. A smoothing factor is introduced to perform a weighted average of the sub-stream proportions at the current time and at historical times to obtain the sub-stream allocation ratio.

[0012] The step of splitting the business data into sub-streams according to the sub-stream allocation ratio and transmitting them respectively through the power line communication link and the low-power wireless communication link in the optimal communication path includes: The service data is split according to the proportion of power line communication sub-stream and low-power wireless communication sub-stream in the sub-stream allocation ratio, generating power line communication sub-stream data and low-power wireless communication sub-stream data; Power line communication substream data is encapsulated into a unified data frame structure and transmitted through the power line communication link in the optimal communication path; Low-power wireless communication sub-stream data is encapsulated into a unified data frame structure and transmitted through a low-power wireless communication link in the optimal communication path; At the receiving end, a reassembly buffer is set up to receive the power line communication sub-stream data and the low-power wireless communication sub-stream data respectively. When the amount of data in the reassembly buffer reaches a preset trigger threshold, a sub-stream synthesis operation is performed to synthesize the power line communication sub-stream data and the low-power wireless communication sub-stream data into complete service data.

[0013] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0014] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing computer program instructions, which, when executed by a processor, implement the steps in any of the aforementioned methods.

[0015] This invention achieves coordinated operation of power line communication and low-power wireless communication through a unified MAC layer protocol design, improving communication efficiency. A path selection mechanism incorporating perturbation parameters avoids local optima and enhances the robustness of routing. Deep learning is used to process the initial channel matrix, and feature extraction and residual connections improve the accuracy of channel state estimation. A gradient-adjusted dynamic bit error rate demand factor adapts to real-time network environment changes, optimizing substream allocation strategies. A multipath transmission mechanism fully utilizes dual-mode communication resources, increasing system throughput. Adaptive data splitting reduces service latency and improves transmission reliability. The reassembly buffer design ensures data integrity and enhances system fault tolerance. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a dual-mode communication method for the power Internet of Things based on deep learning for channel estimation, provided as an embodiment of the present invention; Figure 2 This is a schematic diagram of beacon period time slot division according to an embodiment of the present invention; Figure 3 This is a diagram of the channel estimation model architecture based on the convolutional attention mechanism in an embodiment of the present invention. Detailed Implementation

[0017] like Figure 1 As shown, Figure 1 A flowchart of a dual-mode communication method for the power Internet of Things based on deep learning for channel estimation, provided in an embodiment of the present invention, is included in the following steps: Design a unified MAC layer protocol, set a bit error rate requirement factor field in the unified data frame structure, and output a unified data frame structure containing the bit error rate requirement factor field. Link parameters of power line communication links and low-power wireless communication links are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to service type, and disturbance parameters are introduced to adjust the path search process, outputting the optimal communication path. In the optimal communication path, the power line communication link and the low-power wireless communication link receive pilot signals to obtain the initial channel matrix. The initial channel matrix is ​​then subjected to global average pooling through a convolutional layer to obtain a compressed feature vector. After dimensionality reduction and dimensionality increase processing of the compressed feature vector through a convolutional layer, feature channel weights are generated. The feature channel weights are then multiplied element-wise with the initial channel matrix and residually concatenated with the initial channel matrix to output channel state information. The bit error rate requirement factor is obtained from the bit error rate requirement factor field in the unified data frame structure. The bit error rate requirement factor is dynamically adjusted using a gradient adjustment method based on the channel state information. The sub-stream allocation ratio is calculated based on the adjusted bit error rate requirement factor and the channel state information. The business data is split into sub-streams according to the sub-stream allocation ratio and transmitted separately through the power line communication link and the low-power wireless communication link in the optimal communication path.

[0018] like Figure 2 The diagram illustrates the beacon period time slot division provided by an embodiment of the present invention.

[0019] The design of the unified MAC layer protocol includes setting a bit error rate requirement factor field in the unified data frame structure, and outputting a unified data frame structure containing the bit error rate requirement factor field, including: The beacon period is divided into contention-based and non-contention-based time slots. During the contention-based time slots, a carrier sense multiple access mechanism is used to control node access, while during the non-contention-based time slots, a time division multiple access mechanism is used to allocate transmission time slots. Construct an initial data frame structure that includes a preamble field, a network type field, and a bit error rate requirement factor field; In the initial data frame structure, the preamble field is set to the known sequence required for pilot signal reception and channel matrix acquisition, the network type field is set to the identification code that identifies the power line communication link and the low-power wireless communication link, and the bit error rate requirement factor field is set to the bit error rate parameter required for substream allocation ratio calculation. The output contains a unified data frame structure that includes a bit error rate requirement factor field.

[0020] The entire communication process is organized according to beacon periods, each with a length of 20ms. Contention slots occupy 30% of the beacon period (6ms), while non-contention slots occupy 70% (14ms). During contention slots, nodes compete for channel space using a carrier sense multiple access (CSMA) mechanism. A node can transmit data when the channel is detected as idle; if the channel is detected as busy, it randomly backs off for a period before attempting again. The backoff time range is set from 0 to 3ms, with the specific backoff time value determined by a pseudo-random number generator. During non-contention slots, fixed transmission slots are allocated to each node according to a time division multiple access (TDMA) mechanism. The length of each slot is dynamically adjusted based on the number of nodes. When there are 10 active nodes in the network, each node is allocated an average of 1.4ms of transmission slot space.

[0021] When constructing a unified data frame structure, the initial data frame contains three key fields: a preamble field, a network type field, and a bit error rate (BER) requirement factor field. The preamble field is 32 bits long and consists of alternating sequences of 0s and 1s, in the form of "01010101...01", used for receiver synchronization and channel estimation. The network type field is 8 bits long, with the identification code for power line communication links set to "10101010" and the identification code for low-power wireless communication links set to "01010101". This field enables the identification and management of different physical layer communication technologies. The BER requirement factor field is 16 bits long and is used to identify the transmission quality requirements of the service data. Its value ranges from 0 to 65535, with smaller values ​​indicating stricter BER requirements.

[0022] In the use of the bit error rate requirement factor field, values ​​from 0 to 100 are defined as the high reliability service range, suitable for critical data transmission such as control commands and protection information; 101 to 1000 are defined as the medium reliability service range, suitable for general services such as measurement data and status information; and 1001 to 65535 are defined as the low reliability service range, suitable for non-critical data transmission. When a node needs to send data, it selects the appropriate bit error rate requirement factor value according to the service type and fills it into the corresponding field of the unified data frame.

[0023] The specific implementation of the sub-stream allocation ratio calculation process based on the bit error rate requirement factor is as follows: When the bit error rate of the current power line communication link is detected to be 0.01 and the bit error rate of the low-power wireless communication link is 0.001, for high-reliability services with a bit error rate requirement factor of 50, the sub-stream allocation ratio between the power line communication link and the low-power wireless communication link is calculated to be 1:9; for medium-reliability services with a bit error rate requirement factor of 500, the calculated sub-stream allocation ratio is 3:7; and for low-reliability services with a bit error rate requirement factor of 5000, the calculated sub-stream allocation ratio is 6:4.

[0024] In addition to the three fields mentioned above, the complete unified data frame structure also includes a frame control field, a sequence number field, an address field, a data payload field, and a frame check field. The frame control field is 8 bits long and is used to identify frame type, security settings, and other information. The sequence number field is also 8 bits long and is used for frame order control and duplicate frame detection. The address field includes source and destination addresses, each 16 bits long, supporting up to 65,536 nodes in the network. The data payload field has a variable length, supporting a maximum of 2048 bytes of service data. The frame check field uses a 32-bit cyclic redundancy check (CRC) code with a generator polynomial of 0x10⁴C11DB7, used to detect data errors during transmission.

[0025] In actual communication, the sending node sets the bit error rate requirement factor value according to the service type, constructs a unified data frame, and then distributes the data to the power line communication link and the low-power wireless communication link for transmission according to the calculated sub-stream allocation ratio. The receiving node simultaneously listens to both communication links, merges the received data, and restores the original information.

[0026] This method can maintain stable communication under different electromagnetic environments. When the power line communication link is subjected to strong interference and the bit error rate rises to 0.1, the high-reliability service can still maintain an overall bit error rate of less than 0.001 by adjusting the sub-stream allocation ratio to 0.5:9.5. When the low-power wireless communication link is shielded and the bit error rate rises to 0.2, the service continuity can still be maintained by adjusting the sub-stream allocation ratio to 9:1.

[0027] This invention improves the reliability and stability of power line communication (PIoT) by introducing a bit error rate (BER) requirement factor field into a unified MAC layer protocol, thereby enabling collaborative management of PIoT and low-power wireless communication. The invention can dynamically adjust communication resource allocation based on the transmission quality requirements of different services, effectively addressing communication challenges in complex electromagnetic environments. The unified data frame structure simplifies the management complexity of heterogeneous networks, reduces the difficulty of inter-device collaborative communication, and is suitable for various PIoT application scenarios.

[0028] The link parameters of the power line communication link and the low-power wireless communication link are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to the service type, and disturbance parameters are introduced to adjust the path search process. The output of the optimal communication path includes: Power line communication links and low-power wireless communication links are identified from the network type field of a unified data frame structure, and the delay, bit error rate, communication rate and routing hop count of each link are obtained as link parameters. Based on the business type, select the corresponding weight coefficient combination from the pre-configured weight coefficient library, and construct the path cost function based on the link parameters and the weight coefficient combination; Obtain the current number of network nodes and calculate the perturbation parameter. When the number of consecutive convergences of the path search reaches a preset convergence threshold, introduce the perturbation parameter to perform Brownian motion perturbation on the path cost function. The optimal next-hop node is searched based on the perturbed path cost function, and the optimal communication path is output. The optimal communication path is used for subsequent pilot signal reception and channel matrix acquisition.

[0029] After receiving a unified data frame, the device reads the 8-bit network type field in the data frame to identify the currently available communication link type. When the network type field value is "10101010", it is identified as a power line communication link; when the value is "01010101", it is identified as a low-power wireless communication link; when both types of links exist simultaneously, the network type field value is "11001100". After link identification is completed, key parameters of each link are obtained by sending test data frames, including latency, bit error rate, communication rate, and hop count. The latency parameter is obtained by calculating the time difference between sending the test frame and receiving the acknowledgment frame, in milliseconds; the bit error rate parameter is calculated by continuously sending 100 test frames and counting the proportion of received errors; the communication rate parameter is obtained by measuring the amount of data successfully transmitted per unit time, in kbps; and the hop count parameter is determined by the decrementing value of the time-to-live field in the data frame.

[0030] After obtaining the link parameters, the corresponding weight coefficient combination is selected from a pre-configured weight coefficient library based on the service type. Different weight coefficient combinations are set for different service types in the library. For control services, the latency weight is 0.5, the bit error rate weight is 0.3, the communication rate weight is 0.1, and the hop count weight is 0.1; for monitoring services, the latency weight is 0.2, the bit error rate weight is 0.2, the communication rate weight is 0.5, and the hop count weight is 0.1; for ordinary data services, the latency weight is 0.1, the bit error rate weight is 0.2, the communication rate weight is 0.3, and the hop count weight is 0.4. After selecting the weight coefficient combination, a path cost function is constructed. Specifically, the link parameter values ​​are normalized to a range between 0 and 1, then multiplied by the corresponding weight coefficient and summed to obtain the comprehensive path cost value. The smaller this value, the better the path quality.

[0031] To avoid getting trapped in local optima during path search, a perturbation parameter adjustment mechanism is introduced. The perturbation parameter is calculated as follows: The number of nodes in the current network is obtained, denoted as N. A convergence threshold of 5 iterations is set; that is, when the path search algorithm selects the same path 5 times consecutively, it is considered to be in a convergent state, at which point the perturbation mechanism is triggered. The perturbation is implemented by applying Brownian motion perturbation to the path cost function, adding random fluctuations to the original path cost. The fluctuation amplitude is controlled by the perturbation parameter. For a path cost value of 0.45, after introducing a perturbation parameter of 0.256, the path cost value may become 0.396 or 0.504, with the randomness determined by the Brownian motion characteristics.

[0032] The perturbated path cost function is used to search for the optimal next-hop node. The path search employs a greedy strategy, selecting the neighbor node with the smallest current cost function value as the next hop each time, until the destination node is reached. Considering that some nodes may support both power line communication and low-power wireless communication, the selection of the next hop must consider not only the node selection but also the communication link selection. When there are two communication link options from node A to node B, the path costs of the two links are calculated separately, and the link with the lower cost is selected as the communication path. For example, in a test environment, the parameters of the power line communication link from node A to node B are: delay 15ms, bit error rate 0.02, communication rate 1200kbps, and hop count 2; the parameters of the low-power wireless communication link are: delay 8ms, bit error rate 0.01, communication rate 800kbps, and hop count 1. For control-type services, the calculated path cost of the power line communication link is 0.42, and the path cost of the low-power wireless communication link is 0.35. Therefore, the low-power wireless communication link is selected as the next-hop communication path.

[0033] After path search is completed, the optimal communication path is output, including complete path information from the source node to the destination node and the communication link type used for each hop. This path information is used for subsequent pilot signal reception and channel matrix acquisition. During pilot signal transmission, the transmitting end sends a specific sequence of pilot signals according to the determined optimal communication path, and the receiving end estimates the channel state based on the received pilot signals to generate the channel matrix. The channel matrix contains various characteristic parameters of the communication link, providing a basis for the selection of channel coding and modulation methods for subsequent data transmission.

[0034] As the communication environment changes or node states shift, the path search process needs to be periodically re-executed to update the optimal communication path. The periodic path update interval is set to 30 seconds, and a trigger-based update mechanism is implemented. When a sharp deterioration in link quality is detected, a path reselection is immediately triggered. The criteria for determining link quality deterioration are a 50% increase in bit error rate or a 100% increase in latency. This dynamic path optimization mechanism ensures that the communication path remains in an optimal state, improving overall communication efficiency and reliability.

[0035] This invention achieves coordinated optimization of power line communication and low-power wireless communication through three key steps: link parameter acquisition, weight coefficient matching, and perturbation parameter adjustment. It fully considers the differentiated communication quality requirements of different service types and can dynamically adjust the path selection strategy based on real-time network conditions, effectively avoiding the problem of traditional path selection algorithms easily getting trapped in local optima. By introducing perturbation parameters and Brownian motion mechanisms, the algorithm's ability to explore new paths is enhanced, improving its adaptability in complex power Internet of Things (IoT) environments.

[0036] The perturbation parameter is related to the network size, and its calculation formula is as follows: ; Where η is the perturbation parameter, and N is the number of network nodes, with N>1.

[0037] Considering In this embodiment, the number of network nodes N is limited to be greater than 1. When N=1 (i.e., there is only a single node), there is no routing optimization problem, and the system can automatically skip this step or use the default path. This formula shows that the larger the network size N is, the finer the perturbation step size is required, thereby effectively escaping local extrema while ensuring the convergence speed.

[0038] The formula for the path cost function is: ; in, Let be the path cost function, D be the delay, E be the bit error rate, B be the communication rate, and H be the number of hops. For the weighting coefficients corresponding to the delay, The weighting coefficients corresponding to the bit error rate. Weighting coefficients corresponding to communication rates This represents the weight coefficient corresponding to the number of route hops.

[0039] The optimal communication path, consisting of a power line communication link and a low-power wireless communication link, receives pilot signals to obtain an initial channel matrix. A global average pooling method is then applied to the initial channel matrix via a convolutional layer to obtain a compressed feature vector. This compressed feature vector is then subjected to dimensionality reduction and expansion processing via another convolutional layer to generate feature channel weights. The feature channel weights are then element-wise multiplied with the initial channel matrix, and a residual concatenation is performed between the feature channel weights and the initial channel matrix. The output channel state information includes: Extract the power line communication link and the low-power wireless communication link from the optimal communication path as the target channel link, and control the target channel link to receive pilot signals; The received pilot sequence is extracted from the preamble field of the unified data frame structure, the locally known pilot sequence is obtained, and the ratio of the received pilot sequence to the locally known pilot sequence is calculated using the least squares method to obtain the initial channel matrix. The initial channel matrix is ​​input into a deep residual network, and global average pooling is performed on the initial channel matrix to obtain a compressed feature vector. The compressed feature vector is then sequentially input into a dimensionality reduction convolutional layer and an dimensionality increase convolutional layer for dimensionality reduction and dimensionality increase processing to obtain an dimensionality increase feature vector. The upgraded feature vector is normalized and activated to generate feature channel weights. The feature channel weights are multiplied element-wise with the initial channel matrix to obtain a weighted channel matrix. The weighted channel matrix is ​​then residually concatenated with the initial matrix to output the channel state information. The channel state information includes the bit error rate of the power line communication link, the bit error rate of the low-power wireless communication link, the communication rate of the power line communication link, and the communication rate of the low-power wireless communication link.

[0040] The power line communication link and the low-power wireless communication link are extracted from the communication path as target channel links. These links are controlled to receive specific pilot signals. The pilot signals are generated by the transmitter and transmitted to the receiver through a preset path. The pilot sequence adopts the Zadoff-Chu sequence, which has good autocorrelation and constant amplitude characteristics, and the sequence length is set to 64. The receiver extracts the received pilot sequence from the preamble field of the unified data frame structure. The preamble is located at the beginning of the data frame, occupying 32 bytes of space, of which the pilot sequence occupies the first 16 bytes. At the same time, the receiver has pre-stored locally known pilot sequences as a reference.

[0041] The receiver obtains the initial channel matrix by comparing the received pilot sequence with the locally known pilot sequence, and calculates the ratio of the received sequence to the known sequence using the least squares method. During processing, the received pilot sequence is represented as a 64-dimensional complex vector, and the locally known pilot sequence is also represented as a 64-dimensional complex vector. The ratio of the real part to the imaginary part is calculated using the least squares method to obtain the initial channel response estimate. This process can be viewed as solving a system of linear equations to minimize the sum of squared estimation errors. For a received signal amplitude of 0.85 and a phase offset of 30 degrees, the least squares estimation yields a channel coefficient amplitude of approximately 0.85 and a phase of approximately 30 degrees, which basically reflects the channel's attenuation and phase change characteristics. The resulting initial channel matrix has a dimension of 8×8, where each element is a complex number containing amplitude and phase information, reflecting the channel's transmission characteristics at different frequencies and time points.

[0042] After obtaining the initial channel matrix, to improve the accuracy of channel estimation, it is processed by a deep residual structure. The deep residual structure includes a main path and skip paths. The main path is responsible for channel feature extraction and enhancement, while the skip paths retain the original channel information to prevent information loss. At the beginning of processing, features are extracted from the 8×8 initial channel matrix using convolutional layers, converting it into 64 feature channels, each with a 1×1 feature response. Then, global average pooling is performed on these 64 feature channels, compressing each feature channel into a scalar value, ultimately resulting in a 1×64 compressed feature vector. Global average pooling reduces dimensionality by calculating the average value of each feature channel, preserving the global statistical characteristics of the channel while reducing computational complexity. The compressed feature vector contains an overview of the channel but loses some detailed information.

[0043] The compressed feature vector is then fed into a dimension reduction convolutional layer, which uses 16 1×1 convolutional kernels to reduce the feature dimension from 64 to 16. The kernel weights are obtained through pre-training. Dimension reduction filters out redundant information, highlights key features, and reduces the computational burden of subsequent processing. The dimension-reduced feature vector is then fed into a dimension-up convolutional layer, which uses 64 1×1 convolutional kernels to restore the feature dimension from 16 to 64. Dimension-up convolution reconstructs the feature space, generates richer feature representations, and improves the accuracy of subsequent feature channel weighting. In practical applications, for a channel feature at a certain frequency, the original value is 0.6 + 0.3j. After dimension reduction convolution, it becomes 0.2 + 0.1j, and after dimension-up convolution, it becomes 0.58 + 0.29j. The values ​​change slightly, but the main characteristics are preserved.

[0044] After normalization, the upgraded feature vectors undergo a nonlinear transformation using the Sigmoid activation function, with the output range controlled between 0 and 1, generating feature channel weights. Batch normalization is employed to stabilize the feature distribution, which is beneficial for subsequent activation functions. The generated feature channel weights are used to adjust the importance of each element in the initial channel matrix; a larger weight indicates a greater contribution of the feature to channel estimation. The feature channel weights are then multiplied element-wise with the initial channel matrix to obtain a weighted channel matrix. In this process, channel elements with stronger signals receive higher weights, while those with weaker signals receive lower weights, thus highlighting reliable information and suppressing unreliable information. For frequencies with low signal-to-noise ratios (SNR), the feature channel weight may be only 0.3, while for frequencies with high SNR, the feature channel weight can reach 0.9, effectively distinguishing channel information of different qualities.

[0045] The weighted channel matrix and the initial channel matrix are combined using a residual concatenation method. Residual concatenation adds corresponding elements of the weighted channel matrix and the initial channel matrix to generate the final channel state information output. The introduction of residual concatenation effectively avoids the gradient vanishing problem in deep processing while preserving the initial channel information, thus improving the robustness of channel estimation. The final output channel state information still maintains an 8×8 matrix format, but its accuracy and reliability are significantly improved. Several key parameters can be extracted from the channel state information, including the bit error rate (BER) of the power line communication link, the BER of the low-power wireless communication link, the communication rate of the power line communication link, and the communication rate of the low-power wireless communication link. These parameters are calculated through post-processing of the channel matrix. The BER is determined by the mapping relationship between the signal-to-noise ratio (SNR) and the modulation scheme, and the communication rate is calculated using the channel capacity formula combined with the available bandwidth. Taking a certain test scenario as an example, after processing with a deep residual structure, the estimated bit error rate of the power line communication link is 0.015, the estimated bit error rate of the low-power wireless communication link is 0.008, the estimated communication rate of the power line communication link is 1350kbps, and the estimated communication rate of the low-power wireless communication link is 870kbps.

[0046] The acquired channel state information can be directly used for communication parameter optimization, including modulation scheme selection, coding rate determination, and power allocation. When the bit error rate of the power line communication link exceeds the threshold of 0.05, the modulation order can be reduced or coding redundancy can be increased; when the communication rate of the low-power wireless communication link is lower than 500 kbps, switching to the power line communication link can be considered. This adaptive adjustment mechanism for communication parameters based on accurate channel state information significantly improves the stability and efficiency of power Internet of Things (IoT) communication.

[0047] This invention addresses the issue of insufficient accuracy in traditional channel estimation under complex power IoT environments by employing a deep learning-based channel estimation method. It intelligently enhances the initial channel matrix through a deep residual structure, achieving more accurate channel state awareness. This overcomes the challenges of channel estimation in different environments for power line communication and low-power wireless communication, providing a unified channel processing framework. A feature extraction method combining global average pooling and convolutional layers fully leverages the correlation of channel features, reducing the impact of noise and interference. A feature channel weighting mechanism enables adaptive adjustment of channel information of varying quality, while residual connections ensure the preservation of original channel information.

[0048] The steps include obtaining the bit error rate (BER) requirement factor based on the BER requirement factor field in the unified data frame structure, dynamically adjusting the BER requirement factor using a gradient adjustment method based on channel state information, and calculating the substream allocation ratio based on the adjusted BER requirement factor and channel state information. The initial bit error rate requirement factor is extracted from the bit error rate requirement factor field of the unified data frame structure, and the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link are extracted from the channel state information. The error gradient value is calculated based on the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link. The initial bit error rate requirement factor is updated according to the preset adjustment step size and the error gradient value to obtain the adjusted bit error rate requirement factor. The power line communication link rate and the low-power wireless communication link rate are extracted from the channel state information. The proportion of power line communication sub-streams and the proportion of low-power wireless communication sub-streams are calculated based on the adjusted bit error rate requirement factor, the power line communication link rate, and the low-power wireless communication link rate. A smoothing factor is introduced to perform a weighted average of the sub-stream proportions at the current time and at historical times to obtain the sub-stream allocation ratio.

[0049] In the unified data frame structure, the bit error rate requirement factor field is located in the 12th to 13th bytes of the data frame header, occupying 16 bits. This field uses the IEEE 754 standard half-precision floating-point representation, with a value ranging from 0 to 1. A smaller bit error rate requirement factor value indicates a higher requirement for communication quality; a larger value indicates a lower requirement for communication quality. Dual-mode communication equipment extracts this field value from the received data frame as the initial bit error rate requirement factor, denoted as μ. ini A typical value is 0.3. This value reflects the tolerance of the current service for the communication bit error rate. For control services, μ ini It is usually set to 0.1, for monitoring-related services μ ini It is usually set to 0.3 for ordinary data-related services. ini It is usually set to 0.5.

[0050] After extracting the initial bit error rate (BER) requirement factor, it is necessary to extract the power line communication link BER and the low-power wireless communication link BER from the acquired channel state information. The power line communication link BER is denoted as BER. plc The bit error rate of a low-power wireless communication link is denoted as BER. rfThese two parameters are important indicators for evaluating the quality of each communication link. In a certain test environment, the channel state information shows that the bit error rate (BER) of the power line communication link is 0.018, and the BER of the low-power wireless communication link is 0.009. The communication equipment calculates the error gradient value G based on the relationship between these two BER values ​​and the initial BER requirement factor. The calculation process of the error gradient value G involves comparing the BER of the two links with the initial BER requirement factor. When the BER of both links is greater than the initial BER requirement factor, it indicates that the current communication quality cannot meet the service requirements, and the BER requirement needs to be reduced; G is positive. When the BER of both links is less than the initial BER requirement factor, it indicates that the current communication quality exceeds the service requirements, and the BER requirement can be increased to improve communication efficiency; G is negative. When the BER of one link is greater than the initial BER requirement factor, and the BER of the other link is less than the initial BER requirement factor, G is zero, and the BER requirement factor remains unchanged.

[0051] In the aforementioned test environment, the initial bit error rate (BER) requirement factor was 0.3, which is greater than the BERs of the two links, 0.018 and 0.009, respectively. Therefore, G was set to -0.05. The magnitude of G reflects the required adjustment range of the BER; a larger absolute value indicates a stronger adjustment requirement. The range of G is typically controlled between -0.1 and 0.1 to avoid drastic fluctuations in the BER requirement factor. After the error gradient value is determined, the initial BER requirement factor is updated using a preset adjustment step size α. α is a parameter controlling the adjustment speed of the BER requirement factor, ranging from 0 to 1. A larger α value results in a faster adjustment speed but may lead to system instability; a smaller α value results in a slower adjustment speed and higher system stability. In the power IoT scenario, α is typically set to 0.2 and can be dynamically adjusted according to network conditions. The update process involves adding the initial BER requirement factor to the product of the error gradient value and the adjustment step size to obtain the adjusted BER requirement factor μ. adj .

[0052] In the example environment, μ ini Given 0.3, G = -0.05, and α = 0.2, μ is calculated. adj The value is 0.29. This gradient-based dynamic adjustment method allows the bit error rate demand factor to adapt to changes in the actual communication environment, improving the flexibility and accuracy of communication parameter configuration. Typically, to prevent over-adjustment of the bit error rate demand factor, upper and lower thresholds are set, with an upper limit of 0.8 and a lower limit of 0.1, to ensure μ... adj Always within a reasonable range.

[0053] Before calculating the sub-stream allocation ratio, the communication rates of the power line communication link and the low-power wireless communication link are extracted from the channel state information. Communication rate is a crucial indicator of link transmission capacity and directly impacts data transmission efficiency. The power line communication link rate is denoted as R.plc The unit is kbps; the communication rate of a low-power wireless communication link is denoted as R. rf The unit is kbps. In the aforementioned test environment, R plc For 1350kbps, R rf The speed is 870 kbps. The communication equipment calculates the proportion of power line communication sub-streams and low-power wireless communication sub-streams based on the adjusted bit error rate requirement factor, the communication rates of the two links, and the bit error rate. The calculation process comprehensively considers both bit error rate and communication rate factors, reflecting a trade-off between reliability and efficiency.

[0054] In the example environment, μ adj The BER is 0.29. plc BER is 0.018. rf R is 0.009. plc For 1350kbps, R rf The bandwidth is 870 kbps. Calculations show that the power line communication substream proportion is 0.58, and the low-power wireless communication substream proportion is 0.42. This result indicates that under the current channel conditions, 58% of the data stream should be transmitted via the power line communication link, and 42% via the low-power wireless communication link. The higher allocation ratio of the power line communication link is mainly due to its communication rate advantage, despite its slightly higher bit error rate. To avoid communication instability caused by frequent changes in the substream allocation ratio, a smoothing factor is introduced to perform a weighted average of the substream proportions at the current and historical times. The smoothing factor β ranges from 0 to 1. The larger the β, the greater the weight of the current substream proportion, resulting in a more agile system response but reduced stability; the smaller the β, the greater the weight of the historical substream proportion, resulting in a slower system response but improved stability.

[0055] In power Internet of Things (IoT) applications, β is typically set to 0.3, reflecting the emphasis on stability. During smoothing, a weighted average is calculated between the currently calculated sub-stream proportion and the sub-stream allocation proportion from the previous time step. If the current calculated power line communication sub-stream proportion is 0.58, the previous time step's power line communication sub-stream allocation proportion is 0.52, and β is 0.3, then the smoothed power line communication sub-stream allocation proportion is 0.538. Similarly, the low-power wireless communication sub-stream allocation proportion is 0.462. This smoothing mechanism effectively reduces sub-stream allocation oscillations and improves communication stability. After the sub-stream allocation proportion is determined, the communication device divides the data to be transmitted into two parts according to this proportion, transmitting them separately through the power line communication link and the low-power wireless communication link. To ensure orderly data reception and reassembly, sequence numbers and timestamp information are added during data segmentation.

[0056] The entire substream allocation process is a closed-loop control mechanism. Communication equipment continuously adjusts the bit error rate (BER) requirement factor and substream allocation ratio based on changes in channel conditions. When significant changes occur in the network environment, such as a sharp increase in BER due to strong interference on power line communication links, the system quickly reduces the proportion of power line communication substreams, diverting more data to low-power wireless communication links to ensure communication quality. In extreme cases, such as a complete link failure, the system adjusts the substream allocation ratio to 0 and 1, ensuring all data is transmitted through available links.

[0057] This invention, based on gradient-adjusted dynamic optimization of the bit error rate requirement factor and substream allocation technology, realizes intelligent data stream allocation in dual-mode communication for the power Internet of Things (IoT), effectively solving the problem that traditional fixed-ratio allocation schemes are difficult to adapt to complex and variable network environments. It dynamically adjusts the bit error rate requirement through error gradient calculation, adaptively allocates the substream ratio according to real-time channel conditions, and employs a smoothing factor mechanism to reduce allocation oscillations. This allocation method fully utilizes the complementary advantages of power line communication and low-power wireless communication, improving transmission efficiency while ensuring communication reliability.

[0058] The step of splitting the business data into sub-streams according to the sub-stream allocation ratio and transmitting them respectively through the power line communication link and the low-power wireless communication link in the optimal communication path includes: The service data is split according to the proportion of power line communication sub-stream and low-power wireless communication sub-stream in the sub-stream allocation ratio, generating power line communication sub-stream data and low-power wireless communication sub-stream data; Power line communication substream data is encapsulated into a unified data frame structure and transmitted through the power line communication link in the optimal communication path; Low-power wireless communication sub-stream data is encapsulated into a unified data frame structure and transmitted through a low-power wireless communication link in the optimal communication path; At the receiving end, a reassembly buffer is set up to receive the power line communication sub-stream data and the low-power wireless communication sub-stream data respectively. When the amount of data in the reassembly buffer reaches a preset trigger threshold, a sub-stream synthesis operation is performed to synthesize the power line communication sub-stream data and the low-power wireless communication sub-stream data into complete service data.

[0059] In dual-mode communication for the power Internet of Things, service data is split and transmitted according to the sub-stream allocation ratio. The proportion of the power line communication sub-stream is denoted as β. plc The proportion of low-power wireless communication sub-streams is denoted as β. rf The sum of the two is 1. For example, if the size of the service data to be transmitted is 5MB, the calculated proportion of the power line communication sub-stream is 0.65, and the proportion of the low-power wireless communication sub-stream is 0.35, then the size of the power line communication sub-stream data is 3.25MB, and the size of the low-power wireless communication sub-stream data is 1.75MB.

[0060] Business data splitting must consider the inherent structure and integrity of the data. For structured data, such as database tables and XML documents, the integrity of data elements should be maintained during splitting. For unstructured data, such as video streams and audio streams, splitting should be done according to data frames or data blocks. Dual-mode communication equipment maintains a data splitting buffer. After business data enters the buffer, it is divided according to the sub-stream allocation ratio. To ensure that the sub-stream data can be correctly reassembled at the receiving end, a sequence number and a timestamp are added to each data segment during splitting. The sequence number is 2 bytes long, represented by an unsigned integer, with a value range of 0 to 65535; the timestamp is 4 bytes long, using millisecond precision, representing the absolute time when the data was generated.

[0061] After the business data is split, the sub-stream data needs to be encapsulated into a unified data frame structure, which consists of three parts: a frame header, payload data, and a frame trailer. The frame header is 16 bytes long and includes fields such as frame start flag, frame length, source address, destination address, frame type, bit error rate requirement factor, sub-stream identifier, and fragment identifier. The sub-stream identifier field occupies 1 byte, with a value of 0x01 indicating a power line communication sub-stream and a value of 0x02 indicating a low-power wireless communication sub-stream. The fragment identifier field occupies 1 byte and indicates whether the current frame is part of a fragmented transmission and its position in the fragment sequence. The frame trailer is 4 bytes long and includes a 32-bit CRC checksum, used by the receiver to verify data integrity. The payload data portion has a variable length, with its maximum value limited by the maximum transmission unit (MTU) of 1500 bytes for power line communication and 128 bytes for low-power wireless communication.

[0062] During encapsulation, the total data size of the power line communication substream is 3.25MB, requiring transmission in approximately 2280 frames, with each frame containing 1500 bytes of data. The total data size of the low-power wireless communication substream is 1.75MB, requiring transmission in approximately 14350 frames, with each frame containing 128 bytes of data. The encapsulated data frames are encoded using FEC to enhance transmission reliability. The power line communication substream uses RS(255, 239) encoding, adding 16 bytes of checksum for every 239 bytes of data; the low-power wireless communication substream uses RS(255, 223) encoding, adding 32 bytes of checksum for every 223 bytes of data. The encoding parameters can be dynamically adjusted according to actual channel conditions; redundancy can be reduced when the channel conditions are good, and increased when the channel conditions are poor.

[0063] The encapsulated data frames are transmitted through corresponding communication links. The power line communication link uses orthogonal frequency division multiplexing (OFDM) technology, operating at a frequency range of 2MHz to 12MHz, with a maximum transmission rate of 200Mbps. The low-power wireless communication link uses spread spectrum technology, operating at a frequency of 470MHz, with a maximum transmission rate of 100kbps. The transmitting device hands over the power line communication sub-stream data frames to the power line communication module for processing, and the low-power wireless communication sub-stream data frames to the low-power wireless communication module for processing. Each communication module selects an appropriate modulation / demodulation method and transmission parameters based on the channel conditions. Power line communication can use modulation methods such as QPSK, 16QAM, and 64QAM, while low-power wireless communication can use modulation methods such as FSK and GFSK.

[0064] Two communication links operate in parallel, transmitting their respective sub-streams simultaneously. To reduce transmission latency, an interleaved transmission strategy can be employed, alternating the transmission of large data blocks between the two sub-streams to prevent a single large data block from occupying a single link for an excessively long time. In practical applications, the interleaving threshold is set to 100KB. When the data block to be transmitted exceeds this threshold, the interleaved transmission mechanism is activated, and the data block is divided into multiple 10KB smaller blocks, distributed among the two sub-streams according to the sub-stream allocation ratio. This strategy significantly reduces the tail latency problem caused by transmitting large data blocks on a single link.

[0065] After receiving a data frame, the receiving device first performs a CRC check to verify data integrity. Data frames that fail the check are retransmitted by the sending end. Successful data frames are then placed into either the power line communication sub-stream reassembly buffer or the low-power wireless communication sub-stream reassembly buffer, based on the sub-stream identifier field. The reassembly buffer employs a dual-buffering mechanism, with buffers A and B used alternately. While buffer A receives data, data in buffer B is processed. Each buffer is 10MB in size, and the data in the buffers is sorted by sequence number, awaiting sub-stream reassembly.

[0066] The substream merging operation is triggered when the amount of data in the reassembly buffer reaches a preset trigger threshold or when a timeout condition is met. The preset trigger threshold is typically set to 80% of the original service data size, and the timeout period is set to 1000ms. The substream merging operation is triggered when the amount of data in the reassembly buffer reaches 80% of 5MB (4MB), or 1000ms after the arrival of the first data frame. The merging operation first checks the integrity of the data in the buffer. If there are discontinuous sequence numbers, it indicates that some data is lost and a retransmission request is needed. After confirming the data integrity, the power line communication substream data and the low-power wireless communication substream data are merged in the correct order according to the sequence number and timestamp to restore the original service data.

[0067] Data redundancy may be encountered during the data merging process, meaning the same data segment may exist in both sub-streams. To handle redundant data, the data segment arriving earlier is selected based on arrival time, and duplicate segments arriving later are discarded. If the content of the same data segment in the two sub-streams is inconsistent, the data segment that passes the CRC check is selected; if both data segments pass the check, the data segment transmitted via the more reliable link is selected based on the historical reliability of the transmission link. After the sub-streams are merged, the recovered service data is passed to the application layer for processing.

[0068] This invention, based on a substream allocation-based dual-mode communication transmission and synthesis mechanism for the power Internet of Things, fully leverages the complementary advantages of power line communication and low-power wireless communication, achieving load balancing and fault tolerance in data transmission. By intelligently splitting service data and utilizing dual links in parallel, communication efficiency and reliability are significantly improved. The substream splitting process considers the inherent structure of the data to ensure data integrity; the encapsulation process adopts a unified data frame structure and adds necessary control information to facilitate reassembly at the receiving end; the transmission process dynamically selects modulation parameters based on channel conditions to optimize transmission performance; and the reassembly process employs a double-buffering mechanism for efficient processing of received data.

[0069] In another embodiment of the invention, a deep residual network based on a convolutional attention mechanism is used to perform refined channel estimation for PLC links (and wireless links supporting multi-carrier modulation). Figure 3 As shown, the specific workflow of this model is as follows: The initial channel estimation matrix is ​​obtained using the least squares (LS) method. Since LS estimation is sensitive to noise, its results contain a significant amount of noise interference. The LS estimation formula is as follows: ; in, The received pilot signal, For locally known pilot signals, This represents the noisy channel response. The real and imaginary parts of the complex matrix are separated and constructed as a three-dimensional tensor, which is then used as input to the deep learning model.

[0070] Input data enters the attention-convolutional network. The network consists of multiple cascaded residual blocks with attention mechanisms. To accurately extract channel features in low signal-to-noise ratio environments, this embodiment designs a convolutional attention module in the residual block. This module compresses the spatial dimension of the feature map to 1×1 through global average pooling, obtaining the global receptive field. To reduce the number of parameters and improve computational efficiency, this invention uses 1×1 convolutional layers instead of traditional fully connected layers to learn the correlation between channels. Specifically, channel weights are generated through two 1×1 convolutional layers. s The calculation formula is: Where z is the compressed feature vector, W1 represents the dimension-reducing convolution kernel. W 2 represents the increased-dimensional convolution kernel, δ is the ReLU activation function, and σ is the Sigmoid activation function. The generated weight coefficients s are multiplied back into the original feature map. This mechanism can automatically assign high weights to feature channels carrying effective channel information while suppressing feature channels containing impulse noise.

[0071] After attention-weighted processing and multi-layer residual processing, upsampling is performed using bilinear interpolation to restore the original subcarrier dimension, outputting the denoised and accurate channel matrix. .

[0072] By employing an attention-convolution model, this system can obtain high-precision channel state information in power line environments with extremely low signal-to-noise ratios (compared to fully connected networks) with minimal computational cost. These precise channel parameters are fed back in real-time to the formula in step S4 for calculation. and This ensures that the adjustment of the bit error rate requirement factor I is based on real and reliable channel quality, ultimately achieving highly reliable fusion of dual-mode communication.

[0073] Data is split into sub-streams and distributed to the PLC and wireless communication network according to the business layer requirements. To achieve load balancing, this invention proposes a dynamic load balancing strategy based on the bit error rate requirement factor (I).

[0074] The formulas for PLC sub-stream percentage (CX) and wireless sub-stream percentage (CY) are as follows: ; Where, is the PLC communication rate. This represents the wireless communication rate. The value of factor I is not fixed but dynamically adjusted using gradient descent. Where Inew represents the updated bit error rate requirement factor, and Iold represents the bit error rate requirement factor from the previous step.

[0075] In this embodiment, the adjustment step size ΔI is limited to 0.05. Simultaneously, to prevent network jitter caused by abrupt changes in the substream ratio, a smoothing factor is introduced when calculating the final ratio. =0.1.

[0076] At the receiving end, a reassembly buffer is set up. When the amount of data in the buffer reaches 60% of the total capacity (half full threshold), the synchronous synthesis of the sub-stream is triggered to overcome the jitter problem caused by the difference in transmission delay between the two-mode links.

[0077] One technical solution provided in this embodiment of the invention is an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0078] One technical solution provided in this embodiment of the invention is a computer-readable storage medium storing a computer program, wherein the processor executes the computer program to implement the steps in any of the aforementioned methods.

[0079] The specific embodiments described above are preferred embodiments of the present invention and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A dual-mode communication method for the power Internet of Things based on deep learning for channel estimation, characterized in that, Includes the following steps: Design a unified MAC layer protocol, set a bit error rate requirement factor field in the unified data frame structure, and output a unified data frame structure containing the bit error rate requirement factor field. Link parameters of power line communication links and low-power wireless communication links are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to service type, and disturbance parameters are introduced to adjust the path search process, outputting the optimal communication path. In the optimal communication path, the power line communication link and the low-power wireless communication link receive pilot signals to obtain the initial channel matrix. The initial channel matrix is ​​then subjected to global average pooling through a convolutional layer to obtain a compressed feature vector. After dimensionality reduction and dimensionality increase processing of the compressed feature vector through a convolutional layer, feature channel weights are generated. The feature channel weights are then multiplied element-wise with the initial channel matrix and residually concatenated with the initial channel matrix to output channel state information. The bit error rate requirement factor is obtained from the bit error rate requirement factor field in the unified data frame structure. The bit error rate requirement factor is dynamically adjusted using a gradient adjustment method based on the channel state information. The sub-stream allocation ratio is calculated based on the adjusted bit error rate requirement factor and the channel state information. The business data is split into sub-streams according to the sub-stream allocation ratio and transmitted separately through the power line communication link and the low-power wireless communication link in the optimal communication path.

2. The method according to claim 1, characterized in that, The design of the unified MAC layer protocol includes setting a bit error rate requirement factor field in the unified data frame structure, and outputting a unified data frame structure containing the bit error rate requirement factor field, including: The beacon period is divided into contention-based and non-contention-based time slots. During the contention-based time slots, a carrier sense multiple access mechanism is used to control node access, while during the non-contention-based time slots, a time division multiple access mechanism is used to allocate transmission time slots. Construct an initial data frame structure that includes a preamble field, a network type field, and a bit error rate requirement factor field; In the initial data frame structure, the preamble field is set to the known sequence required for pilot signal reception and channel matrix acquisition, the network type field is set to the identification code that identifies the power line communication link and the low-power wireless communication link, and the bit error rate requirement factor field is set to the bit error rate parameter required for substream allocation ratio calculation. The output contains a unified data frame structure that includes a bit error rate requirement factor field.

3. The method according to claim 1, characterized in that, The link parameters of the power line communication link and the low-power wireless communication link are obtained based on a unified data frame structure. Weight coefficient combinations are matched according to the service type, and disturbance parameters are introduced to adjust the path search process. The output of the optimal communication path includes: Power line communication links and low-power wireless communication links are identified from the network type field of a unified data frame structure, and the delay, bit error rate, communication rate and routing hop count of each link are obtained as link parameters. Based on the business type, select the corresponding weight coefficient combination from the pre-configured weight coefficient library, and construct the path cost function based on the link parameters and the weight coefficient combination; Obtain the current number of network nodes and calculate the perturbation parameter. When the number of consecutive convergences of the path search reaches a preset convergence threshold, introduce the perturbation parameter to perform Brownian motion perturbation on the path cost function. The optimal next-hop node is searched based on the perturbed path cost function, and the optimal communication path is output. The optimal communication path is used for subsequent pilot signal reception and channel matrix acquisition.

4. The method according to claim 3, characterized in that, The perturbation parameter is related to the network size, and its calculation formula is as follows: ； Where η is the perturbation parameter, and N is the number of network nodes, with N>1.

5. The method according to claim 3, characterized in that, The formula for the path cost function is: ； Among them, S C Let be the path cost function, D be the delay, E be the bit error rate, B be the communication rate, and H be the number of hops. For the weighting coefficients corresponding to the delay, The weighting coefficients corresponding to the bit error rate. Weighting coefficients corresponding to communication rates This represents the weight coefficient corresponding to the number of route hops.

6. The method according to claim 1, characterized in that, The optimal communication path, consisting of a power line communication link and a low-power wireless communication link, receives pilot signals to obtain an initial channel matrix. A global average pooling method is then applied to the initial channel matrix via a convolutional layer to obtain a compressed feature vector. This compressed feature vector is then subjected to dimensionality reduction and expansion processing via another convolutional layer to generate feature channel weights. The feature channel weights are then element-wise multiplied with the initial channel matrix, and a residual concatenation is performed between the feature channel weights and the initial channel matrix. The output channel state information includes: Extract the power line communication link and the low-power wireless communication link from the optimal communication path as the target channel link, and control the target channel link to receive pilot signals; The received pilot sequence is extracted from the preamble field of the unified data frame structure, the locally known pilot sequence is obtained, and the ratio of the received pilot sequence to the locally known pilot sequence is calculated using the least squares method to obtain the initial channel matrix. The initial channel matrix is ​​input into a deep residual network, and global average pooling is performed on the initial channel matrix to obtain a compressed feature vector. The compressed feature vector is then sequentially input into a dimensionality reduction convolutional layer and an dimensionality increase convolutional layer for dimensionality reduction and dimensionality increase processing to obtain an dimensionality increase feature vector. The upgraded feature vector is normalized and activated to generate feature channel weights. The feature channel weights are multiplied element-wise with the initial channel matrix to obtain a weighted channel matrix. The weighted channel matrix is ​​then residually concatenated with the initial matrix to output the channel state information. The channel state information includes the bit error rate of the power line communication link, the bit error rate of the low-power wireless communication link, the communication rate of the power line communication link, and the communication rate of the low-power wireless communication link.

7. The method according to claim 1, characterized in that, The steps include obtaining the bit error rate (BER) requirement factor based on the BER requirement factor field in the unified data frame structure, dynamically adjusting the BER requirement factor using a gradient adjustment method based on channel state information, and calculating the substream allocation ratio based on the adjusted BER requirement factor and channel state information. The initial bit error rate requirement factor is extracted from the bit error rate requirement factor field of the unified data frame structure, and the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link are extracted from the channel state information. The error gradient value is calculated based on the bit error rate of the power line communication link and the bit error rate of the low-power wireless communication link. The initial bit error rate requirement factor is updated according to the preset adjustment step size and the error gradient value to obtain the adjusted bit error rate requirement factor. The power line communication link rate and the low-power wireless communication link rate are extracted from the channel state information. The proportion of power line communication sub-streams and the proportion of low-power wireless communication sub-streams are calculated based on the adjusted bit error rate requirement factor, the power line communication link rate, and the low-power wireless communication link rate. A smoothing factor is introduced to perform a weighted average of the sub-stream proportions at the current time and at historical times to obtain the sub-stream allocation ratio.

8. The method according to claim 1, characterized in that, The step of splitting the business data into sub-streams according to the sub-stream allocation ratio and transmitting them respectively through the power line communication link and the low-power wireless communication link in the optimal communication path includes: The service data is split according to the proportion of power line communication sub-stream and low-power wireless communication sub-stream in the sub-stream allocation ratio, generating power line communication sub-stream data and low-power wireless communication sub-stream data; Power line communication substream data is encapsulated into a unified data frame structure and transmitted through the power line communication link in the optimal communication path; Low-power wireless communication sub-stream data is encapsulated into a unified data frame structure and transmitted through a low-power wireless communication link in the optimal communication path; At the receiving end, a reassembly buffer is set up to receive the power line communication sub-stream data and the low-power wireless communication sub-stream data respectively. When the amount of data in the reassembly buffer reaches a preset trigger threshold, a sub-stream synthesis operation is performed to synthesize the power line communication sub-stream data and the low-power wireless communication sub-stream data into complete service data.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method as described in any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 8.

Images