Wireless link transmission rate prediction method and device, storage medium and equipment

Abstract

A wireless link transmission rate prediction method and device, a storage medium and an equipment. The method comprises: collecting channel state data; obtaining a plurality of signal-to-interference-and-noise ratio vectors based on the collected channel state data; performing feature extraction of different dimensions on the plurality of signal-to-interference-and-noise ratio vectors respectively to obtain scalar data corresponding to each dimension respectively; and performing wireless link transmission rate prediction based on the obtained scalar data. The above scheme can effectively improve the prediction accuracy of the wireless link transmission rate.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and specifically to a method, apparatus, storage medium, and device for predicting wireless link transmission rate. Background Technology

[0002] In modern wireless communication systems such as Multiple-Input Multiple-Output Orthogonal Frequency Division Multiplexing (MIMO-OFDM), in order to achieve dynamic resource scheduling and link adaptation, it is necessary to make fast and accurate predictions of the achievable rate of the wireless link.

[0003] However, existing wireless link transmission rate prediction methods have low prediction accuracy and are difficult to meet the requirements of link adaptation in sixth-generation mobile communication and future networks. Summary of the Invention

[0004] The problem this invention aims to solve is: how to improve the prediction accuracy in the process of predicting the transmission rate of wireless links.

[0005] To address the above problems, embodiments of the present invention provide a wireless link transmission rate prediction method, the method comprising: Collect channel status data; Several signal-to-interference-plus-noise ratio (SINR) vectors are obtained based on the collected channel state data; The signal-to-interference-plus-noise ratio vectors are subjected to feature extraction in different dimensions to obtain scalar data corresponding to each dimension. Wireless link transmission rate prediction is performed based on the obtained scalar data.

[0006] In one possible embodiment, the step of extracting features from the plurality of signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension includes: The signal-to-interference-plus-noise ratio (SINR) vectors are respectively input to a preset number of neurons. Each neuron extracts features of the corresponding dimension from the input SINR vectors to obtain scalar data output by each neuron. In this study, the feature dimensions corresponding to each neuron are mutually exclusive.

[0007] In one possible embodiment, the step of extracting features from the input signal-to-interference-plus-noise ratio vector of each neuron to obtain scalar data output by each neuron includes: Each neuron sequentially performs a first scaling operation, a summation operation, an aggregation operation, a second scaling operation, and an activation operation on the input signal-to-interference-plus-noise ratio vector to obtain the corresponding scalar data.

[0008] In one possible embodiment, the scaling factors in the first scaling operation and the second scaling operation are determined in advance after training each neuron using training samples, and the value of the scaling factor is related to the feature dimension corresponding to the neuron.

[0009] In one possible embodiment, the wireless link transmission rate prediction based on the obtained scalar data includes: The scalar data corresponding to each dimension are combined into a scalar data matrix; The scalar data matrix is ​​input into a preset fully connected layer to obtain the output of the fully connected layer; The final modulation and coding strategy index value is predicted based on the output of the fully connected layer. Based on the final modulation and coding strategy index value, the predicted wireless link transmission rate is obtained.

[0010] In one possible embodiment, the fully connected layer is used to perform a linear transformation on the feature vector matrix.

[0011] In one possible embodiment, predicting the final modulation and coding strategy index value based on the output of the fully connected layer includes: Based on the output of the fully connected layer, predict all possible modulation and coding strategy index values ​​and the probability values ​​corresponding to each modulation and coding strategy index; The modulation and coding strategy index value with the highest probability is selected as the final modulation and coding strategy index value.

[0012] In one possible embodiment, obtaining the predicted wireless link transmission rate based on the final modulation and coding strategy index value includes: Based on the final modulation and coding strategy index value, the preset modulation and coding strategy index table is searched to obtain the predicted value of the wireless link transmission rate.

[0013] In one possible embodiment, the method further includes: The difference between the predicted wireless link transmission rate and the actual wireless link transmission rate is obtained, and the parameters in the wireless link transmission rate prediction process are updated based on the difference.

[0014] This invention also provides a wireless link transmission rate prediction device, the device comprising: Acquisition unit, suitable for acquiring channel status data; The data processing unit is adapted to obtain several signal-to-interference-plus-noise ratio vectors based on the collected channel state data; The feature extraction unit is adapted to extract features from the several signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension. The prediction unit is suitable for predicting the transmission rate of a wireless link based on the obtained scalar data.

[0015] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the methods described above.

[0016] This invention also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the steps of any of the methods described above when running the computer program.

[0017] Compared with the prior art, the technical solution of the embodiments of the present invention has the following advantages: The present invention, after obtaining the signal-to-interference-plus-noise ratio (SINR) vector, extracts features from several SINR vectors in different dimensions to obtain scalar data corresponding to each dimension. Then, it combines the obtained scalar data to predict the wireless link transmission rate. Compared to a single mapping scheme between the SINR vector and scalar data, extracting features from the SINR vector in different dimensions can adaptively fit the complex and ever-changing real channel environment. This effectively improves the prediction accuracy when predicting the wireless link transmission rate based on scalar data in different dimensions, and is particularly suitable for MIMO-OFDM systems with complex channel characteristics and multiple interference sources.

[0018] Furthermore, during the feature extraction process of the signal-to-interference-plus-noise ratio (SINR) vector across different dimensions, several SINR vectors are input into multiple preset neurons. Each neuron then extracts features from the input SINR vector in its corresponding dimension, yielding scalar data output by each neuron. The number of neurons can be flexibly configured, adaptively adjusting to the complexity and accuracy requirements of the actual communication prediction task. This ensures both the overall feature extraction capability and business prediction accuracy of the network, while also flexibly adapting to different computing resources and business scenarios, significantly improving the model's data-driven adaptability and scenario generalization ability. Attached Figure Description

[0019] Figure 1 This is a flowchart of a wireless link transmission rate prediction method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of the network structure of the shallow neuron model in an embodiment of the present invention; Figure 3 This is a flowchart of another wireless link transmission rate prediction method in an embodiment of the present invention; Figure 4 This is a flowchart of a training method for a shallow neuron model according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the structure of a wireless link transmission rate prediction device according to an embodiment of the present invention. Detailed Implementation

[0020] In modern wireless communication systems (such as 6G), to cope with complex and ever-changing wireless channel environments and maximize spectral efficiency, real-time sensing and dynamic adjustment of wireless link performance are necessary. Adaptive coding and modulation (OCM), as the core technology of link adaptation, can dynamically select the optimal modulation scheme and coding rate based on instantaneous channel quality, thereby approximating channel capacity while ensuring transmission reliability. The key is that the transmitter relies on the receiver's predicted accurate link rate (usually represented by the MCS index) as a decision-making basis to achieve dynamic matching of the coding and modulation scheme. Therefore, fast and accurate link rate prediction is crucial when wireless communication systems utilize adaptive coding and modulation techniques to optimize system performance.

[0021] To achieve efficient link adaptation, the industry commonly employs methods that predict link rates based on channel state information. Among these, the Exponential Effective Signal-to-Noise Ratio Mapping (EESM) model is the most frequently used classic method due to its simple mathematical form. In addition, there are also improved schemes such as polynomial-based fitting models and piecewise constant function models, as well as purely data-driven schemes such as random forests and machine learning models.

[0022] However, these existing solutions have low prediction accuracy. Faced with the higher demands of 6G mobile communication and future networks for link adaptation, there is an urgent need to explore a wireless link transmission rate prediction method with higher prediction accuracy.

[0023] To address this problem, this invention provides a method for predicting wireless link transmission rate. This method, after obtaining the signal-to-interference-plus-noise ratio (SINR) vector, extracts features from several SINR vectors in different dimensions to obtain scalar data corresponding to each dimension. This scalar data can then be combined to predict the wireless link transmission rate. Compared to single-dimensional feature extraction, multi-dimensional parallel feature extraction can comprehensively characterize the deep information of the channel, accurately fit the complex and dynamically changing actual wireless channel environment, and compensate for the incomplete representation of channel state by single features, thereby significantly improving the prediction accuracy of wireless link transmission rate.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Reference Figure 1 This invention provides a method for predicting the transmission rate of a wireless link, which may include the following steps: Step 11: Collect channel status data.

[0026] In specific implementation, the channel state data refers to relevant data that can be used to calculate the signal-to-interference-plus-noise ratio (SINR). Specifically, a target link (i.e., a transmission channel) typically contains multiple subcarriers, and the collected channel state data can include the channel state data of each subcarrier in the target link. The channel state data of the i-th subcarrier in the target link can include: the noise floor power n of the i-th subcarrier. i Sub-channel matrix H i .

[0027] In practical implementation, when it is necessary to measure the transmission rate of the target link, the channel state data corresponding to each subcarrier of the target link at the current moment can be collected. Based on the channel state data corresponding to each subcarrier at the current moment, the signal-to-interference-plus-noise ratio vector corresponding to each subcarrier can be obtained.

[0028] Step 12: Obtain several signal-to-interference-plus-noise ratio vectors based on the collected channel state data.

[0029] In practical implementation, the signal-to-interference-plus-noise ratio (SINR) of the i-th carrier can be calculated using the following formula. i : (1) Among them, H i Let n be the subchannel matrix corresponding to the i-th carrier; i V represents the noise floor power of the i-th subcarrier; i It is for H i The unitary matrix obtained after singular value decomposition. is the Frobenius norm, representing the sum of squares of the magnitudes of the channel coefficients across all antenna pairs. txbf indicates transmit beamforming is enabled, and notxbf indicates transmit beamforming is disabled.

[0030] By traversing all subcarriers, an I-dimensional SINR eigenvector matrix composed of the signal-to-interference-plus-noise ratio of each subcarrier in the target link is obtained.

[0031] Step 13: Extract features from the several signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension.

[0032] In one embodiment, reference is made to Figure 2A custom neuron layer 21 can be pre-defined. This custom neuron layer can include multiple parallel neurons, each employing differential learning to mine different feature dimensions of the SINR vector. These feature dimensions include, but are not limited to, amplitude features, extreme value features, distribution features, and aggregate statistical features of the SINR vector. The feature dimensions corresponding to each neuron are non-overlapping.

[0033] For example, refer to Figure 3 N parallel neurons are pre-set, namely neuron 1 to neuron N. Neuron 1 is used to mine the amplitude feature of SINR vector, neuron 2 is used to mine the extreme value feature of SINR vector, neuron 3 is used to mine the distribution feature of SINR vector, and so on.

[0034] In the actual transmission rate prediction process, a 1D SINR vector matrix composed of several signal-to-interference-plus-noise ratio (SINR) vectors can be input into multiple pre-set neurons. Each neuron can receive the 1D SINR vector matrix and perform a completely identical data processing procedure to extract deep features of the SINR vector matrix from different dimensions. After independent processing, each neuron outputs a scalar data point, with each scalar data point uniquely corresponding to a feature dimension, used to characterize the channel feature information of the SINR vector matrix in that dimension.

[0035] In one embodiment of the present invention, each neuron can sequentially perform a first scaling operation, a summation operation, an aggregation operation, a second scaling operation, and an activation operation on the input signal-to-interference-plus-noise ratio vector to obtain corresponding scalar data.

[0036] For example, a first scaling operation is performed on the input 1D SINR vector matrix using a first scaling factor p. After the first scaling operation, the result of the first scaling operation can be logarithmically summed and exponentially calculated. For the summation operation, an aggregation operation is then performed to obtain the corresponding aggregated value. Subsequently, the aggregation result is scaled again by a second scaling factor q (i.e., a second scaling operation is performed), and finally processed by the Sigmoid activation function to output the corresponding scalar data.

[0037] The first scaling factor p and the second scaling factor q are determined in advance after training each neuron using training samples, and their values ​​are related to the feature dimension corresponding to the neuron. The first scaling factor p and the second scaling factor q of each neuron are independent and adaptively converge and optimize in different directions through training, ultimately enabling the neuron to extract the defined features from the input SINR vector. Each neuron outputs a scalar data point corresponding to its own feature dimension, comprehensively characterizing the deep state information of the wireless channel.

[0038] Through the first scaling operation, summation operation, aggregation operation, and second scaling operation, the SINR vector can be subjected to multi-order nonlinear projection to construct a preliminary feature combination. Then, the preliminary feature combination is subjected to nonlinear transformation through activation operation to capture the complex characteristics of the channel, improve the expressive power of the neuron, and provide an abstract feature basis for subsequent transmission rate prediction, realizing the transformation from raw channel state data to efficient prediction features.

[0039] Step 14: Predict the wireless link transmission rate based on the obtained scalar data.

[0040] In practice, based on the obtained scalar data, various methods can be used to predict the wireless link transmission rate.

[0041] In existing technologies, shallow networks with fixed structures are typically used for wireless link transmission rate prediction. However, using shallow networks with fixed structures for wireless link transmission rate prediction lacks the ability to deeply map the complex features extracted in previous steps, and cannot accurately characterize the complex decision boundaries from features to discrete link rates, thus affecting prediction accuracy.

[0042] In one embodiment of the present invention, a fully connected decision network with multi-layer nonlinear transformation capabilities is employed to achieve accurate mapping of complex features, thereby effectively improving prediction accuracy. Compared to shallow networks with fixed structures, using a fully connected decision network with multi-layer nonlinear transformation capabilities for wireless link transmission rate prediction can improve feature utilization. It can effectively decouple and recombine the high-dimensional nonlinearities containing rich potential information extracted in previous steps, thereby improving prediction accuracy and avoiding the impact of a lack of multi-layer nonlinear transformation on prediction accuracy. Furthermore, the fully connected decision network with multi-layer nonlinear transformation capabilities, due to its nonlinearity, makes the prediction results at feature boundaries more stable and less prone to abrupt errors.

[0043] In specific implementation, refer to Figure 2 The fully connected decision network with multi-layer nonlinear transformation capability may include a fully connected layer 22, an activation layer 23, and an output layer 24 connected in sequence. This fully connected decision network is a high-dimensional discrimination process that outputs discrimination results based on scalar data.

[0044] Specifically, refer to Figure 3 The specific steps for predicting the wireless link transmission rate based on the obtained scalar data may include: Step 141: Concatenate the scalar data corresponding to each dimension to form a scalar data matrix.

[0045] The scalar data output by each neuron in the custom neuron layer is concatenated to form an N-dimensional scalar data matrix.

[0046] Step 142: Input the scalar data matrix into a preset fully connected layer to obtain the output result of the fully connected layer.

[0047] In practice, neurons in the fully connected layer 22 perform only linear transformations and do not involve any activation or nonlinear operations. All neurons in the fully connected layer 22 share the same set of inputs and need to perform a weighted summation of all input features. The weight and bias of each neuron are determined based on the importance of the features to the current decision, and finally the linear transformation result of N high-dimensional features is output.

[0048] Step 143: Predict the final modulation and coding strategy index value based on the output of the fully connected layer.

[0049] In practice, all possible modulation and coding scheme index values ​​and the probability values ​​corresponding to each modulation and coding scheme (MCS) index can be predicted based on the output of the fully connected layer. Finally, the MCS index value with the highest probability is selected as the final modulation and coding scheme index value.

[0050] Specifically, the decision features output by the fully connected layer can be weighted and combined to calculate the probability for each possible MCS index value. Then, the probability can be converted into a probability distribution using the Sigmoid function. Based on this distribution, the model selects the MCS index value with the highest probability as the prediction output.

[0051] Step 144: Based on the final modulation and coding strategy index value, obtain the predicted value of the wireless link transmission rate.

[0052] Specifically, based on the final MCS index value, the predicted wireless link transmission rate can be obtained by looking up a preset MCS index table. The MCS index table is the core configuration table for link adaptation in wireless communication systems, formulated according to 3GPP communication standards, and used to specify the modulation scheme and coding rate corresponding to different MCS index values.

[0053] Based on the final MCS index value, the standard MCS index table can be directly queried to determine the modulation scheme, coding rate, and spectral efficiency corresponding to the current link, ultimately completing the wireless link transmission rate calculation and link adaptive configuration. This process replaces the traditional threshold-based table lookup method, improving the accuracy of MCS selection in complex channel environments.

[0054] In existing technologies, traditional machine learning models are often applied directly with their parameters unchanged after training. This results in a lack of real-time optimization and scene adaptation capabilities in transmission rate prediction. When the actual channel deviates from the training data due to interference or other factors, the predictive ability of traditional machine learning models will be further reduced, and they cannot maintain long-term prediction reliability.

[0055] To address the above problems, in embodiments of the present invention, the wireless link transmission rate prediction method may further include: obtaining the difference between the predicted wireless link transmission rate and the actual wireless link transmission rate, and updating the parameters in the wireless link transmission rate prediction process based on the difference between the two.

[0056] Specifically, feedback from the wireless link can be received in real time. Based on this feedback and the model's prediction results, the loss function is calculated. The backpropagation algorithm, combined with the gradient descent strategy, is used to update the parameters of the custom neuron layer 21 and the fully connected decision network online. This enables the shallow neuron model composed of the custom neuron layer 21 and the fully connected decision network to have environmental awareness and dynamic adjustment capabilities, effectively overcoming the defect of low reliability. As a result, it can better understand the changing wireless channel patterns in real time, further improving prediction accuracy and enhancing the stability of the communication link.

[0057] This invention also provides a training method for a shallow neuron model, specifically, referring to... Figure 4 The method may include the following steps: Step 41: Collect sample data.

[0058] In specific implementation, the sample data collected may include: channel state information matrix, channel information acquisition time, transmission rate, MCS index value corresponding to each transmission rate, transmission rate acquisition time, channel state acquisition time, beamforming on-state indication, and noise floor power, etc.

[0059] Step 42: Divide the collected sample data to obtain the training sample set and the test sample set.

[0060] In practice, the collected sample data can be preprocessed before it is divided.

[0061] Specifically, preprocessing the collected sample data may include sequentially performing time-series alignment and anomaly cleaning, SINR calculation, and data reconstruction and organization. Among these: Regarding timing alignment and anomaly removal, due to the time-varying nature of wireless channels, timing matching is required, especially in mobile scenarios where the channel matrix changes rapidly over time. The live network data collected during the data acquisition phase is categorized, and the absolute value of the difference between the channel state information acquisition time (csi_time) and the transmission rate acquisition time (dfx_time) within the same data sample is calculated. If the time difference is too large, it means using the "past" channel state to predict the "future" channel transmission rate, which violates the system's causal relationship. Therefore, a physically reasonable threshold based on channel coherence time needs to be set (this threshold is based on the distribution across the entire range or set based on channel coherence time; in this patent, it relies on live network data, for example, setting the threshold to 10ms). This ensures that the absolute value of the difference between the channel state information acquisition time and the transmission rate acquisition time is within this threshold range, guaranteeing alignment between the two acquisition times; otherwise, the corresponding data sample is discarded. This operation ensures that both the training and testing sample sets are under the same quasi-static channel conditions.

[0062] Without time-series alignment and cleaning, the data will contain samples with inconsistent channel states. For example, using the channel state from 10ms ago to predict the current transmission rate may result in the channel having completely changed. This non-causal relationship will mislead the model into learning incorrect mapping relationships, seriously damaging the model's rationality and prediction accuracy.

[0063] After timing alignment and anomaly cleanup, SINR can be calculated using formula (1) to obtain the SINR vector corresponding to each subcarrier of the wireless channel at each data acquisition time. The SINR vectors corresponding to each acquisition time form a 1D SINR vector matrix. Thus, several 1D SINR vector matrices can be obtained. In the process of SINR calculation, the unit of noise power can be uniformly converted to the linear domain to avoid the calculated SINR value and trend being inconsistent with the actual signal-to-noise ratio due to directly using the noise power in dBm units for calculation.

[0064] All valid data samples, after time-series alignment, cleaning, and computation, are organized into a new dataset, which is then stored in a designated table. Without standardized data reconstruction, the cleaned data is disorganized, making efficient batch matrix operations impossible, resulting in low model training efficiency and potential program crashes due to data format errors.

[0065] Without the above-mentioned refined preprocessing operations, the data foundation of the model will be distorted or erroneous, causing prediction defects in subsequent steps during model runtime.

[0066] Step 43: Iteratively train the initial network model using the training sample set and perform performance verification using the test sample set until the model converges, thus obtaining the final network architecture after training.

[0067] Specifically, the initial network model includes an initial custom neuron layer and an initial fully connected decision network. The initial network model is trained iteratively multiple times using a training sample set. During training, the backpropagation algorithm and gradient descent strategy, combined with the Adam optimizer, are used to automatically update all learnable parameters of the network, including the first scaling learnable parameter p and the second scaling learnable parameter q within each individual neuron, as well as the network weights and bias parameters in the fully connected decision network. After the model iterative training is completed, the predictive performance of the trained model is tested and verified using a test sample set. The model's fitting effect is continuously optimized until it meets the preset convergence conditions and prediction accuracy metrics, ultimately yielding the target network architecture that has achieved training convergence.

[0068] In this embodiment of the invention, the aforementioned shallow neuron model is used as the target network architecture. By automatically fitting the nonlinear mapping relationship between the channel and the rate through scientific parameters, not only can the wireless link transmission rate be predicted, but the interpretability gap caused by "black box" deep networks can also be avoided. Furthermore, this target network architecture adopts a shallow parallel neuron structure, which is suitable for rapid deployment of pre-online learning in resource-constrained scenarios.

[0069] The wireless link transmission rate prediction method employed in this invention utilizes a hierarchical parallel custom neuron structure in the feature extraction stage. Each neuron can independently learn the optimal nonlinear mapping relationship from the data, achieving a fusion of physical interpretability and data-driven capabilities. In the final MCS decision and output stage, the traditional regression problem is reconstructed into a classification task, mapping the extracted features to the probability distribution of discrete MCS categories, and finally outputting the most probable transmission rate index, effectively avoiding prediction jitter at the decision boundary of the regression model. This method significantly improves the accuracy and stability of rate prediction while maintaining extremely low computational complexity, providing effective technical support for real-time link adaptation in sixth-generation wireless communication and future networks.

[0070] To enable those skilled in the art to better understand and implement the present invention, the apparatus, testing system, electronic device, and computer-readable storage medium corresponding to the above method are described in detail below.

[0071] Reference Figure 5 This invention provides a wireless link transmission rate prediction device 50, which may include: a data acquisition unit 51, a data processing unit 52, a feature extraction unit 53, and a prediction unit 54. Wherein: The acquisition unit 51 is adapted to acquire channel status data; The data processing unit 52 is adapted to obtain several signal-to-interference-plus-noise ratio vectors based on the collected channel state data; The feature extraction unit 53 is adapted to perform feature extraction on the plurality of signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension. The prediction unit 54 is adapted to predict the wireless link transmission rate based on the obtained scalar data.

[0072] In some embodiments, the feature extraction unit 53 is adapted to input the plurality of signal-to-interference-plus-noise ratio (SINR) vectors into a plurality of preset neurons, and each neuron extracts features of the corresponding dimension from the input SINR vector to obtain scalar data output by each neuron; wherein the feature dimensions corresponding to each neuron do not overlap.

[0073] The acquisition unit 51, data processing unit 52, feature extraction unit 53, and prediction unit 54 can be implemented with reference to the above description of the corresponding method steps, and will not be repeated here.

[0074] This invention also provides a computer-readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the steps of any of the above methods.

[0075] In specific implementations, the computer-readable storage medium may include ROM, RAM, disk, or optical disk, etc.

[0076] This invention also provides an electronic device, which includes a memory and a processor. The memory stores a computer program that can run on the processor, and when the processor runs the computer program, it performs the steps of any of the methods described above.

[0077] Regarding the modules / units included in the various devices and products described in the above embodiments, they can be software modules / units, hardware modules / units, or a combination of both. For example, for various devices and products applied to or integrated into a chip, all of their modules / units can be implemented using hardware methods such as circuits, or at least some modules / units can be implemented using software programs that run on a processor integrated within the chip, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits; for various devices and products applied to or integrated into a chip module, all of their modules / units can be implemented using hardware methods such as circuits, and different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or different components of the chip module, or at least some modules / units can be implemented using hardware methods such as circuits. The components can be implemented using software programs that run on the processor integrated within the chip module. The remaining (if any) modules / units can be implemented using hardware methods such as circuits. For various devices and products applied to or integrated into the terminal, each of its components / units can be implemented using hardware methods such as circuits. Different modules / units can be located in the same component (e.g., chip, circuit module, etc.) or in different components within the terminal. Alternatively, at least some modules / units can be implemented using software programs that run on the processor integrated within the terminal, while the remaining (if any) modules / units can be implemented using hardware methods such as circuits.

[0078] While the present invention has been disclosed above, it is not limited thereto. Any person skilled in the art can make various modifications and alterations without departing from the spirit and scope of the invention; therefore, the scope of protection of the present invention should be determined by the scope defined in the claims.

Claims

1. A method for predicting the transmission rate of a wireless link, characterized in that, include: Collect channel status data; Several signal-to-interference-plus-noise ratio (SINR) vectors are obtained based on the collected channel state data; The signal-to-interference-plus-noise ratio vectors are subjected to feature extraction in different dimensions to obtain scalar data corresponding to each dimension. Wireless link transmission rate prediction is performed based on the obtained scalar data.

2. The wireless link transmission rate prediction method as described in claim 1, characterized in that, The step of extracting features from the plurality of signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension includes: The signal-to-interference-plus-noise ratio (SINR) vectors are respectively input to a preset number of neurons. Each neuron extracts features of the corresponding dimension from the input SINR vectors to obtain scalar data output by each neuron. In this study, the feature dimensions corresponding to each neuron are mutually exclusive.

3. The wireless link transmission rate prediction method as described in claim 2, characterized in that, The step involves extracting features from the input signal-to-interference-plus-noise ratio (SINR) vector for each neuron, resulting in scalar data output by each neuron, including: Each neuron sequentially performs a first scaling operation, a summation operation, an aggregation operation, a second scaling operation, and an activation operation on the input signal-to-interference-plus-noise ratio vector to obtain the corresponding scalar data.

4. The wireless link transmission rate prediction method as described in claim 3, characterized in that, The scaling factors in the first scaling operation and the second scaling operation are determined in advance after training each neuron using training samples, and the value of the scaling factor is related to the feature dimension corresponding to the neuron.

5. The wireless link transmission rate prediction method as described in claim 1, characterized in that, The wireless link transmission rate prediction based on the obtained scalar data includes: The scalar data corresponding to each dimension are combined into a scalar data matrix; The scalar data matrix is ​​input into a preset fully connected layer to obtain the output of the fully connected layer; The final modulation and coding strategy index value is predicted based on the output of the fully connected layer. Based on the final modulation and coding strategy index value, the predicted wireless link transmission rate is obtained.

6. The wireless link transmission rate prediction method as described in claim 5, characterized in that, The fully connected layer is used to perform a linear transformation on the feature vector matrix.

7. The wireless link transmission rate prediction method as described in claim 5, characterized in that, The prediction of the final modulation and coding strategy index value based on the output of the fully connected layer includes: Based on the output of the fully connected layer, predict all possible modulation and coding strategy index values ​​and the probability values ​​corresponding to each modulation and coding strategy index; The modulation and coding strategy index value with the highest probability is selected as the final modulation and coding strategy index value.

8. The wireless link transmission rate prediction method as described in claim 5, characterized in that, The process of obtaining the predicted wireless link transmission rate based on the final modulation and coding strategy index value includes: Based on the final modulation and coding strategy index value, the preset modulation and coding strategy index table is searched to obtain the predicted value of the wireless link transmission rate.

9. The wireless link transmission rate prediction method as described in claim 1, characterized in that, Also includes: The difference between the predicted wireless link transmission rate and the actual wireless link transmission rate is obtained, and the parameters in the wireless link transmission rate prediction process are updated based on the difference.

10. A wireless link transmission rate prediction device, characterized in that, include: Acquisition unit, suitable for acquiring channel status data; The data processing unit is adapted to obtain several signal-to-interference-plus-noise ratio vectors based on the collected channel state data; The feature extraction unit is adapted to extract features from the several signal-to-interference-plus-noise ratio vectors in different dimensions to obtain scalar data corresponding to each dimension. The prediction unit is suitable for predicting the transmission rate of a wireless link based on the obtained scalar data.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, The computer program is executed by a processor to implement the steps of the method according to any one of claims 1 to 9.

12. An electronic device comprising a memory and a processor, wherein the memory stores a computer program capable of running on the processor, characterized in that, When the processor runs the computer program, it performs the steps of the method according to any one of claims 1 to 9.

Images