WO2023029751A1

WO2023029751A1 - Adaptive coding and modulation method and apparatus, electronic device and storage medium

Info

Publication number: WO2023029751A1
Application number: PCT/CN2022/104761
Authority: WO
Inventors: 林志远; 林伟; 刘向凤; 芮华; 黄河
Original assignee: 中兴通讯股份有限公司
Priority date: 2021-08-31
Filing date: 2022-07-08
Publication date: 2023-03-09
Also published as: CN115730676A

Abstract

An adaptive coding and modulation method and apparatus, an electronic device, and a storage medium, the method comprising: acquiring feature parameters, wherein the feature parameters comprise data which represents user transmission capabilities (101); inputting the feature parameters into an inner loop machine learning model to obtain an MCS initial value (102); inputting the MCS initial value and the feature parameters into an outer loop machine learning model to obtain an MCS adjustment value corresponding to the MCS initial value (103); and obtaining an adjusted MCS value according to the adjustment value corresponding to the MCS initial value and the MCS initial value (104).

Description

Adaptive coding and modulation method, device, electronic equipment and storage medium

Cross References to Related Applications

This application is based on the Chinese patent application with the application number "202111014423.5" and the filing date is August 31, 2021, and claims the priority of the Chinese patent application. The entire content of the Chinese patent application is hereby incorporated by reference. Apply.

technical field

The embodiments of the present application relate to the communication field, and in particular to an adaptive coding and modulation method, device, electronic equipment, and storage medium.

Background technique

Adaptive Modulation and Coding (AMC) is an adaptive coding and modulation technology used on wireless channels. It ensures the transmission quality of links by adjusting the modulation mode and coding rate of wireless link transmission. The implementation principle of AMC can be simply described as: using channel-related characteristic parameters to adaptively adjust the Modulation and Coding Scheme (MCS for short).

Since AMC is user-oriented, in the method of obtaining the MCS value, a specific formula will be used to describe the corresponding relationship between the MCS value and the characteristic parameters. When calculating the MCS value, the characteristic parameters, such as Channel Quality Indicator (Channel Quality Indicator, CQI) and the number of streams are substituted into a specific formula to calculate the MCS value. However, the relationship between the MCS value and the characteristic parameters is different in different transmission scenarios. Still using this specific formula to calculate the MCS value cannot fit different transmission scenarios, which will lead to low accuracy of the calculated MCS.

Contents of the invention

An embodiment of the present application provides an adaptive coding and modulation method, including: obtaining characteristic parameters, wherein the characteristic parameters include data representing the user's transmission capability; inputting the characteristic parameters into the inner loop machine learning model to obtain MCS initial value; input the MCS initial value and the characteristic parameters into the outer loop machine learning model to obtain the MCS adjustment value corresponding to the MCS initial value; according to the adjustment value corresponding to the MCS initial value and the MCS initial value value to get the adjusted MCS value.

The embodiment of the present application also provides an adaptive coding and modulation device, including: a parameter acquisition module, configured to acquire a characteristic parameter, wherein the characteristic parameter includes data representing the user's transmission capability; an initial value acquisition module, which obtains the The characteristic parameters are input into the inner loop machine learning model to obtain the initial value of the MCS; the adjustment value acquisition module is used to input the initial value of the MCS and the characteristic parameters into the outer loop machine learning model to obtain the corresponding MCS initial value MCS adjustment value; an adjustment module, configured to obtain an adjusted MCS value according to the adjustment value corresponding to the initial MCS value and the initial MCS value.

The embodiment of the present application also provides an electronic device, at least one processor; and a memory connected in communication with the at least one processor; wherein, the memory stores instructions executable by the at least one processor, The instructions are executed by the at least one processor, so that the at least one processor can execute the above adaptive coding and modulation method.

The embodiment of the present application also provides a computer-readable storage medium storing a computer program, and implementing the above-mentioned adaptive coding and modulation method when the computer program is executed by a processor.

Description of drawings

FIG. 1 is a flow chart of an adaptive coding and modulation method provided according to an embodiment of the present application;

FIG. 2 is a schematic diagram of the interaction of various functional modules in an adaptive coding and modulation method provided according to an embodiment of the present application;

FIG. 3 is a schematic diagram of an inner-loop machine learning model in an adaptive coding and modulation method provided according to an embodiment of the present application;

4 is a schematic diagram of an outer-loop machine learning model in an adaptive coding and modulation method provided according to an embodiment of the present application;

Fig. 5 is a flow chart of an adaptive coding and modulation method for obtaining characteristic parameters according to transmission types according to an embodiment of the present application;

FIG. 6 is a schematic diagram of an adaptive coding and modulation device provided according to an embodiment of the present application;

Fig. 7 is a schematic structural diagram of an electronic device provided according to an embodiment of the present application.

Detailed ways

The main purpose of the embodiments of the present application is to provide an adaptive coding and modulation method, device, electronic equipment and storage medium, which can improve the accuracy of MCS selection.

In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the embodiments of the present application will be described in detail below with reference to the accompanying drawings. However, those of ordinary skill in the art can understand that in each embodiment of the application, many technical details are provided for readers to better understand the application. However, even without these technical details and various changes and modifications based on the following embodiments, the technical solution claimed in this application can also be realized. The division of the following embodiments is for the convenience of description, and should not constitute any limitation to the specific implementation of the present application, and the various embodiments can be combined and referred to each other on the premise of no contradiction.

In the method of obtaining the MCS value, a specific formula is used to describe the corresponding relationship between the MCS value and the characteristic parameters. When calculating the MCS value, the characteristic parameters, such as Channel Quality Indicator (CQI for short) and the number of flows, Substitute into the specific formula to calculate the MCS value.

In different transmission scenarios, the relationship between the MCS value and the characteristic parameters is different. This formula is not universal. If the specific formula is still used to calculate the MCS value, it cannot be adapted to different transmission scenarios, which will lead to the calculated The accuracy of MCS is low.

In some schemes, the characteristic parameters can also be used as the state (state), the MCS value can be used as the behavior (action), and the spectral efficiency or throughput brought by MCS-based transmission can be used as the reward (reward) for reinforcement learning. Get the Q value under different states and behaviors, and choose the MCS value based on the Q value. The Q value will be updated in real time with the income of each feedback.

However, although reinforcement learning can adjust the MCS value online in real time according to the feedback, the value range of the MCS value is large, and the training of the MCS value as a behavior will lead to a high dimension of the MCS behavior space, resulting in high computational complexity, which in turn reduces the adaptive coding. modulation efficiency.

Therefore, this embodiment provides an adaptive coding and modulation method, which can be applied to equipment such as base stations, but is not limited thereto. The adaptive coding and modulation method in this embodiment includes: obtaining characteristic parameters, wherein the characteristic parameters include The data of the user's transmission ability; input the characteristic parameters into the inner loop machine learning model to obtain the initial MCS value; input the MCS initial value and the characteristic parameters into the outer loop machine learning model to obtain the MCS initial value A corresponding MCS adjustment value; an adjusted MCS value is obtained according to the adjustment value corresponding to the initial MCS value and the initial MCS value.

Compared with using a specific formula to calculate the MCS value, the MCS value obtained based on the inner-loop machine learning model and the outer-loop machine learning model in this embodiment does not require a specific calculation formula for the MCS value, so that the input of the inner-loop machine learning model and the outer-loop The characteristic parameters of the machine learning model can be determined according to the transmission scenario, and can be used in different transmission scenarios, which in turn helps to improve the accuracy of the MCS.

In addition, compared with the MCS value obtained by reinforcement learning, this embodiment inputs the characteristic parameters including the data representing the user's transmission capability into the inner-loop machine learning model to obtain the initial value of MCS, and inputs the initial MCS value and the characteristic parameters into the outer-loop machine Learn the model to get the adjusted value of MCS. Since the value range of the MCS adjusted value is smaller than that of the initial value of the MCS, when the MCS adjusted value is used as an action for training, the outer loop machine learning model The dimension of the behavior space is low, the scale of the neural network is small, and the computational complexity of obtaining the MCS adjustment value is low, thereby improving the efficiency of adaptive coding and modulation. Computational complexity of the MCS value.

The implementation details of the adaptive coding and modulation method of this embodiment are described in detail below, and the following content is only implementation details provided for easy understanding, and is not necessary for implementing this solution.

Step 101, acquire feature parameters, where the feature parameters include data representing the user's transmission capability.

In some embodiments, the feature parameters may include: the user's transmission stream number (LayerNum), channel quality indicator (Channel Quality Indicator, referred to as CQI), CQI link calculation SINR (CQI SINR), precoding matrix indicator (Precoding Matrix Indicator, referred to as PMI), rank indication (rank indication, referred to as RI), SINR calculated by the SRS link (SRS SINR) and other data representing the strength of the transmission capability of the user, which is not limited in this embodiment.

Step 102, input the feature parameters into the inner loop machine learning model to obtain the initial value of MCS.

The inner loop means that the base station determines the modulation and coding scheme (Modulation And Coding Scheme, MCS for short) of the uplink and downlink signals of the terminal according to the channel conditions of the terminal.

In some embodiments, the inner loop machine learning model has classification and regression functions, and the inner loop machine learning model may be: a classification neural network, a regression neural network, a decision tree, etc., and this embodiment is not limited thereto.

Exemplary, with reference to shown in Figure 2, feature parameter input is based on the AI-based AMC inner loop module in Figure 2, i.e. inner loop machine learning model, uses multilayer perceptron (Multilayer Perceptron, MLP for short) to come up in this module to the classification function.

The schematic diagram of the MLP neural network is shown in Figure 3, which includes an input layer, multiple hidden layers, and an output layer. The number of input layer nodes is equal to the number of input characteristic parameters. For example, the input characteristic parameters include: the number of transmission streams (LayerNum), the SINR calculated by the CQI link (CQI SINR), the SINR calculated by the PMI, RI, and SRS links (SRS SINR), etc. In this embodiment, the number of hidden layers and the number of nodes in each layer of the MLP neural network can be configured, and the number of nodes in the output layer is equal to the number of different MCS values, for example, 0-27 can be configured.

In this embodiment, the MLP neural network is mainly used for classification tasks: the MLP network outputs predicted values corresponding to 0-27 different MCS values, and the MCS value with the largest predicted value can be used as the initial value of the MCS.

Step 103, input the MCS initial value, characteristic parameters, and the value range of the MCS adjustment value into the outer loop machine learning model to obtain the MCS adjustment value corresponding to the MCS initial value. The outer loop refers to the process of determining the MCS adjustment value.

In some embodiments, the outer loop machine learning model can make decisions based on a series of input parameters, for example, the outer loop machine learning model is: reinforcement learning neural network. The neural network in the AI-based AMC outer-loop module in Figure 2 is the outer-loop machine learning model, and the characteristic parameters can be input into the AI-based AMC outer-loop module to obtain the MCS adjustment value.

Exemplarily, the outer loop machine learning model is a deep Q network (Deep Q network, DQN for short). The DQN network can include multiple convolutional neural networks, fully connected networks, etc. The specific network types, numbers and scales are configurable. The DQN network outputs the Q function value Q(state, action) corresponding to different actions (action) in the state (state). The schematic diagram of the DQN network can be referred to as shown in FIG. 4 .

Referring to Figure 4, the characteristic parameters representing the state, such as LayerNum, CQI SINR, PMI, RI, SRS SINR and MCS initial value, and the MCS adjustment value representing the behavior are input into the enhanced Q network representing the outer ring of AMC (Deep Q Learning, DQN) network. Wherein, the MCS adjustment value may be 0, +1, -1, +2, -2, etc., representing an adjustment value adjusted on the basis of the original MCS initial value. The number of possible values of the input MCS adjustment value is less than the number of possible values of the initial MCS value.

After the DQN network outputs different behaviors, that is, MCS adjustment values 0, -1, +1, -2, +2, etc., and corresponding to different Q function values, the ε-greedy method can be used, that is, a behavior is randomly selected with probability ε Output as the MCS adjustment value; the probability 1-ε selects the behavior with the largest Q function value as the MCS adjustment value output. The ε-greedy method can avoid the selected MCS adjustment value from falling into local optimum.

In some embodiments, after the MCS adjustment value is obtained in the AI-based AMC outer loop, that is, the outer loop machine learning model, both the MCS adjustment value and the MCS initial value can be input into the outer loop module learning module for subsequent outer loop model learning The module obtains the outer ring model sample data according to the MCS adjustment value and the MCS initial value, and trains the outer ring model parameters in the outer ring machine learning model based on the outer ring model data.

Step 104: Obtain an adjusted MCS value according to the adjustment value corresponding to the initial MCS value and the initial MCS value.

In some embodiments, the MCS initial value output by the MLP network is added to the MCS adjustment value output by the DQN network, and the specific formula is MCS=min{max{MCS initial value+MCS adjustment value, MCSmin}, MCSmax}; where, MCSmax is the maximum MCS value, and MCSmin is the minimum MCS value to ensure that the output MCS value will not be greater than MCSmax or less than MCSmin.

After obtaining the adjusted MCS value, the transmission module in Figure 2 performs signal transmission according to the adjusted MCS value, and then feeds back the feedback value of the transmission result, such as Ack/Nack, to the inner loop model learning module, that is, the inner loop machine learning The training module of the model and the learning module of the outer ring model, that is, the training module of the outer ring machine learning model.

In some embodiments, the inner loop model learning module trains the inner loop model parameters in the inner loop machine learning model according to the inner loop model sample data; wherein, the inner loop model sample data includes: adjusted MCS value, A transmission result of signal transmission using the adjusted MCS value, and the characteristic parameter.

In some embodiments, the inner-loop model parameters in the inner-loop machine learning model are trained through offline learning, and the computational complexity of using offline learning is lower than that of online learning. Therefore, this embodiment can further reduce the learning complexity. .

Exemplarily, if the offline update is adopted, the inner-loop model learning module in Figure 2 needs to update the inner-loop model according to the collected samples after receiving a certain amount of sample data of the inner-loop model or reaching a certain period, that is, to update the MLP network parameters , in the training process, the data whose feedback value of the transmission result is Ack is used to generate samples. An inner loop model sample data includes LayerNum, CQI SINR, PMI, RI, SRS SINR and other characteristic parameters, and the adjusted MCS corresponding to these characteristic parameters value, and the transmission result Ack, the label of this sample is the adjusted MCS value.

After obtaining the sample data of the inner ring model, a loss function can be defined, and based on the collected sample data of the inner ring model, the parameters of the inner ring model can be updated using the gradient descent method.

In some embodiments, the outer ring model parameters in the outer ring machine learning model are trained according to the outer ring model sample data; wherein, the outer ring model sample data includes: the characteristic parameters, the MCS initial value, the The MCS adjustment value, the feedback value of the transmission result.

In some embodiments, the parameters of the outer loop model are trained through online learning, and the parameters of the outer loop model are updated in real time through online learning to obtain the MCS adjustment value, so that the MCS can converge faster.

Exemplarily, after receiving the transmission result feedback, it is necessary to update the outer loop model parameters in real time according to the transmission result feedback value, that is, to update the DQN network parameters. The transmission feedback value represents the reward (Reward) of the DQN network after taking a certain behavior. When the transmission result is Ack, the feedback of the transmission result, that is, the return is equal to the spectral efficiency (SE) corresponding to the MCS value; when the transmission result feedback is Nack, the return is 0.

In the process of training the parameters of the outer loop model, according to the sample data of the outer loop model, {the state before taking action, behavior, reward, and the state after taking action} are obtained, and added to an Experience Replay Memory (Experience Replay Memory); Randomly sample some samples from the buffer to form a small-scale training set, and update the DQN network parameters based on these samples. The update method is to calculate the loss function and use the gradient descent method.

The parameters of the inner loop model in the inner loop machine learning model of this embodiment are trained through offline learning; the parameters of the outer loop model in the outer loop machine learning model are trained through online learning. In this embodiment, the artificial intelligence method is applied to the inner loop and the outer loop at the same time, which can not only ensure the online adjustment capability of the AMC outer loop, but also reduce the complexity of the AMC outer loop model through accurate calculation of the AMC inner loop.

In some other embodiments, the outer-loop model learning module and the inner-loop model learning module can also adopt different learning methods. For example, after the outer-loop model learning module converges and stabilizes the parameters of the outer-loop model, it can use an offline learning method to reduce the external The update complexity of ring model parameters. For example, the learning module of the outer ring model can be directly configured as an offline learning method, periodically updating the outer ring model parameters, or triggering the update of the outer ring model parameters, that is, the inner ring model parameters and the outer ring model parameters. The learning method of model parameters can be set according to actual needs.

In this embodiment, the characteristic parameters including the data representing the user's transmission capability are input into the inner-loop machine learning model to obtain the initial value of the MCS, and the initial value of the MCS and the characteristic parameters are input into the outer-loop machine learning model to obtain the adjusted value of the MCS. The value range of the adjustment value is smaller than the value range of the MCS initial value. Therefore, when the value range of the MCS adjustment value is used as the state for training, the behavior space dimension of the outer loop machine learning model is low, and the scale of the neural network is small. , the computational complexity of obtaining the MCS adjustment value is low, thereby improving the efficiency of adaptive coding and modulation. In addition, this embodiment trains the AMC inner-loop model parameters through offline learning, and trains the outer-loop model parameters through online learning, which can ensure The online adjustment capability of the AMC outer loop further reduces the complexity of the AMC outer loop model through the accurate calculation of the AMC inner loop. Moreover, compared with using the characteristic formula to calculate the AMC value, this embodiment calculates the AMC value through the machine learning model. value is more precise.

Embodiments of the present application also provide an adaptive coding and modulation method. The adaptive coding and modulation method in this embodiment can be applied to a multi-user space division scenario. The characteristic parameters obtained in this embodiment also include: interference capability parameters. The flow chart of this embodiment can be seen as shown in Figure 5, including the following steps:

Step 501, detecting the transmission type.

Step 502, if the transmission type is single-user transmission, acquire data representing the transmission capability of the user.

Step 503, if the transmission type is multi-user space division transmission, then acquire data and interference capability parameters representing user transmission capabilities.

In some embodiments, the interference capability parameters include: correlation parameters between interfering users in the space division user pairing combination where the user is located and the user's port, and each interfering user in the space division user pairing combination The number of transport streams interfering with the user. The interference capability parameter is data that can characterize the interference between users in the empty allocation pair combination. The user's inter-port correlation parameter can be the mean value or the maximum value of the inter-port correlation.

In the traditional technical solution, the MCS is often calculated based on a specific formula, that is, the MCS value is obtained by using the CQI link feedback value, the number of interfering user streams, and correlation parameters into a specific calculation formula, and the AMC outer ring maintenance is performed for a single user .

However, in the scenario of multi-user transmission, it is difficult to accurately represent inter-user interference through calculation formulas, resulting in inaccurate calculated MCS values. In addition, the AMC outer loop may not converge when changes in user pairing combinations lead to interference changes.

In view of this, this embodiment can also comprehensively consider the user's own transmission capability and the interference capability between users, not only for single-user transmission scenarios, but also for multi-space division user transmission, so that the output of MCS value and MCS adjustment value is more accurate , in addition, this embodiment does not need to use a specific formula to calculate the inter-user interference value, this embodiment uses the inter-port correlation parameter of the user, and the transmission flow number of each interfering user in the space division user pairing combination as the input of the characteristic parameter, to realize The method is simpler, and the machine learning model can better capture the relationship between the interference between users and the MCS, making the calculated value more accurate and avoiding the interference caused by the AMC outer loop being user-oriented and not distinguishing interference. A situation where the MCS adjustment value does not converge occurs.

Exemplarily, considering multi-user space division transmission, there is interference among space division users, so parameters related to the transmission capability of the primary user and parameters related to the interference capability of the interfering user are selected. Specifically, the input feature combination is constructed as follows:

For a given pairing combination of space-separated users, one user in the combination is regarded as the main user, and other users are regarded as interference users, and the feature combination corresponding to the main user is constructed. For example, the space division combination is user 1, user 2, and user 3. First, user 1 is used as the main user, and other users are used as interference users to construct a feature combination corresponding to user 1.

The number of transmission streams, SRS SINR, CQI SINR, PMI, RI, and the average or maximum value of the correlation between ports of the primary user 1 are selected as the data representing the user's transmission capability, that is, the above data is used as the measurement of the primary user's own transmission capability. Then, select the number of transmission streams of interfering user 2, the average or maximum value of the correlation between the ports of user 2 and user 1; finally, select the number of transmission streams of interfering user 3, the average or maximum value of the correlation between the ports of user 3 and user 1 value. The above interference user parameters are used as data of the interference user interference capability. All the above parameters are arranged in a fixed order as the features of User1.

Using the same method, the features of user 2 and user 3 are respectively constructed, and the features corresponding to different users are input into the subsequent AMC inner loop and outer loop modules one by one.

Step 504, input the feature parameters into the inner loop machine learning model to obtain the initial value of MCS.

Illustratively, if it is multi-user space division transmission, the characteristic parameter input corresponding to a certain main user represents in the multilayer perceptron (MLP, Multilayer Perceptron) neural network of AMC inner loop, the MLP neural network structure and single-user transmission The MLP neural network structure is similar to that shown in Figure 3, except that the input characteristic parameters include the transmission capability parameters of the main user and the interference capability parameters of the interference user.

In this embodiment, the MLP neural network is mainly used for classification tasks, and the MLP network outputs predicted values of different MCS values, and the MCS value with the largest predicted value can be used as the initial value of the primary user's MCS.

Step 505, input the initial MCS value and feature parameters into the outer-loop machine learning model to obtain the adjusted MCS value corresponding to the initial MCS value.

Exemplarily, if it is multi-user space division transmission, the input feature and MCS initial value representing the state, and the MCS adjustment value representing the behavior are input into the strengthened Q network (Deep Q Learning, DQN) network representing the AMC outer ring. Wherein, the MCS adjustment value may be 0, +1, -1, +2, -2, etc., representing an adjustment value adjusted on the basis of the original MCS initial value.

The DQN network structure is similar to the DQN network for single-user transmission, the difference is that the environmental parameters become environmental parameters in a multi-user scenario, including the transmission capability parameters of the primary user and the interference capability parameters of the interfering user.

After the DQN network outputs different Q function values corresponding to different behaviors (ie, MCS adjustment values 0, -1, +1, -2, +2, etc.), we use the ε-greedy method: randomly select a behavior with probability ε as MCS adjustment value output; the probability 1-ε selects the behavior with the largest Q function value as the MCS adjustment value output.

The MCS initial value and MCS adjustment value need to be transmitted to the outer loop model learning module in Figure 2.

In step 506, an adjusted MCS value is obtained according to the adjustment value corresponding to the initial MCS value and the initial MCS value.

In this embodiment, step 505 is substantially the same as step 103, and in order to avoid repetition in expression, details are not repeated in this embodiment.

In some embodiments, according to the inner-ring model sample data of each user in the user pairing group where the user is located, the inner-ring model parameters in the inner-ring machine learning model are trained; The outer ring model sample data of each user in the user pair combination trains the outer ring model parameters in the outer ring machine learning model.

The training method of the inner loop model parameters in this embodiment is roughly the same as the above embodiment. It is worth mentioning that since this embodiment supports multi-user space separation scenarios, feedback results from multiple space separation users will be received after each scheduling , so multiple inner-ring sample data corresponding to different main users in the space-separation user pairing combination will be generated. These inner-ring sample data will be stored by the inner-ring model learning module and wait for subsequent MLP network parameter update algorithm calls.

The training method of the outer ring model parameters in this embodiment is roughly the same. Since this embodiment supports multi-user space separation scenarios, this embodiment will obtain feedback results from multiple space separation users, so multiple outer ring model samples corresponding to different main users will be obtained. These samples will be put into the experience playback buffer at the same time, waiting for the subsequent call of the DQN network parameter update algorithm.

The model data is obtained by training the sample data of each user in the same space separation user pairing combination, which is more conducive to considering the interference factors between users, so that in the case of space separation, the MCS value can also gradually converge to the optimum, and further speed up convergence speed.

It is worth mentioning that in this embodiment, the transmission type is detected, different characteristic parameters are obtained for different transmission types, training is performed, and the adjusted MCS value is obtained. In the actual application process, it can also only be used for single-user transmission scenarios or Only for multi-user air separation scenarios.

In this embodiment, the parameters of the inner ring model are obtained by learning the successful transmission experience in history, and the initial value of the MCS is calculated through the inner ring model parameters and characteristic parameters, which can more accurately estimate the initial value of the MCS under the current environment parameters, and avoid single-user or multi-user In the case of space-division transmission, the conversion of the inner ring is inaccurate; moreover, this embodiment comprehensively considers the impact of various factors on the outer ring, such as the parameters related to the interference ability of the interfering user in a multi-user scenario to obtain the outer ring model parameters, and pass The outer ring model parameters and user characteristic parameters including interference ability parameters calculate the adjustment value, which can more accurately estimate the MCS adjustment value in the current environment, and can avoid the outer ring only focusing on a single user and ignoring other dimensions, such as interference caused by users non-convergence phenomenon; in addition, the AMC outer loop of this embodiment can be adjusted online, and the AMC inner loop can be adjusted in real time, which can not only ensure the online adjustment ability of the AMC outer loop, but also reduce the AMC outer loop model through the accurate calculation of the AMC inner loop. The complexity makes the output of AMC's inner and outer rings more accurate.

The step division of the above various methods is only for the sake of clarity of description. During implementation, it can be combined into one step or some steps can be split and decomposed into multiple steps. As long as they include the same logical relationship, they are all within the scope of protection of this patent. ; Adding insignificant modifications or introducing insignificant designs to the algorithm or process, but not changing the core design of the algorithm and process are all within the scope of protection of this patent.

The embodiment of the present application also provides an adaptive coding and modulation device, as shown in FIG. 6 , including: a parameter acquisition module 601, configured to acquire characteristic parameters, wherein the characteristic parameters include data representing the user's transmission capability; The initial value acquisition module 602 is used to input the characteristic parameters into the inner loop machine learning model to obtain the MCS initial value; the adjustment value acquisition module 603 is used to input the MCS initial value and the characteristic parameters into the outer loop machine learning model to obtain an adjusted MCS value corresponding to the initial MCS value; an adjustment module 604 configured to obtain an adjusted MCS value according to the adjusted value corresponding to the initial MCS value and the initial MCS value.

In some embodiments, the parameter acquisition module 601 is further configured to detect the transmission type; wherein, the transmission type includes: single-user transmission and/or multi-user space division transmission; if the transmission type is multi-user space division transmission, then The characteristic parameter further includes: an interference capability parameter; wherein, the interference capability parameter represents data of an interference capability of an interfering user.

In some embodiments, the interference capability parameters acquired by the parameter acquisition module 601 include: the correlation parameters between the ports of each interfering user in the space division user pairing combination where the user is located and the user, and the space division user pairing The number of transmit streams for each interfering user in the combination.

In some embodiments, the adaptive coding and modulation device further includes a model training module, and the model training module is further used to train the inner-loop model parameters in the inner-loop machine learning model according to the inner-loop model sample data; wherein, the The inner loop model sample data includes: the adjusted MCS initial value, the feedback value of the transmission result of signal transmission using the adjusted MCS value, and the characteristic parameters; the outer loop machine learning is trained according to the outer loop model sample data Outer loop model parameters in the model; wherein, the outer loop model sample data includes: the characteristic parameters, the initial MCS value, the MCS adjustment value, and the feedback value of the transmission result.

In some embodiments, when the transmission type is multi-user space division transmission, the model training module is further used to train the inner loop machine according to the inner loop model sample data of each user in the space division user pairing combination where the user is located Inner-ring model parameters in the learning model; training outer-ring model parameters in the outer-ring machine learning model according to the outer-ring model sample data of each user in the space-separated user pairing combination where the user is located.

In some embodiments, the inner loop machine learning model in the initial value acquisition module 602 is any of the following classification neural network, regression neural network, decision tree; the outer loop machine learning model in the adjustment value acquisition module 603 is reinforcement learning neural network network.

In some embodiments, the parameters of the inner ring model in the inner ring machine learning model in the initial value acquisition module 602 are trained through offline learning; the outer ring model in the outer ring machine learning model in the adjustment value acquisition module 603 The parameters are trained by online learning.

It is not difficult to find that this embodiment is a system embodiment corresponding to the above-mentioned method embodiment, and this embodiment can be implemented in cooperation with the above-mentioned method embodiment. The relevant technical details mentioned in the foregoing method embodiments are still valid in this embodiment, and will not be repeated here in order to reduce repetition. Correspondingly, the relevant technical details mentioned in this embodiment can also be applied in the first embodiment.

It is worth mentioning that all the modules involved in this embodiment are logical modules. In practical applications, a logical unit can be a physical unit, or a part of a physical unit, or multiple physical units. Combination of units. In addition, in order to highlight the innovative part of the present application, units that are not closely related to solving the technical problem proposed in the present application are not introduced in this embodiment, but this does not mean that there are no other units in this embodiment.

The embodiment of the present application also provides an electronic device, as shown in FIG. 7 , including at least one processor 701; and a memory 702 connected in communication with the at least one processor 701; Instructions executed by the at least one processor 701, the instructions executed by the at least one processor 701, so that the at least one processor 701 can execute the above adaptive coding and modulation method.

Wherein, the memory and the processor are connected by a bus, and the bus may include any number of interconnected buses and bridges, and the bus connects one or more processors and various circuits of the memory together. The bus may also connect together various other circuits such as peripherals, voltage regulators, and power management circuits, all of which are well known in the art and therefore will not be further described herein. The bus interface provides an interface between the bus and the transceivers. A transceiver may be a single element or multiple elements, such as multiple receivers and transmitters, providing means for communicating with various other devices over a transmission medium. The data processed by the processor is transmitted on the wireless medium through the antenna, further, the antenna also receives the data and transmits the data to the processor.

The processor is responsible for managing the bus and general processing, and can also provide various functions, including timing, peripheral interface, voltage regulation, power management, and other control functions. Instead, memory can be used to store data that the processor uses when performing operations.

Embodiments of the present application also provide a computer-readable storage medium storing a computer program. The above method embodiments are implemented when the computer program is executed by the processor.

That is, those skilled in the art can understand that all or part of the steps in the method of the above-mentioned embodiments can be completed by instructing related hardware through a program, the program is stored in a storage medium, and includes several instructions to make a device ( It may be a single-chip microcomputer, a chip, etc.) or a processor (processor) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .

Those of ordinary skill in the art can understand that the above-mentioned embodiments are specific embodiments for realizing the present application, and in practical applications, various changes can be made to it in form and details without departing from the spirit and spirit of the present application. scope.

Claims

An adaptive coding and modulation method, comprising:

Acquiring characteristic parameters, wherein the characteristic parameters include data representing the user's transmission capability;

The feature parameter is input into the inner loop machine learning model to obtain the initial value of MCS;

Inputting the MCS initial value and the characteristic parameters into the outer loop machine learning model to obtain the MCS adjustment value corresponding to the MCS initial value;

An adjusted MCS value is obtained according to the adjustment value corresponding to the initial MCS value and the initial MCS value.
The adaptive coding and modulation method according to claim 1, wherein, before obtaining the characteristic parameters, comprising:

Detecting a transmission type; wherein, the transmission type includes: single-user transmission and/or multi-user space division transmission;

If the transmission type is the multi-user space division transmission, the acquired characteristic parameter further includes: an interference capability parameter; wherein the interference capability parameter represents data of an interference capability of an interfering user.
The adaptive coding and modulation method according to claim 2, wherein the interference capability parameter comprises: a correlation parameter between each interfering user in the space-division user pairing combination where the user is located and the port of the user, and The number of transmission streams of each interfering user in the space-division user pair combination.
The adaptive coding and modulation method according to claim 2, wherein, after obtaining the adjusted MCS value according to the adjustment value corresponding to the initial MCS value and the initial MCS value, further comprising:

Train the inner-ring model parameters in the inner-ring machine learning model according to the inner-ring model sample data; wherein, the inner-ring model sample data includes: the adjusted MCS value, and signal using the adjusted MCS value The feedback value of the transmitted transmission result and the characteristic parameters;

The outer ring model parameters in the outer ring machine learning model are trained according to the outer ring model sample data; wherein, the outer ring model sample data includes: the characteristic parameters, the MCS initial value, the MCS adjustment value and the set The feedback value of the above transmission result.
The adaptive coding and modulation method according to claim 4, wherein, if the transmission type is the multi-user space division transmission, the inner loop in the inner loop machine learning model is trained according to the inner loop model sample data Model parameters, including:

According to the inner ring model sample data of each user in the space separation user pairing combination where the user is located, train the inner ring model parameters in the inner ring machine learning model;

The training of the outer ring model parameters in the outer ring machine learning model according to the outer ring model sample data includes:

The outer ring model parameters in the outer ring machine learning model are trained according to the outer ring model sample data of each user in the space division user pair combination where the user is located.
The adaptive coding and modulation method according to any one of claims 1 to 5, wherein the inner loop machine learning model comprises any of the following: a classification neural network, a regression neural network, a decision tree;

The outer loop machine learning model includes a reinforcement learning neural network.
The adaptive coding and modulation method according to any one of claims 1 to 6, wherein, the inner loop model parameters in the inner loop machine learning model are trained by offline learning; the outer loop machine learning model parameters The parameters of the outer ring model are trained through online learning.
An adaptive coding and modulation device, comprising:

A parameter acquisition module, configured to acquire characteristic parameters, wherein the characteristic parameters include data representing the user's transmission capability;

The initial value acquisition module inputs the characteristic parameters into the inner loop machine learning model to obtain the MCS initial value;

An adjustment value acquisition module, configured to input the MCS initial value and the characteristic parameters into the outer loop machine learning model to obtain an MCS adjustment value corresponding to the MCS initial value;

An adjustment module, configured to obtain an adjusted MCS value according to the adjustment value corresponding to the initial MCS value and the initial MCS value.
An electronic device comprising:

at least one processor; and,

a memory communicatively coupled to the at least one processor; wherein,

The memory is stored with instructions executable by the at least one processor, the instructions are executed by the at least one processor, so that the at least one processor can perform any one of claims 1 to 7 Adaptive coding and modulation method.
A computer-readable storage medium storing a computer program, which implements the adaptive coding and modulation method according to any one of claims 1 to 7 when the computer program is executed by a processor.