A channel state prediction method and system based on LSTM

Abstract

The present invention discloses a channel state prediction method and system based on LSTM, which obtains multiple rounds of historical channel state data as input of LSTM predictor; obtains training sets according to the multiple rounds of historical channel state data, and in the LSTM predictor, uses the training sets to train the training model multiple times to obtain a prediction model; in the LSTM predictor, uses the multiple rounds of historical channel state data and the prediction model to obtain the current SNR value, and outputs the current SNR value as the channel state prediction result. The present invention is used to accurately fit the nonlinear mapping relationship between multi-dimensional channel state information and channel SNR values, and realizes that it is not dependent on the channel estimation value at the satellite terminal in real time, so that it is not strictly subject to the ACM loop delay, and can alleviate the influence of the ACM loop delay in DVB‑S2 on the adaptability of the satellite communication system.

Description

Technical Field

[0001] The present invention relates to the technical field of channel state prediction, and in particular to a channel state prediction method and system based on LSTM. Background Art

[0002] To improve the spectrum utilization of the system, the DVB-S2 standard supports adaptive coding and modulation (ACM). The ACM gateway refers to the measured channel status information transmitted by the return channel and dynamically adjusts the coding rate and modulation mode for each data frame.

[0003] The changes in satellite channels and ACM loop delay are the main challenges of ACM technology. In satellite communication systems, loop transmission delay and data processing delay are both over 500ms. If severe rain attenuation and shadowing effects are taken into account, satellite channels will experience severe attenuation in extreme times. At this time, loop delay will cause the misalignment of channel state information, so that the selected coding modulation method does not match the channel state at that moment, causing transmission errors and reduced throughput, which will seriously affect the satellite communication system's ability to track channel changes.

[0004] In view of the above problems, if the signal-to-noise ratio (SNR) value is predicted at the ACM gateway and does not rely on the channel estimation value at the satellite terminal in real time, the influence of the loop delay on the adaptive performance of the satellite communication system can be eliminated. The existing technical solution uses the least mean square algorithm (LMS) to predict the SNR value at the ACM gateway. This technical solution predicts the SNR value at time t+1 by the SNR value at the previous t moments, and obtains the weight vector value through multiple iterations. However, in this technical solution, the channel SNR value is the only decision condition for the modulation and coding method, and the satellite channel capacity is not only affected by the channel SNR value, but also by factors such as spatial propagation distance loss and shadow fading. In response to this, the industry uses multi-dimensional channel state information (CSI) to predict the SNR value, taking into account the influence of various influencing factors on the channel capacity, so that the selection of the modulation and coding method will be more adapted to the channel state, thereby improving the channel capacity. However, there is a nonlinear mapping relationship between multi-dimensional CSI data and SNR values, and there is currently no exact model to characterize it. The traditional numerical method cannot accurately fit the nonlinear mapping relationship between multi-dimensional CSI data and SNR values.

[0005] In summary, the existing technology cannot accurately predict the signal-to-noise ratio value at the ACM gateway, which affects the adaptive performance of the ACM loop delay to the satellite communication system.

[0006] In view of this, this application is hereby filed. Summary of the invention

[0007] The technical problem to be solved by the present invention is that the existing technology cannot accurately predict the signal-to-noise ratio value at the ACM gateway. The purpose is to provide a channel state prediction method based on LSTM, use a deep learning method to predict the channel state, fit the nonlinear mapping relationship between multi-dimensional channel state information and channel SNR value, and accurately predict the SNR value at the ACM gateway, improve the adaptive performance of the ACM loop delay to the satellite communication system, and further improve the system throughput.

[0008] The present invention is achieved through the following technical solutions:

[0009] On the one hand, the present invention provides a channel state prediction method based on LSTM, and the following steps are performed in an ACM link introducing an LSTM predictor:

[0010] Acquire multiple rounds of historical channel state data as input of the LSTM predictor, wherein the historical channel state data includes: SNR value, transmission delay, weather and environmental state values;

[0011] Acquire a training set according to the multiple rounds of historical channel state data, and in the LSTM predictor, use the training set to train the training model multiple times to obtain a prediction model;

[0012] In the LSTM predictor, the multiple rounds of historical channel state data and the prediction model are used to obtain a current SNR value, and the current SNR value is output as a channel state prediction result.

[0013] The present invention proposes an improvement on the existing SNR value prediction using the minimum mean square error algorithm. In view of the fact that the channel SNR value is the only decision condition for the modulation and coding method in the prior art, and the satellite channel capacity is not only affected by the channel SNR value, but also by factors such as space propagation distance loss and shadow fading, the present invention considers using multi-dimensional channel state information to predict the SNR value, taking into account the influence of various influencing factors on the channel capacity, and on this basis, introducing the LSTM predictor into the traditional ACM link, and using the cyclic characteristics of the LSTM predictor to process the correlation between multiple rounds of multi-dimensional channel state information. In addition, considering the nonlinear mapping relationship between the multi-dimensional channel state information and the SNR value, it is proposed to use a deep learning method to solve the problem that the traditional numerical method cannot accurately fit the nonlinear mapping relationship, by storing a certain number of rounds of channel state data in the ACM gateway as the input of the LSTM predictor, repeatedly training the gradient descent algorithm optimization model in the LSTM predictor, obtaining a prediction model, substituting multiple rounds of channel state data into the prediction model for prediction to obtain the SNR value, that is, obtaining the channel state prediction result.

[0014] Therefore, the LSTM-based channel state prediction method provided by the present invention does not rely on the channel estimation value at the satellite terminal in real time, and is thus not strictly subject to the ACM loop delay. It can alleviate the influence of the ACM loop delay in DVB-S2 on the adaptability of the satellite communication system, and predicts the channel state by a deep learning method, which can fit the nonlinear mapping relationship between the multi-dimensional channel state information and the channel SNR value.

[0015] As a further description of the present invention, the process of training the training model multiple times using the training set includes the following steps:

[0016] Acquire a training loss value and a test loss value in real time, determine a stop condition for training according to the training loss value and the test loss value, and stop training according to the stop condition;

[0017] The model with the smallest test loss value is saved in real time.

[0018] As a further description of the present invention, the termination conditions include:

[0019] If the test loss value does not decrease for several consecutive times, the training is stopped;

[0020] If the test loss value is higher than the training loss value for multiple consecutive times, the training is stopped.

[0021] As a further description of the present invention, the loss function used to obtain the training loss value and the test loss value is a mean square error function; the training model model is a gradient descent algorithm optimization model.

[0022] As a further description of the present invention, the LSTM-based channel state prediction method further includes the following steps:

[0023] Obtain new transmission delay, weather and environmental status values, and input the new transmission delay, weather, environmental status values ​​and the current SNR value into the LSTM predictor as a new round of channel status data;

[0024] Using the ACM gateway router to look up a table to obtain a MODCOD value that matches the current SNR value;

[0025] The modulation and coding method is obtained from the MODCOD value, and the ACM modulator modulates and encodes the transmission information.

[0026] As a further description of the present invention, the LSTM-based channel state prediction method further includes the following steps:

[0027] Obtaining a validation set, and using the validation set to perform performance verification on the prediction model;

[0028] The performance verification method is:

[0029] Under rain attenuation conditions, obtaining the system throughput of the ACM link introduced with the LSTM predictor by using the verification set;

[0030] Compare the system throughput of the ACM link introduced with the LSTM predictor with the system throughput of the traditional ACM link to obtain a comparison result;

[0031] If the comparison result is: the system throughput of the ACM link introducing the LSTM predictor is increased by ≥10% compared with the system throughput of the traditional ACM link, the prediction model is determined to be valid; if the comparison result is: the system throughput of the ACM link introducing the LSTM predictor is increased by <10% compared with the system throughput of the traditional ACM link, the data amount of the training set is increased and the number of rounds of obtaining historical channel status data is adjusted.

[0032] As a further description of the present invention, the LSTM-based channel state prediction method further includes the following steps:

[0033] Obtain fine-tuning datasets for different application scenarios, and use the fine-tuning datasets to fine-tune the prediction model.

[0034] Obtain a validation set to verify the performance of the fine-tuned prediction model.

[0035] On the other hand, the present invention provides a LSTM-based channel state prediction system, comprising a traditional ACM link and an LSTM predictor, wherein the LSTM predictor is connected in the traditional ACM link;

[0036] The LSTM predictor includes:

[0037] A data storage module, used to temporarily store multiple rounds of historical channel status data, wherein the historical channel status data includes: SNR value, transmission delay, weather and environmental status values;

[0038] The LSTM model module is used to obtain the current SNR value according to the multiple rounds of historical channel state data, and output the current SNR value as a channel state prediction result.

[0039] As a further description of the present invention, the LSTM predictor includes: a model verification module, which is used to obtain a verification set and use the verification set to perform performance verification on the prediction model;

[0040] The model verification module includes:

[0041] A system throughput calculation unit, used for obtaining the system throughput of the ACM link introduced with the LSTM predictor under rain attenuation conditions by using the verification set;

[0042] A data analysis unit, used for comparing the system throughput of the ACM link introduced with the LSTM predictor with the system throughput of the traditional ACM link to obtain a comparison result;

[0043] A result determination unit is used to determine the effectiveness of the prediction system according to the comparison result.

[0044] As a further description of the present invention, the LSTM predictor includes: a model fine-tuning module, which is used to fine-tune the prediction data according to different application scenarios to obtain prediction models under different application scenarios; the model fine-tuning module is connected to the model verification module.

[0045] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0046] 1. An LSTM-based channel state prediction method and system provided in an embodiment of the present invention uses a deep learning method to predict the channel state, and can fit the nonlinear mapping relationship between multi-dimensional channel state information and channel SNR value;

[0047] 2. An LSTM-based channel state prediction method and system provided in an embodiment of the present invention takes transmission delay, weather conditions, and environmental conditions as the basis for channel state prediction, and takes the influence of free space propagation loss, shadow fading, and rain fading on channel capacity into consideration in the prediction process of the predictor, so that the predicted SNR value is more fully adapted to the channel state, so that the selected MODCOD value is more adapted to the channel state, further improving the throughput;

[0048] 3. An LSTM-based channel state prediction method and system provided in an embodiment of the present invention uses an LSTM network to build a prediction model, and uses the correlation between multiple rounds of channel states to track channel changes, thereby achieving non-real-time dependence on the channel estimation value at the satellite terminal, and thus not strictly subject to the ACM loop delay, which can alleviate the impact of the ACM loop delay in DVB-S2 on the adaptability of the satellite communication system;

[0049] 4. The LSTM-based channel state prediction method and system provided in the embodiment of the present invention can improve the effectiveness of the prediction model and the prediction accuracy by testing the prediction model, thereby improving the adaptability of the selected MODCOD value to the channel state and further improving the throughput;

[0050] 5. The embodiment of the present invention provides a LSTM-based channel state prediction method and system, which can fine-tune the prediction model so that the prediction model can be applied to different practical scenarios, thereby reducing the time consumption of data collection and training process as a whole, and has high flexibility and high adaptability. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0052] Figure 1 ACM technology link diagram using LSTM predictor provided in Example 1 of the present invention;

[0053] Figure 2 A schematic diagram of the structure of the LSTM predictor provided in Example 1 of the present invention;

[0054] Figure 3 A schematic diagram of the expanded calculation logic of the LSTM predictor provided in Example 1 of the present invention. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0056] Example 1

[0057] Since there is delay in loop transmission and data processing in satellite communication systems, if severe rain attenuation and shadowing effects are taken into account, the satellite channel will experience severe attenuation in extreme times. At this time, loop delay will cause the misalignment of channel state information, so that the selected coding modulation method does not match the channel state at that moment, causing transmission errors and reduced throughput, which will seriously affect the satellite communication system's ability to track channel changes.

[0058] Eliminating the influence of loop delay on the adaptive performance of satellite communication system At present, the industry considers predicting the signal-to-noise ratio (SNR) value at the ACM gateway, so as not to rely on the channel estimation value at the satellite terminal in real time. In this regard, the existing technical solution adopted is to use the least mean square algorithm (LMS) to predict the SNR value at the ACM gateway. This technical solution predicts the SNR value at time t+1 by the SNR value at the previous t moments, and obtains the weight vector value through multiple iterations. However, in this technical solution, the channel SNR value is the only decision condition for the modulation and coding method, and the satellite channel capacity is not only affected by the channel SNR value, but also by factors such as space propagation distance loss and shadow fading. In this regard, the industry uses multi-dimensional channel state information (CSI) to predict the SNR value, taking into account the influence of various influencing factors on the channel capacity, so that the selection of the modulation and coding method will be more adapted to the channel state, thereby improving the channel capacity. However, there is a nonlinear mapping relationship between the multi-dimensional CSI data and the SNR value. However, there is no exact model to characterize it, and the traditional numerical method cannot accurately fit the nonlinear mapping relationship between multi-dimensional CSI data and SNR values.

[0059] In response to the above problems, this embodiment provides a channel state prediction method based on LSTM, considers using multi-dimensional channel state information to predict the SNR value, takes into account the influence of various influencing factors on the channel capacity, introduces the LSTM predictor into the traditional ACM link, and uses the cyclic characteristics of the LSTM predictor to process the correlation between multiple rounds of multi-dimensional channel state information. In addition, it is proposed to use a deep learning method to solve the problem that traditional numerical methods cannot accurately fit nonlinear mapping relationships. By storing a certain number of rounds of channel state data in the ACM gateway as the input of the LSTM predictor, the gradient descent algorithm optimization model is repeatedly trained in the LSTM predictor to obtain a prediction model, and the multiple rounds of channel state data are substituted into the prediction model for prediction to obtain the SNR value, that is, the channel state prediction result is obtained.

[0060] Specifically, the method is implemented in an ACM link that introduces an LSTM predictor, and includes the following steps:

[0061] Step 1: Receive multiple rounds of historical channel state data through the satellite terminal as the input of the LSTM predictor.

[0062] The ACM link of the LSTM predictor is as follows Figure 1 As shown. Use the satellite terminal to receive the historical channel state data sent by the satellite, including SNR value, transmission delay, weather and environmental state values. The received historical signal state data is transmitted back to the ACM gateway through the return channel for temporary storage. Repeat the above method for multiple rounds, and input the stored historical channel state data into the LSTM predictor.

[0063] Step 2: In the LSTM predictor, the training model is trained multiple times using the training set to obtain a prediction model.

[0064] The structure diagram of the LSTM predictor is as follows Figure 2 As shown in the figure, the structure of LSTM is represented by LSTM(k, num), where k represents the number of LSTM cycles, which is the same as the number of cycles of the input channel state data. The expanded calculation diagram of the LSTM predictor is shown in Figure 3 As shown, num represents the number of LSTM units in the LSTM predictor.

[0065] In the training phase, the multiple rounds of historical channel state data input into the LSTM predictor are divided into a training set and a test set, and the ratio of the number of samples in the training set to the number of samples in the test set is 4:1. In this embodiment, the initial number of training sets is set to 2000. The training set is represented by the data structure (n, k, 4). Among them, n represents the number of input data, k represents an input data containing the channel state data of the previous k rounds, and the initial value of k is set to 5 here; 4 represents that the channel state data of each round is 4-dimensional data, namely SNR value, transmission delay, weather status, and environmental status; the size of the label is (n, 1), which is the accurate value of the channel SNR for this round.

[0066] It should be noted that the larger the k value, the higher the complexity of the corresponding model, and there is not much correlation between channel state data with longer intervals, so the k value should not be set too large; however, if the k value is too small, the model cannot fully capture the correlation between the channel state data of each round, and thus cannot accurately predict the SNR value. Therefore, the selection of the k value is determined by the model performance corresponding to different k values.

[0067] In the LSTM predictor, the training model is trained using the training set.

[0068] During the training process, the performance of the temporary model obtained through training is tested in real time using the test set to prevent the model from overfitting, and the prediction model with the lowest test loss value is saved in real time during the training process. On the one hand, the training model used in this embodiment is a small batch gradient descent algorithm optimization model; the initial value of the batch is set to 50, which specifically refers to the loss value curve during the training process. If the loss value curve fluctuates greatly, consider increasing the batch value; if the iteration time is too long, consider reducing the batch value. On the other hand, when to end the training is determined by the test loss value and the training loss value. This embodiment stipulates that when the test loss value does not decrease for 5 consecutive times or the test loss value is higher than the training loss value for 5 consecutive times, the training is stopped, and the maximum training test is set to 200 times. On the third hand, the temporary prediction model with the lowest test loss value is saved in real time, and the current optimal prediction model is found in this iterative manner.

[0069] Step 3: In the LSTM predictor, the current SNR value is obtained using the multiple rounds of historical channel state data and the prediction model, and the current SNR value is output as a channel state prediction result.

[0070] As of step 3, this embodiment has obtained the channel state prediction result according to the prediction model and the channel state information. In order to further verify the validity of the prediction model and thus evaluate the accuracy of the channel state prediction result, this embodiment also verifies the validity of the prediction model through step 4.

[0071] Step 4: Obtain a validation set, and use the validation set to verify the performance of the prediction model;

[0072] The performance verification method is:

[0073] Under rain attenuation conditions, obtaining the system throughput of the ACM link introduced with the LSTM predictor by using the verification set;

[0074] Compare the system throughput of the ACM link introduced with the LSTM predictor with the system throughput of the traditional ACM link to obtain a comparison result;

[0075] If the comparison result is: the system throughput of the ACM link introducing the LSTM predictor is increased by ≥10% compared with the system throughput of the traditional ACM link, the prediction model is determined to be valid; if the comparison result is: the system throughput of the ACM link introducing the LSTM predictor is increased by <10% compared with the system throughput of the traditional ACM link, the data amount of the training set is increased and the number of rounds of obtaining historical channel status data is adjusted.

[0076] In this embodiment, the size of the validation set is set to 200, and the performance of the trained LSTM predictor is verified using the validation set. The system throughput of the ACM link using the predictor is calculated and compared with the system throughput of the traditional ACM link not using the predictor. If the throughput is improved by more than 10%, it means that the predictor has excellent performance.

[0077] Furthermore, in order to generalize and apply the prediction models in different transmission scenarios to other similar scenarios, this embodiment further fine-tunes the prediction model through step 5.

[0078] Step 5: Obtain fine-tuning data sets for different application scenarios. At this time, the size of the data set is much smaller than that in the training stage and can be reduced to 500. Through fine-tuning training, the weights of the trained LSTM predictor are adjusted for this application scenario using the training data set, and the LSTM predictor is extended to other transmission scenarios. In addition, the prediction model obtained after fine-tuning is also verified according to the method in step 4, and the system throughput value is calculated and compared with the throughput value of the traditional ACM link.

[0079] Step 6: Obtain new transmission delay, weather and environmental status values, and input the new transmission delay, weather, environmental status values ​​and the current SNR value into the LSTM predictor as a new round of channel status data.

[0080] Step 7: Use the ACM gateway router to look up a table to obtain a MODCOD value that matches the current SNR value.

[0081] Step 8: Use the ACM modulator to modulate the MODCOD value and send the modulated code to the satellite.

[0082] The LSTM predictor having the prediction model obtained through steps 2 to 5 above is used in the ACM link. Compared with the traditional ACM link, since the SNR value is predicted in real time, it does not rely on the estimated SNR value with loop delay, and the modulation and coding method used is selected, which can better achieve the channel capacity at this time; since the throughput improvement threshold set in the verification stage is 10%, the system throughput using the predictor is improved by at least 10%.

[0083] Example 2

[0084] This embodiment provides a LSTM-based channel state prediction system, including a traditional ACM link and an LSTM predictor, wherein the LSTM predictor is connected to the traditional ACM link;

[0085] The LSTM predictor includes:

[0086] A data storage module, used to temporarily store multiple rounds of historical channel status data, wherein the historical channel status data includes: SNR value, transmission delay, weather and environmental status values;

[0087] A model training module, used to obtain a training set according to the multiple rounds of historical channel state data, and train the training model multiple times using the training set to obtain a prediction model;

[0088] The LSTM model module is used to obtain the current SNR value according to the multiple rounds of historical channel state data, and output the current SNR value as a channel state prediction result.

[0089] in,

[0090] The LSTM predictor includes: a model verification module, which is used to obtain a verification set and use the verification set to perform performance verification on the prediction model;

[0091] The model verification module includes:

[0092] A system throughput calculation unit, used for obtaining the system throughput of the ACM link introduced with the LSTM predictor under rain attenuation conditions by using the verification set;

[0093] A data analysis unit, used for comparing the system throughput of the ACM link introduced with the LSTM predictor with the system throughput of the traditional ACM link to obtain a comparison result;

[0094] A result determination unit is used to determine the effectiveness of the prediction system according to the comparison result.

[0095] The LSTM predictor includes: a model fine-tuning module, which is used to fine-tune the prediction data according to different application scenarios to obtain prediction models under different application scenarios; the model fine-tuning module is connected to the model verification module.

[0096] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A channel state prediction method based on LSTM, characterized in that: The following steps are performed in the ACM link that introduces the LSTM predictor: Acquire multiple rounds of historical channel state data as input of the LSTM predictor, wherein the historical channel state data includes: SNR value, transmission delay, weather and environmental state values; Using a portion of the multiple rounds of historical channel state data as a training set, and in the LSTM predictor, training the training model multiple times using the training set to obtain a prediction model; In the LSTM predictor, the current SNR value is obtained by using the multiple rounds of historical channel state data and the prediction model, and the current SNR value is output as a channel state prediction result; The following steps are also included: Obtain new transmission delay, weather and environmental status values, and input the new transmission delay, weather, environmental status values ​​and the current SNR value into the LSTM predictor as a new round of channel status data; Using the ACM gateway router to look up a table to obtain a MODCOD value that matches the current SNR value; The modulation and coding mode is obtained from the MODCOD value, and the ACM modulator modulates and codes the transmission information; Obtaining a validation set, and using the validation set to perform performance verification on the prediction model; The performance verification method is: Under rain attenuation conditions, obtaining a system throughput of the ACM link using the verification set; Comparing the system throughput of the ACM link with the system throughput of a traditional ACM link to obtain a comparison result, wherein the traditional ACM link does not introduce the LSTM predictor; If the comparison result is: the system throughput of the ACM link is increased by ≥10% compared with the system throughput of the traditional ACM link, the prediction model is determined to be valid; if the comparison result is: the system throughput of the ACM link is increased by <10% compared with the system throughput of the traditional ACM link, the data amount of the training set is increased and the number of rounds for obtaining historical channel status data is adjusted.

2. The channel state prediction method based on LSTM according to claim 1, characterized in that: The process of training the training model multiple times using the training set includes the following steps: Acquire a training loss value and a test loss value in real time, determine a stop condition for training according to the training loss value and the test loss value, and stop training according to the stop condition; The model with the smallest test loss value is saved in real time.

3. The channel state prediction method based on LSTM according to claim 2, characterized in that: The stop conditions include: If the test loss value does not decrease for several consecutive times, the training is stopped; If the test loss value is higher than the training loss value for multiple consecutive times, the training is stopped.

4. The channel state prediction method based on LSTM according to claim 2, characterized in that: The loss function used to obtain the training loss value and the test loss value is a mean square error function; the training model is a gradient descent algorithm optimization model.

5. The channel state prediction method based on LSTM according to claim 1, characterized in that: The following steps are involved: Obtain fine-tuning datasets for different application scenarios, and use the fine-tuning datasets to fine-tune the prediction model. Obtain a validation set to verify the performance of the fine-tuned prediction model.

6. A channel state prediction system based on LSTM, including an ACM link, characterized in that: include: An LSTM predictor, the LSTM predictor being connected in the ACM link and configured to execute the LSTM-based channel state prediction method according to any one of claims 1 to 5; The LSTM predictor includes: A data storage module, used to temporarily store multiple rounds of historical channel status data, wherein the historical channel status data includes: SNR value, transmission delay, weather and environmental status values; An LSTM model module is used to obtain a current SNR value according to the multiple rounds of historical channel state data, and output the current SNR value as a channel state prediction result; The LSTM predictor includes: a model verification module, which is used to obtain a verification set and use the verification set to perform performance verification on the prediction model; the model verification module includes: A system throughput calculation unit, configured to obtain the system throughput of the ACM link under rain attenuation conditions using the verification set; A data analysis unit, used for comparing the system throughput of the ACM link with the system throughput of a traditional ACM link to obtain a comparison result; A result determination unit is used to determine the effectiveness of the prediction system according to the comparison result.

7. The LSTM-based channel state prediction system according to claim 6, characterized in that: The LSTM predictor includes: a model fine-tuning module, which is used to fine-tune the prediction model according to the fine-tuning data sets in different application scenarios to obtain the prediction models in different application scenarios; the model fine-tuning module is connected to the model verification module.

Images