CN114580279B - An adaptive coding method for low-orbit satellite communication based on LSTM - Google Patents

Abstract

The invention discloses a low-orbit satellite communication self-adaptive coding method based on LSTM, which is characterized by comprising the following steps: step 1, ground equipment acquires local real-time weather and longitude and latitude; step 2, generating satellite position information according to satellite orbit information stored in ground equipment, and representing the satellite position information by three-dimensional coordinates; step 3, establishing a prediction SNR model based on LSTM by using real-time weather, longitude, latitude and position information; and 4, selecting an optimal coding scheme according to a table look-up method. The invention has the advantages that: predicting channel quality SNR (signal to noise ratio) containing weather information by using LSTM (least squares), considering SNR at the first several moments, and fully utilizing satellite operation rules; as LSTM can learn information such as longitude and latitude, and further learn rain attenuation values of different areas, the global universal model is obtained.

Description

An adaptive coding method for low-orbit satellite communication based on LSTM

Technical Field

The present invention relates to the field of computer network technology, and in particular to an adaptive coding (Adaptive Coding Modulation, ACM) method for low-orbit satellite communication based on a long short-term memory network (Lon gShort Term Memory, LSTM).

Background technique

Low-orbit satellites are tens to hundreds of kilometers away from the earth's surface. Due to their proximity to the surface, their channel capacity is higher than that of medium- and high-orbit satellites. Low-orbit satellites can provide coverage without blind spots and provide high-quality communications for air and ground facilities. Their strategic position is very important. Low-orbit satellite communications face the dual challenges of availability in extreme weather and rapid throughput increases: low-orbit satellites need to maintain high availability in scenarios such as disaster relief, polar regions, rain and snow, and with the increase in network bandwidth and video content, the total throughput of the network will be further improved. When used in areas such as unmanned driving and watching videos, lower latency and higher bandwidth are required.

Adaptive Coding Modulation (ACM) is an important method to improve satellite frequency band utilization and thus throughput. In traditional satellite communications, a 6dB margin is usually provided for rain attenuation. However, due to different cloud thicknesses, rain attenuation values vary across the globe. Moreover, there is only a rain attenuation value on rainy days, and the rain attenuation on sunny days is 0. Therefore, in practice, ACM is usually performed based on the weather to improve throughput. ACM in low-orbit satellite communications has technical difficulties, such as: low-orbit satellites move fast and switch frequently, requiring fast real-time predictions; rain attenuation values vary across the globe, requiring the establishment of a global model; global weather changes dynamically, making it difficult to quickly capture and synchronize.

In the problem of adaptive coding for low-orbit satellite communications, researchers have used methods such as identifying antenna noise temperature, linear regression, exponential regression, and support vector regression (SVR) to predict the signal-to-noise ratio (SNR), but none of them use the information that satellites have strong motion patterns. Long short-term memory (LSTM) is a recurrent neural network with memory. It controls the degree of memory of historical information and the degree of adoption of new information by forgetting gates, input gates, and output gates, and can recognize past satellite information through memory units.

Summary of the invention

In view of the defects of the prior art, the present invention provides a low-orbit satellite communication adaptive coding method based on LSTM.

In order to achieve the above invention object, the technical solution adopted by the present invention is as follows:

An adaptive coding method for low-orbit satellite communication based on LSTM, comprising the following:

Step 1: The ground equipment obtains the local real-time weather and longitude and latitude;

Step 2: Generate satellite position information based on satellite orbit information stored in ground equipment, expressed in three-dimensional coordinates;

Step 3: Establish a LSTM-based prediction SNR model using real-time weather, longitude and latitude, and location information;

Step 4: Select the optimal coding scheme based on the table lookup method.

Furthermore, the specific steps of step 1 are as follows:

Step 1.1 The ground device obtains its own latitude and longitude;

Step 1.2 The ground device requests the website to obtain real-time weather according to the latitude and longitude;

It should be noted that since the single communication time between low-orbit satellites and ground equipment is very short, usually a few minutes, the weather at the start of communication is considered to be the weather for this communication, so it is only necessary to request the weather conditions once at the start of communication;

Furthermore, the specific steps of step 2 are as follows:

Step 2.1 The ground equipment needs to store the orbital information of each satellite and have the ability to find the satellites that can communicate within a specified time. The orbital information is the information that describes and predicts the satellite position through calculation. It is represented by two lines of orbital elements and uses two lines of 80 characters of ASCII code to store data. The storage space is very small.

Step 2.2 converts the satellite position from orbital information into three-dimensional coordinates;

Furthermore, the specific steps of step 3 are as follows:

Step 3.1 Use real-time weather, latitude and longitude, and location information to form a sequence

(UE _lat ,UE _lon ,Sat _lat ,Sat _Lon ,Sat _alt ,Weather);

Where UE _lat and UE _lon are the latitude and longitude of the ground equipment, Sat _lat , Sat _Lon and Sat _alt are the latitude and longitude and altitude of the satellite, and Weather is the local real-time weather.

Step 3.2: Input the sequence in step 3.1 into the LSTM network for training, and use MAE as the error function; MAE is the absolute value of the difference between the actual SNR and the predicted SNR; the error function is used for back propagation to update the parameters in the network;

Step 3.3 Use the trained LSTM for prediction and verification;

The verification step is to prove that the network is not overfitting. The entire data set is divided into three parts: training set, validation set, and test set. The situation on the validation set is observed during training to determine the overfitting of the model.

Furthermore, the specific steps of step 4 are as follows:

Select a coding scheme based on the predicted SNR by looking up the table, and choose the coding scheme with the lowest redundancy rate with a packet error rate (PER) less than 0.1;

Among them, the packet error rate is the probability of a bit error in a frame of data during communication. It is generally believed that normal communication is possible when it is less than 0.1;

The redundancy rate is the ratio of the number of bits used for error correction in the forward error correction coding to the total number of bits, that is, redundancy rate=(number of codes used for error correction/total number of bits)×100%.

Compared with the prior art, the advantages of the present invention are:

1. Use LSTM to predict the channel quality SNR including weather information, taking into account the SNR of the previous moments and making full use of the satellite operation rules;

2. Since LSTM can learn information such as longitude and latitude, it can then learn the rain attenuation values of different regions and obtain a global universal model;

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a diagram showing the structure of an LSTM network according to an embodiment of the present invention;

FIG2 is a diagram of the structure of an LSTM neuron in an embodiment of the present invention;

FIG3 is a diagram of packet error rates for different coding scheme redundancy rates according to an embodiment of the present invention;

FIG4 is a cumulative distribution diagram (Cumulative Distribution Function, CDF) of MAE in different regions around the world for different methods in an embodiment of the present invention;

FIG5 is a CDF diagram of the optimal coding scheme matching ratio (Optimal Match Ration, OMR) of different methods in different regions of the world according to an embodiment of the present invention.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and examples.

As shown in FIG1 , this embodiment discusses the application of an LSTM-based low-orbit satellite communication adaptive coding method in satellite-to-ground communication.

This embodiment is based on an LSTM network, which is responsible for identifying the two-dimensional position of the ground user (User Equipment, UE) and the three-dimensional position of the satellite and the weather conditions where the UE is located, and predicting the SNR at the next moment based on the communication quality at the past moment. The embodiment of the low-orbit satellite communication adaptive coding method based on LSTM described in the present invention is as follows:

Step 1: The ground equipment obtains the local real-time weather and its latitude and longitude:

Step 1.1 The ground device obtains its own longitude and latitude. For example, the real-time longitude and latitude of the UE is (UE _lat , UE _lon ). The longitude and latitude information can be obtained by GPS satellite positioning or base station positioning. If the above information is missing, the last ground base station location interacted by the UE can be used as the default location;

Step 1.2 The ground device requests the website to obtain real-time weather according to the longitude and latitude. After obtaining the longitude and latitude information, the UE can request the weather API to obtain the real-time weather of the location;

API is the abbreviation of Application Programming Interface. Users send requests to API, and API will give corresponding responses.

It should be noted that the local weather needs to be queried only at the beginning of each connection, because the connection time of low-orbit satellites is very short, generally a few minutes. Taking Starlink as an example, its communication diameter is 1,100 kilometers and the longest connection time is 170 seconds. Therefore, querying the weather only at the beginning of the connection will not cause too much burden on the server, nor will it take up too many additional communication resources.

Step 2: Generate satellite position information based on satellite orbit information stored in ground equipment, expressed in three-dimensional coordinates:

Step 2.1 The ground equipment needs to store the orbital information of each satellite and have the ability to find the satellite that can communicate within the specified time:

Satellite orbit information is mostly stored in two lines of orbit information. This storage method is universally used and saves a lot of storage space. Even storing the orbit information of all satellites currently in orbit only requires 800KB of space.

Table 1 Satellite CALSPHERE 1 orbit information

Table 1 shows the orbital parameter information of the satellite CALSPHERE 1 launched on January 4, 2022:

The first row and the first column indicate its NASA unique number, and U indicates a non-confidential satellite;

The first row and second column indicate the international number of the aircraft;

The third column of the first row is the launch time. For example, 22 means 2022, and 004.58268302 means it was launched on the 4.58268302nd day of 2022.

The second column of the second row is the orbital inclination;

The third column of the second row is the longitude of the ascending node;

The second row and fourth column are the orbital eccentricity;

The second row and the fifth column are the perigee angular distance;

The second row and sixth column is the mean anomaly angle;

The second row and the seventh column are the horizontal motion rate;

The above orbital information can be used to calculate the satellites within the communication distance and their specific positions at the current time.

Step 2.2: Convert the satellite position from orbital information to three-dimensional coordinates. Lagrange polynomial interpolation, Newton polynomial interpolation, Neville successive linear interpolation and other methods can be used to convert orbital information into three-dimensional coordinate information. In the simulation scenario, satellite simulation software STK (System Toolkit, STK) can be used to obtain the position of the satellite at different times.

Step 3: Establish a model for predicting SNR based on LSTM using the information in steps 1 and 2;

Step 3.1 Use the information in steps 1 and 2 to form a sequence

Combine the information obtained in step 1 and step 2, and use One-Hot coding to represent the weather information. In practice, (Sunny, Cloudy, Rainy) is used to represent the weather. When it is sunny, it is coded as (1, 0, 0), and when it is cloudy, it is coded as (0, 1, 0). Rainy and snowy weather are all considered precipitation weather, and are uniformly regarded as rainfall weather, coded as (0, 0, 1);

So far, the eight-tuple input to the LSTM network has been constructed

x _t =(UE _lat ,UE _lon ,Sat _lat ,Sat _Lon ,Sat _alt ,Sunny,Cloudy,Rainy), where UE _lat ,UE _lon are the longitude and latitude of the ground equipment, Sat _lat ,Sat _Lon ,Sat _alt are the longitude and latitude and altitude of the satellite, and (Sunny,Cloudy,Rainy) represents the local weather;

During the simulation, there is a corresponding piece of information at each moment, that is, every 1 second.

Step 3.2 Input the sequence in the previous step into the LSTM network for training, using MAE as the error function;

The structure of the LSTM network is shown in Figure 1. The embedding layer is introduced to process weather and location information respectively. There are 3 neurons responsible for processing One-Hot encoded weather information, and 5 neurons responsible for processing location information. The embedding layer is followed by an LSTM layer, which contains 100 neurons and is responsible for memorizing past information; the LSTM layer is followed by a fully connected layer containing 100 neurons, which is responsible for integrating the information of the LSTM layer; and finally, there is an output layer containing one neuron, which outputs the predicted SNR.

The LSTM network solves the problem of long-term dependency by relying on three gate functions. It can remember long-term information by relying on gate functions. These three gate functions are forget gate, input gate and output gate, as shown in Figure 2.

(1) Forget gate: When the network receives new input, if the network needs to forget the old information, this is done through the forget gate. The forget gate is an important component of the LSTM network. It can control which information needs to be retained and which information needs to be forgotten, and avoid the problems of gradient explosion and gradient vanishing caused by the gradient back propagation over time. The forget gate determines what information the LSTM network forgets from the network state C _t-1 at the previous moment. The forget gate reads the output value h _t-1 at the previous moment and the input value x _t at this moment, and then maps it to a value between 0 and 1 through the sigmoid activation function. Finally, this value is multiplied by the network state C _t-1 to determine what information C _t-1 should discard. When the forget gate value is 1, it means that the information of the network state C _t-1 is completely retained. When the value is 0, it means that the information of the network state C _t-1 at the previous moment is completely discarded. The expression of the forget gate is:

_ft =σ( _Wf [ht _-1 , _xt ]+ _bf ),

(2) Input gate: determines which part of the new input information is retained in the network state. The input gate is used to control how much of the network's current input data _xt flows into the memory unit, that is, how much input information _xt can be saved in the current network state _Ct . The input gate consists of two parts. The first part: a control signal i _t between 0 and 1 generated by a sigmoid activation function, which is used to control the network state at this time. The second part is the candidate network state at the current moment generated by the tanh layer/> This value will be added to the network state to a certain extent by _it . The state expression of the input gate is:

_it =σ( _Wi [ht _-1 , _xt ]+ _bi ),

After obtaining the status of the forget gate and the input gate, the status of the network can be updated:

Where _Ct is the final network status at this moment.

(3) Output gate: The output value is based on the current network state, but there will be a filtering and screening process. The output gate also includes two operations: the first part is the control signal o _t between 0 and 1 generated by the sigmoid activation function; the second part is to multiply the output information tanh(C _t ) generated by the tanh function with the control signal o _t to obtain the final output value h _t at this moment. The output gate controls the influence of the memory unit C _t on the current output value h _t , that is, which parts of the memory unit will be output at this moment.

In the process of error back propagation, MAE (Mean Absolute Error) is used as the error metric because compared to MSE (Mean Squared Error), MAE tends to continue to reduce the error when the error is very small, while MSE will have a vanishing gradient due to its square property.

During training, you can obtain real-time weather information from all over the world from the Internet.

Step 3.3 Use the trained LSTM to perform prediction and verification;

Since neural networks tend to overfit, the overall data set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The data of the three data sets do not overlap and are shuffled. During training, the validation set is predicted every 10 epochs to observe whether the network is overfitting.

After the training is completed, the test set is sampled to obtain the final training results.

During training, the learning rate is set to 0.001, the batch size is set to 128, and the related time is set to 5.

After training, the MAE of different methods is compared as shown in Table 2:

Table 2 MAE of different methods

method Linear Smoothing Exponential Smoothing SVR LSTM MAE 0.114 0.062 0.032 0.017

It can be seen that the error between LSTM and the true value is the smallest.

The horizontal axis of Figure 3 is MAE, and the vertical axis is the Cumulative Distribution Function (CDF) of different locations around the world. It can be seen from the figure that LSTM performs better than other methods in different locations around the world.

Step 4: Select the optimal coding scheme based on the table lookup method.

According to the Consultative Committee for Space Data Systems (CCSDS), the communication code for low-Earth orbit satellites is the AR4JA code (accumulate-repeat-4-jagged-accumulate, AR4JA) in the pseudo-random low-density parity check code (QC-LDPC). CCSDS uses the AR4JA code and specifies three different data rates, 50%, 66%, and 80%. As shown in Figure 4, when the signal-to-noise ratio is low, the communication environment is poor, and a coding scheme with a lower data rate and a higher redundancy rate should be selected to ensure the communication quality; when the communication quality is good, a coding scheme that can transmit more data should be selected to improve the overall throughput.

In practice, the coding scheme with the lowest redundancy rate and a packet error rate PER less than 0.1 is selected as the optimal coding scheme, which is also often used in actual low-orbit satellite communication scenarios. The optimal match ratio (OMR) is used to represent the overall performance of the adaptive coding system. OMR is the ratio of the number of optimal coding schemes that can be selected under different prediction methods to the total number of tests.

Table 3 OMR of different methods

method Linear Smoothing Exponential Smoothing SVR LSTM OMR 97.9% 98.8% 99.5% 99.8%

As can be seen in Figure 5, compared with other methods, LSTM has the highest OMR, and the performance variance in different locations around the world is also smaller than other methods.

Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the implementation methods of the present invention, and should be understood that the protection scope of the present invention is not limited to such special statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims (1)

1. An LSTM-based low-orbit satellite communication adaptive coding method is characterized by comprising the following steps:

Step 1, ground equipment acquires local real-time weather and longitude and latitude;

The specific steps of the step 1 are as follows:

step 1.1, ground equipment acquires longitude and latitude of the ground equipment;

step 1.2, the ground equipment requests a website to acquire real-time weather according to longitude and latitude;

it should be noted that, because the single communication time between the low-orbit satellite and the ground equipment is very short, the weather at the communication starting time is considered as the weather of the communication, so that only one weather condition is required at the communication starting time;

step 2, generating satellite position information according to satellite orbit information stored in ground equipment, and representing the satellite position information by three-dimensional coordinates;

the specific steps of the step 2 are as follows:

Step 2.1, the ground equipment needs to store the orbit information of each satellite and has the capability of finding out the satellites which can communicate in a specified time; the orbit information is information describing and predicting satellite positions through calculation, is expressed by two rows of orbit element forms, adopts two rows of ASCII codes of 80 characters to store data, and occupies little space;

step 2.2, converting the position of the satellite from orbit information into three-dimensional coordinates;

step 3, establishing a prediction SNR model based on LSTM by using real-time weather, longitude, latitude and position information;

the specific steps of the step 3 are as follows:

Step 3.1 forming a sequence using real-time weather, latitude and longitude and position information

(UE_lat,UE_lon,Sat_lat,Sat_Lon,Sat_alt，Weather)；

Wherein, the UE _lat,UE_lon is the longitude and latitude of the ground equipment, the Sat _lat,Sat_Lon,Sat_alt is the longitude and latitude and the altitude of the satellite, and the Weather is the local real-time Weather;

Step 3.2, inputting the sequence in the step 3.1 into an LSTM network for training, and using MAE as an error function; MAE is the absolute value of the difference between the true SNR and the predicted SNR; the error function is used for back propagation and updating parameters in the network;

step 3.3, predicting and verifying by using the trained LSTM;

The verification step is to prove that the network is not over fitted, divide all data sets into three parts of a training set, a verification set and a test set, and observe the conditions on the verification set during training to determine the over fitting condition of the model;

Step 4, selecting an optimal coding scheme according to a table look-up method;

the specific steps of the step 4 are as follows:

Selecting a coding scheme according to the predicted SNR table lookup, and selecting a coding scheme with the lowest redundancy rate, wherein the packet error rate of the coding scheme is smaller than 0.1;

The packet error rate is the probability of error code in one frame of data in communication, and is considered to be smaller than 0.1, so that normal communication can be realized;

Here, the redundancy rate is a ratio of the number of bits used for error correction in forward error correction coding to the total number of bits, that is, redundancy rate= (number of codes used for error correction/total number of bits) ×100%.