Adaptive modulation method for laser microwave hybrid communication based on machine learning

Abstract

The application discloses a laser microwave hybrid communication adaptive modulation method based on machine learning, comprising: acquiring a weather data set, and establishing a channel state set according to weather data in the weather data set; training a random forest model RFM by using the weather data set and the channel state set, and obtaining the trained random forest model RFM; acquiring real-time weather data, inputting the weather data into the trained random forest model RFM, and obtaining a current optimal channel state D i ; judging whether an average error rate P i under the current channel state D e is less than a target error rate P e,obj , if not, adjusting the random forest model RFM until the current average error rate P e is less than the target error rate P e,obj , and simultaneously judging whether the currently executed channel state is consistent with the optimal channel state, if not, executing channel state switching, and completing switching of the laser microwave hybrid link. The influence of bad weather on the channel is reduced, and the communication efficiency is improved.

Description

Technical Field

[0001] This invention belongs to the technical field of laser-microwave hybrid networking communication methods, and relates to an adaptive modulation method for laser-microwave hybrid communication based on machine learning. Background Technology

[0002] FSO / RF hybrid links, by leveraging the complementary advantages of FSO and RF links, can significantly improve the reliability of communication systems. However, severe weather conditions still impact the reliability of FSO / RF hybrid links. Currently, the main handover methods for FSO / RF hybrid links are soft handover and hard handover, both of which require the transmitter to adjust the link based on channel state information fed back from the receiver, thus reducing communication efficiency. Summary of the Invention

[0003] The purpose of this invention is to provide an adaptive modulation method for laser-microwave hybrid communication based on machine learning, which solves the problem of low communication efficiency under adverse weather conditions in the existing technology.

[0004] The technical solution adopted in this invention is an adaptive modulation method for laser-microwave hybrid communication based on machine learning, comprising the following steps:

[0005] Step 1: Obtain the weather dataset S1, and establish the channel state set S2 based on the weather data in the weather dataset S1;

[0006] Step 2: Train the random forest model RFM using the weather dataset S1 and the channel state set S2 to obtain the trained random forest model RFM.

[0007] Step 3: Obtain real-time weather data and input the weather data into the trained Random Forest (RFM) model to obtain the current optimal channel state D. i ;

[0008] Step 4: Determine the current channel state D i Average bit error rate Is it less than the target bit error rate P? e,obj If not, adjust the Random Forest (RFM) model until the current average bit error rate is reached. Less than the target bit error rate P e,obj Simultaneously, it determines whether the current channel state is consistent with the optimal channel state. If not, it performs a channel state switch to complete the switching of the laser-microwave hybrid link.

[0009] The invention is further characterized by:

[0010] The weather dataset S1 includes multiple weather data sets, each containing data on six features: temperature, humidity, wind speed, cloud cover, rainfall, and visibility.

[0011] Step 1 is as follows:

[0012] The link channel state is divided according to the modulation method, and the weather condition data is divided into thresholds according to the link channel state to obtain the link margin threshold area of ​​each weather condition data threshold area; the weather dataset S1 is obtained, and the link margin corresponding to the weather dataset S1 is compared with the link margin threshold area to obtain the channel state set S2.

[0013] Step 1 is as follows:

[0014] The link channel state is divided according to the modulation method, and the target bit error rate P is set according to the channel conditions. e,obj Calculate the channel state D for each type of data under different weather conditions. i，i＝1……N average bit error rate In meeting the average bit error rate Less than the target bit error rate P e,obj Under these conditions, the weather condition data is divided into threshold regions to obtain multiple weather condition data threshold areas. Then, the link margin threshold area (LM) for each weather condition data threshold area is calculated. thi，i＝1……N * ; Obtain the weather data from the weather dataset S1, and obtain the link margin LM corresponding to each weather data point. Then, compare each link margin LM with the link margin threshold LM. thi * Compare them; if LM is less than the smallest LM thi * If LM is greater than LM, then communication is interrupted; thi * And less than LM thi+1 * Then the channel state under this weather data is D. i ;

[0015] The channel state under this weather data is determined to be D. i+1 Bit error rate P e,i Is it less than P? e,obj If it is greater than the threshold value, then the link margin threshold region LM is applied. thi Optimize until P e,i Less than P e,obj Then, the channel state is output, and the channel state of each weather data in the weather dataset S1 is obtained, i.e., the channel state set S2.

[0016] In step 1, the FSO link channel states include 16PPM, 8PPM, BPSK, QPSK, 8PSK, 32QAM, 64QAM, 128QAM, and 256QAM; the RF link channel states include BPSK, QPSK, 16QAM, 32QAM, 64QAM, 128QAM, and 256QAM.

[0017] Average bit error rate in step 1 The formula is:

[0018]

[0019] In the above formula, the instantaneous bit error rate formulas for different modulation methods are as follows:

[0020]

[0021]

[0022]

[0023] In the above formula, L and M represent the modulation order of different modulation methods, and γ is the instantaneous signal-to-noise ratio;

[0024]

[0025] In the above formula, This represents the Meijer'G function.

[0026] The probability density function of the FSO link is:

[0027]

[0028] In the above formula, γ FSO The instantaneous signal-to-noise ratio (SNR) and average signal-to-noise ratio (SNR) of the FSO link. P1 is the transmit power of the FSO link, and η is the photodetector responsivity. Let g1 be the noise variance and g1 be the path loss of the FSO link, where g1 = exp(-α). FSO,atmo L), α FSO,atmo The attenuation coefficient of the FSO link related to weather;

[0029] The probability density function of the RF link is:

[0030]

[0031] In the above formula, the average signal-to-noise ratio g2 represents the path loss of the RF link. α RF,atmo This refers to the weather-related attenuation coefficient of the RF link.

[0032] The formula for calculating link margin (LM) is as follows:

[0033] LM FSO =P+|S r |-α FSO,atmo -α geo -α sys (8);

[0034] In the above formula, P is the transmission power, and S... r For receiving sensitivity, α geo α is the geometric attenuation coefficient. sys The system attenuation coefficient;

[0035]

[0036] In the above formula, EIRP is the equivalent isotropic radiated power, and G... r To increase the gain of the receiver antenna, For the lowest normalized signal-to-noise ratio, R is the system bit rate, k is the Boltzmann constant, T is the receiver system noise temperature, and L... s This refers to path loss.

[0037] Weather condition data includes rainfall or visibility; under rainy conditions, α FSO,atmo =1.076R 0.67 α RF,atmo =kR α k is a function related to the frequency f and the polarization direction τ, and R is the rainfall rate; under foggy conditions, α RF,atmo =K l M and V represent visibility, V (km) represents the visibility range, λ represents wavelength, λ0 = 550nm is the visibility range reference, q is the scattering size distribution coefficient, and K... l A coefficient related to ambient temperature and frequency, where M is the density of the liquid water in the fog. 0.65 =0.024 / V.

[0038] The beneficial effects of this invention are as follows: The adaptive modulation method for laser-microwave hybrid communication based on machine learning utilizes the impact of severe weather on the link margin and bit error rate of the hybrid link to obtain the modulation scheme required to achieve the target bit error rate under different weather conditions. A corresponding model is obtained by training existing weather data using machine learning methods. Based on this model, the channel state under the current weather conditions is predicted in real time, and adaptive switching of the channel state is performed to achieve microwave-laser soft switching. This reduces the impact of severe weather on the channel and improves communication efficiency. Furthermore, it does not require increased transmission power, does not compromise system performance, improves the bit error rate of the communication system, and enhances spectrum utilization. Attached Figure Description

[0039] Figure 1 This is a flowchart of the adaptive modulation method for laser-microwave hybrid communication based on machine learning, as described in this invention.

[0040] Figure 2 This is a flowchart of the adaptive modulation method for laser-microwave hybrid communication based on machine learning in this invention for obtaining the channel state set;

[0041] Figure 3 This is a graph showing the relationship between rainfall and bit error rate under different channel conditions in an embodiment of the adaptive modulation method for laser-microwave hybrid communication based on machine learning of the present invention.

[0042] Figure 4 This is a schematic diagram of rainfall threshold division in an embodiment of the adaptive modulation method for laser-microwave hybrid communication based on machine learning of the present invention.

[0043] Figure 5 This is a schematic diagram of the link margin threshold region in an embodiment of the adaptive modulation method for laser-microwave hybrid communication based on machine learning of the present invention.

[0044] Figure 6 This is a flowchart of the training process of the random forest model in the adaptive modulation method for laser-microwave hybrid communication based on machine learning in this invention. Detailed Implementation

[0045] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments.

[0046] Adaptive modulation methods for laser-microwave hybrid communication based on machine learning, such as Figure 1 As shown, it includes the following steps:

[0047] Step 1: Obtain the weather dataset S1. The weather dataset S1 includes multiple weather data, each of which includes data on six features: temperature, humidity, wind speed, cloud cover, rainfall, and visibility. Establish the channel state set S2 based on the weather data in the weather dataset S1.

[0048] Specifically, the link channel states are divided according to the modulation scheme. FSO link channel states include 16PPM, 8PPM, BPSK, QPSK, 8PSK, 32QAM, 64QAM, 128QAM, and 256QAM; RF link channel states include BPSK, QPSK, 16QAM, 32QAM, 64QAM, 128QAM, and 256QAM. Based on the link channel states, weather data is thresholded to obtain the link margin threshold region for each weather data threshold region. Weather dataset S1 is obtained, and the link margin corresponding to weather dataset S1 is compared with the link margin threshold region to obtain the channel state set S2.

[0049] Furthermore, such as Figure 2 As shown, the link channel state is divided according to the modulation method, and the target bit error rate P is set according to the channel conditions. e,obj According to formula (1), calculate the channel state D for each weather condition (rainfall or visibility) data. i，i＝1……N average bit error rate In meeting the average bit error rate Less than the target bit error rate P e,obj Under certain conditions, weather condition data is divided into threshold regions to obtain multiple weather condition data threshold areas; for example, under channel state D i At that time, calculate the average bit error rate under different rainfall amounts. Meets the average bit error rate Less than the target bit error rate P e,obj The rainfall in a region is considered as different channel states D. i This allows us to obtain multiple regions, i.e., multiple weather data threshold regions; then, we calculate the link margin threshold region (LM) for each weather condition data threshold region. thi，i＝1……N * ; Obtain weather data from weather dataset S1 respectively, and obtain the corresponding link margin LM. Then, compare each link margin LM with the link margin threshold LM. thi * Compare them; if LM is less than the smallest LM thi * If LM is greater than LM, then communication is interrupted; thi * And less than LM thi+1 * Then the channel state under this weather data is D. i For example, LM is greater than LM. th2 And less than LM th3 * The channel state under this weather data is the link margin threshold region (LM). th3 * The corresponding channel state D2. In this embodiment, when the weather condition data is rainfall, different channel states D... i Regarding the error rate of rainfall, as follows: Figure 3 As shown, when the average bit error rate is satisfied... Less than the target bit error rate P e,obj Under the condition that the threshold area of ​​rainfall for each channel state of the FSO link is as follows: Figure 4 As shown, the link margin threshold region obtained based on the rainfall threshold region is as follows: Figure 5 As shown.

[0050] The channel state under this weather data is determined to be D. i+1 Average bit error rate Is it less than P? e,obj If it is greater than the threshold value, then the link margin threshold region LM is applied. thi Optimize until P e,i Less than P e,obj Then, the channel state is output, and the channel state of each weather data in the weather dataset S1 is obtained, i.e., the channel state set S2.

[0051] Average bit error rate The formula is:

[0052]

[0053] In the above formula, the instantaneous bit error rate formulas for different modulation methods are as follows:

[0054]

[0055]

[0056]

[0057] In the above formula, L and M represent the modulation order of different modulation methods, and γ is the instantaneous signal-to-noise ratio;

[0058]

[0059] In the above formula, This represents the Meijer'G function.

[0060] The probability density function of the FSO link is:

[0061]

[0062] In the above formula, γ FSO The instantaneous signal-to-noise ratio (SNR) and average signal-to-noise ratio (SNR) of the FSO link. P1 is the transmit power of the FSO link, and η is the photodetector responsivity. Let g be the noise variance, g1 be the path loss of the FSO link, and g1 = exp(-α) / 2. FSO,atmo L), α FSO,atmo The attenuation coefficient of the FSO link related to weather;

[0063] The probability density function of the RF link is:

[0064]

[0065] In the above formula, the average signal-to-noise ratio g2 represents the path loss of the RF link. α RF,atmo This refers to the weather-related attenuation coefficient of the RF link.

[0066] The formula for calculating link margin (LM) is as follows:

[0067] LM FSO =P+|S r |-α FSO,atmo -α geo -α sys (8);

[0068] In the above formula, P is the transmission power, and S... r For receiving sensitivity, α geo α is the geometric attenuation coefficient. sys The system attenuation coefficient;

[0069]

[0070] In the above formula, EIRP is the equivalent isotropic radiated power, and G... r To increase the gain of the receiver antenna, For the lowest normalized signal-to-noise ratio, R is the system bit rate, k is the Boltzmann constant, T is the receiver system noise temperature, and L... s For path loss;

[0071] Under rainy conditions, α FSO,atmo =1.076R 0.67 α RF,atmo =kR α k is a function related to the frequency f and the polarization direction τ, and R is the rainfall rate; under foggy conditions, α RF,atmo =K l M and V represent visibility, V (km) represents the visibility range, λ represents wavelength, λ0 = 550nm is the visibility range reference, q is the scattering size distribution coefficient, and K... l A coefficient related to ambient temperature and frequency, where M is the density of the liquid water in the fog. 0.65 =0.024 / V.

[0072] Step 2: Train the Random Forest Model (RFM) using the weather dataset S1 and channel state set S2 to obtain the trained RFM model; the specific training process is as follows:

[0073] Random forest is an algorithm that integrates multiple decision trees using the concept of ensemble learning; its basic unit is the decision tree. For example... Figure 6As shown, the training set S includes a weather dataset S1 and a channel state set S2. First, L samples are randomly selected from S with replacement. S contains seven elements: temperature, humidity, wind speed, cloud cover, rainfall, visibility, and channel state. The channel state is the output of the decision tree, while the others are feature attributes. The Gini index is used to evaluate feature importance; a smaller Gini index indicates higher feature purity. Next, the Gini index is used to select the feature with higher purity as the optimal splitting feature for the first layer of node splitting. Then, the optimal feature from the remaining features is selected to further split the split sample dataset, and so on, until all feature values ​​in the weather dataset S1 are used up, the sample data is completely separable, or the purity requirement is met. Finally, the corresponding channel state result in S2 is output, resulting in M ​​decision trees forming a Random Forest (RFM) model. Given a weather data sample, the M decision trees will produce M classification results. The RFM model integrates all classification voting results and designates the channel state with the most votes as the final output.

[0074] Step 3: Monitor real-time weather data online using meteorological equipment, input the weather data into the Random Forest Model (RFM), and obtain the current optimal channel state D. i ;

[0075] Step 4: Determine the current channel state D i Average bit error rate Is it less than the target bit error rate P? e,obj If not, adjust the Random Forest (RFM) model until the current average bit error rate is reached. Less than the target bit error rate P e,obj Simultaneously, it determines whether the current channel state is consistent with the optimal channel state. If not, it performs a channel state switch to complete the switching of the laser-microwave hybrid link.

[0076] Through the above methods, the adaptive modulation method for laser-microwave hybrid communication based on machine learning of the present invention utilizes the impact of severe weather on the link margin and bit error rate of the hybrid link to obtain the modulation mode required to achieve the target bit error rate under different weather conditions. A corresponding model is obtained by training existing weather data using machine learning methods. Based on this model, the channel state under the current weather conditions is predicted in real time, and adaptive switching of the channel state is performed to achieve microwave-laser soft switching. This reduces the impact of severe weather on the channel and improves communication efficiency. It does not require increasing transmission power, does not lose system performance, and can improve the bit error rate of the communication system and increase spectrum utilization.

Claims

1. An adaptive modulation method for laser-microwave hybrid communication based on machine learning, characterized in that, Includes the following steps: Step 1: Obtain the weather dataset S1, and establish the channel state set S2 based on the weather data in the weather dataset S1; Step 2: Use the weather dataset S1 and the channel state set S2 to train a random forest model and obtain the trained random forest model RFM. Step 3: Obtain real-time weather data and input the weather data into the trained Random Forest (RFM) model to obtain the current optimal channel state D. i ; Step 4: Determine D i Average bit error rate Is it less than the target bit error rate P? e,obj If not, adjust the trained Random Forest (RFM) model until the current average bit error rate is reached. Less than the target bit error rate P e,obj At the same time, it determines whether the current channel state is consistent with the optimal channel state. If not, it performs a channel state switch to complete the switching of the laser-microwave hybrid link. Step 1 is as follows: The link channel state is divided according to the modulation method, and the target bit error rate P is set according to the channel conditions. e,obj Calculate the channel state D for each type of data under different weather conditions. i Average bit error rate of i=1,……,N In order to meet the average bit error rate Less than the target bit error rate P e,obj Under these conditions, the weather condition data is divided into threshold regions, resulting in multiple weather condition data threshold areas. Then, the link margin threshold area for each weather condition data threshold area is calculated. ; Obtain weather data from weather dataset S1 respectively, and obtain the link margin LM corresponding to each weather data. Then, compare each link margin LM with a link margin threshold. Compare them; if LM is less than the smallest If LM is greater than 1, then communication is interrupted; if LM is greater than 1, then communication is interrupted. and less than Then the channel state under this weather data is D. i ; The channel state under this weather data is determined to be D. i Bit error rate P e,i Is it less than P? e,obj If it is greater than the threshold, then the link margin threshold is applied. Optimize until P e,i Less than P e,obj Then the channel state is output, and the channel state of each weather data in the weather dataset S1 is obtained, that is, the channel state set S2. The average bit error rate mentioned in step 1 The formula is: （1）； In the above formula, Let be the probability density function. Let be the instantaneous bit error rate function. The formulas for the instantaneous bit error rate for different modulation schemes are as follows: （2）； （3）； （4）； In the above formula, M represents the modulation order. Instantaneous signal-to-noise ratio; （5）； In the above formula, Represents the Meijer'G function; The probability density function of the FSO link is: （6）； In the above formula, The instantaneous signal-to-noise ratio of the FSO link. The average signal-to-noise ratio of the FSO link; The probability density function of the RF link is: （7）； In the above formula, It is the instantaneous signal-to-noise ratio of the RF link. The average signal-to-noise ratio of the RF link; FSO link margin LM FSO The calculation formula is as follows: （8）； In the above formula, P For transmission power, S r For receiving sensitivity, α geo The geometric attenuation coefficient, α sys The system attenuation coefficient is... The attenuation coefficient of the FSO link related to weather; RF link margin LM RF The calculation formula is as follows: （9）； In the above formula, EIRP This is the equivalent isotropic radiated power. G r To increase the gain of the receiver antenna, For the lowest normalized signal-to-noise ratio, R For the system bit rate, k Boltzmann's constant, T The receiver system noise temperature, L s For path loss, This represents the attenuation coefficient of the RF link related to weather conditions.

2. The adaptive modulation method for laser-microwave hybrid communication based on machine learning according to claim 1, characterized in that, The weather dataset S1 includes multiple weather data sets, each containing data on six features: temperature, humidity, wind speed, cloud cover, rainfall, and visibility.

3. The adaptive modulation method for laser-microwave hybrid communication based on machine learning according to claim 1, characterized in that, In step 1, the FSO link channel states include 16PPM, 8PPM, BPSK, QPSK, 8PSK, 32QAM, 64QAM, 128QAM, and 256QAM; the RF link channel states include BPSK, QPSK, 16QAM, 32QAM, 64QAM, 128QAM, and 256QAM.

Images