Unmanned aerial vehicle to-ground space optical communication atmospheric channel estimation method

Abstract

The invention discloses an unmanned aerial vehicle to-ground space optical communication atmospheric channel estimation method. The method comprises the following steps: constructing an unmanned aerial vehicle to-ground space optical communication system model; constructing a fifteen-order pseudo-random binary code as the input of a space optical communication system model, receiving the fifteen-order pseudo-random binary code after the fifteen-order pseudo-random binary code passes through the space optical communication system, and then dividing a received data set; performing feature extraction on the data set; verifying performance parameters of the channel estimation method based on the gated neural network, comparing predicted mean square errors and training time consumption of a gated neural network model under different parameters through a verification set, obtaining the best channel parameters, and training a channel estimation model; and carrying out model performance verification on the trained channel estimation model by using test data, and carrying out prediction mean square error verification. The unmanned aerial vehicle to-ground space optical communication atmospheric channel estimation method provided by the invention is a high-reliability and high-robustness unmanned aerial vehicle to-ground space optical communication system channel estimation method.

Description

Technical Field

[0001] The present invention relates to the field of space optical communication, and in particular to a method for estimating an atmospheric channel for space optical communication between a drone and the ground. Background Art

[0002] Space optical communication is a communication technology that uses lasers to transmit data in atmospheric channels. In some areas where optical fiber cannot be easily installed, traditional wireless communication technology is supplemented by radio frequency communication, and space optical communication, as an efficient and rapidly deployable technology, is considered to be a feasible and better choice in this scenario. UAVs themselves have the characteristics of being easy to carry, flexible to deploy, and having a wide coverage range, which can be combined with the advantages of space optical communication. There is great potential for development in scenarios that require high-bandwidth communication, such as aerial photography of high-definition images and videos, and three-dimensional reconstruction. Therefore, space optical communication based on drones has received a lot of attention.

[0003] Although UAV space optical communication has many advantages, its laser beam is easily affected by unevenly distributed humidity, temperature, wind speed and other factors when it is transmitted in the atmospheric channel, resulting in random fluctuations in the refractive index of the light on the laser beam propagation path, causing negative effects such as light spot flickering, beam expansion and light spot drift, which is called atmospheric turbulence. In fact, atmospheric turbulence can cause phase fluctuations and intensity attenuation of the space optical signal received by the ground receiver, seriously affecting the accuracy of communication and reducing the communication quality.

[0004] At present, neural networks based on deep learning are widely used in the field of optical communications. Their ability to extract hidden rules from historical data is also suitable for application in the field of channel estimation. However, in practical applications, the change of atmospheric turbulence intensity in the atmospheric channel will cause the channel estimation method trained under fixed turbulence to fail to accurately obtain channel state information. Summary of the invention

[0005] In view of the above-mentioned defects of the prior art, the technical problem to be solved by the present invention is that the channel estimation method trained under fixed turbulence obtains inaccurate channel state information due to the change of atmospheric turbulence intensity in the prior art. Therefore, the present invention provides an atmospheric channel estimation method for UAV-to-ground space optical communication, which uses a GRU-based channel estimation method and is trained by using channel data under different turbulence and signal-to-noise ratios to obtain accurate channel estimation results. It is a highly reliable and robust channel estimation method for UAV-to-ground space optical communication system.

[0006] To achieve the above object, the present invention provides a method for estimating atmospheric channels for UAV-to-ground space optical communication, comprising the following steps:

[0007] Construct a model of UAV-to-ground space optical communication system;

[0008] A fifteen-order pseudo-random binary code is constructed as the input of the space optical communication system model, the input electrical signal is modulated and encoded, and then received by the space optical communication system, and then the received data set is divided;

[0009] Extract features from the data set;

[0010] The performance parameters of the channel estimation method based on the gated neural network are verified. Then, the prediction mean square error and training time of the gated neural network model under different parameters are compared through the verification set to obtain the best channel parameters and train the channel estimation model.

[0011] The trained channel estimation model is validated using test data and the prediction mean square error is verified.

[0012] Furthermore, a UAV-to-ground space optical communication system model is constructed, including a UAV transmitter, a ground receiver and an atmospheric channel for UAV-to-ground space optical communication. The UAV transmitter and the ground receiver are communicatively connected via the atmospheric channel. The UAV transmitter is used to encode and modulate the electrical signal of the transmission data and then transmit it into the atmospheric channel as a laser signal. The UAV-to-ground space optical communication atmospheric channel is used to simulate the influence of the atmospheric channel on the signal generation. The ground receiver is used to demodulate and decode the laser signal into the electrical signal of the reception data.

[0013] Furthermore, the fifteenth-order pseudo-random binary code is constructed and passed through the UAV transmitter, then modulated to a 1550nm laser carrier through an MZM modulator and sent to the atmospheric channel of the UAV-to-ground space optical communication. After the ground receiver receives the laser signal from the atmospheric channel, it passes it through an APD photodiode and a filter, and compares the data signal power with the noise signal power to estimate the channel state.

[0014] Furthermore, the received data set is divided, wherein 60% of the data set is used as a training set, 20% as a validation set, and 20% as a test set.

[0015] Furthermore, feature extraction is performed on the data set, including extracting the maximum value, minimum value and mean square error of every ten communication data.

[0016] Furthermore, feature extraction is performed on the data set. The maximum value, minimum value and mean square error of each group of ten communication data are extracted and used as the input data set for training the gated neural network channel estimation model.

[0017] Furthermore, a gated neural network is used to construct a channel estimation model, and the gated neural network structure includes a reset gate, an update gate, and candidate hidden states.

[0018] Furthermore, reset gates, update gates, and candidate hidden states are used to process the relationship between previous and subsequent time series.

[0019] Furthermore, when training the channel estimation model, the Adam optimizer is used, the error calculation function is the mean square error function, the training learning rate is 0.001, and the number of iterations is 1000.

[0020] Technical Effects

[0021] The present invention provides an atmospheric channel estimation method for UAV-to-ground space optical communication. The channel estimation model based on GRU has small channel prediction error and saves more computing resources and model training time. In UAV space optical communication where rapid changes may occur, the method is more suitable for channel estimation tasks that require repeated training.

[0022] The concept, specific structure and technical effects of the present invention will be further described below in conjunction with the accompanying drawings to fully understand the purpose, characteristics and effects of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of a method for estimating atmospheric channels of UAV-to-ground space optical communication according to a preferred embodiment of the present invention;

[0024] Figure 2 It is a structural diagram of a channel estimation system based on a gated neural network of a method for estimating an atmospheric channel of a UAV-to-ground space optical communication according to a preferred embodiment of the present invention;

[0025] Figure 3 It is a comparison diagram of prediction errors of a model selected at different epochs in a method for estimating atmospheric channels of UAV-to-ground space optical communication according to a preferred embodiment of the present invention;

[0026] Figure 4 It is a comparison diagram of prediction errors of a method for estimating atmospheric channels of UAV-to-ground space optical communication and a channel estimation system based on LSTM in a preferred embodiment of the present invention;

[0027] Figure 5 This is a comparison chart of the training time consumption of a method for estimating an atmospheric channel of UAV-to-ground space optical communication and an LSTM channel estimation system in UAV space optical communication, which is a preferred embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0029] In the following description, specific details such as specific internal procedures and techniques are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0030] like Figure 1 As shown, the present invention provides a method for estimating an atmospheric channel of UAV-to-ground space optical communication. By taking channel data under different turbulence and signal-to-noise ratios in space optical communication as input for training, a channel estimation model based on deep learning is obtained to accurately estimate the channel state of the actual atmospheric channel of the UAV space optical communication. The method of the embodiment of the present invention includes the following steps:

[0031] Step 1. Construct a UAV-to-ground space optical communication system model; construct a UAV-to-ground space optical communication system model, including a UAV transmitter, a ground receiver and a UAV-to-ground space optical communication atmospheric channel, wherein the UAV transmitter and the ground receiver are connected for communication via the atmospheric channel. The UAV transmitter is used to encode and modulate the electrical signal of the transmission data and then transmit it into the atmospheric channel as a laser signal. The UAV-to-ground space optical communication atmospheric channel is used to simulate the impact of the atmospheric channel on the signal. The ground receiver is used to demodulate and decode the laser signal into an electrical signal for receiving data. In the UAV-to-ground space optical communication atmospheric channel, the signal x sent by the transmitter and the signal y received by the receiver can be expressed as:

[0032] y=ηIx+noise

[0033] Among them, η is the APD photoelectric conversion sensitivity, h is the link gain, and noise is the additive Gaussian white noise with an average power of zero. The atmospheric turbulence attenuation effect obeys the Gamma-Gamma distribution, and the Retov variance is used to represent the turbulence size.

[0034] The channel link gain h is mainly determined by atmospheric attenuation and atmospheric turbulence:

[0035] I=I l I a

[0036] Among them, I l and I a They represent the effects of atmospheric attenuation and atmospheric turbulence respectively.

[0037] The atmospheric turbulence probability model follows the Gamma-Gamma distribution. The model uses small-scale eddy currents I x Scattering effects and large-scale eddies Iy The refraction effect is used to explain the light intensity flicker, and the formula is as follows:

[0038] I a =I x I y

[0039] The probability density functions are as follows:

[0040]

[0041] Therefore, the probability density function of light intensity flicker caused by atmospheric turbulence is:

[0042]

[0043] Where K n (.) is the second kind of n-order modified Bessel function, Γ(.) is the gamma function, α and β are the effective numbers of large-scale vortices and small-scale vortices, respectively, and their formulas are:

[0044]

[0045] in, is the refractive index structure constant, k=2π / λ is the light wave number, and λ is the wavelength.

[0046] The atmospheric channel model that obeys the Gamma-Gamma atmospheric turbulence distribution model can reflect the impact of the drone-to-ground space light model on the signal in actual communication, and simulate the intensity fluctuations and phase fluctuations generated in signal communication, so as to facilitate the subsequent establishment of a data set under real atmospheric channel communication.

[0047] Step 2: Use simulation software Optisystem 15 and MATLAB R2014a to build the atmospheric channel model based on the Gamma-Gamma atmospheric turbulence distribution model described in step 1, and use Optisystem 15 software to construct a fifteen-order pseudo-random binary code as the electrical signal input of the space optical communication system model. The pseudo-random binary code is a binary code sequence with random statistical characteristics that can be predetermined and repeatedly replicated. It is generally used to simulate random signals. It modulates and encodes the input electrical signal and receives it after passing through the space optical communication system built by MATLAB, and then divides the received data set; constructs a fifteen-order pseudo-random binary code electrical signal, which is modulated by the UAV transmitter, modulated to a 1550nm laser carrier through an MZM modulator and sent to the atmospheric channel of the UAV-to-ground space optical communication. After the ground receiver receives the laser signal from the atmospheric channel, it passes it through an APD photodiode and a filter, and compares the data signal power and the noise signal power to estimate the channel state. The received data set is divided, where 60% of the data set is used as a training set, 20% as a validation set, and 20% as a test set.

[0048] Step 3: Perform simple preprocessing on the data set, that is, feature extraction, which can enhance the accurate fitting ability of the gated neural network for the atmospheric channel with less computational effort, so that subsequent model training can easily converge, including extracting the maximum value, minimum value and mean square error of every ten communication data. Specifically, feature extraction is performed on the data set, and the data set is grouped into ten communication data sets, and the maximum value, minimum value and mean square error are extracted as the input data set for gated neural network channel estimation model training.

[0049] Step 4: Verify the impact of different training parameters in the model on the model performance in the channel estimation method based on the gated neural network, and then compare the mean square error and training time of the prediction results obtained by the gated neural network model under different parameters through the validation set, obtain the best channel parameters and train the channel estimation model. The structure of the gated neural network (GRU) is as follows Figure 2 As shown in the figure, GRU introduces a reset gate, an update gate, and a candidate hidden state to handle the relationship between the previous and next time series. The structural principle formulas of each gate of GRU are as follows:

[0050] r t =σ(W xr x t +W hr h t-1 +b r )

[0051] z t =σ(W xz x t +Whz h t-1 +b z )

[0052]

[0053] Among them, r t , z t and They are the reset gate, update gate and candidate hidden state, and their value range is [0,1]. A value of 0 means that all data in the neural network is discarded, and conversely, a value of 1 means that all data is retained. The reset gate uses a sigmoid function to calculate the weight of the input of the current time step and the hidden state of the previous time step. This gate controls the influence of historical inputs and decides whether to ignore past information and reinitialize the hidden state; the update gate uses a sigmoid function to calculate the weight of the input of the current time step and the hidden state of the previous time step. This gate controls the update of the hidden state and determines how much new information is added to the hidden state; the candidate hidden state superimposes the hidden state of the previous time step and the new candidate hidden state to obtain the final hidden state. In addition, σ is the sigmoid function, and h t Represents the final output of the neural network, represented by z t Perform a status update. t is the input time series, ° represents the Hadamard product. Finally, W and b are the parameters updated during GRU model training.

[0054] Different training parameters in the GRU channel estimation method model have a certain impact on the channel estimation performance of the model. Here, taking the epoch of the model as an example, the impact of selecting different epochs on the prediction error of the model is as follows: Figure 3 As shown. The training epoch will affect the accuracy of the channel estimation results. If the epoch of GRU training is too small, it will not achieve the optimal fit of the channel, and if the epoch is too large, it will cause overfitting. Therefore, this method simulates and analyzes the relationship between the epoch and MSE of the GRU channel estimation algorithm when the SNR is 10dB, 15dB and 20dB. It can be seen that when the epoch is less than 13, the error of the GRU channel estimation shows a downward trend, and the overall MSE decreases with the increase of the signal-to-noise ratio. However, when the epoch is greater than 13, the MSE of the GRU channel estimation will increase to a certain extent, and the MSE with an SNR of 20dB will increase the most. After that, the three MSE curves will continue to decrease, but in the end they can only approach the MSE when the epoch is 13. Therefore, considering the accuracy of the GRU channel estimation algorithm and the consumption of training resources, this method sets the epoch to 13.

[0055] In the channel estimation model training, the validation set is used to verify the predicted mean square error of the channel estimation model every 5 trainings. When the error becomes larger or unchanged for 5 trainings, the model training is stopped to prevent the model from overfitting. The training uses the Adaptive Moment (Adam) optimizer, the error calculation function is the mean square error function, the training learning rate is 0.001, and the number of iterations is 1000. The atmospheric channel of the UAV-to-ground space optical communication includes atmospheric turbulence and additive Gaussian white noise, and the atmospheric turbulence obeys the Gamma-Gamma distribution. The channel Retov variance is set to 0.1, 0.5, 1.0, 1.6 and 3.5, and the channel signal-to-noise ratio and channel length can be adjusted. The channel length can be up to 20km, and the zenith angle is within 60 degrees.

[0056] Step 5: Use the test data to verify the model performance of the trained channel estimation model and verify the prediction mean square error.

[0057] By comparing the channel estimation performance of the GRU-based channel estimation system proposed by this method with that of the long short-term memory neural network (LSTM)-based channel estimation system under the same atmospheric channel conditions, the model performance of the channel estimation model was verified.

[0058] The present invention compares the GRU channel estimation system with the LSTM (Long Short Term Memory) based channel estimation system. Figure 4 , as shown in 5.

[0059] Figure 4 This is a comparison of the estimation errors of the GRU channel estimation system and the LSTM channel estimation system under weak turbulence, and the evaluation criterion is the mean square error (MSE) of the channel estimation. With the increase of the signal to noise ratio (SNR), the MSE curves of the embodiment of the present invention and the LSTM-based system both show a downward trend, and the downward trend is similar. At the same SNR, the performance gap between the embodiment of the present invention and the LSTM is about 1dB, which shows that in terms of channel estimation, although the performance of the GRU is slightly weaker than that of the LSTM, the gap is not obvious. When the structure of the embodiment of the present invention is less complex than that of the LSTM, it can be used as a more preferred alternative to the LSTM for channel estimation in FSO communication. When the SNR is 20dB, the channel estimation performance of the two is similar. Figure 5It is the training time of the GRU channel estimation system and the LSTM channel estimation system in the UAV space optical communication. The model size is represented by the number of hidden nodes in the model, Hidden Size. There is a large gap between the embodiment of the present invention and LSTM in terms of training time. As the Hidden Size increases, the upward trend of the model training time curve of the embodiment of the present invention is stable, but the training time curve of LSTM rises faster. The training time of the embodiment of the present invention is less than 1.7 hours when the Hidden Size is 200, while the training time of LSTM is more than 1.7 hours when the Hidden Size is 140. Therefore, based on Figure 3 and Figure 4 It can be seen that, under the premise of almost no loss of performance, the embodiment of the present invention saves more computing resources and model training time than LSTM, which indicates that in the space optical communication of unmanned aerial vehicles where rapid changes may occur, the method of the embodiment of the present invention is more suitable for channel estimation tasks that are repeatedly trained multiple times.

[0060] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for estimating atmospheric channels for UAV-to-ground space optical communication, characterized in that: The following steps are involved: Construct a model of UAV-to-ground space optical communication system; A fifteen-order pseudo-random binary code is constructed as the input of the space optical communication system model, the input electrical signal is modulated and encoded, and then received by the space optical communication system, and then the received data set is divided; Extract features from the data set; The performance parameters of the channel estimation method based on the gated neural network are verified. Then, the prediction mean square error and training time of the gated neural network model under different parameters are compared through the verification set to obtain the best channel parameters and train the channel estimation model. The trained channel estimation model is validated using test data and the prediction mean square error is verified.

2. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 1, characterized in that: A UAV-to-ground space optical communication system model is constructed, including a UAV transmitter, a ground receiver and an atmospheric channel for UAV-to-ground space optical communication. The UAV transmitter and the ground receiver are communicatively connected through the atmospheric channel. The UAV transmitter is used to encode and modulate an electrical signal for sending data and then transmit it to the atmospheric channel as a laser signal. The atmospheric channel for UAV-to-ground space optical communication is used to simulate the influence of the atmospheric channel on signal generation. The ground receiver is used to demodulate and decode the laser signal into an electrical signal for receiving data.

3. The atmospheric channel estimation method for UAV-to-ground space optical communication according to claim 2, characterized in that: The fifteenth-order pseudo-random binary code is constructed and passed through the UAV transmitter, then modulated to a 1550nm laser carrier through an MZM modulator and sent to the atmospheric channel of the UAV-to-ground space optical communication. After the ground receiver receives the laser signal from the atmospheric channel, it passes it through an APD photodiode and filter, and compares the data signal power with the noise signal power to estimate the channel state.

4. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 1, characterized in that: The received data set is divided into 60% as a training set, 20% as a validation set, and 20% as a test set.

5. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 1, characterized in that: Feature extraction is performed on the data set, including extracting the maximum value, minimum value and mean square error of every ten communication data.

6. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 5, characterized in that: Feature extraction is performed on the data set. The data set is grouped into ten communication data. The maximum value, minimum value and mean square error are extracted and used as the input data set for training the gated neural network channel estimation model.

7. The atmospheric channel estimation method for UAV-to-ground space optical communication according to claim 1, characterized in that: A gated neural network is used to construct a channel estimation model. The gated neural network structure includes a reset gate, an update gate, and candidate hidden states.

8. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 7, characterized in that: Reset gates, update gates, and candidate hidden states are used to handle the relationship between previous and subsequent time series.

9. The method for estimating atmospheric channels for UAV-to-ground space optical communication according to claim 1, characterized in that: When training the channel estimation model, the Adam optimizer is used, the error calculation function is the mean square error function, the training learning rate is 0.001, and the number of iterations is 1000.

Images