Sea surface channel prediction method based on pitching motion characteristics of equipment

Abstract

The invention discloses a sea surface channel prediction method based on equipment pitching motion characteristics, which comprises the following steps of: in a modeling stage, setting coordinates according to a general application scene, simulating heaving and pitching motion of a transceiver according to system parameters and sea surface environment parameters, establishing SW-GBSM according to a motion track to model a channel, and obtaining channel impulse response; calculating the antenna position offset of the transceiver at the same time according to the motion trail, and obtaining the sea wave characteristics of the transceiver; and in the prediction stage, frequency domain conversion is performed on the channel impulse response of the time domain, and the sea wave characteristics SWF-TA of the transmitter and the receiver and the frequency domain channel response are spliced and then input into the SWF-GRU channel prediction network. In the SWF-GRU, the cascaded GRU updates the hidden state of itself according to an input sequence, and then the full connection layer maps the final hidden state of the GRU into a channel prediction result. According to the method, the channel modeling effect of a real sea surface environment is met, and the sea surface channel prediction performance is greatly improved on the premise of increasing relatively low calculation complexity.

Description

Technical Field

[0001] The invention belongs to the technical field of radio communications, and in particular relates to a sea surface channel prediction method based on equipment pitch motion characteristics. Background Art

[0002] In recent years, the rapid development of the marine economy has put forward higher requirements for efficient and reliable marine communication technology. As the basis for the design of marine communication systems, it is crucial to establish a channel model that is close to the real marine environment. Traditional sea surface channel modeling methods mainly rely on a large amount of experimental measurement data to fit empirical models. Although they can reflect some environmental characteristics, they have limited applicability and the measurement process is usually accompanied by high experimental costs. In order to improve the modeling accuracy, some deterministic models use methods such as ray tracing and finite difference time domain method to accurately describe the propagation path of signals in the sea surface environment. However, such methods have high computational overhead and are difficult to meet the needs of real-time channel generation. Therefore, statistical models with lower computational complexity are more valuable in real-time channel modeling. Among them, the geometry-based statistical model (GBSM) has obvious advantages in balancing the accuracy of environmental description and computational efficiency. However, the existing sea surface channel model based on GBSM has not fully considered the impact of the pitch motion of the ship caused by the action of waves on the channel characteristics, which may lead to deviations in the description of Doppler frequency shift and channel dynamic changes. Therefore, further research on GBSM that can characterize the wave disturbance effect is of great significance to improving the reliability of marine communication systems.

[0003] In addition, in a sea surface communication system, accurately predicting the Channel State Information (CSI) enables the system to adapt to channel changes caused by the dynamic environment. However, the non-stationary effects caused by ocean waves pose significant challenges to ocean channel prediction. Existing channel prediction research mainly focuses on the following two aspects. On the one hand, traditional methods (such as autoregressive models) can be used for channel prediction, but it is difficult for them to effectively handle the non-stationary changes in the channel caused by factors such as ocean waves. On the other hand, deep learning-based methods have been widely studied in recent years. Among them, recurrent neural networks (RNNs) such as Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) have been widely used in channel prediction tasks and demonstrated superior prediction performance due to their ability to capture the temporal dependencies of time series data. At the same time, combining network enhancement techniques such as attention mechanisms and hyperparameter networks can further improve the prediction performance, but usually with a relatively high computational complexity. In addition, some studies have proposed fusing multi-modal features for channel prediction. However, existing methods generally do not optimize the design of the features used for the dynamic characteristics of ocean channels, limiting their applicability in complex ocean environments. Summary of the Invention

[0004] The main objective of the present invention is to overcome the drawbacks and deficiencies of the prior art and provide a sea surface channel prediction method based on the pitching motion characteristics of the device. By jointly calculating the sea surface height and the pitching angle of the transceiver, the antenna trajectory is determined, and the Doppler frequency shift change caused by ocean wave fluctuations is simulated accordingly, so as to achieve a channel modeling effect that better conforms to the real sea surface environment. At the same time, a channel prediction method based on SWF-GRU is proposed, which can significantly improve the performance of sea surface channel prediction on the premise of increasing a relatively low computational complexity.

[0005] To achieve the above objective, the present invention adopts the following technical solutions:

[0006] In the first aspect, the present invention provides a sea surface channel prediction method based on the pitching motion characteristics of the device, including:

[0007] Modeling stage:

[0008] Determine the positions of the transmitter and receiver on the sea surface, and obtain the system parameters and sea surface environment parameters;

[0009] Calculate the heaving and pitching motion trajectories of the transmitter and receiver on the sea surface based on the acquired parameters. Use SW-GBSM to model the channel according to the motion trajectories to obtain the channel impulse response. At the same time, calculate the antenna position offset of the transmitter and receiver at the same time according to the motion trajectories, and obtain the sea wave characteristics SWF-TA of the transmitter and receiver.

[0010] Prediction stage:

[0011] Convert the channel impulse response into the frequency-domain channel response; construct an SWF-GRU channel prediction network, where the SWF-GRU channel prediction network includes multiple cascaded GRUs and a fully connected layer. Concatenate the sea wave characteristics SWF-TA of the transmitter and receiver with the frequency-domain channel response to form a joint input feature vector. Input P historical joint input feature vectors into the SWF-GRU channel prediction network in sequence. The cascaded GRUs gradually update their hidden states according to the input sequence, and the fully connected layer maps the final hidden state of the GRUs to the predicted frequency-domain channel response; during the training process, construct a loss function using the normalized mean square error between the predicted frequency-domain channel response and SWF-TA and the true value, and calculate the gradient and update the SWF-GRU network parameters through the backpropagation algorithm.

[0012] As a preferred technical solution, the calculating the heaving and pitching motion trajectories of the transmitter and receiver on the sea surface based on the acquired parameters includes:

[0013] Define the height of a certain point on the sea surface and the sea wave inclination angle according to the P-M sea wave spectrum and the linear superposition theory;

[0014] Use the height of a certain point on the sea surface to describe the height of the heaving motion of the transmitter and receiver on the sea surface; obtain the response function from the sea wave inclination angle to the ship's pitching angle according to the system parameters. According to the sea wave inclination angle and the response function from the sea wave inclination angle to the ship's pitching angle, obtain the ship's pitching angle, and the ship's pitching angle is consistent with the pitching angle corresponding to the heaving motion of the transmitter and receiver; the system parameters include the displacement of the ship, the initial metacentric height, the pitching moment of inertia, and the damping coefficient.

[0015] Based on the stereometric theory, use the height of the heaving motion of the transmitter and receiver on the sea surface and the pitching angle corresponding to the pitching motion to calculate and obtain the antenna coordinates of the transmitter and receiver.

[0016] As a preferred technical solution, the using SW-GBSM to model the channel according to the motion trajectories is specifically:

[0017] According to the motion trajectories, calculate the scattering clusters distributed on the sea surface and the positions of the corresponding scattering points, and calculate the amplitudes and time delays corresponding to all signal paths to obtain the channel impulse response, as shown in the following formula:

[0018]

[0019] Among them, f c is the carrier frequency, K is the Rice factor, and τ LoS is the delay of the line-of-sight path, a m,n and τ m,n are respectively the amplitude and delay of the m-th non-line-of-sight path in the n-th cluster pair, and δ(·) is the Dirac function;

[0020] The signal path includes a line-of-sight path and non-line-of-sight paths.

[0021] As a preferred technical solution, it includes the following steps:

[0022] Assume that there are a total of N t ×M n non-line-of-sight paths in the sea surface communication channel, and all non-line-of-sight paths are divided into N t groups, each group contains M n scattering paths with similar delays. Among them, the n-th group corresponds to a pair of scattering clusters n ∈ {1, …, N t}, is the first-hop scattering cluster on the transmitter side, is the last-hop scattering cluster on the receiver side;

[0023] Calculate the positions of the scattering clusters on the transmitter side and the receiver side, and calculate the delays and amplitudes of the corresponding scattering paths;

[0024] According to the motion trajectory and the horizontal angle and elevation angle of the n-th group of scattering cluster pairs calculate the direction vector from the transmitter to the cluster center and the distance

[0025]

[0026] as follows: L where A

[0027] represents the antenna coordinates of the transmitter and the receiver, Z(·) represents taking the z component of this vector, and L ∈ {T, R} represents the transmitter and the receiver, where T corresponds to the transmitter and R corresponds to the receiver; Calculate the cluster center

[0028]

[0029] The scattering point corresponding to the (m, n)-th path is defined as:

[0030]

[0031] Among them, are respectively the offsets of the scattering points relative to the cluster center in the {x, y, z} axis directions, and respectively follow a zero-mean Gaussian distribution with a standard deviation of ;

[0032] Define the distance from the transmitter to the scattering point corresponding to the (m, n)-th path in the first-hop scattering cluster as and the distance from the receiver to the scattering point corresponding to the (m, n)-th path in the last-hop scattering cluster as Then the propagation delay τ of the (m, n)-th path m,n , is as follows:

[0033]

[0034] where c is the speed of light, ‖·‖ represents the L2 norm of a vector, is the virtual link delay that follows an exponential distribution;

[0035] According to the delay τ m,n , obtain the power γ of the non-line-of-sight path m,n and calculate the non-line-of-sight path amplitude

[0036] As a preferred technical solution, the delay τ of the line-of-sight path LoS , is as follows:

[0037]

[0038] d LoS =‖A T -A R ‖;

[0039] where d LoS is the length of the line-of-sight path, A T is the antenna coordinate of the transmitter, and A R is the antenna coordinate of the receiver.

[0040] As a preferred technical solution, the calculating the antenna position offset of the transmitter and the receiver at the same time according to the movement trajectory includes the following steps:

[0041] Obtain the direction vector r from the transmitter to the receiver, as follows:

[0042]

[0043] Among them, represents the departure elevation angle of the line-of-sight path of the signal from the transmitter to the receiver, Represents the departure horizontal angle of the line-of-sight path of the signal from the transmitter to the receiver;

[0044] Calculate the Doppler frequency shift f of the channel D , as follows:

[0045]

[0046] d w = D R sin(θ R ) - D T sin(θ T );

[0047] Where, A T is the antenna coordinate of the transmitter, A R is the antenna coordinate of the receiver, f c represents the carrier frequency, c represents the speed of light, ° represents the dot product operation, v T represents the speed of the transmitter moving along the y-axis direction, v R represents the speed of the receiver moving along the y-axis direction, d w represents the additional moving distance of the antenna under the influence of sea waves, D T represents the height of the transmitter antenna, D R represents the height of the receiver antenna, θ T represents the pitch angle of the transmitter, θ R represents the pitch angle of the receiver;

[0048] In the t-th time interval, perform a differential calculation on d w to obtain the sea wave characteristics of the transmitter and the receiver.

[0049] As a preferred technical solution, the conversion of the channel impulse response to the frequency-domain channel response is specifically:

[0050] Perform a discrete Fourier transform on the time-domain channel impulse response generated based on SW-GBSM to obtain the frequency-domain channel response;

[0051] In the t-th time interval, the received frequency-domain signal is as follows:

[0052] Y t = H t · X t + W t ;

[0053] Where, X t represents the transmitted signal, H t represents the frequency-domain channel response, W t represents complex additive white Gaussian noise;

[0054] The estimated value of the frequency domain channel response is obtained through the least squares channel estimation algorithm, as shown in the following formula:

[0055]

[0056] in, is the noise variance;

[0057] In the tth time interval, the frequency domain channel response sequence composed of the first P frequency domain channel responses is as follows:

[0058]

[0059] As a preferred technical solution, the sea wave features SWF-TA of the transmitter and the receiver are concatenated with the frequency domain channel response to form a joint input feature vector, including:

[0060] The frequency domain channel response at the corresponding time and the wave characteristics of the transmitter and receiver are spliced ​​as follows:

[0061]

[0062] in, is the frequency domain channel response at the tth time interval, and are operations for extracting the real and imaginary parts, respectively. The wave characteristics for the transmitter and receiver.

[0063] As a preferred technical solution, the P historical joint input feature vectors are sequentially input into the SWF-GRU channel prediction network, specifically:

[0064] The historical data sequence x of length P i , i∈{t-P+1,…,t-1,t} is input into the network in time order. According to the input data, N G The cascaded GRU is reset by the gate and update gate Hiding from oneself Update and extract the time-dependent features of the data. For the t-th time interval, the k-th GRU update step is as follows:

[0065]

[0066] k=1,…,N G ;

[0067] in, is a candidate hidden state, σ(·) and tanh(·) are the Sigmoid function and tanh function respectively, ⊙ represents the Hadamard product operation, and are the weights and biases of the GRU respectively;

[0068] The fully connected layer maps the hidden state of the last GRU to the network output as follows:

[0069]

[0070] As a preferred technical solution, when training the SWF-GRU channel prediction network, the following loss function is adopted:

[0071]

[0072] where J represents the number of samples, is the loss term based on the normalized mean square error, which is used to measure the accuracy of channel prediction, represents the auxiliary loss caused by the SWF-TA prediction error, represents the true frequency-domain channel response at time t + 1, represents the predicted frequency-domain channel response at time t + 1, represents the true SWF-TA at time t + 1, represents the predicted SWF-TA at time t + 1, and λ represents the hyperparameter used to balance the contributions of the two losses.

[0073] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0074] (1) The SW-GBSM modeling method proposed by the present invention provides a more accurate description of the antenna trajectory by combining the pitching motion of the transceiver. The antenna trajectory is determined by the joint calculation of the sea surface height and the transceiver pitching angle, and the change of the Doppler frequency shift caused by the sea wave fluctuation is simulated accordingly, so that a channel modeling effect more in line with the real sea surface environment can be achieved.

[0075] (2) Aiming at the problem that the prediction performance of the existing neural network is limited due to only learning the non-stationary characteristics of channel data in an implicit manner, the present invention proposes to calculate the position offset of the transceiver antenna caused by sea waves, define the difference as the sea wave-oriented feature of the transceiver antenna (SWF-TA), and then explicitly incorporate this information into the GRU network, so as to assist the network in capturing the change characteristics of the Doppler frequency shift under sea wave disturbance. This SWF-GRU-based channel prediction method can greatly improve the performance of sea surface channel prediction on the premise of increasing the lower computational complexity.

[0076] (3) The present invention proposes to introduce an auxiliary loss function based on SWF-TA during the training process. This auxiliary loss function helps to improve the robustness of the SWF-GRU prediction network, especially in prediction tasks involving multiple time intervals (TIs), and enhances the adaptability of the SWF-GRU to non-stationary channel changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0077] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0078] Figure 1 It is a flowchart of the sea surface channel prediction method based on the pitch motion characteristics of the device in the embodiments of the present invention;

[0079] Figure 2 It is a schematic diagram of the application scenario of the system in the embodiments of the present invention;

[0080] Figure 3 It is a schematic diagram of the SW-GBSM channel model in the embodiments of the present invention;

[0081] Figure 4 It is a schematic diagram of the SWF-GRU network structure in the embodiments of the present invention;

[0082] Figure 5 It is a comparison between SW-GBSM and GBSM and the measured channel in the embodiments of the present invention;

[0083] Figure 6 It is a comparison of the prediction performance of each method in the embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0084] To enable those skilled in the art to better understand the solutions of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present application.

[0085] References to "embodiments" in this application mean that specific features, structures, or characteristics described in connection with the embodiments can be included in at least one embodiment of the application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described in this application can be combined with other embodiments.

[0086] Embodiment 1

[0087] Please refer to Figure 1 , this embodiment provides a method for predicting the sea surface channel based on the pitching motion characteristics of the device, including: a modeling stage and a prediction stage.

[0088] Modeling stage:

[0089] S1. Determine the positions of the transmitter and receiver on the sea surface, and obtain the system parameters and sea surface environment parameters.

[0090] For the convenience of marking the positions of the transmitter and receiver and intuitive observation, as Figure 2 shown, for the general ship-to-ship communication scenario on the sea surface, the details of the proposed method are described.

[0091] Assume that the distance between the transmitter (Transmitter, Tx) and the receiver (Receiver, Rx) is less than the distance at which the evaporation duct effect occurs. The sea surface communication channel includes a line-of-sight (Light-of-Sight, LoS) path and a non-line-of-sight (Non-Light-of-Sight, NLoS) path generated by sea surface scattering.

[0092] Assume that at the t-th time interval TI, there are a total of N t ×M n NLoS paths. All the NLoS paths are divided into N t groups, and each group contains M n scattering paths with similar time delays. Among them, the n-th group (n = 1,..., N t ) corresponds to a pair of scattering clusters is the first-hop scattering cluster on the Tx side, is the last-hop scattering cluster on the Rx side. The centers of the clusters are distributed on the sea surface, as shown by the black solid dots in Figure 2 . The blue circles in the figure correspond to the scattering points corresponding to the scattering paths divided into the same group. The signal transmitted by Tx first enters the first-hop scattering cluster and undergoes multiple scattering propagations under the action of sea surface scattering, and then enters the last-hop scattering cluster and finally reaches Rx. To simplify the modeling process, and The multiple-scattering propagation between them is equivalent to a virtual link with a specific time delay. By synthesizing the LoS path and the NLoS path, the CIR of the sea surface channel can be modeled as:

[0093]

[0094] where f c is the carrier frequency, K is the Rice factor, τ LoS (t) is the time delay of the LoS path, a m,n (t) and τ m,n (t) are the amplitude and time delay of the m-th NLoS path in the b-th cluster pair respectively, and δ(·) is the Dirac function.

[0095] At this time, in this embodiment, the system parameters and sea surface environment parameters are obtained by calculation or directly through sensors, and subsequent channel prediction is carried out based on these parameters.

[0096] S2. Calculate the motion trajectories of the transmitter and receiver's heaving and pitching on the sea surface based on the obtained parameters, model the channel using SW-GBSM according to the motion trajectories to obtain the channel impulse response, and at the same time calculate the antenna position offset of the transmitter and receiver at the same time according to the motion trajectories to obtain the sea wave characteristics of the transmitter and receiver.

[0097] In this embodiment, the motion trajectory is the change in the motion state of the transmitter or receiver, specifically the connection line of the antenna coordinates in the system. Under the action of random sea waves, the motion states of Tx and Rx will change, which will affect the position trajectory of the antenna and ultimately change the CIR. Different from the existing method that only considers the heaving motion of the transceiver, this embodiment further improves the accuracy of channel modeling by introducing the dynamic changes of the pitching motion of Tx and Rx, making it more suitable for the design of communication systems in the dynamic sea surface environment.

[0098] As Figure 3 shown, it shows the schematic diagram of the SW-GBSM channel model proposed in this embodiment. For the sake of simplicity, only the n-th cluster pair is drawn in the figure. Assume that Tx and Rx move along the y-axis at speeds of {v T ,v R}, the corresponding antenna heights are {D T ,D R}, and the three-dimensional (3D) coordinates of the antennas are {A T ,A R}. For the n-th cluster pair denote the 3D coordinates corresponding to the cluster center as {C T ,C R}, and the signal travels from A T to C TThe azimuth angle of departure (AAoD) and elevation angle of departure (EAoD) are respectively and The azimuth angle of arrival (AAoA) and elevation angle of arrival (EAoA) of the signal from C R to A R are respectively and For simplicity of description, hereinafter, L ∈ {T, R} is used to replace the superscripts regarding Tx and Rx in the variables.

[0099] The specific modeling method is as follows:

[0100] First, the present invention uses the classical Pierson-Moskowitz (P-M) sea wave spectrum and the linear superposition theory to simulate the random fluctuations of sea waves. Specifically, the height η and sea wave inclination α at a certain point on the sea surface can be expressed as:

[0101]

[0102] In the formula, L is the number of sine waves, ω l and ε l are respectively the angular frequency and phase of the l-th sine wave, ε l obeys the uniform distribution in (0, 2π], Δω is the frequency interval between ω l and ω l+1 , S(ω) is the P-M sea wave spectrum, and its specific expression is as follows:

[0103]

[0104] In the formula, u and β are constants, taking u = 8.1×10 -3 and β = 0.74 respectively; g is the gravitational acceleration, with a value of 9.8 m / s 2 ; ω is the angular frequency of the sea wave; U is the magnitude of the wind speed at 19.5 m above the sea level.

[0105] In this embodiment, Equation (2) is used to describe the height of the heaving motion of Tx or Rx on the sea surface, denoted as η L (t).

[0106] Then, according to the displacement ∈ of the ship, the initial metacentric height μ, the pitching moment of inertia J θ and the damping coefficient N θ , the response function from the sea wave inclination α to the pitching angle θ of the ship can be obtained, expressed as:

[0107]

[0108] In the formula Subsequently, through the inverse Laplace transform and convolution theory, according to Equations (3) and (5), θ(t) can be calculated, and then the pitch angle θ when pitching motions occur at Tx and Rx can be obtained L (t). For simplicity of description, the time index t is omitted in the variable description of the SW-GBSM channel modeling hereinafter

[0109] Next, in this embodiment, the heaving height η L and the pitch angle θ L are introduced when calculating the antenna coordinates of Tx and Rx

[0110] Specifically, assuming that the sea level corresponds to the z = 0 plane, the ocean waves propagate along the y-axis, causing the transceiver to have a heaving motion in the z-axis direction and a pitching motion around the x-axis. Let the projections of the Tx and Rx coordinates on the sea level be [x L , y L , 0]. Then, based on the theory of solid geometry, the antenna coordinates A L of Tx and Rx can be calculated by the following formula

[0111] A L = [x L , y L + D L sin(θ L ), η L + D L cos(θ L )] (6)

[0112] In the formula, D L sin(θ L ) and D L cos(θ L ) are the non-stationary displacements of the transceiver antennas caused by the pitching motion, while η L is the influence caused by the heaving motion. It should be noted that Equation (6) combines the pitching motion and the heaving motion to describe the actual positions of the ship antennas, and can accurately simulate the non-stationary Doppler effect generated by the random ocean waves on the channel. In addition, since the ship moves along the y-axis direction and the distance between ships is much larger than the antenna height, compared with the heaving motion, the pitching motion has a more significant impact on the channel

[0113] Subsequently, in this embodiment, according to the antenna coordinates A derived from Equation (6) L, calculate the scattering clusters distributed on the sea surface and the positions of the corresponding scattering points, and further calculate the amplitude and delay corresponding to all paths, so as to construct a complete CIR. First, as the transceiver moves, some NLoS paths may be interrupted, resulting in the disappearance of cluster pairs, and at the same time, new NLoS paths may appear, thus forming new cluster pairs. Therefore, the number N of cluster pairs t will change with the movement of the Tx and Rx. In this embodiment, the birth and death process of cluster pairs is modeled based on the survival probability model of the Poisson process. For the N t cluster pairs at the current moment, the horizontal angle and elevation angle of each pair of clusters are generated according to the Gaussian distribution and the truncated Gaussian distribution respectively. Taking the nth cluster pair as an example, since the centers of the clusters are distributed on the sea surface, according to the antenna coordinates A L and the horizontal angle and elevation angle corresponding to the cluster the direction vectors from the Tx and Rx to the cluster center can be calculated and the distance The specific calculation formulas are as follows:

[0114]

[0115] where Z(·) represents taking the z component of the vector. Then, the center of the cluster can be calculated:

[0116]

[0117] Furthermore, the scattering point corresponding to the (m,n)th path is defined as:

[0118]

[0119] In the formula, are the offsets of the scattering point relative to the cluster center in the {x,y,z} axis directions respectively, and they respectively follow the zero-mean Gaussian distribution with a standard deviation of . Then, define the distance from the transmitter to the scattering point corresponding to the (m,n)th path in the first-hop scattering cluster as the distance from the receiver to the scattering point corresponding to the (m,n)th path in the last-hop scattering cluster as Then the propagation delay τ of the (m,n)th path m,n is calculated by the following formula:

[0120]

[0121] where c is the speed of light, ‖·‖ represents the L2 norm of the vector, is the virtual link delay that follows the exponential distribution. According to the delay τ m,n , the power γ of the path can be generated m,n , and the path amplitude is further calculated In addition, the time delay of the LoS path is calculated by the following formula:

[0122]

[0123] where d LoS = ‖A T - A R ‖ is the length of the LoS path.

[0124] So far, based on the calculated a m,n , τ m,n and τ LoS , the complete CIR in Equation (1) can be obtained, and the modeling of the sea surface communication channel is completed.

[0125] Note that the proposed SW-GBSM channel modeling method can reflect the non-stationary Doppler frequency shift caused by ocean waves, and by predicting the future channel, the transceiver can respond to this in advance, alleviating the impact of this non-stationary effect on the communication system. Different from the existing methods that use complex network enhancement technologies such as attention mechanisms to improve the channel prediction performance, this embodiment proposes a SWF-GRU channel prediction method based on sea wave features SWF-TA. This method can assist the network to better capture the changing characteristics of the Doppler frequency shift of the sea surface channel under the premise of a relatively low computational complexity overhead, and achieve better channel prediction performance. For details, see step S3.

[0126] Before constructing the SWF-GRU channel prediction network, it is necessary to obtain the sea wave features SWF-TA of the transmitter and receiver first, including the following steps:

[0127] Since in the sea surface communication scenario, the power of the LoS path is usually much higher than that of the NLoS path, this embodiment mainly considers the impact of sea wave movement on the LoS path. Define the EAoD and AAoD of the LoS path as and The direction vector from Tx to Rx can be expressed as:

[0128]

[0129] Without loss of generality, assume that d LoS is much larger than the antenna height D L , it can be approximately considered that and v L << c. According to Equation (6) and Equation (11), the Doppler frequency shift of the channel can be calculated:

[0130]

[0131] where ° represents the dot product operation, d wRepresents the additional moving distance of the antenna under the influence of ocean waves. The specific calculation formula is as follows:

[0132] d w =D R sin(θ R ) - D T sin(θ T ) (13)

[0133] Note that d w can be calculated from the {θ T , θ R} measured by the sensors integrated on the Tx and Rx and the known antenna heights {D T , D R}. At the t-th time interval, the first-order difference of d w is:

[0134]

[0135] Describes the non-stationary Doppler frequency shift generated by random ocean waves. Since will affect the channel phase, in this embodiment, it is used as SWF-TA and input into the GRU network to improve the network prediction performance.

[0136] Prediction stage:

[0137] S3. Convert the channel impulse response to the frequency-domain channel response; construct the SWF-GRU channel prediction network, which includes multiple cascaded GRUs and a fully connected layer. Concatenate the ocean wave features SWF-TA of the transmitter and receiver with the frequency-domain channel response to form a joint input feature vector, and sequentially input P historical joint input feature vectors into the SWF-GRU channel prediction network. The cascaded GRUs gradually update their hidden states according to the input sequence, and the fully connected layer maps the final hidden state of the GRU to the predicted frequency-domain channel response; during the training process, construct a loss function using the normalized mean square error between the predicted frequency-domain channel response and SWF-TA and the true value, and calculate the gradient and update the SWF-GRU network parameters through the backpropagation algorithm.

[0138] In this embodiment, first, the channel impulse response in the time domain needs to be subjected to discrete Fourier transform to obtain the frequency-domain channel response FCR.

[0139] Next, at the t-th time interval, the received frequency-domain signal can be expressed as:

[0140] Y t =H t ·X t +Wt (15)

[0141] Where X t represents the transmitted signal, H t represents the frequency-domain channel response FCR, and W t is the additive white Gaussian noise (AWGN). Given the known pilot symbol X t , in this embodiment, the least square (LS) channel estimation algorithm is adopted to obtain an estimated value of the frequency-domain channel response FCR:

[0142]

[0143] where is the noise variance. Furthermore, at the t-th TI, the frequency-domain channel response FCR sequence composed of the first P frequency-domain channel responses FCR can be expressed as:

[0144]

[0145] In this embodiment, the goal of channel prediction is to predict the frequency-domain channel response FCR of the next TI based on , that is Subsequently, the frequency-domain channel responses FCR of the future Q TIs can be sequentially predicted by the sliding time window method.

[0146] Next, as Figure 4 shown, the SWF-GRU channel prediction network includes N G cascaded GRUs and a fully connected layer.

[0147] In this embodiment, the frequency-domain channel response FCR and SWF-TA at the corresponding moment are concatenated as the input of the network, expressed as:

[0148]

[0149] where and are the operations of extracting the real part and the imaginary part respectively. The operation process of SWF-GRU is as follows. First, the historical data sequence x i , i ∈ {t - P + 1, …, t - 1, t} is input into the network in chronological order. Then, according to the input data, N G cascaded GRUs update their own hidden states and through the reset gate to extract the time-dependent features of the data. Specifically, at the t-th TI, the k-th GRU (k = 1, …, N G) The update steps are as follows:

[0150]

[0151] Where is the candidate hidden state, σ(·) and tanh(·) are the Sigmoid function and the tanh function respectively, ⊙ represents the Hadamard product operation, and are the weights and biases of the GRU network respectively. Finally, a fully connected layer maps the hidden state of the last GRU to the network output:

[0152]

[0153] In the network training stage, the following loss function is adopted in this embodiment:

[0154]

[0155] Where J represents the number of samples. In Equation (21), is the loss term based on the Normalized Mean Squared Error (NMSE), which is used to measure the accuracy of channel prediction. In addition, different from the existing work that only considers the channel prediction error as the loss function, in order to improve the robustness of the network in multiple TI predictions, this embodiment introduces This term represents the auxiliary loss caused by the SWF-TA prediction error. And, this embodiment introduces the hyperparameter λ to balance the contributions of these two losses.

[0156] After the training is completed, the final output is the predicted value of the final frequency-domain channel response FCR.

[0157] Embodiment 2

[0158] To more fully elaborate the beneficial effects of this embodiment, the following further illustrates the effectiveness and advancement of this embodiment in combination with the simulation analysis and results of specific embodiments.

[0159] First, the SW-GBSM channel modeling method in this embodiment is used to generate the CIR under the same conditions as the measurement experiment of Maliatso, and further the cumulative distribution function (CDF) of its root mean square delay spread σ τ is compared with the measurement results of Maliatso. In the simulation, f c = 1.9 GHz, and the distance d between the Tx and the Rx T,R ∈[100,4000] m. Figure 5Part (a) shows the σ of the CIR generated by SW-GBSM τ The distribution highly matches the measurement results, verifying the accuracy of SW-GBSM. Further, in this embodiment, the Tx / Rx is tested for different moving speeds {v T , v R} when moving along the trajectory measured in the literature, and the peak Doppler frequency shift f m is measured and compared with the measurement results of the GBSM method and Wang's. As shown in part (b) of Figure 5 , the difference d y = y R - y T in the y-coordinates of the transceiver is used to describe the movement process of the Tx / Rx in this trajectory. It can be seen that the simulation results of SW-GBSM are consistent with Wang's measured data, and local jitters appear on the main curve, indicating that the pitching motion caused by the sea waves will affect the Doppler frequency shift. In contrast, the simulation curve of GBSM does not show the influence of sea waves on the Doppler frequency shift because it does not consider the pitching motion.

[0160] In addition, in this embodiment, the proposed SW-GBSM is used to generate channel data to demonstrate the excellent performance of the proposed SWF-GRU method. In the following tests, LSTM and GRU without combining SWF-TA, Seq2Seq-attn-R using the attention mechanism, and the LSTM method (SWF-LSTM) combining SWF-TA are selected as comparison methods. Unless otherwise specified, the main simulation parameter settings for the following tests are shown in Table 1.

[0161] Table 1

[0162]

[0163] First, this embodiment discusses the selection of the historical channel length P and the hyperparameter λ in the method. Figure 6 Part (a) shows the influence of P on the NMSE performance of the SWF-GRU method under different Rice factors K. The results show that a larger K enhances the correlation between adjacent channel samples, which is beneficial to channel prediction. However, when P is too large, the correlation of channel samples will weaken. Therefore, an appropriate P needs to be selected to optimize the prediction performance. Figure 6 Part (b) shows that the overall performance of SWF-GRU is better than that of SWF-LSTM under different hyperparameters λ. In addition, there is an optimal λ value to achieve the balance of the losses of and the two parts.

[0164] Next, this embodiment compares the performance of different methods. In the following tests, set P = 9 and K = 10 dB. The hyperparameters λ SWF-GRU and λ SWF-LSTM of SWF-GRU and SWF-LSTM are set to 1.2 and 0.6 respectively. Figure 6 (c) shows that the introduction of SWF-TA has obvious gains for both GRU and LSTM methods. In addition, as can be seen from Figure 6 (d), SWF-GRU and SWF-LSTM show strong robustness in the prediction of multiple TIs. The simulation results show that the SWF-TA designed by combining sea wave characteristics can effectively assist the prediction of the sea surface channel, making the network prediction performance better than the Seq2Seq-attn-R method enhanced by the attention mechanism.

[0165] Table 2

[0166]

[0167] As shown in Table 2, the comparison of the floating-point operation counts and the complexity of the number of network parameters of different methods shows that the SWF-TA feature introduced in this embodiment only brings a slight increase in complexity. Compared with the Seq2Seq-attn-R method using the attention mechanism, the SWF-GRU method in this embodiment achieves better prediction performance while maintaining a lower computational complexity.

[0168] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations. However, those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously.

[0169] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0170] The above embodiments are the preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications made without departing from the spirit and principle of the present invention shall be equivalent replacement methods and are all included in the protection scope of the present invention.

Claims

1. A sea surface channel prediction method based on equipment pitch motion characteristics, characterized in that: include: Modeling phase: Determine the location of the transmitter and receiver on the sea surface, and obtain system parameters and sea surface environment parameters; Based on the acquired parameters, the heave and pitch motion trajectories of the transmitter and receiver on the sea surface are calculated. According to the motion trajectories, the channel is modeled using SW-GBSM to obtain the channel impulse response. At the same time, the antenna position offsets of the transmitter and receiver at the same time are calculated according to the motion trajectories to obtain the wave characteristics SWF-TA of the transmitter and receiver. Prediction stage: Convert the channel impulse response into a frequency domain channel response; construct a SWF-GRU channel prediction network, the SWF-GRU channel prediction network includes a plurality of cascaded GRUs and a fully connected layer, concatenate the wave features SWF-TA of the transmitter and the receiver with the frequency domain channel response to form a joint input feature vector, input P historical joint input feature vectors into the SWF-GRU channel prediction network in sequence, the cascaded GRU gradually updates its hidden state according to the input sequence, and the fully connected layer maps the final hidden state of the GRU to the predicted frequency domain channel response; During the training process, the loss function is constructed using the predicted frequency domain channel response and the normalized mean square error between the SWF-TA and the true value, and the gradient is calculated and the SWF-GRU network parameters are updated through the back propagation algorithm.

2. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The step of calculating the heave and pitch motion trajectories of the transmitter and the receiver on the sea surface based on the acquired parameters includes: Define the height of a certain point on the sea surface and the wave inclination angle according to the PM wave spectrum and linear superposition theory; The height of a certain point on the sea surface is used to describe the height at which the transmitter and the receiver on the sea surface undergo heave motion; a response function from the wave inclination angle to the ship pitch angle is obtained according to the system parameters; and the ship pitch angle is obtained according to the wave inclination angle and the response function from the wave inclination angle to the ship pitch angle, wherein the ship pitch angle is consistent with the pitch angle corresponding to the transmitter and the receiver undergoing pitch motion; the system parameters include the ship's displacement, initial stability height, pitch moment of inertia, and damping coefficient; Based on the theory of solid geometry, the antenna coordinates of the transmitter and receiver are calculated using the heights at which the transmitter and receiver occur heave motion above the sea surface, as well as the corresponding pitch angles when pitch motion occurs.

3. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The channel is modeled using SW-GBSM according to the motion trajectory, specifically: According to the motion trajectory, the scattering clusters distributed on the sea surface and the positions of the corresponding scattering points are calculated, and the amplitudes and delays corresponding to all channel paths are calculated to obtain the channel impulse response, as shown in the following formula: Among them, f c is the carrier frequency, K is the Rice factor, τ LoS is the delay of the line-of-sight path, a m,n and τ m,n are the amplitude and delay of the mth non-line-of-sight path in the nth cluster pair, respectively, and δ(·) is the Dirac function; The channel path includes a line-of-sight path and a non-line-of-sight path.

4. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 3, characterized in that: The following steps are involved: Assume that there are N t ×M n All non-line-of-sight paths are divided into N t groups, each containing M n scattering paths that experience similar delays, where the nth group corresponds to a pair of scattering clusters n∈{1,…,N t }, is the first-hop scattering cluster at the transmitter side, is the last hop scattering cluster at the receiver side; Calculate the scattering cluster positions on the transmitter and receiver sides, and calculate the delay and amplitude of the corresponding scattering path, specifically: According to the motion trajectory and the horizontal angle and elevation angle of the nth group of scattering clusters Calculate the direction vector from the transmitter to the cluster center and distance As follows: Among them, A L represents the antenna coordinates of the transmitter and receiver, Z(·) represents taking the z component of the vector, L∈{T,R} represents the transmitter and receiver, where T corresponds to the transmitter and R corresponds to the receiver; Computing Cluster Center As follows: The scattering point corresponding to the (m,n)th path is defined as: in, are the scatter points relative to the cluster center The offsets in the {x, y, z} directions are subject to standard deviations of The zero-mean Gaussian distribution of The distance from the transmitter to the scattering point corresponding to the (m,n)th path in the first-hop scattering cluster is defined as The distance from the receiver to the scattering point corresponding to the (m,n)th path in the last hop scattering cluster is Then the propagation delay τ of the (m,n)th path is m,n , as follows: Where c is the speed of light, ‖·‖ represents the L2 norm of the vector, is the virtual link delay that obeys exponential distribution; According to the time delay τ m,n , get the power γ of the non-line-of-sight path m,n And calculate the non-line-of-sight path amplitude 5. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 3, characterized in that: The delay τ of the line-of-sight path LoS , as follows: d LoS =‖A T -A R ‖; Among them, d LoS is the length of the line-of-sight path, A T is the antenna coordinate of the transmitter, A R are the antenna coordinates of the receiver.

6. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The step of calculating the antenna position offset of the transmitter and the receiver at the same time according to the motion trajectory comprises the following steps: Get the direction vector r from the transmitter to the receiver as follows: in, represents the departure elevation angle of the signal from the transmitter to the receiver’s line-of-sight path, It represents the departure horizontal angle of the line-of-sight path of the signal from the transmitter to the receiver; Calculate the Doppler frequency shift f of the channel D , as follows: d w =D R sin(θ R )-D T sin(θ T ); Among them, A T is the antenna coordinate of the transmitter, A R is the antenna coordinate of the receiver, f c represents the carrier frequency, c represents the speed of light, ° represents the dot product operation, v T represents the speed of the transmitter moving along the y-axis, v R represents the speed of the receiver moving along the y-axis, d w represents the additional movement distance of the antenna under the influence of waves, D T Denotes the transmitter antenna height, D R represents the receiver antenna height, θ T represents the pitch angle of the transmitter, θ R Indicates the pitch angle of the receiver; In the tth time interval, for d w Perform differential calculations to obtain the wave characteristics of the transmitter and receiver.

7. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The channel impulse response is converted into a frequency domain channel response, specifically: Perform discrete Fourier transform on the time domain channel impulse response generated based on SW-GBSM to obtain the frequency domain channel response; In the tth time interval, the received frequency domain signal is as follows: Y t =H t ·X t +W t ; Among them, X t Indicates sending signal, H t represents the frequency domain channel response, W t represents complex additive Gaussian white noise; The estimated value of the frequency domain channel response is obtained through the least squares channel estimation algorithm, as shown in the following formula: in, is the noise variance; At the tth time interval, the frequency domain channel response sequence composed of the first P frequency domain channel responses is as follows:

8. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The step of concatenating the wave characteristics SWF-TA of the transmitter and the receiver with the frequency domain channel response to form a joint input feature vector includes: The frequency domain channel response at the corresponding time and the wave characteristics of the transmitter and receiver are spliced ​​as follows: in, is the frequency domain channel response at the tth time interval, and are operations for extracting the real and imaginary parts, respectively. The wave characteristics for the transmitter and receiver.

9. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The P historical joint input feature vectors are sequentially input into the SWF-GRU channel prediction network, specifically: The historical data sequence x of length P i , i∈{t-P+1,…,t-1,t} is input into the network in time order. According to the input data, N G The cascaded GRU is reset by the gate and update gate Hiding from oneself Update and extract the time-dependent features of the data. For the t-th time interval, the k-th GRU update step is as follows: k=1,…,N G ; in, is a candidate hidden state, σ(·) and tanh(·) are the Sigmoid function and tanh function respectively, b represents the Hadamard product operation, and are the weights and biases of GRU respectively; The fully connected layer maps the hidden state of the last GRU to the network output as follows:

10. The method for predicting sea surface channels based on equipment pitch motion characteristics according to claim 1, characterized in that: The following loss function is used when training the SWF-GRU channel prediction network: Where J represents the number of samples, is a loss term based on normalized mean square error, used to measure the accuracy of channel prediction. represents the auxiliary loss due to SWF-TA prediction error, represents the actual frequency domain channel response at time t+1, represents the frequency domain channel response predicted at time t+1, represents the real SWF-TA at time t+1, represents the SWF-TA predicted at time t+1, and λ represents a hyperparameter used to balance the contribution of the two losses.

Images