LiFi communication control method and device based on deep learning, medium and equipment

Abstract

The invention discloses a deep learning-based LiFi communication control method and device, a medium and equipment, and the method comprises the steps: obtaining a pilot frequency transmission interval time and a channel theoretical value at the current moment based on the flight state data of an unmanned plane at the current moment and the installation information of a transmitter and a detector; calculating a first pilot frequency power to enable the transmitter to transmit light carrying a pilot frequency signal according to the pilot frequency transmission interval time and the first pilot frequency power; acquiring second pilot frequency power of the light carrying the pilot frequency signal received by the detector, and inputting the second pilot frequency power, the first pilot frequency power and the flight state data into a first channel prediction model to obtain a first channel prediction value at the current moment; and determining a beamforming mode of the transmitter based on the first channel predicted value, the channel theoretical value and the channel actual value, and determining a beam weight based on the beamforming mode, so that the transmitter regulates the light emitting power based on the beam weight. According to the method, the beam weight is determined according to the time-varying characteristic of the LiFi channel, and the communication quality is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle communication, in particular to a LiFi communication control method and device based on deep learning, a medium and equipment. BACKGROUND

[0002] With the rapid development of unmanned aerial vehicle technology, unmanned aerial vehicle groups have shown a wide application prospect in the fields of power inspection, emergency rescue, intelligent monitoring, etc., and therefore the cooperative operation, data transmission and task scheduling among unmanned aerial vehicles of the unmanned aerial vehicle groups put forward higher requirements for stable and reliable communication links. The existing communication technologies among unmanned aerial vehicles mainly include Wi-Fi, 4G / 5G, LoRa and LiFi technologies.

[0003] Wi-Fi, 4G / 5G and LoRa perform well in open environments, but perform poorly in restricted environments (such as high-voltage substations and underground tunnels). LiFi communication technology performs well in fixed scenarios such as offices and conference rooms, but there are some defects in mobile platforms, especially unmanned aerial vehicle platforms. For example, it is difficult to accurately capture the time-varying characteristics of the mobile LiFi channel, resulting in a large estimation error, which in turn affects the communication quality. Under the condition of imperfect channel state information, the beamforming performance decreases significantly, which cannot ensure the stability among multiple unmanned aerial vehicles, and in turn affects the quality of communication. SUMMARY

[0004] Therefore, the present application provides a LiFi communication control method based on deep learning, which mainly aims to solve the problem of low communication quality of the existing LiFi communication control method of unmanned aerial vehicles.

[0005] According to one aspect of the present application, a LiFi communication control method based on deep learning is provided, which comprises: obtaining flight state data of an unmanned aerial vehicle, first installation information of a transmitter and second installation information of a detector at a current time, calculating based on the flight state data, the first installation information and the second installation information to obtain a channel correlation time, and calculating a channel theoretical value at the current time based on the first installation information, the second installation information and a preset channel theoretical model; determining a pilot transmission interval time based on the channel correlation time, obtaining a total transmission power of the transmitter, calculating a first pilot power based on the total transmission power, so that the transmitter transmits light carrying a pilot signal at the pilot transmission interval time and the first pilot power; obtaining a second pilot power of the detector receiving the light carrying the pilot signal, inputting the second pilot power, the first pilot power and the flight state data into the trained first channel estimation model to obtain a first channel prediction value at the current time; If a difference between the first channel prediction value of the current moment and the channel theoretical value is less than or equal to a preset difference value, an actual channel value of the current moment is acquired, a beamforming mode of the transmitter is determined based on an error between the first channel prediction value of the current moment and the actual channel value, a beam weight is determined based on the beamforming mode, and the transmitter is controlled to adjust light emitting power based on the beam weight.

[0006] Optionally, the flight state data includes a speed and an angular velocity, the first mounting information includes a first mounting coordinate and a first mounting surface normal vector, the second mounting information includes a second mounting coordinate and a second mounting surface normal vector, and the calculation based on the flight state data, the first mounting information and the second mounting information to obtain a channel correlation time includes: calculating a linear velocity of the probe based on the speed, the angular velocity and the second mounting coordinate; calculating a light link unit vector based on the first mounting coordinate and the second mounting coordinate; calculating a relative linear velocity based on the linear velocity of the probe and the light link unit vector, and calculating the channel correlation time based on the relative linear velocity.

[0007] Optionally, the preset channel theoretical model is constructed by the following method, including: calculating a straight-line distance between the transmitter and the probe based on the first mounting coordinate and the second mounting coordinate, calculating an angle of a transmission angle and an angle of a receiving angle based on the light link unit vector, the first mounting surface normal vector and the second mounting surface normal vector, and determining an occlusion coefficient based on the angle of the transmission angle or the angle of the receiving angle; constructing a direct path channel gain model based on the occlusion coefficient, the straight-line distance between the transmitter and the probe, the angle of the transmission angle and the angle of the receiving angle; acquiring a coordinate of a virtual image point formed by the transmitter on a reference reflection surface, and calculating a first reflection path and a second reflection path based on the first mounting coordinate and the second mounting coordinate; calculating an angle of a reference incidence angle based on the second mounting coordinate, the coordinate of the virtual image point and the second mounting surface normal vector, and constructing a non-direct path channel gain model based on a reflection coefficient of the reference reflection surface, the angle of the transmission angle, the angle of the reference incidence angle, the first reflection path and the second reflection path; adding the direct path channel gain model and the non-direct path channel gain model to obtain the channel theoretical model.

[0008] Optionally, determining the pilot transmission interval based on the channel correlation time, obtaining the total transmit power of the transmitter, and calculating the first pilot power based on the total transmit power includes: Half of the channel correlation time is used as the pilot transmission interval time; The product of the total transmitted power and the power percentage is taken as the first pilot power.

[0009] Optionally, determining the beamforming mode of the transmitter based on the error between the first channel prediction value and the actual channel value at the current time, and determining the beam weight based on the beamforming mode, includes: If the error between the first channel prediction value and the actual channel value at the current time is less than or equal to a preset difference, then the perfect CSI beamforming mode is adopted, and the beam weight in the perfect CSI beamforming mode is calculated based on the first channel prediction value. If the error between the first channel prediction value and the actual channel value at the current time is greater than a preset difference, then imperfect CSI beamforming is adopted. An optimization model is constructed based on the first channel prediction value and the error. The optimization model is solved by semidefinite programming to obtain the beam weights under the imperfect CSI beamforming mode.

[0010] Optionally, the deep learning-based LiFi communication control method further includes: If the difference between the current channel prediction value and the theoretical channel value is greater than the preset difference, then the pilot transmission interval and the first pilot power are adjusted until the difference between the current channel prediction value and the theoretical channel value is less than or equal to the preset difference.

[0011] Optionally, the deep learning-based LiFi communication control method further includes: Obtain the channel state information of the current time and the preset time period before the current time, input the channel state information into the trained second channel prediction model, and obtain the second channel prediction value of the next time.

[0012] According to another aspect of this application, a deep learning-based LiFi communication control device is provided, comprising: The channel feature analysis module is used to acquire the flight status data of the UAV, the first installation information of the transmitter, and the second installation information of the detector at the current moment. Based on the flight status data, the first installation information, and the second installation information, it calculates the channel correlation time. Based on the first installation information, the second installation information, and the preset channel theory model, it calculates the channel theory value at the current moment. The pilot optimization module is used to determine the pilot transmission interval time based on the channel correlation time, obtain the total transmission power of the transmitter, and calculate the first pilot power based on the total transmission power, so that the transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power; The channel prediction module is used to obtain the second pilot power of the light carrying the pilot signal received by the detector, and input the second pilot power, the first pilot power and the flight state data into the trained first channel prediction model to obtain the first channel prediction value at the current time. The beam weight adjustment module is used to obtain the actual channel value at the current time if the difference between the first channel prediction value and the theoretical channel value at the current time is less than or equal to a preset difference; determine the beamforming mode of the transmitter based on the error between the first channel prediction value and the actual channel value at the current time; and determine the beam weight based on the beamforming mode so that the transmitter can adjust the emission power based on the beam weight.

[0013] Optionally, the flight status data includes velocity and angular velocity; the first installation information includes first installation coordinates and a first installation surface normal vector; the second installation information includes second installation coordinates and a second installation surface normal vector; and the channel feature analysis module is further used for: Based on the velocity, angular velocity, and second installation coordinates, the linear velocity of the detector is calculated; Based on the first installation coordinates and the second installation coordinates, the optical link unit vector is calculated; The relative linear velocity is calculated based on the linear velocity of the detector and the unit vector of the optical link, and the channel correlation time is calculated based on the relative linear velocity.

[0014] Optionally, the channel feature analysis module is further configured to: Based on the first installation coordinates and the second installation coordinates, the straight-line distance between the transmitter and the detector is calculated. Based on the optical link unit vector, the first mounting surface normal vector and the second mounting surface normal vector, the transmission angle and the reception angle are calculated. Based on the transmission angle or the reception angle, the blocking coefficient is determined. Based on the blocking coefficient, the straight-line distance between the transmitter and the detector, the angle of the transmission angle and the angle of the reception angle, a direct path channel gain model is constructed. The coordinates of the virtual image point formed by the transmitter on the reference reflective surface are obtained, and the first reflection path and the second reflection path are calculated based on the first installation coordinates and the second installation coordinates. Based on the second installation coordinates, the coordinates of the virtual image point, and the normal vector of the second installation surface, the angle of the reference incident angle is calculated. Based on the reflection coefficient of the reference reflecting surface, the angle of the emission angle, the angle of the reference incident angle, the first reflection path, and the second reflection path, a non-direct path channel gain model is constructed. The channel gain model of the direct path and the channel gain model of the indirect path are added together to form the channel theory model.

[0015] Optionally, the pilot optimization module is further configured to: Half of the channel correlation time is used as the pilot transmission interval time; The product of the total transmitted power and the power percentage is taken as the first pilot power.

[0016] Optionally, the beam weighting adjustment module is further configured to: If the error between the first channel prediction value and the actual channel value at the current time is less than or equal to a preset difference, then the perfect CSI beamforming mode is adopted, and the beam weight in the perfect CSI beamforming mode is calculated based on the first channel prediction value. If the error between the first channel prediction value and the actual channel value at the current time is greater than a preset difference, then imperfect CSI beamforming is adopted. An optimization model is constructed based on the first channel prediction value and the error. The optimization model is solved by semidefinite programming to obtain the beam weights under the imperfect CSI beamforming mode.

[0017] Optionally, the pilot optimization module is further configured to: If the difference between the current channel prediction value and the theoretical channel value is greater than the preset difference, then the pilot transmission interval and the first pilot power are adjusted until the difference between the current channel prediction value and the theoretical channel value is less than or equal to the preset difference.

[0018] Optionally, the channel prediction module is further configured to: Obtain the channel state information of the current time and the preset time period before the current time, input the channel state information into the trained second channel prediction model, and obtain the second channel prediction value of the next time.

[0019] According to another aspect of this application, a storage medium is provided, wherein at least one executable instruction is stored therein, the executable instruction causing a processor to perform the operations corresponding to the above-described deep learning-based LiFi communication control method.

[0020] According to another aspect of this application, a computer device is provided, comprising: a processor, a memory, a communication interface, and a communication bus, wherein the processor, the memory, and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the deep learning-based LiFi communication control method described above.

[0021] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages: This application provides a deep learning-based LiFi communication control method, apparatus, medium, and device. Based on the current UAV flight status data, transmitter installation information, and detector installation information, it calculates the channel correlation time and the theoretical channel value at the current moment. Based on the channel correlation time, it determines the pilot transmission interval. Based on the transmitter's total transmission power, it calculates the first pilot power. The transmitter transmits light carrying pilot signals according to the pilot transmission interval and the first pilot power. It obtains the second pilot power of the light carrying pilot signals received by the detector. The second pilot power, the first pilot power, and the flight status data are used as input data for a trained first channel prediction model to obtain the first channel prediction value at the current moment. Based on the first channel prediction value, the theoretical channel value, and the actual channel value, it determines the transmitter's beamforming mode. Based on the beamforming mode, it determines the beam weights, capturing the time-varying characteristics of the LiFi channel. By determining the beam weights based on these time-varying characteristics, even under imperfect beamforming modes, suitable beam weights can be determined, improving communication quality.

[0022] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0023] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1A flowchart of a deep learning-based LiFi communication control method provided in an embodiment of this application is shown; Figure 2 Another flowchart of a deep learning-based LiFi communication control method provided in an embodiment of this application is shown; Figure 3 This paper illustrates a structural block diagram of a deep learning-based LiFi communication control device according to an embodiment of this application. Figure 4 A schematic diagram of the structure of a computer device provided in an embodiment of the present invention is shown.

[0024] in, Figure 3 In the middle: 302-Channel Feature Analysis Module; 304-Pilot Optimization Module; 306-Channel Prediction Module; 308-Beam Weight Adjustment Module; Figure 4 In Chinese: 402 - Processor; 404 - Communication interface; 406 - Memory; 408 - Communication bus; 410 - Program. Detailed Implementation

[0025] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of the present invention can be combined with each other.

[0026] To further illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the specific embodiments, structures, features, and effects according to the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments. In the following description, different "an embodiment" or "an embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0027] To address the issue of low communication quality in existing LiFi communication control methods for UAVs, this application provides a deep learning-based LiFi communication control method, such as... Figure 1 As shown, the method includes: 102: Obtain the flight status data of the UAV, the first installation information of the transmitter, and the second installation information of the detector at the current moment. Calculate the channel correlation time based on the flight status data, the first installation information, and the second installation information. Calculate the channel theoretical value at the current moment based on the first installation information, the second installation information, and the preset channel theoretical model. 104: Based on the channel correlation time, determine the pilot transmission interval time, obtain the total transmission power of the transmitter, and calculate the first pilot power based on the total transmission power, so that the transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power; 106: Obtain the second pilot power of the light carrying the pilot signal received by the detector, input the second pilot power, the first pilot power and the flight state data into the trained first channel prediction model, and obtain the first channel prediction value at the current time; 108: If the difference between the first channel prediction value and the theoretical channel value at the current moment is less than or equal to a preset difference, obtain the actual channel value at the current moment. Based on the error between the first channel prediction value and the actual channel value at the current moment, determine the beamforming mode of the transmitter. Based on the beamforming mode, determine the beam weight so that the transmitter can adjust the emission power based on the beam weight.

[0028] Specifically, channel correlation time refers to the maximum time during which channel characteristics (such as gain and attenuation) remain stable. The UAV's flight status data includes its flight speed and angular velocity, while installation information includes installation coordinates and the normal vector of the installation surface. Channel correlation time is calculated based on the flight status data, transmitter installation information, and receiver installation information, making the calculation dynamic and accurate. This allows for the subsequent determination of the pilot transmission interval based on the channel correlation time, making the pilot transmission interval dynamic as well, avoiding the traditional use of a fixed pilot transmission interval.

[0029] The preset channel theory model includes a direct path channel gain model and a non-direct path channel gain model. Substituting the first installation information and the second installation information into the preset channel theory model, the channel theory value at the current moment is obtained. The channel theory value reflects the channel value under ideal conditions.

[0030] Pilot signals provide a reference for the receiver to estimate channel states. Their transmission interval and power must be precisely matched to dynamic channel changes and system power limitations. The transmitter sends pilot optical signals at precise pilot transmission intervals and appropriate pilot power. The transmitter's modulation unit generates an electrical signal carrying the pilot signals according to the pilot transmission interval and the first pilot power. The transmitter's LED driver circuit adjusts the drive current of the LED array based on the electrical signal carrying the pilot signals, converting the electrical signal into an optical signal. Combined with the current beam weight, the pilot optical signal energy is focused towards the UAV's receiving direction, further improving the SNR of the pilot signal at the receiver and ensuring estimation accuracy. By dynamically determining the pilot transmission frequency (pilot transmission interval) and energy intensity (first pilot power), the receiver can efficiently extract pilot signals for channel estimation while avoiding resource waste.

[0031] The first channel prediction model is a CNN deep residual network, with a network structure of input layer → 3 convolutional layers → 5 residual blocks → fully connected layer → output layer. The second pilot power of the light carrying the pilot signal received by the detector is obtained. The second pilot power, the first pilot power, and the flight state data are input into the trained first channel prediction model to obtain the first channel prediction value at the current time.

[0032] Calculate the difference between the first channel prediction value and the theoretical channel value at the current moment. If the difference between the first channel prediction value and the theoretical channel value at the current moment is less than or equal to a preset difference, it indicates that the prediction is relatively accurate. Obtain the actual channel value at the current moment. Based on the error between the first channel prediction value and the actual channel value at the current moment, determine the beamforming mode of the transmitter, such as perfect CSI beamforming mode and imperfect CSI beamforming mode. Different beamforming modes use different methods to determine the beam weights. The transmitter adjusts the emission power based on the beam weights.

[0033] If the difference between the current channel prediction value and the theoretical channel value is greater than the preset difference, it indicates that the prediction is inaccurate. The pilot transmission interval and the power of the first pilot are adjusted until the difference between the current channel prediction value and the theoretical channel value is less than or equal to the preset difference.

[0034] This application provides a deep learning-based LiFi communication control method. Compared with existing technologies, it calculates the channel correlation time and the theoretical channel value at the current moment based on the UAV's flight status data, transmitter installation information, and detector installation information. Based on the channel correlation time, it determines the pilot transmission interval time and calculates the first pilot power based on the transmitter's total transmission power. The transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power. It obtains the second pilot power of the light carrying the pilot signal received by the detector. The second pilot power, the first pilot power, and the flight status data are used as input data for the trained first channel prediction model to obtain the first channel prediction value at the current moment. Based on the first channel prediction value, the theoretical channel value, and the actual channel value, it determines the transmitter's beamforming mode. Based on the beamforming mode, it determines the beam weights, captures the time-varying characteristics of the LiFi channel, and determines the beam weights according to the time-varying characteristics. Even under imperfect beamforming modes, it can determine appropriate beam weights, thus improving communication quality.

[0035] In one embodiment, flight status data includes velocity and angular velocity; first installation information includes first installation coordinates and a first installation surface normal vector; second installation information includes second installation coordinates and a second installation surface normal vector; and channel correlation time is calculated based on the flight status data, the first installation information, and the second installation information, including: The linear velocity of the detector is calculated based on the velocity, angular velocity, and second installation coordinates. The optical link unit vector is calculated based on the first and second installation coordinates; The relative linear velocity is calculated based on the detector's linear velocity and the optical link's unit vector. Based on the relative linear velocity, the channel correlation time is then calculated.

[0036] Specifically, real-time data collection of the UAV's flight status, such as the UAV velocity vector v_UAV=[v_x, v_y, v_z] and the UAV angular velocity ω= [ω_x, ω_y, ω_z].

[0037] Obtain the installation information of the transmitter and receiver, such as the mounting coordinates of the transmitter (r_LED = [x_L, y_L, z_L]), the mounting coordinates of the receiver (r_PD = [x_P, y_P, z_P]), the normal vector of the transmitter's mounting surface (n_LED), and the normal vector of the receiver's mounting surface (n_PD).

[0038] Substitute the velocity, angular velocity, and receiver mounting coordinates into the following formula to calculate the receiver's linear velocity: v_PD = v_UAV + ω × r_PD Substitute the installation coordinates of the transmitter and receiver into the following formula to calculate the optical link unit vector: d=(r_PD r_LED) / ‖r_PD r_LED‖, where ‖r_PD r_LED‖ represents the modulus (norm) of the displacement vector, which is the straight-line distance between the transmitter and the receiver.

[0039] The relative linear velocity is obtained by projecting the linear velocity: v_r = ‖v_PD d‖( (For the dot product, take the absolute value) The channel correlation time is calculated by substituting the relative linear velocity into the following formula: T_c = λ / (4π v_r) (light wavelength λ=850 nm).

[0040] Traditional LiFi communication systems often employ a fixed velocity assumption when calculating channel correlation time, neglecting the impact of real-time velocity fluctuations and attitude angle changes on relative motion of the mobile terminal. This leads to significant deviations between the calculated results and actual channel variations. This application utilizes real-time flight speed and angular velocity, making channel correlation time calculation dynamic and accurate. This provides real-time and precise timing data for pilot design and channel tracking, ensuring that the mobile LiFi communication system maintains high channel estimation accuracy, low resource consumption, and robust communication performance even in high-speed, dynamic terminal motion scenarios. This application can accurately predict the channel's stable time window, providing a theoretical basis for subsequent channel estimation and tracking, and significantly improving the system's adaptability to dynamic environments.

[0041] In one embodiment, a pre-defined channel theory model is constructed using the following method: Based on the first installation coordinates and the second installation coordinates, the straight-line distance between the transmitter and the detector is calculated. Based on the optical link unit vector, the first installation surface normal vector, and the second installation surface normal vector, the transmission angle and the reception angle are calculated. Based on the transmission angle or the reception angle, the blocking coefficient is determined. A direct path channel gain model is constructed based on the blocking coefficient, the straight-line distance between the transmitter and the detector, the angle of transmission, and the angle of reception. Obtain the coordinates of the virtual image point formed by the transmitter on the reference reflective surface, and calculate the first reflection path and the second reflection path based on the first installation coordinates and the second installation coordinates; Based on the second installation coordinates, the coordinates of the virtual image point, and the normal vector of the second installation surface, the angle of reference incident angle is calculated. Based on the reflection coefficient of the reference reflecting surface, the angle of emission angle, the angle of reference incident angle, the first reflection path, and the second reflection path, a channel gain model for the non-direct path is constructed. The channel gain model of the direct path and the channel gain model of the indirect path are added together to form the channel theory model.

[0042] Specifically, the channel theory models include the direct path channel gain model h_LOS(t) and the non-direct path channel gain model h_LOS(t).

[0043] The direct path channel gain model h_LOS(t) is constructed using the following method: First, the direct distance between the transmitter and receiver is calculated by substituting the installation coordinates of the transmitter and receiver into the following formula: R(t) = ‖r_PD(t) r_LED(t)‖.

[0044] Substituting the optical link unit vector and the transmitter's mounting surface normal vector into the following formula, the emission angle is calculated: cos _T = ( d) × n_LED; Substituting the optical link unit vector and the receiver mounting surface normal vector into the following formula, the receiving angle is calculated: cos _R = d × n_PD; If the angle of launch _T <cos -1 (0.5) or the angle of reception _R <cos -1 If (0.5), then it is considered complete occlusion, and the occlusion coefficient A_obs = 0; otherwise, the occlusion coefficient A_obs = 1. Substituting the blocking coefficient, the straight-line distance between the transmitter and the detector, the transmission angle, and the reception angle into the following formula, a direct path channel gain model is constructed: h_LOS(t) = A_obs×(m+1) / (2π)×cos^m _T×cos _R / R²(t) (m is the Lambert order, m=1).

[0045] The non-direct path channel gain model h_NLOS(t) is constructed using the following method: Obtain the reference reflector height z_ref (a constant value, calibrated on-site). For example, if the reflector is the ground / ceiling, obtain the coordinates of the virtual image point formed by the transmitter on the reference reflector surface r_img = [x_LED, y_LED, 2z_ref]. z_LED].

[0046] Substitute the coordinates of the virtual image point and the installation coordinates of the receiver into the following formula to calculate the length of the first reflection path R_1 =‖r_PD The second reflection path length R_2 = ‖r_img‖ is calculated by substituting the coordinates of the virtual image point and the transmitter's installation coordinates into the following formula. r_LED‖.

[0047] Obtain the reflection coefficient ρ of the reference reflective surface (e.g., 0.2 for concrete floors and 0.8 for metal ceilings).

[0048] Substituting the receiver's mounting coordinates, the coordinates of the virtual image point, and the receiver's mounting surface normal vector into the following formula, the reference incident angle is calculated: cos _R_ref = (r_PD r_img) / ‖r_PD r_img‖×n_PD.

[0049] Based on the reflection coefficient of the reference reflector, the angle of emission, the angle of reference incident, the first reflection path, and the second reflection path, a channel gain model for the non-direct path is constructed: h_NLOS(t) =ρ×(m+1) / (2π)×cos^m _T×cos _R_ref / (R_1×R_2)² Adding the direct path channel gain model and the non-direct path channel gain model together, we get h(t) = h_LOS(t) + h_NLOS(t).

[0050] The channel theoretical gain h(t) is decomposed into a direct component hLOS(t) and a non-direct component hNLOS(t), which accurately matches the actual propagation scenario of mobile LiFi and is more in line with the physical essence, resulting in a more accurate channel theoretical value.

[0051] In one embodiment, the pilot transmission interval is determined based on the channel correlation time, the total transmit power of the transmitter is obtained, and the first pilot power is calculated based on the total transmit power, including: Use half of the channel correlation time as the pilot transmission interval; The product of the total transmitted power and the power percentage is used as the first pilot power.

[0052] Specifically, based on the current channel signal-to-noise ratio (SNR), a power allocation α is set, and the first pilot power Ppilot = αPtotal is calculated, where Ptotal is the total transmit power. α is set to 0.3~0.5 when the SNR is low, and 0.1~0.2 when the SNR is high. Using the product of the total power and the power allocation as the pilot power allows for dynamic adjustment of the power allocation, enabling the pilot power to change with channel quality (such as SNR and attenuation), ensuring reliable reception of the pilot signal while avoiding power waste.

[0053] Based on the channel correlation time Tc, the pilot transmission interval Tpilot is set, where Tpilot = Tc / 2. Half of the channel correlation time is used as the pilot transmission interval. High-frequency, non-redundant pilot transmission ensures that the receiver can track channel changes in real time, while avoiding excessive time consumption by the pilots.

[0054] In one embodiment, such as Figure 2 As shown, based on the error between the first channel prediction value and the actual channel value at the current moment, the beamforming mode of the transmitter is determined, and based on the beamforming mode, the beam weights are determined, including: 202: If the error between the first channel prediction value and the actual channel value at the current moment is less than or equal to the preset difference, then the perfect CSI beamforming mode is adopted; 204: The beam weights in the perfect CSI beamforming mode are calculated based on the first channel prediction value; 206: If the error between the first channel prediction value and the actual channel value at the current moment is greater than the preset difference, then imperfect CSI beamforming is adopted; 208: Based on the first channel prediction value and error, an optimization model is constructed. The optimization model is solved by semidefinite programming to obtain the beam weights under the imperfect CSI beamforming mode.

[0055] Specifically, calculate the error ΔH = H between the first predicted channel value and the actual channel value at the current moment. H^ (H is the actual value of the channel, H^ is the predicted value of the first channel). If the absolute value of ΔH is less than the preset difference, it is determined to be a perfect CSI beamforming mode; otherwise, it is an imperfect CSI beamforming mode.

[0056] In perfect CSI beamforming mode, the beam weight w is calculated using the maximum ratio combining criterion. / (||H^||²) ( (This is the conjugate transpose of H^), which focuses the optical signal energy in the direction of the receiver, thereby improving the signal-to-noise ratio.

[0057] Under imperfect CSI beamforming mode, a robust optimization method is used to construct an optimization model: The objective function is min max ||w||² The constraint condition is |w^H( +ΔH)|² ≥ γ (γ is the minimum received power threshold).

[0058] The optimal beam weight w is obtained by solving the optimization model using the semidefinite programming (SDP) algorithm, ensuring that the received power still meets the communication requirements even in the presence of CSI error.

[0059] The calculated beam weights are transmitted to the transmitter's drive circuit to adjust the luminous intensity of each LED in the LED array. By adjusting the energy distribution of the optical signal, the link attenuation and occlusion caused by movement are offset, providing a stable optical channel for data transmission.

[0060] This application presents a robust beamforming mode that adapts to both perfect and imperfect CSI conditions. Through robust optimization theory, it can still guarantee communication quality even when there are errors in channel estimation, thus significantly enhancing the system's robustness.

[0061] In one embodiment, the deep learning-based LiFi communication control method further includes: Obtain the channel state information for the current time and the preset time period before the current time, input the channel state information into the trained second channel prediction model, and obtain the second channel prediction value for the next time.

[0062] Specifically, the channel state information of the current time and the preset number of times before the current time is obtained, such as the channel state information (CSI) data (including H^, obstruction status, and SNR) of the current time and the past T times (e.g., T=10, each time corresponding to one pilot transmission interval), and input into the second channel prediction model; The second channel prediction model learns the channel timing variation patterns through two LSTM layers to predict the channel gain matrix H^pred at the next moment. Each time a new pilot signal is received, the historical CSI data window is updated, and the prediction is re-executed to ensure real-time tracking. By capturing channel timing patterns, prediction accuracy is improved, providing a time lead for parameter adjustments. This optimizes resource utilization while enhancing anti-interference capabilities, perfectly adapting to the dynamic channel changes in mobile scenarios. It achieves accurate prediction of future channel states, effectively compensating for performance losses caused by CSI expiration and improving communication continuity.

[0063] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides a LiFi communication control device based on deep learning, such as... Figure 3 As shown, the device includes: The channel feature analysis module 302 is used to acquire the flight status data of the UAV, the first installation information of the transmitter and the second installation information of the detector at the current moment, calculate the channel correlation time based on the flight status data, the first installation information and the second installation information, and calculate the channel theoretical value at the current moment based on the first installation information, the second installation information and the preset channel theoretical model. The pilot optimization module 304 is used to determine the pilot transmission interval time based on the channel correlation time, obtain the total transmission power of the transmitter, and calculate the first pilot power based on the total transmission power, so that the transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power; The channel prediction module 306 is used to obtain the second pilot power of the light carrying the pilot signal received by the detector, and input the second pilot power, the first pilot power and the flight state data into the trained first channel prediction model to obtain the first channel prediction value at the current time. The beam weight adjustment module is used to obtain the actual channel value at the current moment if the difference between the first channel prediction value and the theoretical channel value at the current moment is less than or equal to a preset difference 308. Based on the error between the first channel prediction value and the actual channel value at the current moment, the beamforming mode of the transmitter is determined. Based on the beamforming mode, the beam weight is determined so that the transmitter can adjust the emission power based on the beam weight.

[0064] This application provides a deep learning-based LiFi communication control device. Compared with existing technologies, it calculates the channel correlation time and the theoretical channel value at the current moment based on the UAV's flight status data, transmitter installation information, and detector installation information. Based on the channel correlation time, it determines the pilot transmission interval time and calculates the first pilot power based on the transmitter's total transmission power. The transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power. It obtains the second pilot power of the light carrying the pilot signal received by the detector. The second pilot power, the first pilot power, and the flight status data are used as input data for the trained first channel prediction model to obtain the first channel prediction value at the current moment. Based on the first channel prediction value, the theoretical channel value, and the actual channel value, it determines the transmitter's beamforming mode. Based on the beamforming mode, it determines the beam weights, captures the time-varying characteristics of the LiFi channel, and determines the beam weights according to the time-varying characteristics. Even under imperfect beamforming modes, it can determine appropriate beam weights, thus improving communication quality.

[0065] In one embodiment, the flight status data includes velocity and angular velocity; the first installation information includes first installation coordinates and a first installation surface normal vector; the second installation information includes second installation coordinates and a second installation surface normal vector; and the channel feature analysis module is further used for: The linear velocity of the detector is calculated based on the velocity, angular velocity, and second installation coordinates. The optical link unit vector is calculated based on the first and second installation coordinates; The relative linear velocity is calculated based on the detector's linear velocity and the optical link's unit vector. Based on the relative linear velocity, the channel correlation time is then calculated.

[0066] In one embodiment, the channel feature analysis module is also used for: Based on the first installation coordinates and the second installation coordinates, the straight-line distance between the transmitter and the detector is calculated. Based on the optical link unit vector, the first installation surface normal vector, and the second installation surface normal vector, the transmission angle and the reception angle are calculated. Based on the transmission angle or the reception angle, the blocking coefficient is determined. A direct path channel gain model is constructed based on the blocking coefficient, the straight-line distance between the transmitter and the detector, the angle of transmission, and the angle of reception. Obtain the coordinates of the virtual image point formed by the transmitter on the reference reflective surface, and calculate the first reflection path and the second reflection path based on the first installation coordinates and the second installation coordinates; Based on the second installation coordinates, the coordinates of the virtual image point, and the normal vector of the second installation surface, the angle of reference incident angle is calculated. Based on the reflection coefficient of the reference reflecting surface, the angle of emission angle, the angle of reference incident angle, the first reflection path, and the second reflection path, a channel gain model for the non-direct path is constructed. The channel gain model of the direct path and the channel gain model of the indirect path are added together to form the channel theory model.

[0067] In one embodiment, the pilot optimization module is further configured to: Use half of the channel correlation time as the pilot transmission interval; The product of the total transmitted power and the power percentage is used as the first pilot power.

[0068] In one embodiment, the beam weighting module is further configured to: If the error between the first channel prediction value and the actual channel value at the current moment is less than or equal to the preset difference, then the perfect CSI beamforming mode is adopted, and the beam weight in the perfect CSI beamforming mode is calculated based on the first channel prediction value. If the error between the predicted value of the first channel and the actual value of the channel at the current moment is greater than the preset difference, then imperfect CSI beamforming is adopted. An optimization model is constructed based on the predicted value of the first channel and the error. The optimization model is solved by semidefinite programming to obtain the beam weights under the imperfect CSI beamforming mode.

[0069] In one embodiment, the pilot optimization module is further configured to: If the difference between the current channel prediction value and the theoretical channel value is greater than a preset difference, the pilot transmission interval and the power of the first pilot are adjusted until the difference between the current channel prediction value and the theoretical channel value is less than or equal to the preset difference.

[0070] In one embodiment, the channel prediction module is further configured to: Obtain the channel state information for the current time and the preset time period before the current time, input the channel state information into the trained second channel prediction model, and obtain the second channel prediction value for the next time.

[0071] According to one embodiment of the present invention, a storage medium is provided, the storage medium storing at least one executable instruction, which can execute the deep learning-based LiFi communication control method in any of the above method embodiments.

[0072] Figure 4The diagram illustrates a structural schematic of a computer device according to an embodiment of the present invention. The specific embodiments of the present invention do not limit the specific implementation of the computer device.

[0073] like Figure 4 As shown, the computer device may include: a processor 402, a communications interface 404, a memory 406, and a communications bus 408.

[0074] The processor 402, communication interface 404, and memory 406 communicate with each other via communication bus 408.

[0075] Communication interface 404 is used to communicate with other network elements such as clients or other servers.

[0076] The processor 402 is used to execute program 410, specifically to perform the relevant steps in the above-described embodiment of the deep learning-based LiFi communication control method.

[0077] Specifically, program 410 may include program code that includes computer operation instructions.

[0078] Processor 402 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The computer device includes one or more processors, which may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0079] Memory 406 is used to store program 410. Memory 406 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0080] Specifically, program 410 can be used to cause processor 402 to perform the following operations: The system acquires the flight status data of the UAV, the first installation information of the transmitter, and the second installation information of the detector at the current moment. Based on the flight status data, the first installation information, and the second installation information, it calculates the channel correlation time. Based on the first installation information, the second installation information, and the preset channel theory model, it calculates the channel theory value at the current moment. Based on the channel correlation time, the pilot transmission interval is determined, the total transmission power of the transmitter is obtained, and the first pilot power is calculated based on the total transmission power, so that the transmitter transmits light carrying the pilot signal according to the pilot transmission interval and the first pilot power; The second pilot power of the light carrying the pilot signal received by the detector is obtained. The second pilot power, the first pilot power, and the flight state data are input into the trained first channel prediction model to obtain the first channel prediction value at the current time. If the difference between the first channel prediction value and the theoretical channel value at the current moment is less than or equal to a preset difference, the actual channel value at the current moment is obtained. Based on the error between the first channel prediction value and the actual channel value at the current moment, the beamforming mode of the transmitter is determined. Based on the beamforming mode, the beam weight is determined so that the transmitter can adjust the emission power based on the beam weight.

[0081] It will be apparent to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. In one embodiment, they can be implemented using device-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0082] The above embodiments are merely exemplary embodiments of this application and are not intended to limit this application. The scope of protection of this application is defined by the claims. Those skilled in the art can make various modifications or equivalent substitutions to this application within its substance and scope of protection, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of this application.

Claims

1. A LiFi communication control method based on deep learning, characterized in that, include: The flight status data of the UAV, the first installation information of the transmitter, and the second installation information of the detector at the current moment are obtained. Based on the flight status data, the first installation information, and the second installation information, the channel correlation time is calculated. Based on the first installation information, the second installation information, and the preset channel theory model, the channel theory value at the current moment is calculated. Based on the channel correlation time, the pilot transmission interval is determined, the total transmission power of the transmitter is obtained, and the first pilot power is calculated based on the total transmission power, so that the transmitter transmits light carrying pilot signals according to the pilot transmission interval and the first pilot power; The second pilot power of the light carrying the pilot signal received by the detector is obtained, and the second pilot power, the first pilot power, and the flight state data are input into the trained first channel prediction model to obtain the first channel prediction value at the current time. If the difference between the first channel prediction value and the theoretical channel value at the current moment is less than or equal to a preset difference, the actual channel value at the current moment is obtained. Based on the error between the first channel prediction value and the actual channel value at the current moment, the beamforming mode of the transmitter is determined. Based on the beamforming mode, the beam weight is determined so that the transmitter can adjust the emission power based on the beam weight.

2. The deep learning-based LiFi communication control method as described in claim 1, characterized in that, The flight status data includes velocity and angular velocity; the first installation information includes first installation coordinates and a first installation surface normal vector; the second installation information includes second installation coordinates and a second installation surface normal vector; the calculation of channel correlation time based on the flight status data, the first installation information, and the second installation information includes: Based on the velocity, angular velocity, and second installation coordinates, the linear velocity of the detector is calculated; Based on the first installation coordinates and the second installation coordinates, the optical link unit vector is calculated; The relative linear velocity is calculated based on the linear velocity of the detector and the unit vector of the optical link, and the channel correlation time is calculated based on the relative linear velocity.

3. The deep learning-based LiFi communication control method as described in claim 2, characterized in that, The following methods are used to construct a pre-defined channel theory model, including: Based on the first installation coordinates and the second installation coordinates, the straight-line distance between the transmitter and the detector is calculated. Based on the optical link unit vector, the first mounting surface normal vector and the second mounting surface normal vector, the transmission angle and the reception angle are calculated. Based on the transmission angle or the reception angle, the blocking coefficient is determined. Based on the blocking coefficient, the straight-line distance between the transmitter and the detector, the angle of the transmission angle and the angle of the reception angle, a direct path channel gain model is constructed. The coordinates of the virtual image point formed by the transmitter on the reference reflective surface are obtained, and the first reflection path and the second reflection path are calculated based on the first installation coordinates and the second installation coordinates. Based on the second installation coordinates, the coordinates of the virtual image point, and the normal vector of the second installation surface, the angle of the reference incident angle is calculated. Based on the reflection coefficient of the reference reflecting surface, the angle of the emission angle, the angle of the reference incident angle, the first reflection path, and the second reflection path, a non-direct path channel gain model is constructed. The channel gain model of the direct path and the channel gain model of the indirect path are added together to form the channel theory model.

4. The deep learning-based LiFi communication control method as described in claim 2, characterized in that, The process of determining the pilot transmission interval based on the channel correlation time, obtaining the total transmission power of the transmitter, and calculating the first pilot power based on the total transmission power includes: Half of the channel correlation time is used as the pilot transmission interval time; The product of the total transmitted power and the power percentage is taken as the first pilot power.

5. The deep learning-based LiFi communication control method as described in claim 1, characterized in that, The process of determining the transmitter's beamforming mode based on the error between the first channel prediction value and the actual channel value at the current moment, and determining beam weights based on the beamforming mode, includes: If the error between the first channel prediction value and the actual channel value at the current time is less than or equal to a preset difference, then the perfect CSI beamforming mode is adopted, and the beam weight in the perfect CSI beamforming mode is calculated based on the first channel prediction value. If the error between the first channel prediction value and the actual channel value at the current time is greater than a preset difference, then imperfect CSI beamforming is adopted. An optimization model is constructed based on the first channel prediction value and the error. The optimization model is solved by semidefinite programming to obtain the beam weights under the imperfect CSI beamforming mode.

6. The deep learning-based LiFi communication control method according to any one of claims 1-5, characterized in that, The deep learning-based LiFi communication control method also includes: If the difference between the current channel prediction value and the theoretical channel value is greater than the preset difference, then the pilot transmission interval and the first pilot power are adjusted until the difference between the current channel prediction value and the theoretical channel value is less than or equal to the preset difference.

7. The deep learning-based LiFi communication control method according to any one of claims 1-5, characterized in that, The deep learning-based LiFi communication control method also includes: Obtain the channel state information of the current time and the preset time period before the current time, input the channel state information into the trained second channel prediction model, and obtain the second channel prediction value of the next time.

8. A LiFi communication control device based on deep learning, characterized in that, include: The channel feature analysis module is used to acquire the flight status data of the UAV, the first installation information of the transmitter, and the second installation information of the detector at the current moment. Based on the flight status data, the first installation information, and the second installation information, it calculates the channel correlation time. Based on the first installation information, the second installation information, and the preset channel theory model, it calculates the channel theory value at the current moment. The pilot optimization module is used to determine the pilot transmission interval time based on the channel correlation time, obtain the total transmission power of the transmitter, and calculate the first pilot power based on the total transmission power, so that the transmitter transmits light carrying the pilot signal according to the pilot transmission interval time and the first pilot power; The channel prediction module is used to obtain the second pilot power of the light carrying the pilot signal received by the detector, and input the second pilot power, the first pilot power and the flight state data into the trained first channel prediction model to obtain the first channel prediction value at the current time. The beam weight adjustment module is used to obtain the actual channel value at the current time if the difference between the first channel prediction value and the theoretical channel value at the current time is less than or equal to a preset difference; determine the beamforming mode of the transmitter based on the error between the first channel prediction value and the actual channel value at the current time; and determine the beam weight based on the beamforming mode so that the transmitter can adjust the emission power based on the beam weight.

9. A storage medium storing at least one executable instruction that causes a processor to perform an operation corresponding to the deep learning-based LiFi communication control method as described in any one of claims 1-7.

10. A computer device, comprising: The processor, memory, communication interface, and communication bus are provided, wherein the processor, memory, and communication interface communicate with each other via the communication bus. The memory is used to store at least one executable instruction, which causes the processor to perform the operation corresponding to the deep learning-based LiFi communication control method as described in any one of claims 1-7.

Images