Network-free region video stream transmission method based on air-space cooperation

Abstract

The invention discloses a video stream transmission method for a network-free region based on air-space cooperation, and mainly solves the problem that videos are difficult to collect and transmit in a region without base station coverage or a region with a damaged ground network. Comprising the following steps: 1) constructing an air-space integrated communication system consisting of a satellite, an unmanned aerial vehicle and a user; 2) the user side predicts the demand according to the historical data and uploads the demand to the satellite side; 3) the satellite end performs intention recognition and matches a corresponding DQN reinforcement learning template; 4) constructing a user quality of experience QOE model according to the transmission time delay and the state of the video buffer area; 5) the satellite end adopts the DQN template to adjust a video stream transmission strategy in real time according to the QOE model, and transmits the video blocks to the user end; and 6) the user side receives the video blocks and plays the video blocks in real time. According to the method, subjective preference requirements and objective video requirements of the user are predicted and recognized, a resource allocation strategy is dynamically formulated by adopting a reinforcement learning algorithm, communication resources in the system are fully utilized, consumption of the communication resources is reduced, and the method can be used for video stream transmission service in an air-space integrated communication system.

Description

Technical Field

[0001] The present invention belongs to the technical fields of multimedia transmission and streaming media technology, and further relates to video stream service technology. Specifically, it is a method for transmitting video streams in network-free areas based on space-air collaboration, which can be used for on-demand services in space-air integrated video stream service application scenarios. Background Art

[0002] The space-air integrated network of 6G networks integrates air networks and satellite networks, enabling video stream services to seamlessly cover a wider area. Video stream services will no longer be restricted by geographical locations, and users can enjoy high-quality video content at any time and any place. However, there are the following problems in the process of users enjoying video stream services: 1) Lack of consideration for user needs: The vision of 6G networks is to achieve "digital twin, intelligent omnipresence", which means that the network will be able to understand user intentions and provide various services on demand according to user needs; 2) Insufficient personalized video stream services: It is unable to accurately capture users' interest preferences, affecting the user experience and reducing user satisfaction; 3) Unreasonable network resource allocation: The result of network resource allocation directly affects the user experience in the process of video stream services, and problems such as stuttering and frame dropping may occur. Therefore, how to identify user needs in advance and intelligently adjust resources in the network according to user needs to provide on-demand personalized video stream services has become a challenge.

[0003] Nanjing University of Aeronautics and Astronautics proposed a method for large-scale multi-party interactive real-time video stream transmission in its patent document "A Method for Large-scale Multi-party Interactive Real-time Video Stream Transmission" (Patent Application No.: CN202411373135.2, Publication No.: CN119232983A). This method first establishes a global QOE model; then constructs a two-layer Stackelberg game model. The first-layer Stackelberg game is used for the uplink bandwidth allocation of the sender. Regarding the uplink bandwidth allocation as a bandwidth trading problem, each sender decides the amount of bandwidth to purchase to ensure high-quality video streams and smooth interactions. The second-layer Stackelberg game is used for the downlink bandwidth allocation and bitrate selection of the receiver. The sender selects its own downlink bandwidth and the bitrate of each received video stream according to different viewing preferences; the two-layer Stackelberg game model combines the joint upload-download bandwidth allocation with adaptive bitrate control to manage the complexity of large-scale scenarios and optimize the utility function of the receiver's QoE by maximizing the utility of the three parties. By introducing a new transmission scheme based on a two-layer Stackelberg game and an optimized embedding strategy, this method achieves efficient bandwidth allocation and adaptive bitrate control for large-scale multi-party interactive real-time video streams, significantly improving the overall service performance and user experience. However, there are still the following three deficiencies: First, since this method only considers the real-time interaction of video streams and does not consider the resource changes caused by environmental changes during the video stream transmission, it is easy to fail to provide accurate on-demand services for users; Second, since this method only considers the user's selection of the video stream bitrate but does not consider the user's preferences for video stream resolution, frame rate, latency, and smoothness parameters, it is easy to ignore the objective and subjective personalized needs of users for video streams; Third, since this method only considers the user's delay tolerance and does not consider the video cache capacity of the user device itself, it is easy to cause stuttering or waste of resources during the user's video viewing experience. Summary of the Invention

[0004] The object of the present invention is to address the deficiencies of the above-mentioned existing technologies and propose a method for transmitting video streams in network-free areas based on air-space collaboration, which is used to solve the problem of difficult video stream acquisition and transmission in remote areas without base station coverage or areas where the ground network is damaged due to disasters or other reasons. This method integrates the air network and the space-based network through the 6G air-space integrated network to provide on-demand video stream services for users. First, an air-space integrated communication system composed of satellites, unmanned aerial vehicles (UAVs), and users is constructed; then, the historical data of users' requests for video stream services is used to predict users' needs; then, intention recognition is performed according to users' needs and the corresponding DQN reinforcement learning template is matched; then, a quality of experience (QOE) model for users is constructed based on the transmission delay and the status of the video buffer; finally, the DQN template is used to adjust the transmission strategy of the video stream in real time according to the QOE model of the user, and the user module receives the video blocks and plays them in real time to complete the transmission process. The present invention can meet the service needs of users under limited communication resources, provide on-demand video stream transmission services for users, and effectively improve resource utilization rate.

[0005] To achieve the above object of the invention, the technical solutions adopted by the present invention include the following steps:

[0006] (1) Construct an air-space integrated communication system, including a UAV module, a satellite module, and a user module; the satellite module includes a communication ground station and a satellite, and the satellite contains a pre-trained deep Q-network (DQN) decision template library; wherein the UAV module is used to sense real-time scenes and generate video streams, and at the same time transmit video stream data packets to the satellite module; the satellite module processes them and transmits the results to the user module; the user module is implemented through a head-mounted device.

[0007] (2) According to the historical data sequence of users' requests for video stream services, the user module uses the head-mounted device to predict users' needs and synchronizes the prediction results to the satellite module.

[0008] (3) The satellite module performs templated intention recognition according to the prediction results synchronized by the head-mounted device, translates users' needs into a parameter sequence composed of resolution, frame rate, delay preference, and smoothness preference; and matches the pre-trained DQN template in the DQN decision template library according to this parameter sequence.

[0009] (4) Construct a quality of experience (QOE) model for users:

[0010]

[0011] where t h and s h are the delay preference and smoothness preference of the user respectively, Δt represents the transmission delay of the t-th video block, and N {·}The function means that if the condition of {·} is satisfied, the output is 1, otherwise the output is 0; e1 and e2 are preset thresholds to prevent buffer underflow and overflow; B max is the upper limit value of the video buffer, B t represents the video buffer status when the t-th video block transmission is completed;

[0012] (5) The DQN template is adopted to adjust the video stream transmission strategy in real time according to the user's QOE model. The satellite module transmits the video block to the user module according to the decision of the DQN template, that is, the adjusted transmission strategy;

[0013] (6) The user module puts the received video block into the video buffer, waits to read the RTP data packet from the buffer, and performs unpacking, verification, and error correction operations. The error correction operation is performed when the unpacking and verification in the user module are incorrect; after the verification is correct, the video block is decoded and played in real time on the display device.

[0014] Compared with the prior art, the present invention has the following advantages:

[0015] First, since the present invention accurately predicts user needs, identifies user intentions according to the recorded information of the user-requested video stream history, and translates the user need intention into a parameter sequence to provide data support for the user to provide video stream services. At the same time, a more refined DQN template library is provided, so that a more accurate DQN template can be matched, and its decision-making accuracy is more in line with the actual environmental needs. Therefore, it can meet the user's service needs in the case of limited communication resources and provide an on-demand video stream transmission service for the user.

[0016] Second, the present invention can accurately predict the personalized needs of the user according to the video stream information and preference information of the user-requested video stream history, and select the corresponding video stream and transmission method according to the user's needs to provide personalized on-demand video stream services for the user, improving the user's viewing experience and enhancing the user's participation.

[0017] Third, since the present invention adopts the deep Q-network (DQN) algorithm of reinforcement learning and deploys the pre-trained DQN algorithm model in the satellite module; the algorithm uses the real-time transmission rate of the path and the real-time buffer status of the user as the decision basis of the DQN agent to formulate the satellite routing and bandwidth allocation strategy of the video stream. Finally, a user service satisfaction function is established based on the video transmission delay and the buffer size of the head-mounted device, and this is used as the reward function to instruct the DQN agent to continuously optimize and improve its action strategy, thereby significantly improving the resource utilization rate and enhancing the transmission efficiency and user experience of the video stream. Brief Description of the Drawings

[0018] Figure 1 is the overall implementation flowchart of the method of the present invention;

[0019] Figure 2 This is the architecture diagram of the aerospace network system of the present invention;

[0020] Figure 3 This is the comparison diagram of the average bandwidth simulation results between the method of the present invention and the existing solutions. Specific embodiments

[0021] The following further describes the present invention in detail with reference to the accompanying drawings and embodiments.

[0022] Example 1: Referring to the attached Figure 1 A method for video stream transmission in network-free areas based on aerospace collaboration proposed by the present invention specifically includes the following steps:

[0023] Step 1) Construct an aerospace integrated communication system, including a drone module, a satellite module, and a user module; the satellite module includes a communication ground station and a satellite, and the satellite contains a pre-trained deep Q-network DQN decision template library; the drone module is used to sense real-time scenes and generate video streams, and at the same time transmit video stream data packets to the satellite module; the satellite module processes it and transmits the result to the user module; the user module is implemented through a head-mounted device.

[0024] In the aerospace integrated communication system constructed in this embodiment, the above drone module sends video stream data packets to the satellite module, which encodes and stores the original video stream data in different resolutions and frame rates, and divides the encoded video stream into video blocks of a fixed size, encapsulates and packs them into RTP data packets and sends them to the satellite module; the communication ground station in the above satellite module receives the RTP data packets from the drone module and forwards them to the satellite, and the satellite stores the received RTP data packets; the satellite contains a pre-trained deep Q-network DQN decision template library, and the template is used to adjust the transmission strategy according to the transmission environment; when the satellite receives the user's intention, it recognizes and translates the user's intention, and matches the corresponding RTP data packets of the resolution and frame rate format, that is, video stream data packets, and the corresponding DQN decision template according to the translation result. The above video stream data packets are transmitted to the user module according to the routing result of the DQN decision template and the formulated bandwidth allocation strategy; the user module predicts the user's demand for watching the video stream through the head-mounted device and sends it to the satellite module, and at the same time receives the video stream data packets transmitted by the satellite and displays and plays them; the buffer in the head-mounted device stores different amounts of video streams according to the buffer size.

[0025] Step 2) According to the historical data sequence of the user's requested video stream service, the user module uses the head-mounted device to predict the user's needs and synchronizes the prediction results to the satellite module. The prediction results in this embodiment include the video stream information and preference information requested by the user at a certain place at a future time; specifically, the user's needs F are predicted according to the historical data sequence H of the user's requested video stream service by using the head-mounted device:

[0026] H = { { T 1 , L 1 , V 1 , U 1},..., { T h , L h , V h , U h},..., { T H , L H , V H , U H}},

[0027] F = H + 1;

[0028] Where T h , L h , V h , U h respectively represent the time, location, video stream information, and user preference information of the user's h-th request for the video stream; V h = { r h , f h}, r h , f h respectively represent the resolution and frame rate of the video stream; U h = { t h , s h}, t h , s h respectively represent the user's preferences for latency and smoothness.

[0029] The above historical data includes the time, location, video stream information, and user preference information of the user's requested video stream, where the video stream information includes the resolution and frame rate of the video stream, and the user preference information includes the user's preferences for latency and smoothness.

[0030] Step 3) The satellite module performs templated intention recognition based on the prediction results synchronized by the head-mounted device, translates the user's requirements into a parameter sequence composed of resolution, frame rate, latency preference, and smoothness preference; and matches the pre-trained DQN template in the DQN decision template library according to this parameter sequence. The above-mentioned intention recognition in this embodiment includes recognizing two parts: video stream-related parameters and user personal preferences; among them, the video stream-related parameters include video resolution parameters and frame rate parameters; the video resolution parameter is used to measure the data volume size of a single-frame video image, and the video frame rate parameter is the number of frames displayed by the head-mounted device per second; the user's personal preferences include personal preferences for latency and smoothness; if the latency preference coefficient is larger, the DQN template tends to low latency and selects a large bandwidth when making a decision; on the contrary, when the smoothness preference coefficient is larger, the DQN template tends to smoothness and selects a bandwidth that is large first and then small to keep the buffer size stable within a preset range.

[0031] Step 4) Build a Quality of Experience (QOE) model for the user:

[0032]

[0033] where t h and s h are the user's latency preference and smoothness preference respectively, Δt represents the transmission latency of the t-th video block, N {·} function means outputting 1 if the condition of {·} holds, otherwise outputting 0; e1 and e2 are preset thresholds to prevent buffer underflow and overflow; B max is the upper limit value of the video buffer, and B t represents the video buffer state when the t-th video block transmission is completed, which is expressed as follows:

[0034]

[0035] where B max is the upper limit value of the video buffer, and B t-1 represents the buffer state when the (t - 1)-th video block transmission is completed; Δc t represents the buffer change value from when the (t - 1)-th video block transmission is completed to when the t-th video block transmission is completed.

[0036] Step 5) The satellite module uses the DQN template to adjust the transmission strategy of the video stream in real time according to the user's QOE model, and transmits the video blocks to the user module according to the decision of the DQN template, that is, the adjusted transmission strategy. In this embodiment, adjusting the transmission strategy of the video stream specifically means: the satellite module calculates the transmission latency and video buffer state in the QOE model, and uses the transmission rate and buffer state as the input of the DQN template to output a transmission decision, that is, the routing selection and bandwidth allocation when each video block is transmitted.

[0037] The above transmission delay is calculated according to the following formula:

[0038]

[0039] where D t represents the data volume of the t-th video block, o t represents the satellite routing delay when transmitting the t-th video block, R t represents the transmission rate when transmitting the t-th video block, and is obtained according to the following formula:

[0040] R t = β t Blog2(1 + SINR t )

[0041] where β t is the bandwidth ratio allocated to the user when transmitting the t-th video block, B represents the total bandwidth between the satellite and the user, and SINR t represents the signal-to-noise ratio between the satellite and the user, and is expressed as:

[0042]

[0043] where P is the transmission power of the satellite, and N0 is the Gaussian white noise power; is the space idle loss; c is the speed of light, with the unit of km / s; f0 is the communication center frequency of the inter-satellite link, with the unit of Hz; d is the slant range; is the rainfall attenuation, that is, the environmental impact parameter, is the equivalent effective path length, η is the rainfall intensity, ρ and λ represent two coefficients related to f0, and can be obtained from the ITU-R P.838 report; the factors affecting this rainfall attenuation include at least frequency, elevation angle, altitude, rainfall intensity, etc. is the satellite receiving antenna gain, is the satellite transmitting gain, and is expressed as follows:

[0044]

[0045] where g max is the maximum antenna gain, ω t is the corresponding off-axis angle relative to the maximum radiation, and ω 3dB is the half-power angle of the transmitting antenna.

[0046] Step 6) The user module puts the received video block into the video buffer, waits to read the RTP data packet from the buffer, and performs unpacking, verification, and error correction operations. The error correction operation is performed when the unpacking and verification by the user module are incorrect; after the verification is correct, the video block is decoded and played in real time on the display device.

[0047] Embodiment 2: The overall implementation steps of the video stream transmission method proposed in this embodiment are the same as those in Embodiment 1. Now, in combination with the attached Figure 1-2 The implementation process of the present invention will be further described in detail:

[0048] Step 1, construct an integrated space-air communication system.

[0049] As Figure 2 shown, the integrated space-air communication system constructed in this step includes a drone module, a satellite module, and a user module. Each module collaborates closely to achieve on-demand services for ubiquitous connected video streams; among them:

[0050] The drone module is used to sense the real-time scene and generate a video stream, encode and store the original video stream data at different resolutions and frame rates, divide the encoded video stream into video blocks of a fixed size, encapsulate and package them into RTP data packets, and send them to the satellite module. Since there is no ground network at the drone's location, space-air ubiquitous connection is required;

[0051] The satellite module includes a satellite communication ground station and a satellite. The satellite communication ground station receives the RTP data packets from the drone module and forwards them to the satellite, and the satellite stores the received RTP data packets. The satellite has a pre-trained DQN decision template library. The templates in this library can adjust the transmission strategy in real time according to the transmission delay and the cache size of the head-mounted device in a complex communication environment. When the satellite receives the user's intention, it identifies and translates the user's intention, matches the corresponding RTP data packets in the resolution and frame rate format and the corresponding DQN decision template according to the translated user's intention, and the RTP data packets will be transmitted to the user module according to the routing result of the DQN decision template and the formulated bandwidth allocation strategy;

[0052] The user module includes a head-mounted device and a user. The head-mounted device is used to predict the user's demand for watching the video stream and send the demand to the satellite module, and is also equipped with a video buffer for temporarily storing data packets. The user module first puts the received video blocks into the video buffer, waits to read the RTP data packets from the buffer, and performs operations such as unpacking, verification, and error correction. After the user module unpacks correctly, it decodes the video blocks and plays them in real time on the display device.

[0053] Step 2, the head-mounted device predicts the demand for the user to request video stream services.

[0054] The head-mounted device predicts the user's demand F according to the historical data sequence H of the user's request for video stream services. The historical data sequence is expressed as H = {{T 1 ,L 1 ,V 1 ,U 1},...,{Th ,L h ,V h ,U h},...,{T H ,L H ,V H ,U H}},where the user requirement F = H + 1, and {T h ,L h ,V h ,U h} represents the time, location, video stream information, and user preference information of the user's h-th request for the video stream. Among them, V h = {r h ,f h} represents the resolution and frame rate of the video stream, and U h = {t h ,s h} represents the user's preferences for latency and smoothness. Based on the above historical data, predict the video stream information and preference requirements that the user will request at a certain location at a future moment. The head-mounted device synchronizes the predicted user data F to the satellite module to provide data support for subsequent intention recognition.

[0055] Step 3: Perform intention recognition according to the user requirement and match the corresponding DQN reinforcement learning template.

[0056] The satellite module performs templated intention recognition based on the user data F synchronized by the head-mounted device. The video stream intention mainly includes two parts: video stream-related parameters and user personal preferences. For video stream-related parameters, consider the video stream resolution parameter and the video stream frame rate parameter; among them, the video resolution parameter is used to measure the data volume size of a single-frame video image; the video frame rate parameter is the number of frames displayed by the head-mounted device per second. For user personal preferences, consider the user's personalized preferences for latency and smoothness; among them, when the user's latency preference coefficient is larger, the DQN template is more biased towards low latency during decision-making, so the action selection will be biased towards large bandwidth; when the user's smoothness preference coefficient is larger, the DQN template is more biased towards smoothness during decision-making. Since the user buffer size is fixed, the bandwidth action selection is first large and then small to keep the buffer size stable within an appropriate range;

[0057] Translate the user requirement intention into a parameter sequence I = {r h ,f h ,t h ,s h} composed of resolution, frame rate, latency preference, and smoothness preference;

[0058] The satellite module matches and selects to transmit a video stream with a resolution of r h and a frame rate of f hThe corresponding video stream, and match the pre-trained latency and smoothness preferences in the decision library to be t h and s h of the DQN model to provide a decision basis for subsequent video stream transmission.

[0059] Step 4: Use the DQN template to adjust the video stream transmission strategy in real time according to the user's QOE model.

[0060] (4.1) The satellite module calculates the transmission latency and video buffer status in the QOE model according to the communication model and the video stream model;

[0061] (4.1.1) Communication model:

[0062] According to the Shannon formula, the data transmission rate between the satellite and the user is calculated as follows:

[0063] R t = β t Blog2(1 + SINR t )

[0064] where β t is the bandwidth ratio allocated to the user during the transmission of the t-th video block, B represents the total bandwidth between the satellite and the user, and SINR t represents the signal-to-noise ratio between the satellite and the user, which can be expressed as:

[0065]

[0066] where P is the transmission power of the satellite and N0 is the Gaussian white noise power. is the satellite receiving antenna gain, is the satellite transmitting gain, which is expressed as follows:

[0067]

[0068] where g max is the maximum antenna gain, ω t is the corresponding off-axis angle relative to the maximum radiation, and ω 3dB is the half-power angle of the transmitting antenna. is the space free-space loss, which is expressed as follows:

[0069]

[0070] where c is the speed of light (in km / s), f0 is the communication center frequency of the inter-satellite link (in Hz), and d is the slant range. is mainly affected by factors such as frequency, altitude, altitude, rainfall intensity, etc., and can be expressed as:

[0071]

[0072] wherein the equivalent effective path length, according to ITU-R P.618-12, l e t is a constant. η is the rainfall intensity, and η and λ are related to the frequency f0 and can be obtained from the report of ITU-R P.838;

[0073] The transmission delay of the t-th video block is calculated as follows:

[0074]

[0075] where D t represents the data volume of the t-th video block, and o t represents the satellite routing delay when transmitting the t-th video block;

[0076] (4.1.2) Video stream model:

[0077] The video buffer of the head-mounted device receives the video blocks from the satellite. When the t-th video block is transmitted, the video buffer is regarded as a queue, and the video transmitted from the satellite to the video buffer is regarded as an enqueue process; the head-mounted device playing the video from the video buffer is regarded as a dequeue process. Therefore, the change value of the video buffer is as follows:

[0078] Δc t = D t - y t

[0079] where y t represents the playback smoothness requirement of the head-mounted device, and the expression is as follows:

[0080] y t = u t Δt

[0081] where u t represents the rate at which the head-mounted device plays the video; the state of the video buffer queue is related to the buffer state at the end of the transmission of the previous video block and the change value at the end of the transmission of the current video block. B t represents the buffer length occupied at the end of the transmission of the t-th video block, and the expression is as follows:

[0082]

[0083] where B max is the upper limit value of the video buffer, and B t is between 0 and the maximum value to ensure that the video playback does not stall or the video buffer does not overflow;

[0084] (4.1.3) User QOE model:

[0085] The QOE of the user is determined by the smoothness of video playback. To ensure video smoothness, it is necessary to avoid insufficient video buffers and prevent video buffer overflows. The QOE expression for the user to request video services during the transmission of the t-th video block is as follows:

[0086]

[0087] where t h and s h are the user's delay preference and smoothness preference respectively, Δt represents the delay of the transmission of the t-th video block, and the N {·} function means that if the condition {·} holds, the output is 1, and if the condition does not hold, the output is 0. To prevent buffer underflow or overflow, we set e1 and e2 to 0.1 and 0.9;

[0088] (4.2) Use the transmission rate and buffer status as the input of the DQN template to calculate the routing selection and bandwidth allocation for each video block transmission;

[0089] (4.2.1) DQN model:

[0090] The DQN algorithm is defined by the state, action, and reward functions. Before transmitting each video block with the user, the satellite obtains state information, such as the transmission rate and video buffer status, by interacting with the environment, and makes actions based on the state information, that is, the satellite's bandwidth allocation strategy and routing selection. After executing the action, the satellite agent will receive a reward feedback and optimize the resource allocation decision according to the reward. Let the state be s t represent the user's state space, including the transmission rate and buffer length, that is, s t ={R t ,B t}, where t represents the t-th video block. Let the action be a t ={β t ,o t}, β t represents the decision of the bandwidth allocated to the user by the satellite during the transmission of the t-th video block, and o t represents the satellite routing delay during the transmission of the t-th video block. The reward is defined as:

[0091]

[0092] The goal of the DQN algorithm is to learn the optimal policy π * , to maximize the cumulative reward, which can be expressed as:

[0093]

[0094] Among them, \(0 < \delta < 1\) is the discount factor. To obtain the optimal policy, the Q-function is iteratively updated in a value-based manner. In each iteration, the update process of the Q-value can be given by:

[0095]

[0096] Among them, \(\alpha(\alpha\in[0,1])\) is the learning rate. To reduce the demand for storage resources by high-dimensional states and actions, a neural network is used to learn the effective mapping between states, actions, and rewards, so as to obtain the optimal policy. Specifically, a deep neural network (DNN) with weights \(\{\theta\}\) is used as a function approximator to fit the Q-function, which is expressed as follows:

[0097] Q(s t ,a t )≈Q'(s t ,a t ;θ)

[0098] Among them, the network Q' is trained to converge to the true Q-value, and the optimal policy in the state s t can be expressed as In addition, an action with a random selection probability of \(\epsilon(0 < \epsilon < 1)\) is selected to explore better policies and avoid the DQN algorithm from obtaining local optimal solutions. During the learning process, DQN stores some multiple \(\{s t ,a t ,r(s t ,a t ),s t+1} experiences in the experience pool to improve the learning efficiency. During the update process, a small batch is randomly selected from the experience pool as a sample to train the weights \(\theta\) of the Q network to minimize the following loss function:

[0099] Loss(\(\theta\)) = E(Q tar -Q'(s t ,a t ;\(\theta\)) 2

[0100] Among them

[0101] (4.3) The satellite module transmits the video block to the user module according to the decision of the DQN template;

[0102] Step 5, the user module receives the video block and plays it in real time.

[0103] The user module first puts the received video block into the video buffer, waits to read the RTP data packet from the buffer, and performs operations such as unpacking, verification, and error correction. When the user module unpacks correctly, it decodes the video block and plays it in real time on the display device.

[0104] The technical effects of the present invention will be further described below in combination with simulation experiments.

[0105] 1. Simulation Conditions

[0106] The user requests a video stream service. During this process, it is assumed that the user's head-mounted device has predicted the user's demand F based on the user's historical data and sent the user's demand to the satellite module. At the same time, the satellite module stores video block data packets with different resolution and frame rate combinations collected by the drone.

[0107] The platform for the simulation experiment is: Windows 10 operating system and Python 3.9.

[0108] The parameter settings for the simulation experiment are shown in Table 1.

[0109] Table 1 Simulation Parameters

[0110] Setting item Value <![CDATA[Video stream resolution requirement r h > {1920*1080、2560*1440、3840*2160} <![CDATA[Video stream frame rate requirement f h > {24，30，60} <![CDATA[User delay preference t h > {0.2，0.3，0.4，0.5，0.6，0.7，0.8} <![CDATA[User fluency preference s h > {0.2，0.3，0.4，0.5，0.6，0.7，0.8}

[0111] 2. Simulation Content

[0112] Under the above scenario, the average bandwidth comparison diagrams of the present invention and the existing solution are respectively simulated, and the results are as Figure 3 .

[0113] The average bandwidth consumption of the present solution is calculated by using a refined general template library for the average bandwidth consumption under different resolution and frame rate combinations;

[0114] The service delay of the existing solution is calculated by using the existing template library for the average bandwidth consumption under different resolution and frame rate combinations.

[0115] 3. Analysis of Simulation Results

[0116] From Figure 3 it can be seen that the solution of the present invention is superior to the existing solution in terms of average bandwidth consumption. The present invention has a more refined template library, which can more accurately identify the user's intention, so as to be able to match a more accurate DQN template. Its decision-making accuracy is more in line with the actual environmental needs. Therefore, under the condition of meeting the user's needs, it reduces the consumption of communication resources and improves the resource utilization rate. The comparison solution uses the existing template library to match the user's intention, and cannot accurately identify the user's intention, and thus cannot obtain an accurate DQN template. Therefore, its decision-making accuracy is poor and it cannot meet the actual environmental needs. In order to meet the user's needs, it increases the consumption of communication resources.

[0117] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data that have been authorized by the user or fully authorized by all parties. Moreover, the collection, use, and processing of relevant data need to comply with the laws, regulations, and standards of relevant countries and regions, and corresponding operation entrances are provided for users to choose to authorize or reject.

[0118] The above simulation analysis proves the correctness and effectiveness of the method proposed by the present invention.

[0119] The parts not described in detail in the present invention belong to the common general knowledge of those skilled in the art.

[0120] The above are only the preferred embodiments of the present invention and are not intended to limit the present invention. Obviously, for professionals in the field, after understanding the content and principle of the present invention, various modifications and changes in form and details may be made without departing from the principle and structure of the present invention. However, these modifications and changes based on the idea of the present invention are still within the scope of protection of the claims of the present invention.

Claims

1. A method for transmitting video streams in network - free areas based on space - air cooperation, characterized in that, It includes the following steps: (1) Construct an air and space integrated communication system, including a drone module, a satellite module and a user module; the satellite module includes a communication ground station and a satellite, and the satellite contains a pre-trained deep Q-network DQN decision template library; the drone module is used to sense real-time scenes and generate video streams, and at the same time transmit video stream data packets to the satellite module; the satellite module processes it and transmits the result to the user module; The user module is implemented through a head-mounted device; (2) According to the historical data sequence of the user's requested video stream service, the user module uses the head-mounted device to predict the user's demand and synchronize the prediction result to the satellite module; (3) The satellite module performs templated intention recognition according to the prediction result synchronized by the head-mounted device, translates the user's demand into a parameter sequence composed of resolution, frame rate, latency preference, and smoothness preference; and matches the pre-trained DQN template in the DQN decision template library according to this parameter sequence; (4) Construct a quality of experience QOE model: where t h and s h are the user's latency preference and smoothness preference respectively, Δt represents the transmission latency of the t-th video chunk, N {·} function outputs 1 if the condition of {·} holds, otherwise outputs 0; e1 and e2 are preset thresholds to prevent buffer underflow and overflow; B max is the upper limit of the video buffer, B t represents the video buffer status when the transmission of the t-th video chunk is completed; (5) Use the DQN template to adjust the transmission strategy of the video stream in real time according to the user's QOE model. The satellite module transmits the video blocks to the user module according to the decision of the DQN template, that is, the adjusted transmission strategy; (6) The user module puts the received video blocks into the video buffer, waits to read RTP data packets from the buffer, and performs unpacking, verification and error correction operations. The error correction operation is performed when the unpacking and verification in the user module are incorrect; after the verification is correct, the video blocks are decoded and played in real time on the display device.

2. The method according to claim 1, characterized in that: In the air and space integrated communication system constructed in step (1), when the drone module sends the video stream data packet to the satellite module, it encodes and stores the original video stream data in different resolutions and frame rates, and divides the encoded video stream into video blocks of a fixed size, encapsulates and packs them into RTP data packets and sends them to the satellite module; the communication ground station in the satellite module receives the RTP data packets from the drone module and forwards them to the satellite, and the satellite stores the received RTP data packets; the satellite contains a pre-trained deep Q-network DQN decision template library, where the template is used to adjust the transmission strategy according to the transmission environment; when the satellite receives the user's intention, it recognizes and translates the user's intention, and matches the RTP data packets of the corresponding resolution and frame rate format according to the translation result, that is, the video stream data packets, and the corresponding DQN decision template. The video stream data packets are transmitted to the user module according to the routing result of the DQN template and the formulated bandwidth allocation strategy; The user module predicts the user's demand for watching the video stream through the head-mounted device and sends it to the satellite module, and at the same time receives the video stream data packets transmitted by the satellite and performs display and playback; the buffer in the head-mounted device stores different amounts of video streams according to the buffer size.

3. The method according to claim 1, characterized in that: The prediction result in step (2) includes the video stream information and preference information requested by the user at a certain place in the coming moment; it is to predict the user's demand F using the head-mounted device according to the historical data sequence H of the user's requested video stream service: H = {{T 1 , L 1 , V 1 , U 1},..., {T h , L h , V h , U h},..., {T H , L H , V H , U H}} F = H + 1; where T h , L h , V h , U h respectively represent the time, location, video stream information, and user preference information of the user's h-th request for the video stream; V h = {r h , f h}, where r h and f h respectively represent the resolution and frame rate of the video stream; U h = {t h , s h}, where t h and s h respectively represent the user's preferences for latency and smoothness.

4. The method according to claim 1 or 2, characterized in that: The historical data includes the time and location of the user's request for the video stream, the video stream information, and the user preference information, where the video stream information includes the resolution and frame rate of the video stream, and the user preference information includes the user's preferences for latency and smoothness.

5. The method according to claim 1, wherein: The intention recognition in step (3) includes recognizing two parts: the video stream related parameters and the user's personalized preferences; among them, the video stream related parameters include the video resolution parameter and the frame rate parameter; the video resolution parameter is used to measure the data volume size of a single-frame video image, and the video frame rate parameter is the number of frames displayed by the head-mounted device per second; the user's personalized preferences include the personalized preferences for latency and smoothness; If the latency preference coefficient is larger, the DQN template tends to low latency and selects a large bandwidth when making a decision; On the contrary, when the smoothness preference coefficient is larger, the DQN template tends to smoothness when making a decision and selects a bandwidth that is large first and then small, so as to keep the buffer size stable within a preset range.

6. The method according to claim 1, wherein: The video buffer state when the t-th video block is transmitted in step (4) is expressed as follows: where B max is the upper limit value of the video buffer, and B t-1 represents the buffer state when the (t - 1)-th video block is completely transmitted; Δc t represents the buffer change value from when the (t - 1)-th video block is completely transmitted to when the t-th video block is completely transmitted.

7. The method according to claim 1, wherein: The adjustment of the video stream transmission strategy in step (5) is specifically as follows: The satellite module calculates the transmission latency and the video buffer state in the QOE model, and uses the transmission rate and the buffer state as the input of the DQN template to output a transmission decision, that is, the routing selection and bandwidth allocation when each video block is transmitted.

8. The method according to claim 7, wherein: The transmission latency is calculated according to the following formula: Among which D t represents the data volume of the t-th video block, o t represents the satellite routing delay when transmitting the t-th video block, R t represents the transmission rate when transmitting the t-th video block.

9. The method according to claim 8, characterized in that: The transmission rate is obtained according to the following formula: R t = β t Blog2(1 + SINR t ) where β t is the bandwidth ratio allocated to the user when the t-th video block is transmitted, B represents the total bandwidth between the satellite and the user, and SINR t represents the signal-to-noise ratio between the satellite and the user, expressed as: where P is the transmission power of the satellite and N0 is the Gaussian white noise power; is the space idle loss; c is the speed of light in km / s; f0 is the communication center frequency of the inter-satellite link in Hz; d is the slant range; is the rainfall attenuation; is the satellite receiving antenna gain, is the satellite transmitting gain, expressed as follows: where g max is the maximum antenna gain, ω t is the corresponding off-axis angle relative to the maximum radiation, ω 3dB is the half-power angle of the transmitting antenna.

10. The method according to claim 9, characterized in that: The factors affecting the rainfall attenuation include frequency, elevation angle, altitude, and rainfall intensity.

Images