Data packet adaptive aggregation compression method and device

Abstract

The invention relates to a data packet adaptive aggregation compression method and device, and belongs to the technical field of communication. The method uses a deep reinforcement learning algorithm to optimize data packet transmission between a mobile relay and a base station, and uses an artificial deep neural network to represent a data packet aggregation compression strategy; and obtaining an optimal aggregation compression strategy under different channel states and data packet buffer queue states. The method provided by the invention solves the problem of performance reduction caused by the fact that the existing heuristic data packet aggregation and compression strategy scheduling method cannot well adapt to a dynamic network transmission environment, thereby quickly adapting to a high time-varying transmission environment and making dynamic adjustment according to the environment.

Description

Technical Field

[0001] The present invention belongs to the field of communications, and in particular relates to a method and device for adaptively aggregating and compressing data packets. Background Art

[0002] Due to the development of multimedia technology, the data traffic in the network has exploded. Data compression has been widely studied as an effective means to reduce transmission traffic and alleviate network congestion. In scenarios where transmission bandwidth is limited, it becomes more necessary to compress data to reduce transmission volume. According to the composition structure of the data packet, the compression method is divided into two categories: header compression and payload compression. Among them, the main processing flow of payload compression is: the sender establishes dictionary information based on the common sequence in the payload, and then replaces the sequence with a shorter feature code in the message; the receiver uses the same dictionary for decompression. Representative algorithms include LZ4, LZO, and zlib.

[0003] Since additional compression / decompression delay will be generated when compressing the transmitted data, it is necessary to select a suitable compression algorithm based on factors such as specific channel bandwidth to balance traffic reduction and processing time. In the existing technology, the data compression mechanism adjusts the compression according to pre-set heuristic rules, and requires accurate modeling of the transmission environment to make a correct response, which may result in poor performance in a highly time-varying network environment.

[0004] Aggregation compression is a compression method for high-latency networks. In each round of aggregation compression, if the number of packets in the buffer reaches a preset value or the waiting time exceeds a threshold, aggregation compression is performed. However, in the prior art, since the aggregation parameters are fixed in advance and there is no mechanism for dynamic adjustment to the environment, the performance is poor. Summary of the invention

[0005] In view of the above-mentioned deficiencies in the prior art, the purpose of the invention is to provide a method and device for adaptive aggregation and compression of data packets, which uses a deep reinforcement learning algorithm to optimize the transmission of data packets between mobile relays and base stations, and uses an artificial deep neural network to represent the data packet aggregation and compression strategy, so as to obtain the optimal aggregation and compression strategy under different channel states and data packet buffer queue states, thereby quickly adapting to highly time-varying transmission environments and making dynamic adjustments according to the environment.

[0006] A first aspect of the present invention provides a method for adaptively aggregating and compressing data packets, comprising:

[0007] S1, dividing the process of forwarding data packets from the mobile relay to the ground base station into multiple control cycles with an aggregated compression control cycle T;

[0008] S2, at the beginning of each control cycle, obtaining wireless channel state parameters and relay data packet buffer queue state;

[0009] S3, determining an aggregation compression action according to the wireless channel state parameters and the relay data packet buffer queue state;

[0010] S4, within the control period, according to the aggregation and compression action, performing a data packet aggregation and compression operation and sending the data packet to a ground base station;

[0011] The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

[0012] Furthermore, the wireless channel state parameters and relay data packet buffer queue state include:

[0013] The wireless channel state parameters include: channel quality δ of the channel within a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR;

[0014] The relay data packet buffer queue state includes: the number of relay buffered data packets at the beginning of the control period δ M [t], the number of relay packets arriving at the end of the control period δ N [t], and whether the aggregated compressed data packet of the previous control cycle was successfully transmittedδ T [t-1], and δ T [t-1]∈{0,1};

[0015] Wherein, t represents the number of the control cycle.

[0016] Further, determining the aggregation compression action according to the wireless channel state parameter and the relay data packet buffer queue state includes:

[0017] Inputting the wireless channel state parameters and the relay data packet buffer queue state into a deep deterministic policy gradient model to obtain an aggregation compression action consisting of a data packet aggregation quantity and a compression method;

[0018] The deep deterministic policy gradient model is composed of a first neural network and a second neural network, wherein the first neural network is used to represent the data packet aggregation and compression strategy, and the second neural network is used to represent the quality of the data packet aggregation and compression strategy;

[0019] The data packet aggregation and compression strategy includes the data packet aggregation number α A [t] and compression method α C [t], αA [t] represents the number of data packets aggregated in control period t, α C [t] represents the compression method of the control period t.

[0020] Furthermore, the data packet adaptive aggregation compression method of the present invention further includes:

[0021] Calculate the reward value output by the model in each control cycle according to the set reward function, and observe the state of the network environment after executing the control action;

[0022] The wireless channel state parameters and the relay data packet buffer queue state in the control period are combined into an environment state δ[t] before the control action is executed, and the wireless channel state parameters and the relay data packet buffer queue state after the control action is executed are combined into an environment state δ[t+1] after the control action is executed;

[0023] The data training sample of the four-tuple (δ[t], α[t], R[t], δ[t+1]) consisting of the environment state δ[t] before the control action is executed, the aggregate compression action α[t], the reward value R[t], and the environment state δ[t+1] after the control action is executed is stored in the experience replay pool;

[0024] Randomly sampling data training samples from the experience replay pool, inputting them into the deep deterministic policy gradient model, and updating the first neural network parameters and the second neural network parameters according to the loss function;

[0025] Among them, α[t] represents the compression action of control period t, and R[t] represents the reward value of control period t.

[0026] Furthermore, the channel quality δ Q [t] is based on the previous k Q The historical channel quality data of the control cycle is predicted based on the pre-trained LSTM network, and the transmission bandwidth δ B [t] is based on k B The historical transmission bandwidth of the k control cycles is predicted based on the pre-trained long short-term memory network LSTM network; Q and k B The value is preset.

[0027] A second aspect of the present invention provides a device for implementing a method for adaptively aggregating and compressing data packets, comprising:

[0028] A division module is configured to divide the process of forwarding a data packet from a mobile relay to a ground base station into a plurality of control periods with an aggregated compressed control period T;

[0029] An acquisition module is configured to acquire wireless channel state parameters and relay data packet buffer queue state at the beginning of each control cycle;

[0030] A calculation module, configured to determine an aggregation compression action according to the wireless channel state parameters and the relay data packet buffer queue state;

[0031] A compression module, configured to perform a data packet aggregation compression operation and send the data packet to a ground base station according to the aggregation compression action within the control period;

[0032] The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

[0033] A third aspect of the present invention provides a device for implementing a method for adaptively aggregating and compressing data packets, comprising a memory, a processor, and a user interface;

[0034] The memory is used to store computer programs;

[0035] The user interface is used to interact with the user;

[0036] The processor is used to read the computer program in the memory, and when the processor executes the computer program, the above-mentioned data packet adaptive aggregation compression method is implemented.

[0037] According to a fourth aspect of the present invention, a processor-readable storage medium is provided, wherein the processor-readable storage medium stores a computer program, and when the processor executes the computer program, the above-mentioned data packet adaptive aggregation compression method is implemented.

[0038] The beneficial effects of the present invention are as follows:

[0039] The data packet adaptive aggregation and compression method of the present invention uses a deep reinforcement learning algorithm to optimize the data packet transmission between the mobile relay and the base station, and uses an artificial deep neural network to represent the data packet aggregation and compression strategy to obtain the optimal aggregation and compression strategy under different channel states and data packet buffer queue states. It solves the problem that the existing heuristic data packet aggregation and compression strategy scheduling method cannot adapt well to the performance degradation caused by the dynamic network transmission environment, thereby quickly adapting to the high time-varying transmission environment and making dynamic adjustments according to the environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are only used to illustrate specific embodiments and are not considered to limit the present invention. In the entire drawings, the same reference symbols represent the same components. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings.

[0041] Figure 1 It is a schematic diagram of the connection between the terminal device, the mobile relay and the base station;

[0042] Figure 2 A schematic diagram of a method for adaptively aggregating and compressing data packets according to an embodiment of the present invention;

[0043] Figure 3 A schematic diagram of an adaptive aggregation compression decision structure according to an embodiment of the present invention;

[0044] Figure 4 A schematic diagram of another data packet adaptive aggregation compression method according to an embodiment of the present invention;

[0045] Figure 5 This is a graph showing the average reduction in delay when processing and forwarding each data packet in 100 control cycles between the present invention and the other two solutions;

[0046] Figure 6 This is a comparison chart of the average delay reduction of the present invention and the other two solutions under different network bandwidth transmission conditions when forwarding and transmitting 100 data packets;

[0047] Figure 7 is the average packet loss rate of the present invention and the other two solutions under different network bandwidth transmission conditions and user equipment data packet arrival frequency;

[0048] Figure 8 A schematic diagram of a data packet adaptive aggregation and compression device according to an embodiment of the present invention;

[0049] Fig. 9 Schematic diagram of another data packet adaptive aggregation and compression device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in the field without making creative work should fall within the scope of protection of the present invention.

[0051] Furthermore, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessarily obscuring the concepts disclosed in the present invention.

[0052] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the orientation or position relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inside", "outside", etc. is based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation of the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. The terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0053] Due to the development of multimedia technology, the data traffic in the network has exploded. Data compression has been widely studied as an effective means to reduce transmission traffic and alleviate network congestion. This is especially true in scenarios where user terminals on high-speed moving trains, cars, and aircraft transmit data. Figure 1 As shown, mobile relay devices are often required to forward data to the base station, and data transmission between the terminal device and the base station needs to be forwarded through the relay device. Due to the rapid changes in the transmission environment, the data transmission rate is greatly affected, and the transmission bandwidth is often insufficient, so it is even more necessary to reduce the transmission volume. In scenarios such as high-speed moving trains, cars, and aircraft, the network environment is highly time-varying. In the prior art, the data compression mechanism adjusts the compression based on pre-set heuristic rules, and requires accurate modeling of the transmission environment to make a correct response, which may result in poor performance in a highly time-varying network environment. Aggregation compression is a compression method for high-latency networks. In each round of aggregation compression, if the number of data packets in the buffer reaches a preset value or the waiting time exceeds a threshold, aggregation compression is performed. However, in the prior art, since the aggregation parameters are fixed in advance and there is a lack of a mechanism for dynamic adjustment of the environment, the performance is poor.

[0054] In view of the above problems existing in the prior art, the present invention proposes a method, device and storage medium for adaptive aggregation and compression of data packets. The present invention uses a deep reinforcement learning algorithm to optimize the transmission of data packets between mobile relays and base stations, and uses an artificial deep neural network to represent the data packet aggregation and compression strategy, and obtains the optimal aggregation and compression strategy under different channel states and data packet buffer queue states, thereby solving the problem of performance degradation caused by the inability of existing heuristic data packet aggregation and compression strategy scheduling methods to adapt to dynamic network transmission environments. In the present invention, the Actor-Critic reinforcement learning framework is used to separate the policy output and policy evaluation, thereby improving the efficiency of policy optimization and accelerating the training of the neural network for the adaptive aggregation and compression scheduling strategy of data packets.

[0055] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Instead, they are merely examples of methods and systems consistent with some aspects of the present invention as detailed in the appended claims.

[0056] In a first aspect of the present invention, the present invention provides a method for adaptively aggregating and compressing data packets, such as Figure 2 As shown, the method includes steps S1 to S4:

[0057] S1, dividing the process of forwarding data packets from the mobile relay to the ground base station into multiple control cycles with an aggregated compression control cycle T;

[0058] S2, at the beginning of each control cycle, obtaining wireless channel state parameters and relay data packet buffer queue state;

[0059] S3, determining an aggregation compression action according to the wireless channel state parameters and the relay data packet buffer queue state;

[0060] S4, within the control period, according to the aggregation and compression action, performing a data packet aggregation and compression operation and sending the data packet to a ground base station;

[0061] The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

[0062] The adaptive aggregation compression algorithm of the present invention can be used in a mobile relay device installed in a high-speed mobile device, such as a high-speed train, an aircraft, etc. Taking a high-speed train as an example, in a carriage, assuming that there is δ N [t] data packets are transmitted from the user equipment to the relay in control period t, and the length of each data packet header is size H(Its length is usually fixed), the length of the payload is size i . t represents the number of the control cycle.

[0063] It should be noted that, in the symbols used in the present invention, [t] indicates the value in the control period t, and when the control period t is different, the values ​​indicated by the same symbol may be different. Similarly, [t-1] indicates the value in the control period t-1, and [t+1] indicates the value in the control period t+1. For example, δ N [t] means that in the control cycle t, there is δ N [t] packets are transmitted from the user equipment to the relay in control period t. For another example, α c [t] represents the compression algorithm used in control cycle t, α c [t-1] represents the compression algorithm used in control cycle t-1, α c [t+1] indicates the compression algorithm used in control cycle t+1.

[0064] δ N [t] packets are then stored in a buffer queue with a capacity of M. The mobile relay will make a decision to A [t]∈{1,…,M} packets are aggregated (α A [t] is the number of data packets in the buffer queue that are aggregated and compressed during control period t, ranging from 1 to M), and then selects the appropriate algorithm α from l compression algorithms c [t]∈{0,1,…,j,…,l} is used for compression processing, where j is the number of the algorithm and the compression rate of algorithm j is R j , the compression speed and decompression speed are recorded as and You can also use and Respectively represent the compression algorithm α C [t] compression ratio, compression speed and decompression speed. If α C [t] = 0, no compression is performed on the aggregated data packet.

[0065] After data packet aggregation and payload compression operations, the transmission efficiency can be expressed as follows:

[0066]

[0067] From formula 1, we can see that aggregating more data packets can achieve higher transmission efficiency and help to further improve the compression rate by utilizing the similarity of consecutive data packets in the stream. However, under the influence of bit errors caused by low channel quality, more data packets will need to be retransmitted each time a transmission fails. M[t+1]=min{δ M [t]+δ N [t]-α A [t]×δ T [t],M} represents the number of buffered packets at the end of control period t, where δ T [t]∈{0,1} indicates whether the aggregated data packet is successfully transmitted (1 indicates successful transmission, 0 indicates failed transmission). Multiple consecutive transmission failures will cause too many pending data packets to be accumulated in the buffer queue. When the arriving data packets exceed the capacity M, the relay will no longer be able to receive the user's data packets, and the part exceeding the capacity will be lost.

[0068] The selected compression algorithm will affect the transmission efficiency and transmission delay. An algorithm with a higher compression ratio can effectively reduce the length of the data packet and save more link bandwidth, but this will also introduce a longer compression processing delay. B [t] represents the channel bandwidth within the control period t. When the data packet is numbered α C [t] algorithm compression transmission will be slower than direct transmission without processing. The selection of compression algorithm needs to refer to the specific channel quality and link bandwidth.

[0069] Affected by factors such as the deployment of rail communication facilities and the surrounding environment, the quality of the wireless transmission link between the mobile relay and the ground base station changes dynamically. The channel bandwidth is represented by δ B [t], and select the channel quality indicator (CQI) w t and signal-to-noise ratio (SNR) t The indicator evaluates the channel quality, denoted as δ Q [t]=[w t ,y t ], the bit error rate can be calculated by the following formula, where erfc represents the complementary error function.

[0070]

[0071] Then the probability of successful transmission of the aggregated compressed data packet is:

[0072]

[0073] It should be noted that since the decompression speed is usually faster than the compression speed and occupies less computing resources such as CPU and memory, the ground base station in the present invention can always decompress the aggregated data packet after correctly receiving it and restore the original data packet content.

[0074] It should be noted that the probability of successful data packet transmission expressed by the above formula 3 means that the data packet may fail to be transmitted. If the data packet fails to be transmitted, it is retransmitted. Only when the data packet is successfully transmitted, it is removed from the buffer, otherwise it is not removed.

[0075] In some embodiments, in S1, the process of forwarding data packets from the mobile relay to the ground base station is divided into multiple control periods with an aggregation compression control period T, and the aggregation compression process of forwarding data from the mobile relay to the ground base station is controlled with T as a period.

[0076] In some embodiments, the aggregation compression control period T is pre-set according to system requirements. Preferably, the aggregation compression control period T is equal to 0.02 seconds.

[0077] In some embodiments, the wireless channel state parameters and relay data packet buffer queue state in S2 include:

[0078] The wireless channel state parameters include: channel quality δ of the channel within a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR;

[0079] The relay data packet buffer queue state includes: the number of relay buffered data packets at the beginning of the control period δ M [t], the number of relay packets arriving at the end of the control period δ N [t], and whether the aggregated compressed data packet of the previous control cycle was successfully transmittedδ T [t-1], and δ T [t-1]∈{0,1};

[0080] Wherein, t represents the number of the control cycle.

[0081] In the present invention, δ T [t] indicates whether the aggregated compressed data packet of control period t is transmitted successfully, 0 indicates transmission failure, and 1 indicates transmission success. T [t-1] indicates whether the aggregated compressed data packet of the previous control cycle t-1 of the current control cycle t is transmitted successfully. 1 indicates that the transmission fails, and 0 indicates that the transmission is successful.

[0082] In some embodiments, in S3, determining the aggregation compression action according to the wireless channel state parameter and the relay data packet buffer queue state includes:

[0083] Inputting the wireless channel state parameters and the relay data packet buffer queue state into a deep deterministic policy gradient model to obtain an aggregation compression action consisting of a data packet aggregation quantity and a compression method;

[0084] The deep deterministic policy gradient model consists of a first neural network and a second neural network. The first neural network is used to represent the data packet aggregation and compression strategy, and the second neural network is used to represent the quality of the data packet aggregation and compression strategy.

[0085] Figure 3 It is a schematic diagram of the adaptive aggregation compression decision structure. Figure 3 In the present invention, the agent is a data packet aggregation and compression control program deployed on a relay device, and the environment is a transmission channel, a relay data packet buffer queue, and a user device that transmits data packets to the relay. Figure 3 In the control cycle t, at the beginning, the agent obtains the channel quality δ of the next-hop transmission channel in the relay within the cycle Q [t], bandwidth δ B [t and whether the aggregated compressed data packet within period t-1 is transmitted successfullyδ T [t-1], the number of packets in the relay buffer queue δ M [t]; The agent outputs the data packet aggregation strategy α based on the information obtained A [t] and compression strategy α C [t], and perform data packet aggregation and compression operations. The relay transmits the aggregated compressed data packet to the next hop receiver through the wireless channel. After the above operations are completed, at the end of the control cycle t, the relay obtains feedback from the receiver to confirm whether the data packet is successfully transmitted. T [t], the number of packets received from user equipment is δ N [t], and calculate the number of packets in the update buffer queue δ M [t+1]. Enter the next control cycle.

[0086] In some embodiments, the first neural network is an Actor neural network, which is used to decide which compression strategy to use to aggregate and compress data packets, also known as a strategy network. The second neural network is a Critic neural network, which is used to indicate the quality of data packet aggregation and compression strategies, also known as a value network.

[0087] For example, for the control period t, according to the input wireless channel state parameters and relay packet buffer queue state, a set of Actor neural network outputs representing the number of packet aggregation α is obtained. A [t]∈(1,2,…,10) and compression method α C [t]∈(0,1,2,…,7) control action. Among them, α AThe value of [t] ranges from 1 to 10, indicating that the number of packets aggregated is 1 to 10. C The value of [t] ranges from 0 to 7, corresponding to no compression, compression using snappy, LZO, LZ4, zlib2, zlib4, zlib7, and zlib9 algorithms, respectively. During this period, the mobile relay performs data packet aggregation compression operations according to the control action and sends it to the ground base station.

[0088] It should be noted that, in the present invention, in addition to the above-mentioned snappy, LZO, LZ4, zlib2, zlib4, zlib7, and zlib9 algorithms for compression, other compression algorithms may also be included, and the present invention does not limit them. Each compression algorithm is α C [t] is a unique value, that is, within a certain control cycle t, α C [t] The specific value indicates the serial number of the compression algorithm used in the control cycle t. The compression algorithm used may be different in different control cycles.

[0089] In each control cycle t, the currently observed wireless channel state parameters and relay packet cache queue state information are input into the deep deterministic policy gradient model, and then the algorithm outputs the aggregation compression strategy (α A [t],α C [t]), instructing the mobile relay to aggregate and compress the data packets according to the specified number of aggregated data packets and payload compression algorithm. The reduced transmission delay is used as a reward and the number of packet losses is used as a penalty to optimize the decision strategy.

[0090] In some embodiments, the first neural network and the second neural network are used after training. In the training phase, the entire first neural network and the second neural network are first trained offline using the training data, and then deployed and put into use in a specific implementation scenario. In the training, in each control cycle t, whose duration is fixed, for example, equal to 0.02 seconds, the agent reads the state δ[t]=(δ M [t],δ N [t],δ Q [t],δ B [t],δ T [t-1]), reducing the transmission delay is used as a reward and the number of packet losses is used as a penalty to optimize the decision strategy. Q [t] and bandwidth δ B [t] respectively from the previous k Q Quality information and k B The bandwidth information is predicted. Q =12, k B =7.

[0091] In some embodiments, the training data for offline training is obtained in the following ways, including but not limited to: public data sets, which are usually provided by institutions and organizations such as mobile operators; self-generated data sets. If the public data sets do not meet the requirements, simulators such as the ns3 offline event simulator can be used to simulate the network data transmission process in different environments to generate a large amount of interaction data, or interact with the real environment to collect data.

[0092] In some embodiments, Figure 4 As shown, the present invention provides a data packet adaptive aggregation compression method, the method comprises steps S1 to S7,

[0093] S1 to S4 are the same as those in the above method embodiment and will not be described in detail here.

[0094] S5, calculate the reward value output by the model in each control cycle according to the set reward function, and observe the state of the network environment after executing the control action;

[0095] In this step, the reduced transmission delay is used as a reward and the number of packet losses is used as a penalty to optimize the decision strategy.

[0096] In some embodiments, the reward value is calculated by a reward function, which is:

[0097] Z=βt save -∈m lost ; Formula 4

[0098] Among them, Z is the reward function, β represents the impact weight of transmission delay reduction, and ∈ represents the impact weight of packet loss;

[0099] t save Indicates the latency reduced by the data packet aggregation and compression operation in one cycle;

[0100] m lost Indicates the number of packets lost due to multiple packet transmission failures and / or packet overflow in the relay buffer queue;

[0101] t save Determined by the following formula:

[0102]

[0103]

[0104] m lost Determined by the following formula:

[0105] m lost =max{δ M [t]+δ N [t]-αA [t]×δ T [t]-M,0}; Formula 6

[0106] Among them, size H and size i are the header and body sizes of data packet i respectively;

[0107] and They are compression algorithms α C [t] compression ratio, compression speed and decompression speed;

[0108] P success [t] is the probability of successful transmission of the aggregated compressed data packet in the control period t, and the P success [t]Satisfy:

[0109]

[0110] M is the capacity of the buffer queue;

[0111] α A [t] is the number of compressed data packets, α A [t]∈{1,…,M};

[0112] δ T [t] indicates whether the aggregated data packet is successfully transmitted, δ T [t]∈{0,1};

[0113] δ M [t] is the number of packets cached by the relay at the beginning of control period t;

[0114] i is the number of the data packet;

[0115] erfc represents the complementary error function.

[0116] S6, store the data training sample of the four-tuple (δ[t], α[t], R[t], δ[t+1]) consisting of the environment state δ[t] before the control action is executed, the aggregate compression action α[t], the reward value R[t], and the environment state δ[t+1] after the control action is executed in the experience replay pool.

[0117] The wireless channel state parameters and the relay data packet buffer queue state within the control period constitute the environment state δ[t] before the control action is executed, and the wireless channel state parameters and the relay data packet buffer queue state after the control action is executed constitute the environment state δ[t+1] after the control action is executed.

[0118] S7, randomly sampling data training samples from the experience replay pool, inputting them into the deep deterministic policy gradient model, and updating the first neural network parameters and the second neural network parameters according to the loss function.

[0119] After S7 is executed, the process returns to S2. That is, in each control cycle t, S2, S3, S4, S5, S6 and S7 are executed.

[0120] In some embodiments, the wireless channel state parameter includes: the channel quality δ of the channel in a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR. Where the channel quality δ Q [t] is based on the previous k Q The historical channel quality data of the control cycle is predicted based on the pre-trained LSTM network, and the transmission bandwidth δ B [t] is based on k B The historical transmission bandwidth of the k control cycles is predicted based on the pre-trained long short-term memory network LSTM network; Q and k B The value is preset.

[0121] Affected by factors such as the speed of the moving train and the trackside environment, the bandwidth of the wireless network between the mobile relay and the ground base station changes dramatically. If the historical bandwidth and signal-to-noise ratio data are directly used as the input of the deep deterministic policy gradient model, the model convergence speed may not keep up with the frequency of bandwidth changes, resulting in increased training time and slower model convergence. To this end, a channel state prediction module is added to the aggregation compression control, and a long short-term memory network (LSTM) is used to predict the bandwidth and channel quality. The specific approach is: k B A vector composed of historical bandwidths B and k Q A is composed of historical channel states Q The k is input into two pre-trained LSTM networks respectively to obtain the predicted transmission bandwidth and predicted channel quality value of the control period t. Q and k B The value of is pre-set, for example, k Q =12, k B =7.

[0122] In some embodiments, the data packet adaptive aggregation compression method of the present invention further includes:

[0123] The second neural network outputs a first function value as the estimation information of the value function according to the wireless channel state parameters, the relay data packet buffer queue state and the data packet aggregation compression action in the data training sample;

[0124] The first neural network updates the first neural network parameters by using a policy gradient algorithm according to the reward value and the value function, and the second neural network updates the second neural network parameters by using a temporal difference learning TD algorithm according to the valuation information of the value function;

[0125] The value function is the action value function output by the second neural network.

[0126] The action value function is a function used to represent the quality of the packet aggregation and compression strategy.

[0127] In the present invention, the TD algorithm is Temporal Difference (TD), which is an algorithm in reinforcement learning. The policy gradient algorithm is also an algorithm in reinforcement learning.

[0128] The data packet adaptive aggregation and compression method of the present invention uses a deep reinforcement learning algorithm to optimize the data packet transmission between the mobile relay and the base station, and uses an artificial deep neural network to represent the data packet aggregation and compression strategy to obtain the optimal aggregation and compression strategy under different channel states and data packet buffer queue states. The problem of performance degradation caused by the inability of existing heuristic data packet aggregation and compression strategy scheduling methods to adapt well to the dynamic network transmission environment is solved, thereby quickly adapting to the high time-varying transmission environment and making dynamic adjustments according to the environment.

[0129] Compared with other polymerization compression methods, the polymerization compression method of the present invention has obvious effect. The compression schemes compared are scheme 1 and scheme 2.

[0130] Solution 1: Fixed strategy compression solution: Ignoring the changes in wireless network channel quality and bandwidth, the relay uses a fixed compression algorithm, such as zlib, for all data packets that need to be forwarded from user devices. They are processed one by one at a constant compression rate and speed and then sent to the ground base station.

[0131] Solution 2: Aggregation and compression solution: The relay aggregates and compresses a fixed number of data packets in the buffer. If the number of incoming data packets does not reach the expected aggregation number and the waiting time exceeds the preset threshold, the existing data packets are directly aggregated and compressed.

[0132] Figure 5 This is one of the comparisons between the solution of the present invention and solution 1 and solution 2, which is a comparison of the average reduction of the forwarding delay of each data packet in each control cycle under the dynamic change of the wireless network. Figure 5As shown, the delay reduction of the method of the present invention fluctuates due to the time-varying channel bandwidth, but is always positive and outperforms other schemes in most time periods. The fixed strategy full compression scheme of Scheme 1 uses a single compression algorithm. Under high transmission bandwidth conditions, it still takes a long time to perform unnecessary deep compression on the data packet payload, which increases the transmission delay. The aggregate compression scheme of Scheme 2 uses the similarity of continuous data packets in the stream through a block compression method, which improves the compression rate and increases the effective bandwidth to a certain extent. However, since the number of aggregated data packets is relatively fixed, it takes a long time to perform compression in scenarios where data packets arrive at a lower frequency, which will also have a negative impact on the transmission delay.

[0133] Figure 6 The total delay improvement of the method of the present invention is compared with that of solution 1 and solution 2 under different network transmission environments. Figure 6 As shown, in terms of overall performance, the method of the present application is the best. In high-bandwidth scenarios, it can adaptively adjust the data packet aggregation compression strategy and appropriately select a compression algorithm with a low compression rate and high processing speed to ensure the improvement of latency. The fixed full compression scheme of Scheme 1 and the aggregate compression scheme of Scheme 2 use the same compression algorithm to process all data packets from beginning to end, which will result in additional compression processing time overhead in high-bandwidth scenarios, making the latency reduction negative.

[0134] Figure 7 The packet loss rate comparison between the method of the present application and solution 1 and solution 2 under different network transmission environments and user device data packet arrival frequencies. Figure 7 As shown, in the case of low network bandwidth, since the mobile relay needs a long time to transmit data to the base station, the possibility of user data packets being accumulated in the buffer queue is higher, so the packet loss rate caused by data packet overflow in the three schemes is relatively high. In scheme 2, since the number of data packet aggregation is fixed, the volume of packets that need to be retransmitted due to bit errors is larger, and the retransmission time is longer, making the data packet overflow buffer situation more serious than other schemes. The method of the present application can dynamically adjust the number of aggregated data packets according to the transmission quality of the channel and the remaining capacity of the buffer, so the packet loss rate is lower, which is reduced by 17.2% and 53.4% ​​respectively compared with the other two schemes.

[0135] In a second aspect, based on the same inventive concept, another specific embodiment of the present invention discloses a data packet adaptive aggregation compression device, such as Figure 8 As shown, including:

[0136] The division module 801 is configured to divide the process of forwarding a data packet from the mobile relay to the ground base station into multiple control periods with an aggregated compressed control period T;

[0137] The acquisition module 802 is configured to acquire the wireless channel state parameters and the relay data packet buffer queue state at the beginning of each control cycle;

[0138] The calculation module 803 is configured to determine an aggregation compression action according to the wireless channel state parameter and the relay data packet buffer queue state;

[0139] The compression module 804 is configured to perform a data packet aggregation compression operation and send the data packet to the ground base station according to the aggregation compression action within the control period;

[0140] The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

[0141] In some embodiments, the wireless channel state parameter includes: the channel quality δ of the channel within a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR;

[0142] The relay data packet buffer queue state includes: the number of relay buffered data packets at the beginning of the control period δ M [t], the number of relay packets arriving at the end of the control period δ N [t], and whether the aggregated compressed data packet of the previous control cycle was successfully transmittedδ T [t-1], and δ T [t-1]∈{0,1};

[0143] Wherein, t represents the number of the control cycle.

[0144] In some embodiments, the calculation module 803 is further configured to:

[0145] Inputting the wireless channel state parameters and the relay data packet buffer queue state into a deep deterministic policy gradient model to obtain an aggregation compression action consisting of a data packet aggregation quantity and a compression method;

[0146] The deep deterministic policy gradient model is composed of a first neural network and a second neural network, wherein the first neural network is used to represent the data packet aggregation and compression strategy, and the second neural network is used to represent the quality of the data packet aggregation and compression strategy;

[0147] The data packet aggregation and compression strategy includes the data packet aggregation number α A [t] and compression method α C[t],α A [t] represents the number of data packets aggregated in control period t, α C [t] represents the compression method of the control period t.

[0148] In some embodiments, the data packet adaptive aggregation compression device of the present invention further includes:

[0149] The reward module 805 is configured to calculate the reward value output by the model in each control cycle according to the set reward function, and observe the network environment state after executing the control action.

[0150] In some embodiments, the channel quality δ Q [t] is based on the previous k Q The historical channel quality data of the control cycle is predicted based on the pre-trained LSTM network, and the transmission bandwidth δ B [t is based on k B The historical transmission bandwidth of the k control cycles is predicted based on the pre-trained long short-term memory network LSTM network; Q and k B The value is preset.

[0151] In some embodiments, the reward function is:

[0152] Z=βt save -∈m lost ;

[0153] Among them, Z is the reward function, β represents the impact weight of transmission delay reduction, and ∈ represents the impact weight of packet loss;

[0154] t save Indicates the latency reduced by the data packet aggregation and compression operation in one cycle;

[0155] m lost Indicates the number of packets lost due to multiple packet transmission failures and / or packet overflow in the relay buffer queue;

[0156] t save Determined by the following formula:

[0157]

[0158] m lost Determined by the following formula:

[0159] m lost =max{δ M [t]+δ N [t]-α A [t]×δ T [t]-M,0};

[0160] Among them, size H and size i are the header and body sizes of data packet i respectively;

[0161] and They are compression algorithms α C [t] compression ratio, compression speed and decompression speed;

[0162] P success [t] is the probability of successful transmission of the aggregated compressed data packet in the control period t, and the P success [t]Satisfy:

[0163]

[0164] M is the capacity of the buffer queue;

[0165] α A [t] is the number of compressed data packets, α A [t]∈{1,…,M};

[0166] δ T [t] indicates whether the aggregated data packet is successfully transmitted, δ T [t]∈{0,1};

[0167] δ M [t] is the number of packets cached by the relay at the beginning of control period t;

[0168] i is the number of the data packet;

[0169] erfc represents the complementary error function.

[0170] The storage module 806 is configured to: combine the wireless channel state parameters and the relay data packet buffer queue state in the control period into an environment state δ[t] before the control action is executed, and combine the wireless channel state parameters and the relay data packet buffer queue state after the control action is executed into an environment state δ[t+1] after the control action is executed;

[0171] The data training samples of the four-tuple (δ[t], α[t], R[t], δ[t+1]) consisting of the environment state δ[t] before the control action is executed, the aggregate compression action α[t], the reward value R[t], and the environment state δ[t+1] after the control action is executed are stored in the experience replay pool.

[0172] The updating module 807 is configured to: randomly sample data training samples from the experience replay pool, input them into the deep deterministic policy gradient model, and update the first neural network parameters and the second neural network parameters according to the loss function.

[0173] In some embodiments, the update module 807 is further configured to:

[0174] The second neural network outputs a first function value as the estimation information of the value function according to the wireless channel state parameters, the relay data packet buffer queue state and the data packet aggregation compression action in the data training sample;

[0175] The first neural network updates the first neural network parameters by using a policy gradient algorithm according to the reward value and the value function, and the second neural network updates the second neural network parameters by using a temporal difference learning TD algorithm according to the valuation information of the value function;

[0176] The value function is the action value function output by the second neural network.

[0177] It should be noted that the data packet adaptive aggregation compression device of this embodiment and the data packet adaptive aggregation compression method of the first aspect are based on the same inventive concept, solve the same technical problems, and obtain the same technical effects, and the similarities will not be repeated here.

[0178] In the third aspect, based on the same concept, the present invention proposes a data packet adaptive aggregation compression device, such as Fig. 9 As shown, it includes: a memory and one or more processors.

[0179] One or more applications are stored in the memory, and the one or more applications are suitable for being executed by the one or more processors to implement the data packet adaptive aggregation compression method described in the first aspect.

[0180] like Fig. 9 As shown, the electronic device includes: a processor 901 and a memory 902. The processor 901 and the memory 902 are connected, for example, through a bus interface.

[0181] The structure of the electronic device does not constitute a limitation on the embodiments of the present invention.

[0182] The processor 901 may be a CPU, a general-purpose processor, a DSP, an ASIC, an FPGA or other programmable logic device, a transistor logic device, a hardware component or any combination thereof. It may implement or execute various exemplary logic blocks, modules and circuits described in conjunction with the disclosure of the present invention. The processor 901 may also be a combination that implements computing functions, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0183] The bus interface may include a path to transfer information between the above components. The bus interface may be a PCI bus or an EISA bus, etc. The bus interface may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig. 9 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0184] The memory 402 may be a ROM or other type of static storage device that can store static information and instructions, a RAM or other type of dynamic storage device that can store information and instructions, or an EEPROM, a CD-ROM or other optical disk storage, an optical disk storage (including a compressed optical disk, a laser disk, an optical disk, a digital versatile disk, a Blu-ray disk, etc.), a magnetic disk storage medium or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. In a fourth aspect, the present invention proposes a computer-readable storage medium having a computer program stored thereon, which can be loaded and executed by a processor to construct the scientific and technological big data element method described in the first aspect.

[0185] In some embodiments, the wireless channel state parameter includes: the channel quality δ of the channel in a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR;

[0186] The state of the relay packet buffer queue includes: the number of packets buffered by the relay at the beginning of the control cycle δ M [t], the number of relay packets arriving at the end of the control period δ N [t], and whether the aggregated compressed data packet of the previous control cycle was successfully transmittedδ T [t-1], and δ T [t-1]∈{0,1};

[0187] Wherein, t represents the number of the control cycle.

[0188] In some embodiments, the processor 901 reads one or more applications stored in the memory, and when the processor 901 executes the application, the following is achieved:

[0189] Determining an aggregation compression action according to the wireless channel state parameter and the relay data packet buffer queue state includes:

[0190] Inputting the wireless channel state parameters and the relay data packet buffer queue state into a deep deterministic policy gradient model to obtain an aggregation compression action consisting of a data packet aggregation quantity and a compression method;

[0191] The deep deterministic policy gradient model is composed of a first neural network and a second neural network, wherein the first neural network is used to represent the data packet aggregation and compression strategy, and the second neural network is used to represent the quality of the data packet aggregation and compression strategy;

[0192] The data packet aggregation and compression strategy includes the data packet aggregation number α A [t] and compression method α C [t],α A [t] represents the number of data packets aggregated in control period t, α C [t] represents the compression method of the control period t.

[0193] In some embodiments, the processor 901 reads one or more applications stored in the memory, and when the processor 901 executes the application, the following is achieved:

[0194] Calculate the reward value output by the model in each control cycle according to the set reward function, and observe the state of the network environment after executing the control action;

[0195] The wireless channel state parameters and the relay data packet buffer queue state in the control period are combined into an environment state δ[t] before the control action is executed, and the wireless channel state parameters and the relay data packet buffer queue state after the control action is executed are combined into an environment state δ[t+1] after the control action is executed;

[0196] The data training sample of the four-tuple (δ[t], α[t], R[t], δ[t+1]) consisting of the environment state δ[t] before the control action is executed, the aggregate compression action α[t], the reward value R[t], and the environment state δ[t+1] after the control action is executed is stored in the experience replay pool;

[0197] Data training samples are randomly sampled from the experience replay pool, input into the deep deterministic policy gradient model, and the first neural network parameters and the second neural network parameters are updated according to the loss function.

[0198] In some embodiments, the channel quality δ Q [t] is based on the previous k Q The historical channel quality data of the control cycle is predicted based on the pre-trained LSTM network, and the transmission bandwidth δ B [t] is based on k BThe historical transmission bandwidth of the k control cycles is predicted based on the pre-trained long short-term memory network LSTM network; Q and k B The value is preset.

[0199] In some embodiments, the reward function is:

[0200] Z=βt save -∈m lost ;

[0201] Among them, Z is the reward function, β represents the impact weight of transmission delay reduction, and ∈ represents the impact weight of packet loss;

[0202] t save Indicates the latency reduced by the data packet aggregation and compression operation in one cycle;

[0203] m lost Indicates the number of packets lost due to multiple packet transmission failures and / or packet overflow in the relay buffer queue;

[0204] t save Determined by the following formula:

[0205]

[0206] m lost Determined by the following formula:

[0207] m lost =max{δ M [t]+δ N [t]-α A [t]×δ T [t]-M,0};

[0208] Among them, size H and size i are the header and body sizes of data packet i respectively;

[0209] and They are compression algorithms α C [t] compression ratio, compression speed and decompression speed;

[0210] P success [t] is the probability of successful transmission of the aggregated compressed data packet in the control period t, and the P success [t]Satisfy:

[0211]

[0212] M is the capacity of the buffer queue;

[0213] α A [t] is the number of compressed data packets, α A [t]∈{1,…,M};

[0214] δ T [t] indicates whether the aggregated data packet is successfully transmitted, δ T [t]∈{0,1};

[0215] δ M [t] is the number of packets cached by the relay at the beginning of control period t;

[0216] i is the number of the data packet;

[0217] erfc represents the complementary error function.

[0218] In some embodiments, the processor 901 reads one or more applications stored in the memory, and when the processor 901 executes the application, the following is achieved:

[0219] The second neural network outputs a first function value as the estimation information of the value function according to the wireless channel state parameters, the relay data packet buffer queue state and the data packet aggregation compression action in the data training sample;

[0220] The first neural network updates the first neural network parameters by using a policy gradient algorithm according to the reward value and the value function, and the second neural network updates the second neural network parameters by using a temporal difference learning TD algorithm according to the valuation information of the value function;

[0221] The value function is the action value function output by the second neural network.

[0222] It should be noted that the data packet adaptive aggregation compression device of this embodiment and the data packet adaptive aggregation compression method of the first aspect are based on the same inventive concept, solve the same technical problems, and obtain the same technical effects, and the similarities will not be repeated here.

[0223] In a fourth aspect, the present invention proposes a computer-readable storage medium having a computer program stored thereon, wherein the computer program can be loaded and executed by a processor according to the method for adaptive aggregation and compression of data packets described in the first aspect.

[0224] The applicant of the present invention has made a detailed explanation and description of the implementation examples of the present invention in conjunction with the drawings in the specification. However, those skilled in the art should understand that the above implementation examples are only preferred implementation schemes of the present invention, and the detailed description is only to help readers better understand the spirit of the present invention, but not to limit the scope of protection of the present invention. On the contrary, any improvements or modifications based on the inventive spirit of the present invention should fall within the scope of protection of the present invention.

[0225] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the above embodiments, or replace some of the technical features therein by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or replacements that can be easily thought of by those skilled in the art within the technical scope disclosed in the present invention should be covered within the protection scope of the present invention.

Claims

1. A data packet adaptive aggregation compression method, characterized in that: include: S1, dividing the process of forwarding data packets from the mobile relay to the ground base station into multiple control cycles with an aggregated compression control cycle T; S2, at the beginning of each control cycle, obtaining wireless channel state parameters and relay data packet buffer queue state; S3, determining an aggregation compression action according to the wireless channel state parameters and the relay data packet buffer queue state; S4, within the control period, according to the aggregation and compression action, performing a data packet aggregation and compression operation and sending the data packet to a ground base station; The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

2. The method for adaptively aggregating and compressing data packets according to claim 1, characterized in that: The wireless channel state parameters and relay data packet buffer queue state in S2 include: The wireless channel state parameters include: channel quality δ of the channel within a control period Q [t]=[w t ,y t ] and transmission bandwidth δ B [t], where w t is the channel quality indicator CQI, y t is the signal-to-noise ratio SNR; The relay data packet buffer queue state includes: the number of relay buffered data packets at the beginning of the control period δ M [t], the number of relay packets arriving at the end of the control period δ N [t], and whether the aggregated compressed data packet of the previous control cycle was successfully transmittedδ T [t-1], and δ T [t-1]∈{0,1}; Here, t represents the number of the control cycle.

3. The method for adaptively aggregating and compressing data packets according to claim 2, characterized in that: The determining of the aggregation compression action according to the wireless channel state parameter and the relay data packet buffer queue state comprises: Inputting the wireless channel state parameters and the relay data packet buffer queue state into a deep deterministic policy gradient model to obtain an aggregation compression action consisting of a data packet aggregation quantity and a compression method; The deep deterministic policy gradient model is composed of a first neural network and a second neural network, wherein the first neural network is used to represent the data packet aggregation and compression strategy, and the second neural network is used to represent the quality of the data packet aggregation and compression strategy; The data packet aggregation and compression strategy includes the data packet aggregation number α A [t] and compression method α C [t], α A [t] represents the number of data packets aggregated in control period t, α C [t] represents the compression method of the control period t.

4. The method for adaptively aggregating and compressing data packets according to claim 3, characterized in that: Also includes: Calculate the reward value output by the model in each control cycle according to the set reward function, and observe the state of the network environment after executing the control action; The wireless channel state parameters and the relay data packet buffer queue state in the control period are combined into an environment state δ[t] before the control action is executed, and the wireless channel state parameters and the relay data packet buffer queue state after the control action is executed are combined into an environment state δ[t+1] after the control action is executed; The data training sample of the four-tuple (δ[t], α[t], R[t], δ[t+1]) consisting of the environment state δ[t] before the control action is executed, the aggregate compression action α[t], the reward value R[t], and the environment state δ[t+1] after the control action is executed is stored in the experience replay pool; Randomly sampling data training samples from the experience replay pool, inputting them into the deep deterministic policy gradient model, and updating the first neural network parameters and the second neural network parameters according to the loss function; Among them, α[t] represents the compression action of control period t, and R[t] represents the reward value of control period t.

5. The method for adaptively aggregating and compressing data packets according to claim 4, characterized in that: The channel quality δ Q [t] is based on the previous k Q The historical channel quality data of the control cycle is obtained based on the prediction of the pre-trained LSTM network. B[ t ] According to k B The historical transmission bandwidth of each control cycle is predicted based on the pre-trained long short-term memory network LSTM network; Among them, the k Q and k B The value is preset.

6. The method for adaptively aggregating and compressing data packets according to claim 4, characterized in that: The reward function is: Z=βt save -εm lost ; Where Z is the reward function, β represents the impact weight of transmission delay reduction, and ε represents the impact weight of packet loss; t save Indicates the latency reduced by the data packet aggregation and compression operation in one cycle; m lost Indicates the number of packets lost due to multiple packet transmission failures and / or packet overflow in the relay buffer queue; t save Determined by the following formula: m lost Determined by the following formula: m lost =max{δ M [ t ] +d N [ t ] -a A [ t ] ×d T [ t ] -M,0}; Among them, size H and size i are the header and body sizes of data packet i respectively; and They are compression algorithms α C [t] compression ratio, compression speed and decompression speed; P success[ t ] is the probability of successful transmission of the aggregate compressed data packet in the control period t, and the P success[ t ] satisfy: M is the capacity of the buffer queue; α A[ t ] is the number of compressed data packets, α A[ t ] ∈{1,…,M}; δ T[ t ] Indicates whether the aggregated data packet is successfully transmitted, δ T[ t ] ∈{0,1}; δ M[ t ] is the number of packets cached by the relay at the beginning of control period t; i is the number of the data packet; erfc represents the complementary error function.

7. The method for adaptively aggregating and compressing data packets according to claim 6, characterized in that: Also includes: The second neural network outputs a first function value as the estimation information of the value function according to the wireless channel state parameters, the relay data packet buffer queue state and the data packet aggregation compression action in the data training sample; The first neural network updates the first neural network parameters by using a policy gradient algorithm according to the reward value and the value function, and the second neural network updates the second neural network parameters by using a temporal difference learning TD algorithm according to the valuation information of the value function; The value function is the action value function output by the second neural network.

8. A data packet adaptive aggregation compression device, characterized in that: Used to implement the data packet adaptive aggregation compression method according to any one of claims 1 to 7, the device comprises: A division module is configured to divide the process of forwarding a data packet from a mobile relay to a ground base station into a plurality of control periods with an aggregated compressed control period T; An acquisition module is configured to acquire wireless channel state parameters and relay data packet buffer queue state at the beginning of each control cycle; A calculation module, configured to determine an aggregation compression action according to the wireless channel state parameters and the relay data packet buffer queue state; A compression module, configured to perform a data packet aggregation compression operation and send the data packet to a ground base station according to the aggregation compression action within the control period; The aggregation compression action indicates the aggregation quantity and compression method of data packets, and the value of the control period T is preset.

9. A data packet adaptive aggregation compression device, characterized in that: including a memory, a processor and a user interface; The memory is used to store computer programs; The user interface is used to interact with the user; The processor is used to read the computer program in the memory, and when the processor executes the computer program, it implements the data packet adaptive aggregation compression method as described in any one of claims 1 to 7.

10. A processor-readable storage medium, characterized in that: The processor-readable storage medium stores a computer program, and when the processor executes the computer program, the method for adaptive aggregation and compression of data packets as described in any one of claims 1 to 7 is implemented.

Images