A retransmission system and a method for link adaptation in a mobile environment

Abstract

This invention discloses a method for link adaptation in retransmission systems and mobile environments. The method includes: constructing and initializing a reinforcement learning neural network; acquiring information, including current user feedback on channel conditions and HARQ process information; monitoring whether the channel environment has switched; constructing an observation state vector based on the information; inputting the observation state vector into the reinforcement learning neural network and outputting the MCS corresponding to the current HARQ process; transmitting data packets containing the MCS to the user and receiving user feedback; determining whether the current HARQ process has ended based on the feedback information; if so, calculating the retransmission throughput reward of the current HARQ process, aligning and integrating the state, action, and reward, storing it in an experience pool, and then proceeding to subsequent steps; training the reinforcement learning neural network based on gradient descent using several empirical data points. This invention solves the problem that traditional algorithms cannot adapt to actual 5G networks and complex, ever-changing environments.

Description

Technical Field

[0001] This invention relates to the field of wireless network technology, and in particular to a retransmission system and a method for link adaptation in mobile environments. Background Technology

[0002] Fifth-generation mobile communication technology (5G) is a new generation of communication technology featuring high speed, low latency, and massive connectivity. One of its key technologies is Link Adaptation (LA). In 5G networks, base stations convert user-feedback Channel Quality Indicator (CQI) data into appropriate Mid-Segment Control Strategies (MCS) for packet transmission to adapt to changing channel environments. However, the 4-bit CQI only provides a coarse-grained estimate of downlink channel quality, has quantization errors, and is limited by periodically transmitted CQI signals. Therefore, base station assessments of the channel quickly become outdated, preventing them from providing timely channel quality estimates. Furthermore, complex retransmission mechanisms and feedback processing delays further impact the performance of their algorithms. Moreover, these problems are exacerbated when channel conditions change drastically; incorrect MCS decisions can lead to a sharp drop in throughput.

[0003] Existing methods for MCS (Multi-Channel System) decision-making mainly include traditional inner-loop and outer-loop link adaptation techniques and algorithms based on deep reinforcement learning. Traditional inner-loop link adaptation directly maps the CQI (Constant Quality Indicator) to an MCS according to a pre-prepared mapping table. However, this method suffers from quantization errors, typically choosing a conservative MCS, failing to improve throughput. Furthermore, when the channel changes drastically, the CQI becomes rapidly outdated, resulting in a significant performance degradation of the inner loop. OLLA (Optimal Link Adaptation Algorithm), a classic link adaptation algorithm, has the advantage of continuously adjusting the estimation bias based on HARQ feedback information, enabling the algorithm to adapt to dynamic channels. However, its performance is significantly affected by the algorithm parameters TargetBLER and Step, and the optimal parameter configuration varies considerably under different channels, making it unsuitable for complex real-world environments.

[0004] The link adaptation technique DRLLA, based on deep reinforcement learning, directly stores and trains upon receiving HARQ feedback. However, in real-world 5G networks, complex retransmission mechanisms and processing feedback delays severely impact the algorithm's performance. Firstly, retransmission mechanisms may cause the same HARQ process to receive multiple NACKs. Indiscriminately calculating negative rewards and storing them in the experience pool results in low agent rewards, leading to a conservative selection of the MCS. Secondly, the processing feedback delay manifests in the fact that the ACK / NACK feedback received in the next time slot is not the decision result of the current time slot's MCS. The lack of alignment between the state, reward, and action experience built in this algorithm leads to training chaos and haphazard MCS selection in real-world systems. Furthermore, when the actual environment changes (e.g., LOS / NLOS switching due to movement or occlusion), the experience pool from before the switch is no longer applicable to the new environment. Without timely adjustment of the exploration rate, the algorithm cannot fully adapt to the new environment and may converge to a suboptimal solution. In conclusion, this type of algorithm is no longer suitable for complex real-world environments. Summary of the Invention

[0005] This invention proposes a retransmission system and a link adaptation method in mobile environments to address the problem that existing algorithms are not applicable to complex real-world 5G network environments.

[0006] The present invention achieves the above objectives through the following technical solutions:

[0007] This invention provides a method for link adaptation in retransmission systems and mobile environments, comprising:

[0008] Construct and initialize the reinforcement learning neural network;

[0009] Obtain information, including current user-reported channel status information and HARQ process information;

[0010] If the monitoring channel environment changes based on the information, then the experience pool of the reinforcement learning neural network is reconstructed and the subsequent steps are performed; otherwise, the subsequent steps are performed directly.

[0011] Construct an observation state vector based on the information;

[0012] Determine whether the current HARQ process is in retransmission or is a new process. If the current HARQ process is in retransmission, the MCS of the current HARQ process is consistent with the MCS used in the first HARQ process. If the current HARQ process is a new process, input the observation state vector into the reinforcement learning neural network and output the MCS corresponding to the current HARQ process.

[0013] Transmit a data packet containing the MCS to the user and receive feedback information from the user;

[0014] Based on the feedback information, determine whether the current HARQ process has ended. If yes, calculate the retransmission throughput reward of the current HARQ process, and after aligning and integrating the status, action, and reward according to the throughput reward, store it in the experience pool and proceed to the next step. If no, continue the current HARQ process and directly proceed to the next step.

[0015] Several empirical data points are extracted from the experience pool, and the reinforcement learning neural network is trained based on the gradient descent method using these empirical data points. The trained reinforcement learning neural network is then used to output the MCS corresponding to the next HARQ process.

[0016] Specifically, constructing and initializing the neural network includes:

[0017] Construct an Actor network and a TargetActor network. The Actor network is used to output the MCS selected by the current HARQ process, and the TargetActor network is used to periodically replicate the weights of the Actor network. Both the Actor network and the TargetActor network include a gated recurrent unit and a neural network with several fully connected layers.

[0018] Initialize the Actor network and TargetActor network, and assign default weights to the Actor network and TargetActor network respectively.

[0019] Specifically, inputting the observed state vector into the reinforcement learning neural network includes:

[0020] After the observation state vector is processed by the gated recurrent unit, the value of each MCS is output by several fully connected layers.

[0021] Based on the aforementioned value combination The strategy makes an MCS decision for the HARQ process i at time t. The formula is:

[0022]

[0023] in, The action selection strategy exploration rate for the reinforcement learning neural network. For the value of MCS, The default weights of the Actor network are... Let be the current state vector, and 'a' be the action.

[0024] Specifically, the retransmission-compensated throughput reward of the HARQ process is calculated, and the state, action, and reward are aligned and integrated based on the throughput reward before being stored in the experience pool, including:

[0025] The throughput reward with retransmission discount is calculated for the HARQ process i, and the calculation formula is as follows:

[0026]

[0027] in , To calculate the discount factor for the reward, This represents the number of retransmissions for process i. At time t, the HARQ's IDi is compared with the observed state vector at the time of making the MCS decision. Save the data in a dictionary as key-value pairs, with the key being HARQIDi and the value being the observed state vector at that moment. , is represented as:

[0028]

[0029] When the HARQ process is identified as ending after n time slots, the observation state vector at the time of decision-making is retrieved from the dictionary using the IDi of this HARQ. And the reward calculated after n time slots. and the MCS index for the first decision Current observation status Form a set of aligned empirical values And store it in the experience pool;

[0030] When the HARQ process ends, record the observed state vector at that time. To update the key-value pair information corresponding to HARQ process i.

[0031] Specifically, several pieces of empirical data are extracted from the experience pool, and the reinforcement learning neural network is trained using gradient descent based on these pieces of empirical data.

[0032] N empirical data points are randomly selected from the experience pool to form a data set M. The Target network is used to compute the values ​​of each empirical data point in the data set M. TDTarget value The calculation formula is as follows:

[0033]

[0034] This indicates the discounted return on the current and future rewards of the currently selected MCS. This is used to estimate the true value of a state-action pair, where ρ is the discount factor. Indicates the next state. Indicates the next action. This represents the default weights of the Target Actor network;

[0035] The action value of each piece of empirical data is calculated using an Actor network. :

[0036]

[0037] The loss function of the reinforcement learning neural network is constructed using the gradient descent method. :

[0038] .

[0039] Specifically, monitoring whether the information monitoring channel environment has switched includes:

[0040] The channel condition information is filtered to obtain filtered channel condition information.

[0041] The channel environment is determined to switch based on the filtered channel condition information and preset conditions.

[0042] Specifically, monitoring whether the information monitoring channel environment has switched includes:

[0043] The channel condition information is filtered, and the filtering formula is as follows:

[0044]

[0045]

[0046] Where CQI is the CQI value reported by the user, and SRS is the uplink SRS signal received power. The value after CQI filtering at time t. Let be the filtered value of the uplink SRS signal received power at time t. , These are the filter coefficients;

[0047] when or When this occurs, a scene switching indicator from NLOS to LOS is triggered;

[0048] when or When this occurs, a switch indication from LOS scene to NLOS scene is triggered;

[0049] in, , , , To switch control parameters.

[0050] Specifically, reconstructing the experience pool of the reinforcement learning neural network includes:

[0051] The algorithm's experience pool is cleared, and the action selection strategy exploration rate is increased according to the preset action selection strategy exploration rate change formula, or the experience from the most recent time period is extracted to reconstruct the experience pool, and then the action selection strategy exploration rate is increased according to the preset action selection strategy exploration rate change formula.

[0052] Specifically, after outputting the MCS corresponding to the current HARQ process and before transmitting the data packet containing the MCS to the user, the exploration rate of the action selection strategy is reduced according to a preset action selection strategy exploration rate change formula.

[0053] Specifically, the preset formula for the change in the exploration rate of the action selection strategy is as follows:

[0054]

[0055] in, This represents the initial exploration rate, where t is the current time. The switching trigger time of the LOS / NLOS switching indication. It is the attenuation factor.

[0056] The beneficial effects of this invention are as follows:

[0057] This invention proposes a retransmission system and a link adaptation method in mobile environments, which improves link throughput by adaptively selecting the MCS. This invention considers the retransmission mechanisms and processing latency present in actual 5G communication networks, as well as the complexities of LOS / NLOS handover in real-world environments. By combining deep reinforcement learning techniques and adaptively selecting the MCS, it solves the problem that traditional OLLA or DRLLA algorithms cannot adapt to actual 5G networks and complex, ever-changing environments. Attached Figure Description

[0058] Figure 1 A flowchart illustrating the retransmission system and link adaptation method in a mobile environment as described in this application;

[0059] Figure 2 This is a schematic diagram of the HARQ retransmission mechanism and delay feedback in the 5G network used in this application;

[0060] Figure 3The diagrams illustrate several channel switching scenarios in the 5G mobile scenario of this application ((a) represents a full LOS channel environment, (b) represents a full NLOS channel environment, (c) represents a LOS to NLOS channel environment, and (d) represents a NLOS to LOS channel environment). Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0062] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0063] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0064] In the description of this invention, it should be understood that the terms "upper," "lower," "inner," "outer," "left," "right," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of this invention is in use, or the orientation or positional relationship commonly understood by those skilled in the art. They are only used to facilitate the description of this invention and to simplify the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0065] Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be interpreted as indicating or implying relative importance.

[0066] In the description of this invention, it should also be noted that, unless otherwise explicitly specified and limited, terms such as "set" and "connection" should be interpreted broadly. For example, "connection" 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 an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0067] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0068] like Figure 1 As shown, the present invention provides a retransmission system and a method for link adaptation in a mobile environment, comprising the following stages:

[0069] The initialization phase involves constructing the neural network for selecting the MCS and initializing the neural network structure.

[0070] During the information collection and processing phase, the base station collects CQI and HARQ information from user feedback, integrates and processes them to construct the observation state vector;

[0071] During the monitoring environment handover phase, the base station filters the periodic CQI feedback from users. When the filtered CQI meets the given conditions, it triggers the LOS / NLOS handover indication and performs related operations.

[0072] During the MCS selection phase, the observed state vector is input into the neural network, and the output of the neural network is the MCS prepared for the current HARQ process. Then, the algorithm exploration rate is reduced and data packets are transmitted to the user.

[0073] During the user unpacking phase, the user decodes the received packet and uses CRC check data to generate HARQ feedback information containing ACK / NACK, retransmission count, etc., and sends it back to the base station.

[0074] During the HARQ process monitoring phase, monitor whether the HARQ process has ended. If it has not ended, continue the HARQ retransmission process; if it has ended, proceed to the next step.

[0075] During the reward calculation and state storage phase, the retransmission discounted throughput reward of the HARQ process is calculated, and the state, action, and reward are aligned and integrated before being stored in the experience pool.

[0076] During the training phase of the neural network, batches of experience data are extracted from the experience pool, and the gradient descent method is used to train the neural network.

[0077] Specifically, the method includes the following steps:

[0078] Step 1.1: Construct two neural networks with gated recurrent units (GRUs) and several fully connected layers, designated as the Actor network and the TargetActor network, respectively. The Actor network outputs the MCS selected by the base station in the current HARQ process, while the TargetActor network periodically replicates the weights of the Actor network. During initialization, default weights are assigned to both networks. as well as ;

[0079] Step 2.1 The base station collects channel condition information fed back by users at time t. The HAQR process i contains: the MCS index used in the previous transfer of this HARQ process. The number of times the HARQ process data packets have been retransmitted. And the latest HARQ process data packet instruct The observation information ultimately forms the observation state vector upon which the algorithm makes its decision on the MCS at time t. ;

[0080] Step 2.2 Combines the most recent L observation state vectors and cyclically inputs them into the GRU layer for temporal correlation learning. Specifically, ;

[0081] Step 3.1: Filter the received power of periodically fluctuating CQI or uplink SRS signals to smooth out jitter and increase handover identification accuracy. The filtering formula is as follows:

[0082]

[0083]

[0084] Where CQI is the CQI value reported by the user, and SRS is the uplink SRS signal received power. The value after CQI filtering at time t. Let be the filtered value of the uplink SRS signal received power at time t. , These are the filter coefficients;

[0085] Step 3.2, when or When, trigger the NLOS scene to LOS scene switching indicator; when or When the LOS scene transition indicator is triggered, the time of triggering the scene transition indicator is recorded. .in , , , To switch control parameters. When the above switching indication is triggered, the experience pool is reconstructed, and the exploration rate is increased;

[0086] When the switching indication in step 3.2 is triggered, the reconstructed experience pool performs one of the following operations:

[0087] (1) Clear the algorithm's experience pool and increase the exploration rate to fully adapt to the new environment and prevent convergence to a suboptimal solution. Otherwise, proceed directly to the next step to select the MCS;

[0088] (2) Because the scene switching instruction recorded in step 2.2 of claim 4 has a certain delay, i.e. The time has actually changed, among which To indicate the average delay for scene switching, then in The experience pool already contains some experience from the new environment within the given timeframe, so the most recent experience can be extracted. The experience gained within the specified time period is used to reconstruct the experience pool. The exploration rate is then increased to fully adapt to the new environment and prevent convergence to a suboptimal solution. Otherwise, proceed directly to the next step to select the MCS.

[0089] When the switching indication in step 3.2 is triggered, the exploration rate is increased to perform the following operations:

[0090] Since neural networks no longer require extensive exploration of the environment after convergence, they can be further trained... The strategy exploration rate varies according to the following formula:

[0091]

[0092] in, This represents the initial exploration rate, where t is the current time. The switching trigger time in the monitoring environment switching in claim 4, It is the decay factor, where the initial exploration rate is generally 1 or 0.9.

[0093] Step 4.1: If the HARQ process is in the process of retransmission, keep the MCS of the retransmitted data packet consistent with the MCS used for the first time.

[0094] Step 4.2: If the HARQ process is new, the agent performs neural network inference and outputs the MCS. Specifically, it will... After being processed by the GRU layer, the value of each MCS is output by several fully connected layers, and then combined. The strategy makes an MCS decision for the HARQ process i at time t. :

[0095]

[0096] in The exploration rate is a value that allows the agent to choose actions randomly to explore the environment, thus preventing the algorithm from getting stuck in a suboptimal solution.

[0097] Step 5.1: The user performs MCS demodulation using the MCS indication in the DCI information. Next, the CRC cyclic redundancy check code evaluates the demodulation result. If the CRC result indicates successful demodulation, an ACK feedback message is generated; otherwise, a NACK message is generated and the retransmission count is recorded. Add 1. Finally, package the above information into HARQ information and wait for the uplink control channel to send it back to the base station.

[0098] Step 6.1: When the base station receives HARQ information from the user that includes ACK or has retransmitted 3 times and is still NACK, it determines that the HARQ process has ended. This also means that the HARQ process is about to transmit new data, i.e., using the MCS output by the agent, and simultaneously triggers the reward calculation and state storage phase. If the conditions are not met, it determines that the HARQ process has not ended, and the data packet is retransmitted using the MCS used during the initial data packet transmission.

[0099] Step 7.1, Calculate the reward: Calculate the throughput reward with retransmission discount for a terminated HARQ process i, as follows:

[0100]

[0101] in It is the action taken against HARQ process i at time t. This indicates the effective throughput resulting from the action decision.

[0102] Step 7.2: Record state information: Due to limitations in the user's processing capabilities, HARQ feedback generally has a delay. To avoid confusion between state, reward, and action, the HARQID in the HARQ information is used for discrimination. For example, at time t, HARQIDi is compared with the observed state vector when making the MCS decision. Save the data in a dictionary as key-value pairs, with the key being HARQIDi and the value being the observed state vector at that moment. , is represented as:

[0103]

[0104] As described in claim 7, when the HARQ process is identified to have ended after n time slots, the observation state vector at the time of decision-making is retrieved from the dictionary using the ID of this HARQ process. And the reward calculated using step (1) and the MCS index for the first decision Current observation status Form a set of aligned empirical values And store it in the experience pool.

[0105] Step 7.3, Update key-value pairs: As described in claim 7, when the HARQ process ends, record the observed state vector at this time. To update the key-value pair information corresponding to HARQ process i.

[0106] Step 8.1: Randomly select N empirical data points from the experience pool to form a data set M, and use the Target network to compute the results for each empirical data point. TDTarget value :

[0107]

[0108] This represents the discounted reward of the current selected MCS and the future reward, used to estimate the true value of a state-action pair, where ρ is the discount factor. Indicates the next state. Indicates the next action. This represents the default weights of the TargetActor network;

[0109] Step 8.2: Calculate the action value of each piece of empirical data using an Actor network. :

[0110]

[0111] Step 8.3: Minimize the loss function of the deep neural network using gradient descent. :

[0112]

[0113] like Figure 2 As shown, Figure 1 The diagram illustrates the HARQ retransmission mechanism and delay feedback in the 5G network involved in this invention. The diagram shows that the HAQR feedback has a delay of 4 time slots, and the retransmission mechanism uses the same MCS as the first transmission for retransmitted data.

[0114] like Figure 3 As shown: Figure 3 The document demonstrates several channel switching scenarios in 5G mobile applications: all LOS, all NLOS, LOS to NLOS, and NLOS to LOS. Figure 3 This represents the user's movement on a two-dimensional plane, with the unit being meters. gNB represents a 5G base station, and UE represents the user.

[0115] The reinforcement learning parameters of the reinforcement learning neural network in this embodiment of the invention are shown in Table 1 below:

[0116] Table 1

[0117] Parameter type Parameter value Discount factor 0.99 Learning rate 0.005 Batch size 64 Hidden layer dimensions 128

[0118] This invention proposes a retransmission system and a link adaptation method in mobile environments, which improves link throughput by adaptively selecting the MCS (Multi-Channel Selection). This invention considers the retransmission mechanisms and processing delays present in actual 5G communication networks, as well as the complexities of LOS / NLOS handover in real-world environments. It combines deep reinforcement learning technology to adaptively select the MCS, solving the problem that traditional OLLA or DRLLA algorithms cannot adapt to actual 5G networks and complex, ever-changing environments. This invention successfully addresses the limitations of traditional algorithms in adapting to retransmission mechanisms, processing feedback delays, and mobility channel handover through adaptive MCS selection mechanisms such as retransmission reward discounting, experience alignment, and handover indication operations.

[0119] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method of link adaptation in a retransmission system and in a mobile environment, characterized by, The method comprises the following steps: constructing and initializing a reinforcement learning neural network; obtaining information, which comprises channel condition information and HARQ process information of current user feedback; monitoring whether the channel environment is switched according to the information, and if yes, reconstructing an experience pool of the reinforcement learning neural network and then entering the subsequent step, and if no, directly entering the subsequent step; constructing an observation state vector according to the information; judging whether the current HARQ process is in retransmission or is a new process, if the current HARQ process is in retransmission, the MCS of the current HARQ process is consistent with the MCS used in the first HARQ process, and if the current HARQ process is a new process, inputting the observation state vector into the reinforcement learning neural network to output the MCS corresponding to the current HARQ process; transmitting a data packet containing the MCS to the user and receiving feedback information of the user; judging whether the current HARQ process is ended according to the feedback information, if yes, calculating a retransmission conversion throughput reward of the current HARQ process, and storing the state, action and reward aligned and integrated according to the throughput reward into the experience pool and then entering the subsequent step, and if no, continuing the current HARQ process and directly entering the subsequent step; extracting a plurality of experience data from the experience pool, and training the reinforcement learning neural network based on gradient descent method according to the plurality of experience data; constructing and initializing a neural network, comprising: ； wherein , is a reward computation discount factor, denotes the number of retransmissions of process i, denotes the effective throughput brought by the action decision, at time t, the IDi of HARQ and the observation state vector at the time of making the MCS decision is saved in a dictionary in the form of key-value pairs, with the key being the IDi of HARQ and the value being the observation state vector at this time , is denoted as: ； When the end of the HARQ process is identified after n time slots, the observation state vector at the time of decision is taken out from the dictionary using the IDi of this HARQ , and the reward calculated after n time slots , and the MCS index of the first decision , the current observation state are used to form a set of aligned experience values , and are stored in the experience pool; When the HARQ process termination is triggered, record the state vector observed at this time The key-value pair information corresponding to the HARQ process i is updated.

2. The method of link adaptation in a retransmission system and mobile environment according to claim 1, wherein, constructing an Actor network and a Target Actor network, the Actor network is used to output the MCS selected by the current HARQ process, and the Target Actor network is used to periodically copy the weights of the Actor network, and the Actor network and the Target Actor network both comprise a gated recurrent unit and a neural network with a plurality of fully connected layers; initializing the Actor network and the Target Actor network, and assigning default weights to the Actor network and the Target Actor network respectively. inputting the observation state vector into the reinforcement learning neural network, comprising:

3. The method of link adaptation in a retransmission system and mobile environment according to claim 2, wherein, inputting the observation state vector into the gated recurrent unit for processing, and outputting the value of each MCS from a plurality of fully connected layers; the action selection strategy exploration rate change formula is as follows: According to the value combination The policy makes MCS decision for HARQ process i at time t The formula is: ； wherein, is an action selection policy exploration rate for the reinforcement learning neural network, is a value for the MCS, is a network default weight for the Actor network, is a current state vector, is an action; extracting a plurality of experience data from the experience pool, and training the reinforcement learning neural network through gradient descent method according to the plurality of experience data: ； wherein, represents an initial exploration rate, t is a current time, is a switching trigger time of the LOS / NLOS switching indication, is a decay factor, and the initial exploration rate is 1 or 0.

9.

4. The method of link adaptation in a retransmission system and mobile environment according to claim 1, wherein, monitoring whether the channel environment is switched according to the information, comprising: Randomly draw N pieces of experience data from the experience pool to form a data set M, and calculate the TD Target value of each piece of experience data in the data set M by using the Target network , and the calculation formula is as follows: ​ ； a discounted return representing the immediate reward and future rewards of the currently selected MCS, to estimate the true value of a state-action pair, where is a discount factor, represents the next state, represents the next action, represents the network default weights of the Target Actor network; computing an action value for each piece of experience data using an actor network : ； applying a gradient descent method to construct a loss function of the reinforcement learning neural network : 。 5. The method of link adaptation in a retransmission system and mobile environment according to claim 1, wherein, filtering the channel condition information to obtain filtered channel condition information; judging whether the channel environment is switched based on the filtered channel condition information and a preset condition. reconstructing the experience pool of the reinforcement learning neural network, comprising:

6. The method of link adaptation in a retransmission system and mobile environment according to claim 1, wherein, emptying the algorithm experience pool, and increasing the action selection strategy exploration rate according to a preset action selection strategy exploration rate change formula, or ​ Extracting experience in a recent time period to reconstruct the experience pool, and then increasing the action selection strategy exploration rate according to a preset action selection strategy exploration rate change formula.

7. The method of link adaptation in a retransmission system and mobile environment according to claim 1, wherein, After outputting the MCS corresponding to the current HARQ process and before transmitting the data packet containing the MCS to the user, the action selection strategy exploration rate is decreased according to a preset action selection strategy exploration rate change formula.

Images