Adaptive link control method for multi-dimensional identification system

Abstract

The invention discloses a self-adaptive link control method for a multi-dimensional identification system, and belongs to the technical field of link control. The method comprises the following steps: embedding an environment sensing module in a multi-dimensional identification system to obtain network state information; introducing a reinforcement learning algorithm to carry out adaptive coding; the self-adaptive waveform is realized through discrete Fourier transform (DFT) and inverse discrete Fourier transform (IDFT); a self-adaptive waveform is realized through discrete Fourier transform (DFT) and inverse discrete Fourier transform (IDFT); and establishing a link layer frame format of the multi-dimensional identification system to realize communication link management. According to the invention, through combination of a multi-dimensional identification system and an environment perception technology, a network state and channel characteristics can be perceived in real time, and a link control strategy is dynamically adjusted; the data transmission efficiency can be dynamically optimized according to the network state and the channel condition by introducing the adaptive coding modulation and adaptive waveform technology; the data transmission time delay is obviously reduced, and the data packet disorder and packet loss rate are reduced, so that the user experience is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of link control, and particularly relates to an adaptive link control method for a multi-dimensional identification system. Background Art

[0002] With the continuous expansion of the scale of Internet users and the continuous extension of application fields, the existing Internet identification system based on TCP / IP has shown a "static" and "rigid" trend, and serious drawbacks have emerged in many aspects such as service quality, transmission efficiency, resource utilization rate, and management and control capabilities, making it difficult to support the urgent needs of the increasing complexity and diversification of communication networks and services in the future intelligent and complex network environment.

[0003] The network identification system is the foundation for the Internet to achieve information exchange and the link for data sharing. Network identifiers mark the addresses and storage locations of each node in the network and are the unique symbols of access objects in the fields of network communication and applications, representing a kind of identity symbol. Innovating the identification system can be unrestricted by the current IP network architecture, and a new network system with heterogeneous flexible perception, flexible resource scheduling, high efficiency, and scalability can be constructed according to user needs to solve the problems of the existing network in terms of address space, security, mobility, intelligence, etc. Therefore, how to design a new Internet identification system has become one of the most urgent and core research contents in the information field.

[0004] Based on this, the multi-dimensional identification system emerges as a new type of network identification architecture. The multi-dimensional identification is a network identification based on the spatio-temporal reference, including multi-dimensional attributes such as time and space, and is divided into three types: object, connection, and application, and uses diverse attribute combinations to describe four types of intelligent agents. It is precisely by introducing multi-dimensional attributes such as time and space that the intelligent agents in the network can be described and managed more efficiently. In the three-sided four-layer two-ring architecture, an identification system that realizes diverse addressing and routing is based on the multi-dimensional identification matrix to support the efficient interconnection of intelligent agents. However, there are still many deficiencies in the application of existing network communication technologies in the multi-dimensional identification system.

[0005] At present, although the multi-dimensional identification system is theoretically efficient and flexible, in practical applications, it lacks a close combination with the dynamic network environment. For example, when facing complex network topologies and dynamically changing traffic, the multi-dimensional identification system is difficult to achieve efficient link control. Moreover, most of the existing link control technologies are based on static configuration or simple dynamic adjustment and cannot real-time sense the changes in the network environment. Just like in wireless communication, although the link adaptation technology can adjust transmission parameters according to the channel state, its adaptability and flexibility are still insufficient under the multi-dimensional identification system.

[0006] Although environmental perception technology has been widely used in fields such as smart home and autonomous driving, in the field of network communication, especially in combination with link control, it is still in its infancy. Existing technologies cannot effectively utilize environmental perception information to optimize link control strategies, resulting in insufficient flexibility and adaptability of link control when facing complex network environments. In addition, network communication under the multi-dimensional identification system has high dynamics, and traditional link control methods are difficult to cope with dynamic attacks and faults in the network, leading to a decline in the reliability and stability of the network.

[0007] Therefore, how to solve the problem that when the network topology changes or traffic bursts, the link control method can adjust the strategy in a timely manner, so as to improve the flexibility, reliability and resource utilization efficiency of the network is the technical problem that the present invention wants to solve. Summary of the Invention

[0008] The purpose of the present invention is to provide an adaptive link control method for a multi-dimensional identification system to solve the problems raised in the above background technology.

[0009] The purpose of the present invention is achieved as follows: an adaptive link control method for a multi-dimensional identification system, characterized in that: the method includes the following steps:

[0010] Step S1: Embed an environmental perception module in the multi-dimensional identification system to obtain network status information;

[0011] Combine environmental information with multi-dimensional identification node information by using the environmental perception module;

[0012] Step S2: Introduce a reinforcement learning algorithm for adaptive coding to solve the problem of out-of-order packets;

[0013] Step S3: Implement an adaptive waveform through discrete Fourier transform DFT and inverse discrete Fourier transform IDFT;

[0014] Step S4: Design a traffic control method based on a smart network card to achieve traffic control;

[0015] Obtain information through the environmental perception module in step S1, and feedback the information to the upper-layer load balancing module of the smart network card, and achieve traffic control by sending signal frames;

[0016] Step S5: Establish a link layer frame format for the multi-dimensional identification system to achieve communication link management.

[0017] Preferably, the environmental perception module obtains environmental information and multi-dimensional identification node information, and the network information includes link bandwidth, transmission delay, download delay, transceiver delay, sender response delay, packet loss rate, CSI, packet type, and queue length;

[0018] The multi-dimensional identification node information includes a time identification node, a spatial dimension identification node, an identity dimension identification node, and an attribute dimension identification node. The time identification node obtains the active time of the identification node and the timeliness of the data.

[0019] The spatial dimension identification node obtains the geographical location and the network topology location. The identity dimension identification node obtains the node type and the permission level. The attribute dimension identification node obtains the resource status and the functional characteristics.

[0020] Preferably, in step S2, a reinforcement learning algorithm is constructed for adaptive coding to solve the problem of out-of-order data packets. Specifically:

[0021] Step S2-1: Construct a reinforcement learning algorithm.

[0022] Step S2-1-1: Define the path state set S t , the path state set S t is:

[0023] S t ={Bw t ,Buf t ,(N,R),T,T t ,PI t};

[0024] Among them, Bw t ={Bw t,1 ,Bw t,2 ,…,Bw t,I}, Bw t,i represents the average bandwidth size of path i at time t, Buf t represents the video buffer time size at time t; (N,R) represents the N and R selected at the previous moment; T represents the current packet transmission delay; T t ={T t,1 ,T t,2 ,…,T t,I}, among which, T t,i represents the download delay of the path at time t; PI t ={PI t,1 ,PI t,2 ,…,PI t,I}, PI t,i represents the packet loss rate of the path at time t, and I(1≤i≤I) represents the total number of links;

[0025] Step S2-1-2: Define the action set. The action is defined as: a t ={N,R};

[0026] Among them, N represents the size of the encoded packet, R represents the size of the redundant data packet, and N and R are integer values;

[0027] Step S2-1-3: Define the state transition probability. After executing action a t , the probability that the environmental state transfers from S t to S t+1 is called the state transition probability. Since the task of the reinforcement learning algorithm has the Markov property, that is, the state at the next moment is only determined by the state at the current moment, the state transition probability is expressed as:

[0028]

[0029] Step S2-1-4: Define the reward function. After executing action a t , the reinforcement learning algorithm environment will feedback a reward value r t to measure the quality of the execution of action a t . The reward function r t is defined as:

[0030]

[0031] where, log2(∑ I i＝1 Bw i ×T t,i ) represents maximizing the throughput; α represents the penalty factor for re-buffering, rebuf represents the re-buffering time; β represents the delay penalty factor, T = max(T t,i ) (1≤i≤I) represents the current packet transmission delay; λ represents the buffer penalty factor;

[0032]

[0033] where, B u1 , B u2 are the upper and lower threshold values of the buffer respectively, and θ(1≤θ≤1) is the buffer adjustment factor;

[0034] Step S2-2: Construct an adaptive coding decision based on the reinforcement learning algorithm.

[0035] Preferably, in the step S2-2, constructing an adaptive coding decision based on the reinforcement learning algorithm is specifically:

[0036] The agent learns the policy through continuous attempts, that is, in the current state S t , the probability of executing action a t is π(a t |S t ) ∈ [0, 1]; the update of the policy mainly depends on the state value function State value function that is, in the state S tUnder this circumstance, the long-term cumulative reward brought by the execution strategy, represented by the value function based on discounted rewards, is:

[0037]

[0038] Among them, γ (1 ≤ γ ≤ 1) represents the discount factor;

[0039] In the reinforcement learning algorithm, the Critic network updates the network parameter θ using the gradient descent method based on temporal difference TD, that is: c , namely:

[0040]

[0041] Among them, a represents the update step size of the Critic network, represents the TD error, and θ c represents the neural network parameter of the Critic network; θ a represents the neural network parameter of the Actor network;

[0042] In the reinforcement learning algorithm, the Actor network updates the network parameter θ using the gradient ascent method based on the advantage function, that is: a , namely:

[0043]

[0044] Among them, w represents the update step size of the Actor network; represents the advantage function, that is, the difference between the reward obtained by applying the strategy and the expected reward;

[0045] In the reinforcement learning algorithm, the Actor network is responsible for generating the strategy, that is, determining the coding packet size N and the redundant data packet size R; the Critic network evaluates the strategy generated by the Actor network and provides feedback based on the value function.

[0046] Preferably, in step S3, the adaptive waveform is realized through the discrete Fourier transform DFT and the inverse discrete Fourier transform IDFT, specifically as follows:

[0047] Step S3-1: Spectrum division and frequency classification;

[0048] Total bandwidth division: Assume that the total allocated spectrum bandwidth is C, which is divided into M sub-bands, and each sub-band is called a basic frequency f m , satisfying:

[0049]

[0050] Frequency classification: According to the channel state information CSI, each basic frequency f m is classified as "attenuated" or "non-attenuated", and the operation is as follows:

[0051] Compare the normalized amplitude of the frequency channel response |H(fm)| at the fundamental frequency f m with a predefined threshold If |H(fm)|dB ≥ α, the fundamental frequency f m is "not attenuated" by the channel; otherwise, it is called "attenuated".

[0052] Step S3-2: Perform an encoding operation using the discrete Fourier transform DFT;

[0053] Step S3-2-1: Input a data symbol vector. Assume the input data symbol vector is

[0054] Step S3-2-2: According to the frequency classification result, divide the input vector x into multiple subbands b, each subband containing L b symbols, satisfying:

[0055]

[0056] where B is the number of subbands;

[0057] Step S3-2-3: For each subband b, if the frequency in the subband is non-attenuated, perform a DFT operation on the symbols in the subband. For subband b, its DFT output is:

[0058] u b = DFT(x b );

[0059] where x b is the symbol vector within subband b, with a length of L b ; if the frequency in subband b is attenuated, then L b = 1, that is, no DFT operation is performed on this frequency, and its symbol is directly passed to the next step;

[0060] Step S3-3: Process the symbol vector using the inverse discrete Fourier transform IDFT to obtain the expression of the FAW modulation signal.

[0061] Preferably, in step S3-3, processing the symbol vector using the inverse discrete Fourier transform IDFT specifically includes:

[0062] Step S3-3-1: Combine the DFT outputs u b of all subbands into a total symbol vector u:

[0063]

[0064] Step S3-3-2: Perform IDFT processing on the combined symbol vector u to obtain the final FAW modulation signal X(n):

[0065]

[0066] where n is the time index;

[0067] The expression of the FAW modulation signal is further expanded as:

[0068]

[0069] where l(k) is the index within sub-band b(k), is the m-th symbol within sub-band b(k).

[0070] Preferably, in step S4, a traffic control method based on an intelligent network card is designed, specifically:

[0071] Step S4-1: Use the environment perception module of the intelligent network card for status perception and information feedback;

[0072] The environment perception module perceives the following information: network maximum frame size, transceiver delay measurement, sender response time monitoring, receiver internal buffer structure analysis, packet loss rate statistics, packet type identification, and queue length monitoring;

[0073] The environment perception module feeds back the perceived information to the load balancing module through the upper-layer interface of the intelligent network card;

[0074] Step S4-2: Design a traffic control signal frame and send it;

[0075] Insert a traffic control field FC, one bit (0 / 1), set to 0 under normal circumstances, and when set to 1, define the frame as a traffic control frame.

[0076] Preferably, in step S5, a link layer frame format with a multi-dimensional identification system is established, specifically:

[0077] Define the frame format, which includes the following fields:

[0078] Preamble and SFD: 7-byte preamble field, 1-byte SFD;

[0079] Destination address DST and source address SRC: Identify the sending and receiving nodes of the frame;

[0080] Traffic control field FC: Used for the transmission of traffic control signals;

[0081] Coding control field CC: Used to indicate the coding packet size N and redundancy size R;

[0082] Waveform Control Field WC: used to indicate the dynamic change of the waveform;

[0083] Length / Type Field: used to identify the length or data type of the frame;

[0084] Upper Layer Payload: stores the upper layer protocol data unit PDU;

[0085] Frame Check Sequence FCS: used for frame integrity verification.

[0086] Compared with the prior art, the present invention has the following improvements and advantages:

[0087] 1. By combining the multi-dimensional identification system and the environment perception technology, it can perceive the network state and channel characteristics in real time and dynamically adjust the link control strategy; by introducing adaptive coding modulation (ACM) and flexible adaptive waveform (FAW) technology, it can dynamically optimize the data transmission efficiency according to the network state and channel conditions; at the same time, it can significantly reduce the data transmission delay, reduce the packet out-of-order and packet loss rate, thereby improving the user experience.

[0088] 2. Compared with the traditional link control methods with static or simple dynamic adjustment, the present invention can better adapt to the complex and dynamic network environment, significantly improve the flexibility and adaptability of link control. By transforming the complex link control problem into a dynamic adjustment strategy based on environment perception, the present invention avoids the complex rules and conditional judgments in the traditional methods; by designing a new link layer frame structure and using a multi-dimensional identification intelligent network card to implement an intelligent addressing and traffic control mechanism based on the multi-dimensional identification system, it can achieve more efficient resource allocation and further improve the overall performance of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0089] Figure 1 is the overall flowchart of the method of the present invention.

[0090] Figure 2 is the schematic diagram of the adaptive coding process of the reinforcement learning algorithm.

[0091] Figure 3 is the schematic diagram of the transceiver process of the Fourier-based flexible adaptive waveform method.

[0092] Figure 4 is the schematic diagram of the frame format design. DETAILED DESCRIPTION OF THE INVENTION

[0093] The following further outlines the present invention in conjunction with the accompanying drawings.

[0094] As Figure 1 shown, an adaptive link control method for a multi-dimensional identification system, the method includes the following steps:

[0095] Step S1: Embed an environmental perception module in the multi-dimensional identification system to obtain network status information;

[0096] In the multi-dimensional identification system, each network node and link is assigned multi-dimensional identifiers, including time, space, identity, and attribute dimensions. An environmental perception module is embedded in the multi-dimensional identification system to collect information such as network status in real time;

[0097] The environmental perception module obtains environmental information and multi-dimensional identification node information. The network information includes link bandwidth, transmission delay, download delay, transceiver delay, sender response delay, packet loss rate, CSI, packet type, and queue length;

[0098] The multi-dimensional identification node information includes time identification nodes, space dimension identification nodes, identity dimension identification nodes, and attribute dimension identification nodes. The time identification nodes obtain the active time of the identification nodes and the timeliness of the data;

[0099] The space dimension identification nodes obtain the geographical location and network topology location. The identity dimension identification nodes obtain the node type and permission level. The attribute dimension identification nodes obtain the resource status and functional characteristics.

[0100] The environmental perception module combines the environmental information with the multi-dimensional identification node information to provide a more comprehensive decision-making basis for link control.

[0101] In step S2, a reinforcement learning algorithm is constructed for adaptive coding to solve the problem of out-of-order packets. Specifically:

[0102] Step S2-1: Construct a reinforcement learning algorithm;

[0103] Step S2-1-1: Define the path state set S t , the path state set S t is:

[0104] S t ={Bw t ,Buf t ,(N,R),T,T t ,PI t};

[0105] Among them, Bw t ={Bw t,1 ,Bw t,2 ,…,Bw t,I}, Bw t,i represents the average bandwidth size of path i at time t, Buf t represents the video buffer time size at time t; (N,R) represents the N and R selected at the previous moment; T represents the current packet transmission delay; T t ={T t,1 ,Tt,2 ,…,T t,I}, where T t,i represents the download delay of the path at time t; PI t ={PI t,1 , PI t,2 ,…, PI t,I}, and PI t,i represents the packet loss rate of the path at time t, and I (1≤i≤I) represents the total number of links;

[0106] Step S2-1-2: Define the action set, and the actions are defined as: a t ={N, R};

[0107] where N represents the size of the encoded packet, R represents the size of the redundant data packet, and N and R are integer values;

[0108] Step S2-1-3: Define the state transition probability. After performing the action a t , the probability that the environmental state transfers from S t to S t+1 is called the state transition probability. Since the task of the reinforcement learning algorithm has the Markov property, that is, the state at the next moment is only determined by the state at the current moment, the state transition probability is expressed as:

[0109]

[0110] Step S2-1-4: Define the reward function. After performing the action a t , the reinforcement learning algorithm environment will feedback a reward value r t to measure the quality of the execution of the action a t . The reward function r t is defined as:

[0111]

[0112] where represents maximizing the throughput; α represents the penalty factor for re-buffering, rebuf represents the re-buffering time; β represents the delay penalty factor, T = max(T t,i ) (1≤i≤I) represents the current packet transmission delay; λ represents the buffer penalty factor;

[0113]

[0114] where B u1 , B u2 are the upper and lower threshold values of the buffer respectively, and θ (1≤θ≤1) is the buffer adjustment factor;

[0115] In step S2-2, an adaptive coding decision based on a reinforcement learning algorithm is constructed, specifically as follows:

[0116] The agent learns the policy through continuous attempts, that is, in the current state S t the probability of executing action a t is π(a t |S t ) ∈ [0, 1]; The update of the policy mainly depends on the state value function State value function That is, in state S t the long-term cumulative reward brought by executing the policy, and the value function based on the discounted reward is expressed as:

[0117]

[0118] where γ (1 ≤ γ ≤ 1) represents the discount factor;

[0119] In the reinforcement learning algorithm, the Critic network updates the network parameter θ using the gradient descent method based on temporal difference TD c , that is:

[0120]

[0121] where a represents the update step size of the Critic network, represents the TD error, and θ c represents the neural network parameter of the Critic network; θ a represents the neural network parameter of the Actor network;

[0122] In the reinforcement learning algorithm, the Actor network updates the network parameter θ using the gradient ascent method based on the advantage function a , that is:

[0123]

[0124] where w represents the update step size of the Actor network; represents the advantage function, that is, the difference between the reward obtained by applying the policy and the expected reward; H(.) represents the regularization entropy value to ensure that the decision agent fully explores the environmental space, represents the coefficient of entropy, which is set to a relatively large value at the beginning, so that the parameters of the Actor network change in the direction of increasing entropy;

[0125] Algorithm input: Network parameters θ c and θ a , global iteration number m, global maximum iteration number M;

[0126] Algorithm output: Iterated network parameters θ cand θ a 。

[0127] The adaptive coding in step S2 can intelligently select the coding group size and redundancy size according to the current network condition; the sender runs a reinforcement learning algorithm according to the current network state, and the algorithm selects an action a according to the output of the neural network t , and execute action a t After that, based on the state transition probability, the network state changes from S t to S t+1 . At the same time, the environment generates a reward value r based on the defined reward function t to measure the quality of executing action a t . S t+1 and r t are fed back to the decision-making agent together for the next decision-making

[0128] In the reinforcement learning algorithm, the Actor network is responsible for generating policies, that is, determining the coding group size N and the redundant data packet size R; the Critic network evaluates the policies generated by the Actor network and provides feedback based on the value function; this update mechanism enables the reinforcement learning algorithm to converge efficiently and dynamically adapt to changes in the network condition; through continuous trial and error and feedback, the reinforcement learning module can learn the optimal coding policy, including the best coding group size and redundant data packet size; these optimized network parameters are used to guide the multi-dimensional identification intelligent network card to achieve multi-path concurrent transmission control, thereby improving the overall performance and user experience of the system

[0129] Step S3: Implement adaptive waveform through discrete Fourier transform DFT and inverse discrete Fourier transform IDFT, specifically:

[0130] Step S3-1: Spectrum division and frequency classification

[0131] Total bandwidth division: Assume that the total allocated spectrum bandwidth is C, and it is divided into M sub-bands, and each sub-band is called a basic frequency f m , satisfying:

[0132]

[0133] Frequency classification: According to the channel state information CSI, each basic frequency f m is classified as "attenuated" or "non-attenuated", and the operation is as follows:

[0134] Normalize the magnitude of the frequency channel response |H(fm)| of the basic frequency f m and compare it with the predefined threshold . If |H(fm)|dB ≥ α, then the base frequency f mThe channel is "non - attenuating", otherwise it is called "attenuating";

[0135] Step S3 - 2: Perform encoding operations using the Discrete Fourier Transform (DFT);

[0136] Step S3 - 2 - 1: Input the data symbol vector. Assume the input data symbol vector is

[0137] Step S3 - 2 - 2: According to the frequency classification result, divide the input vector x into multiple sub - bands b, and each sub - band contains L b symbols, satisfying:

[0138]

[0139] where B is the number of sub - bands;

[0140] Step S3 - 2 - 3: For each sub - band b, if the frequency in this sub - band is non - attenuating, perform DFT operations on the symbols in this sub - band. For sub - band b, its DFT output is:

[0141] u b = DFT(x b );

[0142] where x b is the symbol vector within sub - band b, with a length of L b ; if the frequency in sub - band b is attenuating, then L b = 1, that is, do not perform DFT operations on this frequency and directly pass its symbol to the next step;

[0143] Step S3 - 3: Process the symbol vector using the Inverse Discrete Fourier Transform (IDFT) to obtain the expression of the FAW modulation signal;

[0144] Step S3 - 3 - 1: Combine the DFT outputs u b of all sub - bands into a total symbol vector u:

[0145]

[0146] Step S3 - 3 - 2: Perform IDFT processing on the combined symbol vector u to obtain the final FAW modulation signal X(n):

[0147]

[0148] where n is the time index;

[0149] The expression of the FAW modulation signal is further expanded as:

[0150]

[0151] where \(l(k)\) is the index within sub - band \(b(k)\), and \(s_{m}(k)\) is the \(m\) - th symbol within sub - band \(b(k)\).

[0152] Construct a Fourier - based signal that provides good frequency localization around the attenuation frequency. The FAW signal can be effectively implemented using discrete Fourier transform (DFT) - based precoding and inverse discrete Fourier transform (IDFT) processing. The FAW modulation can be represented as adaptive precoding OFDM, where the precoding block takes channel state information (CSI) and attenuation threshold \(\alpha\) as inputs. According to the CSI information obtained in step S1, the fundamental frequencies are labeled as attenuated or non - attenuated relative to the \(\alpha\) value. The precoding applies the global DFT to adjacent non - attenuated fundamental frequencies and then applies the global IDFT to the precoded symbol vector. This results in bandwidth re - allocation, and the attenuated frequencies are well - localized in frequency, thus being isolated from other frequencies.

[0153] Step S4: Design a traffic control method based on an intelligent network card, specifically:

[0154] Step S4 - 1: Use the environmental perception module of the intelligent network card for status perception and information feedback;

[0155] The environmental perception module perceives the following information: the maximum network frame size, transceiver delay measurement, sender response time monitoring, receiver internal cache structure analysis, packet loss rate statistics, packet type identification, and queue length monitoring;

[0156] The environmental perception module feeds back the perceived information to the load - balancing module through the upper - layer interface of the intelligent network card;

[0157] Maximum network frame size: The intelligent network card detects the maximum frame size supported by the network through the network interface, records and stores this value;

[0158] Transceiver delay measurement: Using the timestamp technology, record the time when the packet enters and leaves the transceiver, and calculate the transceiver delay;

[0159] Sender response time monitoring: Record the packet sending time at the sender and the time when the receiver returns a response, and calculate the sender response time;

[0160] Receiver internal cache structure analysis: Through the cache management module, analyze the capacity and organization method of the receiver internal cache to determine the storage unit size;

[0161] Packet loss rate statistics: At the receiver, count the difference between the number of received packets and the expected number of sent packets, and calculate the packet loss rate;

[0162] Packet type identification: The intelligent network card identifies the type of the packet (unicast, multicast, or broadcast) based on the header information of the packet;

[0163] Queue length monitoring: Monitor the number of packets waiting for transmission in the network queue and record the queue length;

[0164] Information feedback: Feed back the above perceived information to the load balancing module through the upper-layer interface of the network card.

[0165] Step S4-2: Design a flow control signal frame and send it;

[0166] Insert a flow control field FC, one bit (0 / 1), set to 0 under normal circumstances, and when set to 1, define the frame as a flow control frame.

[0167] In step S4, it is necessary to obtain the necessary information sensed by the environmental perception module of the intelligent network card in step S1, and feed back this information to the load balancing module on the upper layer of the network card. Finally, flow control is achieved by sending a special signal frame.

[0168] Step S5; Establish the link layer frame format of the multi-dimensional identification system, specifically:

[0169] Define the frame format, and the frame format includes the following fields:

[0170] Preamble and SFD: The preamble field is 7 bytes, and the SFD is 1 byte;

[0171] Destination address DST and source address SRC: Identify the sending and receiving nodes of the frame;

[0172] Flow control field FC: Used for the transmission of flow control signals;

[0173] Coding control field CC: Used to indicate the coding packet size N and the redundancy size R;

[0174] Waveform control field WC: Used to indicate the dynamic change of the waveform;

[0175] Length / type field: Used to identify the length or data type of the frame;

[0176] Upper-layer payload: Store the upper-layer protocol data unit PDU;

[0177] Frame check sequence FCS: Used for frame integrity verification.

[0178] Reserve an extended field in the frame format to support the upgrade of future control methods and the integration of new functions. For example, support new multi-dimensional identification attributes or environmental perception information through the extended field.

[0179] After the above steps, the design of the link layer frame structure under the multi-dimensional identification system is completed and sent to the communication peer equipped with a multi-dimensional identification smart network card. After the parsing and identification of each field information by the multi-dimensional identification smart network card of the peer, corresponding actions are taken, and finally, the adaptive link control under the multi-dimensional identification system is realized.

[0180] An adaptive link control method for a multi-dimensional identification system, which is implemented in the link layer of the multi-dimensional identification system's simplest protocol stack. The adaptive coding and modulation, adaptive waveform technology, and traffic control technology of self-regulating cache are studied. The implementation is mainly based on the dynamic monitoring of the network environment, channel state, and signal characteristics of the multi-dimensional identification smart network card. Through the feedback of link state information, efficient and reliable communication link management is ensured in a complex and dynamic environment.

[0181] The above is only the implementation mode of the present invention and is not used to limit the present invention. For those skilled in the art, various changes and modifications can be made to the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the scope of the claims of the present invention.

Claims

1. An adaptive link control method for a multi-dimensional identification system, characterized in that: The method comprises the following steps: Step S1: embedding an environment perception module in the multi-dimensional identification system to obtain network status information; Using the environment perception module to combine environmental information with multi-dimensional identification node information; Step S2: Introduce reinforcement learning algorithm for adaptive encoding to solve the problem of data packet disorder; Step S3: realizing adaptive waveform by discrete Fourier transform DFT and inverse discrete Fourier transform IDFT; Step S4: designing a flow control method based on the smart network card to implement flow control; The environment perception module in step S1 obtains information and feeds the information back to the upper-layer load balancing module of the smart network card, and implements flow control by sending signal frames; Step S5: Establish a link layer frame format of a multi-dimensional identification system to implement communication link management.

2. The method for adaptive link control based on a multi-dimensional identification system according to claim 1, characterized in that: The environment perception module obtains environment information and multi-dimensional identification node information, and network information includes link bandwidth, transmission delay, download delay, transceiver delay, sender response delay, packet loss rate, CSI, data packet type and queue length; The multi-dimensional identification node information includes a time identification node, a space dimension identification node, an identity dimension identification node, and an attribute dimension identification node. The time identification node obtains the active time of the identification node and the timeliness of the data; The spatial dimension identification node obtains the geographical location and network topological location, the identity dimension identification node obtains the node type and authority level, and the attribute dimension identification node obtains the resource status and functional characteristics.

3. The method for adaptive link control based on a multi-dimensional identification system according to claim 1, characterized in that: In step S2, a reinforcement learning algorithm is constructed to perform adaptive coding to solve the problem of data packet disorder, specifically: Step S2-1: construct a reinforcement learning algorithm; Step S2-1-1: Define the path state set S t , path state set S t for: S t ={Bw t ,Buf t ,(N,R),T,T t ,PI t }; Among them, Bw t = {Bw t,1 ,Bw t,2 ,…,Bw t,I }, Bw t,i represents the average bandwidth of path i at time t, Buf t represents the video buffering time size at time t; (N, R) represents the N and R selected at the previous moment; T represents the current packet transmission delay; T t ={T t,1 ,T t,2 ,…,T t,I }, where T t,i represents the download delay of the path at time t; PI t = {PI t,1 ,PI t,2 ,…,PI t,I }, PI t,i represents the packet loss rate of the path at time t, and I (1≤i≤I) represents the total number of links; Step S2-1-2: Define an action set, where an action is defined as: a t ={N,R}; Where N represents the size of the coded packet, R represents the size of the redundant data packet, and N and R are integer values; Step S2-1-3: Define the state transition probability, when executing action a t After that, the environment state is changed from S t Transfer to S t+1 The probability of is called the state transition probability; since the reinforcement learning algorithm task has the Markov characteristic, that is, the state at the next moment is only determined by the state at the current moment, the state transition probability It is expressed as: Step S2-1-4: Define reward function and execute action a t After that, the reinforcement learning algorithm environment will feedback a reward value r t , to measure action a t The quality of execution, the reward function r t Defined as: in, represents the maximum throughput; α represents the rebuffering penalty factor, rebuf represents the rebuffering time; β represents the delay penalty factor, T = max(T t,i ) (1≤i≤I) represents the current packet transmission delay; λ represents the buffer penalty factor; Among them, B u1 ,B u2 are the upper and lower thresholds of the buffer respectively, and θ (1≤θ≤1) is the buffer adjustment factor; Step S2-2: Construct an adaptive coding decision based on a reinforcement learning algorithm.

4. The method for adaptive link control based on a multi-dimensional identification system according to claim 3, characterized in that: In step S2-2, an adaptive coding decision based on a reinforcement learning algorithm is constructed, specifically: The agent continuously learns strategies, that is, in the current state S t Next, perform action a t The probability of t |S t )∈[0,1]; the update of the strategy mainly depends on the state value function State Value Function That is, in state S t Under this circumstance, the long-term cumulative reward brought by the execution strategy is expressed as follows based on the discounted reward value function: Among them, γ (1≤γ≤1) represents the discount factor; In the reinforcement learning algorithm, the Critic network uses the gradient descent method based on temporal difference TD to update the network parameters θ c ,Right now: Among them, a represents the update step size of the Critic network, represents the TD error, θ c Represents the neural network parameters of the Critic network; θ a Represents the neural network parameters of the Actor network; In the reinforcement learning algorithm, the Actor network uses the gradient ascent method based on the advantage function to update the network parameters θ a ,Right now: Where w represents the update step of the Actor network; represents the advantage function, that is, the difference between the reward obtained by using the strategy and the expected reward; H(.) represents the regularized entropy value to ensure that the decision-making agent fully explores the environment space. represents the coefficient of entropy; In the reinforcement learning algorithm, the Actor network is responsible for generating strategies, that is, determining the encoding packet size N and the redundant data packet size R; the Critic network evaluates the strategies generated by the Actor network and provides feedback based on the value function.

5. The method for adaptive link control based on a multi-dimensional identification system according to claim 1, characterized in that: In step S3, the adaptive waveform is realized by discrete Fourier transform DFT and inverse discrete Fourier transform IDFT, specifically: Step S3-1: spectrum division and frequency classification; Total bandwidth division: Assuming the total allocated spectrum bandwidth is C, divide it into M sub-bands, each of which is called a basic frequency f m ,satisfy: Frequency classification: According to the channel state information CSI, each basic frequency f m Classified as "decay" or "non-decay", the actions are: The fundamental frequency f m The normalized magnitude of the frequency channel response |H(fm)| is equal to the predefined threshold For comparison, if |H(fm)|dB≥α, then the fundamental frequency f m The channel is "not attenuated", otherwise it is called "attenuated"; Step S3-2: Perform encoding operation using discrete Fourier transform DFT; Step S3-2-1: Input data symbol vector. Assume that the input data symbol vector is Step S3-2-2: According to the frequency classification result, the input vector x is divided into multiple sub-bands b, each of which contains L b symbols, satisfying: Where B is the number of subbands; Step S3-2-3: For each sub-band b, if the frequency in the sub-band is non-attenuated, a DFT operation is performed on the symbols in the sub-band. For sub-band b, the DFT output is: u b =DFT(x b ); Among them, x b is the symbol vector in subband b, with length L b ; If the frequencies in subband b are attenuated, then L b =1, that is, do not perform DFT operation on the frequency and directly pass its symbol to the next step; Step S3-3: Process the symbol vector using the inverse discrete Fourier transform IDFT to obtain an expression of the FAW modulated signal.

6. The method for adaptive link control based on a multi-dimensional identification system according to claim 5, characterized in that: In step S3-3, the symbol vector is processed by using the inverse discrete Fourier transform IDFT, specifically: Step S3-3-1: DFT output u of all sub-bands b Combined into a total symbol vector u: Step S3-3-2: Perform IDFT processing on the combined symbol vector u to obtain the final FAW modulated signal X(n): Where n is the time index; The expression of FAW modulation signal is further expanded as: where l(k) is the index within subband b(k), is the mth symbol in subband b(k).

7. The method for adaptive link control based on a multi-dimensional identification system according to claim 1, characterized in that: The flow control method based on the smart network card is designed in step S4, specifically: Step S4-1: Using the environment perception module of the smart network card to perform state perception and information feedback; The environment perception module perceives the following information: network maximum frame size, transceiver latency measurement, transmitter response time monitoring, receiver internal buffer structure analysis, packet loss rate statistics, data packet type identification, and queue length monitoring; The environment perception module feeds back the perceived information to the load balancing module through the upper interface of the smart network card; Step S4-2: Design a flow control signal frame and send it; The flow control field FC is inserted into the link layer frame, one bit (0 / 1), which is usually set to 0. When it is set to 1, the frame is defined as a flow control frame.

8. The method for adaptive link control based on a multi-dimensional identification system according to claim 1, characterized in that: The link layer frame format of the multi-dimensional identification system established in step S5 is specifically: Defines the frame format, which contains the following fields: Preamble and SFD: Preamble field 7 bytes, SFD 1 byte; Destination address DST and source address SRC: identify the sending and receiving nodes of the frame; Flow control field FC: used for transmission of flow control signals; Coding control field CC: used to indicate the coding packet size N and redundancy size R; Waveform control field WC: used to indicate the dynamic change of the waveform; Length / Type field: used to identify the length or data type of the frame; Upper layer payload: stores upper layer protocol data unit PDU; Frame Check Sequence FCS: used to check frame integrity.

Images