A D2D mode selection method for semantic communication

Abstract

The present invention provides a D2D mode selection method for semantic communication, which provides a new resource management method for D2D semantic communication. For the D2D counterpart, the main steps of this method are as follows: establishing a D2D communication model; determining a D2D communication overhead model; establishing an optimization problem; establishing an alternating optimization solution process and introducing a game model; and algorithm implementation. In this method, an overhead model for semantic communication is established, forming a problem of minimizing system overhead. Using an alternating optimization method, by introducing game theory as a research framework, the utility function is continuously and iteratively and distributedly calculated until convergence. This optimization method can be widely applied to the fields of D2D communication and semantic communication.

Description

Technical Field

[0001] The present invention belongs to the technical field of semantic communication, and in particular to a D2D mode selection method for semantic communication. Background Art

[0002] In recent years, with the rapid growth and widespread adoption of smart devices, device-to-device (D2D) communication has gained increasing attention, particularly in crowded venues such as stadiums and shopping malls. Through D2D communication, users can directly share content of mutual interest with each other in a single communication, without the need for a base station. This reduces power consumption and latency while alleviating the burden on base stations. Consequently, D2D communication improves the overall spectrum efficiency of the network, expands network capacity, and enables the network to accommodate more users. While D2D communication offers numerous benefits, it also faces several challenges. On the one hand, the surge in the number of users has led to an increasing shortage of spectrum resources, and the difficulty in resolving inter-channel interference has hindered the development of D2D. On the other hand, the development of traditional communications is currently limited, approaching the Shannon limit, and improvements in traditional communication technologies have had little effect on improving communication performance.

[0003] Fortunately, with the popularization of artificial intelligence, a communication mode that is deeply integrated with artificial intelligence technology - semantic communication, has attracted widespread attention in the industry. Semantic communication breaks through the limitations of traditional Shannon capacity. By understanding the semantics of the context of the transmitted data, the sender and receiver can greatly reduce the amount of transmitted data and save spectrum resources. Therefore, in addition to traditional D2D communication, a novel and efficient way is to combine D2D communication with semantic communication to form a semantic communication mode, that is, to use semantic communication to intelligently reason and understand the true intention of the original data between D2D pairs, and then transmit it through D2D communication. Therefore, this mode only transmits necessary information and has semantic understanding, which greatly improves resource utilization efficiency and user experience. For example, the document [Semantic Communication-Based Dynamic Resource Allocation in D2D Vehicular Networks[J].IEEE Transactions on Vehicular Technology, 2023, 72(8):10784-10796] considers the use of semantic communication in the vehicle network scenario. Vehicles can quickly exchange clear safety alerts, reducing misunderstandings and response time.

[0004] However, semantic communication is more complex than traditional communication models. Both the sender and receiver must have the same semantic knowledge base to ensure mutual understanding of the information. Furthermore, due to the addition of a semantic encoding module, semantic communication often requires more computing resources to complete semantic encoding and decoding, which incurs additional overhead. Therefore, how to achieve efficient semantic communication, how to choose the most appropriate communication model based on task requirements and individual needs, and how to address complex resource management issues remain unresolved. Summary of the Invention

[0005] The purpose of the present invention is to address the problems existing in the above-mentioned prior art and provide a semantic and wireless resource allocation solution, aiming to minimize the total communication overhead of target users in a cell and provide an efficient mode selection method.

[0006] The technical solution to achieve the purpose of the present invention is: a D2D mode selection method for semantic communication, the method comprising the following steps:

[0007] Step 1: Establish a D2D communication model: Establish a random channel model, where the channel state changes over time and is active during D2D communication and silent otherwise.

[0008] Step 2: Determine the D2D communication overhead model: Establish the overhead model of the semantic communication mode and the traditional communication mode in D2D communication;

[0009] Step 3: Establish an optimization problem: Under the constraint of semantic communication accuracy, minimize the total communication overhead of the system by jointly optimizing the channel allocation, semantic extraction ratio, and communication mode selection of the D2D pairs.

[0010] Step 4: Establish an alternating optimization solution process and introduce a stochastic game model: Solve the optimization problem using the alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. A convex optimization method is used for the semantic information extraction subproblem, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0011] Step 5: Based on the random game model established in step 4, a random learning algorithm is constructed to jointly optimize channel allocation, semantic extraction ratio, and communication mode selection, and the Nash equilibrium point is reached through continuous iteration.

[0012] Furthermore, the D2D communication model in step 1 is specifically:

[0013] Assume that D2D pairs are randomly distributed in a given area, expressed as M is the total number of D2D pairs, each Contains two users, a sender and a receiver, where the sender is represented by e i , the receiver is represented by r i ;

[0014] Each D2D pair needs to select an adaptive channel from the resource pool for transmission. Assuming that the number of selectable channels is finite, it can be expressed as N is the total number of selectable channels; at the same time, each D2D pair has two communication modes to choose from, including semantic stream communication mode and traditional bit stream communication mode;

[0015] The state of D2D pair i is represented by on-off distribution, and each D2D pair has θ i The probability of being active is 1-θ i The probability of being in a silent state;

[0016] Through a i ∈{1,2,...,2N-1,2N} represents the transmission mode of D2D pair i, a i =2n-1 represents the sender e i The information to be transmitted is transmitted to the receiver r in the traditional bit stream communication mode on the channel n i , a i =2n means transmission on channel n in semantic stream communication mode;

[0017] The channel between each D2D pair follows large-scale fading and small-scale Rayleigh fading; the channel state information within the D2D pair is expressed as in represents the sender e in D2D pair i i With the receiver r i The distance between them, α is the large-scale fading coefficient, Indicates the fading value when the reference distance is d0 = 1m, Constantly changing over time;

[0018] Define a probability space Where Ψ represents all channel state information in the sample space, Indicates a The probability of a specific event Δ in the sample space, then Θ(Δ)=[v(Δ),h(Δ)]:Ψ→2 M ×R M , where Θ(Δ) is a random vector, Represents the activity state of the D2D pair, and the compliance probability is θ i The on-off distribution, that is, v i =1 means D2D pair i is active, v i =0 is in silent state, It represents the fading state of the channel, h i is the fading state of the D2D channel i.

[0019] Furthermore, in step 2, an overhead model under the traditional communication mode is constructed, which specifically includes:

[0020] Given an event Δ, the set of active D2D pairs is Its transmission mode set is expressed as v i represents the active D2D pair i;

[0021] Considering the transmission overhead in the traditional bitstream communication mode, the transmission delay is expressed as:

[0022]

[0023] Where, is the set of transmission modes under a given event Δ The transmission delay, is the data transmission rate of D2D pair i under a given event Δ, D i The amount of information for the transmitted data;

[0024] The energy consumption of transmission is expressed as:

[0025]

[0026] Where, is the set of transmission modes under a given event Δ transmission energy consumption;

[0027] Considering the transmission delay and energy consumption as the transmission cost, the sender e i The communication cost or total overhead when the traditional bitstream communication mode is selected is:

[0028]

[0029] Where, is the set of transmission modes under a given event Δ The transmission cost, is the overhead weight coefficient for D2D pair i in transmission mode, and

[0030] Furthermore, in step 2, an overhead model is constructed under the semantic communication mode, specifically including:

[0031] Define semantic accuracy as:

[0032]

[0033] Where S irepresents the semantic information that D2D pair i chooses to transmit, |Z(S i )| represents Z(S i ), Z(S i ) represents the semantic information S i The recovered data, s′ ij Indicates Z(S i ), Indicates that in D i middle s′ ij The number of occurrences, Indicates that in Z(S i ) in s′ ij Number of occurrences;

[0034] (2) For sender e i :

[0035] The computational time required for semantic information extraction is expressed as

[0036]

[0037] Where, Indicates that from D i Extract S i The total number of CPU cycles required, Indicates the sender e i The computing power, that is, the CPU's rotation speed in seconds;

[0038] Energy consumption is expressed as

[0039]

[0040] Where κ is a constant coefficient that measures the effective switching capacitance related to its chip architecture;

[0041] At a transmission rate of R i (a A ,Δ), the amount of transmitted semantic data is S i The delay is expressed as

[0042]

[0043] Where, For S i The amount of data bits;

[0044] The transmission energy consumption is expressed as

[0045]

[0046] Where p i Indicates the sender e in the D2D pair i The transmission power;

[0047] (3) For the receiver r i

[0048] The computational time required for semantic information extraction is expressed as

[0049]

[0050] Where, Indicates that from D i Extract S i The total number of CPU cycles required, Represents the receiver r i The computing power, that is, the CPU's rotation speed in seconds;

[0051] Energy consumption is expressed as

[0052]

[0053] The total energy and time cost of semantic communication are obtained as T i S 、

[0054]

[0055] The communication cost or total overhead required to select the semantic communication mode is:

[0056]

[0057] Where, and They represent the weight coefficients of D2D for latency and energy consumption respectively.

[0058] Furthermore, the optimization problem established in step 3 specifically includes:

[0059] for The total overhead of the two communication modes is expressed as W i (a i ,a -i ):

[0060]

[0061] Where a -i Indicates the communication mode of other active D2D pairs except D2D pair i;

[0062] The optimization problem is constructed to minimize the total communication cost of the system:

[0063]

[0064] The first constraint is used to ensure that semantic communication meets the minimum accuracy threshold requirement, the second constraint is used to ensure that the extracted semantic information is included in the semantic information set, and the third constraint is used to ensure that the channel selected by the D2D pair is within the resource pool.

[0065] Where, a=[a1,a2,...,a M ] represents the transmission mode set of all D2D pairs, represents the minimum semantic accuracy of D2D pair i, ε(D i ) represents the data D i The semantic information set obtained by extracting semantic information.

[0066] Furthermore, in step 4, a convex optimization method is used for the semantic information extraction sub-problem, specifically including:

[0067] Step 4-1: Under the premise of given channel selection and communication mode selection strategies, the original optimization problem is reformulated as:

[0068]

[0069] Step 4-2, introduce a new variable γ i To characterize the extraction rate, it is expressed as:

[0070]

[0071] Among them, γ i ∈(0,1], The amount of bit data representing the semantic information set;

[0072] Step 4-3, and Respectively expressed as:

[0073]

[0074] Where, and The sender e i , receiver r i From D i Extract γ i The total number of CPU cycles required; K i,1 , K i,2 , K i,3 , K i,4 , K i,5are all constants, and K i,1 >0,K i,2 ∈(0,1), K i,3 >0,K i,4 >0 and K i,5 >0;

[0075] Step 4-4, based on step 4-3, express the communication overhead of the semantic communication mode as:

[0076]

[0077] Steps 4-5, express the optimization problem as a convex optimization problem:

[0078]

[0079] Where, To minimize the semantic information extraction threshold to meet the accuracy requirement;

[0080] The convex optimization problem is solved by the derivative method.

[0081] Furthermore, in step 4, the D2D pair strategy selection sub-problem is modeled as a stochastic game problem, which specifically includes:

[0082] Steps 4-6, in static case, given probability space Considering an event Δ and the transmission strategy set of the D2D pair, i.e., the communication mode set, the communication overhead of the D2D pair is regarded as a utility function, which is expressed as:

[0083]

[0084] Where W i B and W i S are the communication costs of the traditional communication mode and the semantic communication mode respectively;

[0085] Steps 4-7, construct the game model as follows:

[0086]

[0087] Steps 4-8, define:

[0088]

[0089] Where p j is the D2D centering sender e j The transmission power;

[0090] Step 4-9: Based on steps 4-6 and 4-8, construct a new utility function U i (a i ,a-i ,Δ) to express the utility value of the D2D pair:

[0091]

[0092] in,

[0093]

[0094] Step 4-10: Based on steps 4-7 and 4-9, a new game model is constructed as follows:

[0095]

[0096] Furthermore, in the dynamic case, the utility function in steps 4-9 is redefined as:

[0097]

[0098] in,

[0099]

[0100] Where, represents the expected value of D2D gain for channel i, represents the expected value of the utility function;

[0101] The new game model is constructed as follows:

[0102]

[0103] Furthermore, the random learning algorithm for jointly optimizing channel allocation, semantic extraction ratio, and communication mode selection in step 5 is specifically:

[0104] Input: Set iteration index t = 0, each D2D pair Set average distribution As the initial probability distribution of strategy selection, initialize the semantic extraction ratio Where N' is the number of D2D pairs;

[0105] Cycle the following process:

[0106] (1) Update semantic extraction ratio:

[0107] if Given a strategy selection probability vector Next, solve the semantic information extraction sub-problem and obtain the optimal solution Right now:

[0108]

[0109] (2) Update D2D pair strategy selection

[0110] if D2D pairs are classified according to the probability vector Generate strategy choices Calculate instantaneous gain and evaluate utility function The following situations apply: Then D2D interference to calculation (H Interf ) (t) Otherwise, calculate

[0111] (3) Update the transmission strategy vector

[0112] if Then the D2D pair updates the transmission strategy probability vector according to the following rules:

[0113]

[0114] Where, 0<s<1 is the learning step size, is a 2N-dimensional unit vector, is the reward value received, ξ i is the scale factor;

[0115] Until all D2D pairs no longer change their strategies or the number of iterations reaches the maximum;

[0116] Output: semantic extraction ratio and the probability vector

[0117] In another aspect, a D2D mode selection system for semantic communication is provided, the system comprising:

[0118] The first module is used to establish a D2D communication model: a random channel model is established, in which the channel state changes over time and is active during D2D communication and silent otherwise.

[0119] The second module is used to determine the D2D communication overhead model: establishing overhead models for semantic communication mode and traditional communication mode in D2D communication. The D2D communication overhead model is based on the analysis of energy consumption and latency during communication. In particular, in the semantic communication mode, the communication overhead has a strong correlation with the semantic extraction ratio.

[0120] The third module is used to formulate an optimization problem: Under the constraint of semantic communication accuracy, the total communication overhead of the system is minimized by jointly optimizing the channel allocation, semantic extraction ratio and communication mode selection of D2D pairs;

[0121] The fourth module is used to establish an alternating optimization solution process and introduce a stochastic game model. This module solves the optimization problem using an alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. The semantic information extraction subproblem uses a convex optimization method, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0122] The fifth module is used to construct a random learning algorithm that jointly optimizes channel allocation, semantic extraction ratio and communication mode selection based on the random game model established in the fourth module, and reaches the Nash equilibrium point through continuous iteration.

[0123] Compared with the prior art, the present invention has the following significant advantages:

[0124] (1) The present invention solves the problem of D2D distributed resource management for semantic communication by utilizing a random game model and a random learning algorithm. By optimizing the semantic compression ratio, communication mode selection, and channel allocation, the cost of the entire system is minimized.

[0125] (2) The present invention considers more practical scenarios in scenario selection, taking into account the complex and changeable wireless transmission environment and the dynamic D2D activity issues. It starts with static analysis and then transitions to dynamic analysis, step by step, solving the problem of complex dynamic optimization analysis.

[0126] (3) On the one hand, the algorithm proposed in the present invention, which is based on a distributed perspective, solves the problem of complex information interaction in traditional centralized resource management algorithms; on the other hand, it innovates the content of resource optimization, and solves the resource management problem in complex scenarios where semantic communication and traditional communication coexist.

[0127] The present invention is further described in detail below with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0128] Figure 1 This is a flow chart of the D2D mode selection method for semantic communication according to the present invention.

[0129] Figure 2 Schematic diagram of the D2D mode selection model for semantic communication.

[0130] Figure 3 Schematic diagram of the performance convergence of the algorithm proposed in an embodiment.

[0131] Figure 4 FIG. 4 is a schematic diagram of convergence of strategy selection in an embodiment.

[0132] Figure 5 FIG. 1 is a schematic diagram showing how the total system overhead varies with the number of channels in an embodiment.

[0133] Figure 6 FIG. 4 is a schematic diagram showing how the total system overhead varies with the number of D2D pairs in an embodiment. DETAILED DESCRIPTION

[0134] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0135] It should be noted that if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only for descriptive purposes and cannot be understood as indicating or implying their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of such features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in this field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0136] The present invention provides a D2D mode selection method for semantic communication, which is used to solve the mode selection problem of semantic communication under dynamic conditions in D2D communication. Specifically, the method takes into account the scenario of D2D communication, where D2D completes the pairing work in advance and has two communication modes, namely, semantic communication mode and traditional communication mode. Considering the energy consumption overhead of the D2D pair and the delay overhead of data transmission as the total communication overhead, the total overhead of the system is minimized by optimizing the semantic extraction ratio, channel allocation and mode selection during the communication process. The present invention proposes a distributed method, which converts the original problem into a random game problem. The analysis shows that the game is a weighted potential energy game, and there is at least one Nash equilibrium point. The Nash equilibrium point is found through a double-layer random learning algorithm.

[0137] Combine Figure 1 In one embodiment, a D2D mode selection method for semantic communication is provided, the method comprising the following steps:

[0138] Step 1: Establish a D2D communication model: Establish a random channel model, where the channel state changes over time and is active during D2D communication and silent otherwise.

[0139] Step 2: Determine the D2D communication overhead model: Build overhead models for semantic communication and traditional D2D communication. This model is based on an analysis of energy consumption and latency during communication. In particular, in the semantic communication mode, communication overhead is strongly correlated with the semantic extraction ratio.

[0140] Step 3: Establish an optimization problem: Under the constraint of semantic communication accuracy, minimize the total communication overhead of the system by jointly optimizing the channel allocation, semantic extraction ratio, and communication mode selection of the D2D pairs.

[0141] Step 4: Establish an alternating optimization solution process and introduce a stochastic game model: Solve the optimization problem using the alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. A convex optimization method is used for the semantic information extraction subproblem, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0142] Step 5: Based on the random game model established in step 4, a random learning algorithm is constructed to jointly optimize channel allocation, semantic extraction ratio, and communication mode selection, and the Nash equilibrium point is reached through continuous iteration.

[0143] Furthermore, in one embodiment, the D2D communication model in step 1 is specifically:

[0144] Assume that D2D pairs (DP) are randomly distributed in a given area, expressed as M is the total number of D2D pairs, each Contains two users, a sender and a receiver, where the sender is represented by e i , the receiver is represented by r i ;

[0145] Each D2D pair needs to select an adaptive channel from the resource pool for transmission. Assuming that the number of selectable channels is finite, it can be expressed as N is the total number of selectable channels; at the same time, each D2D pair has two communication modes to choose from, including semantic stream communication mode and traditional bit stream communication mode; it should be noted that the amount of information transmitted is D i , considering D i Large values ​​usually require transmission over several time slots.

[0146] In the proposed model, a general case is considered where the transmission in each D2D pair is sporadic, which means that the sender may be dynamically active or inactive in different time periods. Specifically, if the transmitter has data to transmit, it is in the active state. If not, the transmitter is in the inactive state. For the specific state representation, the state of D2D pair i is represented by the on-off distribution, and each D2D pair has θ i The probability of being active is 1-θ i The probability of being in a silent state;

[0147] Through a i ∈{1,2,...,2N-1,2N} represents the transmission mode of D2D pair i, a i =2n-1 represents the sender e i The information to be transmitted is transmitted to the receiver r in the traditional bit stream communication mode on the channel n i , a i =2n means transmission on channel n in semantic stream communication mode;

[0148] A more practical model is considered to describe the time-varying wireless channel. Rayleigh fading is a realistic and widely adopted D2D channel model. Therefore, in the present invention, the channel between each D2D pair follows large-scale fading and small-scale Rayleigh fading; the channel state information within the D2D pair is expressed as in represents the sender e in D2D pair i i With the receiver r i The distance between them, α is the large-scale fading coefficient, It represents the fading value when the reference distance is d0=1m, considering the dynamic environment. Constantly changing over time;

[0149] Define a probability space Where Ψ represents all channel state information in the sample space, Indicates a The probability of a specific event Δ in the sample space, then Θ(Δ)=[v(Δ),h(Δ)]:Ψ→2 M ×R M , where Θ(Δ) is a random vector, Represents the activity state of the D2D pair, and the compliance probability is θ i The on-off distribution, that is, v i =1 means D2D pair i is active, v i =0 is in silent state, It represents the fading state of the channel, h i is the fading state of the D2D channel i.

[0150] Furthermore, in one embodiment, constructing an overhead model in a traditional communication mode in step 2 specifically includes:

[0151] Given an event Δ, the set of active D2D pairs is Its transmission mode set is expressed as v i represents the active D2D pair i; therefore, the data transmission rate of D2D pair i can be obtained as:

[0152]

[0153] Where B represents the transmission bandwidth, N0 is the noise density, and p i is the D2D transmit power of sender i. Represents the receiver r i It can be seen that if too many D2D pairs choose the same channel for transmission, it will cause greater co-channel interference, resulting in a decrease in transmission rate. Therefore, a suitable channel selection strategy is an important issue.

[0154] Considering the transmission overhead in the traditional bitstream communication mode, the transmission delay is expressed as:

[0155]

[0156] Where, is the set of transmission modes under a given event Δ The transmission delay, is the data transmission rate of D2D pair i under a given event Δ, D i The amount of information for the transmitted data;

[0157] The energy consumption of traditional communication mode is mainly concentrated in transmission consumption, so the energy consumption of transmission is expressed as:

[0158]

[0159] Where, is the set of transmission modes under a given event Δ transmission energy consumption;

[0160] Considering the transmission delay and energy consumption as the transmission cost, the sender e i The communication cost or total overhead when the traditional bitstream communication mode is selected is:

[0161]

[0162] Where, is the set of transmission modes under a given event Δ The transmission cost, is the overhead weight coefficient for D2D pair i in transmission mode, and The values ​​of these two factors depend on the state and requirements of D2D communication. Specifically, when a D2D pair is in a low-energy state, the energy consumption weighting coefficient can be increased to achieve a more energy-efficient strategy. When a D2D pair requires a low-latency experience, the latency weighting coefficient can be increased to meet this requirement. Therefore, the model of this invention has strong applicability in practice.

[0163] Traditional communication methods convert all information into bitstreams for transmission. In contrast, semantic communication extracts small amounts of semantic information from the bitstream and restores the original data at the receiving end. While traditional communication is direct and concise, semantic communication requires conversion, extraction, and restoration, consuming additional energy and time. Therefore, under conditions with relatively good signal-to-noise ratios, traditional communication methods often outperform semantic communication. Conversely, under conditions of deteriorating signal-to-noise ratios, semantic communication often offers unexpected benefits.

[0164] Furthermore, in one of the embodiments, in step 2, a probability directional graph method is used to represent the transmitted information. A vertex represents a semantic entity in the directional probability graph, and the connection between any two vertices represents a probability. In addition, each vertex represents a semantic entity with a different semantic level. The higher the semantic level, the more complex the semantic information. In order to construct the directional probability graph, a deep neural network is needed to train the knowledge base. First, the semantic entities are identified from the data set. Secondly, a convolutional neural network is used to calculate the probability between the two vertices. Finally, semantic information fusion is performed. In this way, semantic communication of small-sized information can be obtained. The extraction process includes two parts. Specifically, the directional probability graph is used to extract semantic information and obtain the output ε(D i ), for efficient transmission, select a part of the information S i For transmission, the last step, the receiver r i With a common direction probability map, the original information can be restored, which is expressed as

[0165] In step 2, the cost model for semantic communication mode is constructed, which specifically includes:

[0166] Define semantic accuracy as:

[0167]

[0168] Where S i represents the semantic information that D2D pair i chooses to transmit, |Z(S i )| represents Z(S i), Z(S i ) represents the semantic information S i The recovered data, s′ ij Indicates Z(S i ), Indicates that in D i Middle S i ' j The number of occurrences, Indicates that in Z(S i ) in s′ ij Number of occurrences;

[0169] When D2D pair i chooses to transmit in the form of semantic communication, the first step is the process of extracting semantic information;

[0170] (1) For sender e i :

[0171] The computational time required for semantic information extraction is expressed as

[0172]

[0173] Where, Indicates that from D i Extract S i The total number of CPU cycles required, Indicates the sender e i The computing power, that is, the CPU's rotation speed in seconds;

[0174] Therefore, the energy consumption is expressed as

[0175]

[0176] Where k is a constant coefficient that measures the effective switching capacitance associated with its chip architecture;

[0177] At a transmission rate of R i (a A ,Δ), the amount of transmitted semantic data is S i The delay is expressed as

[0178]

[0179] Where, For S i The amount of data bits;

[0180] The transmission energy consumption is expressed as

[0181]

[0182] Where p i Indicates the sender e in the D2D pair i The transmission power;

[0183] (2) For the receiver r i

[0184] The computational time required for semantic information extraction is expressed as

[0185]

[0186] Where, Indicates that from D i Extract S i The total number of CPU cycles required, Represents the receiver r i The computing power, that is, the CPU's rotation speed in seconds;

[0187] Energy consumption is expressed as

[0188]

[0189] The total energy and time cost of semantic communication are obtained as T i S 、

[0190]

[0191] The communication cost or total overhead required to select the semantic communication mode is:

[0192]

[0193] Where, and They represent the weight coefficients of D2D for latency and energy consumption respectively.

[0194] Furthermore, in one embodiment, the step 3 of establishing the optimization problem specifically includes:

[0195] for The total overhead of the two communication modes is expressed as W i (a i ,a -i ):

[0196]

[0197] Where a -irepresents the communication mode of all active D2D pairs other than D2D pair i. As can be seen from the above equation, the transmission overhead of D2D communication depends not only on its transmission strategy but also on the mode selection of all other active D2D pairs. From a system perspective, the goal is to minimize the total transmission overhead of all D2D pairs in the entire system by considering the transmission mode selection, channel allocation, and semantic information extraction ratio of D2D pairs.

[0198] The optimization problem is constructed to minimize the total communication cost of the system:

[0199]

[0200] Where, a=[a1,a2,...,a M ] represents the transmission mode set of all D2D pairs, represents the minimum semantic accuracy of D2D pair i, ε(D i ) represents the data D i The semantic information set obtained by extracting semantic information.

[0201] Among them, the first constraint is used to ensure that semantic communication meets the minimum accuracy threshold requirement, the second constraint is used to ensure that the extracted semantic information is in the semantic information set, and the third constraint is used to ensure that the channel selection of the D2D pair is within the resource pool.

[0202] Compared to traditional transmission methods, semantic communication models can transmit larger amounts of data using smaller packets. However, they consume more energy and time resources. Developing optimal strategies for semantic extraction ratio, channel allocation, and transmission mode selection is a challenge that needs to be addressed. Furthermore, considering dynamic channel conditions and the activity of different D2D pairs, the access of multiple active DPs and the dynamic changes in channel conditions can lead to a decrease in transmission rate. Finally, making reasonable strategic choices under dynamic conditions is another issue that needs to be addressed. Currently, popular solutions include machine learning algorithms and convex optimization algorithms. While these methods can produce optimal solutions, they require significant computational resources and time. Furthermore, users often approach problems based on their own self-interest, meaning that even if the solution obtained is globally optimal, it may not be optimal for every D2D pair. Because the proposed problem is NP-hard with a non-convex objective function and constraints, this paper proposes a joint optimization algorithm that leverages game theory and convex optimization methods to solve it.

[0203] Specifically, in some embodiments, step 4 specifically includes:

[0204] Step 4-1: Under the premise of given channel selection and communication mode selection strategies, the original optimization problem is reformulated as:

[0205]

[0206] There are two difficulties in solving the above problem. First, the accuracy function and computational load function of semantic information extraction and conversion are not clearly expressed. Another challenge is the discrete space of variables. This leads to high complexity in finding the optimal solution.

[0207] Step 4-2, introduce a new variable γ i To characterize the extraction rate, it is expressed as:

[0208]

[0209] Among them, γ i ∈(0,1], Represents the amount of bit data in the semantic information set; it can be seen that in the most extreme case, γ i =0 and γ i = 1 has the lowest computation time. Then, considering the accuracy function, a higher extraction rate means a higher accuracy because more information is extracted from the original data. The minimum accuracy constraint is converted into

[0210]

[0211] Indicates the minimum semantic information extraction threshold that meets the accuracy requirement;

[0212] The calculation function consists of two parts. The first part involves i The number of CPU cycles required to obtain the directional probability map depends entirely on the amount of data. The second part involves extracting the content of the selected transmission from the directional probability map, which is expressed as the extraction rate.

[0213] Step 4-3, and Respectively expressed as:

[0214]

[0215] Where, and The sender e i , receiver r i From D i Extract γ i The total number of CPU cycles required for Higher extraction rates are more beneficial to the recipients of DP i Recover semantic information, but also incurs greater computational cost; K i,1 , K i,2 , K i,3 , Ki,4 , K i,5 are constants fitted by simulation experiments, and K i,1 >0,K i,2 ∈(0,1), K i,3 >0,K i,4 >0 and K i,5 >0;

[0216] Step 4-4, based on step 4-3, express the communication overhead of the semantic communication mode as:

[0217]

[0218] Steps 4-5, express the optimization problem as a convex optimization problem:

[0219]

[0220] Where, To minimize the semantic information extraction threshold to meet the accuracy requirement;

[0221] The convex optimization problem is solved by the derivative method.

[0222] Theorem 1: The optimal solution to the above problem is

[0223] Proof: Taking the derivative of the objective function, we can get

[0224]

[0225] Assumptions express The solution, note that For γ i A monotonically increasing function, so the optimal solution can be obtained by binary search

[0226] Due to the dynamic nature of D2D pair initiatives and the time-varying nature of the channel, the state and decisions of each D2D pair impact those of other pairs. Therefore, obtaining a globally optimal solution is difficult. Furthermore, each D2D pair lacks information about other pairs, rendering centralized solutions ineffective. Game theory, a common mathematical tool for solving distributed problems, can be used to analyze the mutual interference and decision-making between D2D pairs. In this section, a stochastic game is proposed to describe the transmission strategy process of D2D pairs.

[0227] Steps 4-6, in static case, given probability space Considering an event Δ and the transmission strategy set of the D2D pair, i.e., the communication mode set, the communication overhead of the D2D pair is regarded as a utility function, which is expressed as:

[0228]

[0229] Where W i B and W i S are the communication costs of the traditional communication mode and the semantic communication mode respectively;

[0230] Steps 4-7, build a game model and Expressed as:

[0231]

[0232] In steps 4-8, the mutual interference between D2D pairs has a key impact on the decision-making of D2D pairs. Therefore, we define:

[0233]

[0234] Where p j is the D2D centering sender e j The transmission power;

[0235] Next, we illustrate the effect of interference by proving Lemma 1.

[0236] Lemma 1:

[0237] Given a transmission strategy If interference satisfy Then the D2D pair tends to choose the transmission mode of semantic communication, where

[0238]

[0239] Proof: When the overhead of semantic communication is small, D2D pairs are more inclined to choose semantic communication as the transmission mode, that is, W i S (a i ,a -i ,Δ)≤W i B (a i ,a -i ,Δ) is satisfied, we can get from the definition of overhead:

[0240]

[0241] After a series of transformations, we can get:

[0242]

[0243] Thus we get:

[0244]

[0245] The proof ends.

[0246] In step 4-9, according to Lemma 1, a new utility function is proposed to express the utility value of a D2D pair, which is expressed as:

[0247]

[0248] Steps 4-10, construct a new game model:

[0249]

[0250] Next we discuss the equivalence between the two games.

[0251] Lemma 2:

[0252] In the game and In the above equation, D2D pairs have the same preference. For any event Δ∈Ψ, And a′ i ≠a i , we can get:

[0253]

[0254] prove:

[0255] The proof is based on the details presented in Lemma 1. Specifically, and Four cases are considered as follows:

[0256] ·a i ′=2n,a i =2n-1, according to the value of the utility function, we can get U i (a i ′,a -i ,Δ)=Ξ i (a i ,a -i ,Δ) and According to Lemma 1, we know that:

[0257]

[0258] ·a i ′=2n-1,a i =2n, according to the value of the utility function, we can get and U i (a i ,a -i ,Δ)=Ξ i(a i ,a -i ,Δ), according to Lemma 1, we can see that:

[0259]

[0260] ·a i ′=2n,a i =2m, according to the value of the utility function, we can get U i (a i ′,a -i ,Δ)=Ξ i (a i ′,a -i ,Δ) and U i (a i ,a -i ,Δ)=Ξ i (a i ,a -i ,Δ), according to the definition, Ξ i (a i ,a -i ,Δ) is independent of the channel selected by the D2D pair, so we can get:

[0261]

[0262] ·a i ′=2n-1,a i =2m-1, according to the value of the utility function, we can get U i (a i ′,a -i ,Δ)=Ξ i (a i ′,a -i ,Δ) and U i (a i ,a -i ,Δ)=Ξ i (a i ,a -i ,Δ), because we get the conclusion R i (a i ,a -i ,Δ) is a disturbance An increasing and decreasing function can be defined as

[0263] The results are as follows:

[0264]

[0265] Each D2D pair minimizes the independent adjustment cost function value, and then proves the game and relationship.

[0266] Theorem 2: Game and have the same Nash equilibrium solution.

[0267] prove:

[0268] Assume that Ω0 represents the game The Nash equilibrium solution set of the game The Nash equilibrium solution set of . Can get According to Lemma 2, we have:

[0269]

[0270] This means Similarly, we can get Therefore, the sets Ω0 and Ω1 are consistent.

[0271] Furthermore, the static scene is extended to all dynamic probability spaces. Specifically, the expected value of the cost function is used to replace the original one. Under dynamic conditions, for D2D pairs The utility function in steps 4-9 is redefined as:

[0272]

[0273] in,

[0274]

[0275] Where, represents the expected value of D2D gain for channel i, represents the expected value of the utility function;

[0276] The new game model is constructed as follows:

[0277]

[0278] Next, we analyze the Nash equilibrium of the proposed game. First, we discuss the Nash equilibrium game. The existence of Nash equilibrium.

[0279] Definition 1 (Nash equilibrium):

[0280] For the game If and only if a D2D pair cannot reduce its utility function by changing its strategy, it can be said that It's a game The Nash equilibrium solution of .

[0281] Definition 2 (weighted potential game):

[0282] For finite strategy game models For D2D pairs and its two strategies a i and a i ′, if there exists a function satisfy:

[0283]

[0284] in is the utility function value of the game, is a series of positive weight parameter values. For weighted potential energy games, the change in the utility function caused by the change in the D2D strategy is proportional to the change in the potential energy function Φ.

[0285] In order to prove the existence of Nash equilibrium in the communication channel and mode selection game model of multiple D2D pairs, the potential energy game method is used to solve it.

[0286] Theorem 3:

[0287] For any realization Δ∈Ψ, the game There exists at least one Nash equilibrium.

[0288] prove:

[0289] The key point of the proof is to design a potential energy function that satisfies the conditions of the weighted potential energy game. Therefore, a potential energy function is designed:

[0290]

[0291] Among them, χ {·} represents the indicator function, specifically, χ {·} =0 means that the condition {·} is wrong, χ {·} =1 means correct. Next, we discuss the case of an independent D2D pair k. The above expression is equivalent to:

[0292]

[0293] Notice Therefore, define a variable for an independent D2D pair k is expressed as:

[0294]

[0295] So far, the final form of the potential function can be derived as:

[0296]

[0297] And, the cost function can be equivalently written as:

[0298]

[0299] According to the definition of weighted potential game, for different strategies a i and a i Any D2D pair of and You can get:

[0300]

[0301] From the above formula, we can see that the change in potential function caused by the change in D2D decision is proportional to the change in the utility function, and the proportional coefficient is Therefore, based on the knowledge of weighted potential game theory, this game has at least one Nash equilibrium solution.

[0302] Theorem 4:

[0303] Game is a weighted potential game with at least one pure strategy NE. In the above equation, the optimal channel and transmission mode selection strategy constitutes the Nash equilibrium solution. The potential function is expressed as

[0304]

[0305] in is the transmission strategy of the D2D pair, represents the activity probability of D2D pair i.

[0306] prove:

[0307] The proof process is similar to the proof of Theorem 3 above, with similar transformations and derivations. Finally, we get:

[0308]

[0309] According to the definition of potential game, it can be concluded that this is a weighted coefficient There is at least one pure strategy Nash equilibrium solution.

[0310] Consider how to achieve a pure-strategy Nash equilibrium solution. First, achieving Nash equilibrium in a dynamic environment is extremely challenging. Current algorithms include spatial adaptive algorithms, optimal response, and superior response. These algorithms are all based on a large amount of interactive information while maintaining the stability of environmental information. In addition, a central processor is required to coordinate the dynamic information of each user, which not only increases the communication burden but also is inefficient. The present invention considers a more practical scenario, which takes into account the activity of the D2D pair and the channel dynamics, and therefore requires a more appropriate algorithm to achieve a Nash equilibrium solution. The D2D pair bases its strategy selection entirely on the dynamic perception of its interference and the calculation of the Ξ value, and continuously updates the strategy over time. Therefore, stochastic learning is applied to the algorithm design. During the algorithm's implementation, the D2D pair gains experience from its own behavior, selects the appropriate transmission method and trust, and ultimately reaches a state that adapts to its own behavior and achieves a pure-strategy Ξ.

[0311] Furthermore, in one embodiment, the random learning algorithm for jointly optimizing channel allocation, semantic extraction ratio, and communication mode selection in step 5 is specifically:

[0312] Input: Set iteration index t = 0, each D2D pair Set average distribution As the initial probability distribution of strategy selection, initialize the semantic extraction ratio Where N' is the number of D2D pairs;

[0313] Cycle the following process:

[0314] (1) Update semantic extraction ratio:

[0315] if Given a strategy selection probability vector Next, solve the semantic information extraction sub-problem and obtain the optimal solution Right now:

[0316]

[0317] (2) Update D2D pair strategy selection

[0318] if D2D pairs are classified according to the probability vector Generate strategy choices Calculate instantaneous gain and evaluate utility function The following situations apply: Then D2D interference to calculation (H Interf ) (t) Otherwise, calculate

[0319] (3) Update the transmission strategy vector

[0320] if Then the D2D pair updates the transmission strategy probability vector according to the following rules:

[0321]

[0322] Where, 0<s<1 is the learning step size, is a 2N-dimensional unit vector, is the reward value received, ξ i is the scale factor;

[0323] Until all D2D pairs no longer change their strategies or the number of iterations reaches the maximum;

[0324] Output: semantic extraction ratio and the probability vector

[0325] In one embodiment, a D2D mode selection system for semantic communication is provided, the system comprising:

[0326] The first module is used to establish a D2D communication model: a random channel model is established, in which the channel state changes over time and is active during D2D communication and silent otherwise.

[0327] The second module is used to determine the D2D communication overhead model: establishing overhead models for semantic communication mode and traditional communication mode in D2D communication. The D2D communication overhead model is based on the analysis of energy consumption and latency during communication. In particular, in the semantic communication mode, the communication overhead has a strong correlation with the semantic extraction ratio.

[0328] The third module is used to formulate an optimization problem: Under the constraint of semantic communication accuracy, the total communication overhead of the system is minimized by jointly optimizing the channel allocation, semantic extraction ratio and communication mode selection of D2D pairs;

[0329] The fourth module is used to establish an alternating optimization solution process and introduce a stochastic game model. This module solves the optimization problem using an alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. The semantic information extraction subproblem uses a convex optimization method, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0330] The fifth module is used to construct a random learning algorithm that jointly optimizes channel allocation, semantic extraction ratio and communication mode selection based on the random game model established in the fourth module, and reaches the Nash equilibrium point through continuous iteration.

[0331] For the specific definition of the D2D mode selection system for semantic communication, please refer to the definition of the D2D mode selection method for semantic communication above, which will not be repeated here. The various modules in the above-mentioned D2D mode selection system for semantic communication can be implemented in whole or in part by software, hardware and their combination. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0332] In one embodiment, a computer device is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the following is achieved:

[0333] Step 1: Establish a D2D communication model: Establish a random channel model, where the channel state changes over time and is active during D2D communication and silent otherwise.

[0334] Step 2: Determine the D2D communication overhead model: Establish the overhead model of the semantic communication mode and the traditional communication mode in D2D communication;

[0335] Step 3: Establish an optimization problem: Under the constraint of semantic communication accuracy, minimize the total communication overhead of the system by jointly optimizing the channel allocation, semantic extraction ratio, and communication mode selection of the D2D pairs.

[0336] Step 4: Establish an alternating optimization solution process and introduce a stochastic game model: Solve the optimization problem using the alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. A convex optimization method is used for the semantic information extraction subproblem, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0337] Step 5: Based on the random game model established in step 4, a random learning algorithm is constructed to jointly optimize channel allocation, semantic extraction ratio, and communication mode selection, and the Nash equilibrium point is reached through continuous iteration.

[0338] For the specific definition of each step, please refer to the definition of the D2D mode selection method for semantic communication above, which will not be repeated here.

[0339] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the computer program implements:

[0340] Step 1: Establish a D2D communication model: Establish a random channel model, where the channel state changes over time and is active during D2D communication and silent otherwise.

[0341] Step 2: Determine the D2D communication overhead model: Establish the overhead model of the semantic communication mode and the traditional communication mode in D2D communication;

[0342] Step 3: Establish an optimization problem: Under the constraint of semantic communication accuracy, minimize the total communication overhead of the system by jointly optimizing the channel allocation, semantic extraction ratio, and communication mode selection of the D2D pairs.

[0343] Step 4: Establish an alternating optimization solution process and introduce a stochastic game model: Solve the optimization problem using the alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. A convex optimization method is used for the semantic information extraction subproblem, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem.

[0344] Step 5: Based on the random game model established in step 4, a random learning algorithm is constructed to jointly optimize channel allocation, semantic extraction ratio, and communication mode selection, and the Nash equilibrium point is reached through continuous iteration.

[0345] For the specific definition of each step, please refer to the definition of the D2D mode selection method for semantic communication above, which will not be repeated here.

[0346] As a specific example, in one of the embodiments, the present invention is further verified and explained with reference to the accompanying drawings.

[0347] Figure 2 The system model diagram of the invention is shown.

[0348] Figure 3 The total transmission cost of the entire system under the proposed algorithm in a dynamic environment is shown. Here, the traditional best response algorithm is used as a benchmark because it best meets the inherent intention of the users, which is to minimize their own costs in each iteration. In each iteration, the selected user adjusts its strategy to the optimal strategy, while the other D2D pairs remain unchanged. Figure 3 As can be seen in the figure, the transmission cost of the optimal response algorithm fluctuates significantly with increasing iterations, far exceeding that of the proposed algorithm. In contrast, the proposed algorithm continuously reduces the total system cost with increasing iterations, maintaining relatively small fluctuations as the wireless environment and the activity level of D2D pairs change dynamically. This is because the proposed algorithm accumulates costs over multiple iterations, while the optimal response algorithm is based on current costs. Therefore, the proposed algorithm exhibits a certain degree of stability and superior performance.

[0349] Figure 4 The iterative process of the strategy selection probability for one D2D pair in this scheme is described. It can be observed that in the initial iterations, the D2D pair selects a strategy with equal probability. As the iterations progress, the strategy selection probability of the D2D pair is continuously updated and finally converges after 160 iterations. Furthermore, it can be seen that when active, this D2D pair chooses to send text messages using traditional communication methods on channel 2.

[0350] Figure 5 The relationship between an increase in available channels and the total system cost is shown. To better demonstrate the advantages of the proposed strategy, two schemes are compared, using two different communication methods, denoted as "semantic communication only mode" and "traditional communication only mode." The "semantic communication only mode" scheme involves all users transmitting exclusively via semantic communication, while the "traditional communication only mode" scheme involves traditional bit-based transmission. First, it can be observed that the total system cost decreases with an increase in the number of channels. This is because D2D pairs allow for more channel selection options, significantly reducing interference within the same channel, thereby reducing transmission delay and cost. Second, when the number of channels is small, the cost difference between the system using semantic communication only and the proposed scheme is small, but significantly greater than that of the system using bit-based transmission. This is because the interference caused by D2D transmission is significant when the number of channels is small. Semantic communication reduces the amount of transmitted data through semantic extraction and recovery, resulting in more significant benefits, while also resulting in significantly lower user communication costs compared to traditional bit-based transmission. Conversely, when a large number of channels are available, interference within the same channel is reduced, leading to faster D2D communication rates. The additional cost incurred for semantic extraction and recovery is relatively high, making the performance of traditional bit-based transmission better.

[0351] Figure 6 The relationship between the number of D2D pairs in the system and the total system cost is shown. As the number of D2D pairs increases, the total system cost also increases. Figure 6 As can be seen, when there are few D2D pairs, the performance of the "semantic communication mode only" scheme is comparable to that of the proposed scheme. This is because the D2D transmission rate is high, and the communication process cost is relatively small compared to the semantic processing cost. However, as the number of D2D pairs increases, the cost of the "semantic communication mode only" scheme increases rapidly. This is because the D2D communication rate decreases significantly, and even reasonable channel allocation cannot prevent the cost increase. On the other hand, the "bit communication only" scheme and the proposed scheme show relatively mild growth. By shifting the cost of the transmission process through semantic processing, the algorithm has good adaptability and robustness even in dense D2D pair scenarios.

[0352] In summary, the algorithm proposed in the present invention has certain advantages in terms of convergence and optimality. This optimization strategy can be widely applied to the scenarios of semantic communication and D2D communication, but is not limited to the scope listed above.

[0353] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only illustrative of the principles of the present invention. Without departing from the spirit and scope of the present invention, any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention shall be included in the scope of protection of the present invention.

Claims

1. A D2D mode selection method for semantic communication, characterized in that: The method comprises the following steps: Step 1: Establish a D2D communication model: Establish a random channel model, where the channel state changes over time and is active during D2D communication and silent otherwise. Step 2: Determine the D2D communication overhead model: Establish the overhead model of the semantic communication mode and the traditional communication mode in D2D communication; Step 3: Establish an optimization problem: Under the constraint of semantic communication accuracy, minimize the total communication overhead of the system by jointly optimizing the channel allocation, semantic extraction ratio, and communication mode selection of the D2D pairs. Step 4: Establish an alternating optimization solution process and introduce a stochastic game model: Solve the optimization problem using the alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. A convex optimization method is used for the semantic information extraction subproblem, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem. Step 5: Based on the random game model established in step 4, a random learning algorithm is constructed to jointly optimize channel allocation, semantic extraction ratio, and communication mode selection, and the Nash equilibrium point is reached through continuous iteration.

2. The D2D mode selection method for semantic communication according to claim 1, characterized in that: The D2D communication model described in step 1 is specifically as follows: Assume that D2D pairs are randomly distributed in a given area, expressed as M is the total number of D2D pairs, each Contains two users, a sender and a receiver, where the sender is represented by e i , the receiver is represented by r i ; Each D2D pair needs to select an adaptive channel from the resource pool for transmission. Assuming that the number of selectable channels is finite, it can be expressed as N is the total number of selectable channels; at the same time, each D2D pair has two communication modes to choose from, including semantic stream communication mode and traditional bit stream communication mode; The state of D2D pair i is represented by on-off distribution, and each D2D pair has θ i The probability of being active is 1-θ i The probability of being in a silent state; Through a i ∈{1,2,...,2N-1,2N} represents the transmission mode of D2D pair i, a i =2n-1 represents the sender e i The information to be transmitted is transmitted to the receiver r in the traditional bit stream communication mode on the channel n i , a i =2n means transmission on channel n in semantic stream communication mode; The channel between each D2D pair follows large-scale fading and small-scale Rayleigh fading; the channel state information within the D2D pair is expressed as in represents the sender e in D2D pair i i With the receiver r i The distance between them, α is the large-scale fading coefficient, Indicates the fading value when the reference distance is d0 = 1m, Constantly changing over time; Define a probability space Where Ψ represents all channel state information in the sample space, Indicates a The probability of a specific event Δ in the sample space, then Θ(Δ)=[v(Δ),h(Δ)]:Ψ→2 M ×R M , where Θ(Δ) is a random vector, Represents the activity state of the D2D pair, and the compliance probability is θ i The on-off distribution, that is, v i =1 means D2D pair i is active, v i =0 is in silent state, It represents the fading state of the channel, h i is the fading state of the D2D channel i.

3. The D2D mode selection method for semantic communication according to claim 2, characterized in that: In step 2, the overhead model under the traditional communication mode is constructed, which specifically includes: Given an event Δ, the set of active D2D pairs is Its transmission mode set is expressed as v i represents the active D2D pair i; Considering the transmission overhead in the traditional bitstream communication mode, the transmission delay is expressed as: Where, T i B (a v ,Δ) is the transmission mode set a under a given event Δ v The transmission delay, R i (a v ,Δ) is the data transmission rate of D2D pair i under a given event Δ, D i The amount of information for the transmitted data; The energy consumption of transmission is expressed as: Where, is the transmission mode set a under a given event Δ v transmission energy consumption; Considering the transmission delay and energy consumption as the transmission cost, the sender e i The communication cost or total overhead when the traditional bitstream communication mode is selected is: Where W i B (a v ,Δ) is the transmission mode set a under a given event Δ v The transmission cost, is the overhead weight coefficient for D2D pair i in transmission mode, and 4. The D2D mode selection method for semantic communication according to claim 3, characterized in that: In step 2, the cost model for semantic communication mode is constructed, which specifically includes: Define semantic accuracy as: Where S i represents the semantic information that D2D pair i chooses to transmit, |Z(S i )| represents Z(S i ), Z(S i ) represents the semantic information S i The recovered data, s′ ij Indicates Z(S i ), Indicates that in D i middle s′ ij The number of occurrences, Indicates that in Z(S i ) in s′ ij Number of occurrences; (1) For sender e i : The computational time required for semantic information extraction is expressed as Where, Indicates that from D i Extract S i The total number of CPU cycles required, Indicates the sender e i The computing power, that is, the CPU's rotation speed in seconds; Energy consumption is expressed as Where κ is a constant coefficient that measures the effective switching capacitance related to its chip architecture; At a transmission rate of R i (a A ,Δ), the amount of transmitted semantic data is S i The delay is expressed as Where, For S i The amount of data bits; The transmission energy consumption is expressed as Where p i Indicates the sender e in the D2D pair i The transmission power; (2) For the receiver r i The computational time required for semantic information extraction is expressed as Where, Indicates that from D i Extract S i The total number of CPU cycles required, Represents the receiver r i The computing power, that is, the CPU's rotation speed in seconds; Energy consumption is expressed as The total energy and time cost of semantic communication are obtained as T i S 、 The communication cost required to select the semantic communication mode is the total cost W i S (a v ,Δ) is: Where, and They represent the weight coefficients of D2D for latency and energy consumption respectively.

5. The D2D mode selection method for semantic communication according to claim 4, characterized in that: Step 3 describes establishing an optimization problem, specifically including: for The total overhead of the two communication modes is expressed as W i (a i ,a -i ): Where a -i Indicates the communication mode of other active D2D pairs except D2D pair i; The optimization problem is constructed to minimize the total communication cost of the system: The first constraint is used to ensure that semantic communication meets the minimum accuracy threshold requirement, the second constraint is used to ensure that the extracted semantic information is included in the semantic information set, and the third constraint is used to ensure that the channel selected by the D2D pair is within the resource pool. Where, a=[a1,a2,...,a M ] represents the transmission mode set of all D2D pairs, represents the minimum semantic accuracy of D2D pair i, ε(D i ) represents the data D i The semantic information set obtained by extracting semantic information.

6. The D2D mode selection method for semantic communication according to claim 5, characterized in that: In step 4, the convex optimization method is used for the semantic information extraction sub-problem, which includes: Step 4-1: Under the premise of given channel selection and communication mode selection strategies, the original optimization problem is reformulated as: Step 4-2, introduce a new variable γ i To characterize the extraction rate, it is expressed as: Among them, γ i ∈(0,1], The amount of bit data representing the semantic information set; Step 4-3, and Respectively expressed as: Where, and The sender e i , receiver r i From D i Extract γ i The total number of CPU cycles required; K i,1 , K i,2 , K i,3 , K i,4 , K i,5 are all constants, and K i,1 >0,K i,2 ∈(0,1), K i,3 >0,K i,4 >0 and K i,5 >0; Step 4-4, based on step 4-3, express the communication overhead of the semantic communication mode as: Steps 4-5, express the optimization problem as a convex optimization problem: Where, To minimize the semantic information extraction threshold to meet the accuracy requirement; The convex optimization problem is solved by the derivative method.

7. The D2D mode selection method for semantic communication according to claim 6, characterized in that: In step 4, the D2D pair strategy selection sub-problem is modeled as a stochastic game problem, which specifically includes: In static condition: Steps 4-6, given the probability space Considering an event Δ and the transmission strategy set of the D2D pair, i.e., the communication mode set, the communication overhead of the D2D pair is regarded as a utility function, which is expressed as: Where W i B and W i S are the communication costs of the traditional communication mode and the semantic communication mode respectively; Steps 4-7, construct the game model as follows: Steps 4-8, define: Where p j is the D2D centering sender e j The transmission power; Step 4-9: Based on steps 4-6 and 4-8, construct a new utility function U i (a i ,a -i ,Δ) to express the utility value of the D2D pair: in, Step 4-10: Based on steps 4-7 and 4-9, a new game model is constructed as follows:

8. The D2D mode selection method for semantic communication according to claim 7, characterized in that: In the dynamic case, the utility function in steps 4-9 is redefined as: in, Where, represents the expected value of D2D gain for channel i, represents the expected value of the utility function; The new game model is constructed as follows:

9. The D2D mode selection method for semantic communication according to claim 8, characterized in that: The random learning algorithm for jointly optimizing channel allocation, semantic extraction ratio, and communication mode selection described in step 5 is specifically: Input: Set iteration index t = 0, each D2D pair Set average distribution As the initial probability distribution of strategy selection, initialize the semantic extraction ratio Where N' is the number of D2D pairs; Cycle the following process: (1) Update semantic extraction ratio: if Given a strategy selection probability vector Next, solve the semantic information extraction sub-problem and obtain the optimal solution Right now: (2) Update D2D pair strategy selection if D2D pairs are classified according to the probability vector Generate strategy choices Calculate instantaneous gain and evaluate utility function The following situations apply: Then D2D interference to calculation (H Interf ) (t) Otherwise, calculate (3) Update the transmission strategy vector if Then the D2D pair updates the transmission strategy probability vector according to the following rules: Where, 0<s<1 is the learning step size, is a 2N-dimensional unit vector, is the reward value received, ξ i is the scale factor; Until all D2D pairs no longer change their strategies or the number of iterations reaches the maximum; Output: semantic extraction ratio s i and the probability vector 10. A D2D mode selection system for semantic communication based on the method according to any one of claims 1 to 9, characterized in that: The system comprises: The first module is used to establish a D2D communication model: a random channel model is established, in which the channel state changes over time and is active during D2D communication and silent otherwise. The second module is used to determine the D2D communication overhead model: establishing the D2D communication overhead model under the semantic communication mode and the traditional communication mode. The D2D communication overhead model is based on the analysis of energy consumption and latency during the communication process. In the semantic communication mode, the communication overhead has a strong correlation with the semantic extraction ratio. The third module is used to formulate an optimization problem: Under the constraint of semantic communication accuracy, the total communication overhead of the system is minimized by jointly optimizing the channel allocation, semantic extraction ratio and communication mode selection of D2D pairs; The fourth module is used to establish an alternating optimization solution process and introduce a stochastic game model. This module solves the optimization problem using an alternating optimization method, decomposing the original optimization problem into two subproblems: a semantic information extraction subproblem and a D2D pairing strategy selection subproblem. The semantic information extraction subproblem uses a convex optimization method, and the D2D pairing strategy selection subproblem is modeled as a stochastic game problem. The fifth module is used to construct a random learning algorithm that jointly optimizes channel allocation, semantic extraction ratio and communication mode selection based on the random game model established in the fourth module, and reaches the Nash equilibrium point through continuous iteration.

Images