IRS-aided NOMA semantic transmission resource allocation method

Abstract

The application belongs to the technical field of wireless communication, and particularly relates to an IRS-assisted NOMA semantic transmission resource allocation method, which comprises the following steps: in an IRS-assisted NOMA semantic transmission model, by optimizing semantic compression rate, power allocation and phase shift at the IRS, the minimum semantic spectrum efficiency of users is maximized. Simulation results show that the application can not only ensure the fairness of information transmission between users, but also has improved transmission reliability and transmission rate compared with the prior art, and the application can accurately map based on the quality of signals and channels.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of wireless communication, and particularly relates to an IRS-aided NOMA semantic transmission resource allocation method. BACKGROUND

[0002] With the wide application of various artificial intelligence (AI) applications such as automatic driving and virtual reality, the number of connected intelligent devices is showing exponential growth. This large-scale connection and the increase in massive data traffic pose unprecedented challenges to existing communication systems. To meet these challenges, semantic communication (SC) emerges as a new communication technology, showing the potential to break through the bottlenecks of traditional communication. Traditional communication systems mainly focus on the transmission of bit streams, while the purpose of semantic communication is to transmit semantic information in the source data. Through semantic extraction and compression of information, SC technology can significantly reduce the amount of data required to be transmitted in the network. This feature benefits from the development of joint source-channel coding technology, which shows better performance than separate coding, thereby significantly improving the reliability of semantic transmission.

[0003] In the field of semantic communication, combining the advantages of non-orthogonal multiple access technology (NOMA) and intelligent reflecting surface technology (IRS), more users can be efficiently supported for connection and data transmission within the same spectrum resource, thereby realizing efficient semantic transmission. This combination not only optimizes the communication performance, but also provides a new idea for the design of future wireless communication systems.

[0004] Compared with traditional orthogonal multiple access technology (OMA), NOMA superimposes multiple user signals with different powers on the same resource block, and decodes the multi-user information stream through successive interference cancellation at the receiving end, thereby realizing power domain multiple access. NOMA technology performs well in user access. IRS integrates a large number of low-cost passive reflecting elements and intelligently reconfigures the wireless propagation environment on the plane, significantly improving the performance of wireless communication networks. The core of IRS technology lies in its ability to optimize the signal propagation path, which can dynamically adjust the direction and phase of the reflected signal to overcome the problems of signal attenuation and short transmission distance in traditional wireless communication, thereby enhancing the signal strength and transmission reliability of the system.

[0005] However, there is still a lack of effective and more efficient application scheme for semantic communication resource allocation in the IRS-NOMA scenario in the prior art. SUMMARY

[0006] In order to guarantee the communication quality of each user without difference and realize a more efficient semantic transmission system, the application provides an IRS-assisted NOMA semantic transmission resource allocation method, which comprises the following steps: in an IRS-assisted NOMA semantic transmission model, by optimizing the semantic compression rate, power allocation and phase shift at the IRS, the minimum semantic spectral efficiency of the user is maximized.

[0007] Further, the optimization problem of maximizing the minimum semantic spectral efficiency of the user is represented as:

[0008] The optimization problem is:

[0009] The constraint condition is:

[0010] 0≤θ m ≤2π,m∈M

[0011]

[0012] Wherein, α is the power allocation of all users, represented as α=[α1,…,α K ] T , α k is the power allocated to the user k, K is the number of users in the IRS-assisted NOMA semantic transmission model; Θ is the reflection matrix of the IRS; τ k is the average semantic symbol number of the word mapping of the user k, B is the maximum value of the average semantic symbol number of a single word mapping; Φ k is the semantic spectral efficiency of the user k; is the set of users; θ m is the reflection phase shift of the mth reflection element, M is the number of reflection elements; γ t→k represents the signal-to-interference-plus-noise ratio of the kth user decoding the tth data stream; P is the total power allocated by the base station to all users; is the conjugate transpose of the channel gain from the IRS to the user; f is the channel gain from the base station to the IRS; v t is the channel gain from the base station to the tth user; σ 2 is the noise power; γ t is the SINR of the tth data stream; ξ k is the semantic similarity of the user k; ξ th is the threshold of the semantic similarity.

[0013] Further, the semantic spectral efficiency of the user k is represented as:

[0014]

[0015] Wherein, ξ(τk ,γ k represents the semantic similarity of user k, I represents the semantic information amount of the text S k requested by user k to receive, and L represents the average number of words of each text.

[0016] Further, the semantic similarity of user k, ξ(τ k ,γ k ) is represented as:

[0017]

[0018] wherein, is a parameter determined using a nonlinear least square criterion.

[0019] Further, the process of solving the optimization problem includes:

[0020] The decoding order is determined by optimizing the IRS phase shift to maximize the combined channel strength of all K users that can be achieved;

[0021] For a given decoding order, the SIC decoding constraint in the optimization problem is simplified, and an optimization problem is constructed to maximize the minimum signal-to-noise ratio of all users;

[0022] The optimization problem of maximizing the minimum signal-to-noise ratio of all users is solved alternately, i.e., first fixing the power allocation coefficient to optimize the reflection phase shift, and then fixing the reflection phase shift to optimize the power allocation coefficient;

[0023] After determining the minimum signal-to-noise ratio of all users, the optimization problem of maximizing the minimum semantic spectral efficiency of the user is converted to solve the semantic symbol number, and the final solution is obtained by looking up the table.

[0024] The present application considers an IRS-assisted NOMA semantic transmission framework, in which the base station can effectively transmit semantic information to multiple users by sending superimposed downlink signals to these users. In order to further improve the spectral efficiency and ensure user fairness, the present application jointly optimizes the semantic symbol number, the transmit power at the base station and the phase shift at the IRS to maximize the minimum semantic spectral efficiency of all users; in order to solve this problem, the present application proposes a new three-step resource allocation algorithm, and the simulation results clearly confirm the effectiveness of the resource allocation method proposed by the present application, promoting the coexistence of traditional communication and semantic communication in future networks. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 is a schematic diagram of the IRS-assisted downlink NOMA semantic transmission framework of the present application;

[0026] Figure 2 is a schematic diagram of the convergence of the joint power allocation and phase shift design algorithm of the present application;

[0027] Figure 3 For the present application scheme and other prior art schemes with other th The contrast curve diagram of the present application scheme and other prior art schemes with power P changes when

[0028] Figure 4 For when th The contrast curve diagram of the present application scheme and other prior art schemes with power P changes when DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0030] The present application proposes an IRS-assisted NOMA semantic transmission resource allocation method, comprising: in an IRS-assisted NOMA semantic transmission model, by optimizing the semantic compression rate, power allocation and phase shift at the IRS, the minimum semantic spectral efficiency of the user is maximized.

[0031] In order to ensure the communication quality of each user without difference, and at the same time realize a more efficient semantic transmission system, the present application considers an IRS-assisted NOMA semantic transmission framework, in which the present application establishes an optimization problem with the goal of maximizing the minimum semantic spectral efficiency (Semantic Spectral efficiency, S-SE) of the user, and optimizes the optimization problem by optimizing the semantic compression rate, power allocation and phase shift at the IRS. Due to various coupled variables and non-convexity in the problem, it is challenging to solve the problem, therefore the present application designs a three-step algorithm for the joint optimization problem of the transmission power of the semantic communication base station, the IRS phase shift and the number of semantic symbols, specifically including:

[0032] Firstly, a low-complexity NOMA decoding scheme is proposed to decouple the decoding order design and the IRS reflection coefficient design;

[0033] Then, a joint optimization algorithm is proposed for solving the joint power allocation and reflection coefficient design problem under the given decoding order;

[0034] Finally, under the given phase shift and power allocation, the number of transmission semantic symbols is designed by the exhaustive search method;

[0035] The resource allocation method of the present application effectively optimizes the S-SE while reducing the computational complexity.

[0036] As Figure 1 shown, this embodiment considers an IRS-assisted downlink NOMA semantic transmission framework, which consists of one base station, an RIS with M independent passive reflecting elements, and K users. This embodiment defines the set of reflecting elements as The set of users is The DeepSC model is first trained at the base station to improve its ability to extract underlying semantics, and then the trained DeepSC model is broadcast to the users. The transmitter performs feature extraction on the source information, followed by semantic compression encoding and channel encoding. In this process, the base station serves as the coordination center, processing the extracted key task information. The base station uses non-orthogonal multiple access (NOMA) technology to superimpose these information, ensuring that multiple users can efficiently share channel resources. During transmission, the IRS is used to enhance and optimize the signal to ensure effective transmission of the signal in complex propagation environments. The receiver performs channel decoding and semantic decoding to ensure that the key task information can be recovered to support the decision and execution of downstream applications.

[0037] 1. IRS-assisted NOMA semantic transmission model

[0038] User k requests to receive a text file, and the text data can be represented as S k = [s1, s2,... s i ,...,s L ], where s i represents the i-th word, and L represents the average number of words per text. The file is mapped to a semantic symbol vector x k by the DeepSC transmitter, where x k contains the semantic information of S k , and x k has a length of τ k L, where τ k represents the average number of semantic symbols mapped per word, and simultaneously:

[0039]

[0040] where B represents the maximum value that the average number of semantic symbols mapped per word can take.

[0041] The single-antenna base station transmits superimposed signals to K users on the same time and frequency block, and the intelligent controller connected to the IRS can intelligently adjust the phase shift to assist NOMA transmission, defining as the reflection matrix of the IRS, where β m represents the reflection coefficient of the m-th reflecting element, and θ m represents the reflection phase shift of the m-th reflecting element. In this case, a fixed amplitude reflection coefficient βm = 1, the phase shift at the IRS needs to satisfy the following constraint:

[0042] 0 < θ m ≤ 2π, m e M (2)

[0043] The baseband transmit signal at the base station can be represented as where a k is the power allocation factor that allocates a portion of the total power P to the kth user, i.e., the power allocated to user k. To accommodate the power resource constraint at the base station, the embodiment imposes the following restriction on the vector a k :

[0044]

[0045] Let the channel from the IRS to user k be denoted as The channel from the base station to the IRS is denoted as The channel from the base station to user k is denoted as v k , The additive white Gaussian noise (AWGN) at user k is denoted as n The received signal at user k can be represented as:

[0046]

[0047] The Successive Interference Cancellation (SIC) decoding order is an important issue in NOMA systems, where the optimal decoding order is determined by the channel gain. The channel gain of user k is denoted as Without loss of generality, the embodiment assumes that the ordering of users is 0 < |h1| 2 ≤ |h2| 2 ≤... ≤ |h K | 2 .

[0048] The corresponding Signal to Interference and Noise Ratio (SINR) of the kth user decoding the tth data stream can be represented as:

[0049]

[0050] To ensure that SIC can be performed correctly, the SINR of the kth user decoding the tth data stream (i.e., γ t→k ) needs to be no less than the SINR of the tth data stream, which can be represented as:

[0051]

[0052] The Semantic Spectral Efficiency (S-SE) of user k can be defined as:

[0053]

[0054] where I denotes the semantic information amount of the text S k , and ξ(τ k ,γ k ) denotes the semantic similarity. The semantic similarity lacks a clear form. The relationship between the semantic similarity with a given semantic compression degree and the SINR can be modeled as a generalized logistic function, i.e.,

[0055]

[0056] where are parameters determined using a nonlinear least squares criterion.

[0057] In order to achieve the optimization and coordination between effectiveness and transmission efficiency, the semantic similarity is constrained, and is expressed as:

[0058]

[0059] 2. Construction of optimization problem

[0060] According to the above analysis, the present embodiment maximizes the resource allocation problem of the minimum S-SE of users by optimizing the semantic compression rate (the number of semantic symbols) τ k , the power allocation (i.e., α = [α1,...,α K ] T ) at the base station, and the phase shift (i.e., Θ) at the IRS. The optimization problem of maximizing the minimum semantic spectral efficiency of users is expressed as:

[0061] Optimization problem (P0)

[0062] Constraints: (1), (2), (3), (6), (9)

[0063] The optimization problem P0 is a mixed integer nonlinear programming problem, and solving the problem P0 faces three main challenges: first, due to the existence of integer variables k, the optimization problem P0 is an NP-hard problem; second, since the IRS reflection coefficient can control the decoding order, it is difficult for NOMA users to obtain the optimal decoding order; third, the transmission power and the reflection coefficient are highly coupled, which makes the problem more challenging. In order to find a feasible solution, the present embodiment proposes a three-step algorithm, which specifically includes:

[0064] (1) Determine the decoding order

[0065] The determination of the SIC decoding order depends on the combined channel gain of the direct link and the reflected link. The decoding order is determined by optimizing the IRS phase shift to maximize the maximum achievable combined channel strength of all K users. Specifically, the maximum achievable strength of the combined channel of the kth user, denoted as, can be obtained by solving the following problem:

[0066] Optimization problem (P1)

[0067] Constraint: (2)

[0068] The decoding order is determined by sorting the achievable combined channel strength of the users after solving the K optimization problems.

[0069] (2) Joint power allocation and reflection coefficient design

[0070] For a given decoding order, the SIC decoding constraint (6) in optimization problem P0 can be further simplified as:

[0071]

[0072] From equation (8), it is known that ξ(τ k ,γ k ) is monotonically non-decreasing with γ k , so to solve problem P0, we need to maximize the minimum signal-to-noise ratio of all users, i.e.:

[0073]

[0074] Problem P0 can be simplified as follows:

[0075] Optimization problem (P2)

[0076] Constraints: (1), (2), (3), (9), (10), (11)

[0077] The present application ensures that the signal-to-noise ratio of each user exceeds Q, where Q is a relaxation variable representing the minimum SINR to be maximized. Since there are two variable blocks (i.e. α and Θ) coupled in optimization problem (P2), the present embodiment alternately optimizes the power allocation coefficient and the reflection phase shift, and P2 can be decomposed into two sub-problems P2.1 and P2.2 for solving. Specifically:

[0078] Given the power allocation coefficient, the reflection phase shift is optimized, i.e.

[0079] Optimization problem (P2.1)

[0080] Constraints: (2), (10), (11)

[0081] Given the reflection phase shift, the power allocation coefficients are optimized, i.e.

[0082] Optimization problem (P2.2)

[0083] Constraints: (3), (11)

[0084] In optimization problem P2.1, the process can be optimized and solved by using the solver CVX, specifically: first, a bisection search method is used to solve the non-convex constraint and SDR (semi-definite relaxation) technology is used for approximate solution, but the obtained phase shift solution usually does not satisfy the rank-one constraint; then, a Gaussian randomization scheme is introduced to obtain a rank-one solution.

[0085] In optimization problem P2.1, the SINR of each user at the optimal solution must be equal, resulting in a second-order nonlinear equation, which can be effectively solved by existing software such as matlab.

[0086] (3) Semantic symbol number design

[0087] When the SINR is known, the semantic spectral efficiency of the user is maximized, and the optimization variable is the number of semantic symbols, resulting in the following optimization problem:

[0088] Optimization problem (P3)

[0089] Constraints: (1), (9)

[0090] where ξ(τ k ,γ k ) is affected by the number of semantic symbols τ k and the signal-to-noise ratio γ k ; I / L is the amount of semantic information contained in each word or word, in this invention I / L depends on the type of source, this term is constant for a specific type of source and does not affect the optimization of resources, and can be obtained by looking up the table; after the minimum signal-to-noise ratio is determined, the problem of maximizing the minimum semantic spectral efficiency only needs to maximize the value of , so that τ k can be obtained, so the exhaustive method is used to solve P3.

[0091] Next, the effectiveness of the present application is verified by simulation. In the simulation, two users are considered, i.e. K=2, the distance from the BS to the IRS is 10m, the distances from the BS and the IRS to user 1 are set to 40m and 20m respectively, and the distances from the BS and the IRS to user 2 are set to 35m and 25m respectively; the IRS has 2 reflecting elements, and the noise power at the user is: σ 2= -80 dBm. In addition, considering that the link from the base station to the user is Rayleigh fading, and the links from the base station to the IRS and from the IRS to the user are Rician fading, a distance-dependent path loss model P L = C0(d / d0) -η is adopted, where C0= -20 dB represents the path loss at a reference distance d0= 1 m, and η represents the path loss exponent, and d represents the distance between the transmitting end and the receiving end. The path loss coefficients of the base station-user, base station-IRS, and IRS-user are 3.5, 2, and 2, respectively. The present application needs to make restrictions on Φ k to ensure that the experimental results are meaningful, i.e., Φ k ≥ Φ th , Φ th = 0.05, and the number of Gaussian randomized random vectors is 400 times.

[0092] Figure 2 The convergence of the joint power allocation and phase shift design algorithm of the present application is shown. Initially, the value of the Max-min SINR is low, and then through the alternating iteration of the phase and the power, the optimal phase and power values are constantly found, and as the number of iterations increases, the growth value gradually tends to be stable at 11.4, which embodies the guarantee of user fairness of the algorithm.

[0093] Figure 3 The curve of different algorithms with the change of semantic similarity threshold ξ th is depicted. As ξ th increases, the max min S-SE of the RMS and MMS schemes decreases slightly, while the proposed scheme (PS) and the least mapping scheme (LMS) of the present application decrease significantly from 0.8 to 0.9. For the random mapping scheme (RMS) and the maximum mapping scheme (MMS) scheme, the average number of semantic symbols of each word is fixed, so the max min S-SE only depends on the SINR, and the increase in the semantic similarity threshold can be achieved by increasing the average number of semantic symbols of each word, but this also leads to the decrease of the max min S-SE. In short, the improvement of transmission reliability is at the expense of transmission rate. From Figure 3 , it can be observed that the performance of the scheme of the present application is significantly better than the other three schemes. In the algorithm proposed in the present application, an optimal mapping strategy is adopted, and the other three schemes lack this ability, which can accurately map based on the quality of the signal and the channel.

[0094] Figure 4 The curve of different algorithms with the change of semantic similarity threshold ξ thFig. 6 is a comparative curve diagram of the max min S-SE of different algorithms varying with power P when =0.7. It can be observed from the figure that the max min S-SE of each scheme increases with the increase of power. This is because the increase of power enhances the signal and thus improves the SINR, so that a relatively low code rate can be used to ensure the task performance, thus reducing the average number of semantic symbols per word. Figure 4 It can be observed from Fig. 6 that the algorithm proposed in the present application achieves a performance close to the optimal performance and is significantly better than other schemes.

[0095] Although the embodiments of the present application have been shown and described, it can be understood by those of ordinary skill in the art that various changes, modifications, replacements and variations can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A IRS-aided NOMA semantic transmission resource allocation method, characterized in that, The optimization problem of maximizing the minimum semantic spectral efficiency of users is expressed as: In the IRS-aided NOMA semantic transmission model, the minimum semantic spectral efficiency of users is maximized by optimizing semantic compression rate, power allocation and phase shift at IRS, wherein the IRS-aided NOMA semantic transmission model comprises a base station, an IRS with M independent passive reflecting elements and K users, the set of reflecting elements is defined as , and the set of users is defined as The DeepSC model is first trained at the base station, and then the trained DeepSC model is broadcast to the users. The transmitter extracts features from the source information, and then performs semantic compression encoding and channel encoding. In this process, the base station serves as a coordination center to process the extracted key task information. The base station uses non-orthogonal multiple access (NOMA) technology to superimpose these information. In the transmission process, the IRS is used to enhance and optimize the signal. The receiver recovers the key task information through channel decoding and semantic decoding. The semantic spectral efficiency of user k is expressed as: Optimization problem: Constraints: in, Power allocation for all users is represented as , The power allocated to user k, where K is the number of users in the IRS-assisted NOMA semantic transmission model; Here is the reflection matrix of the IRS; Let B be the average number of semantic symbols mapped to the words of user k, where the number of semantic symbols is the semantic compression ratio; B is the maximum value of the average number of semantic symbols mapped to a single word. For user k, the semantic spectrum efficiency; A collection of users; For the first The reflection phase shift of each reflecting element, where M is the number of reflecting elements; This represents the signal-to-interference-plus-noise ratio when the k-th user decodes the t-th data stream; The total power allocated by the base station to all users; Let be the conjugate transpose of the channel gain from the IRS to the user for the t-th user; The channel gain from the base station to the IRS; Let be the channel gain from the base station to the t-th user; Noise power; Let SINR be the value of the t-th data stream. Let k be the semantic similarity to user k. The threshold for semantic similarity; The process of solving the optimization problem includes: wherein, represents the semantic similarity of user k, I represents the semantic information amount of the text requested by user k to receive L represents the average number of words of each text; Semantic similarity of user k is represented as: wherein , , , are parameters determined using a non-linear least squares criterion; The decoding order is determined by optimizing the IRS phase shift to maximize the maximum achievable combined channel strength of all K users; For a given decoding order, the SIC decoding constraint in the optimization problem is simplified, and an optimization problem of maximizing the minimum signal-to-noise ratio of all users is constructed; The optimization problem of maximizing the minimum signal-to-noise ratio of all users is solved alternately, i.e., first fix the power allocation coefficient to optimize the reflection phase shift, and then fix the reflection phase shift to optimize the power allocation coefficient; After determining the minimum signal-to-noise ratio of all users, the optimization problem of maximizing the minimum semantic spectral efficiency of users is converted to solve the semantic symbol number, and the final solution is obtained by looking up the table. For each user, the maximum achievable strength of the combined channel is determined, and the optimization problem of the maximum achievable strength of the combined channel of the kth user is expressed as:

2. The IRS-assisted NOMA semantic transmission resource allocation method according to claim 1, wherein, The optimization problem of maximizing the minimum signal-to-noise ratio of all users is expressed as: Optimization problem: Constraints: wherein, denotes the conjugate transpose of the channel from the IRS to user k, denotes the channel from the base station to user k, .

3. The IRS-assisted NOMA semantic transmission resource allocation method according to claim 1, wherein, The process of solving the optimization problem of maximizing the minimum signal-to-noise ratio of all users includes: Optimization problem: Constraints: Where Q represents the user's minimum signal-to-noise ratio; Indicates IRS to user channel The conjugate transpose of . This indicates the distance from the base station to the user. The channel, .

4. The IRS-assisted NOMA semantic transmission resource allocation method according to claim 3, wherein, First, fix the power allocation coefficient to optimize the reflection phase shift, i.e., The process is optimized and solved by using the solver CVX, i.e., first use the bisection search method to solve the non-convex constraint and the semi-positive definite relaxation technique for approximate solution, and then introduce the Gaussian randomization scheme to obtain a rank-one solution; Optimization problem: Constraints: In the fixed reflection phase shift, the power allocation coefficient is optimized, i.e., The signal-to-noise ratio of each user at the optimal solution must be equal, resulting in a second-order nonlinear equation, which is solved by matlab to obtain the power allocation coefficient. Optimization problem: Constraints: After determining the minimum signal-to-noise ratio of all users, the optimization problem of maximizing the minimum semantic spectral efficiency of users is converted to solve the semantic symbol number, which is expressed as:

5. The IRS-assisted NOMA semantic transmission resource allocation method according to claim 1, wherein, ​ Optimization problem: Constraints: 。

Images