Robust semantic communication beam forming method of MISO channel

Abstract

The invention relates to the technical field of wireless communication, in particular to a robust semantic communication beam forming method of an MISO channel. Constructing a digital robust semantic communication and beam forming network for a multi-input single-output (MISO) channel; establishing an ABG expression according to semantic transmission performance in the network; designing and considering an optimal beam forming scheme under different channel state information conditions, wherein the optimal beam forming scheme comprises optimal beam design of an MISO channel under an ideal channel state information condition; carrying out MISO channel optimal beam design with unbounded channel estimation errors and MISO channel optimal beam design with bounded channel estimation errors; and respectively solving the robust beam forming schemes under different channel state information conditions, and outputting optimal beams under different channel state information conditions. According to the method, an optimal design scheme of the MISO channel under different channel state information conditions is provided, and the transmission power can be minimized while the service quality of a semantic communication system can be effectively met.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communications, and in particular to a robust semantic communication beamforming method for a MISO channel. Background Art

[0002] With the rapid deployment of 5G wireless networks, global attention is increasingly turning to the prospects for the development of 6G technology, especially in user-centric applications such as Internet of Vehicles (IoV), autonomous driving, and telemedicine; these innovative services not only require a large amount of data transmission and resource allocation, but also impose strict requirements on high reliability and ultra-low latency. Traditional data-driven communication systems are now considered to be the main bottleneck for achieving high quality of service (QoS). To address these challenges, semantic communication has emerged as an emerging paradigm that aims to deeply understand and utilize the semantic content of transmitted information. Unlike traditional methods, semantic communication enables transmitters and receivers to collaborate more efficiently by leveraging shared knowledge, experience, and syntactic, semantic, and reasoning rules to ensure the accurate transmission of the conveyed intent.

[0003] The development of 6G technology is leading to new changes in the field of communications, especially in user-centric applications such as Internet of Vehicles (IoV), autonomous driving, and telemedicine. These applications have extremely high requirements for data transmission, not only requiring large capacity and efficient resource allocation, but also extremely high reliability and extremely low latency. Traditional communication systems face bottlenecks in achieving high quality of service (QoS), so semantic communication has emerged as an emerging communication paradigm. The core of semantic communication lies in the in-depth understanding and utilization of the semantic content of the transmitted information. Unlike traditional communication methods, semantic communication focuses not only on the transmission of data, but also on the meaning and intention carried by the data. By sharing knowledge, experience, and specific syntax, semantics, and reasoning rules between transmitters and receivers, semantic communication enables more efficient collaboration. This approach not only improves communication efficiency, but also ensures the accurate transmission of information intent, providing users with a better service experience.

[0004] Multiple-input single-output (MISO) is an important technology in wireless communication systems, especially suitable for scenarios where there are multiple antennas at the transmitter and only one antenna at the receiver. By deploying multiple antenna arrays at the transmitter and combining beamforming technology, the MISO system can significantly enhance the reliability and anti-interference capability of signal transmission; in addition, MISO technology not only improves the transmission data rate, but also helps to reduce the impact of channel fading, especially in scenarios that require high data rates and low latency. Therefore, combining semantic communication with MISO technology, giving full play to the ability of semantic communication to extract key semantics at the source, and the advantages of MISO in high spectral efficiency and signal processing capabilities, provides a highly potential solution for 6G large-scale communication systems.

[0005] In the prior art, some scholars have proposed a RIS-assisted multi-user multiple-input single-output (MISO) system, which uses maximum ratio transmission (MRT) and minimum mean square error (MMSE) precoding, as well as water injection power allocation; they have also proposed a novel deep reinforcement learning strategy for jointly optimizing transmission beamforming in a multi-user MISO environment, effectively solving the challenges associated with non-ideal channel state information (CSI); at the same time, the prior art focuses on minimizing the precoding design of information transmission in multi-user MISO networks, and proposes a generalized power iteration (GPI) precoding algorithm to generate the precoding vector of the base station. However, due to the complexity of the MISO system, the above prior art does not take into account the impact of CSI errors on the performance of the entire communication system. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the present invention provides a robust semantic communication beamforming method for a MISO channel. The method includes: constructing a digital robust semantic communication and beamforming network for a multi-input single-output MISO channel; establishing an ABG expression according to the semantic transmission performance in the network; designing an optimal beamforming scheme under different channel state information conditions, including an optimal beam design for a MISO channel under perfect channel state information conditions; an optimal beam design for a MISO channel with an unbounded channel estimation error and an optimal beam design for a MISO channel with a bounded channel estimation error; respectively solving the robust beamforming schemes under different channel state information conditions, and outputting the optimal beams under different channel state information conditions. The present invention proposes an optimized design scheme for a MISO channel under different channel state information conditions, which can effectively meet the service quality of the semantic communication system while minimizing the transmission power.

[0007] The present invention adopts the following technical solution, a robust semantic communication beamforming method for a MISO channel, comprising: Constructing digital robust semantic communication and beamforming networks for multiple-input single-output (MISO) channels; Establishing ABG expressions according to the semantic transmission performance in the network; Based on the established ABG expression, the optimal beamforming scheme considering different channel state information conditions is designed; The optimal beamforming schemes under different channel state information conditions include: Optimal beam design for MISO channels under ideal channel state information conditions; Optimal beam design for MISO channels with unbounded channel estimation error; and Optimal beam design for MISO channels with bounded channel estimation error; The robust beamforming schemes under different channel state information conditions are solved respectively, and the optimal beams under different channel state information conditions are output.

[0008] Further, the digital robust semantic communication and beamforming network is: Assume that a transmitter with N antennas transmits semantic information to a single-antenna receiver through a MISO semantic communication system. The semantic encoder on the transmitter is derived from the input image Extract and encode semantic features , expressed as: ; in represents the semantic encoder, represents the learnable parameters of the semantic encoder; The semantic features are converted into binary vectors and sent to the receiver based on the beamforming design. The received signal is expressed as: ; in, , represents the channel vector between the transmitter and the receiver in the MISO semantic communication system; It means that through beamforming design, Semantic features represented as binary vectors; Through a fully connected layer, the equalized signal is transformed into semantic features Vectors with the same dimensions ; The reconstructed image is obtained through the semantic decoder; among them, the signal-to-noise ratio received by the MISO semantic communication system is It is expressed as: ; in, is the noise power of the complex Gaussian additive white noise.

[0009] Furthermore, an ABG expression is established according to the semantic transmission performance in the network, specifically: The semantic transfer performance is modeled as: ; in, represents the compression ratio of the semantic communication network, and Represent the quantization bit number and original bit number of the image respectively. are the learnable parameters of the semantic encoder; The ABG expression based on the semantic transfer performance is: ; in, Indicates the upper limit of image reconstruction quality, parameter Depends on semantic encoder and decoder.

[0010] Furthermore, the optimal beam design of the MISO channel under the ideal channel state information condition includes: The optimization problem of satisfying the quality of service of semantic communication while minimizing power consumption is proposed, which can be expressed as: ; in, Representing multi-scale structural similarity thresholds in semantic transfer performance; Based on the matched filtering method, the optimal beamformer is obtained for: ; The optimal beamformer is used to solve the problem and output the optimal robust beam.

[0011] Furthermore, the optimal beam design of the MISO channel with unbounded channel estimation error includes: The unbounded channel state information error is expressed as: ; in, represents the channel state information estimated by the receiver, represents the estimation error of the channel state information, and ; The problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under the MISO channel with unbounded channel estimation error is proposed, which can be expressed as: ; in, represents the minimum required multi-scale structural similarity threshold in semantic communication systems, Represents the maximum interruption probability in the semantic communication system.

[0012] Defined using a positive semidefinite matrix , and define for: ; According to the ABG formula, the constraint problem is transformed into: ; in, , ; The constraint problem is transformed into a convex constraint problem by using the semi-positive definite relaxation method and Bernstein inequality, which can be expressed as: ; The interior point method is used to solve the problem and the optimal beam in the MISO channel with unbounded channel estimation error is obtained.

[0013] Furthermore, the optimal beam design of the MISO channel with bounded channel estimation error includes: The bounded channel state information error is expressed as: ; in, represents a Hermitian matrix; Under the condition of bounded channel state information error, the signal-to-noise ratio received by the receiver is expressed as: ; The problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under a MISO channel with bounded channel estimation error is proposed, which can be expressed as: ; According to the ABG formula, the constraint problem is transformed into: ; in represents the minimum required signal-to-noise ratio, The S-lemma and semi-positive definite relaxation method are used in turn to transform the constraint problem into a convex semi-positive definite programming subproblem, which can be expressed as: ; The transformed convex semidefinite programming subproblem is solved using the CVX method to obtain the optimal beam in the MISO channel with bounded channel estimation error.

[0014] The beneficial effects of the present invention are: (1) The present invention extends the ABG formula to the MISO channel under ideal CSI conditions and proposes a corresponding beamforming design scheme. Compared with the random and omnidirectional beamforming methods, the method proposed in the present invention is significantly superior to them while ensuring the QoS requirements of the communication, highlighting the advantages and effectiveness of the beamforming design based on the ABG formula.

[0015] (2) The present invention proposes a robust beamforming design with chance constraints for the DRSC-BF network in a MISO channel with unbounded channel estimation error. Due to the difficulty of probability constraints, the present invention uses the SDR method and Bernstein type inequalities to convert these probability constraints into deterministic convex constraints, and efficiently solves them through the interior point method, effectively reducing the channel estimation error while maintaining the user's QoS requirements.

[0016] (3) The present invention proposes a robust beamforming design for RSCBF networks in MISO channels with bounded channel estimation errors. By sequentially applying the S-lemma, orthogonal relaxation and SDR methods, the non-convex joint optimization problem is decomposed and converted into convex SDP sub-problems, thereby achieving robust semantic communication under MISO channel conditions and effectively guaranteeing the user's QoS requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 A schematic flow chart of a robust semantic communication beamforming method for a MISO channel according to an embodiment of the present invention; Figure 2 A schematic diagram of a digital robust semantic communication and beamforming network structure for a MISO channel according to an embodiment of the present invention; Figure 3 A schematic diagram of an image semantic extraction network structure according to an embodiment of the present invention; Figure 4 A schematic diagram of a visual converter model test curve and an ABG formula fitting curve according to an embodiment of the present invention; Figure 5 A schematic diagram of a cumulative distribution function comparing a robust design with a non-robust design according to an embodiment of the present invention; Figure 6 A schematic diagram showing the comparison between a cumulative distribution function of a beamforming design according to an embodiment of the present invention and a fixed phase and a random phase; Figure 7 A schematic diagram of a cumulative distribution function comparing a robust design and a non-robust design according to an embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] A flowchart of a robust semantic communication beamforming method for a MISO channel according to an embodiment of the present invention is shown in FIG. Figure 1 As shown, including: Constructing digital robust semantic communication and beamforming networks for multiple-input single-output (MISO) channels; Establishing ABG expressions according to the semantic transmission performance in the network; Based on the established ABG expression, the optimal beamforming scheme considering different channel state information conditions is designed; The optimal beamforming schemes under different channel state information conditions include: Optimal beam design for MISO channels under ideal channel state information conditions; Optimal beam design for MISO channels with unbounded channel estimation error; and Optimal beam design for MISO channels with bounded channel estimation error; The robust beamforming schemes under different channel state information conditions are solved respectively, and the optimal beams under different channel state information conditions are output.

[0021] In another specific embodiment of the present invention: A digital robust semantic communication and beamforming network (DRSC-BF) for multiple-input single-output MISO channels is constructed. Figure 2 As shown in Figure 1, a transmitter with N antennas is designed to transmit semantic information to a single-antenna receiver through an efficient beamforming design, specifically: The semantic encoder on the transmitter is trained from the input image Extract and encode semantic information , expressed as: ; in represents the semantic encoder, represents the learnable parameters of the semantic encoder; Convert the semantic features into binary vectors, ,in is the number of bits per image, and the quantization process is expressed as , its output can be expressed as: ; Based on beamforming design , and send it to the receiver. The received signal is expressed as: ; in, , represents the channel vector between the transmitter and the receiver in the MISO semantic communication system; It means that through beamforming design, Semantic features represented as binary vectors; Dequantization Then, a fully connected layer is used to transform the equalized signal into semantic features. Vectors with the same dimensions ; ; The reconstructed image is obtained through the semantic decoder: ; Among them, the signal-to-noise ratio received by the MISO semantic communication system It is expressed as: ; in, is the noise power of the complex Gaussian additive white noise.

[0022] So far, the performance of semantic communication mainly relies on the generalization ability of deep learning networks and empirical parameter tuning, and lacks the quality of service (QoS) guarantee of network communication, which hinders its practical application. The main reason is that semantic communication lacks theoretical tools for analysis and design. For example, the performance measurement of semantic encoding and semantic information transmission, as well as end-to-end latency, are unresolved issues.

[0023] To date, the performance of semantic communication mainly depends on the generalization ability of deep learning networks and the tuning of empirical parameters. However, this dependence leads to a lack of effective guarantee of the quality of service (QoS) of network communication, which hinders the application of semantic communication in practice. The main reason is that semantic communication lacks theoretical tools for analysis and design. For example, how to measure the performance of semantic encoding and semantic information transmission, and how to evaluate end-to-end latency, are still unresolved issues.

[0024] Due to the high nonlinearity of deep learning-based encoders, it is difficult to derive their performance measurement formulas from a theoretical perspective. This application uses a nonlinear least squares (NLS) method to fit the increase of the signal-to-noise ratio (SNR). The image reconstruction quality (MS-SSIM and PSNR) first increases rapidly, then slowly increases to the upper limit, and finally stabilizes. Inspired by this phenomenon, the theoretical relationship between channel state information in semantic coding: multi-scale structural similarity (MS-SSIM) and compression ratio It can be modeled as: ; in, represents the compression ratio of the semantic communication network, and Represent the quantization bit number and original bit number of the image respectively. are the learnable parameters of the semantic encoder; The ABG expression based on the semantic transfer performance is: ; in, Indicates the upper limit of image reconstruction quality, parameter Depends on semantic encoder and decoder.

[0025] It should be noted that the ABG formula can also be applied to Figure 3 As shown in Figure 2, in image reconstruction tasks with PSNR and semantic communication with reasoning tasks, Figure 3 ×L in the figure indicates that the module is repeated L times.

[0026] As the signal-to-noise ratio (SNR) increases, the performance measures of semantic communication in the image reconstruction task (MS-SSIM and PSNR) and the reasoning task (accuracy) first increase rapidly, then increase slowly to an upper limit, and finally stabilize; the reasons for the regularity of performance measurements are as follows: (1) Effect of noise reduction: At low SNR, the received signal is dominated by noise, which severely degrades the quality of the transmitted signal. The semantic encoder and decoder have difficulty extracting and reconstructing meaningful information from noisy signals, resulting in lower performance. As the SNR increases, the noise decreases and the signal becomes clearer. This enables the decoder to decode and reconstruct semantic information more efficiently, resulting in rapid improvement in inference accuracy and reconstruction performance. Specifically, as the SNR increases, the reduction in noise has a significant impact, allowing the system to recover more semantic information from the received signal, resulting in a rapid improvement in performance.

[0027] (2) Transition from noise-dominated to signal-dominated transmission: When the SNR is low, noise has a large impact on transmission, and the performance of the semantic communication system is limited by its ability to handle this noise. As the SNR increases, semantic communication transitions from a noise-dominated state to a signal-dominated state. During this transition phase, even a small increase in SNR can significantly improve performance because the system is better at extracting the underlying semantic information. Once the signal begins to dominate the noise, the semantic communication system can more effectively utilize the transmitted information, thereby significantly improving performance. However, once the semantic communication system reaches this stage, further increases in SNR will provide diminishing returns.

[0028] (3) Performance saturation: As the SNR continues to increase, the noise becomes negligible, and the semantic communication system operates in an almost noise-free environment. The performance of the semantic communication system is mainly limited by the capabilities of the encoder, decoder, and underlying model architecture, rather than channel noise. When noise is no longer a limiting factor, the performance of the semantic communication system reaches a saturation point, which is mainly limited by the model design and learning capabilities. Since the system has achieved near-optimal performance given the model architecture, further increases in SNR will not bring substantial performance improvements.

[0029] (4) Nonlinear performance gain: The initial rapid improvement can also be attributed to the nonlinear behavior of neural networks and other components used in semantic communication systems. These models are usually very sensitive to changes in input quality. As the SNR increases and the input signal quality improves, these models can extract features and semantic information more effectively, resulting in a sharp improvement in performance. However, as the signal quality continues to improve, this nonlinear improvement gradually fades, resulting in a gradual improvement in performance. Neural networks with nonlinear activations can show rapid performance gains when the input quality improves, but the rate of performance improvement slows down as the input becomes sufficiently clean (high SNR).

[0030] make Represents the end-to-end latency of the semantic communication system, including semantic encoding time , Transmission time and semantic decoding time ,Right now: ; The semantic encoding time can be estimated by analyzing the computational complexity of each layer in the semantic encoding network. The computational complexity of the semantic encoding network is usually expressed in floating point operations (FLOPs). Suppose the semantic encoder contains layer, and let (FLOPs) represents the semantic encoder The computational amount of the layer, ;make (FLOPs per second) represents the computing power of the semantic encoder hardware and the semantic encoding time It is expressed as: ; Similarly, the semantic decoding time It is expressed as: ; in (FLOPs) is the semantic decoder The amount of computation of the layer, is the total number of layers in the semantic decoder network, (FLOPs per second) represents the semantic decoder computing power.

[0031] Transfer time It is expressed as: ; Therefore, the end-to-end latency of the semantic communication system is expressed as: ; Therefore, the present application proposes an optimal beam design for a MISO channel under ideal channel state information conditions, including: The optimization problem of satisfying the quality of service of semantic communication while minimizing power consumption is proposed, which can be expressed as: ; in, Representing multi-scale structural similarity thresholds in semantic transfer performance; The optimization problem is non-convex, so the optimization problem can be equivalently expressed as: ; Based on the matched filtering method, the optimal beamformer is obtained for: ; The optimal beamformer is used to solve the problem and output the optimal robust beam.

[0032] In practical applications, due to the limited length of pilot sequences and the existence of quantization errors, channel estimation errors are inevitable. In addition, the performance of semantic communication is sensitive to channel state information (CSI) errors, which will seriously deteriorate the inference accuracy and data reconstruction quality performance. Based on this fact, this application further proposes a robust beamforming design for semantic communication. Specifically, the estimated semantic transmission performance (CSI) error is non-ideal, that is: ; in represents the CSI estimation error.

[0033] Since it is difficult to obtain accurate channel state information (CSI) in practical environments, this application proposes an unbounded channel state information error representation as:

[0034] in, represents the channel state information estimated by the receiver, represents the estimation error of the channel state information, and .

[0035] The problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under the MISO channel with unbounded channel estimation error is proposed, which can be expressed as: ; in, represents the semantic similarity threshold in the semantic communication system, which is set as the minimum semantic similarity threshold in the present invention. represents the interruption probability in the semantic communication system, which is also set to the maximum allowed interruption probability in the present invention; According to the ABG formula, the constraint problem is transformed into: ; definition , and define for: ; The constraint problem is transformed into: ; in, ; This application further replaces the positive semidefinite matrix ,in ; To handle non-convex constraints , the SDR method is used to simplify the optimization problem and relax the above constraints as follows: ; Due to the rank-1 relaxation, the expression of CSI error can be rewritten as: ; in, It obeys the standard complex Gaussian distribution.

[0036] therefore, It can be expressed as: ; in, ; get: ; in, .

[0037] In order to transform the probabilistic constraints into deterministic constraints, this application further uses the following Bernstein-type inequality: set up ,in represents the complex conjugate matrix, ,for , the following inequality holds: ; in Representation Matrix The maximum eigenvalue of .

[0038] Based on the Bernstein-type inequality, the probabilistic constraint can be transformed into the following deterministic constraint: ; in, , which can be further transformed into the following convex constraint: ; in, represents the slack variable. The constraint contains a linear constraint, an SDP constraint and a convex PSD constraint, so it can be conservatively transformed into: ; The interior point method is used to solve the problem and the optimal beam in the MISO channel with unbounded channel estimation error is obtained.

[0039] In another specific embodiment of the present invention: The optimal beam design for MISO channel with bounded channel estimation error is further performed, including: The bounded channel state information error is expressed as: ; in, represents a Hermitian matrix; Under the condition of bounded channel state information error, the signal-to-noise ratio received by the receiver is expressed as: ; The problem of minimizing the transmission power while satisfying the semantic similarity constraint under a MISO channel with bounded channel estimation error is proposed, which can be expressed as: ; According to the ABG formula, the constraint problem is transformed into: ; in represents the minimum required signal-to-noise ratio; because and , so there are infinitely many possible and The implementation form of , which leads to the existence of infinite constraints, makes the solution difficult. In order to transform these infinite constraints into a finite set of linear matrix inequalities (LMIs), this application applies the following lemma: set up ; in ; Suppose there exists a vector x such that Established, we get the following two equivalent conditions: (1) If x satisfies ,but ; (2) There is a constant , so that .

[0040] Therefore, the problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under the MISO channel with bounded channel estimation error is transformed into: ; in, .

[0041] Applying the S-lemma to the above equation, we can reformulate it as a linear matrix inequality (LMI) as follows: ; To overcome the non-convexity problem, this application adopts the SDR method by using the following equation: ; Ignoring the rank-one constraint, the constraint problem is transformed into a convex semi-positive definite programming subproblem, expressed as: ; The transformed convex semidefinite programming subproblem is solved using the CVX method to obtain the optimal beam in the MISO channel with bounded channel estimation error.

[0042] In an experimental embodiment of the present invention: This application verifies the effectiveness of the proposed robust MISO semantic communication system in reconstruction tasks compared to the robust MISO semantic communication system with unbounded CSI estimation error through a set of comprehensive simulation experiments. The experiments are conducted using NVIDIA GeForce RTX 4090 GPU and rely on PyTorch supported by CUDA 11.4 for calculations.

[0043] like Figure 4 As shown in the figure, the reconstruction quality of the fitted MISO semantic communication system MS-SSIM varies with the signal-to-noise ratio The changing curve illustrates the relationship between the end-to-end performance of the deep learning model and the physical resources. In addition, the proposed ABG formula is highly consistent with the data obtained from simulation measurements.

[0044] Figure 5 Zuohe Figure 5 The right side shows the random phase and fixed phase under different reconstruction MS-SSIM thresholds based on the ABG formula under ideal channel conditions. From the simulation comparison of the three methods, it can be seen that the proposed ABG beamforming scheme under ideal channel conditions is better than the random phase and fixed phase beamforming schemes. The MS-SSIM reconstruction index of the ABG beamforming scheme proposed in this application always meets the threshold requirement, which proves the effectiveness of the beamforming design based on the ABG formula.

[0045] like Figure 6As shown in Figure 2, the CDF comparison between the robust MISO semantic communication system with bounded CSI estimation and the non-robust semantic communication system is shown, and the upper limit of the estimation error is set to , the threshold of the reconstruction index MS-SSIM It is 0.98, and the interruption probability of the non-robust semantic communication system is 0.2378. In comparison, the robust semantic communication system designed under the bounded CSI estimation error condition in this application achieves zero interruption probability and meets the design constraints, which shows that the proposed robust communication system provides better QoS than the non-robust system.

[0046] The comparison results between the ideal CSI and the channel estimation error following the complex Gaussian distribution are shown in Figure 7 As shown, the channel estimation error The average values ​​of 1%, 4%, 7% and 10% of the simulation results. From the comparison of the simulation results of the ideal CSI and the case with estimation error, it can be seen that the image reconstruction quality of the system with channel estimation error is lower than that of the ideal CSI scenario. This shows that the robust design of the MISO semantic communication system is necessary.

[0047] Figure 7 The cumulative distribution function (CDF) and MS-SSIM threshold of robust and non-robust systems under unbounded channel estimation error are shown. are set to 0.97 and 0.98 respectively, and the chance constraint are 0.07 and 0.10 respectively. Figure 7 Middle left: MS-SSIM threshold is 0.97, chance constraint is 0.07, it can be observed that the outage probability of the non-robust communication system is 0.09, while the outage probability of the robustly designed communication system is 0.04; Figure 7 Middle right, MS-SSIM threshold is 0.98, chance constraint is 0.10, it can be observed that the outage probability of the non-robust communication system is 0.36, while the outage probability of the robustly designed communication system is 0.07; from the above simulation, it can be seen that the outage probability of the robustly designed system always meets the constraint probability and is lower than the outage probability of the non-robust communication system, which indicates that the robust design can meet the QoS requirements and is effective.

[0048] In summary, the present application innovatively proposes a design of a robust beamforming semantic communication solution for MISO channels under CSI error conditions. The design based on the ABG company's formula describes the relationship between end-to-end communication performance indicators and physical layer resources. First, a beamforming solution is formulated under ideal CSI conditions. The simulation results show that the proposed solution effectively ensures the user's QoS. In order to solve the bounded channel estimation error in the MISO semantic communication network, the present application successively uses the S lemma, orthogonal relaxation and SDR to decompose the non-convex joint optimization problem into convex SDP sub-problems, thereby realizing robust semantic communication in the MISO channel. The simulation results confirm that the proposed solution meets the user's QoS requirements. In addition, the present application also proposes a robust beamforming design specifically for MISO semantic communication networks with bounded channel estimation errors. By adopting the S lemma, orthogonal relaxation and SDR, the non-convex joint optimization problem is decomposed into convex SDP sub-problems, thereby realizing robust semantic communication under MISO channel conditions. The simulation results verify that the proposed solution can effectively guarantee the user's QoS requirements.

[0049] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A robust semantic communication beamforming method for a MISO channel, characterized in that: include: Constructing digital robust semantic communication and beamforming networks for multiple-input single-output (MISO) channels; Establishing ABG expressions according to the semantic transmission performance in the network; Based on the established ABG expression, the optimal beamforming scheme considering different channel state information conditions is designed; The optimal beamforming schemes under different channel state information conditions include: Optimal beam design for MISO channels under ideal channel state information conditions; Optimal beam design for MISO channels with unbounded channel estimation error; and Optimal beam design for MISO channels with bounded channel estimation error; The robust beamforming schemes under different channel state information conditions are solved respectively, and the optimal beams under different channel state information conditions are output.

2. The robust semantic communication beamforming method for a MISO channel according to claim 1, characterized in that: The digital robust semantic communication and beamforming network is: Assume that a transmitter with N antennas transmits semantic information to a single-antenna receiver through a MISO semantic communication system. The semantic encoder on the transmitter extracts and encodes the semantic feature x from the input image s, which is expressed as: x=f θ (s) where f θ (·) represents the semantic encoder, θ represents the learnable parameters of the semantic encoder; The semantic features are converted into binary vectors and sent to the receiver based on the beamforming design. The received signal is expressed as: y=h H wz+n Through a fully connected layer, the equalized signal is transformed into semantic features Vectors with the same dimensions Superscript ^ indicates estimated values; The reconstructed image is obtained through the semantic decoder; wherein the signal-to-noise ratio ρ(w) received by the MISO semantic communication system is expressed as: In the above formula, h=[h1,h2,…,h N ] T ~CN(0,δ 2 I), represents the channel vector between the transmitter and the receiver in the MISO semantic communication system, I represents the identity matrix, δ 2 represents the complex Gaussian additive white noise power; w represents the beamforming design, z represents the semantic feature converted into a binary vector, and the superscript H represents the conjugate transpose.

3. The robust semantic communication beamforming method for a MISO channel according to claim 1, characterized in that: The ABG expression is established according to the semantic transmission performance in the network, specifically: The semantic transfer performance is modeled as: in, Denotes the compression ratio of the semantic communication network, d and n b Represent the quantized bits and original bits of the image respectively, c1, c2, c3 are the learnable parameters of the semantic encoder; The ABG expression based on the semantic transfer performance is: Among them, α>0 represents the upper limit of image reconstruction quality, parameters β>0, τ>0, γ>0 depend on the semantic encoder and decoder, and ρ(w) is the signal-to-noise ratio received by the MISO semantic communication system.

4. The robust semantic communication beamforming method for a MISO channel according to claim 1, characterized in that: The optimal beam design of the MISO channel under the ideal channel state information condition includes: The optimization problem of satisfying the quality of service of semantic communication while minimizing power consumption is proposed, which can be expressed as: stΓ(w)≥μ Among them, μ represents the semantic similarity threshold in semantic transmission performance, and w represents the beamforming design; Based on the matched filtering method, the optimal beamformer w is obtained opt for: Where h=[h1,h2,...,h N ] T ~CN(0,δ 2 I), represents the channel vector between the transmitter and the receiver in the MISO semantic communication system, α>0 represents the upper limit of the image reconstruction quality, the parameters β>0, τ>0, γ>0 depend on the semantic encoder and decoder, μ represents the semantic similarity threshold in the semantic transmission performance, ||·|| represents the Euclidean norm; The optimal beamformer is used to solve the problem and output the optimal robust beam.

5. The robust semantic communication beamforming method for a MISO channel according to claim 1, characterized in that: The optimal beam design for a MISO channel with unbounded channel estimation error comprises: The unbounded channel state information error is expressed as: in, represents the channel state information estimated by the receiver, represents a complex number, the superscript N represents the number of antennas, Δh represents the estimated error of the channel state information, and Δh~CN(0,E),Ef0; The problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under the MISO channel with unbounded channel estimation error is proposed, which can be expressed as: stPr{Γ(w)≤μ}≤p out Where μ represents the semantic similarity threshold in the semantic communication system, w represents the beamforming design, and p out ∈(0,1) represents the interruption probability of the semantic communication system; Defined using a positive semidefinite matrix And define f(W) as: According to the ABG formula, the constraint problem is transformed into: rank(W)=1, in, rank(ww H )=1; The constraint problem is transformed into a convex constraint problem by using the semi-positive definite relaxation method and Bernstein inequality, which can be expressed as: ε≥0 The interior point method is used to solve the problem and the optimal beam in the MISO channel with unbounded channel estimation error is obtained.

6. The robust semantic communication beamforming method for a MISO channel according to claim 1, characterized in that: The optimal beam design for a MISO channel with bounded channel estimation error includes: The bounded channel state information error is expressed as: Δh H AΔh≤c Where A represents the Hermitian matrix; Under the condition of bounded channel state information error, the signal-to-noise ratio received by the receiver is expressed as: The problem of minimizing the transmission power while satisfying the multi-scale structural similarity constraint under a MISO channel with bounded channel estimation error is proposed, which can be expressed as: Δh H AΔh≤c According to the ABG formula, the constraint problem is transformed into: Δh H AΔh≤c in, represents the minimum required signal-to-noise ratio; The S-lemma and semi-positive definite relaxation method are used in turn to transform the constraint problem into a convex semi-positive definite programming subproblem, which can be expressed as: λ ≥ 0 The CVX toolbox is used to solve the transformed convex semidefinite programming subproblem and obtain the optimal beam in the MISO channel with bounded channel estimation error.

Images