A robust semantic communication method, device and equipment based on sparse vector coding

Abstract

The application relates to the technical field of wireless communication, and discloses a robust semantic communication method, device and equipment based on sparse vector coding, which comprises the following steps: extracting continuous semantic features of original data through a semantic encoder model; discretizing the continuous semantic features into index bits through a vector quantization method by using a learnable feature dictionary; mapping the index bits into sparse vectors by using sparse vector coding, and transmitting the sparse vectors after codebook extension; and identifying the index and reconstructing the original data by using a multi-path matching tracking algorithm at a receiving end. By fusing a semantic feature extraction model, a vector quantization technology and sparse vector coding, and by designing a gradient approximation, a joint loss function optimization and a transmission parameter dynamic adjustment mechanism, the application effectively solves the problems of poor compatibility and weak channel adaptability of traditional semantic communication digital transmission, and improves the semantic transmission reliability and system resource utilization efficiency under dynamic channel conditions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless communication, and in particular to a robust semantic communication method, device and equipment based on sparse vector coding. BACKGROUND

[0002] In view of the massive communication demand in a dynamic channel environment, semantic communication is a key technology to improve the efficiency of intelligent task execution and the efficiency of system resource allocation. In typical dynamic scenarios such as industrial Internet of Things and Internet of Vehicles, the channel state often shows significant time-varying and non-stationary characteristics. Therefore, how to design a robust semantic communication scheme compatible with existing digital communication systems and adaptive to channel changes is an important link to realize high-reliability and high-efficiency information transmission.

[0003] The existing mainstream semantic communication scheme mainly adopts Deep Joint Source-Channel Coding (DJSCC) based on deep learning, which directly learns the mapping relationship between source information and channel symbols through a deep neural network to realize efficient transmission of semantics. However, such methods have obvious defects: first, they rely on analog transmission methods, which hinder their integration in actual digital communication systems; second, they have poor channel adaptability, as the model parameters are fixed after training and cannot respond to dynamic changes in channel state in real time. SUMMARY

[0004] Therefore, the present application provides a robust semantic communication method, device and equipment based on sparse vector coding to solve the problems of digital communication incompatibility and poor adaptability to channel changes in semantic communication.

[0005] In a first aspect, the present application provides a robust semantic communication method based on sparse vector coding, which is applied to a sending end, and the method comprises:

[0006] Obtaining original data and encoding the original data using a semantic encoder to extract continuous semantic features;

[0007] Based on a learnable feature dictionary, the continuous semantic features are vector quantized to obtain discrete code word indexes corresponding to the continuous semantic features;

[0008] Sparse vectors are constructed according to the discrete code word indexes, and the sparse vectors are randomly spread using a non-orthogonal codebook to obtain spread spectrum signals;

[0009] The spread spectrum signals are loaded onto multiple subcarriers and sent to a receiving end through a target channel.

[0010] The robust semantic communication method based on sparse vector coding provided by the application can solve the technical challenge that the traditional semantic communication scheme is difficult to deploy integration in the actual digital network by introducing vector quantization technology to discretize continuous semantic features into bit streams, so that the semantic information can be compatible with the modulation and demodulation and coding and decoding processes of the existing digital communication system, and the key parameters of the sparse vector transmission technology can be dynamically adjusted according to the real-time channel state without relying on complex neural network retraining or fine-tuning, thereby solving the technical challenge that the traditional semantic communication scheme cannot adapt to the dynamically changing channel conditions in real time.

[0011] In an optional implementation, the continuous semantic features are vector quantized based on a learnable feature dictionary to obtain discrete code word indexes corresponding to the continuous semantic features, including:

[0012] The learnable feature dictionary and the semantic encoder are jointly trained and optimized to obtain a discretized feature dictionary representation.

[0013] Based on the discretized feature dictionary representation, the continuous semantic features are discretized into discrete code word indexes by using a nearest neighbor method.

[0014] The robust semantic communication method based on sparse vector coding provided by the application can solve the technical challenge that the traditional semantic communication scheme is difficult to deploy integration in the actual digital network by introducing vector quantization technology to discretize continuous semantic features into bit streams, so that the semantic information can be compatible with the modulation and demodulation and coding and decoding processes of the existing digital communication system, and the key parameters of the sparse vector transmission technology can be dynamically adjusted according to the real-time channel state without relying on complex neural network retraining or fine-tuning, thereby solving the technical challenge that the traditional semantic communication scheme cannot adapt to the dynamically changing channel conditions in real time.

[0015] In an optional implementation, the discrete code word index includes an index bit and a modulation bit, and the sparse vector is constructed according to the discrete code word index, including:

[0016] According to the index bit, the position index of the non-zero element in the sparse vector is determined by using a sparse transformation mapping table, and the modulation constellation value and constellation rotation angle of the non-zero element are determined according to the modulation bit.

[0017] The sparse vector is constructed according to the position index, the modulation constellation value and the constellation rotation angle of the non-zero element.

[0018] In an optional implementation, the method further includes:

[0019] The real-time channel state of the target channel is obtained, and at least one parameter of the sparse vector is adjusted according to the real-time channel state.

[0020] Based on the adjusted parameter values, a sparse vector is constructed by combining the position index of the non-zero elements, the modulation constellation value, and the constellation rotation angle.

[0021] The robust semantic communication method based on sparse vector coding provided by this invention uses sparse vector coding to transmit discrete index values, carries information in the form of sparse indexes, and decodes it at the receiving end using compressed sensing. Key transmission parameters can be flexibly adjusted according to channel conditions, which can improve the model's adaptability to dynamically changing channel environments.

[0022] Secondly, this invention provides a robust semantic communication method based on sparse vector coding, applied at the receiving end, the method comprising:

[0023] The spread spectrum signal transmitted through the target channel is received, and the spread spectrum signal is sparsely recovered using a multipath matching algorithm to obtain a discrete codeword index.

[0024] Based on the discrete codeword index and the learnable feature dictionary, the continuous semantic features corresponding to the discrete codeword index are determined.

[0025] The continuous semantic features obtained are reconstructed using a semantic decoder to obtain reconstructed data. The semantic decoder, semantic encoder, and learnable feature dictionary are then jointly trained and optimized to obtain the reconstructed data.

[0026] In one optional implementation, a multipath matching algorithm is used to perform sparse recovery of the spread spectrum signal to obtain a discrete codeword index, including:

[0027] Obtain the transmission configuration parameters and determine the search path order list based on the transmission configuration parameters;

[0028] Based on the search path sequence list, iterative calculations are performed on each search path to determine the index of non-zero elements and the corresponding non-zero element values ​​under each search path.

[0029] Calculate and compare the residuals of each search path based on the non-zero element indices and corresponding non-zero element values.

[0030] The search path with the smallest residual is selected as the optimal solution, and the index bits and modulation bits are reconstructed based on the optimal solution.

[0031] The robust semantic communication method based on sparse vector coding provided by this invention, combined with sparse vector transmission technology, allows its key parameters to be dynamically adjusted according to the real-time channel state without relying on complex neural network retraining or fine-tuning. This solves the technical challenge of traditional semantic communication schemes being unable to adapt to dynamically changing channel conditions in real time. Through the design of vector quantization and gradient stopping, the problem of non-differentiability of the quantization stage in neural network training is solved. By using dictionary learning and basis vector representation, only the index needs to be transmitted instead of the complete feature vector, which can reduce the amount of data transmitted and reduce the semantic loss caused by quantization.

[0032] Thirdly, the present invention provides a robust semantic communication device based on sparse vector coding, the device being located at the transmitting end, comprising:

[0033] The data encoding module is used to acquire raw data and encode the raw data using a semantic encoder to extract continuous semantic features;

[0034] The vector quantization module is used to perform vector quantization on continuous semantic features based on a learnable feature dictionary to obtain the discrete codeword index corresponding to the continuous semantic features.

[0035] The sparse vector construction module is used to construct sparse vectors based on discrete codeword indices and to randomly spread the sparse vectors using a non-orthogonal codebook to obtain a spread spectrum signal.

[0036] The data transmission module is used to load the spread spectrum signal onto multiple subcarriers and transmit it to the receiving end through the target channel.

[0037] Fourthly, the present invention provides a robust semantic communication device based on sparse vector coding, the device being located at the receiving end, comprising:

[0038] The data receiving module is used to receive the spread spectrum signal transmitted through the target channel and use a multipath matching algorithm to perform sparse recovery of the spread spectrum signal to obtain a discrete codeword index.

[0039] The codeword indexing module is used to determine the continuous semantic features corresponding to the discrete codeword index based on the discrete codeword index and the learnable feature dictionary;

[0040] The data decoding module is used to reconstruct the obtained continuous semantic features using the semantic decoder to obtain reconstructed data.

[0041] Fifthly, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the method of the first aspect or any corresponding embodiment thereof, or the method of the second aspect or any corresponding embodiment thereof.

[0042] In a sixth aspect, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to perform the method of the first aspect or any corresponding embodiment thereof, or the method of the second aspect or any corresponding embodiment thereof. Attached Figure Description

[0043] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the application of the robust semantic communication method based on sparse vector coding according to an embodiment of the present invention to the sending end.

[0045] Figure 2 This is a schematic diagram illustrating data transmission using a robust semantic communication method based on sparse vector coding according to an embodiment of the present invention.

[0046] Figure 3 This is a schematic diagram illustrating the principle of the codebook vector quantization model in the robust semantic communication method based on sparse vector coding according to an embodiment of the present invention.

[0047] Figure 4 This is a schematic diagram illustrating the performance relationship between the minimum number of transmitted subcarriers and the channel signal-to-noise ratio in a robust semantic communication method based on sparse vector coding according to an embodiment of the present invention.

[0048] Figure 5 This is a flowchart illustrating the application of a robust semantic communication method based on sparse vector coding according to an embodiment of the present invention to the receiving end.

[0049] Figure 6 This is a schematic diagram comparing the peak signal-to-noise ratio performance of different transmission methods in a specific embodiment of the robust semantic communication method based on sparse vector coding according to an embodiment of the present invention.

[0050] Figure 7 This is a structural block diagram of a robust semantic communication device based on sparse vector coding located at the transmitting end according to an embodiment of the present invention;

[0051] Figure 8 This is a structural block diagram of a robust semantic communication device based on sparse vector coding located at the receiving end according to an embodiment of the present invention;

[0052] Figure 9This is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] For semantic communication scenarios in complex dynamic channel environments, such as the Industrial Internet of Things (IIoT) and dynamic unmanned swarms in the Internet of Vehicles (IoV), achieving highly reliable and efficient transmission of semantic information is a core requirement supporting such communication scenarios. In these scenarios, on the one hand, the key semantic information contained in the transmitted content, such as device status, control commands, and environmental understanding, often has highly generalized characteristics, and semantic coding can significantly reduce the amount of data transmitted; on the other hand, wireless channel conditions, such as signal-to-noise ratio and multipath fading characteristics, exhibit significant time-varying characteristics. Therefore, directly deploying an end-to-end semantic communication scheme trained under specific channel conditions and based on continuous analog value transmission into actual dynamic channels can easily lead to severe performance degradation, failing to simultaneously meet the dual requirements of efficient and highly reliable transmission in dynamic environments.

[0055] The relevant technologies and existing technical problems include:

[0056] a. Using a convolutional neural network to directly map the original image into channel transmission symbols can achieve a significant performance improvement over traditional methods when channel conditions are poor. However, this method uses analog transmission, which lacks compatibility with existing mature digital communication systems, thus limiting the practical deployment of semantic communication; moreover, the semantic transmission model needs to be trained under fixed channel conditions, and the model performance degrades significantly when the actual deployment environment differs from the training environment.

[0057] b. A feature attention module is added to the network to dynamically adjust feature weights based on the channel signal-to-noise ratio and the mean of transmission features, enabling a single model to adapt to different channel states. This method uses an attention mechanism for channel adaptation, but the model has a large computational load and many parameters, making it unsuitable for deployment on resource-constrained devices.

[0058] c. A predictive network is used to combine features, channel conditions, and compression ratio to predict image transmission quality, thereby achieving optimal bitrate control under a given objective. This method employs analog transmission, which lacks compatibility with existing mature digital communication systems, thus limiting the practical deployment of semantic communication.

[0059] Based on the problems existing in the above-mentioned related technologies, the embodiments of the present invention provide a robust semantic communication method based on sparse vector coding. By discretizing continuous semantic features into a bit stream through vector quantization technology and dynamically adjusting key parameters in combination with sparse vector transmission technology, the method achieves digital compatibility of semantic transmission and adaptive capability for dynamic channels.

[0060] According to an embodiment of the present invention, a robust semantic communication method based on sparse vector coding is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0061] This embodiment provides a robust semantic communication method based on sparse vector coding, which can be used in the aforementioned computer system. Figure 1 This is a flowchart of a robust semantic communication method based on sparse vector coding according to an embodiment of the present invention. The method is applied to the sending end of semantic communication, such as... Figure 1 As shown, the process includes the following steps:

[0062] Step S101: Obtain the original data and encode the original data using a semantic encoder to extract continuous semantic features.

[0063] Specifically, such as Figure 2 The diagram illustrates data transmission using a robust semantic communication method based on sparse vector coding. The sending end determines the original data to be sent, including but not limited to image data, audio data, and text data. A semantic encoder encodes the original data and extracts semantic features. The semantic encoder includes a semantic feature extraction module constructed using a convolutional neural network, used to map the original data into low-dimensional continuous semantic features. Both the semantic feature extraction module in the semantic encoder and the semantic feature reconstruction module in the semantic decoder use convolutional neural network models as the semantic transmission model to achieve compressed mapping from original information to low-dimensional continuous semantic features, and semantic recovery from quantized features to reconstructed information.

[0064] Step S102: Based on the learnable feature dictionary, the continuous semantic features are vector quantized to obtain the discrete codeword index corresponding to the continuous semantic features.

[0065] Specifically, to ensure the semantic transmission model is compatible with digital communication systems, a dimension is constructed. Learnable feature dictionary, where, The number of codebooks, i.e., basis vectors The number of values ​​corresponds to the range of output indices; The feature dimension is the number of output channels of the semantic feature extraction module. correspond.

[0066] The semantic feature extraction process involves finding the discrete codewords representing the corresponding features from the learnable feature dictionary using a nearest-neighbor approach, and outputting the discrete codeword index corresponding to the discrete codeword. Used for transmission, enabling the transfer of continuous semantic features To discrete index The transformation, the specific process of vector quantization is as follows Figure 3 As shown, only the index represented by low-dimensional bits needs to be transmitted. Instead of transmitting the complete feature vector The codewords in the feature dictionary are shared as priors at both the sending and receiving ends, significantly reducing the amount of data transmitted.

[0067] Step S103: Construct a sparse vector based on the discrete codeword index, and randomly spread the sparse vector using a non-orthogonal codebook to obtain a spread spectrum signal.

[0068] Specifically, to ensure highly reliable transmission and dynamic adaptability of semantic features, sparse vector coding, which has significant advantages in short packet communication, is introduced during the actual deployment and transmission phase. By carrying information in the form of an index, transmission performance under adverse channel conditions is improved. Furthermore, since the key transmission parameters of sparse vector coding are not determined by model training, the transmission parameters can be flexibly adjusted according to channel conditions to achieve dynamic environment adaptation.

[0069] The transmitting end first splits the original data bits to be transmitted into two parts. One part is used as the index part bits to determine the position index of the non-zero element in the sparse vector (the mapping relationship between the index and the bit is preset by the mapping table and must satisfy a specific logarithmic relationship constraint); the other part is used as the modulation part bit group to determine the modulation symbol and load it onto the non-zero element value to complete the construction of the sparse vector.

[0070] After the sparse vectors are constructed, they are distributed using a Bernoulli codebook matrix (a non-orthogonal matrix). Implementing sparse vectors Random spread spectrum, and loaded into The spread spectrum signal transmitted on each subcarrier can be represented as:

[0071] (1)

[0072] in, For non-zero indices in the codebook The corresponding spreading codeword, For the corresponding modulation constellation value, The corresponding constellation rotation angle.

[0073] Step S104: Load the spread spectrum signal onto multiple subcarriers and transmit it to the receiving end through the target channel.

[0074] Specifically, the signal obtained by non-orthogonal codebook spread spectrum is distributed to multiple subcarriers and transmitted to the receiving end through the target channel based on the set channel transmission parameters.

[0075] The robust semantic communication method based on sparse vector coding provided in this embodiment discretizes continuous semantic features into a bit stream by introducing vector quantization technology. This enables semantic information to be compatible with the modulation, demodulation, encoding, and decoding processes of existing digital communication systems, solving the technical challenge of traditional semantic communication schemes being difficult to deploy and integrate in actual digital networks. Combined with sparse vector transmission technology, its key parameters do not need to rely on complex neural network retraining or fine-tuning, and can be dynamically adjusted according to the real-time channel state, solving the technical challenge of traditional semantic communication schemes being unable to adapt to dynamically changing channel conditions in real time. By adopting an index-based information carrying format and a compression sensing-based decoding method, it effectively combats channel fading and noise interference, enhances the reliability of semantic transmission, and solves the technical challenge of the sharp performance degradation of traditional digital semantic communication schemes in harsh channel environments.

[0076] In some optional implementations, step S102 above includes:

[0077] Step S1021: Jointly train and optimize the learnable feature dictionary and the semantic encoder to obtain a discretized feature dictionary representation.

[0078] Specifically, the semantic feature extraction module, the semantic feature reconstruction module, and the vector quantized feature dictionary are jointly trained and optimized to obtain efficient low-dimensional discretized semantic feature representations. The sparse vector encoding transmission module does not participate in model training to ensure the model's differentiability.

[0079] The loss function of the semantic transport model is designed to be jointly optimized by the semantic encoder, semantic decoder, and learnable feature dictionary, while maintaining different learning rates, which significantly reduces the semantic quantization loss.

[0080] The loss function setting during training considers the following two aspects: first, minimizing the difference between the original image and the reconstructed image, using mean squared error for evaluation; that is... This item is used to optimize the feature extraction and reconstruction modules; secondly, it aims to minimize the loss caused by quantization, and optimize the output of the feature extraction module. and basis vectors in the feature dictionary To be as close as possible, that is Here, considering the inconsistency in learning rates between the feature extraction module and the feature dictionary, this item is divided into... Approaching Obtain a better feature dictionary and Approaching This ensures that the output of the feature extraction module does not deviate excessively from the two parts of the feature dictionary. and That is, the total loss function for the training process is set as follows:

[0081] (2)

[0082] in, Indicates the loss value. Indicates the transmission of signals. This represents the feature vector to be encoded. This represents the basis vectors in the feature dictionary, and sg is the gradient stopping operator. During forward propagation, However, during reverse gradient descent, .

[0083] After the training process is completed, the semantic transmission model can be used to obtain better semantic encoding and decoding and discrete representation results.

[0084] Step S1022: Based on the discretized feature dictionary representation, the continuous semantic features are discretized into discrete codeword indexes using the nearest neighbor method.

[0085] Specifically, the stopping gradient operator sg is used in the quantization phase to solve the non-differentiability problem, i.e., the forward propagation output... Backpropagation transmits the gradient values ​​to Since the quantization process is considered gradient-free, the input to the semantic decoder can be represented as:

[0086] (3)

[0087] in, This represents the input to the semantic decoder, during forward transmission. However, during reverse gradient descent, By defining the gradient function To meet the required functionality, since the quantization process itself is not differentiable, its gradient is directly set to 0, skipping the gradient update of the quantization process and ensuring the overall model's differentiability.

[0088] Assuming the transmitted signal The input image is processed by a semantic feature extraction module, which outputs low-dimensional features. The quantization process refers to searching for features in the feature dictionary that match the given features. The basis vector with the smallest distance and output its index. The transformation formula for discretized semantic feature representation is:

[0089] (4)

[0090] in, The feature vector to be encoded. For learnable feature dictionary and The corresponding code words, Index to be transmitted (from 1 to ...) (discrete numerical values) This represents the number of codewords in the feature dictionary. Discretizing continuous semantic features into discrete codeword indices achieves a discretized representation of semantic features; only the indices need to be transmitted during transmission. The basis vectors, as prior knowledge of both the sender and receiver, do not need to be transmitted.

[0091] The robust semantic communication method based on sparse vector coding provided in this embodiment obtains an efficient low-dimensional discretized semantic feature representation by jointly training and optimizing a learnable feature dictionary and a semantic encoder. This effectively solves the problem of poor compatibility of traditional semantic communication digital transmission. The sparse vector coding transmission does not participate in model training to ensure the model's differentiability, thereby improving the reliability of semantic transmission.

[0092] In some optional implementations, the discrete codeword index includes: index bits and modulation bits. The step S103 above, which constructs a sparse vector based on the discrete codeword index, includes:

[0093] Step S1031: Based on the index bits, determine the position index of the non-zero element in the sparse vector using the sparse transform mapping table, and determine the modulation constellation value and constellation rotation angle of the non-zero element based on the modulation bits.

[0094] Step S1032: Construct a sparse vector based on the position index of the non-zero elements, the modulation constellation value, and the constellation rotation angle.

[0095] The sending end will transmit Each information bit is divided into index bits. and modulation bits The index bits determine the position index of the non-zero element in the sparse vector through a sparse transformation mapping table, and satisfy the relationship... ,in For the sparse vector length, The number of non-zero elements; modulation bits The modulation is based on the constellation values ​​of quadrature amplitude modulation (M-order QAM modulation), and these values ​​are used as non-zero element values. And satisfy .

[0096] In some alternative implementations, the method further includes:

[0097] Obtain the real-time channel state of the target channel and adjust at least one parameter of the sparse vector based on the real-time channel state.

[0098] Based on the adjusted parameter values, a sparse vector is constructed by combining the position index of the non-zero elements, the modulation constellation value, and the constellation rotation angle.

[0099] Specifically, the sparse vector length is dynamically optimized based on real-time channel conditions, such as signal-to-noise ratio. Number of non-zero elements Number of bits transmitted Number of transmitted subcarriers These parameters are illustrated using signal-to-noise ratio as an example in this embodiment. Figure 4 The diagram illustrates the relationship between the minimum number of transmitted subcarriers required for the system to reach a fixed peak signal-to-noise ratio (SNR) threshold of 20 dB and the SNR. Figure 4 It can be seen that as the signal-to-noise ratio (SNR) increases, the required number of subcarriers decreases significantly. At an SNR of 10 dB, only 16% of the required number of transmission subcarriers is needed to achieve the same performance as at an SNR of -5 dB. Therefore, the number of transmission subcarriers can be dynamically optimized based on real-time channel conditions (including but not limited to SNR). Transmission parameters are adjusted such that the coding rate is reduced at low signal-to-noise ratios (SNR) to improve transmission reliability, while the coding rate is increased at high SNRs to reduce transmission resource consumption. This flexible adjustment of transmission resources dynamically adapts to changing channel environments, improving system adaptability and optimizing system resource allocation efficiency.

[0100] The robust semantic communication method based on sparse vector coding provided in this embodiment uses sparse vector coding to transmit discrete index values, carries information in the form of sparse indexes, and decodes it at the receiving end using compressed sensing. Key transmission parameters can be flexibly adjusted according to channel conditions, which can improve the model's adaptability to dynamically changing channel environments.

[0101] This embodiment provides a robust semantic communication method based on sparse vector coding, which can be used in the aforementioned computer system. Figure 5 This is a flowchart of a robust semantic communication method based on sparse vector coding according to an embodiment of the present invention, applied to the receiving end of semantic communication, such as... Figure 5 As shown, the process includes the following steps:

[0102] Step S201: Receive the spread spectrum signal transmitted through the target channel, and use the multipath matching algorithm to perform sparse recovery on the spread spectrum signal to obtain the discrete codeword index.

[0103] Specifically, the multipath matching pursuit (MMP) algorithm is used to perform sparse recovery of the spread spectrum signal, after passing through a channel with the following characteristics: The received signal after the channel can be represented as the observation matrix. Zhong Mou List by non-zero element values For a linear combination of coefficients, the expression is as follows:

[0104] (5)

[0105] in, In Orthogonal Frequency Division Multiplexing (OFDM) channels Channel gain of each subcarrier The variance is Additive white Gaussian noise.

[0106] The decoding process can be modeled as a sparse recovery problem in compressed sensing, expressed as:

[0107] (6)

[0108] in, For the estimated sparse vector index, In order to receive signals, For the candidate index set, This is the observation matrix.

[0109] By constructing multiple parallel search paths, and calculating the observation matrix in each iteration... By correlating with the residuals, the non-zero position indices in the sparse vector are estimated, and the least squares method is used to iteratively update the residuals to obtain all the non-zero position indices and element values.

[0110] Step S202: Based on the discrete codeword index and the learnable feature dictionary, determine the continuous semantic features corresponding to the discrete codeword index.

[0111] Specifically, such as Figure 2 As shown, the receiving end uses the learnable feature dictionary in the semantic database to dequantize the discrete codeword index, and retrieves the corresponding codeword from the learnable feature dictionary according to the transmitted discrete codeword index. Replace the original feature vector As input to the semantic feature reconstruction module in the semantic decoder, it completes the inverse quantization process.

[0112] The quantization and dequantization phases use the stopping gradient operator sg to solve the non-differentiability problem, i.e., the forward propagation output... Backpropagation transmits the gradient values ​​to The quantization process is considered to be gradient-free, so the input of the semantic decoder is as shown in formula (3), which will not be elaborated here.

[0113] Step S203: The obtained continuous semantic features are reconstructed using the semantic decoder to obtain reconstructed data. The semantic decoder, semantic encoder, and learnable feature dictionary are jointly trained and optimized to obtain the reconstructed data.

[0114] Specifically, in the end-to-end model training phase, a semantic feature extraction model and a semantic feature reconstruction module are constructed using convolutional neural networks, enabling the sending end to extract semantic features from the raw data. To low-dimensional continuous semantic features The receiving end uses low-dimensional continuous semantic features To reconstruct data The transformation.

[0115] The robust semantic communication method based on sparse vector coding provided in this embodiment, combined with sparse vector transmission technology, allows its key parameters to be dynamically adjusted according to the real-time channel state without relying on complex neural network retraining or fine-tuning. This solves the technical challenge of traditional semantic communication schemes being unable to adapt to dynamically changing channel conditions in real time. Through the design of vector quantization and gradient stopping, the problem of non-differentiability in the quantization stage during neural network training is solved. By using dictionary learning and basis vector representation, only the index needs to be transmitted instead of the complete feature vector, which can reduce the amount of data transmitted and reduce the semantic loss caused by quantization.

[0116] In an optional implementation, step S201 above includes:

[0117] Step S2011: Obtain transmission configuration parameters and determine the search path sequence table based on the transmission configuration parameters.

[0118] Specifically, before the iteration, based on the number of non-zero elements Number of sub-paths at each level and maximum number of search paths The search path sequence list is calculated, and several promising search paths are pre-determined as candidate search path sequence lists:

[0119] (7)

[0120] in, For the order of the search paths, The sub-path number selected for each iteration indicates that the inner product is sorted in descending order and then the sorting is performed. The corresponding index.

[0121] Step S2012: Iteratively calculate each search path based on the search path sequence list to determine the non-zero element index and corresponding non-zero element value under each search path.

[0122] Specifically, iterative calculations are performed on each search path based on the search path sequence list, including: calculating the observation matrix. Each column and residual The inner product of the residuals, where the initial values ​​of the residuals are... .

[0123] Select the items arranged in descending order of their inner products. The index of the corresponding column is used as an estimate of the index of the non-zero element in the sparse vector. This process can be expressed as:

[0124] (8)

[0125] in, The index of the non-zero element estimated for this layer. For the observation matrix, This is the residual.

[0126] Update the obtained non-zero element values ​​using the least squares method:

[0127] (9)

[0128] in, For the front Estimates of non-zero elements, For the front The non-zero element indices correspond to The combination of the middle column, Representation matrix The pseudo-inverse matrix, i.e. .

[0129] Update residuals .

[0130] Repeat the above process until all non-zero element indices and corresponding non-zero element values ​​are found under the corresponding search path.

[0131] Step S2013: Calculate and compare the residuals of each search path based on the non-zero element indices and corresponding non-zero element values ​​under each search path.

[0132] Specifically, for each search path, the residual of the index set (including all non-zero element indices and corresponding non-zero element values ​​under the corresponding search path) is calculated, and the index set with the smallest residual is selected to obtain the final sparse vector, i.e.:

[0133] (10)

[0134] in, This represents the sparse vector set obtained from each path search.

[0135] Step S2014: Select the search path with the smallest residual as the optimal solution, and restore the index bit and modulation bit based on the optimal solution.

[0136] Specifically, in the multipath matching pursuit sparse recovery algorithm, the observation matrix is ​​calculated separately by constructing a multipath search space. The inner product of the residuals is used to iteratively estimate the non-zero index, and the least squares method is used to update the non-zero value estimate and the residuals. Finally, the path with the minimum residuals is selected as the optimal solution.

[0137] After obtaining the sparse vector estimate through the MMP algorithm, the index of the non-zero element is obtained by inverse mapping with the aid of the sparse transformation table, and the non-zero element value can be demodulated to obtain the modulation bit part, thus completing the sparse vector transmission process.

[0138] In one specific embodiment, the performance of the semantic communication system was verified through simulation experiments. The system was tested using the Modified National Institute of Standards and Technology (MNIST) database dataset under Rayleigh fading channel conditions. The system compression ratio was set to 1 / 16, the codebook size to 16, the codebook dimension to 128, the number of bits represented by a single sparse vector to 16, the number of non-zero elements to 2, and the modulation scheme to QPSK. The relationship between the peak signal-to-noise ratio (PSNR) of the transmitted image and the channel signal-to-noise ratio was obtained by changing the transmission signal-to-noise ratio (SNR) as follows: Figure 6 As shown, the number of transmission subcarriers is fixed at 32 at this time. From Figure 6 As can be seen, the semantic transmission method proposed in this invention outperforms both traditional digital semantic communication methods without sparse vector coding and traditional methods without semantic communication, and exhibits significant advantages at different signal-to-noise ratios. For example, at a signal-to-noise ratio of 10 dB, the method proposed in this invention improves performance by 26% compared to traditional digital semantic communication methods and by 80% compared to traditional non-semantic methods, fully demonstrating the advantages of the proposed method in digital communication compatibility and improved semantic transmission reliability.

[0139] This embodiment also provides a robust semantic communication device based on sparse vector coding, which is used to implement the above embodiments and preferred embodiments; details already described will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0140] This embodiment provides a robust semantic communication device based on sparse vector coding. The device is located at the transmitting end, such as... Figure 7 As shown, it includes:

[0141] The data encoding module 701 is used to acquire raw data and encode the raw data using a semantic encoder to extract continuous semantic features.

[0142] The vector quantization module 702 is used to perform vector quantization on continuous semantic features based on a learnable feature dictionary to obtain discrete codeword indices corresponding to the continuous semantic features.

[0143] The sparse vector construction module 703 is used to construct sparse vectors based on discrete codeword indices and to randomly spread the sparse vectors using a non-orthogonal codebook to obtain a spread spectrum signal.

[0144] The data transmission module 704 is used to load the spread spectrum signal onto multiple subcarriers and transmit it to the receiving end through the target channel.

[0145] This embodiment provides a robust semantic communication device based on sparse vector coding. The device is located at the receiving end, such as... Figure 8 As shown, it includes:

[0146] The data receiving module 801 is used to receive the spread spectrum signal transmitted through the target channel and use a multipath matching algorithm to perform sparse recovery of the spread spectrum signal to obtain a discrete codeword index.

[0147] The codeword index module 802 is used to determine the continuous semantic features corresponding to the discrete codeword index based on the discrete codeword index and the learnable feature dictionary.

[0148] The data decoding module 803 is used to reconstruct the obtained continuous semantic features using the semantic decoder to obtain reconstructed data.

[0149] Further functional descriptions of the above modules and units are the same as those in the corresponding embodiments described above, and will not be repeated here.

[0150] In this embodiment, the robust semantic communication device based on sparse vector coding is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0151] This invention also provides a computer device having the above-described features. Figure 7 or Figure 8 The robust semantic communication device based on sparse vector coding is shown.

[0152] Please see Figure 9 , Figure 9 This is a schematic diagram of the structure of a computer device provided in an optional embodiment of the present invention, such as... Figure 9 As shown, the computer device includes one or more processors 10, memory 20, and interfaces for connecting the components, including high-speed interfaces and low-speed interfaces. The components communicate with each other via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on external input / output devices (such as display devices coupled to the interfaces). In some alternative implementations, multiple processors and / or multiple buses can be used with multiple memories and multiple memory modules, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 9 Take a processor 10 as an example.

[0153] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GPA), or any combination thereof.

[0154] The memory 20 stores instructions executable by at least one processor 10 to cause the at least one processor 10 to perform the method shown in the above embodiments.

[0155] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function; the data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0156] The memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk or solid-state drive; the memory 20 may also include a combination of the above types of memory.

[0157] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0158] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0159] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A robust semantic communication method based on sparse vector coding, characterized in that, The method is applied at the sending end, and the method includes: Obtain the raw data and encode the raw data using a semantic encoder to extract continuous semantic features; Based on a learnable feature dictionary, the continuous semantic features are vector-quantized to obtain the discrete codeword index corresponding to the continuous semantic features; A sparse vector is constructed based on the discrete codeword index, and the sparse vector is randomly spread using a non-orthogonal codebook to obtain a spread spectrum signal. The spread spectrum signal is loaded onto multiple subcarriers and transmitted to the receiving end through the target channel.

2. The method according to claim 1, characterized in that, Based on a learnable feature dictionary, the continuous semantic features are vector-quantized to obtain discrete codeword indices corresponding to the continuous semantic features, including: The learnable feature dictionary and the semantic encoder are jointly trained and optimized to obtain a discretized feature dictionary representation. Based on the discretized feature dictionary representation, the continuous semantic features are discretized into discrete codeword indices using the nearest neighbor method.

3. The method according to claim 1, characterized in that, The discrete codeword index includes: index bits and modulation bits, and the construction of a sparse vector based on the discrete codeword index includes: Based on the index bits, the position index of the non-zero element in the sparse vector is determined using a sparse transform mapping table, and the modulation constellation value and constellation rotation angle of the non-zero element are determined based on the modulation bits. A sparse vector is constructed based on the position index of the non-zero element, the modulation constellation value, and the constellation rotation angle.

4. The method according to claim 3, characterized in that, The method further includes: Obtain the real-time channel state of the target channel, and adjust at least one parameter of the sparse vector according to the real-time channel state; Based on the adjusted parameter values, and combined with the position index of the non-zero elements, the modulation constellation value, and the constellation rotation angle, a sparse vector is constructed.

5. A robust semantic communication method based on sparse vector coding, characterized in that, The method is applied at the receiving end, and the method includes: The spread spectrum signal transmitted through the target channel is received, and the spread spectrum signal is sparsely recovered using a multipath matching algorithm to obtain a discrete codeword index. Based on the discrete codeword index and the learnable feature dictionary, the continuous semantic features corresponding to the discrete codeword index are determined; The obtained continuous semantic features are reconstructed using a semantic decoder to obtain reconstructed data. The semantic decoder is jointly trained and optimized with the semantic encoder and the learnable feature dictionary.

6. The method according to claim 5, characterized in that, The spread spectrum signal is sparsely recovered using a multipath matching algorithm to obtain a discrete codeword index, including: Obtain the transmission configuration parameters and determine the search path order list based on the transmission configuration parameters; Based on the search path sequence list, iterative calculations are performed on each search path to determine the non-zero element index and the corresponding non-zero element value under each search path. Calculate and compare the residuals of each search path based on the non-zero element indices and corresponding non-zero element values. The search path with the smallest residual is selected as the optimal solution, and the index bits and modulation bits are then reconstructed based on the optimal solution.

7. A robust semantic communication device based on sparse vector coding, characterized in that, The device is located at the transmitting end and includes: The data encoding module is used to acquire raw data and encode the raw data using a semantic encoder to extract continuous semantic features; The vector quantization module is used to perform vector quantization on the continuous semantic features based on a learnable feature dictionary to obtain the discrete codeword index corresponding to the continuous semantic features; A sparse vector construction module is used to construct sparse vectors based on the discrete codeword index, and to randomly spread the sparse vectors using a non-orthogonal codebook to obtain a spread spectrum signal. The data transmission module is used to load the spread spectrum signal onto multiple subcarriers and transmit it to the receiving end through the target channel.

8. A robust semantic communication device based on sparse vector coding, characterized in that, The device is located at the receiving end and includes: The data receiving module is used to receive the spread spectrum signal transmitted through the target channel and perform sparse recovery on the spread spectrum signal using a multipath matching algorithm to obtain a discrete codeword index. The codeword indexing module is used to determine the continuous semantic features corresponding to the discrete codeword index based on the discrete codeword index and the learnable feature dictionary; The data decoding module is used to reconstruct the obtained continuous semantic features using the semantic decoder to obtain reconstructed data.

9. A computer device, characterized in that, include: A memory and a processor are communicatively connected, the memory storing computer instructions, and the processor executing the computer instructions to perform the robust semantic communication method based on sparse vector coding as described in any one of claims 1 to 4 or 5 to 6.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the robust semantic communication method based on sparse vector coding as described in any one of claims 1 to 4 or 5 to 6.

Images