Phase shift feedback method for reconfigurable intelligent surface
By using the feature index feedback method in the RIS system, the problems of large data volume and time delay in RIS phase shift feedback are solved, and efficient phase accuracy feedback is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- PENG CHENG LAB
- Filing Date
- 2023-04-13
- Publication Date
- 2026-07-03
AI Technical Summary
In RIS-assisted wireless communication systems, the feedback RIS phase shift requires a huge amount of data, resulting in excessive consumption of spectrum and time resources, and making it difficult to guarantee phase accuracy.
By acquiring the phase shift matrix to be transmitted, the feature sequence number is determined using the target encoder and the shared knowledge base, and then sent as a bit stream to the reconfigurable smart surface device. The target decoder is used to reconstruct the received phase shift matrix, thereby reducing the amount of feedback data.
While ensuring phase accuracy, the amount of feedback data and communication resource consumption are significantly reduced, and the feedback latency is lowered.
Smart Images
Figure CN116760441B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication technology, and in particular to a phase shift feedback method for a reconfigurable smart surface. Background Technology
[0002] Reconfigurable Intelligence Surface (RIS) is a promising technology in next-generation wireless communication systems, capable of enhancing millimeter-wave signal coverage and improving spectral and energy efficiency with low power consumption and low hardware cost. RIS can intelligently control the reflected phase of incident electromagnetic waves using a large number of low-cost passive components, thereby enabling wireless propagation environment reconfiguration. When the line-of-sight link between the User Equipment (UE) and the Base Station (BS) is blocked, RIS can improve the link quality of the UE and enhance the performance of the communication system by rationally configuring the phase shift of passive RIS sub-units.
[0003] Phase control of the RIS sub-units is a key technology in RIS-assisted wireless communication systems. To achieve optimal performance, RIS-assisted wireless communication systems need to find the optimal phase shift of the RIS in the cascaded channel and feed this optimal phase shift back to the RIS end via a feedback channel. However, due to the large number of RIS sub-units (typically hundreds or thousands) and the high quantization precision of each sub-unit's phase value, the amount of bit data requiring feedback after analog-to-digital conversion is enormous. This necessitates a huge amount of feedback data, consuming significant spectrum and time resources, and resulting in substantial latency. Therefore, the urgent problem to be solved is to significantly reduce the amount of feedback data while maintaining RIS phase accuracy during phase shift feedback. Summary of the Invention
[0004] The main objective of this invention is to provide a phase shift feedback method for reconfigurable smart surfaces, aiming to solve the technical problem in the prior art of how to ensure the phase accuracy of RIS while greatly reducing the amount of feedback data during phase shift feedback.
[0005] To achieve the above objectives, the present invention provides a phase shift feedback method for a reconfigurable smart surface, the phase shift feedback method for the reconfigurable smart surface comprising:
[0006] Obtain the phase shift matrix to be transmitted;
[0007] The feature number is determined based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device;
[0008] The bit stream of the feature sequence number is sent to the reconfigurable smart surface device so that the reconfigurable smart surface device can determine the received phase shift matrix based on the bit stream of the feature sequence number, the shared knowledge base, and the target decoder.
[0009] Optionally, determining the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base includes:
[0010] The phase shift matrix to be transmitted is preprocessed to obtain a normalized matrix;
[0011] Multiple feature information vectors are determined based on the normalized matrix and the target encoder;
[0012] The similarity calculation is performed between each feature information vector and each preset feature vector in the shared knowledge base to obtain the similarity calculation result;
[0013] The feature index of each feature information vector is determined based on the similarity calculation results.
[0014] Optionally, determining multiple feature information vectors based on the normalized matrix and the target encoder includes:
[0015] Based on the target encoder, high-dimensional feature extraction is performed on the normalized matrix to determine multiple feature information matrices;
[0016] The target encoder compresses each feature information matrix to determine multiple feature information vectors, and the vector length of each feature information vector is the same as the vector length of each preset feature vector.
[0017] Optionally, before determining the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, the method further includes:
[0018] Acquire a learnable knowledge base and a set of sample phase shift matrices;
[0019] The learning knowledge base is initialized to obtain an initial learning knowledge base;
[0020] The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set to obtain a shared knowledge base, a target encoder, and a target decoder, and the shared knowledge base and the target decoder are sent to the reconfigurable smart surface device.
[0021] Optionally, the step of training the initial learning knowledge base, the initial encoder, and the initial decoder based on the sample phase shift matrix set to obtain the shared knowledge base, the target encoder, and the target decoder includes:
[0022] Determine the target loss function based on reconstruction loss and output loss.
[0023] The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set and the target loss function to obtain a shared knowledge base, a trained encoder, and a trained decoder.
[0024] The target encoder and target decoder are determined based on the trained encoder and the trained decoder.
[0025] Optionally, determining the target encoder and target decoder based on the trained encoder and the trained decoder includes:
[0026] The output sample phase shift is determined based on the sample phase shift matrix in the sample phase shift matrix set, the trained encoder, and the trained decoder;
[0027] The reconstruction quality values of the trained encoder and the trained decoder are determined based on the preset evaluation function, the output sample phase shift, and the output sample phase shift.
[0028] The target encoder and target decoder are determined based on the reconstructed quality values.
[0029] Furthermore, to achieve the above objectives, the present invention also proposes a transmitting device applied to the phase shift feedback method for reconfigurable smart surfaces described above, the transmitting device comprising:
[0030] The acquisition module is used to acquire the phase shift matrix to be transmitted;
[0031] The determining module is used to determine the feature number based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device;
[0032] The transmitting module is used to transmit the bit stream of the feature sequence number to the reconfigurable smart surface device, so that the reconfigurable smart surface device can determine the receiving phase shift matrix based on the bit stream of the feature sequence number, the shared knowledge base, and the target decoder.
[0033] Furthermore, to achieve the above objectives, this invention also proposes a phase shift feedback method for reconfigurable smart surfaces, applied to reconfigurable smart surface devices. The phase shift feedback method for reconfigurable smart surfaces includes:
[0034] Upon receiving a bitstream of feature sequence numbers sent by a transmitting device, multiple received feature vectors are determined based on the bitstreams of multiple feature sequence numbers and a shared knowledge base, wherein the feature sequence numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base;
[0035] Multiple receive feature matrices are determined based on the target decoder and multiple receive feature vectors;
[0036] The receiving phase shift matrix is determined by performing matrix reconstruction based on the target decoder and multiple received feature matrices.
[0037] Furthermore, to achieve the above objectives, the present invention also proposes a reconfigurable smart surface device, which is applied to the phase shift feedback method of the reconfigurable smart surface described above. The reconfigurable smart surface device includes:
[0038] The determination module is used to determine multiple received feature vectors based on the bit stream of multiple feature numbers and a shared knowledge base when receiving a bit stream of feature numbers sent by the transmitting device, wherein the feature numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base;
[0039] The determining module is also used to determine multiple receiving feature matrices based on the target decoder and multiple receiving feature vectors;
[0040] The reconstruction module is used to perform matrix reconstruction based on the target decoder and multiple received feature matrices to determine the received phase shift matrix.
[0041] Furthermore, to achieve the above objectives, the present invention also proposes a phase shift feedback system for a reconfigurable smart surface, the phase shift feedback system for the reconfigurable smart surface including a transmitting device applied to the above-described device and a reconfigurable smart surface device applied to the above-described device.
[0042] This invention is applied to a transmitting device. It acquires the phase shift matrix to be transmitted; determines feature indices based on the phase shift matrix, a target encoder, and a shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device; and transmits a bitstream of the feature indices to the reconfigurable smart surface device, enabling the reconfigurable smart surface device to determine the received phase shift matrix based on the bitstream of the feature indices, the shared knowledge base, and the target decoder. By determining the feature indices of the phase shift matrix to be transmitted that require feedback, and transmitting only the bitstream of the feature indices during transmission, the communication resources and feedback data volume occupied by the reconfigurable smart surface feedback are significantly reduced, as is the time consumed during feedback, while ensuring the phase accuracy of the feedback. Attached Figure Description
[0043] Figure 1 This is a schematic flowchart of the first embodiment of the phase shift feedback method for reconfigurable smart surfaces of the present invention;
[0044] Figure 2 This is a schematic diagram of the target encoder and target decoder structure of an embodiment of the phase shift feedback method for reconfigurable smart surfaces of the present invention;
[0045] Figure 3 This is a schematic flowchart of the second embodiment of the phase shift feedback method for reconfigurable smart surfaces of the present invention;
[0046] Figure 4 This is a schematic diagram of the overall structure of an embodiment of the phase shift feedback method for reconfigurable smart surfaces according to the present invention;
[0047] Figure 5 This is a schematic diagram of the overall process of an embodiment of the phase shift feedback method for reconfigurable smart surfaces of the present invention;
[0048] Figure 6 This is a structural block diagram of the first embodiment of the transmitting device of the present invention;
[0049] Figure 7 This is a structural block diagram of the first embodiment of the reconfigurable smart surface device of the present invention;
[0050] Figure 8 This is a structural block diagram of the first embodiment of the phase shift feedback system of the reconfigurable smart surface of the present invention.
[0051] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0052] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.
[0053] This invention provides a phase shift feedback method for reconfigurable smart surfaces, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the phase shift feedback method for a reconfigurable smart surface according to the present invention.
[0054] The phase shift feedback method for reconfigurable smart surfaces in this embodiment is applied to a transmitting device. The phase shift feedback method for reconfigurable smart surfaces includes the following steps:
[0055] Step S10: Obtain the phase shift matrix to be transmitted.
[0056] It should be noted that the execution subject of this embodiment is the transmitting device in the phase shift feedback system of the reconfigurable smart surface. The phase shift feedback system of the reconfigurable smart surface includes the transmitting device and the reconfigurable smart surface device. The transmitting device includes the UE end and the BS end. This embodiment does not limit this. The transmitting device obtains the phase shift matrix to be transmitted, determines the feature number according to the phase shift matrix to be transmitted, the target encoder and the shared knowledge base, and sends the bit stream of the feature number to the reconfigurable smart surface device, so that the RIS end can determine the received phase shift matrix according to the bit stream of the feature number, the shared knowledge base and the target decoder.
[0057] It is understandable that the phase shift matrix to be transmitted is the phase shift matrix that needs to be transmitted to the RIS end. The phase shift matrix to be transmitted is obtained by maximizing the rate of the UE end in the transmitting device.
[0058] Step S20: Determine the feature number based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device.
[0059] It should be noted that the target encoder is constructed using a neural network, implemented in the PyTorch 1.11 framework using a set of sample phase shift matrices, trained using an Nvidia A40 GPU, and tuned using the Adam optimizer and cosine annealing strategy as learning rate adjustment strategies. The target encoder resides on the transmitting device and can compress the phase shift matrices to be transmitted. The shared knowledge base stores a large number of preset feature vectors and their corresponding data indices. The preset feature vectors are feature information vectors extracted from a large number of sample phase shift matrices, with each set of data indices corresponding to one preset feature vector. The shared knowledge base can be updated during training by learning the feature information vectors of a large number of sample phase shift matrices. The shared knowledge base is stored on both the transmitting device and the reconfigurable smart surface device. Updates to the shared knowledge base occur on the transmitting device, typically at the BS (Browser / Server) end. After the update, the transmitting device transmits the information to the BS / UE (User Equipment) end and the RIS (Reference System) end via wired fiber optic cable or wirelessly.
[0060] It is understandable that after determining the phase shift matrix to be transmitted, the phase shift matrix to be transmitted is input into the target encoder so that the target encoder can compress the phase shift matrix to be transmitted and further compress the output feature information vector using a shared knowledge base to obtain the feature index of each feature information vector.
[0061] In a specific implementation, to obtain accurate feature numbers, the step of determining feature numbers based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base further includes: performing data preprocessing on the phase shift matrix to be transmitted to obtain a normalized matrix; determining multiple feature information vectors based on the normalized matrix and the target encoder; performing similarity calculations on each feature information vector and each preset feature vector in the shared knowledge base to obtain similarity calculation results; and determining the feature number of each feature information vector based on the similarity calculation results.
[0062] It should be noted that the transmitting device performs a preprocessing operation on the phase shift matrix to be transmitted, that is, it normalizes the phase shift matrix to be transmitted to obtain a normalized phase shift matrix. The target encoder located at the BS / UE end extracts features from the normalized matrix, thereby outputting multiple feature information vectors.
[0063] Understandably, the transmitting device uses Euclidean distance to calculate the similarity between multiple feature information vectors output by the target encoder and each preset feature vector in the shared knowledge base, thereby determining the similarity between each feature information vector and each preset feature vector. The similarity between each feature information vector and each preset feature vector is the similarity calculation result. Based on the similarity calculation result, the data sequence number corresponding to the preset feature vector with the highest similarity to each feature information vector is determined. The data sequence number corresponding to the preset feature vector with the highest similarity to each feature information vector is the feature sequence number of each feature information vector.
[0064] In practical implementation, the dimensions of the shared knowledge base are: Where Z is the size of the shared knowledge base space, and K is the size of each preset feature vector. The vector length is Z, and there are a total of Z embedded vectors. .
[0065] It should be noted that, in order to output accurate feature information vectors based on the target encoder, the step of determining multiple feature information vectors based on the normalization matrix and the target encoder further includes: performing high-dimensional feature extraction on the normalization matrix based on the target encoder to determine multiple feature information matrices; and compressing each feature information matrix based on the target encoder to determine multiple feature information vectors, wherein the vector length of each feature information vector is the same as the vector length of each preset feature vector.
[0066] Understandably, the normalized matrix is input into the target encoder, which extracts high-dimensional features from the normalized matrix to obtain several feature information matrices. These feature information matrices are then converted into feature information vectors, resulting in multiple feature information vectors for the phase shift matrix to be transmitted. The length of these feature information vectors is the same as the length of each preset feature vector in the shared knowledge base. For example, ... Figure 2 As shown, taking M=32 (M is the number of sub-units in each row of a square RIS) as an example, the target encoder input normalization matrix... The matrix is processed through three convolutional and ReLU combination layers, each with a 4×4 kernel, a stride of 2, and padding of 1. The designed convolutional kernels compress the feature information matrix to one-quarter of its original size after each convolutional layer, allowing for simultaneous feature extraction and compression of the phase-shifted matrix using a fully convolutional module. The number of convolutional channels is C / 2, C / 2, and C, respectively. After flattening, C K-dimensional feature information vectors are output. .
[0067] In a specific implementation, in order to obtain accurate target encoders and target decoders, before determining the feature sequence number based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, the method further includes: acquiring a learnable knowledge base and a set of sample phase shift matrices; initializing the learnable knowledge base to obtain an initial learnable knowledge base; training the initial learnable knowledge base, the initial encoder, and the initial decoder based on the set of sample phase shift matrices to obtain a shared knowledge base, a target encoder, and a target decoder, and sending the shared knowledge base and the target decoder to the reconfigurable smart surface device.
[0068] It should be noted that the untrained shared knowledge base in the learnable knowledge base, the sample phase shift matrix set refers to a collection containing a large number of sample phase shift matrices. The sample phase shift matrices are all phase shift matrices obtained historically to maximize the rate of the UE.
[0069] It is understandable that the learnable knowledge base is initialized to obtain the initial learning knowledge base. The initial learning knowledge base refers to setting the learnable knowledge base to a uniform distribution of (0, 1 / K) during the network initialization stage, where K is the length of the knowledge base vector. The purpose is to control the model loss from being too large during the initial knowledge base learning stage, thereby accelerating training and convergence.
[0070] In the specific implementation, the sample phase shift matrix set is divided into a training set, a validation set, and a test set in a 10:1:1 ratio. The initial learning knowledge base, initial encoder, and initial decoder are trained to obtain a shared knowledge base, a target encoder, and a target decoder. The shared knowledge base and target decoder are then sent to the reconfigurable smart surface device. When the training process is performed at the BS (Browser / Server) end, the target encoder and shared knowledge base must be sent to the UE (User Equipment) end upon completion. When the training process is performed at the UE end, the target decoder and shared knowledge base must be sent to the BS end upon completion.
[0071] It should be noted that, to ensure the accuracy of the training process, the step of training the initial learning knowledge base, initial encoder, and initial decoder based on the sample phase shift matrix set to obtain the shared knowledge base, target encoder, and target decoder further includes: determining the target loss function based on the reconstruction loss and output loss; training the initial learning knowledge base, initial encoder, and initial decoder based on the sample phase shift matrix set and the target loss function to obtain the shared knowledge base, trained encoder, and trained decoder; and determining the target encoder and target decoder based on the trained encoder and the trained decoder.
[0072] Understandably, the reconstruction loss refers to the encoder's sample phase shift matrix. and the output reconstruction matrix of the decoder The MSE loss between the preset feature vector and the encoder output is used to characterize the knowledge base's update learning, and is denoted as kb loss. This determines the target loss function, which is: The kb loss can be further divided into two terms, which can be expressed as: , where sg represents the stopping gradient operator, which is defined as an operator with identity and zero gradient during forward computation, thus effectively constraining the operand to a non-updating constant. As a penalty factor, it controls the rate of knowledge base and encoder gradient updates in kb loss. This represents the feature information vector output by the initial encoder during training. This represents the preset feature vectors in the shared knowledge base. The first term in the kb loss is used to train the shared knowledge base, making the vector distribution in the initial knowledge base more closely approximate the distribution of the encoder's output feature information vectors. The second term is used to fix the initial knowledge base, making the distribution of the encoder's output feature information vectors approximate the distribution of the vectors in the initial knowledge base. The kb loss aims to make the distribution of the encoder's output feature information vectors closer to the distribution of the vectors in the initial knowledge base.
[0073] In the specific implementation, the sample phase shift matrix set is divided into training, validation, and test sets in a 10:1:1 ratio. A gradient descent-based algorithm is used to train the model, composed of a shared knowledge base, encoder, and decoder, for several rounds. The model is implemented in the PyTorch 1.11 framework, trained using an Nvidia A40 GPU, and the Adam optimizer and cosine annealing strategy are selected as the learning rate adjustment strategies. The objective loss function is used to calculate the similarity between the model's predicted data and the true values. The model mentioned here refers to the model consisting of the initial learning knowledge base, initial encoder, and initial decoder. Finally, multiple training models are determined, each containing a trained encoder, a trained decoder, and a shared knowledge base. The shared knowledge base included in all training models is the same.
[0074] It should be noted that, in order to select the best-performing target decoder and target encoder among multiple models, the step of determining the target encoder and target decoder based on the trained encoder and the trained decoder further includes: determining the output sample phase shift based on the sample phase shift matrix in the sample phase shift matrix set, the trained encoder, and the trained decoder; determining the reconstruction quality value of the trained encoder and the trained decoder based on a preset evaluation function, the output sample phase shift, and the output sample phase shift; and determining the target encoder and target decoder based on the reconstruction quality value.
[0075] Understandably, the preset evaluation function is the normalized mean square error function, used to evaluate the quality of phase-shift compression reconstruction, specifically in the form of... .in The input is the sample phase shift matrix used to train the encoder. To train the phase shift of the output samples of the decoder, the phase shift matrix of the samples in the validation set is input into the training encoder of each model, so that the training encoder outputs sample feature vectors. The training decoder performs matrix reconstruction based on the bit stream of the data sequence number of the output sample feature vector and the shared knowledge base to determine the corresponding output sample phase shift, thereby determining the reconstruction quality value of each model. The model with the lowest reconstruction quality value is selected, which has the best performance. The training encoder and training decoder contained in this model are the target encoder and target decoder, and the generalization of the target encoder and target decoder is tested using the test set.
[0076] Step S30: Send the bit stream of the feature sequence number to the reconfigurable smart surface device, so that the reconfigurable smart surface device can determine the received phase shift matrix based on the bit stream of the feature sequence number, the shared knowledge base, and the target decoder.
[0077] It should be noted that the reconfigurable smart surface device is the RIS end. The transmitting device converts the feature index of each feature information vector into a bit stream and sends it to the RIS end. The RIS end performs matrix reconstruction based on the bit stream of feature indexes, the shared knowledge base, and the target decoder. The phase shift matrix output by the target decoder is the received phase shift matrix. Ideally, the received phase shift matrix and the phase shift matrix to be transmitted are the same.
[0078] This embodiment is applied to a transmitting device. It acquires the phase shift matrix to be transmitted; determines feature numbers based on the phase shift matrix, the target encoder, and a shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device; and sends a bitstream of the feature numbers to the reconfigurable smart surface device, enabling the reconfigurable smart surface device to determine the received phase shift matrix based on the bitstream of the feature numbers, the shared knowledge base, and the target decoder. By determining the feature numbers of the phase shift matrix to be transmitted that require feedback in this way, and transmitting only the bitstream of the feature numbers, the communication resources and feedback data volume occupied by the reconfigurable smart surface feedback are greatly reduced, as is the time consumed during feedback, while ensuring the phase accuracy of the feedback.
[0079] refer to Figure 3 , Figure 3 This is a schematic flowchart of a second embodiment of a phase shift feedback method for a reconfigurable smart surface according to the present invention.
[0080] Based on the first embodiment described above, the phase shift feedback method for reconfigurable smart surfaces in this embodiment is applied to a reconfigurable smart surface device. The phase shift feedback method for reconfigurable smart surfaces includes:
[0081] Step S01: Upon receiving a bit stream of feature sequence numbers sent by the transmitting device, determine multiple received feature vectors based on the bit streams of multiple feature sequence numbers and a shared knowledge base, wherein the feature sequence numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base.
[0082] It should be noted that the execution subject of this embodiment is the reconfigurable smart surface device in the phase shift feedback system of the reconfigurable smart surface, also known as the RIS end. When the RIS end receives the bit stream of feature numbers of each feature information vector sent by the transmitting device, it converts the bit stream of multiple feature numbers to obtain feature numbers, extracts the corresponding received feature vectors in the shared knowledge base using multiple feature numbers, determines multiple received feature matrices based on the target decoder and multiple received feature vectors, and performs matrix reconstruction based on the target decoder and multiple received feature matrices to determine the received phase shift matrix.
[0083] It can be explained that after receiving a bit stream of multiple feature numbers, the RIS end retrieves the corresponding preset feature vector from the shared knowledge base according to the bit stream of each feature number. The preset feature vector corresponding to the bit stream of each feature number is the received feature vector.
[0084] Step S02: Determine multiple receive feature matrices based on the target decoder and multiple receive feature vectors.
[0085] It should be noted that the target decoder is constructed using a neural network, implemented in the PyTorch 1.11 framework using a sample phase shift matrix set, trained using an Nvidia A40 GPU, and tuned using the Adam optimizer and cosine annealing strategy as learning rate adjustment strategies. The target decoder resides at the RIS end and can perform matrix reconstruction based on multiple feature vectors. Multiple received feature vectors are input to the target decoder, and the target encoder determines multiple received feature matrices based on these feature vectors.
[0086] Step S03: Perform matrix reconstruction based on the target decoder and multiple received feature matrices to determine the received phase shift matrix.
[0087] It should be noted that the target decoder first determines several received feature matrices based on multiple received feature vectors, and then uses these received feature matrices to perform matrix reconstruction, outputting the final received phase shift matrix. The RIS end obtains the received phase shift matrix, thereby completing the feedback of the phase shift matrix to be transmitted.
[0088] It is understandable that, such as Figure 2 As shown, taking M=32 as an example, the input to the target decoder is C K-dimensional received feature vectors retrieved from the shared knowledge base. These feature vectors are then transformed into a matrix through a reshaping operation. First, the signal passes through a residual module, consisting of two layers of convolutions and ReLU. The convolution kernels are 3×3 with a stride of 1 and padding of 1. The addition of the residual module makes the forward and backward propagation of information smoother. Because it includes a natural identity mapping, it can solve problems such as network degradation and gradient vanishing to some extent. After the residual module, the target decoder is designed with a structure symmetrical to the target encoder. Then, it passes through three upsampling layers consisting of transposed convolutions (TConv) and ReLU. These layers also have transposed convolution kernels of 4×4 with a stride of 2 and padding of 1, with convolution channels of C / 2, C / 2, and 1 respectively. The output of the target decoder is the received phase shift matrix. .
[0089] In specific implementations, such as Figure 4 and Figure 5 As shown, the learnable knowledge base is initialized, and the phase shift matrix of the samples in the training set is trained for several rounds using a gradient descent algorithm. The similarity between the preset data and the true value is calculated using the target loss function. The generalization of the model is tested using the test set. The phase shift compression reconstruction quality is evaluated using a preset evaluation function. The target encoder, target decoder, and shared knowledge base are determined. A shared knowledge base is shared by the BE / UE / RIS terminals. The phase shift matrix to be transmitted is normalized to obtain a normalized matrix. The high-dimensional features of the normalized matrix are extracted by the target encoder located at the BS / UE end, resulting in several feature matrices. These feature matrices are then converted into feature vectors with the same length as the vectors in the shared knowledge base. The BS / UE end compares the feature vectors output by the target encoder with the pre-learned shared knowledge base to obtain the data index of the shared knowledge base vector that is most similar to the feature vector output by the target encoder. This data index is then converted into a bit stream and transmitted to the RIS end. The RIS end dequantizes the received signal to obtain the data index, extracts the vector corresponding to the data index from the shared knowledge base, and inputs the vector into the target decoder located at the RIS end. Finally, the phase shift matrix of the RIS is reconstructed by the target decoder.
[0090] This embodiment is applied to a reconfigurable smart surface device. Upon receiving a bitstream of feature numbers transmitted by a transmitting device, multiple received feature vectors are determined based on the bitstream of multiple feature numbers and a shared knowledge base. The feature numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base. Multiple received feature matrices are determined based on the target decoder and the multiple received feature vectors. Matrix reconstruction is performed based on the target decoder and the multiple received feature matrices to determine the received phase shift matrix. Through this method, multiple received feature vectors are determined using the bitstream of feature numbers transmitted by the transmitting device and the shared knowledge base. Matrix reconstruction is then performed using the decoder and the received feature vectors to obtain the received phase shift matrix. This eliminates the need to receive a large number of phase shift parameters during reception, significantly reducing transmission latency. Furthermore, matrix reconstruction using the target decoder ensures the RIS phase accuracy.
[0091] In addition, refer to Figure 6 The present invention also proposes a transmitting device, the transmitting device comprising:
[0092] The acquisition module 11 is used to acquire the phase shift matrix to be transmitted.
[0093] The determining module 12 is used to determine the feature number based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device.
[0094] The sending module 13 is used to send the bit stream of the feature sequence number to the reconfigurable smart surface device, so that the reconfigurable smart surface device can determine the receiving phase shift matrix based on the bit stream of the feature sequence number, the shared knowledge base, and the target decoder.
[0095] This embodiment obtains the phase shift matrix to be transmitted; determines feature numbers based on the phase shift matrix, the target encoder, and a shared knowledge base, wherein the shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device; and sends a bitstream of the feature numbers to the reconfigurable smart surface device, enabling the reconfigurable smart surface device to determine the received phase shift matrix based on the bitstream of the feature numbers, the shared knowledge base, and the target decoder. By determining the feature numbers of the phase shift matrix to be transmitted that require feedback in this way, and transmitting only the bitstream of the feature numbers, the communication resources and feedback data volume occupied by the reconfigurable smart surface feedback are greatly reduced, the time consumed during feedback is decreased, and the phase accuracy of the feedback is guaranteed.
[0096] In one embodiment, the determining module 12 is further configured to perform data preprocessing on the phase shift matrix to be transmitted to obtain a normalized matrix;
[0097] Multiple feature information vectors are determined based on the normalized matrix and the target encoder;
[0098] The similarity calculation is performed between each feature information vector and each preset feature vector in the shared knowledge base to obtain the similarity calculation result;
[0099] The feature index of each feature information vector is determined based on the similarity calculation results.
[0100] In one embodiment, the determining module 12 is further configured to perform high-dimensional feature extraction on the normalized matrix based on the target encoder to determine multiple feature information matrices;
[0101] The target encoder compresses each feature information matrix to determine multiple feature information vectors, and the vector length of each feature information vector is the same as the vector length of each preset feature vector.
[0102] In one embodiment, the determining module 12 is further configured to acquire a learnable knowledge base and a sample phase shift matrix set;
[0103] The learning knowledge base is initialized to obtain an initial learning knowledge base;
[0104] The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set to obtain a shared knowledge base, a target encoder, and a target decoder, and the shared knowledge base and the target decoder are sent to the reconfigurable smart surface device.
[0105] In one embodiment, the determining module 12 is further configured to determine a target loss function based on the reconstruction loss and the output loss;
[0106] The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set and the target loss function to obtain a shared knowledge base, a trained encoder, and a trained decoder.
[0107] The target encoder and target decoder are determined based on the trained encoder and the trained decoder.
[0108] In one embodiment, the determining module 12 is further configured to determine the output sample phase shift based on the sample phase shift matrix in the sample phase shift matrix set, the training encoder, and the training decoder;
[0109] The reconstruction quality values of the trained encoder and the trained decoder are determined based on the preset evaluation function, the output sample phase shift, and the output sample phase shift.
[0110] The target encoder and target decoder are determined based on the reconstructed quality values.
[0111] Since this transmitting device adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be described in detail here.
[0112] In addition, refer to Figure 7 This invention also proposes a reconfigurable smart surface device, which includes:
[0113] The determining module 21 is used to determine multiple received feature vectors based on the bit stream of multiple feature numbers and a shared knowledge base when receiving a bit stream of feature numbers sent by the transmitting device, wherein the feature numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base.
[0114] The determining module 21 is also used to determine multiple receiving feature matrices based on the target decoder and multiple receiving feature vectors.
[0115] The reconstruction module 22 is used to perform matrix reconstruction based on the target decoder and multiple received feature matrices to determine the received phase shift matrix.
[0116] This embodiment determines multiple received feature vectors based on the bitstream of feature numbers transmitted by the transmitting device and a shared knowledge base upon receiving the bitstream of feature numbers. The feature numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base. Multiple received feature matrices are then determined based on the target decoder and the multiple received feature vectors. Finally, matrix reconstruction is performed using the target decoder and the multiple received feature matrices to determine the received phase shift matrix. By utilizing the bitstream of feature numbers transmitted by the transmitting device and the shared knowledge base to determine multiple received feature vectors, and then using the decoder and the received feature vectors to perform matrix reconstruction to obtain the received phase shift matrix, the transmission latency is significantly reduced by eliminating the need to receive a large number of phase shift parameters during reception. Furthermore, the use of the target decoder for matrix reconstruction ensures the RIS phase accuracy.
[0117] In addition, refer to Figure 8 The present invention also proposes a phase shift feedback system for a reconfigurable smart surface, the phase shift feedback system for the reconfigurable smart surface including a transmitting device 10 as described above and a reconfigurable smart surface device 20 as described above.
[0118] Since the phase shift feedback system of the reconfigurable smart surface adopts all the technical solutions of all the above embodiments, it has at least all the beneficial effects brought about by the technical solutions of the above embodiments, which will not be repeated here.
[0119] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of this invention. In practical applications, those skilled in the art can select some or all of the workflow to achieve the purpose of this embodiment according to actual needs, and no restrictions are imposed here.
[0120] In addition, for technical details not described in detail in this embodiment, please refer to the phase shift feedback method for reconfigurable smart surfaces provided in any embodiment of the present invention, which will not be repeated here.
[0121] Furthermore, it should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.
[0122] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.
[0123] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory (ROM) / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.
[0124] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.
Claims
1. A phase shift feedback method for a reconfigurable smart surface, characterized in that, The phase shift feedback method for the reconfigurable smart surface, applied to a transmitting device, includes: Obtain the phase shift matrix to be transmitted; The feature index is determined based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base. The shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device, and stores multiple preset feature vectors and their corresponding data indices. The determination of the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base includes: preprocessing the phase shift matrix to be transmitted to obtain a normalized matrix; determining multiple feature information vectors based on the normalized matrix and the target encoder; performing similarity calculations between each feature information vector and each preset feature vector in the shared knowledge base to obtain a similarity calculation result; and obtaining the data indices of the corresponding preset feature vectors from the shared knowledge base based on the similarity calculation results as the feature indices of the feature information vectors. The bit stream of the feature sequence number is sent to the reconfigurable smart surface device, so that the reconfigurable smart surface device obtains the corresponding preset feature vector from the shared knowledge base according to the feature sequence number as the received feature vector, determines multiple received feature matrices according to the target decoder and multiple received feature vectors, and performs matrix reconstruction according to the target decoder and multiple received feature matrices to determine the received phase shift matrix.
2. The phase shift feedback method for reconfigurable smart surfaces as described in claim 1, characterized in that, The step of determining multiple feature information vectors based on the normalized matrix and the target encoder includes: Based on the target encoder, high-dimensional feature extraction is performed on the normalized matrix to determine multiple feature information matrices; The target encoder compresses each feature information matrix to determine multiple feature information vectors, and the vector length of each feature information vector is the same as the vector length of each preset feature vector.
3. The phase shift feedback method for reconfigurable smart surfaces as described in any one of claims 1 to 2, characterized in that, Before determining the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base, the method further includes: Acquire a learnable knowledge base and a set of sample phase shift matrices; The learnable knowledge base is initialized to obtain an initial learning knowledge base; The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set to obtain a shared knowledge base, a target encoder, and a target decoder, and the shared knowledge base and the target decoder are sent to the reconfigurable smart surface device.
4. The phase shift feedback method for reconfigurable smart surfaces as described in claim 3, characterized in that, The step of training the initial learning knowledge base, initial encoder, and initial decoder based on the sample phase shift matrix set to obtain a shared knowledge base, target encoder, and target decoder includes: The target loss function is determined based on the reconstruction loss and the output loss. The initial learning knowledge base, initial encoder, and initial decoder are trained based on the sample phase shift matrix set and the target loss function to obtain a shared knowledge base, a trained encoder, and a trained decoder. The target encoder and target decoder are determined based on the trained encoder and the trained decoder.
5. The phase shift feedback method for reconfigurable smart surfaces as described in claim 4, characterized in that, The step of determining the target encoder and target decoder based on the trained encoder and the trained decoder includes: The output sample phase shift is determined based on the sample phase shift matrix in the sample phase shift matrix set, the trained encoder, and the trained decoder; The reconstruction quality values of the trained encoder and the trained decoder are determined based on the preset evaluation function, the sample phase shift matrix, and the output sample phase shift. The target encoder and target decoder are determined based on the reconstructed quality values.
6. A transmitting device, characterized in that, The transmitting device performs the phase shift feedback method for the reconfigurable smart surface as described in any one of claims 1 to 5, the transmitting device comprising: The acquisition module is used to acquire the phase shift matrix to be transmitted; The determining module is used to determine the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base. The shared knowledge base is shared by the transmitting device and the reconfigurable smart surface device, and stores multiple preset feature vectors and their corresponding data indices. The determination of the feature index based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base includes: preprocessing the phase shift matrix to be transmitted to obtain a normalized matrix; determining multiple feature information vectors based on the normalized matrix and the target encoder; performing similarity calculations between each feature information vector and each preset feature vector in the shared knowledge base to obtain a similarity calculation result; and obtaining the data indices of the corresponding preset feature vectors from the shared knowledge base based on the similarity calculation results as the feature indices of the feature information vectors. The transmitting module is used to transmit the bit stream of the feature sequence number to the reconfigurable smart surface device, so that the reconfigurable smart surface device can obtain the corresponding preset feature vector from the shared knowledge base according to the feature sequence number as the received feature vector, determine multiple received feature matrices according to the target decoder and multiple received feature vectors, and perform matrix reconstruction according to the target decoder and multiple received feature matrices to determine the received phase shift matrix.
7. A phase shift feedback method for a reconfigurable smart surface, characterized in that, A phase shift feedback method for a reconfigurable smart surface, applicable to reconfigurable smart surface devices, includes: Upon receiving a bitstream of feature sequence numbers sent by a transmitting device, multiple received feature vectors are determined based on the bitstreams of multiple feature sequence numbers and a shared knowledge base. The feature sequence numbers are determined by the transmitting device based on the phase shift matrix to be transmitted, the target encoder, and the shared knowledge base. The shared knowledge base stores multiple preset feature vectors and their corresponding data sequence numbers. The transmitting device performs data preprocessing on the phase shift matrix to be transmitted to obtain a normalized matrix. Based on the normalized matrix and the target encoder, multiple feature information vectors are determined. A similarity calculation is performed between each feature information vector and each preset feature vector in the shared knowledge base to obtain a similarity calculation result. Based on the similarity calculation result, the data sequence number of the corresponding preset feature vector is obtained from the shared knowledge base as the feature sequence number of the feature information vector. The determination of multiple received feature vectors based on the bitstreams of multiple feature sequence numbers and the shared knowledge base includes: obtaining the corresponding preset feature vector from the shared knowledge base as the received feature vector based on the feature sequence number. Multiple receive feature matrices are determined based on the target decoder and multiple receive feature vectors; The receiving phase shift matrix is determined by performing matrix reconstruction based on the target decoder and multiple received feature matrices.
8. A reconfigurable smart surface device, characterized in that, The reconfigurable smart surface device is applied to the phase shift feedback method of the reconfigurable smart surface as described in claim 7, wherein the reconfigurable smart surface device comprises: A determination module is used to determine multiple received feature vectors based on the bit stream of multiple feature numbers and a shared knowledge base when receiving a bit stream of feature numbers sent by a transmitting device. The feature numbers are determined by the transmitting device based on a phase shift matrix to be transmitted, a target encoder, and the shared knowledge base. The shared knowledge base stores multiple preset feature vectors and their corresponding data numbers. The transmitting device preprocesses the phase shift matrix to be transmitted to obtain a normalized matrix, determines multiple feature information vectors based on the normalized matrix and the target encoder, performs similarity calculations between each feature information vector and each preset feature vector in the shared knowledge base to obtain a similarity calculation result, and obtains the data number of the corresponding preset feature vector from the shared knowledge base as the feature number of the feature information vector based on the similarity calculation result. The determination of multiple received feature vectors based on the bit stream of multiple feature numbers and the shared knowledge base includes: obtaining the corresponding preset feature vector from the shared knowledge base as the received feature vector based on the feature number. The determining module is also used to determine multiple receiving feature matrices based on the target decoder and multiple receiving feature vectors; The reconstruction module is used to perform matrix reconstruction based on the target decoder and multiple received feature matrices to determine the received phase shift matrix.
9. A phase shift feedback system for a reconfigurable smart surface, characterized in that, The phase shift feedback system of the reconfigurable smart surface includes the transmitting device as described in claim 6 and the reconfigurable smart surface device as described in claim 8.