6G full-decoupling network representative channel generation method and system based on comparative learning
Through the comparative learning method, the multipath channel characterization information is extracted, and the problem of difficulty in selecting transmission parameters in the 6G fully decoupled network is solved, efficient acquisition of channel characterization is achieved, and the establishment efficiency of spectrum maps and channel description quality is improved.
Patent Information
- Application Number
- CN202410994129.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-24
- Publication Date
- 2024-11-12
AI Technical Summary
In 6G fully decoupled wireless access network, it is difficult to select fixed transmission parameters to establish a spectrum map, and the existing feedback mechanism is not applicable to hardware-isolated uplink and downlink base stations, resulting in difficulty in obtaining channel information.
Using a method based on contrast learning, the representation information of the multipath channel is extracted through the Transformer encoder and decoder, and the representative channel is restored to determine the transmission parameters of the spectrum map. The self-attention mechanism and positive and negative sample comparison learning are used to obtain channel representation.
It effectively avoids feedback defects caused by hardware isolation, provides a feasible representative channel generation method, improves the spectrum map establishment efficiency of 6G fully decoupled network, and improves the acquisition quality of channel characterization.
Smart Images

Figure CN118944790A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of wireless mobile communications, relates to contrastive learning and multi-input multi-output transmission, and specifically relates to a 6G fully decoupled network representative channel generation method and system based on contrastive learning. Background Art
[0002] In wireless communications, the channel describes the signal propagation in the physical environment and determines the transmission quality. In a MIMO OFDM system, the channel is usually represented in the space, frequency, and time domains. The space domain describes the multipath signal scattering in the angular domain, involving the departure angle at the transmitter and the arrival angle at the receiver. The frequency domain describes the multipath fading scattering in the delay domain, capturing the frequency selective fading of the channel. The time domain reflects the multipath effect of the environment dynamics. Here, we refer to the channel at a fixed geographical location, thus excluding the Doppler effect. By studying the channel, its characteristics can be effectively utilized to facilitate MIMO transmission, such as beamforming or spatial multiplexing. The channel exhibits correlation in the space, frequency, and time domains. When using uniform linear array or uniform planar array antennas, the electromagnetic waves arrive at different antennas with a fixed phase difference. In addition, there is a fixed spacing between different carriers, resulting in a fixed phase difference in the channels on different subcarriers. At a specific location, the channel exhibits time correlation because the main electromagnetic waves roughly follow the same propagation path. The correlation means that the channel can be represented in a form with higher information entropy, i.e., channel characterization, which is crucial to understanding the intrinsic characteristics of the channel.
[0003] In addition, in the next generation mobile communication network (6G), Academician Yu Quan and others proposed a 6G fully decoupled wireless access network architecture in "A Fully-Decoupled RAN Architecture for 6G Inspired by Neurotransmission". In a fully decoupled network, traditional base stations are decoupled into control base stations, uplink base stations and downlink base stations. Through hardware decoupling, uplink and downlink transmissions can be completely independent and can be flexibly adjusted to meet users' diverse and constantly changing service and coverage requirements. However, the fully decoupled network also brings new challenges to downlink MIMO transmission. Because the uplink and downlink base stations are hardware isolated, the feedback mechanism of the existing network is no longer applicable to the fully decoupled wireless access network architecture. Therefore, the fully decoupled wireless access network uses a spectrum map to select transmission parameters, that is, fixed transmission parameters are used for multiple-input and multiple-output transmission corresponding to a fixed geographical location. But how to select fixed transmission parameters is still a problem.
[0004] In summary, the problems existing in current wireless communications and 6G fully decoupled wireless access network architectures include: (1) Channel information can be represented as a channel representation with higher information entropy, but there is no good method for obtaining channel representations. (2) In 6G fully decoupled networks, it is difficult to select fixed transmission parameters for establishing spectrum maps. Summary of the invention
[0005] Purpose of the invention: The purpose of the present invention is to provide a 6G fully decoupled network representative channel generation method and system based on contrastive learning, which uses contrastive learning to extract the characterization information of the multipath channel, and restores the multipath channel characterization to the multipath channel through a decoder, thereby obtaining the representative channel for determining the fixed transmission parameters of the spectrum map.
[0006] Technical solution: To achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a 6G fully decoupled network representative channel generation method based on contrastive learning, comprising the following steps:
[0007] Step 1: Establish a multipath channel model and a MIMO OFDM system transmission model;
[0008] Step 2: providing a representative channel generation target according to the MIMO OFDM system transmission model;
[0009] Step 3: Establish a multipath channel information dataset based on the multipath channel model, and use the contrastive learning framework to train the Transformer-based encoder, which takes the multipath channel information as input and outputs the multipath channel representation information;
[0010] Step 4: training a Transformer-based decoder according to the multipath channel information and the multipath channel representation information, wherein the decoder takes the multipath channel representation information as input and outputs the multipath channel information;
[0011] Step 5: In the actual deployment phase, the multipath channel information is input, the encoder outputs the multipath channel characterization information, the multipath channel characterization information averaged in the time dimension is input into the decoder, and the representative channel is output. Further, the multipath channel model is:
[0012]
[0013] Among them, t is the sampling time, e is the natural constant; N k is the number of subcarriers of the MIMO OFDM system transmission model, N p is the number of multipaths in the multipath channel, B is the bandwidth of the MIMO OFDM system transmission model; b, u, k are the serial numbers of the base station, user and subcarrier; They represent the power, phase and delay of the multipath channel between base station b and user u at time t respectively. are the departure angle, zenith arrival angle and directional arrival angle of the multipath between base station b and user u at time t; H b,u ,t,k represents the multipath channel on subcarrier k between base station b and user u at time t;
[0014] is the array response matrix of users and base stations;
[0015] The multipath channel between base station b and user u at sampling time t ∪ represents concatenating matrices along a new dimension, N R is the number of physical antenna ports at the receiving end, N T is the number of physical antenna ports at the transmitting end; a fixed base station b and user u are represented by a base station user pair {b,u}, and N t multipath channels, and the multipath channel information is
[0016] Furthermore, for a user using a uniform linear antenna and a base station using a uniform planar array antenna, the array response matrix is expressed as:
[0017]
[0018] in is the Kronecker product; T is the matrix transpose operation; a x (·),a y (·),a z (·) are the array response matrices in the x, y, and z directions respectively:
[0019]
[0020] Among them, N x ,N y ,N z are the number of physical antenna ports of the base station in the x, y, and z directions respectively.
[0021] Furthermore, the MIMO OFDM system transmission model is:
[0022]
[0023] Among them, x b,u,t,k is the signal sent by base station b to user u on subcarrier k at time t; u,t,k is the received signal of user u on subcarrier k at time t; W b,u,t,k is the precoding matrix of base station b to user u on subcarrier k at time t; nb,u,t,k is the additive complex Gaussian white noise from base station b to user u on subcarrier k at time t.
[0024] Furthermore, the representative channel The build targets are:
[0025]
[0026] in, is with H b,i ,t,k are matrices with the same dimensions but do not vary with t; N l is the number of transmission layers; is the variance of the additive complex Gaussian white noise; ||·|| F is the Frobenius norm; ||·||2 is the 2-norm; It is to take the singular vector operation, that is, to take the first N after the singular value decomposition. l The singular vectors corresponding to the largest singular values.
[0027] Furthermore, the Transformer-based encoder decomposes the multipath channel information into multiple multipath channel information blocks, and maps the multipath channel information blocks into a size of d through a fully connected network. model Vector, then add sine / cosine position encoding to the vector:
[0028]
[0029] Where pos is the position code number, i is the vector element number, PE, They are the position encoding single element and the overall output respectively; the vector then undergoes multiple self-attention mechanisms; the self-attention mechanism maps the input to a vector of size d through three independent fully connected layers atten The weights of these three fully connected layers are represented by Ω Θ ,Ω K ,Ω V , the outputs of the three fully connected layers are represented by Θ, K, and V respectively, and the self-attention It is expressed as:
[0030] Θ=Ω Θ ~input,K=Ω K ~input,v=Ω V ~input,
[0031]
[0032] Where input represents input, and softmax is a normalized exponential function. Finally, the output of the self-attention mechanism is average pooled and subjected to a fully connected layer to obtain the output of the Transformer-based encoder, which represents the multipath channel information.
[0033] Furthermore, the contrastive learning framework uses positive and negative samples to achieve self-supervised learning. Consider a positive sample of multipath channel information and Negative samples of multipath channel information The contrastive learning training framework is used to train the two Transformer-based encoders, which make the output of the multipath channel information at different sampling times in the positive samples as similar as possible, while the output of the positive samples and the negative samples are as dissimilar as possible, respectively. pos ,ε neg It means that ε pos The loss function is:
[0034]
[0035] Where exp is the natural exponential function; t′ is another sampling time different from t, and γ is the temperature parameter. neg The parameter θ neg According to ε pos The parameter θ pos ,renew:
[0036] θ neg ←βθ neg +(1-β)θ pos ,
[0037] Among them, β represents the update factor. The multipath channel information negative samples are maintained by a first-in-first-out queue, and each training neg Output a batch of channel characterization information into the queue, and make the batch of channel characterization information that first enters the queue exit the queue.
[0038] Furthermore, the input of the Transformer-based decoder is the multipath channel characterization information, and the output is the multipath channel information. Its purpose is to restore the input according to the output of the Transformer-based encoder. The loss function of the Transformer-based decoder is:
[0039]
[0040] in is the Transformer-based decoder, F b,u , t is the multipath information channel representation output by the encoder, N uis the total number of users.
[0041] Based on the same inventive concept, the present invention provides a 6G fully decoupled network representative channel generation system based on contrastive learning, including: a modeling module, used to establish a multipath channel model and a multi-input multi-output orthogonal frequency division multiplexing system transmission model, and give a representative channel generation target according to the multi-input multi-output orthogonal frequency division multiplexing system transmission model; an encoder training unit, used to establish a multipath channel information data set according to the multipath channel model, and use a contrastive learning framework to train a Transformer-based encoder, the encoder takes the multipath channel information as input, and outputs multipath channel characterization information; a decoder training unit, used to train a Transformer-based decoder according to the multipath channel information and multipath channel characterization information, the decoder takes the multipath channel characterization information as input, and outputs the multipath channel information; and a representative channel generation unit, used to input multipath channel information in the actual deployment stage, the encoder outputs the multipath channel characterization information, the multipath channel characterization information averaged in the time dimension is input into the decoder, and the representative channel is output.
[0042] Based on the same inventive concept, the present invention provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a 6G fully decoupled network representative channel generation method based on contrastive learning.
[0043] Beneficial effects: Compared with the prior art, the present invention has the following advantages: 1. The present invention provides a representative channel generation method that is feasible in a 6G fully decoupled network, avoiding the feedback defects caused by the hardware isolation of the 6G fully decoupled network. 2. The present invention proposes a contrastive learning method to obtain the characterization of the wireless channel. By utilizing the small-scale time domain changes of the channel, a contrastive learning agent task is designed to train an encoder that can map channel information to channel characterization information. 3. The present invention uses deep learning to fit the relationship between channel characterization and channel information, and obtains a decoder that can restore the channel characterization to the channel. Through experimental comparison, the method of the present invention can obtain representative channels for the establishment of spectrum maps of 6G fully decoupled access networks. BRIEF DESCRIPTION OF THE DRAWINGS
[0044] Figure 1 It is a flow chart of the 6G fully decoupled network representative channel generation method based on contrastive learning adopted in an embodiment of the present invention.
[0045] Figure 2 4 is a flow chart of a comparative learning framework according to an embodiment of the present invention.
[0046] Figure 3 Schematic diagram of the neural network structure of the encoder and decoder of an embodiment of the present invention.
[0047] Figure 4 It is a bar chart comparing the performance of the representative channel and the original channel in the embodiment of the present invention.
[0048] Figure 5 It is a line chart comparing the performance of the representative channel and the original channel according to the embodiment of the present invention.
[0049] Figure 6 It is a cumulative distribution function diagram comparing the performance of the representative channel and the original channel according to an embodiment of the present invention. DETAILED DESCRIPTION
[0050] In order to make the purpose, technical method and advantages of the present invention more clear, the embodiments of the present invention are described in detail in conjunction with the accompanying drawings: This embodiment is implemented on the premise of the technical method of the present invention, and a detailed implementation method and a specific operation process are given. It should be understood that the specific examples described here are only used to explain the present invention, but the protection scope of the present invention is not limited to the following embodiments.
[0051] The embodiment of the present invention discloses a novel 6G fully decoupled network representative channel generation method based on contrastive learning. Figure 1 The process of representative channel generation for a 6G fully decoupled network based on contrastive learning is shown. First, a multipath channel model and a multi-input multi-output orthogonal frequency division multiplexing system transmission model are established; secondly, a representative channel generation target is given based on the multi-input multi-output orthogonal frequency division multiplexing system transmission model; thirdly, a multipath channel information dataset is established based on the multipath channel model, and a Transformer-based encoder is trained using a contrastive learning framework. The encoder takes the multipath channel information as input and outputs the multipath channel characterization information. Then, based on the multipath channel information and the multipath channel characterization information, a Transformer-based decoder is trained. The decoder takes the multipath channel characterization information as input and outputs the multipath channel information. Finally, in the actual deployment stage, the multipath channel information is input, the encoder outputs the multipath channel characterization information, the averaged multipath channel characterization information is input into the decoder, and the representative channel is output.
[0052] Specifically, the embodiments of the present invention consider a single-user downlink scenario in a fully decoupled network.
[0053] The specific steps of channel generation are given below.
[0054] In this example, the transmitter antenna is a dual-polarized uniform planar array antenna with N x =8,N y =2,N z = 1 physical antenna port, with a total of N T =2N x N y Nz = 32 physical antenna ports. The receiver antenna is a single-polarized uniform linear antenna with a total of N R = 4 physical antenna ports. The frequency selective fading channel is:
[0055]
[0056] Among them, t is the sampling time, e is the natural constant; N k is the number of subcarriers of the MIMO OFDM system transmission model, N p is the number of multipaths in the multipath channel, B is the bandwidth of the MIMO OFDM system transmission model; b, u, k are the serial numbers of the base station, user and subcarrier; They represent the power, phase and delay of the multipath channel between base station b and user u at time t respectively. are the departure angle, zenith arrival angle and directional arrival angle of the multipath between base station b and user u at time t; H b,u ,t ,k represents the multipath channel on subcarrier k between base station b and user u at time t; is the array response matrix of the user and base station:
[0057]
[0058] in is the Kronecker product; T is the matrix transpose operation; a x (·),a y (·),a z (·) are the array response matrices in the x, y, and z directions respectively
[0059]
[0060] The multipath channel between base station b and user u at sampling time t ∪ represents the concatenation matrix along a new dimension. A fixed base station b and user u are represented by a base station user pair {b,u}. N t multipath channels, and the multipath channel information is
[0061]
[0062] The specific steps for establishing the transmission model of the MIMO OFDM system are given below:
[0063]
[0064] Among them, x b,u,t,kis the signal sent by base station b to user u on subcarrier k at time t; u,t,k is the received signal of user u on subcarrier k at time t; W b,u,t,k is the precoding matrix of base station b to user u on subcarrier k at time t; n b,u,t,k is the additive complex Gaussian white noise from base station n to user u on subcarrier k at time t.
[0065] The representative channels are given below Build Target:
[0066]
[0067] in, is with H b,i ,t ,k A matrix with the same dimensions but not varying with t; N l is the number of transmission layers; is the variance of the additive complex Gaussian white noise; ||·|| F is the Frobenius norm; ||·||2 is the 2-norm; It is to take the singular vector operation, that is, to take the first N after the singular value decomposition. l The singular vector corresponding to the largest singular value. Since the objective function is a nonlinear logarithmic sum form, it is difficult to solve mathematically.
[0068] The specific implementation steps of the Transformer-based encoder are given below.
[0069] like Figure 3 As shown in the figure, the Transformer-based encoder decomposes the multipath channel information into multiple multipath channel information blocks, and maps the multipath channel information blocks into a size of d through a fully connected network. model Vector, then add sine / cosine position encoding to the vector:
[0070]
[0071] Where pos is the position code number, i is the vector element number, They are the position encoding single element and the overall output respectively. The vector then undergoes L self-attention mechanisms. The self-attention mechanism maps the input to a vector of size d through three independent fully connected layers. atten The weights of these three fully connected layers are represented by Ω Θ ,Ω K ,Ω V The outputs of the three fully connected layers are represented by Θ, K, and V respectively. It is expressed as:
[0072] Θ=ΩΘ ·input,K=Ω K ·input,V=Ω V ·input,
[0073]
[0074] Where input represents input, and softmax is a normalized exponential function. Finally, the output of the self-attention mechanism is average pooled and a fully connected layer is used to obtain the output of the Transformer-based encoder, which represents the multipath channel information.
[0075] The specific implementation steps of the contrastive learning framework are given below.
[0076] like Figure 2 As shown, consider a positive sample of multipath channel information and Negative samples of multipath channel information The contrastive learning framework trains two Transformer-based encoders, which make the output of multipath channel information at different sampling times in positive samples as similar as possible, while the output of positive samples and negative samples as dissimilar as possible, using ε pos ,ε neg It means that ε pos The loss function is designed as:
[0077]
[0078] Where exp is the natural exponential function; t′ is another sampling time different from t, and γ is the temperature parameter. neg The parameter θ neg According to ε pos The parameter θ pos ,renew:
[0079] θ neg ←βθ neg +(1-β)θ pos ,
[0080] Among them, β represents the update factor. The multipath channel information negative samples are maintained by a first-in-first-out queue, and each training neg Output a batch of channel characterization information into the queue, and make the batch of channel characterization information that first enters the queue exit the queue.
[0081] The specific implementation steps of the Transformer-based decoder are given below.
[0082] The input of the Transformer-based decoder is the multipath channel characterization information, and the output is the multipath channel information. Its purpose is to restore the input according to the output of the Transformer-based encoder. The loss function of the Transformer-based decoder is:
[0083]
[0084] in is the Transformer-based decoder, F b,u , t is the multipath information channel representation output by the encoder, N u is the total number of users.
[0085] In order to make this embodiment more intuitive and reflect the performance advantages of the method, the present invention first generates multipath channel information in matlab2023a version. The present invention sets the number of multipaths N p =10, number of subcarriers N k =128, bandwidth B=192MHz, and the remaining channel parameters are determined by the ray tracing model. 50 channel data are collected between 2 base stations and 3030 users respectively, and a total of 303,000 multipath channel information is collected. Secondly, the present invention trains the encoder and decoder in Python3.8 and Pytorch1.13.0 environments. The hardware platform used in the present invention is the 12th generation Intel Core i7-12700H. The learning rate of the encoder is designed to be 5e-5. The encoder will divide the input into 16*16 blocks and take d mdel =512, and undergoes L = 4 self-attention mechanisms, and finally maps the output to a vector of size 128 through a fully connected layer. The decoder and encoder use the same d mdel ,L and learning rate, but the block size is 1*1.
[0086] The present invention uses throughput as a performance indicator to verify the method proposed by the present invention:
[0087] 1. If Figure 4 As shown, we will transmit the number of layers N l They are set to 1, 2, 3, and 4 respectively to test the performance of representative channels under different transmission settings, and the average throughput of 3030 users is obtained. Figure 4 It shows that directly averaging the original channel cannot obtain a good representative channel, and the throughput is lower than that of the original channel. However, the representative channel obtained by the method proposed in this invention averages the multipath channel characterization information and then restores it to the multipath channel through the decoder, thereby achieving better performance than the original channel, which shows that our method is effective. In addition, as N lAs the number of transmission layers increases, the performance improvement of the present invention will become smaller. This is because a higher number of transmission layers requires a more accurate description of the multipath. This also shows that the representative channel of the present invention has an accurate description of the multipath characteristics.
[0088] 2. If Figure 5 , Figure 6 As shown, the present invention is directed to N l = 1, the throughput performance of each user is specifically observed. It can be seen from the figure that the representative channel obtained by the present invention is generally better than the representative channel performance obtained by directly averaging the original channel, which shows that the method of the present invention is universal and can be extended to other scenarios.
[0089] Based on the same inventive concept, an embodiment of the present invention discloses a 6G fully decoupled network representative channel generation system based on contrastive learning, including: a modeling module, used to establish a multipath channel model and a multi-input multi-output orthogonal frequency division multiplexing system transmission model, and give a representative channel generation target according to the multi-input multi-output orthogonal frequency division multiplexing system transmission model; an encoder training unit, used to establish a multipath channel information data set according to the multipath channel model, and use a contrastive learning framework to train a Transformer-based encoder, the encoder takes the multipath channel information as input, and outputs multipath channel characterization information; a decoder training unit, used to train a Transformer-based decoder according to the multipath channel information and the multipath channel characterization information, the decoder takes the multipath channel characterization information as input, and outputs the multipath channel information; and a representative channel generation unit, used to input the multipath channel information in the actual deployment stage, the encoder outputs the multipath channel characterization information, the multipath channel characterization information averaged in the time dimension is input into the decoder, and the representative channel is output.
[0090] Based on the same inventive concept, an embodiment of the present invention discloses a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of a method for generating representative channels of a 6G fully decoupled network based on contrastive learning.
[0091] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A 6G fully decoupled network representative channel generation method based on contrastive learning, It is characterized in that The following steps are involved: Step 1: Establish a multipath channel model and a MIMO OFDM system transmission model; Step 2: providing a representative channel generation target according to the MIMO OFDM system transmission model; Step 3: Establish a multipath channel information dataset based on the multipath channel model, and use the contrastive learning framework to train the Transformer-based encoder, which takes the multipath channel information as input and outputs the multipath channel representation information; Step 4: training a Transformer-based decoder according to the multipath channel information and the multipath channel representation information, wherein the decoder takes the multipath channel representation information as input and outputs the multipath channel information; Step 5: In the actual deployment phase, the multipath channel information is input, the encoder outputs the multipath channel representation information, the multipath channel representation information averaged in the time dimension is input into the decoder, and the representative channel is output.
2. According to claim 1, a 6G fully decoupled network representative channel generation method based on contrastive learning, Features: The multipath channel model is: Among them, t is the sampling time, e is the natural constant; N k is the number of subcarriers of the MIMO OFDM system transmission model, N p is the number of multipaths in the multipath channel, B is the bandwidth of the MIMO OFDM system transmission model; b, u, k are the serial numbers of the base station, user and subcarrier; Respectively represent the power, phase and delay of the multipath channel between base station b and user u at time t; θ b,u,t , are the departure angle, zenith arrival angle and directional arrival angle of the multipath between base station b and user u at time t; H b,u,t,k represents the multipath channel on subcarrier k between base station b and user u at time t; a UE (θ b,u,t ), is the array response matrix of users and base stations; The multipath channel between base station b and user u at sampling time t ∪ represents concatenating matrices along a new dimension, N R is the number of physical antenna ports at the receiving end, N T is the number of physical antenna ports at the transmitting end; a fixed base station b and user u are represented by a base station user pair {b,u}, and N t multipath channels, and the multipath channel information is 3. According to the 6G fully decoupled network representative channel generation method based on contrastive learning according to claim 2, Features: For a user using a uniform linear antenna and a base station using a uniform planar array antenna, the array response matrix is expressed as: in is the Kronecker product; T is the matrix transpose operation; a x (·),a y (·),a z (·) are the array response matrices in the x, y, and z directions respectively: Among them, N x ,N y ,N z are the number of physical antenna ports of the base station in the x, y, and z directions respectively.
4. According to the 6G fully decoupled network representative channel generation method based on contrastive learning according to claim 1, Features: The MIMO OFDM system transmission model is: Among them, x b,u,t,k is the signal sent by base station b to user u on subcarrier k at time t; u,t,k is the received signal of user u on subcarrier k at time t; W b,u,t,kl is the precoding matrix of base station b to user u on subcarrier k at time t; H b,u,t,k is the multipath channel on subcarrier k between base station t and user u at time t; n b,u,t,k is the additive complex Gaussian white noise from base station b to user u on subcarrier k at time t; N b is the total number of base stations.
5. According to claim 1, a 6G fully decoupled network representative channel generation method based on contrastive learning, Features: The representative channel The build targets are: in, is with H b,u,t,k A matrix with the same dimensions but not varying with t; H b,u,t,k is the multipath channel on subcarrier k between base station b and user u at time t; N l is the number of transmission layers; is the variance of the additive complex Gaussian white noise; ∥·∥ F is the Frobenius norm; ∥·∥ 2 is the 2-norm; v Nl It is to take the singular vector operation, that is, to take the first N after the singular value decomposition. l The singular vectors corresponding to the largest singular values.
6. According to claim 1, a 6G fully decoupled network representative channel generation method based on contrastive learning, Features: The Transformer-based encoder decomposes the multipath channel information into multiple multipath channel information blocks, and maps the multipath channel information blocks into blocks of size d through a fully connected network. model Vector, then add sine / cosine position encoding to the vector: Where pos is the position code number, i is the vector element number, PE, They are the position encoding single element and the overall output respectively; the vector then undergoes multiple self-attention mechanisms; the self-attention mechanism maps the input to a vector of size d through three independent fully connected layers atten The weights of these three fully connected layers are represented by Ω Θ ,Ω K ,Ω V , the outputs of the three fully connected layers are represented by Θ, K, and V respectively, and the self-attention It is expressed as: I = O Θ ·input,K=Ω K ·input,V=Ω V ·input, Where input represents input, and softmax is a normalized exponential function. Finally, the output of the self-attention mechanism is average pooled and subjected to a fully connected layer to obtain the output of the Transformer-based encoder, which represents the multipath channel information.
7. According to claim 2, a 6G fully decoupled network representative channel generation method based on contrastive learning, Features: The contrastive learning framework uses positive and negative samples to achieve self-supervised learning. Consider a positive sample of multipath channel information. and Q multipath channel information negative samples The contrastive learning training framework is used to train the two Transformer-based encoders, which make the output of the multipath channel information at different sampling times in the positive samples as similar as possible, while the output of the positive samples and the negative samples are as dissimilar as possible, respectively. pos ,ε neg It means that ε pos The loss function is: Where exp is the natural exponential function; t′ is another sampling time different from t, γ is the temperature parameter; ε neg The parameter θ neg According to ε pos The parameter θ pos ,renew: i neg ←bth neg +(1-β)θ pos , Among them, β represents the update factor; Q multipath channel information negative samples are maintained by a first-in-first-out queue, and each training ε neg Output a batch of channel characterization information into the queue, and make the batch of channel characterization information that first enters the queue exit the queue.
8. According to claim 1, a 6G fully decoupled network representative channel generation method based on contrastive learning, Features: The input of the Transformer-based decoder is multipath channel representation information, and the output is multipath channel information; The loss function of the Transformer-based decoder is: in is the Transformer-based decoder, F b,u,t is the multipath information channel representation output by the encoder, H b ,u,t is the multipath channel between base station b and user u; N b is the total number of base stations, N u is the total number of users, N t is the number of adopted moments.
9. A 6G fully decoupled network representative channel generation system based on contrastive learning, It is characterized in that include: A modeling module, used to establish a multipath channel model and a multiple-input multiple-output orthogonal frequency division multiplexing system transmission model, and to provide a representative channel generation target according to the multiple-input multiple-output orthogonal frequency division multiplexing system transmission model; An encoder training unit is used to establish a multipath channel information dataset according to a multipath channel model, and train a Transformer-based encoder using a contrastive learning framework. The encoder takes the multipath channel information as input and outputs multipath channel representation information; A decoder training unit, used for training a Transformer-based decoder according to the multipath channel information and the multipath channel characterization information, wherein the decoder takes the multipath channel characterization information as input and outputs the multipath channel information; And a representative channel generation unit is used to input multipath channel information in the actual deployment stage, the encoder outputs multipath channel characterization information, the multipath channel characterization information averaged in the time dimension is input into the decoder, and the representative channel is output.
10. A computer program product comprising a computer program / instructions, It is characterized in that When the computer program / instructions are executed by the processor, the steps of a 6G fully decoupled network representative channel generation method based on contrastive learning are implemented according to any one of claims 1-8.
Citation Information
Cited By
Contrastive learning-based 6g fully-decoupled network representative channel generation method and system
WO2026020873A1
