Channel Estimation Method in RIS-Assisted Communication System Based on Piecewise Pilot Transmission

Abstract

The present invention discloses a channel estimation method in a RIS-assisted communication system based on segmented pilot transmission. This estimation method is realized by two parts: an offline channel estimation stage and an online channel estimation stage. In the offline channel estimation stage, a channel reconstruction network is built, a training set is generated by using a selection matrix to train the channel reconstruction network, and a frame structure for segmented pilot transmission is designed. In the online channel estimation stage, the selected antennas at the base station and the selected reflection elements at the RIS are respectively connected to the radio frequency chains by using the trained selection matrix. The RIS obtains the sampled channel through segmented pilot transmission, and inputs the sampled channel into the trained channel reconstruction network to output the estimated complete channel. The antennas and reflection elements selected in the present invention correspond to the optimal channel reconstruction performance, and the designed frame structure for segmented pilot transmission realizes the estimation of the direct link channel by the RIS.

Description

Technical Field

[0001] The present invention belongs to the field of communication technologies, and further relates to a channel estimation method in a reconfigurable intelligent surface (RIS)-assisted communication system based on segmented pilot transmission in the field of digital information transmission technologies. The present invention can be used in radio scattering scenarios to meet the communication requirements of multiple users. Background Art

[0002] The reconfigurable intelligent surface (RIS) can achieve intelligent regulation of the communication scattering environment. By deploying a large number of passive reflection units between the transceiver, the RIS can provide additional reflection paths for blocked shadow areas, and is thus considered a key technology for enhancing coverage in millimeter-wave communications. Generally, to fully exploit the above advantages of the RIS, accurate channel state information needs to be obtained first. However, the introduction of the RIS poses numerous challenges to channel estimation in wireless communication systems. First, due to the passive nature of the RIS, the two independent channels associated with the RIS are difficult to be directly estimated, and existing technical solutions often focus on the estimation of the channel obtained after cascading the two independent channels. At the same time, due to the deployment of a large number of RIS reflection units, the dimension of the cascaded channel is large, resulting in a huge pilot training overhead.

[0003] Nantong University disclosed a channel estimation method in an RIS-assisted communication system based on deep learning in its patent document "A Channel Estimation Method for a Passive Intelligent Reflective Surface Based on Deep Learning" (Application No.: 202110436060.8, Publication No.: CN 113179232A, Application Date: April 22, 2021). The implementation steps of this method are as follows: First, generate a fully cascaded channel sample. Then, use the method of equi-probability uniform sampling to obtain a sampled cascaded channel sample. Secondly, train a residual neural network to learn the mapping relationship from the sampled cascaded channel to the fully cascaded channel. Finally, input the online estimated sampled cascaded channel into the trained residual neural network to reconstruct the fully cascaded channel. This method combines the residual neural network in deep learning with channel estimation, and the designed equi-probability uniform selection method significantly reduces the pilot training overhead for estimating the cascaded channel. However, there are still deficiencies in the channel estimation method for this RIS-assisted communication system: In this method, the reflection units at the RIS are selected using channel coherence, and the selection result cannot be dynamically adjusted according to the loss error of channel reconstruction during the network training process, resulting in the reconstruction performance not being able to reach the optimal.

[0004] Zhejiang University published a channel estimation method in its patent application document "Channel Estimation Method for an Intelligent Reflecting Surface System Assisted by Semi-Passive Reflecting Elements" (Application No.: 202110036638.0, Publication No.: CN 112929302 A, Application Date: January 12, 2021) for a RIS-assisted communication system assisted by semi-passive reflecting elements. The implementation steps of this method are as follows. First, build an intelligent reflecting surface system assisted by semi-passive reflecting elements. The direct link between the base station and the user is blocked, and a small number of semi-passive reflecting elements are configured in the RIS. These semi-passive reflecting elements can be connected to the radio frequency link and tuned to the receiving mode to receive pilot signals and achieve channel estimation. Then, divide a transmission frame into two transmission stages. In the first transmission stage, the base station sends pilot signals to the RIS, and the RIS estimates the channel from the base station to the RIS. In the second transmission stage, the user sends pilot signals to the RIS, and the RIS estimates the channel from the user to the RIS. This method provides limited signal processing capabilities for the RIS by configuring semi-passive reflecting elements. The designed transmission frame enables the RIS to estimate two independent channels, decouples the high-dimensional cascaded channel, and solves the problem of large pilot training overhead. However, there are still deficiencies in the channel estimation method in this RIS-assisted communication system: the application scenario of this method assumes that the direct link is blocked, and the designed transmission frame cannot enable the RIS to estimate the direct channel, making it inapplicable to the autonomous beam design of the RIS in scenarios where the direct link is not blocked. Summary of the Invention

[0005] The object of the present invention is to address the deficiencies of the above-mentioned existing technologies and propose a channel estimation method for a RIS-assisted communication system based on segmented pilot transmission, aiming to solve the problems in the existing technologies where the reconstruction performance cannot reach the optimal level and it is inapplicable to the autonomous beam design of the RIS in scenarios where the direct link is not blocked.

[0006] The idea for achieving the object of the present invention is as follows: The present invention provides a channel estimation method for a RIS-assisted communication system based on segmented pilot transmission. Based on a channel extrapolation network constructed by probability selection training, use the trained selection matrix to complete the selection of antennas at the base station and the selection of reflection units at the RIS respectively. The selection matrix is iteratively updated according to the loss error of channel reconstruction during the network training process, so that the selected antennas and reflection units correspond to the optimal channel reconstruction performance, overcoming the problem in the existing technology that the reconstruction performance cannot reach the optimal level; use segmented pilot transmission to perform channel estimation at the RIS. The segmented pilot transmission takes the base station as a relay communication node, enabling the RIS to estimate the direct channel, overcoming the problem in the existing technology that it is inapplicable to the autonomous beam design of the RIS in scenarios where the direct link is not blocked; the estimation method is implemented by two parts: offline channel estimation and online signal estimation; among them, the offline channel estimation includes the following steps:

[0007] Step 1: Build a channel reconstruction network and set the parameters of the channel reconstruction network;

[0008] Step 2: Generate the training set of the channel reconstruction network:

[0009] Generate the sampled channel training set of the user communication node and the sampled channel training set of the RIS node respectively; Combine the sampled channel training set of the user communication node with the label set, and the sampled channel training set of the RIS node with the label set to form the training set of the channel reconstruction network;

[0010] Step 3: Train the channel reconstruction network:

[0011] Input the training set of the channel reconstruction network into the channel reconstruction network, calculate the error between the output data and the label data of the channel reconstruction network, and through the backpropagation algorithm, iteratively update the parameters of the channel reconstruction network until the network function converges, obtaining the trained selection matrix T at the base station B , the selection matrix T at the RIS I , the channel reconstruction network;

[0012] Step 4: Design the frame structure of segmented pilot transmission:

[0013] Step 4.1: Divide the transmission frame of channel estimation in the RIS-assisted communication system into P + 1 block time slots, where the value of P is equal to the total number of users in the RIS-assisted communication system;

[0014] Step 4.2: In the i-th block time slot, the j-th user sends its pilot signal to the base station, and the base station, as a relay communication node, forwards its received signal to the RIS, where the values of i and j correspond equally; The RIS uses the LS algorithm to estimate the sampled cascaded channel in the first P block time slots

[0015] Step 4.3: In the (P + 1)-th block time slot, the base station sends its pilot signal to the RIS, and the RIS uses the LS algorithm to estimate the sampled channel of the RIS node in the (P + 1)-th block time slot Divide element-wise division to obtain the sampled channel of the user communication node in the first P block time slots

[0016] Online channel estimation includes the following steps:

[0017] Step 5: Connect the selected antennas at the base station to the radio frequency chain, and connect the selected reflection units at the RIS to the radio frequency chain;

[0018] Step 6: The RIS uses the frame structure of segmented pilot transmission to obtain the sampled channel of the user communication node Sampled channels of RIS nodes Combine with and input the combined result into the trained channel reconstruction network to output the estimated complete channel.

[0019] Compared with the prior art, the present invention has the following advantages:

[0020] First, the channel extrapolation network constructed by the present invention based on probability selection training uses the trained selection matrix to complete the selection of antennas at the base station and reflection units at RIS respectively. The selection matrix is iteratively updated according to the loss error of channel reconstruction during the network training process, so that the selected antennas and reflection units correspond to the optimal channel reconstruction performance, overcoming the problem that the reconstruction performance in the prior art cannot reach the optimal; making the selected antennas and reflection units of the present invention correspond to the optimal channel reconstruction performance.

[0021] Second, the present invention uses segmented pilot transmission to estimate the direct channel at RIS. The segmented pilot transmission uses the base station as a relay communication node, enabling RIS to estimate the direct channel, overcoming the problem in the prior art that the autonomous beam design of RIS cannot be applied to the scenario where the direct link is not blocked, making the frame structure of the segmented pilot transmission designed by the present invention achieve the estimation of the direct channel by RIS. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a schematic diagram of the application scenario of the present invention;

[0023] Figure 2 is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0024] The present invention will be further described below with reference to the drawings and embodiments.

[0025] Refer to the attached Figure 1 , and further describe the application scenario of the embodiments of the present invention.

[0026] In the RIS-assisted communication scenario of the embodiments of the present invention, there is 1 base station BS, 1 RIS, and 3 communication users assisted by the RIS. Each communication user is configured with a single antenna. The base station is configured with a uniform cylindrical antenna array UCyA (Uniform Cylindrical Array) containing 32 antennas. The number of concentric arcs in the UCyA is 4, and the number of antennas in each concentric arc is 8. The 16 radio frequency chains equipped by the base station can connect 16 antennas to the baseband processing unit. The RIS is configured with a uniform planar array UPA (Uniform Plain Array) containing 64 reflection units, and the number of reflection units in both the vertical and horizontal directions of this array is 8. The 32 radio frequency chains equipped by the RIS can connect 32 reflection units to the baseband processing unit. The channel between the base station and the RIS is denoted as H, and the channels between the base station and the users are denoted as F = [f1, f2, f3] T , where f q represents the channel between the base station and the q-th user.

[0027] The following further describes the steps of implementing the present invention with reference to the attached Figure 2 drawings.

[0028] The embodiments of the present invention provide a channel estimation method in a RIS-assisted communication system based on segmented pilot transmission. A channel extrapolation network constructed based on probabilistic selection training is used, and the trained selection matrix is utilized to complete the selection of antennas at the base station and the selection of reflection units at the RIS. The segmented pilot training strategy is used to estimate the direct channel at the RIS; the estimation method is implemented by two parts: offline channel estimation and online channel estimation; among them, the offline channel estimation includes the following steps:

[0029] Step 1: Build a channel reconstruction network and set the parameters of the channel reconstruction network.

[0030] The structure of the built channel reconstruction network is successively composed of a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer, a third convolutional layer, a third activation layer, a fourth convolutional layer, a fourth activation layer, a fifth convolutional layer, a fifth activation layer, and a sixth convolutional layer connected in series.

[0031] The parameters of the set channel reconstruction network are as follows: The kernel size of the first to sixth convolutional layers is set to 5; the number of kernels in the first to fifth convolutional layers is set to 128, and the number of kernels in the sixth convolutional layer is set to 2; the activation functions of the first to fifth activation layers are all implemented by the ReLU function.

[0032] Step 2: Generate the training set of the channel reconstruction network.

[0033] Generate the sampled channel training sets of the user communication nodes and the RIS nodes respectively; combine the sampled channel training set of the user communication nodes with the label set, and the sampled channel training set of the RIS nodes with the label set to form the training set of the channel reconstruction network;

[0034] The steps for generating the sampled channel training set of the user communication nodes are as follows:

[0035] First step, generate user communication node samples: In the embodiments of the present invention, the base station grid point BS1 is selected from the publicly available DeepMIMO dataset as the base station location of the embodiments of the present invention; from all the grid points from the 301st row to the 500th row in this dataset, the first 150 grid point positions in each row are selected, obtaining a total of 30,000 grid point positions to form a grid point position set. Randomly select 3 grid point positions from this grid point position set as the positions of the 3 users in the embodiments of the present invention. After associating these 3 positions with BS1, the 3 user communication node samples F are obtained. After 20,000 random selections, 20,000 user communication node samples are obtained;

[0036] Second step, generate the selection matrix T at the base station with dimensions Q×P B , where the number of rows of this selection matrix is the total number of antennas at the base station, and the number of columns is the total number of antennas at the base station to be selected when training the channel reconstruction network. In the embodiments of the present invention, Q = 32, P = 16, and T B is obtained by the following formula:

[0037] T B = [one_hot{s1},..., one_hot{s d},..., one_hot{s 16}]+[r1,..., r d ,..., r 16

[0038] -detach{[r1,..., r d ,..., r 16},

[0039]

[0040]

[0041] where, there are 16 non-zero elements in T B with different rows and columns. one_hot{·} represents the one-hot encoding operation. The mth non-zero element is located in the sth B row and the dth column of T d , and r d represents the one-hot vector one_hot{s d ​} corresponding probability soft value vector, detach{·} represents the operation of removing the gradient; argmax{·} represents the index of taking the maximum value. Indicates that the m-th non-zero element appears in T B in the probability parameter of the row, indicating interference T of the random noise; exp{·} represents the exponential function with base e, (·)

[0042] In the third step, using the formula, a sampled channel sample of the user communication node is obtained

[0043] In the fourth step, 20,000 sampled channel samples are used to form the sampled channel training set of the user communication node.

[0044] The steps to generate the sampled channel training set of the RIS node are as follows:

[0045] In the first step, RIS node samples are generated: In the embodiment of the present invention, the base station grid point BS1 is selected from the publicly available DeepMIMO dataset as the base station location of the embodiment of the present invention; among all the grid points from the 301st row to the 500th row in this dataset, the last 10 grid point positions in each row are selected. A total of 2,000 grid point positions form the positions of the RIS in the embodiment of the present invention. After associating one grid point position with BS1, an RIS node sample H is obtained. A total of 2,000 RIS node samples can be obtained, and each sample is amplified 10 times to obtain 20,000 RIS node samples;

[0046] In the second step, an M×N-dimensional selection matrix T I is generated at the RIS. The number of rows of this selection matrix is the total number of reflection units at the RIS, and the number of columns is the total number of reflection units to be selected at the RIS. In the embodiment of the present invention, M = 64 and N = 32. T I is obtained from the following formula:

[0047] T I = [one_hot{k1},..., one_hot{k g},..., one_hot{k 32}] + [l1,..., l g ,..., l 32

[0048] -detach{[l1,..., l g ,..., l 32},

[0049] ​

[0050]

[0051] Among them, T I has 32 non-zero elements in different rows and columns. The nth non-zero element is located in T I in the k g th row and the gth column, and l g represents the probability soft value vector corresponding to the one-hot vector one_hot{k g}; represents the probability parameter that the nth non-zero element appears in the th row, represents the interference of the random noise; the value of g corresponds to the value of n.

[0052] Step 3, use the formula to obtain a sampled channel sample of the RIS node

[0053] Step 4, form a sampled channel training set of the RIS node with 20,000 sampled channel samples.

[0054] Step 3, train the channel reconstruction network.

[0055] Input the training set of the channel reconstruction network into the channel reconstruction network, calculate the error between the output data and the label data of the channel reconstruction network, and through the backpropagation algorithm, iteratively update the parameters of the channel reconstruction network until the network function converges, and obtain the trained selection matrix T B at the base station, the selection matrix T I at the RIS, and the channel reconstruction network.

[0056] The loss function is as follows:

[0057]

[0058] Among them, represents the loss value between the predicted label output by the channel reconstruction network and the true label G of the channel reconstruction network. Q(·) represents the entropy value of the matrix. In the embodiments of the present invention

[0059] Step 4, design the frame structure of segmented pilot transmission.

[0060] Step 4.1, divide the transmission frame of channel estimation in the embodiments of the present invention into P + 1 block time slots, and the value of P is equal to the total number of users in the embodiments of the present invention.

[0061] Step 4.2: In the \(i\)-th block time slot, the \(j\)-th user sends its pilot signal to the base station. The base station, acting as a relay communication node, forwards its received signal to the RIS. The values of \(i\) and \(j\) are correspondingly equal. The RIS uses the LS algorithm to estimate the sampled cascaded channel in the first \(P\) block time slots.

[0062] Step 4.3: In the \((P + 1)\)-th block time slot, the base station sends its pilot signal to the RIS. The RIS uses the LS algorithm to estimate the sampled channel of the RIS node in the \((P + 1)\)-th block time slot. Divide point by to obtain the sampled channel of the user communication node in the first \(P\) block time slots.

[0063] Online channel estimation includes the following steps:

[0064] Step 5: Connect the selected antenna at the base station to the radio frequency chain, and connect the selected reflection unit at the RIS to the radio frequency chain.

[0065] The connection of the selected antenna at the base station to the radio frequency chain means extracting the non-zero elements in the trained selection matrix \(T\) B The base station uses the row and column indices of the non-zero elements to connect the \(x\)-th antenna to the \(y\)-th radio frequency chain. Among them, the total number of non-zero elements corresponds to the total number of radio frequency chains at the base station. The value of \(x\) corresponds to the row index of the \(m\)-th non-zero element, and the value of \(y\) corresponds to the column index of the \(m\)-th non-zero element.

[0066] The connection of the selected reflection unit at the RIS to the radio frequency chain means extracting the non-zero elements in the trained selection matrix \(T\) I The RIS uses the row and column indices of the non-zero elements to connect the \(u\)-th reflection unit to the \(v\)-th radio frequency chain. Among them, the total number of non-zero elements corresponds to the total number of radio frequency chains at the RIS. The value of \(u\) corresponds to the row index of the \(n\)-th non-zero element, and the value of \(v\) corresponds to the column index of the \(n\)-th non-zero element.

[0067] Step 6: The RIS uses the frame structure of segmented pilot transmission to obtain the sampled channel of the user communication node and the sampled channel of the RIS node Combine and input them into the trained channel reconstruction network, and output the estimated complete channel.

Claims

1. A channel estimation method in a RIS-assisted communication system based on segmented pilot transmission, characterized in that The channel extrapolation network constructed based on probability selection training uses the trained selection matrix to complete the selection of antennas at the base station and the selection of reflection units at the RIS, and estimates the direct channel at the RIS using segmented pilot transmission; this estimation method is implemented in two parts: offline channel estimation and online channel estimation; among them, the offline channel estimation includes the following steps: Step 1, build a channel reconstruction network and set the parameters of the channel reconstruction network; Step 2, generate the training set of the channel reconstruction network: Generate the sampled channel training set of the user communication node and the sampled channel training set of the RIS node respectively; combine the sampled channel training set of the user communication node with the label set, and the sampled channel training set of the RIS node with the label set to form the training set of the channel reconstruction network; Step 3, train the channel reconstruction network: Input the training set of the channel reconstruction network into the channel reconstruction network, calculate the error between the output data and the label data of the channel reconstruction network, and through the backpropagation algorithm, iteratively update the parameters of the channel reconstruction network until the network function converges, so as to obtain the trained selection matrix T at the base station B , the selection matrix T at the RIS I , the channel reconstruction network; Step 4, design the frame structure of segmented pilot transmission: Step 4.1, divide the transmission frame of channel estimation in the RIS-assisted communication system into P + 1 block time slots, where the value of P is equal to the total number of users in the RIS-assisted communication system; Step 4.2, within the \(i\)th block time slot, the \(j\)th user sends its pilot signal to the base station. The base station, acting as a relay communication node, forwards its received signal to the RIS, where the values of \(i\) and \(j\) are correspondingly equal. The RIS estimates the sampled cascaded channel within the first \(P\) block time slots using the LS algorithm Step 4.3, within the (P + 1)-th block time slot, the base station transmits its pilot signal to the RIS, and the RIS estimates the sampled channel of the RIS nodes within the (P + 1)-th block time slot using the LS algorithm Divide point by to obtain the sampled channels of the user communication nodes within the first P block time slots The online channel estimation includes the following steps: Step 5, connect the selected antennas at the base station to the radio frequency chain, and connect the selected reflection units at the RIS to the radio frequency chain; Step 6, the RIS obtains the sampled channels of the user communication nodes using the frame structure of the segmented pilot transmission and the sampled channels of the RIS nodes Combine and input them into the trained channel reconstruction network, and output the estimated complete channels.

2. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, wherein The structure of the channel reconstruction network built in Step 1 is successively composed of a first convolutional layer, a first activation layer, a second convolutional layer, a second activation layer, a third convolutional layer, a third activation layer, a fourth convolutional layer, a fourth activation layer, a fifth convolutional layer, a fifth activation layer, and a sixth convolutional layer connected in series.

3. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, wherein The parameters of the channel reconstruction network set in Step 1 are as follows: set the convolutional kernel size of the first to sixth convolutional layers to 5; set the number of convolutional kernels in the first to fifth convolutional layers to 128, and the number of convolutional kernels in the sixth convolutional layer to 2; implement the activation functions of the first to fifth activation layers with the ReLU function.

4. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, wherein The steps for generating the sampled channel training set of the user communication node in Step 2 are as follows: The first step, select the base station location, select at least 20,000 user locations, associate each user location with the base station to obtain the node sample F of this user, and obtain all user communication node samples; Step 2: Generate a selection matrix T of size Q×P at the base station, where the number of rows of the selection matrix is the total number of antennas at the base station, and the number of columns is the total number of antennas to be selected at the base station when training the channel reconstruction network; B , where the number of rows of the selection matrix is the total number of antennas at the base station, and the number of columns is the total number of antennas to be selected at the base station when training the channel reconstruction network; Step 3: Using the formula, obtain a sampled channel sample corresponding to each user communication node The fourth step, combine all the sampled channel samples to form the sampled channel training set of the user communication node.

5. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, characterized in that, The steps for generating the sampled channel training set of the RIS node in Step 2 are as follows: The first step, generate RIS node samples: select the base station location, select the same number of RIS locations as the number of user locations, associate the base station location with the RIS, and obtain each RIS node sample H; Step 2: Generate a selection matrix \(T\) at the RIS with dimensions \(M\times N\). I The number of rows of this selection matrix is the total number of reflection elements at the RIS, and the number of columns is the total number of reflection elements to be selected at the RIS. Step 3: Use the formula to obtain a sampled channel sample corresponding to each RIS node where (·) T denotes the transpose operation; The fourth step, combine all the sampled channel samples to form the sampled channel training set of the RIS node.

6. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, characterized in that, The loss function in Step 3 is as follows: Among them, represents the predicted label output by the channel reconstruction network The loss value between and the true label G of the channel reconstruction network, and Q(·) represents the entropy value of the matrix.

7. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, wherein The connection between the antenna selected at the base station in step 5 and the radio frequency chain means extracting the non-zero elements in the trained selection matrix T B at the base station. The base station uses the row and column indices of the non-zero elements to connect the x-th antenna to the y-th radio frequency chain, where the total number of non-zero elements corresponds to the total number of radio frequency chains at the base station, the value of x corresponds to the row index of the m-th non-zero element, and the value of y corresponds to the column index of the m-th non-zero element.

8. The channel estimation method in the RIS-assisted communication system based on segmented pilot transmission according to claim 1, characterized in that The connection between the selected reflection unit at the RIS described in step 5 and the RF chain means extracting the trained selection matrix T I The non-zero elements in are used. The RIS connects the u-th reflection unit to the v-th RF chain using the row and column indices of the non-zero elements. The total number of non-zero elements corresponds to the total number of RF chains at the RIS. The value of u corresponds to the row index of the n-th non-zero element, and the value of v corresponds to the column index of the n-th non-zero element.

Images