CN118353744A - Packet channel estimation method and device based on heavy parameter and coordinate attention - Google Patents
Packet channel estimation method and device based on heavy parameter and coordinate attention Download PDFInfo
- Publication number
- CN118353744A CN118353744A CN202410485366.6A CN202410485366A CN118353744A CN 118353744 A CN118353744 A CN 118353744A CN 202410485366 A CN202410485366 A CN 202410485366A CN 118353744 A CN118353744 A CN 118353744A
- Authority
- CN
- China
- Prior art keywords
- channel
- ris
- matrix
- heavy
- channel estimation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 239000011159 matrix material Substances 0.000 claims abstract description 108
- 238000004590 computer program Methods 0.000 claims description 14
- 238000012549 training Methods 0.000 claims description 12
- 230000008569 process Effects 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 6
- 239000013598 vector Substances 0.000 claims description 5
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000011176 pooling Methods 0.000 claims description 2
- 238000004891 communication Methods 0.000 abstract description 11
- 230000006870 function Effects 0.000 abstract description 10
- 238000013507 mapping Methods 0.000 abstract description 9
- 238000013528 artificial neural network Methods 0.000 abstract description 8
- 238000001829 resonance ionisation spectroscopy Methods 0.000 description 59
- 238000010586 diagram Methods 0.000 description 12
- 238000000605 extraction Methods 0.000 description 9
- 238000011084 recovery Methods 0.000 description 8
- 238000013527 convolutional neural network Methods 0.000 description 4
- 230000007423 decrease Effects 0.000 description 4
- 238000012545 processing Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 230000004927 fusion Effects 0.000 description 3
- 238000001228 spectrum Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000007667 floating Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 description 1
- 241001168730 Simo Species 0.000 description 1
- 239000000654 additive Substances 0.000 description 1
- 230000000996 additive effect Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 238000004806 packaging method and process Methods 0.000 description 1
- 230000010363 phase shift Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention belongs to the technical field of wireless communication, and provides a packet channel estimation method and a packet channel estimation device based on heavy parameters and coordinate attention, wherein the technical scheme is as follows: based on the strong learning ability of the neural network and the correlation characteristics between the RIS channels, a grouping channel estimation method based on heavy parameters and coordinate attention is provided, and pilot frequency overhead is effectively reduced by grouping RIS adjacent reflection units. Further, the nonlinear mapping function of the neural network is utilized to extend the low-dimensional matrix to the high-dimensional matrix, so that the complete CSI is obtained. By applying the method, the channel estimation precision can be improved with low pilot frequency overhead and low complexity.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a packet channel estimation method and device based on heavy parameters and coordinate attention.
Background
The statements in this section merely provide background information related to the present disclosure and may not necessarily constitute prior art.
In future 6G research, low power consumption and low cost are critical, and researchers are continually looking for suitable approaches to keep low cost and low power consumption at high communication capacity. RIS is receiving extensive attention in academia and industry as a 6G potential technology. In 5G, millimeter waves (MILLIMETER WAVE, mmWave) can solve the problem of spectrum shortage, improve communication capacity, but have poor coverage capability. The coverage capacity of signals can be increased by arranging RIS, and the defects of mmWave are overcome. To fully exploit RIS performance, accurate Channel State Information (CSI) needs to be obtained by using a channel estimation technique, but because RIS is composed of a large number of reflection units, huge pilot overhead is caused, and spectrum efficiency is reduced, so that higher pilot overhead is a great challenge.
The huge pilot frequency expense is caused by the large number of the reflection units of the RIS, so that the pilot frequency expense can be effectively reduced and the frequency spectrum efficiency can be improved by grouping the reflection units of the RIS and then estimating the channel. The grouped channel matrix is a low-dimensional matrix, and if all CSI is wanted, a linear interpolation method can be adopted, but huge estimation errors are brought. In recent years, deep learning has been attracting attention as a powerful nonlinear mapping method, and a mapping relation between a low-dimensional channel matrix and a high-dimensional channel matrix can be constructed through a neural network, so that complete and accurate CSI can be obtained.
Literature [1](Channel estimation in IRS-enhanced mmWave system with super-resolution network[J].IEEE Communications Letters,2021,25(8):2599-2603.) uses Super-resolution convolutional neural networks (Super-Resolution Convolutional Neural Network, SRCNN) in OFDM systems to extend CSI estimated from partial pilots to full CSI. SRCNN is poor in recovery quality and high in complexity. The fast super-resolution convolutional neural network (FAST SRCNN, FSRCNN) further proposed in literature [2](Dong C,Loy C,Tang X.Accelerating the super-resolution convolutional neural network[C].Proceeding of the European Conference on Computer Vision,Amsterdam,The Netherlands,2016:391-407.) can effectively improve the problem and recover super-resolution images by upsampling, but the recovery quality is still limited. The literature [3](Jin Y,Zhang J,Zhang X,et al.Channel estimation for semi-passive reconfigurable intelligent surfaces with enhanced deep residual networks[J].IEEE Transactions on Vehicular Technology,2021,70(10):11083-11088.) adopts semi-passive RIS, part of RIS reflecting units adopt active RIS, which can process signals, but power consumption and cost are correspondingly brought, and the literature [3] adopts enhanced deep Super-Resolution (ENHANCED DEEP Super-Resolution, EDSR) network, so that complete CSI can be obtained effectively. The multi-scale supervised learning is performed by using the laplacian pyramid wide residual error network (LAPLACIAN PYRAMID WIDE residual network, lapWRes) proposed by the literature [4](Xiao J,Wang J,Xie W,et al.Multi-scale supervised learning-based channel estimation for RIS-aided communication systems[C].Proceeding of the IEEE Wireless Communications and Networking Conference,Glasgow,United Kingdom,2023:1-6,article no.10119023.), and the recovery precision can be effectively improved by adopting multi-scale recovery, but greater complexity is brought. Literature [5](Feng H,Zhao Y.MmWave RIS-assisted SIMO channel estimation based on global attention residual network[J].IEEE Wireless Communications Letters,2023,12(7):1179-1183) proposes a global attention residual network (Global Attention Residual Network, GARN) based on the global attention residual network structure, finding the importance between features, which further improves the channel estimation accuracy.
The method keeps low training expenditure to different degrees and can effectively improve the channel estimation precision, but has larger complexity, and how to finish high-precision channel estimation with low complexity under the condition of ensuring low training expenditure is still one of the problems to be solved at present.
Disclosure of Invention
In order to solve at least one technical problem in the background art, the invention provides a grouping channel estimation method and a grouping channel estimation device based on heavy parameters and coordinate attention, which are based on the strong learning ability of a neural network and the correlation characteristics between the channels of RIS.
In order to achieve the above purpose, the present invention adopts the following technical scheme:
a first aspect of the present invention provides a packet channel estimation method based on heavy parameters and coordinate attention, comprising the steps of:
Establishing an uplink RIS-mmWave multi-input multi-output system model by using a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
Calculating according to the received signal matrix to obtain a channel initial estimation matrix;
obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
A second aspect of the present invention provides a packet channel estimation apparatus based on heavy parameters and coordinate attention, comprising:
The receiving signal matrix acquisition module is used for establishing an uplink RIS-mmWave multi-input multi-output system model by utilizing a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
the channel initial estimation matrix calculation module is used for calculating a channel initial estimation matrix according to the received signal matrix;
the grouping channel estimation module is used for obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
A third aspect of the present invention provides a computer-readable storage medium.
A computer readable storage medium having stored thereon a computer program which when executed by a processor performs the steps in a packet channel estimation method based on heavy parameters and coordinate attention as described above.
A fourth aspect of the invention provides a computer device.
A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in a method of packet channel estimation based on heavy parameters and coordinate attention as described above when the program is executed.
Compared with the prior art, the invention has the beneficial effects that:
The invention effectively reduces pilot frequency overhead by grouping RIS adjacent reflection units. Further, the nonlinear mapping function of the neural network is utilized to extend the low-dimensional matrix to the high-dimensional matrix, so that the complete CSI is obtained. By applying the method, the channel estimation precision can be improved with low pilot frequency overhead and low complexity.
Additional aspects of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention.
Fig. 1 is a flowchart of a packet channel estimation method based on heavy parameters and coordinate attention according to an embodiment of the present invention;
FIG. 2 is a RIS-mmWave communication system provided by an embodiment of the present invention;
FIG. 3 is a schematic diagram of a RIS reflection unit grouping provided by an embodiment of the present invention;
FIG. 4 is a schematic diagram of a heavy parameter and coordinate attention network architecture provided by an embodiment of the present invention;
FIG. 5 is a schematic view of an RFB structure provided in an embodiment of the present invention;
Fig. 6 is a schematic structural diagram of an RB structure provided by an embodiment of the present invention;
FIG. 7 is a schematic diagram of a coordinate attention block structure according to an embodiment of the present invention;
Fig. 8 is an NMSE versus SNR curve for different packet channel estimation methods provided by embodiments of the present invention;
Fig. 9 is a convergence curve of different packet channel estimation methods according to the embodiment of the present invention;
fig. 10 is a NMSE curve for three cases where the number V of packet reflection units provided in the embodiment of the present invention is 2, 4, and 8, respectively.
Detailed Description
The invention will be further described with reference to the drawings and examples.
It should be noted that the following detailed description is illustrative and is intended to provide further explanation of the invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
It is noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of exemplary embodiments according to the present invention. As used herein, the singular is also intended to include the plural unless the context clearly indicates otherwise, and furthermore, it is to be understood that the terms "comprises" and/or "comprising" when used in this specification are taken to specify the presence of stated features, steps, operations, devices, components, and/or combinations thereof.
The invention discloses a packet channel estimation method based on heavy parameters and coordinate attention, which comprises the following steps: first, RIS adjacent reflection units are grouped, reducing pilot overhead. The LS algorithm can obtain a low-dimensional initial estimation matrix. The matrix is preprocessed by the separation of the virtual part and the real part and then is input into a heavy parameter and coordinate attention network, and the matrix is restored into a high-dimensional matrix. The heavy parameter and coordinate attention network comprises a shallow feature extraction module, a deep feature extraction module and a feature fusion and recovery module. The shallow feature module consists of 1 convolution block and initially extracts shallow features. The deep feature extraction module consists of 4 heavy parameter feature blocks, wherein the heavy parameter feature blocks consist of 3 heavy parameter blocks, 1 convolution block and 1 coordinate attention module. The feature is effectively and further extracted by utilizing the heavy parameter structure, the coordinate attention block can effectively extract the cross-channel information and the position information of the channel matrix to obtain the deep feature, and then the shallow feature and the deep feature are fused through the residual structure, so that the feature expression capability is improved. Finally, a high-dimensional matrix is obtained through a recovery module formed by transpose convolution. And finally, combining the real parts of the imaginary parts to obtain complete channel state information. The invention realizes high estimation precision with low training overhead and low complexity, and greatly improves the channel estimation performance.
Example 1
As shown in fig. 1, the present embodiment provides a packet channel estimation method based on heavy parameters and coordinate attention, including the steps of:
step 1: establishing an uplink RIS-mmWave multi-input multi-output system model by using a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
Step2: calculating according to the received signal matrix to obtain a channel initial estimation matrix;
Step 3: obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
According to the invention, the RIS adjacent reflection units are grouped, so that the pilot frequency overhead is effectively reduced; further, the nonlinear mapping function of the neural network is utilized to extend the low-dimensional matrix to the high-dimensional matrix, so that the complete CSI is obtained. By applying the method, the channel estimation precision can be improved with low pilot frequency overhead and low complexity.
The channel estimation method provided by the invention mainly comprises two stages: initial estimation is performed using RIS reflection unit packets and LS, after which the full CSI is recovered using the heavy parameter and coordinate attention network.
FIG. 2 shows an RIS-mmWave communication system, which includes a Base Station (BS) with a fixed location, a RIS with a fixed location, and a plurality of single-antenna User Equipments (UEs);
The system is provided with an RIS-UE channel and a BS-RIS channel, and the two channels are combined into a reflection cascade channel;
the direct channel between the BS and the UE is blocked by an obstacle, and cannot directly transfer signals;
The RIS adjacent reflection units are grouped, and the reflection units in each group share the same reflection coefficient;
The user sends an orthogonal pilot signal at a certain moment, and the orthogonal pilot signal reaches a BS receiving end after RIS reflection, and a receiving signal is obtained based on a cascade channel matrix.
Establishing an uplink RIS-mmWave multiple-input multiple-output (Multiple Input Multiple Output, MIMO) system model by using a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and enabling a user side to send orthogonal pilot frequency to reach a BS side through the RIS;
The UE transmits pilot signals to the BS through the RIS, adjacent reflection units on the RIS are grouped, and the same group has common reflection coefficient. It is assumed that the direct links of the UE and BS sides are blocked. BS and RIS set m=m 1×M2 antennas and n=n 1×N2 reflection units, respectively, with K single antenna users. The system has two channels, namely RIS-UE and BS-RIS.
In step 1, the step of establishing an uplink RIS-mmWave mimo system model, grouping RIS adjacent reflection units, and obtaining a received signal matrix of a base station based on a reflection cascade channel matrix formed by combining an RIS-UE channel and a BS-RIS channel, includes:
step 101: calculating a channel matrix between the RIS and the BS, and a channel model of the RIS and the UE;
The BS antenna adopts a Uniform planar array (uniformity PLANAR ARRAY, UPA), the RIS reflection unit also adopts UPA, and a Saleh-Valenzuela model is utilized to establish a channel.
Channel matrix between RIS and BSAs shown in formula (1):
wherein L 1 is the number of paths between BS and RIS, For the first 1 -path complex gain,The horizontal angle and the pitch angle of the arrival angle of the first 1 paths,The horizontal angle and pitch angle are the exit angles of the first 1 paths. (. Cndot.) H is the conjugate transpose, a M (-) is the M-dimensional array steering vector, and a N (-) is the N-dimensional array steering vector.
Channel model for RIS and kth user U k As shown in formula (2):
Where L 2 is the number of paths between BS and U k, For the first 2 -path complex gain,The horizontal angle and the pitch angle are the arrival angles of the first 2 paths.
Array steering vectors of formulas (1) and (2) can be usedThe representation is made of a combination of a first and a second color,As shown in formula (3):
Wherein, And xi is the elevation angle of the light beam,Η is the horizontal angle of the plane,Is Kronecker product, (. Cndot.) T is transposed, P is the number of antennas or reflecting elements, P 1 is the number of antennas or reflecting elements in the vertical direction, P 2 is the number of antennas or reflecting elements in the horizontal direction, λ c is the carrier wavelength, d is the element spacing, and d is assumed to be less than or equal to λ c/2.
Step 102: determining a reflection matrix of the RIS;
As shown in fig. 2, a direct channel between the BS and the UE may be blocked by an obstacle, and a signal cannot be directly transmitted. The signal is reflected by RIS, the amplitude and phase of the incident signal are adjusted, and the reflection matrix of RIS As shown in formula (4):
Wherein, beta n epsilon [0,1] and alpha n epsilon [0,2 pi ] are the amplitude and phase shift of the nth (N epsilon [1,2, …, N ]) passive reflection unit in RIS respectively.
Step 103: combining the channel matrix between the RIS and the BS obtained in the step 101, the channel matrix between the RIS and the user and the reflection matrix of the RIS obtained in the step 102 to obtain a receiving signal of the BS end;
The whole channel estimation process comprises B subframes, and the user U k sends orthogonal pilot frequency symbol sequence with length T in the B (B E B) th subframe Receiving signals at the BS end through two channels H and H k generated in the step 101 and through reflection of RIS in the middle of the two channelsAs shown in formula (5):
Wherein, For Additive White Gaussian Noise (AWGN) with Noise power σ 2, I M is an all 1 matrix with dimension M, and g= Hdiag (h k) is a concatenated channel.
Step 104: grouping RIS adjacent reflecting units, and acquiring a receiving signal matrix of a base station based on the cascade channels after grouping;
FIG. 3 is a group of RIS reflection units, N RIS adjacent reflection units being divided into groups Groups, each group havingThe same reflection coefficient is shared by adjacent reflection units, and the RIS reflection unit can be expressed asThe concatenated channel after grouping is shown in equation (6):
Wherein, And is also provided with
The received signal may be expressed as:
Wherein, the orthogonal pilot frequency can effectively eliminate the inter-group interference. After the reflection units are grouped, the reflection units in the group share the same reflection coefficient, and the CSI of one group is estimated, so that the pilot frequency overhead of V times can be effectively reduced.
Step 105: the orthogonal pilot signal in step 104 hasWherein P t is the transmit power. Using the orthogonal pilot characteristic, right-hand multiplication of x k,b for equation (7) eliminates the pilot sequence as shown in equation (8):
Wherein, AndThe received signals of b subframes are superimposed, and Y k is obtained by equation (8) as shown in equation (9):
Wherein, In order to receive the matrix of signals,In the form of a matrix of reflection coefficients,Is a noise matrix.
Step 106: if the RIS unit in step 104 is not divisible, it is divided intoA group, in which the reflection units share the same reflection coefficient,For rounding down, the RIS unit not in the packet obtains the CSI by adopting the channel estimation method of LS, and the RIS unit in the packet enters step 2.
Step 2: and (3) obtaining a low-dimensional channel matrix which is estimated initially by using the received signal matrix obtained in the step (1) through LS, and constructing a data set by using the Saleh-Valenzuela channel model in the step (1).
Step 201: this step corresponds to the initially estimated LS portion of FIG. 1. Obtaining an initial estimated low-dimensional matrix by LS method from equation (9) in step 105As shown in formula (10):
step 202: using G generated in step 103, Y k generated in step 105, and initial estimate in step 201 Generating corresponding data sets
Step 3: this step corresponds to stage 2 of fig. 1, aiming at obtaining a high-dimensional channel estimation matrix through nonlinear mapping of the deep learning network. The method comprises the steps of constructing a heavy parameter and coordinate attention network, and forming the heavy parameter and coordinate attention network by a shallow feature extraction module, a deep feature extraction module and a feature fusion and recovery module.
Step 301: initially estimating step 201Is divided into two real matricesAndThe combined into 2-channel tensors are used as network inputs.
Step 302: heavy parameter and coordinate attention network as shown in fig. 4, this step corresponds to the shallow feature extraction section of fig. 4. The 2-channel tensor obtained in the step 301 is input into a shallow feature extraction module composed of 3×3 convolution layers and mapped into a 36×h×w shallow feature map, and the step can effectively extract shallow features.
Step 303: the shallow feature map obtained in step 302 is input into a deep feature extraction module composed of four heavy parameter feature blocks (Reparameterized Feature Block, RFB) and mapped into a 36×h×w deep feature map.
Step 3031: an RFB module was constructed, which corresponds to the RFB of fig. 5, with a residual structure consisting of 3 heavy parameter blocks (Reparameterized Block, RB) and 13 x 3 convolutional layers, followed by a coordinate attention module.
Step 3032: and building an RB module, wherein the step corresponds to the RB of fig. 6. A heavy parameter structure consisting of 13 x 3 convolutional layer, 13 x 1 convolutional layer and 1 x 3 convolutional layer, and a parallel residual structure consisting of one 1 x 1 convolutional layer and one block of 1 x 1 convolutional layer and 3 x 3 convolutional layer in series. And by utilizing the re-parameterization and parallel structure, the feature extraction capability is effectively improved.
Step 3033: the correlation of the horizontal direction and the vertical direction of the channel matrix is utilized to build a coordinate attention block, the horizontal direction of the channel matrix represents channels of different RISs, the vertical direction represents channels of a BS antenna, and the correlation can be utilized to better recover the complete CSI.
This step corresponds to the coordinate attention block of fig. 7. And (3) carrying out pooling on an input of C multiplied by H multiplied by W in the horizontal direction and the vertical direction respectively, mapping the input of C multiplied by H multiplied by 1 and C multiplied by 1 multiplied by W into characteristic diagrams, carrying out Concat and carrying out 1 multiplied by 1 convolution layer compression channels, carrying out weight normalization and nonlinear activation layer coding on spatial information in the vertical and horizontal directions, carrying out 1 multiplied by 1 convolution layer to restore the spatial information to the original channel number, carrying out sigmoid activation respectively, and carrying out normalization weighting to obtain the C multiplied by H multiplied by W output characteristic diagram. The location information is embedded into the channel attention so that the network obtains information over a larger area without introducing significant overhead.
Step 304: the shallow features obtained in step 302 are fused with the deep features obtained in step 303.
Step 305: inputting the fusion characteristic obtained in step 307 into a recovery module composed of 1×9 transpose convolutions to obtain 2× (v×h) × (v×w) tensor, and changing the matrix composed of real part and imaginary part into complex matrix form to obtain channel estimation matrix
Step 4: and (3) packaging the modules in the step 3 into an integral network for end-to-end training and testing. The nonlinear mapping relation from the channel initial estimation matrix to the estimated channel matrix is shown in the formula (11):
Wherein, the xi is a weight parameter.
Step 5: training the network packaged in the step 4, and training by using the data set in the step 2. Parameters are set. The parameter settings of the RIS-mmWave communication system are shown in Table 1.
TABLE 1 parameter settings
And (3) generating 20000 group training data set by utilizing the step (2), wherein the training purpose is to reach accurate channel estimation by optimizing the minimum loss function of the Xi. The loss function is represented by mean square error (Mean Square Error, MSE) as shown in equation (12):
Wherein D is the number of training data. Other training parameter settings are shown in table 2.
Table 2 parameter settings
The data is trained with a Signal-to-Noise Ratio (SNR) of 20 dB. Comparing normalized mean square errors (Normalized Mean Square Error, NMSE) at different SNRs and different packet numbers, respectively, the NMSE being shown in equation (13):
Wherein, For the purpose of desire.
The method can ensure that the channel estimation precision is improved with low complexity under the condition of low pilot frequency overhead, and the channel estimation performance is improved. To demonstrate that the present invention has higher channel estimation accuracy and lower training overhead without significantly increasing complexity, NMSEs are compared at different SNRs and different packets. And comparing model parameters and floating point numbers (FLOPs) of the model with those of other models, proving that the complexity of the method is low, and comparing the convergence of the method with other methods, and verifying the performance of the network of the invention.
The number of pilot frequencies required for linear estimation is greater than or equal to the number of RIS reflection units, the RIS reflection units are grouped, each group contains V reflection units, and pilot frequency overhead can be effectively reduced by V times.
Fig. 9 compares the present invention with LS, FSRCNN, EDSR, GARN, lapWRes different packet channel estimation methods in case of packet V of 4. LS is to recover the complete CSI by linear interpolation after initial estimation. As can be seen from fig. 9, as the SNR increases, the NMSE decreases, and the present invention has a smaller NMSE with higher estimation accuracy at the same SNR.
Fig. 9 compares the convergence of the method for estimating the deep learning packet channel of the present invention with that of FSRCNN, EDSR, GARN, lapWRes, and as can be seen from fig. 9, the proposed network has better convergence.
Fig. 10 is a NMSE curve for the present invention in three cases where the number of packet reflection units V is 2, 4 and 8, respectively. As can be seen from fig. 10, as V decreases, the estimation accuracy increases. Conversely, as the V number increases, the estimation accuracy decreases, but the pilot overhead decreases. Therefore, under the high signal-to-noise ratio, the pilot frequency overhead can be greatly reduced, the high estimation precision is kept, and at the same time, under the low signal-to-noise ratio, the high estimation precision can be kept with lower pilot frequency overhead.
Table 3 shows FLOPs and model parameters for different deep learning models, FLOPs is the number of floating point operations, representing the computational complexity of the model, and model parameters are the total number of parameters that need to be trained, representing the spatial complexity of the model. Compared with EDSR, GARN and LapWRes, the invention has lower FLOPs and model parameters, which shows that the invention can keep low complexity under the condition of effectively improving the channel estimation accuracy, and FSRCNN has low complexity but lower estimation accuracy compared with the invention.
Based on the strong learning ability of the neural network and the correlation characteristics between the channels of the RIS, the invention provides a grouping channel estimation method based on heavy parameters and coordinate attention. The method effectively reduces pilot frequency overhead by grouping RIS adjacent reflection units. Further, the nonlinear mapping function of the neural network is utilized to extend the low-dimensional matrix to the high-dimensional matrix, so that the complete CSI is obtained. By applying the method, the channel estimation precision can be improved with low pilot frequency overhead and low complexity.
Table 3 complexity comparison
Example two
The present embodiment provides a packet channel estimation apparatus based on a heavy parameter and a coordinate attention, including:
The receiving signal matrix acquisition module is used for establishing an uplink RIS-mmWave multi-input multi-output system model by utilizing a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
the channel initial estimation matrix calculation module is used for calculating a channel initial estimation matrix according to the received signal matrix;
the grouping channel estimation module is used for obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
Example III
The present embodiment provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, implements the steps in a packet channel estimation method based on heavy parameters and coordinate attention as described above.
Example IV
The present embodiment provides a computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the steps in a method of packet channel estimation based on heavy parameters and coordinate attention as described above when executing the program.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
Those skilled in the art will appreciate that implementing all or part of the above-described methods in accordance with the embodiments may be accomplished by way of a computer program stored on a computer readable storage medium, which when executed may comprise the steps of the embodiments of the methods described above. The storage medium may be a magnetic disk, an optical disk, a Read-Only Memory (ROM), a random-access Memory (Random Access Memory, RAM), or the like.
The above description is only of the preferred embodiments of the present invention and is not intended to limit the present invention, but various modifications and variations can be made to the present invention by those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A method for packet channel estimation based on heavy parameters and coordinate attention, comprising:
Establishing an uplink RIS-mmWave multi-input multi-output system model by using a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
Calculating according to the received signal matrix to obtain a channel initial estimation matrix;
obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
2. The method for grouped channel estimation based on heavy parameter and coordinate attention of claim 1, wherein when RIS neighboring reflection units are grouped, reflection units within each group share the same reflection coefficient.
3. The method for packet channel estimation based on heavy parameters and coordinate attention according to claim 1, wherein when shallow features are extracted based on a channel initial estimation matrix, real and imaginary parts of the channel initial estimation matrix are separated into two real matrices, and after fused channel estimation features are obtained, the two real matrices are transformed into complex matrices.
4. The method for packet channel estimation based on heavy parameter and coordinate attention of claim 1, wherein the uplink RIS-mmWave multiple-input multiple-output system model comprises: a channel matrix between RIS and BS and a channel model of RIS and UE;
Wherein, the channel matrix between RIS and BS is:
wherein L 1 is the number of paths between BS and RIS, For the first 1 -path complex gain,The horizontal angle and the pitch angle of the arrival angle of the first 1 paths,The horizontal angle and pitch angle are the exit angles of the first 1 paths. (. Cndot.) H is the conjugate transpose, a M (-) is the M-dimensional array steering vector, and a N (-) is the N-dimensional array steering vector;
the channel model for RIS and kth user U k is:
Where L 2 is the number of paths between BS and U k, For the first 2 -path complex gain,The horizontal angle and the pitch angle are the arrival angles of the first 2 paths.
5. The method for estimating a packet channel based on heavy parameters and coordinate attention according to claim 1, wherein when the channel primary estimation matrix is calculated based on the received signal, if the RIS units cannot be divided by each other when the RIS adjacent reflection units are grouped, the RIS units not in the group use a least square channel estimation method to obtain the channel primary estimation matrix.
6. The method for estimating a packet channel based on a heavy parameter and a coordinate attention as recited in claim 1, wherein the extracting deep features using cross-channel information and physical location information of a channel matrix based on a plurality of heavy parameter blocks and coordinate attention blocks comprises:
Each heavy parameter characteristic block comprises a residual structure formed by 3 heavy parameter blocks and 1 convolution layer, and then a coordinate attention module is added, wherein each heavy parameter block comprises a heavy parameter structure formed by a plurality of convolution layers with different sizes and a parallel residual structure formed by one convolution layer and a block formed by two convolution layers with different sizes in series; the correlation of the horizontal direction and the vertical direction of a channel matrix is utilized to build a coordinate attention block, the horizontal direction of the channel matrix represents channels of different RIS, the vertical direction represents channels of a BS antenna, the characteristics output by a residual structure are respectively subjected to average pooling in the horizontal direction and the vertical direction, then pass through Concat and pass through a 1X 1 convolution layer laminated channel, then pass through a weight normalization and a nonlinear activation layer to encode space information in the vertical direction and the horizontal direction, then Split, respectively pass through the 1X 1 convolution layer to restore to the original channel number, respectively activate through sigmoid, and finally normalize and weight to obtain deep characteristics.
7. The method for packet channel estimation based on heavy parameters and coordinate attention as recited in claim 1, wherein the training data set is generated by using a Saleh-Valenzuela channel model and a calculated channel preliminary estimation matrix when the packet channel estimation model is trained.
8. A packet channel estimation device based on a heavy parameter and a coordinate attention, comprising:
The receiving signal matrix acquisition module is used for establishing an uplink RIS-mmWave multi-input multi-output system model by utilizing a Saleh-Valenzuela channel model, grouping RIS adjacent reflection units, and acquiring a receiving signal matrix based on a reflection cascade channel matrix formed by combining a grouped RIS-UE channel and a BS-RIS channel;
the channel initial estimation matrix calculation module is used for calculating a channel initial estimation matrix according to the received signal matrix;
the grouping channel estimation module is used for obtaining a channel estimation matrix based on the channel initial estimation matrix and the trained grouping channel estimation model; the construction process of the grouping channel estimation model comprises the following steps:
The method comprises the steps of extracting shallow features based on a channel initial estimation matrix, extracting deep features based on the shallow features and a plurality of heavy parameter feature blocks, wherein each heavy parameter feature block comprises a plurality of heavy parameter blocks and a coordinate attention block, extracting deep features based on the cross-channel information and physical position information of the channel matrix based on the heavy parameter blocks and the coordinate attention block, and fusing the shallow features and the deep features to obtain fused channel estimation features.
9. A computer readable storage medium, on which a computer program is stored, which program, when being executed by a processor, implements the steps in the method for packet channel estimation based on heavy parameters and coordinate attention as claimed in any one of claims 1-7.
10. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the steps in the method of packet channel estimation based on heavy parameters and coordinate attention as claimed in any of claims 1-7 when the program is executed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410485366.6A CN118353744A (en) | 2024-04-22 | 2024-04-22 | Packet channel estimation method and device based on heavy parameter and coordinate attention |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410485366.6A CN118353744A (en) | 2024-04-22 | 2024-04-22 | Packet channel estimation method and device based on heavy parameter and coordinate attention |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118353744A true CN118353744A (en) | 2024-07-16 |
Family
ID=91820706
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410485366.6A Pending CN118353744A (en) | 2024-04-22 | 2024-04-22 | Packet channel estimation method and device based on heavy parameter and coordinate attention |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118353744A (en) |
-
2024
- 2024-04-22 CN CN202410485366.6A patent/CN118353744A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111181619B (en) | Millimeter wave hybrid beam forming design method based on deep reinforcement learning | |
| JP4840608B2 (en) | MIMO decoder and MIMO decoding method | |
| CN107332598A (en) | A kind of precoding of mimo system joint and antenna selecting method based on deep learning | |
| CN111713035B (en) | MIMO multi-antenna signal transmission and detection technology based on artificial intelligence | |
| CN109714091B (en) | Iterative hybrid precoding method based on hierarchical design in millimeter wave MIMO system | |
| Elbir et al. | Federated learning for physical layer design | |
| CN111555781B (en) | Large-scale MIMO channel state information compression and reconstruction method based on deep learning attention mechanism | |
| CN106160831B (en) | A kind of signal detecting method and device | |
| CN113078932B (en) | Intelligent reflection surface assisted downlink transmission precoding design method | |
| CN113193893B (en) | Millimeter wave large-scale MIMO intelligent hybrid beam forming design method | |
| CN109951214A (en) | A kind of signal detecting method suitable for extensive mimo system | |
| CN114884549A (en) | Large-scale MIMO channel state information feedback method based on deep learning | |
| CN116192307A (en) | Distributed cooperative multi-antenna cooperative spectrum intelligent sensing method, system, equipment and medium under non-Gaussian noise | |
| CN114157331B (en) | Large-scale MIMO channel state information feedback method based on pseudo complex value input | |
| CN114785642A (en) | RIS-MIMO system channel estimation method based on sparse tensor decomposition | |
| CN114567359A (en) | CSI feedback method based on multi-resolution fusion convolution feedback network in large-scale MIMO system | |
| CN107733487B (en) | Signal detection method and device for large-scale multi-input multi-output system | |
| Huang et al. | Self-attention reinforcement learning for multi-beam combining in mmWave 3D-MIMO systems | |
| CN118353744A (en) | Packet channel estimation method and device based on heavy parameter and coordinate attention | |
| CN107346985B (en) | Interference alignment method combined with transmitting antenna selection technology | |
| CN115276878A (en) | Intelligent reflecting surface safety rate determination method and device based on inverse iteration and manifold optimization | |
| CN110855338B (en) | FD-MIMO downlink self-adaptive transmission method based on two-layer precoding | |
| Couras et al. | Semi-blind receivers based on a coupled nested Tucker-PARAFAC model for dual-polarized MIMO systems using combined TST and MSMKron codings | |
| Kumagai et al. | Deep unfolding-assisted fully decentralized projected gradient MIMO detection algorithm | |
| CN101795180A (en) | MIMO detection pretreatment method with optimal sequencing feature and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |