CN113726376A - 1bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion - Google Patents

1bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion Download PDF

Info

Publication number
CN113726376A
CN113726376A CN202111011688.XA CN202111011688A CN113726376A CN 113726376 A CN113726376 A CN 113726376A CN 202111011688 A CN202111011688 A CN 202111011688A CN 113726376 A CN113726376 A CN 113726376A
Authority
CN
China
Prior art keywords
csi
vector
downlink csi
amplitude
downlink
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.)
Granted
Application number
CN202111011688.XA
Other languages
Chinese (zh)
Other versions
CN113726376B (en
Inventor
卿朝进
叶青
刘文慧
黄小莉
曹太强
黄永茂
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xihua University
Original Assignee
Xihua University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xihua University filed Critical Xihua University
Priority to CN202111011688.XA priority Critical patent/CN113726376B/en
Publication of CN113726376A publication Critical patent/CN113726376A/en
Application granted granted Critical
Publication of CN113726376B publication Critical patent/CN113726376B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/02—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas
    • H04B7/04—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas
    • H04B7/0413—MIMO systems
    • H04B7/0417—Feedback systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/14—Two-way operation using the same type of signal, i.e. duplex
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D30/00—Reducing energy consumption in communication networks
    • Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Signal Processing (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Health & Medical Sciences (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本发明公开了基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法，包括：根据上行CSI估计矢量的幅度，通过第一神经网络获得对应的下行CSI的学习幅度；根据基站得到的恢复反馈向量，利用专家知识进行CSI特征提取，恢复出下行CSI的特征幅度与特征角度；根据对所述下行CSI的特征幅度与下行CSI的学习幅度拼接得到的下行CSI拼接幅度，通过第二神经网络获得下行CSI的融合幅度；根据所述下行CSI的融合幅度与所述特征角度，恢复得到下行CSI重构矢量。与单比特CS叠加CSI反馈相比，本发明可根据上下行信道双向互异性恢复出单比特CS丢失的下行CSI幅度，极大的提高了CSI的重构精度，同时显著提高了CSI的重构效率。

Figure 202111011688

The invention discloses a 1-bit compression and superposition CSI feedback method based on feature extraction and mutual fusion, comprising: obtaining the corresponding learning amplitude of downlink CSI through a first neural network according to the amplitude of the uplink CSI estimation vector; , using expert knowledge to extract CSI features, and recover the feature amplitude and feature angle of downlink CSI; according to the downlink CSI splicing amplitude obtained by splicing the feature amplitude of the downlink CSI and the learning amplitude of the downlink CSI, the downlink CSI is obtained through the second neural network. The fusion amplitude of the CSI; according to the fusion amplitude of the downlink CSI and the characteristic angle, the downlink CSI reconstruction vector is recovered and obtained. Compared with the single-bit CS superimposed CSI feedback, the present invention can recover the downlink CSI amplitude of the single-bit CS lost according to the bidirectional mutuality of the uplink and downlink channels, which greatly improves the CSI reconstruction accuracy and also significantly improves the CSI reconstruction. efficiency.

Figure 202111011688

Description

基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion

技术领域technical field

本发明涉及FDD(frequency division duplex)大规模MIMO(multiple inputmultiple output)系统的叠加反馈技术领域，特别涉及基于特征抽取与互异性融合的1bit压缩叠加信道状态信息(CSI，Channel State Information)反馈方法。The invention relates to the technical field of superposition feedback of FDD (frequency division duplex) massive MIMO (multiple input multiple output) system, in particular to a 1-bit compressed superposition channel state information (CSI, Channel State Information) feedback method based on feature extraction and mutual fusion.

背景技术Background technique

作为满足未来5G(the fifth generation wireless communication)网络高效频谱效率和能量效率的关键技术，FDD大规模MIMO系统通过基站端部署的上百根天线，能在不增加发射功率和系统带宽的情况下为更多用户提供无线数据服务。同时，FDD大规模MIMO系统中诸多带来性能提升的操作(如多用户调度、速率分配、发射端预编码等)依赖于准确的下行信道状态信息(CSI,channel state information)的获取。在频分双工(FDD，frequency division duplex)大规模MIMO系统中信道间的存在微弱的互惠性，CSI只能由用户端反馈回基站。As a key technology to meet the high spectral efficiency and energy efficiency of the future 5G (the fifth generation wireless communication) network, the FDD massive MIMO system can provide hundreds of antennas deployed at the base station without increasing the transmit power and system bandwidth. More users offer wireless data services. At the same time, many operations (such as multi-user scheduling, rate allocation, transmitter precoding, etc.) that bring about performance improvement in the FDD massive MIMO system depend on the acquisition of accurate downlink channel state information (CSI, channel state information). In a frequency division duplex (FDD, frequency division duplex) massive MIMO system, there is weak reciprocity between channels, and CSI can only be fed back to the base station by the user end.

传统利用信号稀疏性的压缩感知(CS,compressed sensing)反馈技术虽能在一定程度降低系统反馈开销，但在其重构过程中有大量的计算开销；基于深度学习(DL,DeepLearning)的反馈技术因其结构简单，训练速度快等优点，引起广泛关注，但在其反馈过程中仍占用一定的频谱资源。Although the traditional compressed sensing (CS, compressed sensing) feedback technology using signal sparsity can reduce the system feedback overhead to a certain extent, there is a large amount of computational overhead in the reconstruction process; the feedback technology based on deep learning (DL, DeepLearning) Because of its simple structure and fast training speed, it has attracted widespread attention, but it still occupies certain spectrum resources in its feedback process.

近年来，叠加编码(SC,superimposed coding)技术以其能够高效利用频谱资源的特性，被广泛应用于无线通信各领域，但叠加编码会造成相互干扰，单比特CS可以通过丢弃CSI的幅度信息减少这种相互干扰。CSI幅度信息的丢失造成CSI重构精度不高。In recent years, superimposed coding (SC) technology has been widely used in various fields of wireless communication because of its efficient use of spectrum resources. However, superimposed coding will cause mutual interference. Single-bit CS can be reduced by discarding the amplitude information of CSI. this mutual interference. The loss of CSI amplitude information results in low CSI reconstruction accuracy.

发明内容SUMMARY OF THE INVENTION

本发明的目的在于提供基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法，与单比特CS叠加CSI反馈相比，本发明根据上下行信道双向互异性恢复出单比特CS丢失的下行CSI幅度，极大的提高了CSI的重构精度，并且利用浅层神经网络与简化版本的CSI重构方法进行CSI重构，提高了CSI的重构效率。The purpose of the present invention is to provide a 1-bit compression superimposed CSI feedback method based on feature extraction and mutuality fusion. Compared with the single-bit CS superimposed CSI feedback, the present invention recovers the downlink CSI amplitude of the single-bit CS loss according to the bidirectional mutuality of the uplink and downlink channels. , which greatly improves the reconstruction accuracy of CSI, and uses a shallow neural network and a simplified version of the CSI reconstruction method to perform CSI reconstruction, which improves the reconstruction efficiency of CSI.

本发明的技术方案如下：The technical scheme of the present invention is as follows:

基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法，包括：A 1-bit compression and overlay CSI feedback method based on feature extraction and mutual fusion, including:

根据上行CSI估计矢量

Figure BDA0003239194990000021
的幅度
Figure BDA0003239194990000022
通过第一神经网络获得对应的下行CSI的学习幅度
Figure BDA0003239194990000023
Estimating vector based on uplink CSI
Figure BDA0003239194990000021
Amplitude
Figure BDA0003239194990000022
Obtain the learning amplitude of the corresponding downlink CSI through the first neural network
Figure BDA0003239194990000023

根据恢复反馈向量

Figure BDA0003239194990000024
利用专家知识进行CSI特征提取，恢复出下行CSI的特征幅度
Figure BDA0003239194990000025
与特征角度
Figure BDA0003239194990000026
其中，所述恢复反馈向量
Figure BDA0003239194990000027
是指基站接收端根据发射端发出的反馈向量w进行恢复后得到的对应反馈向量；According to the restored feedback vector
Figure BDA0003239194990000024
Use expert knowledge to extract CSI features and recover the feature magnitude of downlink CSI
Figure BDA0003239194990000025
with characteristic angle
Figure BDA0003239194990000026
Wherein, the recovery feedback vector
Figure BDA0003239194990000027
Refers to the corresponding feedback vector obtained by the base station receiving end after recovery according to the feedback vector w sent by the transmitting end;

根据对所述下行CSI的特征幅度

Figure BDA0003239194990000028
与下行CSI的学习幅度
Figure BDA0003239194990000029
拼接得到的下行CSI拼接幅度
Figure BDA00032391949900000210
通过第二神经网络获得下行CSI的融合幅度
Figure BDA00032391949900000211
According to the characteristic magnitude of the downlink CSI
Figure BDA0003239194990000028
Learning magnitude with downlink CSI
Figure BDA0003239194990000029
Downlink CSI splicing amplitude obtained by splicing
Figure BDA00032391949900000210
Obtaining the fusion magnitude of the downlink CSI through the second neural network
Figure BDA00032391949900000211

根据所述下行CSI的融合幅度

Figure BDA00032391949900000212
与所述特征角度
Figure BDA00032391949900000213
恢复得到下行CSI重构矢量
Figure BDA00032391949900000214
According to the fusion amplitude of the downlink CSI
Figure BDA00032391949900000212
with the characteristic angle
Figure BDA00032391949900000213
Recovery to obtain downlink CSI reconstruction vector
Figure BDA00032391949900000214

在一些具体实施方式中，所述上行CSI估计矢量幅度

Figure BDA00032391949900000215
通过以下模型获得：In some specific implementations, the uplink CSI estimated vector magnitude is
Figure BDA00032391949900000215
Obtained by the following models:

Figure BDA00032391949900000216
Figure BDA00032391949900000216

Figure BDA00032391949900000217
Figure BDA00032391949900000217

其中，

Figure BDA00032391949900000218
表示所述上行CSI的估计矢量
Figure BDA00032391949900000219
的第N个元素，|·|表示复数的取模操作，上标T表示转置运算；in,
Figure BDA00032391949900000218
the estimated vector representing the uplink CSI
Figure BDA00032391949900000219
The Nth element of , |·| represents the modulo operation of complex numbers, and the superscript T represents the transpose operation;

在一些具体实施方式中，所述上行CSI估计矢量

Figure BDA00032391949900000220
通过对用于进行信道估计的基站端接收序列
Figure BDA00032391949900000221
进行上行CSI估计得到，估计方法选自LS估计、MMSE估计、ML估计、MAP估计、导频辅助估计中的一种或多种。In some specific embodiments, the uplink CSI estimation vector
Figure BDA00032391949900000220
By receiving the sequence at the base station side for channel estimation
Figure BDA00032391949900000221
The uplink CSI estimation is obtained, and the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation, and pilot-assisted estimation.

优选的，所述估计方法选自LS估计法，且所述上行CSI的估计矢量

Figure BDA00032391949900000222
满足：Preferably, the estimation method is selected from the LS estimation method, and the estimation vector of the uplink CSI
Figure BDA00032391949900000222
Satisfy:

Figure BDA00032391949900000223
Figure BDA00032391949900000223

其中，s表示用户端发送的基站已知信号序列，

Figure BDA00032391949900000224
表示取Moore-Penrose伪逆操作，
Figure BDA00032391949900000225
表示用于进行信道估计的基站端接收序列，且满足：Among them, s represents the known signal sequence of the base station sent by the user terminal,
Figure BDA00032391949900000224
Represents the Moore-Penrose pseudo-inverse operation,
Figure BDA00032391949900000225
represents the base station-side reception sequence used for channel estimation, and satisfies:

Figure BDA0003239194990000031
Figure BDA0003239194990000031

其中，N表示信道噪声，g表示实际的上行CSI矢量，即上行CSI的估计矢量

Figure BDA0003239194990000032
是矢量g的估计值。Among them, N represents the channel noise, and g represents the actual uplink CSI vector, that is, the estimated vector of uplink CSI
Figure BDA0003239194990000032
is an estimate of the vector g.

在一些优选实施方式中，所述第一神经网络包括：In some preferred embodiments, the first neural network includes:

一个含有线性激活函数的输入层、一个含有LeakyReLU激活函数的隐藏层和一个含有线性激活函数的输出层；其中，输入层、隐藏层和输出层节点数分别为N、mN和N，m表示根据工程预设确定的隐藏层节点系数。An input layer with a linear activation function, a hidden layer with a LeakyReLU activation function, and an output layer with a linear activation function; among them, the number of nodes in the input layer, hidden layer, and output layer are N, mN, and N, respectively, where m represents according to The hidden layer node coefficient determined by the engineering preset.

在具体应用中，该第一神经网络的输入为所述上行CSI估计矢量幅度

Figure BDA0003239194990000033
输出为长度为N的所述下行CSI的学习幅度
Figure BDA0003239194990000034
In a specific application, the input of the first neural network is the estimated vector magnitude of the uplink CSI
Figure BDA0003239194990000033
The output is the learning amplitude of the downlink CSI of length N
Figure BDA0003239194990000034

更优选的，该第一神经网络的训练损失函数采用均方误差损失函数。More preferably, the training loss function of the first neural network adopts a mean square error loss function.

在具体应用中，所述专家知识为现有的基于压缩感知的重构方法。In a specific application, the expert knowledge is an existing reconstruction method based on compressed sensing.

在一些具体的实施方式中，所述的现有的基于压缩感知的重构方法包括如基于L1范数最小化、基追踪算法、匹配追踪算法、正交匹配追踪算法、BIHT算法、SCA-BIHT算法等。In some specific embodiments, the existing compressive sensing-based reconstruction methods include, for example, based on L1 norm minimization, basis pursuit algorithm, matching pursuit algorithm, orthogonal matching pursuit algorithm, BIHT algorithm, SCA-BIHT algorithm etc.

在一些具体实施方式中，所述恢复反馈向量

Figure BDA0003239194990000035
为在基站重新恢复的1bit压缩叠加矢量x中的反馈向量，且：In some embodiments, the restored feedback vector
Figure BDA0003239194990000035
is the feedback vector in the 1-bit compressed superposition vector x restored by the base station, and:

所述的1-bit压缩叠加矢量

Figure BDA0003239194990000036
在发射端通过以下模型获得：The 1-bit compressed overlay vector
Figure BDA0003239194990000036
Obtained at the transmitter by the following model:

Figure BDA0003239194990000037
Figure BDA0003239194990000037

其中，d表示长度为P的上行用户数据序列，E表示用户发送功率，ρ∈[0,1]表示叠加因子，可根据工程经验设定，Q表示P×L维的为扩频矩阵，满足QTQ＝P·IL，其中上标“T”表示转置运算，L表示调制信号长度，IL表示L维单位矩阵，r表示长度为L的调制信号序列，且：Among them, d represents the uplink user data sequence of length P, E represents the transmission power of the user, ρ∈[0,1] represents the superposition factor, which can be set according to engineering experience, and Q represents the P×L-dimensional spread spectrum matrix, which satisfies Q T Q=P IL , where the superscript "T" represents the transpose operation, L represents the length of the modulated signal, IL represents the L-dimensional identity matrix, r represents the modulated signal sequence of length L, and:

r＝fmode(w)；r= fmode (w);

其中，w表示长度为K的发射端的反馈向量，所述恢复反馈向量

Figure BDA0003239194990000038
为反馈向量w在基站的估计值，K＝2L，且：Among them, w represents the feedback vector of the transmitting end of length K, and the recovery feedback vector
Figure BDA0003239194990000038
is the estimated value of the feedback vector w at the base station, K=2L, and:

w＝[preal,pimag,z]；w=[p real , p imag , z];

其中，z∈{0,1}1×N表示下行CSI矢量h的长度为N的支撑集向量，preal与pimag分别表示下行CSI矢量长度为M的采用1-bit压缩感知技术压缩量化后的实部与虚部。Among them, z∈{0,1} 1×N represents the downlink CSI vector h with a length of N support set vector, p real and p imag respectively represent the downlink CSI vector with a length of M after compression and quantization using 1-bit compressed sensing technology The real and imaginary parts of .

优选的，preal与pimag满足：Preferably, p real and p imag satisfy:

Figure BDA0003239194990000041
Figure BDA0003239194990000041

其中，h表示下行CSI矢量，所述下行CSI重构矢量

Figure BDA0003239194990000042
表示在基站重构得到的下行CSI，Φ表示采用1-bit压缩感知技术的压缩矩阵，sign(·)、Re(·)和Im(·)分别表示取符号操作、取实部操作和取虚部操作。Among them, h represents the downlink CSI vector, the downlink CSI reconstruction vector
Figure BDA0003239194990000042
Represents the downlink CSI reconstructed at the base station, Φ represents the compression matrix using 1-bit compressed sensing technology, sign( ), Re( ) and Im( ) represent the sign operation, the real part operation and the imaginary operation, respectively Department operation.

在一些优选实施方式中，基于上述应用中，所述恢复反馈向量

Figure BDA0003239194990000043
的获得进一步包括：对接收序列Y进行解扩处理，再通过MMSE检测以及干扰消除得到所述恢复反馈向量
Figure BDA0003239194990000044
In some preferred embodiments, based on the above application, the restored feedback vector
Figure BDA0003239194990000043
The acquisition further includes: despreading the received sequence Y, and then obtaining the recovery feedback vector through MMSE detection and interference cancellation
Figure BDA0003239194990000044

如，其包括：For example, it includes:

通过以下过信道模型获得所述接收序列

Figure BDA0003239194990000045
The received sequence is obtained by the following over-channel model
Figure BDA0003239194990000045

Y＝gx+N；Y=gx+N;

其中，Y表示基站端的接收序列，x表示用户端发送的1-bit压缩叠加矢量，N表示信道噪声，g表示上行信道矢量。Among them, Y represents the sequence received by the base station, x represents the 1-bit compressed superposition vector sent by the user terminal, N represents the channel noise, and g represents the uplink channel vector.

通过以下解扩处理模型获得解扩信号

Figure BDA0003239194990000046
The despread signal is obtained by the following despreading processing model
Figure BDA0003239194990000046

Figure BDA0003239194990000047
Figure BDA0003239194990000047

通过以下MMSE检测模型获得检测信号

Figure BDA0003239194990000048
The detection signal is obtained by the following MMSE detection model
Figure BDA0003239194990000048

Figure BDA0003239194990000049
Figure BDA0003239194990000049

其中，dec(·)表示硬判决操作，(·)-1表示取矩阵的逆操作，(·)H表示矩阵的共轭转置操作，

Figure BDA00032391949900000410
表示上行信道的方差；Among them, dec( ) represents the hard decision operation, ( ) -1 represents the inverse operation of the matrix, ( ) H represents the conjugate transpose operation of the matrix,
Figure BDA00032391949900000410
represents the variance of the uplink channel;

通过以下干扰消除模型获得去干扰数据序列

Figure BDA00032391949900000411
De-interference data sequences are obtained by the following interference cancellation model
Figure BDA00032391949900000411

Figure BDA00032391949900000412
Figure BDA00032391949900000412

通过以下反馈向量恢复模型，获得恢复反馈向量

Figure BDA00032391949900000413
Through the following feedback vector restoration model, the restored feedback vector is obtained
Figure BDA00032391949900000413

Figure BDA00032391949900000414
Figure BDA00032391949900000414

其中，fdemo(·)表示解调处理，则可得到恢复反馈向量

Figure BDA00032391949900000415
为：Among them, f demo ( ) represents the demodulation process, then the recovery feedback vector can be obtained
Figure BDA00032391949900000415
for:

Figure BDA00032391949900000416
Figure BDA00032391949900000416

其中，

Figure BDA0003239194990000051
表示恢复出的下行CSI矢量h的长度为N的支撑集，
Figure BDA0003239194990000052
与
Figure BDA0003239194990000053
分别表示恢复出的采用1-bit压缩感知技术压缩量化的下行CSI矢量h的长度均为M的实部与虚部。in,
Figure BDA0003239194990000051
represents a support set of length N of the recovered downlink CSI vector h,
Figure BDA0003239194990000052
and
Figure BDA0003239194990000053
respectively indicate that the length of the recovered downlink CSI vector h compressed and quantized using the 1-bit compressed sensing technology is both the real part and the imaginary part of M.

在一些优选实施方式中，根据所得基站端恢复反馈向量

Figure BDA0003239194990000054
进行下行CSI的特征幅度
Figure BDA0003239194990000055
与特征角度
Figure BDA0003239194990000056
的恢复的过程包括：In some preferred embodiments, the feedback vector is recovered according to the obtained base station
Figure BDA0003239194990000054
Feature magnitude for downlink CSI
Figure BDA0003239194990000055
with characteristic angle
Figure BDA0003239194990000056
The recovery process includes:

将所得参数

Figure BDA0003239194990000057
作为SCA-BIHT算法的输入，其中，SCA-BIHT算法即为专家知识；the resulting parameters
Figure BDA0003239194990000057
As the input of the SCA-BIHT algorithm, the SCA-BIHT algorithm is the expert knowledge;

通过SCA-BIHT算法循环n次后输出下行CSI的特征值

Figure BDA0003239194990000058
其中，n可根据工程经验预先设定；Output the eigenvalues of downlink CSI after looping through the SCA-BIHT algorithm for n times
Figure BDA0003239194990000058
Among them, n can be preset according to engineering experience;

根据下行CSI的特征值

Figure BDA0003239194990000059
通过下式获得下行CSI特征幅度
Figure BDA00032391949900000510
与特征角度
Figure BDA00032391949900000511
According to the eigenvalues of downlink CSI
Figure BDA0003239194990000059
The downlink CSI feature magnitude is obtained by the following formula
Figure BDA00032391949900000510
with characteristic angle
Figure BDA00032391949900000511

Figure BDA00032391949900000512
Figure BDA00032391949900000512

其中，famp(·)表示对复数进行取幅度操作，fang(·)表示对复数进行取角度操作。Among them, f amp (·) represents the operation of taking the amplitude of the complex number, and f ang (·) represents the operation of taking the angle of the complex number.

在一些优选实施方式中，所述第二神经网络包括：In some preferred embodiments, the second neural network includes:

一个含有线性激活函数的输入层、一个含有LeakyReLU激活函数的隐藏层和一个含有线性激活函数的输出层；其中，输入层、隐藏层和输出层节点数分别为2N、kN和N，k表示根据工程预设确定的隐藏层节点系数。An input layer with a linear activation function, a hidden layer with a LeakyReLU activation function, and an output layer with a linear activation function; where the number of nodes in the input layer, hidden layer, and output layer are 2N, kN, and N, respectively, where k represents the The hidden layer node coefficient determined by the engineering preset.

在具体应用中，该第二神经网络的输入为所述下行CSI拼接幅度

Figure BDA00032391949900000513
输出为长度为N的所述下行CSI的融合幅度
Figure BDA00032391949900000514
In a specific application, the input of the second neural network is the downlink CSI splicing amplitude
Figure BDA00032391949900000513
The output is the fusion amplitude of the downlink CSI with length N
Figure BDA00032391949900000514

其中，所述下行CSI拼接幅度

Figure BDA00032391949900000515
通过以下模型获得：Wherein, the downlink CSI splicing range is
Figure BDA00032391949900000515
Obtained by the following models:

Figure BDA00032391949900000516
Figure BDA00032391949900000516

其中，[·]表示矢量的拼接操作。Among them, [ ] represents the concatenation operation of the vector.

更优选的，该第二神经网络的训练损失函数采用均方误差损失函数。More preferably, the training loss function of the second neural network adopts a mean square error loss function.

在一些优选实施方式中，所述下行CSI重构矢量

Figure BDA00032391949900000517
的恢复通过以下模型获得：In some preferred embodiments, the downlink CSI reconstruction vector
Figure BDA00032391949900000517
The recovery of is obtained by the following model:

Figure BDA0003239194990000061
Figure BDA0003239194990000061

其中，⊙表示Hadamard乘积，e表示自然指数，j表示虚数单位。Among them, ⊙ represents the Hadamard product, e represents the natural exponent, and j represents the imaginary unit.

本发明利用上下行信道的双向互异性，通过浅层幅度学习网络恢复下行CSI的幅度，并结合单比特CS、SC和DL技术优点，将经过单比特CS处理的下行CSI扩频叠加到上行用户序列上反馈回基站，在基站端利用传统UL-US检测与简化版本的传统下行CSI重构方法恢复下行CSI，进而再通过浅层幅度融合网络对传统方法与幅度学习网络得到的下行CSI的幅度进行融合，提高了下行CSI的幅度的精度，在不增加频谱开销的情况下，可有效降低叠加编码造成的相互干扰以保证CSI的重构精度。The invention utilizes the bidirectional mutuality of the uplink and downlink channels, recovers the amplitude of the downlink CSI through a shallow amplitude learning network, and combines the technical advantages of single-bit CS, SC and DL to spread and superimpose the downlink CSI processed by the single-bit CS to the uplink user. The sequence is fed back to the base station, and the traditional UL-US detection and simplified version of the traditional downlink CSI reconstruction method is used at the base station to restore the downlink CSI, and then the amplitude of the downlink CSI obtained by the traditional method and the amplitude learning network is analyzed by the shallow amplitude fusion network. The fusion improves the accuracy of the amplitude of the downlink CSI, and can effectively reduce the mutual interference caused by superposition coding without increasing the spectrum overhead, so as to ensure the reconstruction accuracy of the CSI.

与单比特CS叠加CSI反馈相比，本发明根据上下行信道双向互异性恢复出单比特CS丢失的下行CSI幅度，极大的提高了CSI的重构精度，并且利用浅层神经网络与简化版本的传统下行CSI重构方法，提高了CSI的重构效率。Compared with the single-bit CS superimposed CSI feedback, the present invention recovers the downlink CSI amplitude lost by the single-bit CS according to the bidirectional reciprocity of the uplink and downlink channels, which greatly improves the reconstruction accuracy of the CSI, and uses the shallow neural network and simplified version. The traditional downlink CSI reconstruction method has improved the reconstruction efficiency of CSI.

附图说明Description of drawings

图1为本发明的总体流程示意图；Fig. 1 is the overall flow schematic diagram of the present invention;

图2为本发明的第一神经网络结构示意图；Fig. 2 is the first neural network structure schematic diagram of the present invention;

图3为本发明的第二神经网络结构示意图。FIG. 3 is a schematic structural diagram of the second neural network of the present invention.

具体实施方式Detailed ways

以下结合实施例和附图对本发明进行详细描述，但需要理解的是，所述实施例和附图仅用于对本发明进行示例性的描述，而并不能对本发明的保护范围构成任何限制。所有包含在本发明的发明宗旨范围内的合理的变换和组合均落入本发明的保护范围。The present invention will be described in detail below with reference to the embodiments and drawings, but it should be understood that the embodiments and drawings are only used to describe the present invention by way of example, but do not limit the protection scope of the present invention. All reasonable transformations and combinations included within the scope of the inventive concept of the present invention fall into the protection scope of the present invention.

参照图1，一种具体的基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法包括：Referring to FIG. 1, a specific 1-bit compression and superposition CSI feedback method based on feature extraction and mutual fusion includes:

a1)根据上行CSI估计矢量

Figure BDA0003239194990000062
的幅度
Figure BDA0003239194990000063
通过第一神经网络获得对应的下行CSI的学习幅度
Figure BDA0003239194990000071
a1) Estimating the vector according to the uplink CSI
Figure BDA0003239194990000062
Amplitude
Figure BDA0003239194990000063
Obtain the learning amplitude of the corresponding downlink CSI through the first neural network
Figure BDA0003239194990000071

其中，更具体的一些实施方式如：Among them, some more specific implementations are as follows:

所述上行CSI估计矢量

Figure BDA0003239194990000072
通过对用于进行信道估计的基站端接收序列
Figure BDA0003239194990000073
进行上行CSI估计得到。the uplink CSI estimation vector
Figure BDA0003239194990000072
By receiving the sequence at the base station side for channel estimation
Figure BDA0003239194990000073
Obtained by performing uplink CSI estimation.

该上行CSI的估计可进一步通过现有技术中的上行CSI估计方法如LS估计、MMSE估计、ML估计、MAP估计、导频辅助估计等实现，如在一种具体实施例中，以LS估计进行上行CSI估计如下：The estimation of the uplink CSI can be further realized by the uplink CSI estimation methods in the prior art, such as LS estimation, MMSE estimation, ML estimation, MAP estimation, pilot-assisted estimation, etc. For example, in a specific embodiment, LS estimation is used to perform estimation. The uplink CSI is estimated as follows:

Figure BDA0003239194990000074
Figure BDA0003239194990000074

其中，

Figure BDA0003239194990000075
表示基站端的接收估计序列，s表示用户端发送的基站已知信号，
Figure BDA0003239194990000076
表示取Moore-Penrose伪逆操作。in,
Figure BDA0003239194990000075
represents the received estimation sequence of the base station, s represents the known signal of the base station sent by the user,
Figure BDA0003239194990000076
Represents the Moore-Penrose pseudo-inverse operation.

参照图2，所述第一神经网络包括以下的神经网络结构：2, the first neural network includes the following neural network structure:

一个输入层、一个隐藏层和一个输出层，其中输入层采用线性激活函数，隐藏层采用激活函数LeakyReLU，输出层采用线性激活函数。An input layer, a hidden layer and an output layer, where the input layer adopts a linear activation function, the hidden layer adopts the activation function LeakyReLU, and the output layer adopts a linear activation function.

该第一神经网络的输入层、隐藏层和输出层节点数分别为N、mN和N，m表示隐藏层节点系数，可根据工程预设得到。The number of nodes in the input layer, hidden layer and output layer of the first neural network are N, mN and N respectively, where m represents the node coefficient of the hidden layer, which can be obtained according to engineering presets.

通过该第一神经网络获得下行CSI的学习幅度

Figure BDA0003239194990000077
的过程包括：The learning amplitude of downlink CSI is obtained through the first neural network
Figure BDA0003239194990000077
The process includes:

根据所述上行CSI估计矢量

Figure BDA0003239194990000078
通过下式获得上行CSI幅度
Figure BDA0003239194990000079
Estimate the vector according to the uplink CSI
Figure BDA0003239194990000078
The uplink CSI amplitude is obtained by the following formula
Figure BDA0003239194990000079

Figure BDA00032391949900000710
Figure BDA00032391949900000710

其中，

Figure BDA00032391949900000711
表示矢量
Figure BDA00032391949900000712
的第N个元素，|·|表示复数的取模操作；in,
Figure BDA00032391949900000711
representation vector
Figure BDA00032391949900000712
The Nth element of , |·| represents the modulo operation of complex numbers;

将所得上行CSI估计矢量幅度

Figure BDA00032391949900000713
自输入层输入该第一神经网络，输出即为长度为N的下行CSI的学习幅度
Figure BDA00032391949900000714
The obtained uplink CSI estimated vector magnitude
Figure BDA00032391949900000713
The first neural network is input from the input layer, and the output is the learning range of the downlink CSI of length N
Figure BDA00032391949900000714

所述第一神经网络的训练损失函数采用均方误差损失函数。The training loss function of the first neural network adopts the mean square error loss function.

a2)根据基站得到的恢复反馈向量

Figure BDA00032391949900000715
利用专家知识进行CSI特征提取，恢复出下行CSI的特征幅度
Figure BDA00032391949900000716
与特征角度
Figure BDA00032391949900000717
a2) According to the recovery feedback vector obtained by the base station
Figure BDA00032391949900000715
Use expert knowledge to extract CSI features and recover the feature magnitude of downlink CSI
Figure BDA00032391949900000716
with characteristic angle
Figure BDA00032391949900000717

a3)根据对所述下行CSI的特征幅度

Figure BDA00032391949900000718
与下行CSI的学习幅度
Figure BDA00032391949900000719
拼接得到的下行CSI拼接幅度
Figure BDA00032391949900000720
通过第二神经网络获得下行CSI的融合幅度
Figure BDA00032391949900000721
a3) According to the characteristic magnitude of the downlink CSI
Figure BDA00032391949900000718
Learning magnitude with downlink CSI
Figure BDA00032391949900000719
Downlink CSI splicing amplitude obtained by splicing
Figure BDA00032391949900000720
Obtaining the fusion magnitude of the downlink CSI through the second neural network
Figure BDA00032391949900000721

其中，更具体的一些实施方式如：Among them, some more specific implementations are as follows:

参照图3，所述第二神经网络为包括以下的神经网络结构：3, the second neural network is a neural network structure including the following:

一个输入层、一个隐藏层和一个输出层，其中，输入层采用线性激活函数，隐藏层采用LeakyReLU激活函数，输出层采用线性激活函数。An input layer, a hidden layer and an output layer, where the input layer adopts a linear activation function, the hidden layer adopts the LeakyReLU activation function, and the output layer adopts a linear activation function.

该第二神经网络的输入层、隐藏层和输出层节点数分别为2N、kN和N，k表示隐藏层节点系数，可根据工程预设得到。The number of nodes in the input layer, hidden layer and output layer of the second neural network are 2N, kN and N respectively, where k represents the node coefficient of the hidden layer, which can be obtained according to engineering presets.

通过该第二神经网络获得下行CSI的融合幅度

Figure BDA0003239194990000081
的过程包括：Obtain the fusion magnitude of downlink CSI through the second neural network
Figure BDA0003239194990000081
The process includes:

通过所述下行CSI的特征幅度

Figure BDA0003239194990000082
与所述下行CSI的学习幅度
Figure BDA0003239194990000083
获得下行CSI拼接幅度
Figure BDA0003239194990000084
through the characteristic magnitude of the downlink CSI
Figure BDA0003239194990000082
with the learning amplitude of the downlink CSI
Figure BDA0003239194990000083
Get downlink CSI splicing amplitude
Figure BDA0003239194990000084

Figure BDA0003239194990000085
Figure BDA0003239194990000085

其中，[·]表示矢量的拼接操作，

Figure BDA0003239194990000086
表示维度为1×2N的实数集；Among them, [ ] represents the splicing operation of the vector,
Figure BDA0003239194990000086
Represents a set of real numbers of dimension 1×2N;

将所得下行CSI拼接幅度

Figure BDA0003239194990000087
自输入层输入该第二神经网络，输出即为长度为N的下行CSI的融合幅度
Figure BDA0003239194990000088
Splicing the obtained downlink CSI amplitude
Figure BDA0003239194990000087
The second neural network is input from the input layer, and the output is the fusion amplitude of the downlink CSI of length N
Figure BDA0003239194990000088

所述幅度融合网络的训练损失函数采用均方误差损失函数。The training loss function of the amplitude fusion network adopts the mean square error loss function.

a4)根据所述下行CSI的融合幅度

Figure BDA0003239194990000089
与所述特征角度
Figure BDA00032391949900000810
恢复得到下行CSI重构矢量
Figure BDA00032391949900000811
a4) According to the fusion amplitude of the downlink CSI
Figure BDA0003239194990000089
with the characteristic angle
Figure BDA00032391949900000810
Recovery to obtain downlink CSI reconstruction vector
Figure BDA00032391949900000811

其中，更具体的一些实施方式如：Among them, some more specific implementations are as follows:

所述恢复得到下行CSI重构矢量

Figure BDA00032391949900000812
的方式如下：The recovery obtains the downlink CSI reconstruction vector
Figure BDA00032391949900000812
The way is as follows:

Figure BDA00032391949900000813
Figure BDA00032391949900000813

其中，⊙表示Hadamard乘积，e表示自然指数，j表示虚数单位。Among them, ⊙ represents the Hadamard product, e represents the natural exponent, and j represents the imaginary unit.

实施例1Example 1

步骤a1)中，通过LS估计得到上行CSI估计矢量

Figure BDA00032391949900000814
的一个具体实施例如下：In step a1), the uplink CSI estimation vector is obtained through LS estimation
Figure BDA00032391949900000814
A specific example is as follows:

假设：N＝2，P＝4，用于进行信道估计的基站端接收序列

Figure BDA00032391949900000815
为：Suppose: N=2, P=4, the base station receiving sequence used for channel estimation
Figure BDA00032391949900000815
for:

Figure BDA00032391949900000816
Figure BDA00032391949900000816

用户端发送的基站已知信号s为：The known signal s of the base station sent by the user terminal is:

s＝[0.7528-0.6083i -0.1666-0.1308i 0.9869+0.4514i 0.4556+0.2695i]；s=[0.7528-0.6083i-0.1666-0.1308i 0.9869+0.4514i 0.4556+0.2695i];

用户端发送的基站已知信号s的伪逆矩阵

Figure BDA0003239194990000091
为：The pseudo-inverse matrix of the known signal s of the base station sent by the user terminal
Figure BDA0003239194990000091
for:

Figure BDA0003239194990000092
Figure BDA0003239194990000092

按照LS估计处理公式

Figure BDA0003239194990000093
可计算出上行CSI估计矢量
Figure BDA0003239194990000094
为：According to the LS estimation processing formula
Figure BDA0003239194990000093
The uplink CSI estimation vector can be calculated
Figure BDA0003239194990000094
for:

Figure BDA0003239194990000095
Figure BDA0003239194990000095

实施例2Example 2

步骤a2)中，通过恢复出的采用1-bit压缩感知技术压缩量化的下行CSI矢量h的长度均为M的实部

Figure BDA0003239194990000096
与虚部
Figure BDA0003239194990000097
和恢复反馈向量
Figure BDA0003239194990000098
获得恢复出的下行CSI矢量h的长度为N的支撑集
Figure BDA0003239194990000099
的一个具体实施例如下：In step a2), the length of the recovered downlink CSI vector h compressed and quantized using the 1-bit compressed sensing technology is the real part of M.
Figure BDA0003239194990000096
with the imaginary part
Figure BDA0003239194990000097
and recovery feedback vector
Figure BDA0003239194990000098
Obtain a support set with length N of the recovered downlink CSI vector h
Figure BDA0003239194990000099
A specific example is as follows:

假设：N＝2，M＝3，恢复反馈向量

Figure BDA00032391949900000910
为：Assumption: N=2, M=3, restore the feedback vector
Figure BDA00032391949900000910
for:

Figure BDA00032391949900000911
Figure BDA00032391949900000911

根据公式

Figure BDA00032391949900000912
可得到恢复出的采用1-bit压缩感知技术压缩量化的下行CSI矢量h的长度均为M的实部
Figure BDA00032391949900000913
为：According to the formula
Figure BDA00032391949900000912
It can be obtained that the length of the recovered downlink CSI vector h compressed and quantized using 1-bit compressed sensing technology is the real part of M
Figure BDA00032391949900000913
for:

Figure BDA00032391949900000914
Figure BDA00032391949900000914

恢复出的采用1-bit压缩感知技术压缩量化的下行CSI矢量h的长度均为M的虚部

Figure BDA00032391949900000915
为：The length of the recovered downlink CSI vector h compressed and quantized using 1-bit compressed sensing technology is the imaginary part of M
Figure BDA00032391949900000915
for:

Figure BDA00032391949900000916
Figure BDA00032391949900000916

恢复出的下行CSI矢量h的长度为N的支撑集

Figure BDA00032391949900000917
为：The recovered downlink CSI vector h has a support set of length N
Figure BDA00032391949900000917
for:

Figure BDA00032391949900000918
Figure BDA00032391949900000918

实施例3Example 3

步骤a1)中，通过上行CSI估计矢量

Figure BDA00032391949900000919
获得上行CSI估计矢量幅度
Figure BDA00032391949900000920
的一个具体实施例如下：In step a1), the vector is estimated by the uplink CSI
Figure BDA00032391949900000919
Get the estimated vector magnitude of the uplink CSI
Figure BDA00032391949900000920
A specific example is as follows:

将实施例1所得CSI估计矢量

Figure BDA00032391949900000921
按照公式
Figure BDA00032391949900000922
计算出幅度学习网络中的输入
Figure BDA00032391949900000923
为：The CSI estimation vector obtained in Example 1
Figure BDA00032391949900000921
According to the formula
Figure BDA00032391949900000922
Calculate the input in the magnitude learning network
Figure BDA00032391949900000923
for:

Figure BDA0003239194990000101
Figure BDA0003239194990000101

实施例4Example 4

步骤a3)中，获得下行CSI拼接幅度

Figure BDA0003239194990000102
的一个具体实施例如下：In step a3), obtain the downlink CSI splicing amplitude
Figure BDA0003239194990000102
A specific example is as follows:

假设：N＝2，下行CSI的特征幅度

Figure BDA0003239194990000103
下行CSI的学习幅度
Figure BDA0003239194990000104
Assumption: N=2, the characteristic magnitude of downlink CSI
Figure BDA0003239194990000103
Learning range of downlink CSI
Figure BDA0003239194990000104

根据模型

Figure BDA0003239194990000105
可得到幅度融合网络中的输入即下行CSI拼接幅度
Figure BDA0003239194990000106
为：According to the model
Figure BDA0003239194990000105
The input in the amplitude fusion network, that is, the downlink CSI splicing amplitude, can be obtained.
Figure BDA0003239194990000106
for:

Figure BDA0003239194990000107
Figure BDA0003239194990000107

实施例5Example 5

步骤a4)中，恢复得到下行CSI重构矢量

Figure BDA0003239194990000108
的一个具体实施例如下：In step a4), recover the downlink CSI reconstruction vector
Figure BDA0003239194990000108
A specific example is as follows:

假设：N＝2，下行CSI的融合幅度

Figure BDA0003239194990000109
下行CSI的特征角度
Figure BDA00032391949900001010
Assumption: N=2, the fusion range of downlink CSI
Figure BDA0003239194990000109
The characteristic angle of downlink CSI
Figure BDA00032391949900001010

根据模型

Figure BDA00032391949900001011
可以计算出恢复得到的下行CSI重构矢量
Figure BDA00032391949900001012
为：According to the model
Figure BDA00032391949900001011
The recovered downlink CSI reconstruction vector can be calculated
Figure BDA00032391949900001012
for:

Figure BDA00032391949900001013
Figure BDA00032391949900001013

以上实施例仅是本发明的优选实施方式，本发明的保护范围并不仅局限于上述实施例。凡属于本发明思路下的技术方案均属于本发明的保护范围。应该指出，对于本技术领域的普通技术人员来说，在不脱离本发明原理的前提下的改进和润饰，这些改进和润饰也应视为本发明的保护范围。The above embodiments are only preferred embodiments of the present invention, and the protection scope of the present invention is not limited to the above embodiments. All the technical solutions under the idea of the present invention belong to the protection scope of the present invention. It should be pointed out that for those skilled in the art, improvements and modifications without departing from the principles of the present invention should also be regarded as the protection scope of the present invention.

Claims (10)

1.基于特征抽取与互异性融合的1bit压缩叠加CSI反馈方法，其特征在于，包括：1. based on the 1bit compression and superposition CSI feedback method of feature extraction and mutual fusion, it is characterized in that, comprising: 根据上行CSI的估计矢量
Figure FDA0003239194980000011
的幅度
Figure FDA0003239194980000012
通过第一神经网络获得对应的下行CSI的学习幅度
Figure FDA0003239194980000013
Estimated vector based on uplink CSI
Figure FDA0003239194980000011
Amplitude
Figure FDA0003239194980000012
Obtain the learning amplitude of the corresponding downlink CSI through the first neural network
Figure FDA0003239194980000013
 根据恢复反馈向量
Figure FDA0003239194980000014
利用专家知识进行CSI特征提取，恢复出下行CSI的特征幅度
Figure FDA0003239194980000015
与特征角度
Figure FDA0003239194980000016
其中，所述恢复反馈向量
Figure FDA0003239194980000017
是指基站接收端根据发射端发出的反馈向量w进行恢复后得到的对应反馈向量；According to the restored feedback vector
Figure FDA0003239194980000014
Use expert knowledge to extract CSI features and recover the feature magnitude of downlink CSI
Figure FDA0003239194980000015
with characteristic angle
Figure FDA0003239194980000016
Wherein, the recovery feedback vector
Figure FDA0003239194980000017
Refers to the corresponding feedback vector obtained by the base station receiving end after recovery according to the feedback vector w sent by the transmitting end; 根据对所述下行CSI的特征幅度
Figure FDA0003239194980000018
与下行CSI的学习幅度
Figure FDA0003239194980000019
拼接得到的下行CSI拼接幅度
Figure FDA00032391949800000110
通过第二神经网络获得下行CSI的融合幅度
Figure FDA00032391949800000111
According to the characteristic magnitude of the downlink CSI
Figure FDA0003239194980000018
Learning magnitude with downlink CSI
Figure FDA0003239194980000019
Downlink CSI splicing amplitude obtained by splicing
Figure FDA00032391949800000110
Obtaining the fusion magnitude of the downlink CSI through the second neural network
Figure FDA00032391949800000111
 根据所述下行CSI的融合幅度
Figure FDA00032391949800000112
与所述特征角度
Figure FDA00032391949800000113
恢复得到下行CSI重构矢量
Figure FDA00032391949800000114
According to the fusion amplitude of the downlink CSI
Figure FDA00032391949800000112
with the characteristic angle
Figure FDA00032391949800000113
Recovery to obtain downlink CSI reconstruction vector
Figure FDA00032391949800000114
 2.根据权利要求1所述的方法，其特征在于，所述上行CSI估计矢量幅度
Figure FDA00032391949800000115
通过以下模型获得：2. The method according to claim 1, wherein the uplink CSI estimates vector magnitude
Figure FDA00032391949800000115
Obtained by the following models:
Figure FDA00032391949800000116
Figure FDA00032391949800000116
Figure FDA00032391949800000117
Figure FDA00032391949800000117
 其中，
Figure FDA00032391949800000118
表示所述上行CSI的估计矢量
Figure FDA00032391949800000119
的第N个元素，|·|表示复数的取模操作，上标T表示转置运算。in,
Figure FDA00032391949800000118
the estimated vector representing the uplink CSI
Figure FDA00032391949800000119
The Nth element of , |·| represents the modulo operation of complex numbers, and the superscript T represents the transpose operation. 3.根据权利要求2所述的方法，其特征在于，所述上行CSI的估计矢量
Figure FDA00032391949800000120
通过以下模型得到：3. The method according to claim 2, wherein the estimated vector of the uplink CSI
Figure FDA00032391949800000120
Obtained by the following model:
Figure FDA00032391949800000121
Figure FDA00032391949800000121
 其中，s表示用户端发送的基站已知信号序列，
Figure FDA00032391949800000122
表示取Moore-Penrose伪逆操作，
Figure FDA00032391949800000123
表示用于进行信道估计的基站端接收序列，且满足：Among them, s represents the known signal sequence of the base station sent by the user terminal,
Figure FDA00032391949800000122
Represents the Moore-Penrose pseudo-inverse operation,
Figure FDA00032391949800000123
represents the base station-side reception sequence used for channel estimation, and satisfies:
Figure FDA00032391949800000124
Figure FDA00032391949800000124
 其中，N表示信道噪声，g表示实际的上行CSI矢量。Among them, N represents the channel noise, and g represents the actual uplink CSI vector. 4.根据权利要求2所述的方法，其特征在于，所述第一神经网络包括：4. The method of claim 2, wherein the first neural network comprises: 一个含有线性激活函数的输入层、一个含有Leaky ReLU激活函数的隐藏层和一个含有线性激活函数的输出层；其中，输入层、隐藏层和输出层节点数分别为N、mN和N，m表示根据工程预设确定的隐藏层节点系数。An input layer with a linear activation function, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function; the number of nodes in the input layer, hidden layer, and output layer are N, mN, and N, respectively, where m represents The hidden layer node coefficients are determined according to engineering presets. 5.根据权利要求1所述的方法，其特征在于，所述恢复反馈向量
Figure FDA0003239194980000021
为在基站恢复的1bit压缩叠加矢量x中的反馈向量，且：5. The method of claim 1, wherein the recovery feedback vector is
Figure FDA0003239194980000021
is the feedback vector in the 1-bit compressed superposition vector x recovered by the base station, and: 所述的1-bit压缩叠加矢量
Figure FDA0003239194980000022
在发射端通过以下模型获得：The 1-bit compressed overlay vector
Figure FDA0003239194980000022
Obtained at the transmitter by the following model:
Figure FDA0003239194980000023
Figure FDA0003239194980000023
 其中，d表示长度为P的上行用户数据序列，E表示用户发送功率，ρ∈[0,1]表示叠加因子，可根据工程经验设定，Q表示P×L维的为扩频矩阵，满足QTQ＝P·IL，其中上标T表示转置运算，L表示调制信号长度，IL表示L维单位矩阵，r表示长度为L的调制信号序列。Among them, d represents the uplink user data sequence of length P, E represents the user transmit power, ρ∈[0,1] represents the superposition factor, which can be set according to engineering experience, and Q represents the P×L-dimensional spread spectrum matrix, which satisfies Q T Q=P· IL , where the superscript T represents the transposition operation, L represents the length of the modulated signal, IL represents the L-dimensional unit matrix, and r represents the modulated signal sequence of length L. 6.根据权利要求5所述的方法，其特征在于，获得所述恢复反馈向量
Figure FDA0003239194980000024
的过程包括：6. The method of claim 5, wherein the recovery feedback vector is obtained
Figure FDA0003239194980000024
The process includes: 通过以下解扩处理模型获得解扩信号
Figure FDA0003239194980000025
The despread signal is obtained by the following despreading processing model
Figure FDA0003239194980000025
Figure FDA0003239194980000026
Figure FDA0003239194980000026
 其中，Y表示基站端的N×P维的接收序列；Among them, Y represents the N×P-dimensional receiving sequence of the base station; 通过以下MMSE检测模型获得检测信号
Figure FDA0003239194980000027
The detection signal is obtained by the following MMSE detection model
Figure FDA0003239194980000027
Figure FDA0003239194980000028
Figure FDA0003239194980000028
 其中，dec(·)表示硬判决操作，g表示上行信道矢量，(·)-1表示取矩阵的逆操作，(·)H表示矩阵的共轭转置操作，
Figure FDA0003239194980000029
表示上行信道的方差；Among them, dec( ) represents the hard decision operation, g represents the upstream channel vector, ( ) -1 represents the inverse operation of the matrix, ( ) H represents the conjugate transpose operation of the matrix,
Figure FDA0003239194980000029
represents the variance of the uplink channel; 通过以下干扰消除模型获得去干扰数据序列
Figure FDA00032391949800000210
De-interference data sequences are obtained by the following interference cancellation model
Figure FDA00032391949800000210
Figure FDA00032391949800000211
Figure FDA00032391949800000211
 通过以下反馈向量恢复模型，获得恢复反馈向量
Figure FDA00032391949800000212
Through the following feedback vector restoration model, the restored feedback vector is obtained
Figure FDA00032391949800000212
Figure FDA00032391949800000213
Figure FDA00032391949800000213
 其中，fdemo(·)表示解调处理，则可得到恢复反馈向量
Figure FDA00032391949800000214
为：Among them, f demo ( ) represents the demodulation process, then the recovery feedback vector can be obtained
Figure FDA00032391949800000214
for:
Figure FDA00032391949800000215
Figure FDA00032391949800000215
 其中，
Figure FDA0003239194980000031
表示恢复出的下行CSI矢量h的长度为N的支撑集，
Figure FDA0003239194980000032
与
Figure FDA0003239194980000033
分别表示恢复出的采用1-bit压缩感知技术压缩量化的下行CSI矢量h的长度均为M的实部与虚部。in,
Figure FDA0003239194980000031
represents a support set of length N of the recovered downlink CSI vector h,
Figure FDA0003239194980000032
and
Figure FDA0003239194980000033
respectively indicate that the length of the recovered downlink CSI vector h compressed and quantized using the 1-bit compressed sensing technology is both the real part and the imaginary part of M. 7.根据权利要求6所述的方法，其特征在于，所述下行CSI的特征幅度
Figure FDA0003239194980000034
与特征角度
Figure FDA0003239194980000035
的恢复的过程包括：7. The method according to claim 6, wherein the characteristic magnitude of the downlink CSI is
Figure FDA0003239194980000034
with characteristic angle
Figure FDA0003239194980000035
The recovery process includes: 将所得参数
Figure FDA0003239194980000036
作为SCA-BIHT算法的输入，其中，SCA-BIHT算法即为专家知识；the resulting parameters
Figure FDA0003239194980000036
As the input of the SCA-BIHT algorithm, the SCA-BIHT algorithm is the expert knowledge; 通过SCA-BIHT算法循环n次后输出下行CSI的特征值
Figure FDA0003239194980000037
其中，n可根据工程经验预先设定；The eigenvalues of downlink CSI are output after looping through the SCA-BIHT algorithm for n times
Figure FDA0003239194980000037
Among them, n can be preset according to engineering experience; 根据下行CSI的特征值
Figure FDA0003239194980000038
通过下式获得下行CSI特征幅度
Figure FDA0003239194980000039
与特征角度
Figure FDA00032391949800000310
According to the eigenvalues of downlink CSI
Figure FDA0003239194980000038
The downlink CSI feature magnitude is obtained by the following formula
Figure FDA0003239194980000039
with characteristic angle
Figure FDA00032391949800000310
Figure FDA00032391949800000311
Figure FDA00032391949800000311
 其中，famp(·)表示对复数进行取幅度操作，fang(·)表示对复数进行取角度操作。Among them, f amp (·) represents the operation of taking the magnitude of the complex number, and f ang (·) represents the operation of taking the angle of the complex number. 8.根据权利要求2所述的方法，其特征在于，所述第二神经网络包括：8. The method of claim 2, wherein the second neural network comprises: 一个含有线性激活函数的输入层、一个含有Leaky ReLU激活函数的隐藏层和一个含有线性激活函数的输出层；其中，输入层、隐藏层和输出层节点数分别为2N、kN和N，k表示根据工程预设确定的隐藏层节点系数。An input layer with a linear activation function, a hidden layer with a Leaky ReLU activation function, and an output layer with a linear activation function; the number of nodes in the input layer, hidden layer and output layer are 2N, kN and N, respectively, where k represents The hidden layer node coefficients are determined according to engineering presets. 9.根据权利要求1所述的方法，其特征在于，所述神经网络通过均方误差损失函数进行训练。9. The method of claim 1, wherein the neural network is trained by a mean squared error loss function. 10.根据权利要求1所述的方法，其特征在于，所述下行CSI重构矢量
Figure FDA00032391949800000312
的恢复通过以下模型获得：10. The method according to claim 1, wherein the downlink CSI reconstruction vector
Figure FDA00032391949800000312
The recovery of is obtained by the following model:
Figure FDA00032391949800000313
Figure FDA00032391949800000313
 其中，⊙表示Hadamard乘积，e表示自然指数，j表示虚数单位。Among them, ⊙ represents the Hadamard product, e represents the natural exponent, and j represents the imaginary unit.
CN202111011688.XA 2021-08-31 2021-08-31 1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion Active CN113726376B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111011688.XA CN113726376B (en) 2021-08-31 2021-08-31 1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111011688.XA CN113726376B (en) 2021-08-31 2021-08-31 1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion

Publications (2)

Publication Number Publication Date
CN113726376A true CN113726376A (en) 2021-11-30
CN113726376B CN113726376B (en) 2022-05-20

Family

ID=78679649

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111011688.XA Active CN113726376B (en) 2021-08-31 2021-08-31 1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion

Country Status (1)

Country Link
CN (1) CN113726376B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114826343A (en) * 2022-04-26 2022-07-29 西华大学 Superimposed channel state information feedback method and device for AI enabling data emptying
CN117807383A (en) * 2024-03-01 2024-04-02 深圳市大数据研究院 Channel state information recovery method and device, equipment and storage medium

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108737032A (en) * 2018-05-22 2018-11-02 西华大学 A kind of compression superposition sequence C SI feedback methods
CN109687897A (en) * 2019-02-25 2019-04-26 西华大学 Superposition CSI feedback method based on the extensive mimo system of deep learning
CN110289898A (en) * 2019-07-18 2019-09-27 中国人民解放军空军预警学院 A kind of channel feedback method based on the perception of 1 bit compression in extensive mimo system
US20200220593A1 (en) * 2019-01-04 2020-07-09 Industrial Technology Research Institute Communication system and codec method based on deep learning and known channel state information
CN112564757A (en) * 2020-12-03 2021-03-26 西华大学 Deep learning 1-bit compression superposition channel state information feedback method

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108737032A (en) * 2018-05-22 2018-11-02 西华大学 A kind of compression superposition sequence C SI feedback methods
US20200220593A1 (en) * 2019-01-04 2020-07-09 Industrial Technology Research Institute Communication system and codec method based on deep learning and known channel state information
CN109687897A (en) * 2019-02-25 2019-04-26 西华大学 Superposition CSI feedback method based on the extensive mimo system of deep learning
CN110289898A (en) * 2019-07-18 2019-09-27 中国人民解放军空军预警学院 A kind of channel feedback method based on the perception of 1 bit compression in extensive mimo system
CN112564757A (en) * 2020-12-03 2021-03-26 西华大学 Deep learning 1-bit compression superposition channel state information feedback method

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
卿朝进 等: "部分支撑集辅助的压缩感知CSI 反馈方法", 《电子学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114826343A (en) * 2022-04-26 2022-07-29 西华大学 Superimposed channel state information feedback method and device for AI enabling data emptying
CN117807383A (en) * 2024-03-01 2024-04-02 深圳市大数据研究院 Channel state information recovery method and device, equipment and storage medium

Also Published As

Publication number Publication date
CN113726376B (en) 2022-05-20

Similar Documents

Publication Publication Date Title
CN110166089B (en) Deep learning-based superimposed coding CSI feedback method
Sun et al. AnciNet: An efficient deep learning approach for feedback compression of estimated CSI in massive MIMO systems
CN109687897B (en) Superposition CSI feedback method based on deep learning large-scale MIMO system
CN113726376B (en) 1bit Compression Superposition CSI Feedback Method Based on Feature Extraction and Mutual Dissimilarity Fusion
WO2011076110A1 (en) Method for feeding back channel state information, method and device for obtaining channel state information
CN112564757A (en) Deep learning 1-bit compression superposition channel state information feedback method
Aref et al. Deep learning-aided successive interference cancellation for MIMO-NOMA
CN102647259A (en) Uplink Transmission Method of Multi-user MIMO System Based on Singular Value Decomposition
CN109818645B (en) Superimposed CSI feedback method based on signal detection and support set assistance
Qing et al. ELM-based superimposed CSI feedback for FDD massive MIMO system
CN105610479A (en) Channel estimation method for large-scale MU-MIMO (Multi-User-Multi-Input Multi-Output) system
CN104065462B (en) There is the process of signal transmission method of diversity gain under relaying interference channel
Zhang et al. Continuous online learning-based CSI feedback in massive MIMO systems
CN114124168B (en) Deep learning-based MIMO-NOMA system signal detection method and system
TWI669921B (en) Feedback method for use as a channel information based on deep learning
CN111193535B (en) A feedback method for superimposing CSI based on ELM in FDD massive MIMO system
CN106911443A (en) Pilot tone Optimization Design in M2M communication system based on compressed sensing
WO2008151518A1 (en) The method and device for detecting information in the ofdm system
CN112821928A (en) Rate splitting precoding method of multi-user MIMO system
CN104253638A (en) MIMO (multiple input multiple output) interference alignment algorithm based on Stiefel manifold upper conjugate gradient method
CN103166687B (en) A kind of collaboration diversity transmission method based on random unitary matrix modulation
CN104935546A (en) A Blind Separation Method of MIMO-OFDM Signals to Improve the Convergence Speed of Natural Gradient Algorithm
Bai et al. Joint precoding using successive over-relaxation matrix inversion and Newton iteration for massive MIMO systems
CN115333900A (en) Unmanned aerial vehicle cooperative channel estimation and CSI feedback method based on deep learning
CN115133971A (en) A low-complexity hybrid iterative signal detection method for massive MIMO systems

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
GR01 Patent grant
GR01 Patent grant
EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20211130

Assignee: Chengdu Suyouyun Information Technology Co.,Ltd.

Assignor: XIHUA University

Contract record no.: X2023510000030

Denomination of invention: A 1-bit compressed overlay CSI feedback method based on feature extraction and mutual anisotropy fusion

Granted publication date: 20220520

License type: Common License

Record date: 20231201

EE01 Entry into force of recordation of patent licensing contract
EE01 Entry into force of recordation of patent licensing contract

Application publication date: 20211130

Assignee: Chengdu Yingling Feifan Technology Co.,Ltd.

Assignor: XIHUA University

Contract record no.: X2023510000032

Denomination of invention: A 1-bit compressed overlay CSI feedback method based on feature extraction and mutual anisotropy fusion

Granted publication date: 20220520

License type: Common License

Record date: 20231212

Application publication date: 20211130

Assignee: Sichuan Shenglongxing Technology Co.,Ltd.

Assignor: XIHUA University

Contract record no.: X2023510000031

Denomination of invention: A 1-bit compressed overlay CSI feedback method based on feature extraction and mutual anisotropy fusion

Granted publication date: 20220520

License type: Common License

Record date: 20231211