CN113726376B - 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 PDFInfo
- Publication number
- CN113726376B CN113726376B CN202111011688.XA CN202111011688A CN113726376B CN 113726376 B CN113726376 B CN 113726376B CN 202111011688 A CN202111011688 A CN 202111011688A CN 113726376 B CN113726376 B CN 113726376B
- Authority
- CN
- China
- Prior art keywords
- csi
- vector
- amplitude
- downlink csi
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000004927 fusion Effects 0.000 title claims abstract description 22
- 238000000605 extraction Methods 0.000 title claims abstract description 11
- 230000006835 compression Effects 0.000 title claims abstract description 8
- 238000007906 compression Methods 0.000 title claims abstract description 8
- 239000013598 vector Substances 0.000 claims abstract description 87
- 238000013528 artificial neural network Methods 0.000 claims abstract description 31
- 238000011084 recovery Methods 0.000 claims abstract description 17
- 230000004913 activation Effects 0.000 claims description 18
- 238000005516 engineering process Methods 0.000 claims description 11
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000008569 process Effects 0.000 claims description 7
- 238000001514 detection method Methods 0.000 claims description 6
- 239000000126 substance Substances 0.000 claims description 6
- 238000012545 processing Methods 0.000 claims description 5
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 4
- 230000005540 biological transmission Effects 0.000 claims description 3
- 230000008030 elimination Effects 0.000 claims description 3
- 238000003379 elimination reaction Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 230000007480 spreading Effects 0.000 claims description 2
- 230000001537 neural effect Effects 0.000 claims 1
- 238000011144 upstream manufacturing Methods 0.000 claims 1
- 230000002457 bidirectional effect Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 20
- 238000012549 training Methods 0.000 description 5
- 238000001228 spectrum Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 101001005269 Arabidopsis thaliana Ceramide synthase 1 LOH3 Proteins 0.000 description 1
- 101001005312 Arabidopsis thaliana Ceramide synthase LOH1 Proteins 0.000 description 1
- 101001089091 Cytisus scoparius 2-acetamido-2-deoxy-D-galactose-binding seed lectin 2 Proteins 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
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
The invention discloses a 1bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion, which comprises the following steps: obtaining the learning amplitude of the corresponding downlink CSI through a first neural network according to the amplitude of the uplink CSI estimation vector; according to the recovery feedback vector obtained by the base station, CSI (channel state information) feature extraction is carried out by using expert knowledge, and the feature amplitude and the feature angle of downlink CSI are recovered; obtaining a fusion amplitude of the downlink CSI through a second neural network according to a downlink CSI splicing amplitude obtained by splicing the characteristic amplitude of the downlink CSI and the learning amplitude of the downlink CSI; and recovering to obtain a downlink CSI reconstruction vector according to the fusion amplitude of the downlink CSI and the characteristic angle. Compared with the single-bit CS superposition CSI feedback, the method can recover the amplitude of the downlink CSI lost by the single-bit CS according to the bidirectional reciprocity of the uplink and downlink channels, greatly improve the reconstruction precision of the CSI, and obviously improve the reconstruction efficiency of the CSI at the same time.
Description
Technical Field
The invention relates to the technical field of superposition feedback of an FDD (frequency division duplex) large-scale MIMO (multiple input multiple output) system, in particular to a 1bit compression superposition Channel State Information (CSI) feedback method based on feature extraction and mutual-anisotropy fusion.
Background
As a key technology for meeting the efficient spectrum efficiency and energy efficiency of a future 5g (radio generation wireless communication) network, an FDD massive MIMO system can provide wireless data services for more users without increasing the transmission power and system bandwidth through hundreds of antennas deployed at a base station end. Meanwhile, many operations (such as multi-user scheduling, rate allocation, transmitting end precoding, and the like) which bring performance improvement in the FDD massive MIMO system depend on the acquisition of accurate downlink Channel State Information (CSI). In Frequency Division Duplex (FDD) massive MIMO systems, there is weak reciprocity between channels, and CSI can only be fed back from the user end to the base station.
Although the traditional Compressed Sensing (CS) feedback technology using signal sparsity can reduce the system feedback overhead to a certain extent, a large amount of calculation overhead is generated in the reconstruction process; the feedback technology based on Deep Learning (DL) draws attention due to its advantages such as simple structure and fast training speed, but still occupies certain spectrum resources in its feedback process.
In recent years, Superposition Coding (SC) technology has been widely used in various fields of wireless communication due to its characteristic of being able to efficiently use spectrum resources, but superposition coding causes mutual interference, which can be reduced by discarding amplitude information of CSI by a single-bit CS. The loss of CSI amplitude information results in a low accuracy of CSI reconstruction.
Disclosure of Invention
Compared with single-bit CS (circuit switched) superposed CSI feedback, the method recovers the downlink CSI amplitude lost by a single-bit CS according to the bidirectional reciprocity of an uplink channel and a downlink channel, greatly improves the CSI reconstruction precision, and improves the CSI reconstruction efficiency by utilizing a shallow neural network and a simplified version CSI reconstruction method to reconstruct the CSI.
The technical scheme of the invention is as follows:
the 1bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion comprises the following steps:
estimating vectors from uplink CSI
Amplitude of (2)
Obtaining learning amplitude of corresponding downlink CSI through a first neural network
Based on recovered feedback vector
CSI characteristic extraction is carried out by utilizing expert knowledge, and the characteristic amplitude of downlink CSI is recovered
Angle to characteristic
Wherein the recovery feedback vector
The feedback vector is a corresponding feedback vector obtained by a base station receiving end after recovery according to a feedback vector w sent by a transmitting end;
according to the characteristic amplitude of the downlink CSI
Learning amplitude with downlink CSI
Splicing amplitude of downlink CSI (channel state information) obtained by splicing
Obtaining fusion amplitude of downlink CSI through second neural network
According to the fusion amplitude of the downlink CSI
Angle to said characteristic
Recovering to obtain downlink CSI reconstruction vector
In some embodiments, the uplink CSI estimation vector magnitude
Obtained by the following model:
wherein the content of the first and second substances,
an estimation vector representing the uplink CSI
The nth element of (1), i.e., | represents the modulo operation of the complex number, and superscript T represents the transposition operation;
in some embodiments, the uplink CSI estimation vector
Receiving sequences by base station for channel estimation
And performing uplink CSI estimation, wherein the estimation method is selected from one or more of LS estimation, MMSE estimation, ML estimation, MAP estimation and pilot auxiliary estimation.
Preferably, the estimation method is selected from an LS estimation method, and the estimation vector of the uplink CSI
Satisfies the following conditions:
wherein s represents a base station known signal sequence transmitted by the user terminal,
shows that Moore-Penrose pseudo-inverse operation is taken,
represents a base station receiving sequence used for channel estimation, and satisfies the following conditions:
where N denotes channel noise, and g denotes an actual uplink CSI vector, i.e., an estimated vector of uplink CSI
Is an estimate of the vector g.
In some preferred embodiments, the first neural network comprises:
an input layer containing a linear activation function, a hidden layer containing a LeakyReLU activation function, and an output layer containing a linear activation function; the number of nodes of the input layer, the hidden layer and the output layer is N, mN and N respectively, and m represents a hidden layer node coefficient determined according to engineering presetting.
In a specific application, the input of the first neural network is the vector magnitude of the uplink CSI estimation
Outputting the learning amplitude of the downlink CSI with the length of N
More preferably, the training loss function of the first neural network is a mean square error loss function.
In a specific application, the expert knowledge is an existing reconstruction method based on compressed sensing.
In some embodiments, the existing compressed sensing-based reconstruction methods include, for example, minimization based on the norm of L1, basis pursuit algorithm, matching pursuit algorithm, orthogonal matching pursuit algorithm, BIHT algorithm, SCA-BIHT algorithm, and the like.
In some embodiments, the recovery feedback vector
Compress the feedback vector in the superposition vector x for the 1bit recovered at the base station, and:
the 1-bit compressed superposition vector
Obtained at the transmitting end by the following model:
wherein d represents an uplink user data sequence with the length of P, E represents user transmission power, and rho belongs to [0,1 ]]The representation superposition factor can be set according to engineering experience, Q represents a spreading matrix with dimension of P multiplied by L, and Q is satisfiedTQ＝P·ILWherein the superscript "T" denotes the transposition operation, L denotes the modulation signal length, ILRepresents an L-dimensional identity matrix, r represents a modulated signal sequence of length L, and:
r＝fmode(w)；
wherein w represents a feedback vector of a transmitting end with length K, and the recovered feedback vector
For the estimate of the feedback vector w at the base station, K is 2L, and:
w＝[preal,pimag,z]；
wherein z ∈ {0,1}1×NRepresenting a length N support set vector, p, of a downlink CSI vector hrealAnd pimagRespectively representing the real part and the imaginary part of the downlink CSI with the vector length of M after being compressed and quantized by adopting a 1-bit compressed sensing technology.
Preferably, p isrealAnd pimagSatisfies the following conditions:
wherein h represents a downlink CSI vector, and the downlink CSI reconstructed vector
The method comprises the steps of representing downlink CSI reconstructed at a base station, phi representing a compression matrix adopting a 1-bit compressed sensing technology, and sign (·), Re (·) and Im (·) respectively representing a sign taking operation, a real part taking operation and an imaginary part taking operation.
In some preferred embodiments, based on the above application, the recovery feedback vector
Further comprising: de-spreading the receiving sequence Y, and obtaining the recovery feedback vector by MMSE detection and interference elimination
Such as, it includes:
obtaining the received sequence by following an over-channel model
Y＝gx+N；
Wherein, Y represents the receiving sequence of the base station end, x represents the 1-bit compressed superposition vector sent by the user end, N represents the channel noise, and g represents the uplink channel vector.
Despread signal obtained by despreading processing model
Obtaining a detection signal by the following MMSE detection model
Where dec (-) denotes a hard decision operation (-)-1Representing the inverse operation of the matrix, (-)HRepresents the conjugate transpose operation of the matrix,
representing the variance of the uplink channel;
obtaining a de-interfered data sequence by the following interference elimination model
Obtaining a recovered feedback vector by the following feedback vector recovery model
Wherein f isdemoDenotes demodulation processing, and a recovery feedback vector is obtained
Comprises the following steps:
wherein the content of the first and second substances,
a support set with length N representing the recovered downlink CSI vector h,
and
and respectively representing the real part and the imaginary part of the length M of the recovered downlink CSI vector h compressed and quantized by adopting the 1-bit compressed sensing technology.
In some preferred embodiments, the feedback vector is recovered according to the obtained base station side
Performing characteristic amplitude of downlink CSI
Angle of the characteristic
The process of recovering (1) comprises:
the obtained parameters
The method comprises the following steps of (1) taking the SCA-BIHT algorithm as an input, wherein the SCA-BIHT algorithm is expert knowledge;
outputting the characteristic value of the downlink CSI after circulating n times through an SCA-BIHT algorithm
Wherein n can be preset according to engineering experience;
according to the characteristic value of downlink CSI
Obtaining the downlink CSI characteristic amplitude by the following formula
Angle of the characteristic
Wherein f isampDenotes the amplitude operation on a complex number, fang(. cndot.) denotes performing an angle operation on a complex number.
In some preferred embodiments, the second neural network comprises:
an input layer containing a linear activation function, a hidden layer containing a LeakyReLU activation function, and an output layer containing a linear activation function; the number of nodes of the input layer, the hidden layer and the output layer is 2N, kN and N respectively, and k represents a hidden layer node coefficient determined according to engineering presetting.
In a specific application, the input of the second neural network is the downlink CSI splicing amplitude
Outputting the fusion amplitude of the downlink CSI with the length of N
Wherein the downlink CSI splicing amplitude
Obtained by the following model:
wherein [ · ] denotes the splicing operation of the vectors.
More preferably, the training loss function of the second neural network is a mean square error loss function.
At one endIn some preferred embodiments, the downlink CSI reconstruction vector
Recovery of (c) is obtained by the following model:
wherein, e represents a Hadamard product, e represents a natural index, and j represents an imaginary number unit.
The invention utilizes the bidirectional reciprocity of the uplink and downlink channels, recovers the amplitude of the downlink CSI through a shallow amplitude learning network, combines the technical advantages of single-bit CS, SC and DL, superimposes the downlink CSI processed by the single-bit CS to the uplink user sequence to be fed back to the base station, recovers the downlink CSI at the base station end by using the traditional UL-US detection and the simplified version of the traditional downlink CSI reconstruction method, and then fuses the amplitude of the downlink CSI obtained by the traditional method and the amplitude learning network through the shallow amplitude fusion network, improves the accuracy of the amplitude of the downlink CSI, and can effectively reduce the mutual interference caused by superposition coding to ensure the reconstruction accuracy of the CSI under the condition of not increasing the frequency spectrum overhead.
Compared with the single-bit CS overlapped CSI feedback, the method recovers the downlink CSI amplitude lost by the single-bit CS according to the bidirectional reciprocity of the uplink and downlink channels, greatly improves the CSI reconstruction precision, and improves the CSI reconstruction efficiency by utilizing a shallow neural network and a simplified version of the traditional downlink CSI reconstruction method.
Drawings
FIG. 1 is a schematic overall flow diagram of the present invention;
FIG. 2 is a schematic diagram of a first neural network structure according to the present invention;
FIG. 3 is a structural diagram of a second neural network according to the present invention.
Detailed Description
The present invention is described in detail below with reference to the following embodiments and the attached drawings, but it should be understood that the embodiments and the attached drawings are only used for the illustrative description of the present invention and do not limit the protection scope of the present invention in any way. All reasonable variations and combinations that fall within the spirit of the invention are intended to be within the scope of the invention.
Referring to fig. 1, a specific 1-bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion includes:
a1) estimating vectors from uplink CSI
Amplitude of (2)
Obtaining learning amplitude of corresponding downlink CSI through first neural network
More specific embodiments are as follows:
the uplink CSI estimation vector
Receiving sequences by base station for channel estimation
And performing uplink CSI estimation.
The estimation of the uplink CSI may be further implemented by using an uplink CSI estimation method in the prior art, such as LS estimation, MMSE estimation, ML estimation, MAP estimation, pilot-assisted estimation, and the like, for example, in a specific embodiment, the uplink CSI estimation performed by using LS estimation is as follows:
wherein the content of the first and second substances,
indicating the received estimation sequence at the base station end, s indicating the base station known signal transmitted by the user end,
represents the Moore-Penrose pseudo-inverse operation.
Referring to fig. 2, the first neural network includes the following neural network structure:
the device comprises an input layer, a hidden layer and an output layer, wherein the input layer adopts a linear activation function, the hidden layer adopts an activation function LeakyReLU, and the output layer adopts a linear activation function.
The number of nodes of an input layer, a hidden layer and an output layer of the first neural network is N, mN and N respectively, m represents a node coefficient of the hidden layer and can be obtained according to engineering presetting.
Obtaining a learning amplitude of downlink CSI through the first neural network
The process comprises the following steps:
estimating a vector according to the uplink CSI
Obtaining uplink CSI amplitude by
Wherein the content of the first and second substances,
representing vectors
The nth element of (a), i |, represents a modulo operation of a complex number;
estimating vector magnitude of the obtained uplink CSI
Self-transfusionInputting the first neural network in the layer, and outputting the learning amplitude of the downlink CSI with the length of N
The training loss function of the first neural network adopts a mean square error loss function.
a2) Recovering feedback vector obtained from base station
CSI characteristic extraction is carried out by utilizing expert knowledge, and the characteristic amplitude of downlink CSI is recovered
Angle of the characteristic
a3) According to the characteristic amplitude of the downlink CSI
Learning amplitude with downlink CSI
Splicing amplitude of downlink CSI (channel state information) obtained by splicing
Obtaining a fusion amplitude of downlink CSI through a second neural network
More specific embodiments are as follows:
referring to fig. 3, the second neural network is a neural network structure including:
the device comprises an input layer, a hidden layer and an output layer, wherein the input layer adopts a linear activation function, the hidden layer adopts a LeakyReLU activation function, and the output layer adopts a linear activation function.
The number of nodes of an input layer, a hidden layer and an output layer of the second neural network is 2N, kN and N respectively, k represents a node coefficient of the hidden layer and can be obtained according to engineering presetting.
Obtaining a fusion amplitude of downlink CSI through the second neural network
The process of (2) comprises:
by the characteristic amplitude of the downlink CSI
Learning range with the downlink CSI
Obtaining downlink CSI splicing amplitude
Wherein [ ·]A stitching operation that represents a vector is performed,
a real number set with dimension of 1 × 2N is represented;
splicing the obtained downlink CSI
Inputting the second neural network from the input layer, and outputting the fusion amplitude of the downlink CSI with the length of N
The training loss function of the amplitude fusion network adopts a mean square error loss function.
a4) According to the fusion amplitude of the downlink CSI
Angle to said characteristic
Recovering to obtain downlink CSI reconstruction vector
More specific embodiments are as follows:
the recovery obtains a downlink CSI reconstruction vector
The method of (1) is as follows:
wherein, e represents a Hadamard product, e represents a natural index, and j represents an imaginary number unit.
Example 1
In step a1), obtaining an uplink CSI estimation vector through LS estimation
One specific example of (a) is as follows:
suppose that: n-2, P-4, base station-side received sequence for channel estimation
Comprises the following steps:
the base station known signal s sent by the user side is:
s＝[0.7528-0.6083i -0.1666-0.1308i 0.9869+0.4514i 0.4556+0.2695i]；
pseudo inverse matrix of base station known signal s transmitted by user terminal
Comprises the following steps:
according to LS estimation processing formula
Can calculate the uplink CSI estimation vector
Comprises the following steps:
example 2
In the step a2), the lengths of the recovered downlink CSI vectors h compressed and quantized by adopting the 1-bit compressed sensing technology are all the real parts of M
And imaginary part
And recovering the feedback vector
Obtaining a support set with the length of N of the recovered downlink CSI vector h
One specific example of (a) is as follows:
suppose that: n2 and M3, recovering the feedback vector
Comprises the following steps:
according to the formula
The recovered real part of the length M of the downlink CSI vector h compressed and quantized by the 1-bit compressed sensing technology can be obtained
Comprises the following steps:
the lengths of the recovered downlink CSI vectors h compressed and quantized by the 1-bit compressed sensing technology are M imaginary parts
Comprises the following steps:
support set with length N of recovered downlink CSI vector h
Comprises the following steps:
example 3
In step a1), the vector is estimated by uplink CSI
Obtaining uplink CSI estimation vector magnitude
One specific example of (a) is as follows:
estimating the CSI obtained in example 1Vector
According to the formula
Computing inputs in an amplitude learning network
Comprises the following steps:
example 4
In step a3), obtaining downlink CSI splicing amplitude
One specific example of (a) is as follows:
suppose that: n2, characteristic amplitude of downlink CSI
Learning amplitude of downlink CSI
According to the model
The input namely downlink CSI splicing amplitude in the amplitude fusion network can be obtained
Comprises the following steps:
example 5
In the step a4), recovering to obtain the downlink CSI reconstruction vector
One specific example of (a) is as follows:
suppose that: n2, fusion amplitude of downlink CSI
Characteristic angle of downlink CSI
According to the model
The recovered downlink CSI reconstruction vector can be calculated
Comprises the following steps:
the above examples are merely preferred embodiments of the present invention, and the scope of the present invention is not limited to the above examples. All technical schemes belonging to the idea of the invention belong to the protection scope of the invention. It should be noted that modifications and embellishments within the scope of the invention may be made by those skilled in the art without departing from the principle of the invention, and such modifications and embellishments should also be considered as within the scope of the invention.
Claims (7)
1. The 1bit compression superposition CSI feedback method based on feature extraction and mutual-difference fusion is characterized by comprising the following steps:
estimation vector based on uplink CSI
Amplitude of (2)
Obtaining correspondences via a first neural networkLearning range of downlink CSI
Based on recovered feedback vector
CSI characteristic extraction is carried out by utilizing expert knowledge, and the characteristic amplitude of downlink CSI is recovered
Angle of the characteristic
Wherein the recovery feedback vector
The feedback vector is a corresponding feedback vector obtained by a base station receiving end after recovery according to a feedback vector w sent by a transmitting end;
according to the characteristic amplitude of the downlink CSI
Learning range with downlink CSI
Splicing amplitude of downlink CSI (channel state information) obtained by splicing
Obtaining a fusion amplitude of downlink CSI through a second neural network
According to the fusion amplitude of the downlink CSI
Angle to said characteristic
Recovering to obtain downlink CSI reconstruction vector
Wherein the recovery feedback vector
The feedback vector in the superposition vector x is compressed for 1bit recovered at the base station, and:
the 1-bit compressed superposition vector
Obtained at the transmitting end by the following model:
wherein d represents an uplink user data sequence with the length of P, E represents user transmission power, and rho belongs to [0,1 ]]The superposition factor is expressed and can be set according to engineering experience, Q represents a spreading matrix with dimension of P multiplied by L and satisfies QTQ＝P·ILWhere the superscript T denotes the transposition operation, L denotes the modulation signal length, ILExpressing an L-dimensional identity matrix, and r expressing a modulation signal sequence with the length of L;
obtaining the recovery feedback vector
The process of (2) comprises:
despread signal obtained by despreading processing model
Wherein, Y represents a receiving sequence of NxP dimension at the base station end;
obtaining a detection signal by the following MMSE detection model
Where dec (-) denotes a hard decision operation and g denotes an upstream channel vector, (.)-1Representing the inverse operation of the matrix, (-)HRepresents the conjugate transpose operation of the matrix,
representing the variance of the uplink channel;
obtaining a de-interfered data sequence by the following interference elimination model
Obtaining a recovered feedback vector by the following feedback vector recovery model
Wherein f isdemoDenotes demodulation processing, and a recovery feedback vector is obtained
Comprises the following steps:
wherein the content of the first and second substances,
a support set with length N representing the recovered downlink CSI vector h,
and
respectively representing the real part and the imaginary part of the length M of the recovered downlink CSI vector h compressed and quantized by adopting a 1-bit compressed sensing technology;
characteristic amplitude of the downlink CSI
Angle of the characteristic
The process of recovering (1) comprises:
the obtained parameters
The SCA-BIHT algorithm is used as the input of the SCA-BIHT algorithm, wherein the SCA-BIHT algorithm is expert knowledge;
outputting the characteristic value of the downlink CSI after circulating n times through an SCA-BIHT algorithm
Wherein n can be preset according to engineering experience;
according to the characteristic value of downlink CSI
Obtaining the downlink CSI characteristic amplitude by the following formula
Angle of the characteristic
Wherein f isampDenotes the amplitude operation on a complex number, fang(. cndot.) denotes performing an angle operation on a complex number.
2. The method of claim 1, wherein the uplink CSI estimation vector magnitude
Obtained by the following model:
wherein the content of the first and second substances,
an estimation vector representing the uplink CSI
The nth element, |, represents the modulo operation of the complex number, and the superscript T represents the transpose operation.
3. The method of claim 2, wherein the estimation vector of the uplink CSI is estimated based on a Channel State Information (CSI)
Obtained by the following model:
wherein s represents a base station known signal sequence transmitted by the user terminal,
showing that Moore-Penrose pseudo-inverse operation is taken,
represents a base station receiving sequence used for channel estimation, and satisfies the following conditions:
where N denotes channel noise and g denotes an actual uplink CSI vector.
4. The method of claim 2, wherein the first neural network comprises:
an input layer containing a linear activation function, a hidden layer containing a Leaky ReLU activation function, and an output layer containing a linear activation function; the number of nodes of the input layer, the hidden layer and the output layer is N, mN and N respectively, and m represents a hidden layer node coefficient determined according to engineering presetting.
5. The method of claim 2, wherein the second neural network comprises:
an input layer containing a linear activation function, a hidden layer containing a Leaky ReLU activation function, and an output layer containing a linear activation function; the number of nodes of the input layer, the hidden layer and the output layer is 2N, kN and N respectively, and k represents a hidden layer node coefficient determined according to engineering presetting.
6. The method of claim 1, wherein the neural network is trained by a mean square error loss function.
7. The method of claim 1, wherein the downlink CSI reconstruction vector
Recovery of (c) is obtained by the following model:
wherein, e represents a Hadamard product, e represents a natural index, and j represents an imaginary number unit.
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-difference 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-difference fusion |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113726376A CN113726376A (en) | 2021-11-30 |
| CN113726376B true 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-difference fusion |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113726376B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114826343B (en) * | 2022-04-26 | 2023-06-27 | 西华大学 | Method and device for feeding back information of superimposed channel state with AI (advanced technology attachment) energized data being empty |
| 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 (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108737032B (en) * | 2018-05-22 | 2020-11-24 | 西华大学 | Compressed superposition sequence CSI feedback method |
| TWI687063B (en) * | 2019-01-04 | 2020-03-01 | 財團法人工業技術研究院 | A communication system and codec method based on deep learning and channel state information |
| CN110289898B (en) * | 2019-07-18 | 2022-05-06 | 中国人民解放军空军预警学院 | Channel feedback method based on 1-bit compressed sensing in large-scale 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 |
-
2021
- 2021-08-31 CN CN202111011688.XA patent/CN113726376B/en active Active
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113726376A (en) | 2021-11-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| 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-difference fusion | |
| CN100403596C (en) | Efficient multiple input multiple output system for multi-path fading channels | |
| CN112564757A (en) | Deep learning 1-bit compression superposition channel state information feedback method | |
| CN102474314B (en) | Precoding method and device | |
| Khan et al. | A robust channel estimation scheme for 5G massive MIMO systems | |
| JP2007534250A (en) | Method and apparatus for transmitting a signal in a multi-antenna system, signal and method for estimating a corresponding transmission channel | |
| JP2009049604A (en) | Transmitter, and beam forming matrix generating method | |
| CN101394216B (en) | Pre-coding method and codebook constructing method based on code book mode | |
| CN109818645B (en) | Superposition CSI feedback method based on signal detection and support set assistance | |
| Oyerinde | Comparative study of Overloaded and Underloaded NOMA schemes with two Multiuser Detectors | |
| US20150098402A1 (en) | Wireless reception device, wireless transmission device, wireless communication system, program, and integrated circuit | |
| CN102340336A (en) | User pairing method of MU-MIMO based on SLNR | |
| CN114124168A (en) | MIMO-NOMA system signal detection method and system based on deep learning | |
| CN111193535B (en) | Feedback method based on ELM superposition CSI in FDD large-scale MIMO system | |
| CN101488775B (en) | Space multiplexing method and apparatus in TD-SCDMA system | |
| WO2010146774A1 (en) | Wireless communication apparatus, wireless communication system and communication method | |
| CN102075220B (en) | Channel estimating device and method based on time domain noise reduction | |
| Zhao et al. | A novel signal detection algorithm for underwater mimo-OFDM systems based on generalized MMSE | |
| CN101459645B (en) | Detection method based on sub-band in multi-antenna OFDM system | |
| TW201944745A (en) | Feedback method for use as a channel information based on deep learning | |
| Liu et al. | Robust and energy efficient sparse-Coded OFDM-DCSK system via matrix recovery | |
| CN100544327C (en) | A kind of detector for serial interference deletion in minimum mean square error of low complex degree | |
| CN105812310B (en) | Single carrier MIMO communication means and device based on ZCZ sequence and phase rotation | |
| Wei et al. | Optimal binary/quaternary adaptive signature design for code-division multiplexing |
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 |