CN110099016A - A kind of sparse front channel estimation methods of millimeter wave based on deep learning network - Google Patents
A kind of sparse front channel estimation methods of millimeter wave based on deep learning network Download PDFInfo
- Publication number
- CN110099016A CN110099016A CN201910397076.5A CN201910397076A CN110099016A CN 110099016 A CN110099016 A CN 110099016A CN 201910397076 A CN201910397076 A CN 201910397076A CN 110099016 A CN110099016 A CN 110099016A
- Authority
- CN
- China
- Prior art keywords
- matrix
- channel
- indicate
- sparse
- dimension
- 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
Links
Classifications
-
- 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
- G06N3/045—Combinations of networks
-
- 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
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L25/00—Baseband systems
- H04L25/02—Details ; arrangements for supplying electrical power along data transmission lines
- H04L25/0202—Channel estimation
- H04L25/024—Channel estimation channel estimation algorithms
- H04L25/0254—Channel estimation channel estimation algorithms using neural network algorithms
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/26—Systems using multi-frequency codes
- H04L27/2601—Multicarrier modulation systems
- H04L27/2626—Arrangements specific to the transmitter only
-
- 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)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Artificial Intelligence (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Power Engineering (AREA)
- Radio Transmission System (AREA)
Abstract
The invention discloses a kind of sparse front channel estimation methods of millimeter wave based on deep learning network, and using millimeter wave channel sparse characteristic as prior information, the full connection deep neural network of training design is used for millimeter wave surface communications channel estimation.First using full connection phase shifter network, isotropic analog transceiver is designed by configuring each phase shifter uniform phase distribution;Then using the channel sparse information of acquisition and the optimal digital estimator of design as the training data of full connection deep learning network.For the condition of sparse channel under each signal-to-noise ratio, the sparse information of channel is inputted into network, obtains corresponding digital estimator, and then obtain channel estimation results.The condition of sparse channel estimator that the present invention provides can reduce low precision analog-digital converter nonlinear quantization bring error, and use deep learning network implementations, to reduce channel estimation complexity, channel estimation methods that inventive can be optimal on approximation theory.
Description
Technical field
The present invention relates to the communications fields, are related to channel estimation methods, are based on depth more particularly relate to one kind
Practise the sparse front channel estimation methods of millimeter wave of network.
Background technique
In recent years, the communication technology made breakthrough progress and gradually tend to mature mobile communication in global range
Swift and violent development has been obtained in industry.Multiple-input and multiple-output (MIMO) technology is one of key technology of development communication technologies,
Improve the message transmission rate of system.More antennas are equipped with by the transmitting terminal and receiving end of system, utilize transmitting-receiving both ends
Multiple antennas formed diversity, system stability can be improved.At the same time, due to the independent channel number between transmitting-receiving both ends antenna
Greatly increase, the data volume that system is sent in the unit time is also promoted, and then improves the spectrum utilization efficiency of system.
Massive MIMO (i.e. extensive antenna) technology is in transmission rate, energy efficiency, transmission reliability etc. phase
Have compared with MIMO technology and is greatly improved, and the extensive MIMO technology of millimeter wave substantially reduces the configuration of extensive antenna array
Difficulty, and extensive MIMO technology then solves the problems, such as that the loss of millimeter-wave signal height is easily blocked.It is big in order to reduce millimeter wave
The numerical model analysis framework of a small amount of radio frequency link can be used in power consumption and hardware complexity in scale mimo system.
In order to carry out high-property transmission, need to carry out channel estimation first.Traditional extensive multi-antenna technology system
Channel estimation inherently a major challenge, the use of low precision analog-digital converter is reducing the same of cost and power consumption under mixed architecture
When allow channel estimation to become more difficult.Meanwhile how estimation complexity is reduced while using millimeter wave channel sparsity
And pilot-frequency expense, and obtaining high-precision channel estimation is also a major challenge.
Summary of the invention
To solve the above problems, the present invention proposes a kind of condition of sparse channel estimation method, it to be used for low precision analog-digital converter
And in the extensive multi-antenna technology system of broadband mmWave of hybrid structure.Believe by priori of millimeter wave channel sparse characteristic
Breath, using the selection matrix of condition of sparse channel and its corresponding digital estimator as input to the full connection deep neural network of design
It is trained, obtains the deep neural network suitable for different signal-to-noise ratio, be used for millimeter wave surface communications channel estimation.It adopts first
With full connection phase shifter network, isotropic analog transceiver is designed by configuring each phase shifter uniform phase distribution;So
Afterwards using the channel sparse information of acquisition as priori knowledge, optimal digital estimator is designed, it will be both as full connection depth
The training data of learning network.For the condition of sparse channel under each signal-to-noise ratio, the sparse information of channel is inputted into network, can be obtained
Corresponding digital estimator, to obtain channel estimation results.The condition of sparse channel estimator that the present invention provides can reduce low essence
Analog-digital converter nonlinear quantization bring error is spent, and using deep learning network implementations to reduce channel estimation complexity
Degree.In the hybrid structure multiaerial system using low precision analog-digital converter, inventive can be optimal on approximation theory
Channel estimation methods.
In order to achieve the above object, the invention provides the following technical scheme:
A kind of sparse front channel estimation methods of millimeter wave based on deep learning network, comprising:
Transmitting terminal pilot signal transmitted connects by the simulation precoder of the phase shifter composition connected entirely via channel arrival
Receipts machine, receiving end obtain channel estimation using simulated estimation device, low precision analog-digital converter quantizer and deep learning network;
It is characterized by: channel estimation methods on the basis of the hardware structure the following steps are included:
Step 1: base station design simulates precoding and pilot signal transmitted, simulates precoding FAmIt designs according to the following formula:
Wherein, []ijThe element that i-th row of representing matrix, jth arrange;FAmExpression dimension is Nt×NRFtSimulation precoding
Matrix, NtIndicate transmitting antenna number, NRFtIndicate transmitting terminal rf chain number;Indicate the i-th row, the jth of simulation pre-coding matrix
The phase of column element;
Step 2: receiving end optimization design simulated estimation device WAm, WAmIt calculates according to the following formula:
Wherein, WAmExpression dimension is Nr×NRFrSimulated estimation matrix, NrIndicate receiving antenna number, NRFrIndicate receiving end
Rf chain number;Indicate the i-th row of simulated estimation matrix, the phase of jth column element;
Step 3: receiving end obtains channel by compressed sensing technology, such as classical orthogonal matching pursuit (OMP) technology
Sparse features information, i.e. nonzero element position in condition of sparse channel matrix, and in this, as the prior information of subsequent channel estimation;
Step 4: receiving end is for all possible selection matrix P corresponding to condition of sparse channel characteristic informationvThe collection of [k]
Conjunction is stored, and according to each selection matrix PvThe corresponding digital estimated matrix of [k] design
Step 5: above-mentioned selection matrix PvThe set of [k] and digital estimated matrix correspondinglyAs complete
The training set of Connection Neural Network obtains the feed-forward net connected entirely being suitable under each signal-to-noise ratio by neural metwork training
Network.In coherence time, receiving end is according to the corresponding selection matrix P of condition of sparse channelv[k], application network structure obtain network output
Digital estimated matrixIt is detected with this data estimated matrix to signal is received, to obtain channel estimation value
Further, selection matrix P in the step 4v[k] is indicated with following formula:
Wherein, eπ(i)(π (i) ∈ { 1,2 ..., NrNt) expression dimension be NrNt× 1 a element of π (i) be 1, remaining
The vector that element is 0.NvIndicate that dimension is N on k-th of subcarrierrNt× 1 channel vector projects to the vector in angle domain point
Measure hvThe number of nonzero element in [k].
For the nonzero channel element in channel matrix, the selection matrix P of channelvThe form of [k] has N kind possibility, for
All possible set { Pv1[k], Pv2[k] ..., PvN[k] }, each Pvi[k] is equal a possibility that (i=1 ..., N) occurs
ForI.e.
Wherein, N is all possible number,C indicates number of combinations formula.NvIt indicates on k-th of subcarrier
Dimension is NrNt× 1 channel vector projects to the vector component h in angle domainvThe number of nonzero element in [k].
Further, channel is projected in virtual angle domain and obtains channel vector component hv[k], hv[k] as follows
It calculates:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix,Indicate square
Battle array AtConjugation;Indicate Kronecker product;ArExpression dimension is Nr×NrReceive front response vector composition reception dictionary
Matrix;H [k] indicates that the dimension on k-th of subcarrier is Nr×NtPhysical channel matrix, vec (H [k]) representing matrix H's [k]
Vector quantization.
Wherein, AtIt indicates as follows:
Wherein,(p ∈ { 1,2 ..., Nt) expression dimension be Nt× 1 transmitting front response vector,
InWherein, NtThe horizontal axis antenna of=P × Q, P expression transmitting antenna front
Number, Q indicate the longitudinal axis antenna number of transmitting antenna front;
ArIt indicates as follows:
Wherein,(q ∈ { 1,2 ..., NrExpression dimension be Nr× 1 reception front response vector,
In,Wherein, NrThe horizontal axis antenna of=I × J, I expression receiving antenna front
Number, J indicate the longitudinal axis antenna number of receiving antenna front;
Further, optimal number estimated matrix in the step 4Use Minimum Mean Square Error of Channel Estimation
Criterion,It can be calculated as follows:
Wherein, k ∈ { 1,2 ..., K } indicates that k-th of subcarrier, K indicate total number of sub-carriers;Indicate k-th of son
N on carrier waveRFr×NvOptimal digital estimated matrix, M indicates channel access times, i.e. channel estimation number in coherence time.ηb
Indicate distortion factor related with analog-digital converter (ADC) quantizing bit number b;Ω [k] indicates that the dimension on k-th of subcarrier is
MNRFr×NvCalculation matrix;ΩHThe conjugate transposition of [k] representing matrix Ω [k];Indicate the large-scale fading coefficient of channel;Expression dimension is Nv×NvUnit matrix;Indicate that equivalent noise vector is each
The variance of element,Indicate additive white Gaussian noise (AWGN) variance;P indicates transmitting pilot power.
Further, channel estimation number M is indicated with following formula:
Expression rounds up.
The calculation matrix Ω [k] is indicated with following formula:
Ω [k]=Φ [k] Ψ Pv[k],
Wherein, dimension NRFr×NrNtPilot tone correlation matrixsm[k](m
∈ { 1,2 ..., M }) indicate the m time it is trained when dimension be NRFr× 1 transmitting pilot tone vector.
Wherein, Ψ indicates that dimension is NrNt×NrNtSpace conversion matrix, Ψ are indicated as follows:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix, ArIndicate dimension
For Nr×NrReceive front response vector composition reception dictionary matrix.
Wherein, Pv[k] indicates that dimension is NrNt×NvSelection matrix.
Further, in step 5Pass through the full connection deep learning network of feed-forward as follows
It obtains:
(1) nonzero element number N known for onevCondition of sparse channel, channel matrix H [k] indicate k-th of subcarrier on
Dimension be Nr×NtPhysical channel matrix, under each signal-to-noise ratio, receiving end calculates all possibility according to sparse features information
PvCovariance matrix C [the k]=Ω of [k] compositionH[k] Ω [k], wherein Ω [k]=Φ [k] Ψ Pv[k].Meanwhile according to right
It is required that method described in 4, calculates and each PvThe covariance matrix C [k] of [k] composition is correspondingIt designs newly
Digital estimated matrixAnd by all possible C [k] and corresponding
It stores in database.
(2) data are extracted from database, are divided into training data and two groups of test data.Training data progress complex value is torn open
Divide operation, it will training set C [k] and W 'D[k] is split as real-part matrix CR[k]、With imaginary-part matrix CI[k]、
Two parts.
(3) then by CR[k]、And CI[k]、Matrix vector operation is carried out, obtaining dimension is Nv 2×
1 column vector cR[k]、And cI[k]、By cR[k]、As the defeated of real part deep learning network
Enter and training objective, by cI[k]、Input and training objective as imaginary part deep learning network.
(4) fully-connected network of two deep learnings is constructed, network structure is identical.It is all the full connection of two layers of feed-forward
Neural network, first layer neuron number are N, whereinBiasing (bias) connection is set in first layer simultaneously, and
The transfer function of first layer is set as softmax function;First layer is exported and is connect with second layer neuron, neuron number is
N, and biasing (bias) connection is equally set in the second layer.
Softmax function is defined as follows:
Wherein, the output of each layer of fully-connected network such as following formula:
Wherein, W and b indicates the parameter of full Connection Neural Network, yiAnd biIndicate i-th of element of y and b, xjIndicate x's
J-th of element, WI, jIndicate that position is the element of (i, j) in W.
(5) such full Connection Neural Network is trained, and passes through the fractionation of the complex value of test data and matrix arrow
Quantization tests the deep learning network as input, to obtain the stabilizing network structure under each signal-to-noise ratio.
(6) in coherence time, for the condition of sparse channel under each signal-to-noise ratio, by the channel association side of the sparse information containing channel
Poor Matrix C [k] inputs network, can obtain corresponding digital estimator
Further, the channel estimation value in step 5It obtains as follows:
Wherein, ()HThe conjugate transposition operation of representing matrix;()-1The inversion operation of representing matrix.
Compared with prior art, the invention has the advantages that and the utility model has the advantages that
1) broad-band channel should be estimated, the present invention is modulated using OFDM, to carry out frequency domain letter to each narrow-band sub-carriers
Road estimation, the present invention is using millimeter wave channel sparse characteristic as prior information, by the selection matrix of condition of sparse channel and its corresponding number
Word estimator is trained the full connection deep neural network of design as input, obtains the depth suitable for different signal-to-noise ratio
Neural network is used for millimeter wave surface communications channel estimation;2) program is using full connection phase shifter network, by configuring each shifting
Phase device uniform phase is distributed to design isotropic analog transceiver;The channel sparse features information of acquisition is known as priori
Know, designs optimal digital estimator, it will be both as the training data of full connection deep learning network.For under each signal-to-noise ratio
Condition of sparse channel, the sparse information of channel is inputted into network, corresponding digital estimator can be obtained, to obtain channel estimation
As a result;3) the condition of sparse channel estimator that the present invention provides can reduce low precision analog-digital converter nonlinear quantization bring and miss
Difference, and using deep learning network implementations to reduce channel estimation complexity.In the mixing using low precision analog-digital converter
In framework multiaerial system, channel estimation methods that inventive can be optimal on approximation theory.
Detailed description of the invention
Fig. 1 is present system block diagram;
Fig. 2 is the millimeter wave antenna front of 8*4, under the analog-digital converter quantization of 4 bit of receiving end, channel estimation normalization
The curve graph that mean square error (NMSE) changes with signal-to-noise ratio (SNR);
Specific embodiment
Technical solution provided by the invention is described in detail below with reference to specific embodiment, it should be understood that following specific
Embodiment is only illustrative of the invention and is not intended to limit the scope of the invention.
The invention proposes a kind of sparse front channel estimation methods of millimeter wave based on deep learning network, with millimeter wave
Channel sparse characteristic is prior information, using the selection matrix of condition of sparse channel and its corresponding digital estimator as input to design
Full connection deep neural network be trained, obtain the deep neural network suitable for different signal-to-noise ratio, be used for millimeter wave battle array
Face communications channel estimation.
It is isotropic to design by configuring each phase shifter uniform phase distribution first using full connection phase shifter network
Analog transceiver;Then using the channel sparse information of acquisition as priori knowledge, optimal digital estimator is designed, the two is made
For the training data for connecting deep learning network entirely.For the condition of sparse channel under each signal-to-noise ratio, the sparse information of channel is inputted
Network can obtain corresponding digital estimator, to obtain channel estimation results.The condition of sparse channel estimator that the present invention provides
Can reduce low precision analog-digital converter (ADC) nonlinear quantization bring error, and using deep learning network implementations to
Reduce channel estimation complexity.In the hybrid structure multiaerial system using low precision analog-digital converter, inventive energy energy
Optimal channel estimation methods on enough approximation theories.
As shown in Figure 1, under each signal-to-noise ratio, by compressed sensing technology acquisition channel priori sparse information, and according to
All possible condition of sparse channel design the optimal digital estimator under corresponding least mean-square error, connect entirely as deep learning
The training input of network.By the way that deep learning network is trained and is tested, the network that can be used under each signal-to-noise ratio is obtained,
For channel estimation.Under hybrid structure, transmitting terminal pilot signal transmitted prelists by the simulation of the phase shifter composition connected entirely
Code device reaches receiver via channel, by simulated estimation device, low precision analog-digital converter quantizer, under each signal-to-noise ratio
The sparse information of channel is inputted network, corresponding digital estimator can be obtained, to obtain channel estimation knot by condition of sparse channel
Fruit.The condition of sparse channel estimator that the present invention provides can reduce low precision analog-digital converter nonlinear quantization bring error, and
Using deep learning network implementations to reduce channel estimation complexity.More using the hybrid structure of low precision analog-digital converter
In antenna system, channel estimation methods that inventive can be optimal on approximation theory.
Channel estimation methods proposed by the present invention include the following steps:
Step 1: base station design simulates precoding and pilot signal transmitted, simulates precoding FAmIt designs according to the following formula:
Wherein, []ijThe element that i-th row of representing matrix, jth arrange;FAmExpression dimension is Nt×NRFtSimulation precoding
Matrix, NtIndicate transmitting antenna number, NRFtIndicate transmitting terminal rf chain number;Indicate the i-th row, the jth of simulation pre-coding matrix
The phase of column element;
Step 2: receiving end optimization design simulated estimation device WAm, WAmIt calculates according to the following formula:
Wherein, WAmExpression dimension is Nr×NRFrSimulated estimation matrix, NrIndicate receiving antenna number, NRFrIndicate receiving end
Rf chain number;Indicate the i-th row of simulated estimation matrix, the phase of jth column element;
Step 3: receiving end obtains channel by compressed sensing technology, such as classical orthogonal matching pursuit (OMP) technology
Sparse features information, i.e. nonzero element position in condition of sparse channel matrix, and in this, as the prior information of subsequent channel estimation.
Step 4: receiving end is for all possible selection matrix P corresponding to condition of sparse channel characteristic informationvThe collection of [k]
Conjunction is stored, and according to each selection matrix PvThe corresponding digital estimated matrix of [k] design
Step 5: above-mentioned selection matrix PvThe set of [k] and digital estimated matrix correspondinglyAs complete
The training set of Connection Neural Network obtains the feed-forward net connected entirely being suitable under each signal-to-noise ratio by neural metwork training
Network.In coherence time, receiving end is according to the corresponding selection matrix P of condition of sparse channelv[k], application network structure obtain network output
Digital estimated matrixIt is detected with this data estimated matrix to signal is received, to obtain channel estimation value
Further, selection matrix P in the step 4v[k] is indicated with following formula:
Wherein, eπ(i)(π (i) ∈ { 1,2 ..., NrNt) expression dimension be NrNt× 1 a element of π (i) be 1, remaining
The vector that element is 0.NvIndicate that dimension is N on k-th of subcarrierrNt× 1 channel vector projects to the vector in angle domain point
Measure hvThe number of nonzero element in [k].
For the nonzero channel element in channel matrix, the selection matrix P of channelvThe form of [k] has N kind possibility, for
All possible set { Pv1[k], Pv2[k] ..., PvN[k] }, each Pvi[k] is equal a possibility that (i=1 ..., N) occurs
ForI.e.
Wherein, N is all possible number,C indicates number of combinations formula.NvIt indicates to tie up on k-th of subcarrier
Degree is NrNt× 1 channel vector projects to the vector component h in angle domainvThe number of nonzero element in [k].
Further, channel is projected in virtual angle domain and obtains channel vector component hv[k], hv[k] as follows
It calculates:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix,Representing matrix
AtConjugation;Indicate Kronecker product;ArExpression dimension is Nr×NrReceive front response vector composition reception dictionary square
Battle array;H [k] indicates that the dimension on k-th of subcarrier is Nr×NtPhysical channel matrix, the arrow of vec (H [k]) representing matrix H [k]
Quantization.
Wherein, AtIt indicates as follows:
Wherein,(p ∈ { 1,2 ..., Nt) expression dimension be Nt× 1 transmitting front response vector,
InWherein, NtThe horizontal axis antenna of=P × Q, P expression transmitting antenna front
Number, Q indicate the longitudinal axis antenna number of transmitting antenna front.
ArIt indicates as follows:
Wherein,(q ∈ { 1,2 ..., NrExpression dimension be Nr× 1 reception front response vector,
In,Wherein, NrThe horizontal axis antenna of=I × J, I expression receiving antenna front
Number, J indicate the longitudinal axis antenna number of receiving antenna front.
Further, optimal number estimated matrix in the step 4Use Minimum Mean Square Error of Channel Estimation
Criterion,It can be calculated as follows:
Wherein, k ∈ { 1,2 ..., K } indicates that k-th of subcarrier, K indicate total number of sub-carriers;Indicate k-th of son
N on carrier waveRFr×NvOptimal digital estimated matrix, M indicates channel access times, i.e. channel estimation number in coherence time.ηb
Indicate distortion factor related with analog-digital converter (ADC) quantizing bit number b;Ω [k] indicates that the dimension on k-th of subcarrier is
MNRFr×NvCalculation matrix;ΩHThe conjugate transposition of [k] representing matrix Ω [k];Indicate the large-scale fading coefficient of channel;Expression dimension is Nv×NvUnit matrix;Indicate that equivalent noise vector is each
The variance of element,Indicate additive white Gaussian noise (AWGN) variance;P indicates transmitting pilot power.
Further, channel estimation number M is indicated with following formula:
Expression rounds up.
The calculation matrix Ω [k] is indicated with following formula:
Ω [k]=Φ [k] Ψ Pv[k],
Wherein, dimension NRFr×NrNtPilot tone correlation matrixsm[k] (m ∈ 1,
2 ..., M }) indicate the m time it is trained when dimension be NRFr× 1 transmitting pilot tone vector.
Wherein, Ψ indicates that dimension is NrNt×NrNtSpace conversion matrix, Ψ are indicated as follows:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix, ArIndicate dimension
For Nr×NrReceive front response vector composition reception dictionary matrix.
Wherein, Pv[k] indicates that dimension is NrNt×NvSelection matrix.
Further, in step 5Pass through the full connection deep learning network of feed-forward as follows
It obtains:
(1) nonzero element number N known for onevCondition of sparse channel, channel matrix H [k] indicate k-th of subcarrier on
Dimension be Nr×NtPhysical channel matrix, under each signal-to-noise ratio, receiving end calculates all possibility according to sparse features information
PvCovariance matrix C [the k]=Ω of [k] compositionH[k] Ω [k], wherein Ω [k]=Φ [k] Ψ Pv[k].Meanwhile according to right
It is required that method described in 4, calculates and each PvThe covariance matrix C [k] of [k] composition is correspondingIt designs newly
Digital estimated matrixAnd by all possible C [k] and corresponding
It stores in database.
(2) data are extracted from database, are divided into training data and two groups of test data.Training data progress complex value is torn open
Divide operation, it will training set C [k] and W 'D[k] is split as real-part matrix CR[k]、With imaginary-part matrix CI[k]、
Two parts.
(3) then by CR[k]、And CI[k]、Matrix vector operation is carried out, obtaining dimension is Nv 2×
1 column vector cR[k]、And cI[k]、By cR[k]、As the defeated of real part deep learning network
Enter and training objective, by cI[k]、Input and training objective as imaginary part deep learning network.
(4) fully-connected network of two deep learnings is constructed, network structure is identical.It is all the full connection of two layers of feed-forward
Neural network, first layer neuron number are N, whereinBiasing (bias) connection is set in first layer simultaneously, and
The transfer function of first layer is set as softmax function;First layer is exported and is connect with second layer neuron, neuron number is
N, and biasing (bias) connection is equally set in the second layer.
Softmax function is defined as follows:
Wherein, the output of each layer of fully-connected network such as following formula:
Wherein, W and b indicates the parameter of full Connection Neural Network, yiAnd biIndicate i-th of element of y and b, xjIndicate x's
J-th of element, WI, jIndicate that position is the element of (i, j) in W.
(5) such full Connection Neural Network is trained, and passes through the fractionation of the complex value of test data and matrix arrow
Quantization tests the deep learning network as input, to obtain the stabilizing network structure under each signal-to-noise ratio.
(6) in coherence time, for the condition of sparse channel under each signal-to-noise ratio, by the channel association side of the sparse information containing channel
Poor Matrix C [k] inputs network, can obtain corresponding digital estimator
Further, the channel estimation value in step 5It obtains as follows:
Wherein, ()HThe conjugate transposition operation of representing matrix;()-1The inversion operation of representing matrix.
As shown in Fig. 2, channel estimation methods proposed by the present invention are extremely close to theoretical optimal minimum under condition of sparse channel
Mean square error (MMSE) estimator, especially at low signal-to-noise ratio (SNR).Under condition of sparse channel, the present invention proposes to use depth
Learning network further improves the precision of channel estimation and reduces complexity.
The technical means disclosed in the embodiments of the present invention is not limited only to technological means disclosed in above embodiment, further includes
Technical solution consisting of any combination of the above technical features.It should be pointed out that for those skilled in the art
For, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also considered as
Protection scope of the present invention.
Claims (7)
1. a kind of sparse front channel estimation methods of millimeter wave based on deep learning network, which is characterized in that the method packet
Include following steps:
Step 1: base station design simulates precoding and pilot signal transmitted, simulates precoding FAmIt designs according to the following formula:
Wherein, []ijThe element that i-th row of representing matrix, jth arrange;FAmExpression dimension is Nt×NRFtSimulation pre-coding matrix,
NtIndicate transmitting antenna number, NRFtIndicate transmitting terminal rf chain number;Indicate the i-th row, the jth column member of simulation pre-coding matrix
The phase of element;
Step 2: receiving end optimization design simulated estimation device WAm,WAmIt calculates according to the following formula:
Wherein, WAmExpression dimension is Nr×NRFrSimulated estimation matrix, NrIndicate receiving antenna number, NRFrIndicate receiving end radio frequency
Number of links;Indicate the i-th row of simulated estimation matrix, the phase of jth column element;
Step 3: receiving end obtains the sparse features information of channel by compressed sensing technology, i.e., non-in condition of sparse channel matrix
Neutral element position, and in this, as the prior information of subsequent channel estimation;
Step 4: receiving end is for all possible selection matrix P corresponding to condition of sparse channel characteristic informationvThe set of [k] carries out
Storage, and according to each selection matrix PvThe corresponding digital estimated matrix of [k] design
Step 5: above-mentioned selection matrix PvThe set of [k] and digital estimated matrix correspondinglyAs full connection
The training set of neural network obtains the feed-forward network connected entirely being suitable under each signal-to-noise ratio by neural metwork training,
Receiving end is according to the corresponding selection matrix P of condition of sparse channelv[k], application network structure obtain the digital estimated matrix of network outputIt is detected with this data estimated matrix to signal is received, to obtain channel estimation value
2. the millimeter wave surface channel estimation methods according to claim 1 based on deep learning network, which is characterized in that
In the step 4, the sparse features information of channel, corresponding selection matrix P are obtained using compressed sensing technologyvUnder [k] is used
Face formula indicates:
Wherein, eπ(i)(π(i)∈{1,2,…,NrNt) expression dimension be NrNt× 1 a element of π (i) is 1, remaining element is
0 vector, NvIndicate that dimension is N on k-th of subcarrierrNt× 1 channel vector projects to the vector component h in angle domainv
The number of nonzero element in [k],
For the nonzero channel element in channel matrix, the selection matrix P of channelvThe form of [k] have N kind may, for it is all can
Set { the P of energyv1[k],Pv2[k],…,PvN[k] }, each PviA possibility that [k] (i=1 ..., N) occurs beI.e.
Wherein, N is all possible number,C indicates number of combinations formula.NvIndicate that dimension is on k-th of subcarrier
NrNt× 1 channel vector projects to the vector component h in angle domainvThe number of nonzero element in [k].
3. the sparse front channel estimation methods of the millimeter wave according to claim 2 based on deep learning network, feature
It is, channel is projected in virtual angle domain and obtains channel vector component hv[k], hv[k] is calculated as follows:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix,Representing matrix At's
Conjugation;Indicate Kronecker product;ArExpression dimension is Nr×NrReceive front response vector composition reception dictionary matrix;H
[k] indicates that the dimension on k-th of subcarrier is Nr×NtPhysical channel matrix, the vector of vec (H [k]) representing matrix H [k]
Change;
Wherein, AtIt indicates as follows:
Wherein,Expression dimension is Nt× 1 transmitting front response vector, whereinWherein, NtThe horizontal axis of=P × Q, P expression transmitting antenna front
Antenna number, Q indicate the longitudinal axis antenna number of transmitting antenna front;
ArIt indicates as follows:
Wherein,Expression dimension is Nr× 1 reception front response vector, whereinWherein, NrThe horizontal axis day of=I × J, I expression receiving antenna front
Line number, J indicate the longitudinal axis antenna number of receiving antenna front.
4. the millimeter wave surface channel estimation methods according to claim 3 based on deep learning network, which is characterized in that
Optimal number estimated matrix in the step 4Using Minimum Mean Square Error of Channel Estimation criterion,By as follows
Formula calculates:
Wherein, k ∈ { 1,2 ..., K } indicates that k-th of subcarrier, K indicate total number of sub-carriers;Indicate k-th of subcarrier
Upper NRFr×NvOptimal digital estimated matrix, M indicates channel access times, i.e. channel estimation number in coherence time;ηbIt indicates
Distortion factor related with analog-digital converter (ADC) quantizing bit number b;Ω [k] indicates that the dimension on k-th of subcarrier is MNRFr
×NvCalculation matrix;ΩHThe conjugate transposition of [k] representing matrix Ω [k];Indicate the large-scale fading coefficient of channel;Table
Show that dimension is Nv×NvUnit matrix;Indicate that equivalent noise vector is each
The variance of element,Indicate additive white Gaussian noise (AWGN) variance;P indicates transmitting pilot power.
5. the millimeter wave surface channel estimation methods according to claim 4 based on deep learning network, which is characterized in that
Channel estimation number M is indicated with following formula:
Expression rounds up;
The calculation matrix Ω [k] is indicated with following formula:
Ω [k]=Φ [k] Ψ Pv[k],
Wherein, dimension NRFr×NrNtPilot tone correlation matrixsm[k](m∈{1,
2 ..., M) indicate the m time it is trained when dimension be NRFr× 1 transmitting pilot tone vector.
Wherein, Ψ indicates that dimension is NrNt×NrNtSpace conversion matrix, Ψ are indicated as follows:
Wherein, AtExpression dimension is Nt×NtTransmitting front response vector composition transmitting dictionary matrix, ArExpression dimension is Nr
×NrReceive front response vector composition reception dictionary matrix;
Wherein, Pv[k] indicates that dimension is NrNt×NvSelection matrix.
6. the millimeter wave surface channel estimation methods according to claim 1 based on deep learning network, which is characterized in that
In step 5It is obtained by the fully-connected network of feed-forward, steps are as follows:
(1) nonzero element number N known for onevCondition of sparse channel, channel matrix H [k] indicate k-th of subcarrier on dimension
Degree is Nr×NtPhysical channel matrix, under each signal-to-noise ratio, receiving end calculates all possible P according to sparse features informationv
Covariance matrix C [the k]=Ω of [k] compositionH[k] Ω [k], wherein Ω [k]=Φ [k] Ψ Pv[k].Meanwhile according to claim
Method described in 4 calculates and each PvThe covariance matrix C [k] of [k] composition is correspondingDesign new number
Estimated matrixAnd by all possible C [k] and correspondingStorage
Into database;
(2) data are extracted from database, are divided into training data and two groups of test data, and training data is subjected to complex value and splits behaviour
Make, by training set C [k] and W 'D[k] is split as real-part matrix CR[k]、With imaginary-part matrix CI[k]、Two
Point;
(3) then by CR[k]、And CI[k]、Matrix vector operation is carried out, obtaining dimension is Nv 2× 1 column
Vector cR[k]、And cI[k]、By cR[k]、Input and instruction as real part deep learning network
Practice target, by cI[k]、Input and training objective as imaginary part deep learning network;
(4) fully-connected network of two deep learnings is constructed, network structure is identical, is all the full connection nerve of two layers of feed-forward
Network, first layer neuron number are N, whereinBiasing (bias) connection, and first are set in first layer simultaneously
The transfer function of layer is set as softmax function;First layer is exported and is connect with second layer neuron, neuron number N, and
And biasing (bias) connection is equally set in the second layer;
Softmax function is defined as follows:
Wherein, the output of each layer of fully-connected network such as following formula:
Wherein, W and b indicates the parameter of full Connection Neural Network, yiAnd biIndicate i-th of element of y and b, xjIndicate j-th of x
Element, Wi,jIndicate that position is the element of (i, j) in W;
(5) such full Connection Neural Network is trained, and passes through the fractionation of the complex value of test data and matrix vector
The deep learning network is tested as input, to obtain the stabilizing network structure under each signal-to-noise ratio;
(6) in coherence time, for the condition of sparse channel under each signal-to-noise ratio, by the channel covariancc square of the sparse information containing channel
Battle array C [k] inputs network, can obtain corresponding digital estimator
7. the millimeter wave surface channel estimation methods according to claim 1 based on deep learning network, which is characterized in that
Channel estimation value in step (5)It is calculated as follows:
Wherein, ()HThe conjugate transposition operation of representing matrix;()-1The inversion operation of representing matrix.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910397076.5A CN110099016B (en) | 2019-05-14 | 2019-05-14 | Millimeter wave sparse array surface channel estimation method based on deep learning network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910397076.5A CN110099016B (en) | 2019-05-14 | 2019-05-14 | Millimeter wave sparse array surface channel estimation method based on deep learning network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110099016A true CN110099016A (en) | 2019-08-06 |
CN110099016B CN110099016B (en) | 2022-05-31 |
Family
ID=67447898
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910397076.5A Active CN110099016B (en) | 2019-05-14 | 2019-05-14 | Millimeter wave sparse array surface channel estimation method based on deep learning network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110099016B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110519188A (en) * | 2019-08-20 | 2019-11-29 | 电子科技大学 | A kind of compressed sensing based multi-user's time-varying millimeter wave channel estimation methods |
CN110535500A (en) * | 2019-09-03 | 2019-12-03 | 电子科技大学 | A kind of millimeter wave MIMO mixed-beam forming optimization method based on deep learning |
CN111713035A (en) * | 2020-04-07 | 2020-09-25 | 东莞理工学院 | MIMO multi-antenna signal transmission and detection technology based on artificial intelligence |
CN111835351A (en) * | 2020-06-01 | 2020-10-27 | 东南大学 | Deep learning-based quantitative signal reconstruction method and arrival angle estimation method |
CN112187323A (en) * | 2020-09-29 | 2021-01-05 | 国网江苏省电力有限公司丹阳市供电分公司 | IRS-based large-scale MIMO (multiple input multiple output) cascade channel estimation method under mixed low-precision architecture |
CN112615801A (en) * | 2020-12-16 | 2021-04-06 | 西安交通大学 | Channel estimation method, medium, and apparatus based on compressed sensing and deep learning |
CN113193893A (en) * | 2021-04-30 | 2021-07-30 | 东南大学 | Millimeter wave large-scale MIMO intelligent hybrid beam forming design method |
CN113972939A (en) * | 2021-09-09 | 2022-01-25 | 浙江大学 | Antenna system precoding method and device based on double time scales and deep learning |
CN114584448A (en) * | 2022-02-16 | 2022-06-03 | 山东大学 | SM-OFDM signal grouping detection method based on deep neural network |
CN114759956A (en) * | 2022-04-13 | 2022-07-15 | 东南大学 | Single-bit ADC uplink multi-user MIMO deep unfolding pre-coding method |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107566303A (en) * | 2017-07-27 | 2018-01-09 | 东华大学 | A kind of millimeter wave channel estimation methods based on Bayes's compressed sensing |
US9924880B2 (en) * | 2015-02-11 | 2018-03-27 | Samsung Electronics Co., Ltd. | RF doppler bio-signal sensor for continuous heart rate variability and blood pressure monitoring |
WO2018093411A2 (en) * | 2016-11-16 | 2018-05-24 | Intel IP Corporation | Station (sta), access point (ap) and methods for beam refinement and related signaling |
CN108881074A (en) * | 2018-05-08 | 2018-11-23 | 东南大学 | Broadband millimeter-wave channel estimation methods under a kind of low precision mixed architecture |
CN109005133A (en) * | 2018-07-12 | 2018-12-14 | 南京邮电大学 | Double Sparse multi-path channel models and the channel estimation methods based on this model |
WO2019014264A1 (en) * | 2017-07-12 | 2019-01-17 | Intel IP Corporation | Enhanced frame format for single user 2.16 ghz channel wireless communications |
CN109743268A (en) * | 2018-12-06 | 2019-05-10 | 东南大学 | Millimeter wave channel estimation and compression method based on deep neural network |
-
2019
- 2019-05-14 CN CN201910397076.5A patent/CN110099016B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9924880B2 (en) * | 2015-02-11 | 2018-03-27 | Samsung Electronics Co., Ltd. | RF doppler bio-signal sensor for continuous heart rate variability and blood pressure monitoring |
WO2018093411A2 (en) * | 2016-11-16 | 2018-05-24 | Intel IP Corporation | Station (sta), access point (ap) and methods for beam refinement and related signaling |
WO2019014264A1 (en) * | 2017-07-12 | 2019-01-17 | Intel IP Corporation | Enhanced frame format for single user 2.16 ghz channel wireless communications |
CN107566303A (en) * | 2017-07-27 | 2018-01-09 | 东华大学 | A kind of millimeter wave channel estimation methods based on Bayes's compressed sensing |
CN108881074A (en) * | 2018-05-08 | 2018-11-23 | 东南大学 | Broadband millimeter-wave channel estimation methods under a kind of low precision mixed architecture |
CN109005133A (en) * | 2018-07-12 | 2018-12-14 | 南京邮电大学 | Double Sparse multi-path channel models and the channel estimation methods based on this model |
CN109743268A (en) * | 2018-12-06 | 2019-05-10 | 东南大学 | Millimeter wave channel estimation and compression method based on deep neural network |
Non-Patent Citations (2)
Title |
---|
YUCHENG WANG: "Wideband mmWave Channel Estimation for Hybrid Massive MIMO With Low-Precision ADCs", 《IEEE》 * |
杜若宸: "基于毫米波的5G通信定位一体化***", 《通信设计与应用》 * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110519188A (en) * | 2019-08-20 | 2019-11-29 | 电子科技大学 | A kind of compressed sensing based multi-user's time-varying millimeter wave channel estimation methods |
CN110535500A (en) * | 2019-09-03 | 2019-12-03 | 电子科技大学 | A kind of millimeter wave MIMO mixed-beam forming optimization method based on deep learning |
CN110535500B (en) * | 2019-09-03 | 2021-09-21 | 电子科技大学 | Millimeter wave MIMO mixed beam forming optimization method based on deep learning |
CN111713035A (en) * | 2020-04-07 | 2020-09-25 | 东莞理工学院 | MIMO multi-antenna signal transmission and detection technology based on artificial intelligence |
CN111835351B (en) * | 2020-06-01 | 2023-07-28 | 东南大学 | Quantized signal reconstruction method and arrival angle estimation method based on deep learning |
CN111835351A (en) * | 2020-06-01 | 2020-10-27 | 东南大学 | Deep learning-based quantitative signal reconstruction method and arrival angle estimation method |
CN112187323A (en) * | 2020-09-29 | 2021-01-05 | 国网江苏省电力有限公司丹阳市供电分公司 | IRS-based large-scale MIMO (multiple input multiple output) cascade channel estimation method under mixed low-precision architecture |
CN112615801A (en) * | 2020-12-16 | 2021-04-06 | 西安交通大学 | Channel estimation method, medium, and apparatus based on compressed sensing and deep learning |
CN113193893A (en) * | 2021-04-30 | 2021-07-30 | 东南大学 | Millimeter wave large-scale MIMO intelligent hybrid beam forming design method |
CN113972939A (en) * | 2021-09-09 | 2022-01-25 | 浙江大学 | Antenna system precoding method and device based on double time scales and deep learning |
CN113972939B (en) * | 2021-09-09 | 2022-07-12 | 浙江大学 | Antenna system precoding method and device based on double time scales and deep learning |
WO2023035736A1 (en) * | 2021-09-09 | 2023-03-16 | 浙江大学 | Antenna system precoding method and apparatus based on two time scales and deep learning |
CN114584448A (en) * | 2022-02-16 | 2022-06-03 | 山东大学 | SM-OFDM signal grouping detection method based on deep neural network |
CN114584448B (en) * | 2022-02-16 | 2023-09-22 | 山东大学 | SM-OFDM signal grouping detection method based on deep neural network |
CN114759956A (en) * | 2022-04-13 | 2022-07-15 | 东南大学 | Single-bit ADC uplink multi-user MIMO deep unfolding pre-coding method |
CN114759956B (en) * | 2022-04-13 | 2024-01-30 | 东南大学 | Single-bit ADC uplink multi-user MIMO deep spreading precoding method |
Also Published As
Publication number | Publication date |
---|---|
CN110099016B (en) | 2022-05-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110099016A (en) | A kind of sparse front channel estimation methods of millimeter wave based on deep learning network | |
CN108933745A (en) | A kind of broad-band channel estimation method estimated based on super-resolution angle and time delay | |
Yan et al. | A dynamic array-of-subarrays architecture and hybrid precoding algorithms for terahertz wireless communications | |
CN107135024B (en) | Low-complexity hybrid beam forming iterative design method | |
CN106559367B (en) | MIMO-OFDM system millimeter waves channel estimation methods based on low-rank tensor resolution | |
CN110401476B (en) | Codebook-based millimeter wave communication multi-user parallel beam training method | |
CN109743268A (en) | Millimeter wave channel estimation and compression method based on deep neural network | |
CN108964726A (en) | A kind of extensive MIMO uplink transmission channels estimation method of low complex degree | |
CN110213185B (en) | Three-dimensional channel parameter estimation method based on atomic norm minimization | |
CN106059972A (en) | Modulation identification method under MIMO related channel based on machine learning algorithm | |
CN101222458B (en) | Low-level recursion minimum mean-square error evaluation of MIMO-OFDM channel | |
CN106571858B (en) | Hybrid beam forming transmission system | |
CN108199753A (en) | A kind of method for precoding based on iteration minimum in millimetre-wave attenuator | |
CN108599825A (en) | A kind of hybrid coding method based on MIMO-OFDM millimeters of wave structures | |
CN109302224A (en) | Mixed-beam forming algorithm for extensive MIMO | |
CN107634787A (en) | A kind of method of extensive MIMO millimeter wave channel estimations | |
CN107566305A (en) | A kind of millimeter-wave systems channel estimation methods of low complex degree | |
CN106603130A (en) | Mixed digital-analog pre-coding method for large-scale MIMO system | |
CN109379122A (en) | A kind of millimetre-wave attenuator multipath channel dynamic beam training method | |
CN103929396B (en) | processing method of MIMO-OFDM system downlink information data | |
CN106921422A (en) | For the method and system of the channel estimation of the mixed-beam shaping in MMW channels | |
CN108881074A (en) | Broadband millimeter-wave channel estimation methods under a kind of low precision mixed architecture | |
CN109412983A (en) | A kind of extensive mimo channel algorithm for estimating of mesh freeization based on the domain DFT | |
CN109347529B (en) | Channel estimation and hybrid beam forming method for resisting non-ideality of phase shifter | |
CN114567525A (en) | Channel estimation method and device |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |