CN110113088A - A kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method - Google Patents
A kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method Download PDFInfo
- Publication number
- CN110113088A CN110113088A CN201910376026.9A CN201910376026A CN110113088A CN 110113088 A CN110113088 A CN 110113088A CN 201910376026 A CN201910376026 A CN 201910376026A CN 110113088 A CN110113088 A CN 110113088A
- Authority
- CN
- China
- Prior art keywords
- signal
- arrival
- submatrix
- divergence type
- model
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/391—Modelling the propagation channel
- H04B17/3913—Predictive models, e.g. based on neural network models
-
- 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
- H04B7/0456—Selection of precoding matrices or codebooks, e.g. using matrices antenna weighting
-
- 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
- H04B7/06—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station
- H04B7/0613—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission
- H04B7/0615—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal
- H04B7/0617—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the transmitting station using simultaneous transmission of weighted versions of same signal for beam forming
-
- 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
- H04B7/08—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the receiving station
- H04B7/0837—Diversity systems; Multi-antenna system, i.e. transmission or reception using multiple antennas using two or more spaced independent antennas at the receiving station using pre-detection combining
- H04B7/0842—Weighted combining
- H04B7/086—Weighted combining using weights depending on external parameters, e.g. direction of arrival [DOA], predetermined weights or beamforming
Abstract
The invention discloses a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation methods; signal source emits signal in this method; receiving end uses divergence type numerical model analysis antenna frame, while entire technical solution is divided into two processes: off-line training process and angle the prediction process of model.Compared with prior art, scheme provided by the invention takes full advantage of the room and time resource of system, in the case where any prior information of unknown incoming wave, accurate DOA estimate in the wide scope of low complex degree can be achieved, while solving the problems, such as phase ambiguity present in traditional algorithm.
Description
Technical field
The present invention relates to wireless communication technology fields, more particularly to a kind of divergence type numerical model analysis antenna system direction of arrival
Intelligent estimation method.
Background technique
The fast development of wireless communication technique profoundly changes raw the mode of production and life, has greatly pushed society
The change and innovation of meeting, become the indispensable a part in the world today.MIMO technique (Multiple Input
And Multiple Output, MIMO) it is the important breakthrough for wirelessly communicating smart antenna field, bandwidth can not increased
In the case of double up the capacity of communication system.With flourishing and mobile data amount, access device for mobile Internet
The sharp increase of quantity, how further satisfaction wireless communication sustainable growth rate requirement, become future mobile and face
Critical issue.
Due to the shortage of low-frequency spectra resource in existing cellular system, implements wireless communication using millimeter wave frequency band and attract
The extensive concern and research of academia and industry.It is limited by the biggish radio wave propagation loss of millimeter wave frequency band, millimeter wave is wireless
The research of transmission technology lays particular emphasis on short haul connection scene mostly.But it is big to advise in view of millimeter wave frequency band upper ripple length is relatively short
Mould aerial array can be installed to base station and user side simultaneously.In turn, beam forming gain provided by large-scale antenna array
It can compensate for propagation loss relatively high on millimeter wave frequency band.In view of pure digi-tal beam-forming method bring power consumption burden and
Performance loss caused by pure analog beam manufacturing process, numerical model analysis beam-forming method have become research hotspot.
The sparse characteristic for benefiting from millimeter wave propagation channel estimates that the angle information of propagation channel would be beneficial for millimeter wave biography
The design of broadcast mode and the expense for reducing acquisition channel information.However, unique numerical model analysis antenna frame in millimeter-wave systems
Traditional direction of arrival (Direction of Arrival, DOA) algorithm for estimating is substantially increased with large-scale antenna array
Complexity, such as multiple signal classification algorithm (Multiple Signal Classification, MUSIC) and by invariable rotary
Technology parameter estimation algorithm (Estimating Signal Via Rotational Invariance Techniques,
ESPRIT), and these traditional algorithms have the defects that a universal i.e. phase ambiguity problem, this is because between adjacent submatrix
After being greater than half-wavelength, the integral multiple for 2 π that will appear during solving antitrigonometric function.Therefore, based on mixed using digital-to-analogue
The divergence type submatrix system for closing antenna frame, comprehensively considers the demand of computation complexity and estimated accuracy, the invention proposes
A kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method and system and device, be used in combination space resources and
Time resource.Compared with traditional algorithm, algorithm complexity proposed by the invention is lower, and accuracy is higher.
Summary of the invention
In order to solve problem above, the present invention provides a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation
Method, this method implementation complexity is low, since the training process of model is completed offline, required angle predicted time
The problem of short, while there is no phase ambiguities.It is intelligent that the present invention provides a kind of divergence type numerical model analysis antenna system direction of arrival
Estimation method, which is characterized in that the system and device of the divergence type numerical model analysis antenna system direction of arrival intelligence estimation method
Include: signal source emits signal and receiving end using divergence type numerical model analysis beamforming architectures；
The divergence type numerical model analysis antenna system direction of arrival intelligence estimation method includes the off-line training process of model
Two processes of process are predicted with angle；
The off-line training process of the model is as follows；
Entire model training process is completed under off-line state, is first divided entire space according to code book collection multiple
Subregion；Then for any spatial sub-area, training sample set is constructed；Then it is carried out using proposed new neural network
Training, finally obtains the direction of arrival prediction model of multiple and different subregions；
The angle prediction process of the model is as follows；
Firstly, each submatrix in receiving end selects different code words to receive from unknown bearing signal from known reception code book
The transmitting signal in source selects the corresponding reception strongest code word of signal energy, and then orients the subregion where incoming wave signal；
Then, all submatrixs are all made of choose corresponding and receive the strongest code word of signal energy to receive signal, and from each submatrix
Reception signal in extract the characteristic vector comprising incoming wave sense；Then the characteristic vector extracted is input to offline
In the direction of arrival prediction model of the trained correspondence subregion, further fining direction of arrival is estimated.
As a further improvement of that present invention, the new neural network framework is referred to as " MircoFishNet ", described
Seemingly, the MircoFishNet character is that front head is relatively narrow, and intermediate torso part is first wide for MircoFishNet shape and fish
Narrow afterwards, last wider part is similar to the tail of fish；Meanwhile whole network framework does not need too many layer and can be achieved with comparing
Good estimated performance, the MircoFishNet network architecture include four parts, are narrow network layer first, purpose
It is the thick feature i.e. global characteristics for extracting input data；Followed by wider network layer, the purpose is to sufficiently extract input data
Local feature；Followed by relatively narrow network layer, main purpose are to reduce calculation amount；It is finally wider network, purpose
It is the further feature for sufficiently extracting data again.
As a further improvement of that present invention, the receiving end includes by mould using divergence type numerical model analysis beamforming architectures
Quasi- beam forming, radio frequency link, analog-digital conversion, baseband digital signal handle the receiving module of composition, using neural network
The off-line training module and intelligent direction of arrival prediction module, the receiving end for constructing model are equipped with P × Q subarray,
P submatrix of middle horizontal direction, Q submatrix of vertical direction, each subarray is by M × N root antenna, wherein horizontal direction M root antenna,
Vertical direction N root antenna) composition, for analog beam forming, each submatrix connects every antenna one independent phase-shifter of connection
Connect a radio frequency link.
As a further improvement of that present invention, specific step is as follows for the off-line training process of the model；
Step 1: dividing area of space；
According to known reception code bookSpatial dimension is divided into K sub-regions,
WhereinWithThe vertical phase and horizontal phase of k-th of code word are respectively indicated, K is the number of code word, it is clear that each sub-district
Domain corresponds to certain Vertical Square parallactic angle and level orientation angular region, this is that the main lobe width for corresponding to wave beam by each code word determines
's；
Step 2: generating training sample set；
K sub-regions need to generate K training sample set；
By taking k-th of subregion as an example, R direction sample is randomly generated in the spatial dimension, i.e.,WhereinIndicate r-th of side in k-th of subregion
To the Vertical Square parallactic angle and horizontal azimuth of sample；
For r-th of direction sample, signal source is located atPlace's transmitting signal, each submatrix in receiving end are all made of k-th
Code word receives signal, and the real and imaginary parts that each submatrix receives signal are extracted and are spliced into a vectorBy the vector
As characterization direction sampleCharacteristic vector.Finally obtain the training sample set of k-th of subregion For the characteristic vector of r-th of direction sample,For r-th of direction sample
This label；
Remaining K-1 sub-regions generates respective training sample set after the same method；
Step 3: off-line training model；
It is instructed offline using the training sample set in the region using intelligentized machine learning algorithm for each subregion
Practise the direction of arrival prediction model suitable for one's respective area；
By taking k-th of subregion as an example, by training sample set obtained in step 2
Itd is proposed new neural network is inputted, the parameter of each neuron of training obtains the Vertical Square parallactic angle suitable for this subregion
With horizontal azimuth prediction model；
Remaining K-1 sub-regions trains Vertical Square parallactic angle and level suitable for respective region after the same method
Azimuth prediction model.
As a further improvement of that present invention, specific step is as follows for the angle prediction process of the model；
Step 1: beam scanning, the range of coarse localization incoming wave sense；
Utilize known reception code bookEach submatrix selects mutually different code word
It forms transmitting signal of the wave beam reception from unknown orientation signal source and corresponding reception is selected according to the reception signal of each submatrix
The strongest code word of signal energyAnd then it orients incoming wave sense and is located at kth*In sub-regions；
Step 2: extracting the direction character vector x of incoming wave signal；
All submatrixs are all made of the corresponding reception strongest code word of signal energy chosen in step 1Make
To receive wave beam, and the real and imaginary parts that each submatrix receives signal are extracted and are spliced into a vector as incoming wave signal
Direction character vector x；
Step 3: the angle of fining estimation incoming wave signal；
The direction character vector x in the unknown signaling source obtained in step 2 is input to the kth good through off-line training*Height
In the Vertical Square parallactic angle and horizontal azimuth prediction model in region, the angle of incoming wave signal is further estimated to fining.
A kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method of the application, its advantages are as follows:
Divergence type numerical model analysis antenna system direction of arrival intelligence estimation method proposed by the present invention, including the offline of model
Training process and intelligent angle predict process.Compared with traditional direction of arrival estimation method such as MUSIC, ESPRIT algorithm, this hair
The direction of arrival intelligence estimation method computation complexity of bright proposition is low, can effectively solve the problems, such as phase ambiguity simultaneously, for big rule
Transmission plan under mould divergence type numerical model analysis antenna system framework provides useful information, more with practical value.
Detailed description of the invention
Fig. 1 is divergence type numerical model analysis antenna system direction of arrival intelligence estimation method flow chart；
Fig. 2 is the off-line training procedure chart of model；
Fig. 3 is that intelligent angle predicts procedure chart；
Fig. 4 is to receive end face battle array architecture diagram；
Fig. 5 be MicroFishNet neural network simplified schematic diagram, wherein (a) be MicroFishNet schematic diagram, (b) be
MicroFishNet neuron connection figure；
Fig. 6 is the MircoFishNet neural network architecture diagram used in example；
Angle estimation root-mean-square error that Fig. 7 is sampling number when being 1 with signal-to-noise ratio change curve (MUSIC algorithm and this
Intelligent estimation method proposed in patent)；
Fig. 8 is that 0dB lower angle estimates root-mean-square error with the change curve of sampling number (in MUSIC algorithm and this patent
The intelligent estimation method proposed)；
Fig. 9 is that 10dB lower angle estimates root-mean-square error with the change curve of sampling number (in MUSIC algorithm and this patent
The intelligent estimation method proposed).
Specific embodiment
Present invention is further described in detail with specific embodiment with reference to the accompanying drawing:
The present invention provides a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method, this method is realized
Complexity is low, and since the training process of model is completed offline, required angle predicted time is short, while phase is not present
The problem of position ambiguity.
The embodiment of the invention discloses a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method, this sides
Method is divided into two processes: off-line training process and angle the prediction process of model, as shown in Figure 1.The off-line training process of model
It is: entire space is divided by multiple subregions according to code book collection first；Then for each spatial sub-area, training sample is constructed
Collection；Then it is trained using proposed new neural network, finally obtains the direction of arrival prediction mould of multiple and different subregions
Type.Angle prediction process is: firstly, each submatrix in receiving end selects different code words to receive from not from known reception code book
The transmitting signal for knowing bearing signal source selects the corresponding reception strongest code word of signal energy, and then orients incoming wave signal institute
Subregion；Then, all submatrixs are all made of choose corresponding and receive the strongest code word of signal energy to receive signal,
And the characteristic vector comprising incoming wave sense is extracted from the reception signal of each submatrix；Then the characteristic vector that will be extracted
It is input in the direction of arrival prediction model of the good correspondence subregion of off-line training, further estimates direction of arrival to fining.
Compared with prior art, scheme provided by the invention takes full advantage of the room and time resource of system, appoints in unknown incoming wave
, it can be achieved that accurate DOA estimate in the wide scope of low complex degree, efficiently solves simultaneously in the case where what prior information
Present in traditional algorithm the problem of phase ambiguity.
The array response of receiving plane battle array according to the present invention is described below in content to facilitate the understanding of the present invention
The prior arts relevant knowledges such as vector, face horizontal and vertical parity check code word.It should be noted that method of the invention is not only restricted to lower mask body formula
Representation.
As shown in figure 4, multi-panel array antenna is put in xoy plane, wherein Vertical Square parallactic angle and horizontal azimuth use θ respectively
WithIt indicates, upper along the x-axis direction to place P submatrix, the distance between adjacent two submatrixs respective antenna element is Dx；Along the y-axis direction
Upper Q submatrix of placement, the distance between adjacent two submatrixs respective antenna element are Dy.In every sub- face battle array, along the x-axis direction on put
M root antenna is set, the distance between adjacent antenna element is dx；It goes up along the y-axis direction and places N root antenna, between adjacent antenna element
Distance be dy.Using the antenna element of coordinate origin as reference point, the array response vector of pth row q column virgin face battle array can table
It is shown as:
Wherein, ()TIndicate transposition operation,λ indicates that signal carries
Wave wavelength,Indicate Kronecker Product operation.
According to face battle array receiving array response vector, the design of face horizontal and vertical parity check code word be may be expressed as:
Wherein,WithRespectively indicate the vertical phase and horizontal phase of k-th of code word.
Neural network is by input layer (input layer), hidden layer (hidden layer) and output layer (output
Layer) form, complete wherein hidden layer according to function and neuron connection type Bu Tong and be divided into articulamentum, active coating and with
Machine disconnection layer etc..
Each node and upper one of full articulamentum (Fully connected dence layer, Dense layers of abbreviation)
All nodes of layer are connected, for the characteristic synthetic that front is extracted.
In neural network, activation primitive (Activation Function) is the function run on neuron, is responsible for
The input of neuron is mapped to output end, the purpose is to increase the non-linear of neural network model.Relu function
(Recitified Linear Unit, Relu), also known as line rectification function, expression formula are f (x)=max (0, x).
Random disconnection layer is also known as Dropout layers and refers in the training process of neural network, for neural unit, according to one
Fixed probability temporarily abandons it from network, and the purpose is to prevent over-fitting.
Based on above-mentioned technical backgrounder, a kind of divergence type numerical model analysis antenna system wave disclosed by the embodiments of the present invention reaches
Intelligence estimation method in angle obtains the approximate range of arrival bearing by beam scanning first, further good using off-line training
Direction of arrival prediction model intelligently estimate arrival bearing.
In the embodiment of the present invention, it is assumed that carrier frequency 60GHz, signal source send unknown signaling, to its structure without spy
Provisioning request.Receiving end use divergence type numerical model analysis beamforming architectures, mainly include by analog beam forming, radio frequency link,
The receiving modules of compositions such as analog-digital conversion, baseband digital signal processing, the off-line training that model is constructed using neural network
Module and intelligent direction of arrival prediction module.Receiving end is equipped with 8 × 8 subarrays, and each subarray is equipped with 4 × 4 antennas,
Every antenna connects an independent phase-shifter and shapes for analog beam, and each subarray connects a radio frequency link.It is different
The device of different antennae number can be obtained by the example modified in the present embodiment on submatrix columns, each subarray.
The embodiment of the invention discloses a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation methods, specifically
Steps are as follows:
The off-line training process of model:
Step 1: dividing area of space.
In this example receiving end use DFT code book, it should be noted that this patent to receive code book with no restriction.Horizontal DFT
The l of code bookhA code word can indicate are as follows:
Wherein, the number of antennas M=4, L in each submatrix in the horizontal directionhFor the number of horizontal direction DFT code word, this
In Lh=2M, the spacing of two antennas in the horizontal direction in submatrixλ is carrier wavelength, and j is imaginary unit,The l of vertical DFT code bookvA code word can indicate are as follows:
Wherein, the number of antennas N=4, L in each submatrix verticallyvFor the number of vertical direction DFT code word, here
Lv=2N, the vertically spacing of two antennas in each submatrixAt this point, the three-dimensional DFT code book of receiving end planar array
ForWherein k=(lh-1)Lv+lv,K=LhLv.It is aobvious
So, the vertical phase of k-th of code wordAnd horizontal phaseMeet
According to known reception code bookSpatial dimension is divided into K=64 sub-district
Domain.Obviously, each subregion corresponds to certain Vertical Square parallactic angle and level orientation angular region, this is to correspond to wave beam by each code word
Main lobe width determine.
Step 2: generating training sample set.K sub-regions need to generate K training sample set.
By taking k-th of subregion as an example, R=10000 direction sample is randomly generated in the spatial dimension, i.e.,WhereinIndicate r-th of side in k-th of subregion
To the Vertical Square parallactic angle and horizontal azimuth of sample.
For r-th of direction sample, i.e. signal source is located atPlace's transmitting signal, each submatrix in receiving end are all made of kth
A code wordSignal is received, then the reception signal of PQ submatrix of first of sampling instant can be expressed asWherein P=8 is the submatrix number of receiving end horizontal direction, Q
=8 be the submatrix number of receiving end vertical direction；When being located at r-th of direction sample for signal source in k-th of subregion,
Reception signal of the receiving end pth row q column virgin battle array in first of sampling instant.
Next, averagely being obtained to L sampling instantIt extracts each
The real part of submatrix averaged Received Signal forms vectorAnd imaginary part forms vector
The vector that real and imaginary parts are splicedAs characterization direction sampleCharacteristic vector, wherein
Real () indicates to extract the operation of real part, the operation of imag () expression extraction imaginary part.
Remaining R-1 direction sample obtains characterizing after the same method the characteristic vector in respective direction, and then obtains kth
The training sample set of sub-regions For r-th of direction sample in k-th of subregion
This characteristic vector,For the label of r-th of direction sample.
Remaining K-1 sub-regions generates respective training sample set after the same method.
Step 3: each subregion is directed to, using the training sample set in the region, using the new neural network proposed
It is trained, finally obtains the direction of arrival prediction model of multiple and different subregions.
By taking k-th of subregion as an example, by training sample set obtained in step 2
As input data, pass through MircoFishNet neural network off-line training model.The MircoFishNet used in this example
Neural network framework in the network architecture six full articulamentums as shown in fig. 6, be made of, the number of every layer of neuron is respectively
128,512,256,256,256,512.It is respectively connected with active coating behind each full articulamentum, activation primitive selects Relu function；
And Dropout layers are connected after each active coating, the ratio disconnected at random is 0.5.Two in the last layer of network model
Neuron exports Vertical Square parallactic angle θ and horizontal azimuth
Remaining K-1 sub-regions trains Vertical Square parallactic angle and level suitable for respective region after the same method
Azimuth prediction model.
Angle predicts process:
Step 1: beam scanning, the range of coarse localization incoming wave sense.
Utilize three-dimensional DFT code bookEach submatrix therefrom selects mutually different code word
It forms wave beam and receives the transmitting signal from unknown orientation signal source.It can be scanned simultaneously since receiving end shares PQ=64 submatrix
64 different directions, that is, different submatrixs select the wave beam of different code word formation pointing space different directions, thus can
Beam scanning is carried out using 64 code words received in code book simultaneously.According to the reception signal of each submatrix, corresponding reception is selected
The strongest code word of signal energyAnd then it orients incoming wave sense and is located at kth*In sub-regions.
Step 2: extracting the direction character vector x of incoming wave signal.
All submatrixs are all made of the corresponding reception strongest code word of signal energy chosen in step 1Shape
Signal y (l)=[y is received at wave beam1,1(l),y1,2(l),…,y1,Q(l),y2,1(l),…,yP,Q(l)], then by each submatrix
Signal is received to be averaging about sampling numberFinally by the reality of each submatrix averaged Received Signal
PortionAnd imaginary partExtract the direction character for being spliced into a vector as incoming wave signal
Vector
Step 3: the angle of fining estimation incoming wave signal.
The direction character vector x in the unknown signaling source obtained in step 2 is input to the kth good through off-line training*Height
In the direction of arrival prediction model in region, the angle of incoming wave signal is further estimated to fining.
In order to illustrate a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method proposed by the present invention
Accuracy and validity, embodiment additionally provide the root mean square of the DOA estimate under traditional MUSIC algorithm and Intelligentized method
Error (Root Mean Squared Error, RMSE) and the relationship of signal-to-noise ratio (Signal Noise Ratio, SNR) emulate
Figure, as shown in fig. 7, sampled point number L=1 at this time.It can be seen from figure 7 that the performance of intelligent angle estimating method wants excellent
In traditional MUSIC algorithm.In addition to this, embodiment, which is additionally provided, estimates in traditional MUSIC algorithm with direction of arrival under Intelligentized method
The root-mean-square error of meter and the relationship analogous diagram of sampled point number L, as shown in Figure 8 and Figure 9, wherein signal-to-noise ratio is SNR=in Fig. 8
Signal to Noise Ratio (SNR)=10dB in 0dB, Fig. 9.From Fig. 8 and Fig. 9 as can be seen that as sampling number purpose gradually increases, MUSIC
The estimated accuracy of algorithm and Intelligentized method is all stepping up, but in general, intelligent angle proposed by the invention
The performance of estimation method will be far superior to traditional algorithm.
The above described is only a preferred embodiment of the present invention, being not the limit for making any other form to the present invention
System, and made any modification or equivalent variations according to the technical essence of the invention, still fall within present invention model claimed
It encloses.
Claims (5)
1. a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method, which is characterized in that the divergence type number
The system and device of mould hybrid antenna system direction of arrival intelligence estimation method includes: signal source emits signal and receiving end uses and divides
Release numerical model analysis beamforming architectures；
The divergence type numerical model analysis antenna system direction of arrival intelligence estimation method includes off-line training process and the angle of model
Spend two processes of prediction process；
The off-line training process of the model is as follows；
Entire model training process is completed under off-line state, and entire space is divided multiple sub-districts according to code book collection first
Domain；Then for any spatial sub-area, training sample set is constructed；Then it is instructed using proposed new neural network
Practice, finally obtains the direction of arrival prediction model of multiple and different subregions；
The angle prediction process of the model is as follows；
Firstly, each submatrix in receiving end selects different code words to receive from unknown orientation signal source from known reception code book
Emit signal, selects the corresponding reception strongest code word of signal energy, and then orient the subregion where incoming wave signal；It connects
, all submatrixs are all made of choose corresponding and receive the strongest code word of signal energy to receive signal, and from each submatrix
It receives in signal and extracts the characteristic vector comprising incoming wave sense；Then the characteristic vector extracted is input to offline instruction
In the direction of arrival prediction model for the correspondence subregion perfected, further fining direction of arrival is estimated.
2. a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method according to claim 1, special
Sign is: the new neural network framework is referred to as " MircoFishNet ", the MircoFishNet shape and fish seemingly,
The MircoFishNet character is that front head is relatively narrow, and narrow after intermediate torso part is first wide, last wider part is similar to
The tail of fish；Meanwhile whole network framework does not need too many layer can be achieved with relatively good estimated performance, it is described
The MircoFishNet network architecture includes four parts, is narrow network layer first, the purpose is to extract input data
Thick feature, that is, global characteristics；Followed by wider network layer, the purpose is to sufficiently extract the local feature of input data；Followed by
Relatively narrow network layer, main purpose are to reduce calculation amount；It is finally wider network, the purpose is to sufficiently extract data again
Further feature.
3. a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method according to claim 1, special
Sign is: the receiving end includes by analog beam forming, radio frequency link, mould using divergence type numerical model analysis beamforming architectures
Quasi--number conversion, baseband digital signal handle the receiving module of composition, using the off-line training module of neural network building model
And intelligent direction of arrival prediction module, the receiving end are equipped with P × Q subarray, wherein P submatrix of horizontal direction, vertically
Q, direction submatrix, each subarray is by M × N root antenna, wherein horizontal direction M root antenna, vertical direction N root antenna) composition,
Every antenna connects an independent phase-shifter and shapes for analog beam, and each submatrix connects a radio frequency link.
4. a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method according to claim 1, special
Sign is: specific step is as follows for the off-line training process of the model；
Step 1: dividing area of space；
According to known reception code bookSpatial dimension is divided into K sub-regions, whereinWithThe vertical phase and horizontal phase of k-th of code word are respectively indicated, K is the number of code word, it is clear that each subregion pair
Certain Vertical Square parallactic angle and level orientation angular region are answered, this is that the main lobe width for corresponding to wave beam by each code word determines；
Step 2: generating training sample set；
K sub-regions need to generate K training sample set；
By taking k-th of subregion as an example, R direction sample is randomly generated in the spatial dimension, i.e.,WhereinIndicate r-th of side in k-th of subregion
To the Vertical Square parallactic angle and horizontal azimuth of sample；
For r-th of direction sample, signal source is located atPlace's transmitting signal, each submatrix in receiving end are all made of k-th of code word and receive
Signal, and the real and imaginary parts that each submatrix receives signal are extracted and are spliced into a vectorUsing the vector as characterization direction sample
ThisCharacteristic vector.Finally obtain the training sample set of k-th of subregion For the characteristic vector of r-th of direction sample,For the label of r-th of direction sample；
Remaining K-1 sub-regions generates respective training sample set after the same method；
Step 3: off-line training model；
Gone out using the training sample set in the region using intelligentized machine learning algorithm off-line training for each subregion
Direction of arrival prediction model suitable for one's respective area；
By taking k-th of subregion as an example, by training sample set obtained in step 2It is defeated
Enter proposed new neural network, the parameter of each neuron of training, obtain suitable for this subregion Vertical Square parallactic angle and
Horizontal azimuth prediction model；
Remaining K-1 sub-regions trains Vertical Square parallactic angle and level orientation suitable for respective region after the same method
Angle prediction model.
5. a kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method according to claim 1, special
Sign is: specific step is as follows for the angle prediction process of the model；
Step 1: beam scanning, the range of coarse localization incoming wave sense；
Utilize known reception code bookEach submatrix selects mutually different code word to be formed
Wave beam receives the transmitting signal from unknown orientation signal source, according to the reception signal of each submatrix, selects corresponding reception signal
The strongest code word of energyAnd then it orients incoming wave sense and is located at kth*In sub-regions；
Step 2: extracting the direction character vector x of incoming wave signal；
All submatrixs are all made of the corresponding reception strongest code word of signal energy chosen in step 1As connecing
Wave beam is received, and the real and imaginary parts that each submatrix receives signal are extracted and are spliced into a vector as the direction of incoming wave signal
Characteristic vector x；
Step 3: the angle of fining estimation incoming wave signal；
The direction character vector x in the unknown signaling source obtained in step 2 is input to the kth good through off-line training*Sub-regions
In Vertical Square parallactic angle and horizontal azimuth prediction model, the angle of incoming wave signal is further estimated to fining.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376026.9A CN110113088B (en) | 2019-05-07 | 2019-05-07 | Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910376026.9A CN110113088B (en) | 2019-05-07 | 2019-05-07 | Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110113088A true CN110113088A (en) | 2019-08-09 |
| CN110113088B CN110113088B (en) | 2021-06-11 |
Family
ID=67488614
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910376026.9A Active CN110113088B (en) | 2019-05-07 | 2019-05-07 | Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110113088B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN111372195A (en) * | 2020-01-23 | 2020-07-03 | 鹏城实验室 | Method, apparatus and storage medium for tracking position of mobile terminal in mobile communication network |
| CN112731279A (en) * | 2020-12-09 | 2021-04-30 | 复旦大学 | Arrival angle estimation method based on hybrid antenna subarray |
| CN112803976A (en) * | 2020-12-24 | 2021-05-14 | 浙江香农通信科技有限公司 | Large-scale MIMO precoding method and system and electronic equipment |
| WO2022174642A1 (en) * | 2021-02-22 | 2022-08-25 | 华为技术有限公司 | Space division-based data processing method and communication device |
| WO2022231486A1 (en) * | 2021-04-29 | 2022-11-03 | Telefonaktiebolaget Lm Ericsson (Publ) | Machine learning for phase ambiguity limitation |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9445282B2 (en) * | 2014-11-17 | 2016-09-13 | Mediatek Inc. | Transceiver architecture for multiple antenna systems |
| CN108828505A (en) * | 2018-04-16 | 2018-11-16 | 南京理工大学 | Angle-of- arrival estimation algorithm research and application based on machine learning |
| CN108933745A (en) * | 2018-07-16 | 2018-12-04 | 北京理工大学 | A kind of broad-band channel estimation method estimated based on super-resolution angle and time delay |
| CN109085531A (en) * | 2018-08-27 | 2018-12-25 | 西安电子科技大学 | Near field sources angle-of- arrival estimation method neural network based |
-
2019
- 2019-05-07 CN CN201910376026.9A patent/CN110113088B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9445282B2 (en) * | 2014-11-17 | 2016-09-13 | Mediatek Inc. | Transceiver architecture for multiple antenna systems |
| CN108828505A (en) * | 2018-04-16 | 2018-11-16 | 南京理工大学 | Angle-of- arrival estimation algorithm research and application based on machine learning |
| CN108933745A (en) * | 2018-07-16 | 2018-12-04 | 北京理工大学 | A kind of broad-band channel estimation method estimated based on super-resolution angle and time delay |
| CN109085531A (en) * | 2018-08-27 | 2018-12-25 | 西安电子科技大学 | Near field sources angle-of- arrival estimation method neural network based |
Non-Patent Citations (6)
| Title |
|---|
| DIAN FAN ET AL: "Angle Domain Channel Estimation in Hybrid Millimeter Wave Massive MIMO Systems", 《IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS》 * |
| EVGENY EFIMOV ET AL: "Angle of arrival estimator based on artificial neural networks", 《2016 17TH INTERNATIONAL RADAR SYMPOSIUM》 * |
| HONGJI HUANG ET AL: "Deep Learning for Super-Resolution Channel Estimation and DOA Estimation Based Massive MIMO System", 《IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY》 * |
| KAI WU ET AL: "Robust Unambiguous Estimation of Angle-of-Arrival in Hybrid Array With Localized Analog Subarrays", 《IEEE TRANSACTIONS ON WIRELESS COMMUNICATIONS》 * |
| 孙菲艳 等: "采用混合核支持向量机的DOA估计", 《电讯技术》 * |
| 孟非，王旭: "基于PSO-BP神经网络的DOA估计方法", 《电讯技术》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN111372195A (en) * | 2020-01-23 | 2020-07-03 | 鹏城实验室 | Method, apparatus and storage medium for tracking position of mobile terminal in mobile communication network |
| CN111372195B (en) * | 2020-01-23 | 2022-07-01 | 鹏城实验室 | Method, apparatus and storage medium for tracking position of mobile terminal in mobile communication network |
| CN112731279A (en) * | 2020-12-09 | 2021-04-30 | 复旦大学 | Arrival angle estimation method based on hybrid antenna subarray |
| CN112803976A (en) * | 2020-12-24 | 2021-05-14 | 浙江香农通信科技有限公司 | Large-scale MIMO precoding method and system and electronic equipment |
| WO2022174642A1 (en) * | 2021-02-22 | 2022-08-25 | 华为技术有限公司 | Space division-based data processing method and communication device |
| WO2022231486A1 (en) * | 2021-04-29 | 2022-11-03 | Telefonaktiebolaget Lm Ericsson (Publ) | Machine learning for phase ambiguity limitation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110113088B (en) | 2021-06-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110113088A (en) | A kind of divergence type numerical model analysis antenna system direction of arrival intelligence estimation method | |
| Xiao et al. | UAV communications with millimeter-wave beamforming: Potentials, scenarios, and challenges | |
| Shafi et al. | Microwave vs. millimeter-wave propagation channels: Key differences and impact on 5G cellular systems | |
| Alkhateeb et al. | Deep learning coordinated beamforming for highly-mobile millimeter wave systems | |
| Zhang et al. | Research challenges and opportunities of UAV millimeter-wave communications | |
| Zhou et al. | Beam-forecast: Facilitating mobile 60 GHz networks via model-driven beam steering | |
| KR102154481B1 (en) | Apparatus for beamforming massive mimo system using deep learning | |
| CN110492911A (en) | A kind of beam tracking method and system for UAV Communication | |
| CN106355245A (en) | Method for integrating array antenna directional images on basis of neural network algorithms | |
| CN109412983A (en) | A kind of extensive mimo channel algorithm for estimating of mesh freeization based on the domain DFT | |
| CN107592611A (en) | The extensive mimo system wireless location method in broadband and system | |
| CN108111208A (en) | A kind of millimeter-wave systems angle domain channel estimation methods of flexible codebook configuration | |
| Zhou et al. | AoA-based positioning for aerial intelligent reflecting surface-aided wireless communications: An angle-domain approach | |
| CN107135023A (en) | Three-dimensional training code book design method and beam alignment for millimeter-wave communication system | |
| WO2020248443A1 (en) | Two-dimensional phased antenna array-based rapid and accurate beam tracking method | |
| CN108650200A (en) | The low frequency assisted channel estimation method of low-and high-frequency mixed networking system | |
| Ji et al. | Networking based ISAC hardware testbed and performance evaluation | |
| Jiang et al. | Improve radar sensing performance of multiple roadside units cooperation via space registration | |
| CN104994041B (en) | Three-dimensional extensive Bo Dabo digression combined estimation method of the antenna system based on pilot tone | |
| Wang et al. | CAP-Net: A deep learning-based angle prediction approach for ISAC-enabled RIS-assisted V2I communications | |
| Moon et al. | Coordinated Millimeter Wave Beam Selection Using Fingerprint for Cellular-Connected Unmanned Aerial Vehicle. | |
| CN113075649A (en) | Signal level direct positioning method suitable for distributed networked radar | |
| Yang et al. | 3-D positioning and environment mapping for mmWave communication systems | |
| Kao et al. | AI-aided 3-D beamforming for millimeter wave communications | |
| CN110213186A (en) | The angle of arrival and signal-noise ratio estimation method of ESPRIT based on mixed structure |
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 |