CN110133577A - The relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network - Google Patents
The relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network Download PDFInfo
- Publication number
- CN110133577A CN110133577A CN201910436027.8A CN201910436027A CN110133577A CN 110133577 A CN110133577 A CN 110133577A CN 201910436027 A CN201910436027 A CN 201910436027A CN 110133577 A CN110133577 A CN 110133577A
- Authority
- CN
- China
- Prior art keywords
- matrix
- signal
- imaginary
- data
- frequency
- 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
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/02—Systems using reflection of radio waves, e.g. primary radar systems; Analogous systems
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G01S3/00—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received
- G01S3/02—Direction-finders for determining the direction from which infrasonic, sonic, ultrasonic, or electromagnetic waves, or particle emission, not having a directional significance, are being received using radio waves
- G01S3/74—Multi-channel systems specially adapted for direction-finding, i.e. having a single antenna system capable of giving simultaneous indications of the directions of different signals
Abstract
The invention discloses a kind of relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network, this method carries out DOA estimation to multiple target using the list relatively prime MIMO of base, the estimation including the estimation to target numbers, to multiple azimuth of target.This method improves the part of data processing in traditional relatively prime MIMO of single base.Emission array transmitting electromagnetic wave detects space, and several target echoes are received array received, and the extraction of property field is carried out to received signal, deep neural network is re-fed into and is trained.Target state estimator is carried out with the network for completing training and exports the spatial spectrum of multiple target.It can be very good removal interference signal by residual error neural network, and quickly export high-precision spatial spectrum.
Description
Technical field
The present invention relates to the DOA estimation technique fields of relatively prime MIMO array, especially a kind of to be based on time-frequency residual error nerve net
The relatively prime MIMO array DOA algorithm for estimating of network.
Background technique
Direction of arrival DOA estimates to determine the azimuth position information of multiple extraterrestrial targets have high-resolution, extensively
Applied to fields such as communication, radar, sonar, seismic sensors.In recent years the relatively prime array proposed is in element position determination, adjacent array element
Coupling mutual interference etc. has advantage outstanding, is increasingly becoming the hot spot of concern.
The relatively prime MIMO array of multiple-input multiple-output list base be mainly used for do not have radiation signal ability or there is no stablizing it is outer
The target of portion's radiation source is detected, and target normally behaves as relevant the case where mixing with incoherent target at this time.Based on single base
The freedom degree of the target DOA estimation of relatively prime MIMO array is decided by empty in " virtual and collaboration battle array " corresponding " virtual poor collaboration battle array "
The position number of matroid member, when receiving and dispatching array and being all made of uniformly densely covered array structure, " virtual and collaboration battle array " and " virtually poor association
The case where being located at same position there are multiple Virtual arrays in same battle array ", that is, there is bulk redundancy, causes DOA estimation freedom degree
Loss.For this purpose, the design of sparse MIMO array is to consider to reduce Virtual array under same physical array element number with position
Situation, it is contemplated that " physical array " arrives the complex mapping relation that " virtual and collaboration battle array " arrives " virtual poor collaboration battle array " again, often at present
The sparse MIMO array used is nested MIMO array and the relatively prime MIMO array of single base, and two kinds of MIMO arrays are in " virtual and collaboration
The case where Virtual array is with position is not present in battle array " level, but still has Virtual array with position in " virtual poor collaboration battle array " level
Situation, although there is certain freedom degree loss, Array Design is relatively easy, it has also become the sparse MIMO array knot of mainstream at present
Structure.
Based on the DOA estimation method of the relatively prime MIMO receiving array of single base, building is suitable for the depth of non-homogeneous relatively prime array
Learn orientation recognition network, retains azimuth resolution height, the maximum distinguishable number of targets that classical signal processing class algorithm has
Mesh is broken through under the advantage of physics array element number limitation, further enhances non-homogeneous single relatively prime MIMO array of base to target Bearing Estimation
Timeliness and environmental suitability, and robustness under the conditions of low signal-to-noise ratio, low snap, big azimuth, Coherent Targets.By
In in view of classical deep neural network is in the training process it is possible that the reason of gradient explosion and disappearance, present invention choosing
Residual error neural network is selected.
The algorithm mainly solve it is towards the relatively prime MIMO array of single base, be suitable for narrowband/broadband and multiple relevant/incoherent
Clarification of objective expresses the formation problem in domain and the Construct question of corresponding deep learning network.Its essence is based on " and collaboration
Battle array " concept, be used in combination time domain how soon two feature representation domains of beat of data matrix and frequency domain multi-frequency list snapshot data matrix,
From the compatible narrowband/broadband of original time-frequency data plane and relevant/incoherent, by full convolutional neural networks from two feature representations
It extracts spatial feature relevant to target DOA and maps to form spatial spectrum in domain.
Summary of the invention
In view of this, the purpose of the present invention is to provide a kind of relatively prime MIMO arrays of single base based on time-frequency residual error network
DOA algorithm for estimating.
In order to achieve the above objectives, the invention provides the following technical scheme:
The relatively prime MIMO array DOA algorithm for estimating of single base provided by the invention based on time-frequency residual error network, including following step
It is rapid:
(1) the relatively prime MIMO array of single base is constructed, the list relatively prime MIMO array of base is by possessing the transmitting battle array of N number of array element and gathering around
It is made of the reception battle array of 2M-1 array element；
The physical location coordinate for emitting array element is T={ ti| i=0, Md, 2Md ... (N-1) Md }；
The physical location coordinate for receiving array element is R={ ri| i=Nd, 2Nd, 3Nd ... (2M-1) Nd }；
Wherein, M and N is relatively prime relationship, and M < N, λ are the wavelength for emitting signal,For basic spacing；
(2) emission array tranmitting frequency is that the electromagnetic wave signal of c/2d detects extraterrestrial target；The signal passes through
It is received after several target reflections by receiving array in space, then signal is separated by matched filtering, obtain (2M-1) × N
The echo-signal in a channel；The sampling for carrying out K snap to obtained signal obtains the data matrix C of [(2M-1) × N] × K；
Wherein, c is the light velocity in air；
(3) data matrix C is decomposed into corresponding time domain real-part matrixrC and pure imaginary number matrixiC, the time domain real part square
Battle arrayrC and time domain imaginary-part matrixiThe size of C is identical as data matrix C；
(4) to time domain imaginary-part matrixiC dot product-i, as the input of neural network, is obtained with removing the imaginary number components in matrix
To with time domain imaginary-part matrixiThe identical imaginary-part matrix of C sizeIC；
Leaf transformation in L point discrete Fourier is carried out to same channel signal in data Matrix C, that is, each row signal and obtains data square
Each row in battle array X, data matrix X is the signal that same array element is connected to different frequent points, each to be classified as the institute for possessing same frequency
There is channel signal；
(5) data matrix X is decomposed into frequency domain real-part matrixrX and frequency domain imaginary-part matrixiX, the frequency domain real-part matrixrX
With frequency domain imaginary-part matrixiThe size of X is identical as data X；
To frequency domain imaginary-part matrixiX dot product-i is to remove the imaginary number in matrix, using the input as neural network, obtain with
Frequency domain imaginary-part matrixiThe identical imaginary-part matrix of X sizeIX；
(6) by time domain real-part matrixrC, imaginary-part matrixIC, frequency domain real-part matrixrX, imaginary-part matrixIX is input to depth nerve
In network；Deep neural network is constructed, the deep neural network is multi input residual error neural network；Completion DOA estimation is simultaneously defeated
The spatial spectrum of signal out.
Further, the data matrix C is arranged in the following way:
Wherein, in each behavior (2M-1) × N number of signal path of Matrix C one of channel K sampled signal, number
According in Matrix C each column represent be the identical snap ordinal number that all channel receptions arrive signal；Indicate that the 1st transmitting-receiving is logical
The signal that road snap ordinal number is 1；Indicate that (2M-1) N number of transceiver channel snap ordinal number is the signal of K.
Further, the time domain real-part matrixrC is arranged in the following way:
The pure imaginary number matrix is arranged in the following way:
Wherein,Indicate that the 1st transceiver channel snap ordinal number is the real part of 1 signal；
Indicate that (2M-1) N number of transceiver channel snap ordinal number is the real part of the signal of K；
Indicate that the 1st transceiver channel snap ordinal number is the imaginary part of 1 signal；
Indicate that (2M-1) N number of transceiver channel snap ordinal number is the imaginary part of the signal of K.
Further, the imaginary-part matrixIC is calculated according to following formula:
IC=iC.*(-i)；Wherein, pure imaginary number matrixiIt is calculated between C and-i using dot product；
The imaginary-part matrixIC is indicated in the following way:
Wherein,The signal that the 1st transceiver channel snap ordinal number of pure imaginary number after indicating real number is 1；
The signal that the N number of transceiver channel snap ordinal number of pure imaginary number (2M-1) after indicating real number is K；
Leaf transformation in L point discrete Fourier is carried out to channel signal same in Matrix C, that is, each row signal and obtains data matrix X；
The data matrix X is indicated in the following way:
Wherein, each row in the data matrix X is the signal that same array element is connected to different frequent points, and each be classified as possesses
All channel signals of same frequency；
Wherein,Indicate that the 1st transceiver channel frequency point is 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the signal data of L.
Further, the purely real matrixrX is as follows:
The pure imaginary number matrixiX is as follows:
Wherein,Indicate that the 1st transceiver channel frequency point is the real part of 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the real part of the signal data of L；
Indicate that the 1st transceiver channel frequency point is the imaginary part of 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the imaginary part of the signal data of L.
Further, the imaginary-part matrixIX is calculated according to following formula:
IX=iX.*(-i)；
The imaginary-part matrixIX is as follows:
Wherein,Imaginary part after indicating the signal data real number that the 1st transceiver channel frequency point is 1；
Imaginary part after indicating the signal data real number that (2M-1) N number of transceiver channel frequency point is L.
Further, the parallel residual error neural network structure is provided with three layers, described in detail below:
First big layer is four parallel inputs, and it is later two residual blocks that network front end, which is convolutional layer,；
The second largest layer is two parallel residual error neural networks, and there are two residual blocks for each network；
There are six residual blocks for the third-largest layer, connect an average pond layer later, possess in the last connection one of network
The full articulamentum of 1600-2000 neuron completes the classification problem based on angle, is finally completed DOA estimation and output signal
Spatial spectrum, resolving power be 0.1 ° -0.2 °.
Further, input data includes corresponding time-domain matrix and frequency domain two-by-two in the parallel residual error neural network structure
Matrix.
The beneficial effects of the present invention are:
The invention proposes a kind of relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network, this method
DOA estimation is carried out to multiple target using the relatively prime MIMO of single base to estimate multiple azimuth of target including the estimation to target numbers
Meter.This method improves the part of data processing in traditional relatively prime MIMO of single base.Emission array emits electromagnetic wave pair
Space is detected, and several target echoes are received array received, and the extraction of property field is carried out to received signal, then is sent
Enter deep neural network to be trained.Target state estimator is carried out with the network for completing training and exports the spatial spectrum of multiple target.Pass through
Residual error neural network can be very good removal interference signal, and quickly export high-precision spatial spectrum.
The present invention can efficiently extract target angle and mesh from a large amount of interference using residual error neural network structure
Quantity is marked, realizes and needs extracting for information useful from property field according to network, while avoiding in the training process
There is the case where gradient explosion and gradient disappearance, it is demonstrated experimentally that carrying out DOA estimation using new method, can be very good to carry out
The estimation of target numbers and the generation of spatial spectrum.
Other advantages, target and feature of the invention will be illustrated in the following description to a certain extent, and
And to a certain extent, based on will be apparent to those skilled in the art to investigating hereafter, Huo Zheke
To be instructed from the practice of the present invention.Target of the invention and other advantages can be realized by following specification and
It obtains.
Detailed description of the invention
In order to keep the purpose of the present invention, technical scheme and beneficial effects clearer, the present invention provides following attached drawing and carries out
Illustrate:
Fig. 1 is the relatively prime MIMO array structure of single base.
Fig. 2 is data processing overall flow.
Fig. 3 is parallel residual error neural network structure.
Specific embodiment
The present invention will be further explained below with reference to the attached drawings and specific examples, so that those skilled in the art can be with
It better understands the present invention and can be practiced, but illustrated embodiment is not as a limitation of the invention.
Matrix letter meaning involved in provided in this embodiment is specific as follows: time domain real-part matrixrC (purely real square
Battle arrayr) and pure imaginary number matrix CiC, imaginary-part matrixIC；Frequency domain real-part matrixrX (purely real matrixr) and pure imaginary number matrix XiX, void
Portion's matrixIX。
It is provided in this embodiment based on time-frequency residual error network as shown in FIG. 1, FIG. 1 is the relatively prime MIMO array structure of single base
Single relatively prime MIMO array DOA algorithm for estimating of base, comprising the following steps:
(1) the relatively prime MIMO array of single base is constructed, single relatively prime MIMO array of base is by possessing the transmitting battle array of N number of array element and possessing
The reception battle array composition of 2M-1 array element,
The physical location coordinate for emitting array element is T={ ti| i=0, Md, 2Md ... (N-1) Md },
The physical location coordinate for receiving array element is R={ ri| i=Nd, 2Nd, 3Nd ... (2M-1) Nd }；
Wherein, M and N is relatively prime relationship, and M < N, λ are the wavelength for emitting signal,For basic spacing,.
(2) emission array tranmitting frequency is that the electromagnetic wave signal of c/2d detects extraterrestrial target,
Wherein, c is the light velocity in air；
Signal is received after being reflected by targets several in space by receiving array, then is divided by matched filtering signal
From obtaining the echo-signal in (2M-1) × N number of channel；
The sampling for carrying out K snap to obtained signal obtains the data matrix C of [(2M-1) × N] × K；
Wherein, the K sampling in one of channel is believed in each behavior (2M-1) × N number of signal path of data matrix C
Number, what each column in data matrix C represented is the signal for the identical snap ordinal number that all channel receptions arrive；
Corresponding angle label vector T is generated simultaneously, and label vector size is 1 × 1800, is mentioned for later period neural metwork training
For label；Configure vector process are as follows: dispose 1 remaining angle disposition 0 in respective angles in vector.
(3) data matrix C is decomposed into corresponding purely real matrixrC and pure imaginary number matrixiC, size are identical as C.
Purely real matrixrC is as follows:
Pure imaginary number matrixiC is as follows:
(4) to matrixiC dot product-i is to remove the imaginary number components in matrix, as the input of neural network, obtains and matrixiThe identical matrix of C sizeIC；
IC=iC.*(-i)
Leaf transformation in L point discrete Fourier is carried out to channel signal same in Matrix C, that is, each row signal and obtains data matrix X,
Each row in data matrix X is the signal that same array element is connected to different frequent points, and each be classified as possesses all logical of same frequency
Road signal；
(5) matrix X is decomposed into purely real matrixrX and pure imaginary number matrixiX, matrix size are identical as X；
Purely real matrixrX is as follows:
Pure imaginary number matrixiX is as follows:
To matrixiX dot product-i is to remove the imaginary number in matrix, using the input as neural network, obtains and matrixiX size
Identical matrixIX；
IX=iX.*(-i)
It is sent into the time domain real-part matrix of deep neural network at this timerC, imaginary-part matrixIC, frequency domain real-part matrixrX, imaginary part square
Battle arrayITotally four data matrixes are ready for X, these matrixes have co-expressed temporal signatures expression domain；Data processing bulk flow
Journey is as shown in Fig. 2, Fig. 2 is data processing overall flow.
(6) building of deep neural network, be sent into residual error neural network is four data matrixes, constructs a multi input
Residual error neural network, as shown in figure 3, Fig. 3 is parallel residual error neural network structure.
Wherein, 3 × 3 size for representing convolution kernel, 64,128,256,512,1024 respectively represent convolution in corresponding residual block
Nuclear volume, Conv are convolutional layer./ 2 be the quantity adjustment to convolution kernel, and to match residual block output, Avg pool is average
Pond layer, FC1800 layers are one layer of full articulamentums for having 1800 neurons.
This method uses parallel residual error neural network, specific structure are as follows: whole network is divided into three big layers.First is big
Layer is four parallel inputs, and it is later two residual blocks that network front end, which is convolutional layer,.The second largest layer is two parallel residual errors
Neural network, there are two residual blocks for each network.The third-largest layer has 6 residual blocks, connects an average pond layer later,
The last connection one of network possesses the full articulamentum of 1800 neurons, completes the classification problem based on angle, is finally completed
The spatial spectrum of DOA estimation and output signal, resolving power are 0.1 °.
The data of input network have 4 matrixes, and wherein time-domain matrix and frequency domain matrix correspond to two-by-two, only time domain or frequency domain
Purely real matrix and pure imaginary number matrix joint Shi Caineng embody the angle information of target in data.4 inputs of the first big layer
It is divided into 1. and is inputted there are two 2. two groups, each group, is 1. time-domain matrixrC andI2. the input terminal of C is frequency domain matrixrX andIX
Input terminal.rC andIThe information that C is carried can be converged in 3., and joint embodies angle information together, identicalrX andIX can be 4.
In converge.Data in 3. 4. can be extracted by network association, effect can than 4 inputs directly converge much in the 3rd big layer.
(7) data set that this method uses is emulated by matlab and is generated, and data set specific composition is as follows:
1, mono signal source: target incident angle ° uniform 1800 angles traversal from 0 ° to 180 generates emulation data, and
Corresponding angle label is generated to store in the form of txt file together.It traverses 15 times altogether, single signal source data set includes
27000 data.
2, multisignal source: generating Multiple Target Signals of the number of targets less than 20 by matlab at random, target incident angle with
Machine is distributed in ° uniform 1800 angles from 0 ° to 180, and generates corresponding angle label and stored in the form of txt file together.It is more
Signal source data set includes 40000 data.
Mono signal source data in above-mentioned data set and multisignal source data are respectively separated into training according to the ratio of 7:3 at random
Collection and test set.Respective training set and test set are merged, training set and test set have been ready for.
Training set data is sent into deep neural network and completes training.The angle judgement accuracy rate of output layer reaches
95.6%, it is more than preset value 95%, obtains the deep neural network for completing training.
Test set data are sent into deep neural network and are tested, experiment shows that trained deep neural network is quasi-
Exactness can reach 95.9%, and real-time and anti-interference ability are fine, reach default effect of the invention.
It (8) is matlab, Python3.7.1, Pycharm and pytorch used in this method as experiment porch,
Programming realizes design scheme of the invention.
MIMO array provided in this embodiment can be to not having radiation signal ability or there is no stablize external radiation
The target in source is detected.Relatively prime array can estimate freedom degree by building " virtual and collaboration battle array " Lai Zengjia DOA, to improve
Array identifies the ability of more multiple target simultaneously.Deep neural network can learn from a large amount of training set data to angle information
Feature and extracted.Compared to traditional DOA algorithm for estimating in the environment of lot of interfering factors is possessed, it is based on depth
The DOA estimation of study has very big advantage, and the error etc. of ambient noise, array original part in environment all will be again and again
Training in learnt by deep neural network to corresponding feature, finally rapidly output has the spatial spectrum of very high accurancy and precision.
Embodiment described above is only to absolutely prove preferred embodiment that is of the invention and being lifted, protection model of the invention
It encloses without being limited thereto.Those skilled in the art's made equivalent substitute or transformation on the basis of the present invention, in the present invention
Protection scope within.Protection scope of the present invention is subject to claims.
Claims (8)
1. the relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network, it is characterised in that: the following steps are included:
(1) the relatively prime MIMO array of single base is constructed, the list relatively prime MIMO array of base is by possessing the transmitting battle array of N number of array element and possessing
The reception battle array composition of 2M-1 array element；
The physical location coordinate for emitting array element is T={ ti| i=0, Md, 2Md ... (N-1) Md }；
The physical location coordinate for receiving array element is R={ ri| i=Nd, 2Nd, 3Nd ... (2M-1) Nd }；
Wherein, M and N is relatively prime relationship, and M < N, λ are the wavelength for emitting signal,For basic spacing；
(2) emission array tranmitting frequency is that the electromagnetic wave signal of c/2d detects extraterrestrial target；The signal passes through space
In received by receiving array after the reflection of several targets, then signal is separated by matched filtering, obtains (2M-1) × N number of and lead to
The echo-signal in road；The sampling for carrying out K snap to obtained signal obtains the data matrix C of [(2M-1) × N] × K；Wherein,
C is the light velocity in air；
(3) data matrix C is decomposed into corresponding time domain real-part matrixrC and pure imaginary number matrixiC, the time domain real-part matrixrC
With time domain imaginary-part matrixiThe size of C is identical as data matrix C；
(4) to time domain imaginary-part matrixiC dot product-i is to remove the imaginary number components in matrix, as the input of neural network, obtain with
Time domain imaginary-part matrixiThe identical imaginary-part matrix of C sizeIC；
Leaf transformation in L point discrete Fourier is carried out to same channel signal in data Matrix C, that is, each row signal and obtains data matrix X,
Each row in data matrix X is the signal that same array element is connected to different frequent points, and each be classified as possesses all logical of same frequency
Road signal；
(5) data matrix X is decomposed into frequency domain real-part matrixrX and frequency domain imaginary-part matrixiX, the frequency domain real-part matrixrX and frequency
Domain imaginary-part matrixiThe size of X is identical as data X；
To frequency domain imaginary-part matrixiX dot product-i is to remove the imaginary number in matrix, using the input as neural network, obtains and frequency domain void
Portion's matrixiThe identical imaginary-part matrix of X sizeIX；
(6) by time domain real-part matrixrC, imaginary-part matrixIC, frequency domain real-part matrixrX, imaginary-part matrixIX is input to deep neural network
In；Deep neural network is constructed, the deep neural network is multi input residual error neural network；DOA is completed to estimate and export letter
Number spatial spectrum.
2. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: the data matrix C is arranged in the following way:
Wherein, in each behavior (2M-1) × N number of signal path of Matrix C one of channel K sampled signal, data square
What each column in battle array C represented is the signal for the identical snap ordinal number that all channel receptions arrive；Indicate that the 1st transceiver channel is fast
Clap the signal that ordinal number is 1；Indicate that (2M-1) N number of transceiver channel snap ordinal number is the signal of K.
3. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: the time domain real-part matrixrC is arranged in the following way:
The pure imaginary number matrix is arranged in the following way:
Wherein,Indicate that the 1st transceiver channel snap ordinal number is the real part of 1 signal；
Indicate that (2M-1) N number of transceiver channel snap ordinal number is the real part of the signal of K；
Indicate that the 1st transceiver channel snap ordinal number is the imaginary part of 1 signal；
Indicate that (2M-1) N number of transceiver channel snap ordinal number is the imaginary part of the signal of K.
4. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: the imaginary-part matrixIC is calculated according to following formula:
IC=iC.*(-i)；Wherein, pure imaginary number matrixiIt is calculated between C and-i using dot product；
The imaginary-part matrixIC is indicated in the following way:
Wherein,The signal that the 1st transceiver channel snap ordinal number of pure imaginary number after indicating real number is 1；
The signal that the N number of transceiver channel snap ordinal number of pure imaginary number (2M-1) after indicating real number is K；
Leaf transformation in L point discrete Fourier is carried out to channel signal same in Matrix C, that is, each row signal and obtains data matrix X；It is described
Data matrix X is indicated in the following way:
Wherein, each row in the data matrix X is that same array element is connected to the signals of different frequent points, it is each be classified as possess it is same
All channel signals of frequency；
Wherein,Indicate that the 1st transceiver channel frequency point is 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the signal data of L.
5. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: the purely real matrixrX is as follows:
The pure imaginary number matrixiX is as follows:
Wherein,Indicate that the 1st transceiver channel frequency point is the real part of 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the real part of the signal data of L；
Indicate that the 1st transceiver channel frequency point is the imaginary part of 1 signal data；
Indicate that (2M-1) N number of transceiver channel frequency point is the imaginary part of the signal data of L.
6. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: the imaginary-part matrixIX is calculated according to following formula:
IX=iX.*(-i)；
The imaginary-part matrixIX is as follows:
Wherein,Imaginary part after indicating the signal data real number that the 1st transceiver channel frequency point is 1；
Imaginary part after indicating the signal data real number that (2M-1) N number of transceiver channel frequency point is L.
7. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
: the parallel residual error neural network structure is provided with three layers, described in detail below:
First big layer is four parallel inputs, and it is later two residual blocks that network front end, which is convolutional layer,；
The second largest layer is two parallel residual error neural networks, and there are two residual blocks for each network；
There are six residual blocks for the third-largest layer, connect an average pond layer later, possess 1600- in the last connection one of network
The full articulamentum of 2000 neurons completes the classification problem based on angle, is finally completed the space of DOA estimation and output signal
Spectrum, resolving power are 0.1 ° -0.2 °.
8. the relatively prime MIMO array DOA algorithm for estimating of single base as described in claim 1 based on time-frequency residual error network, feature exist
In: input data includes corresponding time-domain matrix and frequency domain matrix two-by-two in the parallel residual error neural network structure.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910436027.8A CN110133577B (en) | 2019-05-23 | 2019-05-23 | Single-basis co-prime MIMO array DOA estimation algorithm based on time-frequency residual error network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910436027.8A CN110133577B (en) | 2019-05-23 | 2019-05-23 | Single-basis co-prime MIMO array DOA estimation algorithm based on time-frequency residual error network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110133577A true CN110133577A (en) | 2019-08-16 |
| CN110133577B CN110133577B (en) | 2020-11-10 |
Family
ID=67572957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910436027.8A Expired - Fee Related CN110133577B (en) | 2019-05-23 | 2019-05-23 | Single-basis co-prime MIMO array DOA estimation algorithm based on time-frequency residual error network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110133577B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111446998A (en) * | 2019-10-21 | 2020-07-24 | 清华大学 | Direction-of-arrival estimation method based on deep learning |
| CN113030847A (en) * | 2021-04-13 | 2021-06-25 | 中国民用航空飞行学院 | Deep learning data set generation method for dual-channel direction finding system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105182293A (en) * | 2015-08-25 | 2015-12-23 | 西安电子科技大学 | Method for estimating DOA and DOD of MIMO radar based on co-prime array |
| CN106772224A (en) * | 2016-12-28 | 2017-05-31 | 武汉大学 | A kind of L-type array estimating two-dimensional direction-of-arrival algorithm of use time frequency analysis |
| CN109557502A (en) * | 2018-12-19 | 2019-04-02 | 成都理工大学 | Sparse nested MIMO array DOA estimation method based on relatively prime double frequency |
| CN109684995A (en) * | 2018-12-22 | 2019-04-26 | 中国人民解放军战略支援部队信息工程大学 | Specific Emitter Identification method and device based on depth residual error network |
| CN109752687A (en) * | 2019-01-15 | 2019-05-14 | 哈尔滨工程大学 | A kind of relatively prime array setting method of modified reducing array element mutual coupling effect |
-
2019
- 2019-05-23 CN CN201910436027.8A patent/CN110133577B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105182293A (en) * | 2015-08-25 | 2015-12-23 | 西安电子科技大学 | Method for estimating DOA and DOD of MIMO radar based on co-prime array |
| CN106772224A (en) * | 2016-12-28 | 2017-05-31 | 武汉大学 | A kind of L-type array estimating two-dimensional direction-of-arrival algorithm of use time frequency analysis |
| CN109557502A (en) * | 2018-12-19 | 2019-04-02 | 成都理工大学 | Sparse nested MIMO array DOA estimation method based on relatively prime double frequency |
| CN109684995A (en) * | 2018-12-22 | 2019-04-26 | 中国人民解放军战略支援部队信息工程大学 | Specific Emitter Identification method and device based on depth residual error network |
| CN109752687A (en) * | 2019-01-15 | 2019-05-14 | 哈尔滨工程大学 | A kind of relatively prime array setting method of modified reducing array element mutual coupling effect |
Non-Patent Citations (3)
| Title |
|---|
| YONG JIA ET AL.: "DOA and DOD Estimation Based on Bistatic MIMO", 《2017 IEEE RADAR CONFERENCE》 * |
| 张昊 等: "基于协方差矩阵重构的互质阵列DOA估计方法", 《探测与控制学报》 * |
| 陈川 等: "基于双频互质阵列的DOA估计方法", 《计算机应用研究》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111446998A (en) * | 2019-10-21 | 2020-07-24 | 清华大学 | Direction-of-arrival estimation method based on deep learning |
| CN111446998B (en) * | 2019-10-21 | 2021-09-14 | 清华大学 | Direction-of-arrival estimation method based on deep learning |
| CN113030847A (en) * | 2021-04-13 | 2021-06-25 | 中国民用航空飞行学院 | Deep learning data set generation method for dual-channel direction finding system |
| CN113030847B (en) * | 2021-04-13 | 2023-04-25 | 中国民用航空飞行学院 | Deep learning data set generation method for dual-channel direction finding system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110133577B (en) | 2020-11-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111220958B (en) | Radar target Doppler image classification and identification method based on one-dimensional convolutional neural network | |
| CN109086700B (en) | Radar one-dimensional range profile target identification method based on deep convolutional neural network | |
| CN108229404A (en) | A kind of radar echo signal target identification method based on deep learning | |
| CN108182450A (en) | A kind of airborne Ground Penetrating Radar target identification method based on depth convolutional network | |
| CN110109059B (en) | Radar radiation source signal identification method based on deep learning network | |
| CN107728142B (en) | Radar high-resolution range profile target identification method based on two-dimensional convolutional network | |
| CN110133576A (en) | Biradical relatively prime MIMO array orientation algorithm for estimating based on cascade residual error network | |
| CN110365612A (en) | A kind of deep learning Beam Domain channel estimation methods based on approximate Message Passing Algorithm | |
| CN106355151A (en) | Recognition method, based on deep belief network, of three-dimensional SAR images | |
| CN104849698A (en) | Radar signal parallel processing method and system based on heterogeneous multinucleated system | |
| CN109683126A (en) | Direction of arrival measurement method, signal handling equipment and storage medium | |
| CN110348288A (en) | A kind of gesture identification method based on 77GHz MMW RADAR SIGNAL USING | |
| CN110161453A (en) | Relatively prime array wide and narrow strip target DOA estimation algorithm based on multi input residual error network | |
| CN105182313A (en) | MIMO-STAP steady waveform design method based on incomplete clutter prior knowledge | |
| CN111693975A (en) | MIMO radar sparse array design method based on deep neural network | |
| CN110133577A (en) | The relatively prime MIMO array DOA algorithm for estimating of single base based on time-frequency residual error network | |
| CN112859014A (en) | Radar interference suppression method, device and medium based on radar signal sorting | |
| CN113759323B (en) | Signal sorting method and device based on improved K-Means joint convolution self-encoder | |
| CN110232342A (en) | Sea situation level determination method and device based on convolutional neural networks | |
| CN111639746A (en) | GNSS-R sea surface wind speed inversion method and system based on CNN neural network | |
| CN114758230A (en) | Underground target body classification and identification method based on attention mechanism | |
| CN104076360A (en) | Two-dimensional SAR sparse target imaging method based on compression sensing | |
| CN111368930B (en) | Radar human body posture identification method and system based on multi-class spectrogram fusion and hierarchical learning | |
| CN113221631A (en) | Sequence pulse anti-interference target detection method based on convolutional neural network | |
| CN101470195B (en) | Non-parameter type high-resolution beam forming method and apparatus |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20201110 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |