CN110113088B - Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system - Google Patents
Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system Download PDFInfo
- Publication number
- CN110113088B CN110113088B CN201910376026.9A CN201910376026A CN110113088B CN 110113088 B CN110113088 B CN 110113088B CN 201910376026 A CN201910376026 A CN 201910376026A CN 110113088 B CN110113088 B CN 110113088B
- Authority
- CN
- China
- Prior art keywords
- sub
- training
- angle
- signal
- region
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
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 an intelligent estimation method for the arrival angle of a separated digital-analog hybrid antenna system, wherein a signal source transmits signals, a receiving end adopts a separated digital-analog hybrid antenna framework, and the whole technical scheme is divided into two processes: an off-line training process and an angle prediction process of the model. Compared with the prior art, the scheme provided by the invention fully utilizes the space and time resources of the system, can realize accurate estimation of the arrival angle within a low-complexity wide area range under the condition of unknown any prior information of the incoming wave, and effectively solves the problem of phase ambiguity in the traditional algorithm.
Description
Technical Field
The invention relates to the technical field of wireless communication, in particular to an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system.
Background
The rapid development of wireless communication technology changes the production and living modes of people deeply, greatly promotes the revolution and innovation of society, and becomes an indispensable part of the world at present. Multiple Input and Multiple Output (MIMO) technology is a major breakthrough in the field of wireless communication smart antennas, and can improve the capacity of a communication system by a Multiple without increasing the bandwidth. With the rapid development of mobile internet and the dramatic increase of mobile data volume and access device number, how to further satisfy the continuously increasing rate requirement of wireless communication becomes a key problem faced by the mobile communication technology in the future.
Due to the shortage of low-frequency spectrum resources in the existing cellular system, the implementation of wireless communication by using millimeter wave frequency bands attracts extensive attention and research in academia and industry. Due to the large propagation loss of millimeter wave in the millimeter wave band, the research on the millimeter wave wireless transmission technology is mostly focused on the short-distance communication scenario. However, considering the relatively short wavelength in the millimeter wave band, a large-scale antenna array can be equipped to both the base station and the user side. Furthermore, the beamforming gain provided by the large-scale antenna array can compensate for relatively high propagation loss in the millimeter wave frequency band. In view of the power consumption burden caused by the pure digital beamforming method and the performance loss caused by the pure analog beamforming method, the digital-analog hybrid beamforming method has become a research hotspot.
By virtue of the sparse characteristic of the millimeter wave propagation channel, the estimation of the angle information of the propagation channel is beneficial to the design of a millimeter wave propagation mode and the reduction of the cost for acquiring the channel information. However, the unique digital-analog hybrid antenna architecture and large-scale antenna array in the millimeter wave system greatly increase the complexity of the conventional Direction of Arrival (DOA) estimation algorithms, such as Multiple Signal Classification (MUSIC) and parameter estimation with rotation invariant technology (ESPRIT), and these conventional algorithms have a common drawback, i.e., a phase ambiguity problem, which is an integer Multiple of 2 pi that occurs in the process of solving the inverse trigonometric function after the distance between adjacent sub-arrays is greater than a half wavelength. Therefore, based on the separated subarray system adopting the digital-analog hybrid antenna architecture, and comprehensively considering the requirements of calculation complexity and estimation precision, the invention provides the method and the system device for intelligently estimating the wave arrival angle of the separated digital-analog hybrid antenna system, and jointly uses space resources and time resources. Compared with the traditional algorithm, the algorithm provided by the invention has lower complexity and higher accuracy.
Disclosure of Invention
In order to solve the problems, the invention provides an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which has low implementation complexity, and has short required angle prediction time and no phase ambiguity problem because the training process of a model is finished off-line. The invention provides an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which is characterized in that a system device of the intelligent estimation method for the wave arrival angle of the separated digital-analog hybrid antenna system comprises the following steps: a signal source transmitting signal and a receiving end adopt a separated digital-analog hybrid beam forming framework;
the method for intelligently estimating the wave arrival angle of the separated digital-analog hybrid antenna system comprises an offline training process and an angle prediction process of a model;
the off-line training process of the model is as follows;
the whole model training process is completed in an off-line state, and the whole space is divided into a plurality of sub-areas according to a codebook set; then aiming at any space subregion, constructing a training sample set; then, training by adopting the proposed novel neural network to finally obtain a plurality of wave arrival angle prediction models of different sub-regions;
the angle prediction process of the model is as follows;
firstly, each subarray of a receiving end selects different code words from a known receiving codebook to receive a transmitting signal from an unknown azimuth signal source, selects the code word with the strongest energy corresponding to the received signal, and further positions a sub-region where a wave signal is located; secondly, all the sub-arrays receive signals by using the selected code words with the strongest energy corresponding to the received signals, and feature vectors containing incoming wave signal directions are extracted from the received signals of all the sub-arrays; and then inputting the extracted feature vector into an offline trained wave arrival angle prediction model corresponding to the sub-region, and further finely estimating the wave arrival angle.
As a further improvement of the invention, the new neural network architecture is called "mircofishernet", which is similar in shape to fish, and has the characters that the front head is narrower, the middle trunk part is wider first and then narrower, and the last wider part is similar to the tail of fish; meanwhile, the whole network architecture can realize better prediction performance without too many layers, and the MircoFishNet network architecture comprises four parts, namely a narrower network layer, and aims to extract the coarse features, namely the global features, of input data; then a wider network layer, which aims to fully extract the local characteristics of the input data; secondly, a narrower network layer, whose main purpose is to reduce the amount of computation; finally, a broader network, whose purpose is to extract the deep features of the data sufficiently again.
As a further improvement of the present invention, the receiving end using a separate digital-analog hybrid beamforming architecture includes a receiving module composed of analog beamforming, radio frequency link, analog-digital conversion, and baseband digital signal processing, an offline training module using a neural network to construct a model, and an intelligent arrival angle prediction module, the receiving end is equipped with P × Q sub-arrays, where P sub-arrays in the horizontal direction and Q sub-arrays in the vertical direction are formed, each sub-array is composed of M × N antennas, where M antennas in the horizontal direction and N antennas in the vertical direction), each antenna is connected to an independent phase shifter for analog beamforming, and each sub-array is connected to a radio frequency link.
As a further improvement of the invention, the off-line training process of the model comprises the following specific steps;
step 1: dividing a space area;
according to a known receiving codebookDividing the spatial range into K sub-regions, whereinAndthe vertical phase and the horizontal phase of the kth code word are respectively represented, K is the number of the code words, obviously, each subregion corresponds to a certain range of vertical azimuth angles and horizontal azimuth angles, which is determined by the main lobe width of the beam corresponding to each code word;
step 2: generating a training sample set;
k training sample sets need to be generated in K sub-regions;
taking the kth sub-region as an example, R directional samples are randomly generated in the space range, i.e.WhereinRepresenting a vertical azimuth and a horizontal azimuth of an r-th directional sample in a k-th sub-region;
for the r directional sample, the signal source is locatedTransmitting signals, receiving signals by each subarray of a receiving end by using a k-th code word, extracting and splicing real parts and imaginary parts of the received signals of each subarray into a vectorUsing the vector as a sample of the characterizing directionThe feature vector of (2). Finally, a training sample set of the kth sub-region is obtained Is the feature vector of the r-th directional sample,a label for the sample of the r-th direction;
generating respective training sample sets of the other K-1 sub-regions according to the same method;
and step 3: training a model off line;
aiming at each sub-region, training a wave arrival angle prediction model suitable for the region by using a training sample set of the region and adopting an intelligent machine learning algorithm in an off-line manner;
taking the kth sub-region as an example, the training sample set obtained in step 2 is usedInputting the proposed novel neural network, training the parameters of each neuron, and obtaining a vertical azimuth angle and a horizontal azimuth angle prediction model suitable for the sub-region;
and training the other K-1 sub-regions according to the same method to obtain a vertical azimuth angle prediction model and a horizontal azimuth angle prediction model suitable for the respective regions.
As a further improvement of the invention, the specific steps of the angle prediction process of the model are as follows;
step 1: beam scanning, roughly positioning the range of incoming wave signal directions;
using known receive codebooksEach subarray selects different code words to form wave beams to receive the transmitted signals from the unknown azimuth signal source, and selects the code word with the strongest energy corresponding to the received signals according to the received signals of each subarrayFurther locate the incoming wave signal direction at the kth*Within each sub-region;
step 2: extracting a direction characteristic vector x of an incoming wave signal;
all the subarrays adopt the code word which is selected in the step 1 and corresponds to the received signal with the strongest energyAs receiving beams, extracting and splicing the real part and the imaginary part of each subarray receiving signal into a vector serving as a direction characteristic vector x of an incoming wave signal;
and step 3: finely estimating the angle of an incoming wave signal;
inputting the direction characteristic vector x of the unknown signal source obtained in the step 2 into a k < th > trained off-line*And in the prediction model of the vertical azimuth angle and the horizontal azimuth angle of each sub-region, the angle of the wave signal is further estimated in a refined mode.
The application discloses a method for intelligently estimating the arrival angle of a separated digital-analog hybrid antenna system, which has the following beneficial effects:
the invention provides an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which comprises an off-line training process and an intelligent angle prediction process of a model. Compared with the traditional wave angle estimation methods such as MUSIC and ESPRIT algorithms, the wave angle intelligent estimation method provided by the invention has low calculation complexity, can effectively solve the problem of phase ambiguity, provides useful information for a transmission scheme under a large-scale separation type digital-analog hybrid antenna system architecture, and has more practical value.
Drawings
FIG. 1 is a flow chart of an intelligent estimation method for the angle of arrival of a separated digital-analog hybrid antenna system;
FIG. 2 is a diagram of an off-line training process of a model;
FIG. 3 is a diagram of an intelligent angle prediction process;
FIG. 4 is a diagram of a receiving end area array architecture;
FIG. 5 is a schematic diagram of a MicroFishNet neural network architecture, in which (a) is a schematic diagram of MicroFishNet and (b) is a schematic diagram of MicroFishNet neuron connections;
FIG. 6 is a MircoFishNet neural network architecture diagram used in the example;
FIG. 7 is a graph showing the variation of the RMS error with the SNR for an angle estimation with 1 sampling point (MUSIC algorithm and the intelligent estimation method proposed in this patent);
FIG. 8 is a variation curve of the root mean square error of the angle estimation with the number of sampling points under 0dB (MUSIC algorithm and the intelligent estimation method proposed in this patent);
fig. 9 is a curve of the variation of the root mean square error of the angle estimation with the number of sampling points under 10dB (MUSIC algorithm and the intelligent estimation method proposed in this patent).
Detailed Description
The invention is described in further detail below with reference to the following detailed description and accompanying drawings:
the invention provides an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which has low realization complexity, short required angle prediction time and no phase ambiguity problem because the training process of a model is finished off line.
The embodiment of the invention discloses an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which comprises the following two processes: the off-line training process and the angle prediction process of the model are shown in fig. 1. The off-line training process of the model is as follows: firstly, dividing the whole space into a plurality of subregions according to a codebook set; then, aiming at each space subregion, constructing a training sample set; and then training by adopting the proposed novel neural network to finally obtain a plurality of wave arrival angle prediction models of different sub-regions. The angle prediction process is as follows: firstly, each subarray of a receiving end selects different code words from a known receiving codebook to receive a transmitting signal from an unknown azimuth signal source, selects the code word with the strongest energy corresponding to the received signal, and further positions a sub-region where a wave signal is located; secondly, all the sub-arrays receive signals by using the selected code words with the strongest energy corresponding to the received signals, and feature vectors containing incoming wave signal directions are extracted from the received signals of all the sub-arrays; and then inputting the extracted feature vector into an offline trained wave arrival angle prediction model corresponding to the sub-region, and further finely estimating the wave arrival angle. Compared with the prior art, the scheme provided by the invention fully utilizes the space and time resources of the system, can realize accurate estimation of the arrival angle within a low-complexity wide area range under the condition of unknown any prior information of the incoming wave, and effectively solves the problem of phase ambiguity in the traditional algorithm.
For the convenience of understanding the present invention, the following describes the related knowledge of the prior art, such as the array response vector and the area array code word of the receiving area array according to the present invention. It should be noted that the method of the present invention is not limited to the following expression of the specific formula.
As shown in FIG. 4, the multi-planar array antenna is placed in the xoy plane with the vertical azimuth angle and the horizontal azimuth angle being theta and theta, respectivelyShowing that P sub-arrays are arranged along the direction of the x-axis, and the distance between the corresponding antenna elements of two adjacent sub-arrays is Dx(ii) a Q sub-arrays are arranged along the y-axis direction, and the distance between the corresponding antenna elements of two adjacent sub-arrays is Dy. In each sub-area array, M antennas are arranged along the direction of the x axis, and the distance between adjacent antenna elements is dx(ii) a N antennas are arranged along the y-axis direction, and the distance between adjacent antenna elements is dy. With the antenna element of the coordinate origin as a reference point, the array response vector of the sub-area array at the p-th row and the q-th column can be expressed as:
wherein, (.)TWhich represents the operation of transposition by means of a transposition operation,λ represents the wavelength of the signal carrier wave,representing the Kronecker Product operation.
According to the area array receiving array response vector, the design of the area array code word can be expressed as:
wherein the content of the first and second substances,andrespectively representing the vertical phase and the horizontal phase of the k-th codeword.
The neural network is composed of an input layer, a hidden layer and an output layer, wherein the hidden layer is divided into a full connection layer, an activation layer, a random disconnection layer and the like according to different functions and neuron connection modes.
Each node of the Fully connected Dense layer (referred to as "Dense layer") is connected to all nodes of the previous layer for integrating the extracted features.
In neural networks, Activation functions (activations functions) are functions that run on neurons and are responsible for mapping the inputs of the neurons to the outputs with the aim of increasing the non-linearity of the neural network model. The Relu function (called a Linear rectification function) is expressed by f (x) max (0, x).
The random disconnection layer, also called Dropout layer, means that in the training process of the neural network, for the neural unit, the neural unit is temporarily discarded from the network according to a certain probability, and the purpose is to prevent overfitting.
Based on the above technical background, in the method for intelligently estimating the angle of arrival of the separated digital-analog hybrid antenna system disclosed in the embodiments of the present invention, an approximate range of the direction of an incoming wave is obtained through beam scanning, and the direction of the incoming wave is intelligently estimated by further using an offline trained angle of arrival prediction model.
In the embodiment of the invention, the carrier frequency is assumed to be 60GHz, and the signal source sends an unknown signal without specific requirements on the structure of the unknown signal. The receiving end adopts a separated digital-analog hybrid beam forming framework and mainly comprises a receiving module, an off-line training module and an intelligent DOA prediction module, wherein the receiving module consists of analog beam forming, a radio frequency link, analog-digital conversion, baseband digital signal processing and the like, and the off-line training module adopts a neural network to construct a model. The receiving end is provided with 8 x 8 sub-arrays, each sub-array is provided with 4 x 4 antennas, each antenna is connected with an independent phase shifter for analog beam forming, and each sub-array is connected with a radio frequency link. Devices with different numbers of sub-arrays, and different numbers of antennas on each sub-array, can be obtained by modifying the example in the present embodiment.
The embodiment of the invention discloses an intelligent estimation method for the wave arrival angle of a separated digital-analog hybrid antenna system, which comprises the following specific steps:
the off-line training process of the model comprises the following steps:
step 1: and dividing a space region.
In this example, the receiving end uses DFT codebook, and it should be noted that the patent does not limit the receiving codebook. L th of horizontal DFT codebookhThe individual codewords may be represented as:
wherein, the number M of the antennas in each subarray along the horizontal direction is 4, LhIs the number of DFT codewords in the horizontal direction, where Lh2M, the distance between two antennas in the horizontal direction in the sub-arrayλ is the carrier wavelength, j is the imaginary unit,l th of vertical DFT codebookvThe individual codewords may be represented as:
wherein, the number of the antennas in each subarray along the vertical direction is N-4, LvIs the number of DFT codewords in the vertical direction, where Lv2N, the distance between two antennas in each subarray along the vertical directionAt this time, the three-dimensional DFT codebook of the receiving-end planar array isWherein k is (l)h-1)Lv+lv,K=LhLv. Obviously, the vertical phase of the k-th codewordAnd horizontal phaseSatisfy the requirement of
According to a known receiving codebookThe spatial range is divided into 64 sub-regions. Obviously, each sub-region corresponds to a certain range of vertical and horizontal azimuth angles, which is determined by the main lobe width of the beam corresponding to each codeword.
Step 2: a training sample set is generated. K sub-regions require the generation of K training sample sets.
Taking the k-th sub-region as an example, R10000 directional samples are randomly generated in the space range, that isWhereinIndicating the vertical azimuth and the horizontal azimuth of the r-th directional sample in the k-th sub-region.
For the r-th directional sample, i.e. the signal source is located atTransmitting signals, and adopting the k-th code word by each subarray at a receiving endThe received signal of the first sampling time PQ sub-array can be expressed asWherein, P is 8, and Q is 8, the number of sub-arrays in the horizontal direction of the receiving end, and the number of sub-arrays in the vertical direction of the receiving end;when the signal source in the kth sub-region is located at the r-th directional sample, the receiving end receives the signal at the l-th sampling time by the sub-array at the p-th row and the q-th column.
Then, the L sampling time points are averaged to obtainExtracting real component vector of each subarray average received signalAnd imaginary component vectorsVector obtained by splicing real part and imaginary partAs a sample of the characteristic orientationOf (2), wherein real (·)) Denotes an operation of extracting a real part, and imag (·) denotes an operation of extracting an imaginary part.
Obtaining feature vectors representing respective directions of the rest R-1 direction samples according to the same method, and further obtaining a training sample set of the kth sub-region Is the feature vector of the r direction sample in the k sub-region,is the label of the sample of the r-th direction.
The remaining K-1 sub-regions are used to generate respective training sample sets in the same way.
And step 3: and aiming at each sub-region, training by using the training sample set of the region and adopting the proposed novel neural network to finally obtain the DOA prediction models of a plurality of different sub-regions.
Taking the kth sub-region as an example, the training sample set obtained in step 2 is usedAs input data, the model was trained offline by the mircofishernet neural network. The architecture of the mircofishernet neural network used in this example is shown in fig. 6, and the network architecture is composed of six fully-connected layers, and the number of neurons in each layer is 128, 512, 256, 256, 256, 512. The back of each full connection layer is connected with an activation layer, and the activation function is a Relu function; and a Dropout layer is connected after each active layer, with a random break ratio of 0.5. Two neurons in the last layer of the network model output a vertical azimuth angle theta and a horizontal azimuth angle theta
And training the other K-1 sub-regions according to the same method to obtain a vertical azimuth angle prediction model and a horizontal azimuth angle prediction model suitable for the respective regions.
Angle prediction process:
step 1: and beam scanning, namely roughly positioning the range of the incoming wave signal direction.
Using three-dimensional DFT codebooksEach subarray selects different code words to form beams to receive the transmission signals from the unknown azimuth signal source. Because the receiving end has 64 total PQ ═ 64 sub-arrays and can scan 64 different directions at the same time, that is, different sub-arrays select different code words to form beams pointing to different directions in space, 64 code words in the receiving codebook can be used for beam scanning at the same time. Selecting the code word with the strongest energy corresponding to the received signal according to the received signal of each subarrayFurther locate the incoming wave signal direction at the kth*Within each sub-region.
Step 2: and extracting a direction characteristic vector x of the incoming wave signal.
All the subarrays adopt the code word which is selected in the step 1 and corresponds to the received signal with the strongest energyBeam forming reception signal y (l) [ y ]1,1(l),y1,2(l),…,y1,Q(l),y2,1(l),…,yP,Q(l)]Then averaging the received signals of each sub-array with respect to the number of sampling pointsFinally, the real part of each subarray average received signalAnd imaginary partExtracted and spliced into a vector as a square of an incoming wave signalVector of directional features
And step 3: and refining the angle of the estimated incoming wave signal.
Inputting the direction characteristic vector x of the unknown signal source obtained in the step 2 into a k < th > trained off-line*In the prediction model of the arrival angle of the sub-region, the angle of the wave signal is further refined and estimated.
In order to illustrate the accuracy and effectiveness of the method for intelligently estimating the angle of arrival of the separated digital-analog hybrid antenna system provided by the present invention, the embodiment further provides a simulation graph of the relationship between Root Mean Square Error (RMSE) and Signal-to-Noise Ratio (SNR) of the angle of arrival estimation under the conventional MUSIC algorithm and the intelligent method, as shown in fig. 7, where the number L of sampling points is 1. As can be seen from fig. 7, the performance of the intelligent angle estimation method is better than that of the conventional MUSIC algorithm. Besides, the embodiment also provides a simulation graph of the relation between the root mean square error of the estimation of the arrival angle and the number L of the sampling points under the conventional MUSIC algorithm and the intelligent method, as shown in fig. 8 and fig. 9, wherein the signal-to-noise ratio in fig. 8 is SNR 0dB, and the signal-to-noise ratio in fig. 9 is SNR 10 dB. As can be seen from fig. 8 and fig. 9, as the number of sampling points increases, the estimation accuracy of both the MUSIC algorithm and the intelligent method is gradually improved, but in general, the performance of the intelligent angle estimation method provided by the present invention is far better than that of the conventional algorithm.
The above description is only a preferred embodiment of the present invention, and is not intended to limit the present invention in any way, but any modifications or equivalent variations made according to the technical spirit of the present invention are within the scope of the present invention as claimed.
Claims (4)
1. A method for intelligently estimating the wave arrival angle of a separated digital-analog hybrid antenna system is characterized in that a system device of the method for intelligently estimating the wave arrival angle of the separated digital-analog hybrid antenna system comprises the following steps: the signal source is used for transmitting signals, and the receiving end adopts a separated digital-analog mixed antenna architecture;
the method for intelligently estimating the wave arrival angle of the separated digital-analog hybrid antenna system comprises an offline training process and an angle prediction process of a model;
the off-line training process of the model is as follows;
the whole model training process is completed in an off-line state, and the whole space is divided into a plurality of sub-areas according to a codebook set; then aiming at any space subregion, constructing a training sample set; then, training by adopting the proposed novel neural network to finally obtain a plurality of wave arrival angle prediction models of different sub-regions;
the novel neural network architecture is called as 'MircoFishNet', the shape of the MircoFishNet is similar to that of a fish, the MircoFishNet has the characteristics that the front head is narrow, the middle trunk part is wide firstly and then narrow, and the last wide part is similar to the tail of the fish; meanwhile, the whole network architecture can realize better prediction performance without too many layers, and the MircoFishNet network architecture comprises four parts, namely a narrower network layer, and aims to extract the coarse features, namely the global features, of input data; then a wider network layer, which aims to fully extract the local characteristics of the input data; secondly, a narrower network layer, whose main purpose is to reduce the amount of computation; finally, a wider network layer is used for fully extracting deep features of the data again;
the network architecture consists of six full-connection layers, the number of neurons in each layer is 128, 512, 256, 256, 256 and 512, an activation layer is connected behind each full-connection layer, and the activation function is a Relu function; and a Dropout layer is connected behind each activation layer, the random disconnection ratio is 0.5, and two neurons in the last layer of the network model output a vertical azimuth angle theta and a horizontal azimuth angle theta
The angle prediction process of the model is as follows;
firstly, each subarray of a receiving end selects different code words from a known receiving codebook to receive a transmitting signal from an unknown azimuth signal source, selects the code word with the strongest energy corresponding to the received signal, and further positions a sub-region where a wave signal is located; secondly, all the sub-arrays receive signals by using the selected code words with the strongest energy corresponding to the received signals, and feature vectors containing incoming wave signal directions are extracted from the received signals of all the sub-arrays;
all the submatrices select the code word with the strongest energy corresponding to the received signalBeam forming reception signal y (l) [ y ]1,1(l),y1,2(l),…,y1,Q(l),y2,1(l),…,yP,Q(l)]Then averaging the received signals of each sub-array with respect to the number of sampling pointsFinally, the real part of each subarray average received signalAnd imaginary partExtracted and spliced into a vector serving as a direction characteristic vector of an incoming wave signal
And then inputting the extracted feature vector into an offline trained wave arrival angle prediction model corresponding to the sub-region, and further finely estimating the wave arrival angle.
2. The method according to claim 1, wherein the method comprises: the receiving end adopts a separated digital-analog hybrid beam forming framework and comprises a receiving module, an offline training module and an intelligent DOA (angle of arrival) prediction module, wherein the receiving module consists of analog beam forming, a radio frequency link, analog-digital conversion and baseband digital signal processing, the offline training module adopts a neural network to construct a model, the receiving end is provided with P × Q sub-arrays, P sub-arrays in the horizontal direction and Q sub-arrays in the vertical direction, each sub-array consists of M × N antennas, M antennas in the horizontal direction and N antennas in the vertical direction, each antenna is connected with an independent phase shifter for analog beam forming, and each sub-array is connected with a radio frequency link.
3. The method according to claim 1, wherein the method comprises: the off-line training process of the model comprises the following specific steps;
step 1: dividing a space area;
according to a known receiving codebookDividing the spatial range into K sub-regions, whereinAndthe vertical phase and the horizontal phase of the kth code word are respectively represented, K is the number of the code words, obviously, each subregion corresponds to a certain range of vertical azimuth angles and horizontal azimuth angles, which is determined by the main lobe width of the beam corresponding to each code word;
step 2: generating a training sample set;
k training sample sets need to be generated in K sub-regions;
taking the kth sub-region as an example, R directional samples are randomly generated in the space range, i.e.WhereinRepresenting a vertical azimuth and a horizontal azimuth of an r-th directional sample in a k-th sub-region;
for the r directional sample, the signal source is locatedTransmitting signals, receiving signals by each subarray of a receiving end by using a k-th code word, extracting and splicing real parts and imaginary parts of the received signals of each subarray into a vectorUsing the vector as a sample of the characterizing directionFinally obtaining a training sample set of the kth sub-region Is the feature vector of the r-th directional sample,a label for the sample of the r-th direction;
generating respective training sample sets of the other K-1 sub-regions according to the same method;
and step 3: training a model off line;
aiming at each sub-region, training a wave arrival angle prediction model suitable for the region by using a training sample set of the region and adopting an intelligent machine learning algorithm in an off-line manner;
taking the kth sub-region as an example, the training sample set obtained in step 2 is usedInputting the new neural network, training the parameters of each neuron, obtainingObtaining a vertical azimuth angle and horizontal azimuth angle prediction model suitable for the sub-region;
and training the other K-1 sub-regions according to the same method to obtain a vertical azimuth angle prediction model and a horizontal azimuth angle prediction model suitable for the respective regions.
4. The method according to claim 1, wherein the method comprises: the angle prediction process of the model comprises the following specific steps;
step 1: beam scanning, roughly positioning the range of incoming wave signal directions;
using known receive codebooksEach subarray selects different code words to form wave beams to receive the transmitted signals from the unknown azimuth signal source, and selects the code word with the strongest energy corresponding to the received signals according to the received signals of each subarrayFurther locate the incoming wave signal direction at the kth*Within each sub-region;
step 2: extracting a direction characteristic vector x of an incoming wave signal;
all the subarrays adopt the code word which is selected in the step 1 and corresponds to the received signal with the strongest energyAs receiving beams, extracting and splicing the real part and the imaginary part of each subarray receiving signal into a vector serving as a direction characteristic vector x of an incoming wave signal;
and step 3: finely estimating the angle of an incoming wave signal;
inputting the direction characteristic vector x of the unknown signal source obtained in the step 2 into a k < th > trained off-line*And in the prediction model of the vertical azimuth angle and the horizontal azimuth angle of each sub-region, the angle of the wave signal is further estimated in a refined mode.
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 CN110113088A (en) | 2019-08-09 |
CN110113088B true 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) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
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 |
CN112731279B (en) * | 2020-12-09 | 2022-04-12 | 复旦大学 | Arrival angle estimation method based on hybrid antenna subarray |
CN112803976B (en) * | 2020-12-24 | 2022-07-08 | 浙江香农通信科技有限公司 | Large-scale MIMO precoding method and system and electronic equipment |
CN114978382A (en) * | 2021-02-22 | 2022-08-30 | 华为技术有限公司 | Data processing method and communication device based on space division |
EP4331127A1 (en) * | 2021-04-29 | 2024-03-06 | 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 |
---|
Angle Domain Channel Estimation in Hybrid Millimeter Wave Massive MIMO Systems;Dian Fan et al;《IEEE Transactions on Wireless Communications》;20181231;第17卷(第12期);第8165-8179页 * |
Angle of arrival estimator based on artificial neural networks;Evgeny Efimov et al;《2016 17th International Radar Symposium》;20160623;全文 * |
Deep Learning for Super-Resolution Channel Estimation and DOA Estimation Based Massive MIMO System;Hongji Huang et al;《IEEE Transactions on Vehicular Technology》;20180930;第67卷(第9期);第8549-8560页 * |
Robust Unambiguous Estimation of Angle-of-Arrival in Hybrid Array With Localized Analog Subarrays;Kai Wu et al;《IEEE Transactions on Wireless Communications》;20180531;第17卷(第5期);第2987-3002页 * |
基于PSO-BP神经网络的DOA估计方法;孟非,王旭;《电讯技术》;20120531;第52卷(第5期);第694-698页 * |
采用混合核支持向量机的DOA估计;孙菲艳 等;《电讯技术》;20160331;第56卷(第3期);第302-307页 * |
Also Published As
Publication number | Publication date |
---|---|
CN110113088A (en) | 2019-08-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110113088B (en) | Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system | |
CN113411110B (en) | Millimeter wave communication beam training method based on deep reinforcement learning | |
CN106031068A (en) | Beamforming-based communication method and apparatus | |
CN106355245A (en) | Method for integrating array antenna directional images on basis of neural network algorithms | |
CN109714091B (en) | Iterative hybrid precoding method based on hierarchical design in millimeter wave MIMO system | |
CN108650200B (en) | Low-frequency auxiliary channel estimation method of high-frequency and low-frequency hybrid networking system | |
CN115085774B (en) | Common sense fusion hybrid beamforming method based on Keramelteon | |
CN108337031A (en) | Mixing angle-of- arrival estimation based on Root-MUSIC in extensive MIMO | |
CN107104720A (en) | The relatively prime array adaptive beamforming method rebuild based on covariance matrix virtual Domain discretization | |
Ma et al. | Deep learning assisted beam prediction using out-of-band information | |
CN107135023A (en) | Three-dimensional training code book design method and beam alignment for millimeter-wave communication system | |
Zhuang et al. | Machine-learning-based high-resolution DOA measurement and robust directional modulation for hybrid analog-digital massive MIMO transceiver | |
CN110233649B (en) | Dynamic subarray design method based on millimeter wave MIMO system | |
Elbir et al. | Robust hybrid beamforming with quantized deep neural networks | |
CN113644941A (en) | Fuzzy phase rapid elimination method based on large-scale MIMO receiving array structure | |
Kao et al. | AI-aided 3-D beamforming for millimeter wave communications | |
WO2023103356A1 (en) | Near-field broadband beam forming method and apparatus, electronic device, and storage medium | |
Chen et al. | Multi‐beam receive scheme for millimetre wave wireless communication system | |
CN114499615B (en) | Near-far field unified transmitting beam forming method in terahertz communication system | |
CN113242069B (en) | Codebook design method based on neural network | |
CN112731279B (en) | Arrival angle estimation method based on hybrid antenna subarray | |
CN113917388A (en) | DOA estimation and positioning method using hybrid ADC | |
Song et al. | Deep learning based low-rank channel recovery for hybrid beamforming in millimeter-wave massive MIMO | |
Thazeen et al. | An efficient reconfigurable optimal source detection and beam allocation algorithm for signal subspace factorization | |
Bing | A simplified genetic algorithm in multi-beam combination for mmWave communication system |
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 |