CN102868432A - Blind beam forming device and method under dual-stage neural network - Google Patents
Blind beam forming device and method under dual-stage neural network Download PDFInfo
- Publication number
- CN102868432A CN102868432A CN2012103313935A CN201210331393A CN102868432A CN 102868432 A CN102868432 A CN 102868432A CN 2012103313935 A CN2012103313935 A CN 2012103313935A CN 201210331393 A CN201210331393 A CN 201210331393A CN 102868432 A CN102868432 A CN 102868432A
- Authority
- CN
- China
- Prior art keywords
- signal
- stage
- processing module
- output
- unit
- 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
Landscapes
- Variable-Direction Aerials And Aerial Arrays (AREA)
- Radio Transmission System (AREA)
Abstract
The invention discloses a blind beam forming device under a dual-stage neural network, which comprises a signal preprocessor, a dual-stage neural network signal processing module and a signal post-processing module. A blind beam forming method comprises the steps of: receiving a signal, processing a vector, carrying out judging drive by sectors, generating a weight vector, and completing mapping of an optimal weight. The blind beam forming device has the advantages that the structure is simple, the convenience is brought for use, a blind beam can be rapidly approached to form the optimal weight by adopting the blind beam forming method, the channel attenuation can be lightened, and an interference signal is inhibited so that the communication quality and the channel capacity are improved.
Description
(1) technical field:
The present invention relates to automation field and blind wave beam and form the field, especially blind beam-forming device under a kind of pair of stage neural net and forming method thereof.
(2) background technology:
Use intelligent antenna technology that channel capacity is improved in the mobile communication.One of main task in intelligent antenna technology is exactly how to realize in real time wave beam forming.Only have and to realize the intelligent antenna beam shaping algorithm after having solved this task.Smart antenna can adaptive judgement multiple source quantity and direction, and trace into desired signal, then can in down link, produce maximum gain to the wave beam of launching in the desired signal direction by beamforming algorithm, produce darker zero falling at interference radiating way.
In the process of wave beam forming, the character of signal or channel is unknown and does not need the transmitting training signal or know the priori such as array direction vector, then is referred to as blind wave beam and forms.At the receiving terminal of smart antenna, the signal of receiving is multipath signal often, and signal source quantity is often greater than array elements quantity.So higher requirement has been proposed blind beamforming algorithm.
In the collinear array of M array element, establishing from the space has the individual narrow band signal of K(K<M), and the frequency of these narrow band signals is at ω
0Near, incident angle is { θ
1, θ
2, θ
3..., θ
K}:
Formula 1:
i=1,2,…,M
Wherein, s
m(t) be m signal source, interference signal is e
r(t), n
i(t) be the white Gaussian noise of the zero-mean that receives of i array element, λ is carrier wavelength, and d is the array element interval.Also formula one can be expressed as the form of matrix:
Formula 2:X (t)=AS (t)+A
EE (t)+N (t),
Wherein, X (t)=[x
1(t) x
2(t) ... x
M(t)]
T, e (t)=[e
1(t) e
2(t) ... e
K(t)]
T,
, N (t)=[n
1(t) n
2(t) ... n
M(t)]
T, S (t)=[s
1(t) s
2(t) ... s
K(t)]
T, A=[a (θ
1) a (θ
2) ... a (θ
k)]
T
In above-mentioned formula, subscript T represents transposed matrix.A (θ
k)---desired signal S (t) gain of k array element on direction of arrival; T---sampling instant, once sampling is called snap one time, usually can make ARRAY PROCESSING to each time snap first, again the result of snap is many times carried out statistical average.
For establishment disturbs and noise, with the signal weighting summation of obtaining on each bay, make the gain on the desired signal direction maximum, namely wave beam forms:
Formula 3:y (k)=W
HX (k)
Wherein: W=[w
1, w
2... w
M]
TBe weighing vector.And the target that wave beam forms is according to the needed property indices of real-time system, forms the optimal allocation to base band (intermediate frequency) signal.This distribution is used for compensating in the radio transmission process because signal fadeout and distortion that the factors such as space loss, multipath effect are introduced reduce the cochannel interference between the user simultaneously.The weighing vector that W in formula three then forms for wave beam, namely adaptive receiver needs the parameter of regulating in the detection signal process.
(3) summary of the invention:
The object of the present invention is to provide a kind of pair of blind beam-forming device under the stage neural net and forming method thereof, it can overcome the deficiencies in the prior art, is a kind of device of simple in structure, strong operability, and its to form method applicability strong, simple.
Technical scheme of the present invention: the blind beam-forming device under a kind of pair of stage neural net is characterized in that it comprises signal preprocessor, two stage neural net signal processing module and signal post-processing module; Wherein, the input of described signal preprocessor receives from aerial array and through the digital signal after the A/D conversion, and its output connects the input of two stage neural net signal processing modules; The output of described pair of stage neural net signal processing module connects the input of signal post-processing module; The output output best weight value vector signal of described signal post-processing module.
Described pair of stage neural net signal processing module is by being no less than 2 PNN(Probabilistic Neural Network) the sub-probabilistic neural network module of phase I of unit with by the GRNN(General Regression Neural Network identical with phase I PNN element number) the second stage generalized regression nerve networks module composition of unit; Wherein, the input of the PNN unit in the sub-probabilistic neural network module of described phase I is connected by the output of bus with the signal processing module, and the input of the GRNN unit in its output and the second stage generalized regression nerve networks module is and connects one to one; The input of the GRNN unit in the described second stage generalized regression nerve networks module is connected by the output of bus with the signal processing module, and its output is connected by the input of bus with the signal post-processing module.
Blind Beamforming Method under a kind of above-mentioned pair of stage neural net is characterized in that it may further comprise the steps:
1. aerial array receives the signal source from 0 ° to 180 °, and when receiving certain signal source and be incident to antenna, antenna inputs to signal pre-processing module with the signal vector X (t) that receives, to obtain the autocorrelation matrix R of signal
XX, and extract autocorrelation matrix R
XXUpper triangle element, and each element is divided into two elements according to real part and imaginary part, form dimension and be the new vectorial b of M * (M-1), and then the b vector is carried out normalized:
2. the z vector that step is obtained in 1. inputs in the sub-probabilistic neural network module of phase I; Each PNN unit in this mixed-media network modules mixed-media will to the input signal adjudicate, if the signal source that is: is incident to antenna in its corresponding sector, then PNN unit corresponding to this sector is output as " 1 "; Otherwise, output " 0 ", send this signal to second stage generalized regression nerve networks module, if " 1 " signal then second stage generalized regression nerve networks module is activated, when the PNN unit that is to say the phase I is output as " 1 ", just can work in the second stage GRNN unit corresponding with it;
3. 0/1 signal that the z vector sum step that step is produced in 1. produces in 2. inputs to second stage generalized regression nerve networks module; The GRNN unit that is activated is processed the z vector, produces weight vector W
0
4. the signal post-processing module is finished by weight vector W
0To optimum weights W
OptMapping, that is:
, wherein
。
Operation principle of the present invention: signal pre-processing module is used for generation and inputs to one-phase neural net signal; Post-processing module becomes the best weight value vector with the output corresponding conversion in order to network that neural net is output as input; Pretreatment module is carried out corresponding preliminary treatment with the digital signal that aerial array receives, and the signal of this module output enters neural net and is for further processing.
Preliminary treatment work comprises: the autocorrelation matrix R that 1, obtains signal X (t)
XX2, extract autocorrelation matrix R
XXUpper triangle element; 3, with the real part of the element that extracts and the vectorial b of imaginary component open form Cheng Xin, obtain its normalized vector
Phase I is made of the PNN that is no less than two.In this stage, whole angular range (field range of aerial array) is no less than two sub-PNN and divides for L sector.Whether the purpose of every sub-PNN exists one or more signal sources for judgement in corresponding sector, and produces output.By this work can effectively reduce sub-PNN the training set size so that the training time shorten dramatically.In this stage, the input layer number of PNN is that M * (M-1), the neuron number of output layer is 1.Therefore the work of phase I network is in fact to utilize PNN to adjudicate classification.
Second stage is made of a plurality of GRNN identical with one-phase PNN quantity.The reason of selecting GRNN is that the study of network all relies on data sample because its artificial parameter of regulating is few.These characteristics have determined that network avoided the impact of artificial subjective supposition on predicting the outcome to greatest extent.The GRNN in this stage then is the key component that realizes this function.
Post-processing module will be made corresponding processing for the signal of part of neural network output and be produced final needed optimum weights.
Superiority of the present invention is: simple in structure, easy to operate, blind Beamforming Method can approach fast blind wave beam and form optimum weights; Can alleviate channel fading and suppress interference signal and improve communication quality and channel capacity.
(4) description of drawings:
Fig. 1 is the structured flowchart of the blind beam-forming device under the related a kind of pair of stage neural net of the present invention;
Fig. 2 is step schematic diagram 2. in the blind Beamforming Method under the related a kind of pair of stage neural net of the present invention.
(5) embodiment:
Embodiment: the blind beam-forming device (see figure 1) under a kind of pair of stage neural net is characterized in that it comprises signal preprocessor, two stage neural net signal processing module and signal post-processing module; Wherein, the input of described signal preprocessor receives from aerial array and through the digital signal after the A/D conversion, and its output connects the input of two stage neural net signal processing modules; The output of described pair of stage neural net signal processing module connects the input of signal post-processing module; The output output best weight value vector signal of described signal post-processing module.
Described pair of stage neural net signal processing module (see figure 1) is by sub-probabilistic neural network module of phase I and second stage generalized regression nerve networks module composition; Sub-probabilistic neural network module of wherein said phase I is by 3 PNN(Probabilistic Neural Network) unit forms, and is respectively PNN unit 1A, PNN unit 1B, PNN unit 1C; Described second stage generalized regression nerve networks module is by 3 GRNN(General Regression Neural Network) cell formation, be respectively GRNN unit 2A, GRNN unit 2B and GRNN unit 2C; Wherein, the input of described 3 PNN unit is connected by the output of bus with the signal processing module, and the input of its output and 3 GRNN unit is and connects one to one; The input of the GRNN unit in the described second stage generalized regression nerve networks module is connected by the output of bus with the signal processing module, and its output is connected by the input of bus with the signal post-processing module.
Blind Beamforming Method under a kind of above-mentioned pair of stage neural net is characterized in that it may further comprise the steps:
1. aerial array receives the signal source from 0 ° to 180 °, and when receiving certain signal source and be incident to antenna, antenna inputs to signal pre-processing module with the signal vector X (t) that receives, to obtain the autocorrelation matrix R of signal
XX, and extract autocorrelation matrix R
XXUpper triangle element, and each element is divided into two elements according to real part and imaginary part, form dimension and be the new vectorial b of M * (M-1), and then the b vector is carried out normalized:
2. the z vector that step is obtained in 1. inputs in the sub-probabilistic neural network module of phase I; The sector that three PNN unit in this mixed-media network modules mixed-media are administered is respectively the sector of responsible 0 ° to 60 ° of 1A, and 1B is responsible for 61 ° to 120 ° sector, and 1C is responsible for 121 ° to 180 ° sector; Each sector will be adjudicated the signal of input, be 90 ° if that is: be incident to the signal source of antenna, and then the PNN unit of 1B then exports 1, and activate the GRNN unit 2B corresponding with it in the second stage generalized regression nerve networks module; (see figure 2)
3. simultaneously, the z vector that step is produced in 1. inputs to second stage generalized regression nerve networks module; The GRNN unit 2B that is activated starts working, and it is 61 ° to 120 ° GRNN network that 2B is trained to quick output signal source, and 2B exports corresponding weights W
0
4. the signal post-processing module is finished by weight vector W
0To optimum weights W
OptMapping, that is:
, wherein
。
Claims (3)
1. the blind beam-forming device under two stage neural nets is characterized in that it comprises signal preprocessor, two stage neural net signal processing module and signal post-processing module; Wherein, the input of described signal preprocessor receives from aerial array and through the digital signal after the A/D conversion, and its output connects the input of two stage neural net signal processing modules; The output of described pair of stage neural net signal processing module connects the input of signal post-processing module; The output output best weight value vector signal of described signal post-processing module.
2. the blind beam-forming device under the described a kind of pair of stage neural net according to claim 1 is characterized in that described pair of stage neural net signal processing module is by sub-probabilistic neural network module of the phase I that is no less than 2 PNN unit and second stage generalized regression nerve networks module composition by the GRNN unit identical with phase I PNN element number; Wherein, the input of the PNN unit in the sub-probabilistic neural network module of described phase I is connected by the output of bus with the signal processing module, and the input of the GRNN unit in its output and the second stage generalized regression nerve networks module is and connects one to one; The input of the GRNN unit in the described second stage generalized regression nerve networks module is connected by the output of bus with the signal processing module, and its output is connected by the input of bus with the signal post-processing module.
3. the blind Beamforming Method under described pair of stage neural net of a claim 1 is characterized in that it may further comprise the steps:
1. aerial array receives the signal source from 0 ° to 180 °, and when receiving certain signal source and be incident to antenna, antenna inputs to signal pre-processing module with the signal vector X (t) that receives, to obtain the autocorrelation matrix R of signal
XX, and extract autocorrelation matrix R
XXUpper triangle element, and each element is divided into two elements according to real part and imaginary part, form dimension and be the new vectorial b of M * (M-1), and then the b vector is carried out normalized:
2. the z vector that step is obtained in 1. inputs in the sub-probabilistic neural network module of phase I; Each PNN unit in this mixed-media network modules mixed-media will to the input signal adjudicate, if the signal source that is: is incident to antenna in its corresponding sector, then PNN unit corresponding to this sector is output as " 1 "; Otherwise, output " 0 ", send this signal to second stage generalized regression nerve networks module, if " 1 " signal then second stage generalized regression nerve networks module is activated, when the PNN unit that is to say the phase I is output as " 1 ", just can work in the second stage GRNN unit corresponding with it;
3. 0/1 signal that the z vector sum step that step is produced in 1. produces in 2. inputs to second stage generalized regression nerve networks module; The GRNN unit that is activated is processed the z vector, produces weight vector W
0
4. the signal post-processing module is finished by weight vector W
0To optimum weights W
OptMapping, that is:
, wherein
。
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210331393.5A CN102868432B (en) | 2012-09-07 | 2012-09-07 | Blind adaptive beamforming device under a kind of pair of stage neural net and forming method thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201210331393.5A CN102868432B (en) | 2012-09-07 | 2012-09-07 | Blind adaptive beamforming device under a kind of pair of stage neural net and forming method thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102868432A true CN102868432A (en) | 2013-01-09 |
CN102868432B CN102868432B (en) | 2015-08-19 |
Family
ID=47447062
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201210331393.5A Active CN102868432B (en) | 2012-09-07 | 2012-09-07 | Blind adaptive beamforming device under a kind of pair of stage neural net and forming method thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102868432B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103106350A (en) * | 2013-03-08 | 2013-05-15 | 上海市电力公司 | Line loss analysis data reduction method based on artificial neural network |
CN105488565A (en) * | 2015-11-17 | 2016-04-13 | 中国科学院计算技术研究所 | Calculation apparatus and method for accelerator chip accelerating deep neural network algorithm |
CN108599809A (en) * | 2018-03-14 | 2018-09-28 | 中国信息通信研究院 | Full duplex self-interference signal number removing method and device |
CN109547083A (en) * | 2018-11-08 | 2019-03-29 | 广东工业大学 | A kind of flat-topped beam shaping method neural network based |
CN109617584A (en) * | 2019-01-08 | 2019-04-12 | 南京邮电大学 | A kind of mimo system beamforming matrix design method based on deep learning |
CN110690930A (en) * | 2019-09-27 | 2020-01-14 | 清华大学 | Information source number detection method and device |
CN110995327A (en) * | 2019-12-17 | 2020-04-10 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
CN112640325A (en) * | 2018-08-29 | 2021-04-09 | 华为技术有限公司 | Apparatus and method for CSI-assisted beamforming using learning network entities |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0979027A2 (en) * | 1998-08-07 | 2000-02-09 | General Electric Company | Neural network prediction for radiographic x-ray exposures |
CN102638296A (en) * | 2012-03-03 | 2012-08-15 | 天津理工大学 | Anti-interference method based on intelligent antenna and neural network algorithm |
-
2012
- 2012-09-07 CN CN201210331393.5A patent/CN102868432B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0979027A2 (en) * | 1998-08-07 | 2000-02-09 | General Electric Company | Neural network prediction for radiographic x-ray exposures |
CN102638296A (en) * | 2012-03-03 | 2012-08-15 | 天津理工大学 | Anti-interference method based on intelligent antenna and neural network algorithm |
Non-Patent Citations (2)
Title |
---|
孙绪宝、钟顺时: "基于神经网络的盲波束形成", 《电波科学学报》 * |
田丹、汪晋宽、薛桂芹: "一种基于神经网络的盲波束形成算法", 《仪器仪表学报》 * |
Cited By (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103106350A (en) * | 2013-03-08 | 2013-05-15 | 上海市电力公司 | Line loss analysis data reduction method based on artificial neural network |
CN103106350B (en) * | 2013-03-08 | 2016-01-20 | 上海市电力公司 | Based on the line loss analyzing data restoration method of artificial neural network |
CN105488565A (en) * | 2015-11-17 | 2016-04-13 | 中国科学院计算技术研究所 | Calculation apparatus and method for accelerator chip accelerating deep neural network algorithm |
CN108599809A (en) * | 2018-03-14 | 2018-09-28 | 中国信息通信研究院 | Full duplex self-interference signal number removing method and device |
CN112640325A (en) * | 2018-08-29 | 2021-04-09 | 华为技术有限公司 | Apparatus and method for CSI-assisted beamforming using learning network entities |
CN112640325B (en) * | 2018-08-29 | 2022-09-02 | 华为技术有限公司 | Apparatus and method for CSI-assisted beamforming using learning network entities |
CN109547083A (en) * | 2018-11-08 | 2019-03-29 | 广东工业大学 | A kind of flat-topped beam shaping method neural network based |
CN109547083B (en) * | 2018-11-08 | 2022-01-25 | 广东工业大学 | Flat-top beam forming method based on neural network |
CN109617584A (en) * | 2019-01-08 | 2019-04-12 | 南京邮电大学 | A kind of mimo system beamforming matrix design method based on deep learning |
CN109617584B (en) * | 2019-01-08 | 2021-12-21 | 南京邮电大学 | MIMO system beam forming matrix design method based on deep learning |
CN110690930A (en) * | 2019-09-27 | 2020-01-14 | 清华大学 | Information source number detection method and device |
CN110690930B (en) * | 2019-09-27 | 2020-12-15 | 清华大学 | Information source number detection method and device |
CN110995327A (en) * | 2019-12-17 | 2020-04-10 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
CN110995327B (en) * | 2019-12-17 | 2021-05-04 | 电子科技大学 | Mixed beam forming optimization method and system for multi-carrier MIMO system |
Also Published As
Publication number | Publication date |
---|---|
CN102868432B (en) | 2015-08-19 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102868432A (en) | Blind beam forming device and method under dual-stage neural network | |
CN110113088B (en) | Intelligent estimation method for wave arrival angle of separated digital-analog hybrid antenna system | |
CN103765940B (en) | Sector configuration method and device, system | |
CN104917554A (en) | Base station and beam forming method | |
CN105158735B (en) | Null tone Two-Dimensional Spectral Estimation method based on compression sampling array | |
CN110708125B (en) | Method for constructing dynamic target channel and air interface test system | |
Shubair et al. | A Setup for the Evaluation of MUSIC and LMS Algorithms for a Smart Antenna System. | |
CN107104720A (en) | The relatively prime array adaptive beamforming method rebuild based on covariance matrix virtual Domain discretization | |
CN107017931A (en) | The method and device that a kind of beam side lobe suppresses | |
CN114051225B (en) | Resource allocation method and device based on RIS auxiliary D2D secret communication | |
CN110071747B (en) | Low-complexity quantization bit selection method for uplink of large-scale MIMO system | |
WO2017121175A1 (en) | Data processing method and device | |
Kadur et al. | Effective beam alignment algorithm for low cost millimeter wave communication | |
Kang et al. | Smart antenna for emerging 5G and application | |
CN107276657B (en) | Millimeter wave mixed beam forming method based on parallel algorithm | |
CN117240331A (en) | No-cellular network downlink precoding design method based on graph neural network | |
JP3440298B2 (en) | Array antenna directivity control method | |
WO2023103356A1 (en) | Near-field broadband beam forming method and apparatus, electronic device, and storage medium | |
CN113726390A (en) | Energy efficiency maximization transmission system and method based on reconfigurable reflection super surface | |
CN113726389A (en) | Cell-free networking system and method based on reconfigurable reflection super surface | |
WO2016201984A1 (en) | Data transmission method and base station | |
CN108347265B (en) | Beam receiving diversity method and device by using space sparse characteristic | |
CN101771197A (en) | Method for determining order of cables and device therefor | |
CN105656528B (en) | A kind of information processing method and electronic equipment | |
CN113644945B (en) | Codebook determination method, device, equipment and storage medium |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
TR01 | Transfer of patent right |
Effective date of registration: 20230801 Address after: 710000, Room 6501, 2nd Floor, Building 4, Free Trade Industrial Park, No. 2168 Zhenghe Fourth Road, Fengdong New City, Xixian New District, Xi'an City, Shaanxi Province Patentee after: Xixian New Area Yachuang Taiyuan Network Technology Co.,Ltd. Address before: 300384 No. 391 Binshui West Road, Xiqing District, Tianjin Patentee before: TIANJIN University OF TECHNOLOGY |
|
TR01 | Transfer of patent right |