CN105915299B - Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band - Google Patents
Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band Download PDFInfo
- Publication number
- CN105915299B CN105915299B CN201610149087.8A CN201610149087A CN105915299B CN 105915299 B CN105915299 B CN 105915299B CN 201610149087 A CN201610149087 A CN 201610149087A CN 105915299 B CN105915299 B CN 105915299B
- Authority
- CN
- China
- Prior art keywords
- frequency
- time
- vector
- ism band
- csi
- 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
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/373—Predicting channel quality or other radio frequency [RF] parameters
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Electromagnetism (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Quality & Reliability (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
A kind of spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM (2.4GHz) frequency range is claimed in the present invention, including:The calculating of ISM band correlation, by obtaining ISM band time domain and the correlation of frequency domain to the actual measurement, quantization and correlation analysis of ISM band;Second step:Based on ISM band time-frequency correlation, time-frequency two-dimensional LMBP neural networks are built to realize the prediction of ISM band;Third walks:Time-frequency input vector using measured data as neural network and object vector are realized the repetitive exercise of time-frequency two-dimensional LMBP neural networks with Newton method learning rules, obtained by the optimal solution of weight w between neural network node and threshold value b the parameter vector u constituted;4th step:The spectrum prediction of ISM band is realized to train the two-dimentional LMBP neural networks completed.On the basis of calculating ISM band time-frequency correlation, by the spectrum prediction of time-frequency two-dimensional LMBP neural fusion ISM bands, this method has precision of prediction high compared with time domain LMBP neural networks and Markov algorithms, the short advantage of training convergence time.
Description
Technical field
The invention belongs to cognitive radio frequency spectrums to predict field, and time-frequency two-dimensional LMBP is based on specifically in ISM band
The spectrum prediction method of neural network.
Background technology
The fast development of Internet of Things industry causes the internet of things equipment for being operated in ISM (2.4GHz) frequency range increasingly to increase, should
The problem of co-channel interference of frequency range is on the rise, it is contemplated that ISM band is unauthorized frequency range, the pole small-power network (such as ZigBee)
It is easily fallen into oblivion by the interference of this kind of high power equipment of similar WLAN, causes the paralysis of the data frame losing or even whole network between node
Paralysis.Know that the occupied information of frequency range is to solve ISM band equipment room to be compatible with coexistence problems in advance by spectrum prediction algorithm
A kind of effective way.
From document " the Reliable Open Spectrum of Kumar Acharya PA, Singh S, Zheng H
Communications through Proactive Spectrum”(PROCEEDINGS OF FIRST INTERNATIONAL
WORKSHOP ON TECHNOLOGY AND POLICY FOR ACCESSING SPECTRUM, AUGUST, 2006) it is put forward for the first time
Forecasting mechanism is applied to deduce time-domain spectral hole go out current moment and duration after, all kinds of prediction algorithms are in frequency spectrum
It is widely used in prediction.
Sixing Yin, Dawei Chen, Qian Zhang, Mingyan Liu and Shufang are disclosed in the prior art
Document " the Mining spectrum usage data of Li:a large-scale spectrum measurement study”
(IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL.11, NO.6, JUNE 2012), this article proposes one
Kind detailed frequency spectrum data measurement method, to the letter of the channel idle data of 20MHz -3GHz frequency ranges, each independent wireless service
Road utilization rate and the time-frequency correlation of channel have carried out statistical analysis, and propose the 2D based on frequency spectrum correlation with this
Frequent pattern mining algorithms realize the prediction of channel availability, but the algorithm occupies in rule from frequency spectrum and dig
Excavate prediction rule it is minimum trained when it is 2 hours a length of, be otherwise difficult to ensure the precision of prediction and omission factor of algorithm, the algorithm pair
The larger feature of ISM band channel time variation is not met in the limitation of minimum trained duration.
Document " the spectrum prediction algorithm research in the cognitive radio system " (master of Chen Binhua is disclosed in the prior art
Academic dissertation Beijing University of Post & Telecommunications, 2011), single order is respectively adopted in time domain related features of this article based on GSM frequency ranges
Markov chain and time domain LMBP neural networks realize the spectrum prediction of GSM900, GSM1800 frequency range, and by establishing more than one
The spectral model of channels associated, to realize multi channel joint spectrum prediction, but this method only accounts for GSM frequency ranges time slot correlation
Property for spectrum prediction usable value, the frequency domain correlation properties of frequency range are not verified, and use single order Markov chain
Realize that the spectrum prediction of ISM band, the training convergence time and precision of prediction of algorithm are difficult to reach with time domain LMBP neural networks
To ideal equilibrium.
By correlative study it is found that the basic skills of spectrum prediction algorithm is the frequency spectrum measured data based on specific frequency range, obtain
The frequency spectrum correlation properties for knowing specific frequency range are occupied in rule from frequency spectrum and excavate feasible prediction rule, proposed to existing with this
The improvement of prediction algorithm.The present invention is based on the time-frequency correlation properties that measured data obtains ISM band, propose a kind of time-frequency two-dimensional
LMBP neural networks realize the spectrum prediction of ISM band by the parallel training of time-frequency input vector.
Invention content
For the above existing deficiency, it is proposed that a kind of method.Technical scheme is as follows:In a kind of ISM band
Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks comprising step:
Step 1:ISM band measured data is collected, the time-frequency correlation between ISM band time domain and frequency domain is calculated;
Step 2:Time-frequency correlation based on ISM band builds two-dimensional prediction matrix, with parallel in the first layer network
Mode realizes the common training of time domain and frequency domain, obtains the first layer network output vector Y1t,Y2t, by the first layer network export to
Measure Y1t,Y2tAs the input vector of the second layer network, final predicted value Y is obtainedt, that is, build time-frequency two-dimensional LMBP neural networks
Realize the spectrum prediction of ISM band;
Step 3:Using the measured data of the ISM band obtained in step 1 as training sequence, by adjusting formula, with
Error function is condition, completes the repetitive exercise of time-frequency two-dimensional LMBP neural networks, the optimal solution of parameter vector u is obtained, with this
Obtain the weight vector w and threshold vector b of neural network;
Step 4:Build time domain and frequency domain input vector matrix Xt, Xf:By Xt, XfThe time-frequency two-dimensional built by step 2
LMBP neural networks obtain output vector Y=[Y1,Y2,Y3,...,Ym]T, YmAs CSI (tm,cm) predicted value, complete frequency spectrum
Prediction.
Further, the time domain X of the step 1 and step 21=[CSI (t-1, c), CSI (t-2, c) ..., CSI (t-
Δt,c)]TWith frequency domain X2=[CSI (t, c ± 1), CSI (t, c ± 2) ..., CSI (t, c ± Δ f)]T, shown CSI (t- Δ t,
C) indicate that the moment, (channel state information of t- Δ t) channels c, (t, c ± Δ f) indicated the moment t channel (channel of c ± Δs f) to CSI
Status information.
Further, the adjustment formula of the step 3 is:
Adjust formulauk+1Indicate the parameter that next step iteration obtains
Vector, ukIndicate the parameter vector of current iteration, J (uk) indicate e (uk) Jacobian matrixes,Indicate Tiny increment dt unit square
Battle array, e (uk) indicating error vector, for adjustment formula using error function F (u)≤ε as condition, ε indicates preset target error.
Further, in step 1, after collecting ISM band frequency spectrum measured data, the correlation R (Δs of the frequency range time domain
T), (calculating of Δ f) is as follows by the correlation R of frequency domain:
Wherein, the quantitative formula of CSI (t, c) is:
T=1,2 ..., T;C=1,2 ..., C, R
(t, c) and CSI (t, c) indicate the measured power value and channel state information of moment t lower channel c respectively;
The degree of correlation of two 0-1 sequencesIt is defined as follows:
The formula is usually used in assessing the correlation of two binary sequences, and wherein I (A) is discriminant function, if A values are true, then I
(A)=1, otherwise I (A)=0, by R, ((correlation curve of Δ f) obtains ISM band between time slot and neighbouring frequency point by Δ t), R
Correlation.
Further, in step 2, frequency domain prediction point is added in neural network input vector, builds two-dimensional prediction square
Battle array, realizes time domain X in a parallel fashion in the first layer network1=[CSI (t-1, c), CSI (t-2, c) ..., CSI (t- Δs
t,c)]TWith frequency domain X2=[CSI (t, c ± 1), CSI (t, c ± 2) ..., CSI (t, c ± Δ f)]TCommon training, obtain defeated
Outgoing vector Y1t,Y2t, by Y1t,Y2tAs the input vector of the second layer network, final predicted value Y is obtainedt, with time domain and frequency domain
In conjunction with method structure time-frequency two-dimensional LMBP neural networks realize the spectrum prediction of ISM band.
Further, step 3 is specially:, using the measured data of ISM band as training sequence input time-frequency two-dimensional LMBP
In neural network, output vector Y=[Y are obtained1,Y2,Y3,...,Ym]T, and constitute error function with object vector:
Wherein, the parameter vector that u is made of neural network weight w and threshold values b:
The Newton method learning rules of parameter vector u:WhereinFor Hessian inverse of a matrix squares
Battle array;gkFor the gradient of F (u), gk=▽ F (u)=2JT(u) e (u), J (u) are the Jacobian matrix of e (u):
Further, it is specially in step 4:Time domain and frequency domain input vector matrix are built with ISM band measured data
Xt, Xf:
By Xt, XfAs neural network input vector, the time-frequency of optimal value is had reached by weight vector w and threshold vector b
Two-dimentional LMBP neural networks obtain output vector Y=[Y1,Y2,Y3,...,Ym]T, YmAs CSI (tm,cm) predicted value.
It advantages of the present invention and has the beneficial effect that:
Present invention employs the time-frequency correlation properties based on ISM band, build a kind of time-frequency two-dimensional LMBP neural networks, will
Frequency domain parameter is added in input vector, builds two-dimensional prediction matrix, highly relevant based on adjacent frequency point (△ f=1, △ f=2)
Property replaces part time domain input vector with the stronger frequency domain input vector of correlation, on the basis of ensureing precision of prediction, subtracts
The computation complexity of few neural network repetitive exercise, and then shorten the training convergence time of network, therefore it is proposed by the invention
ISM band spectrum prediction method has training convergence time short, the high advantage of precision of prediction.
Description of the drawings
Fig. 1 is that the present invention is provided in ISM (2.4GHz) frequency range that preferred embodiment proposes based on time-frequency two-dimensional LMBP nerves
The spectrum prediction method flow diagram of network;
When Fig. 2 is the 2.4GHz frequency range neighbours of the invention obtained according to measured data/frequency correlogram;
Fig. 3 is present invention two dimension LMBP Prediction Accuracy curve graphs in different time-frequency input vectors;
Fig. 4 is precision of prediction curve comparison figure of the present invention with time domain LMBP neural networks and Markov algorithms;
Table 1 is comparison (N=of the present invention with time domain LMBP neural networks and the training convergence in mean time of Markov algorithms
9)。
Specific implementation mode
Below in conjunction with attached drawing, the invention will be further described:
The present embodiment is the training program of time-frequency two-dimensional LMBP neural networks, and input vector data source is in ISM band
Quantized data is surveyed, it is respectively 1,2,3 ..., 10 that neural network time domain, which inputs △ t, and it is respectively 0,1,2 that frequency domain, which inputs △ f, defeated
The target error of outgoing vector N=16, error function F (u) are 0.01, reach what target error was completed as network training using F (u)
Condition obtains the optimal solution for the parameter vector u being made of neural network weight and threshold value.
The first step:Calculate ISM band time domain and frequency domain correlation:
Wherein, the quantitative formula of CSI (t, c) is:
T=1,2 ..., T;C=1,2 ..., C, R
(t, c) and CSI (t, c) indicate the measured power value and channel state information of moment t lower channel c respectively.
The degree of correlation of two 0-1 sequencesIt is defined as follows:
The formula is usually used in assessing the correlation of two binary sequences, and wherein I (A) is discriminant function, if A values are true, then I
(A)=1, otherwise I (A)=0.By R, ((correlation curve of Δ f) is it is found that ISM frequency spectrums have height phase adjacent to time slot by Δ t), R
Close characteristic;Neighbouring frequency point has stronger correlation properties, and in f=± 1 △ and f=± 2 △ adjacent to frequency point, correlation is respectively
0.87 and 0.86.
Second step:Based on ISM band time-frequency correlation conclusion, time-frequency two-dimensional LMBP neural networks are built to realize ISM frequencies
The spectrum prediction of section;
Third walks:It is obtained using the measured data of ISM band as training sequence input time-frequency two-dimensional LMBP neural networks defeated
Outgoing vector Y=[Y1,Y2,Y3,...,Ym]T, and constitute error function with object vector:
Wherein, the neural network parameter vector that u is made of weights and threshold values:
The Newton method training rules of parameter vector u:WhereinFor Hessian inverse of a matrix squares
Battle array, gkFor the gradient of F (u).Wherein, J (u) is the Jacobian matrix of e (u):
Since F (u) has the form of error of sum square, HkCan approximate expression be:For
EnsureIt is reversible for positive definite, a Tiny increment dt is addedThus the adjustment formula for obtaining u is:Using the measured data of ISM band as training sequence, work as F
(u) when > ε, training is iterated to adjust formula;As F (u)≤ε, terminates training, obtains the optimal solution of parameter vector u,
Obtain the weight vector w and threshold vector b of neural network.
4th step:The time-frequency two-dimensional LMBP neural networks that are constituted with parameter vector u realize the spectrum prediction of ISM band,
Time domain and frequency domain input vector matrix X are built with ISM band measured datat, Xf:
By Xt, XfAs neural network input vector, the time-frequency of optimal value is had reached by weight vector w and threshold vector b
Two-dimentional LMBP neural networks obtain output vector Y=[Y1,Y2,Y3,...,Ym]T, YmAs CSI (tm,cm) predicted value.It will be defeated
Outgoing vector Y is the same as actual measurement object vector Z=[Z1,Z2,Z3,...,Zm]TIt is compared, obtains the pre- of time-frequency two-dimensional LMBP neural networks
Survey precision.
In the present embodiment, Fig. 2 gives time domain, the frequency domain correlation properties being calculated based on ISM band measured data
Curve graph;Fig. 3 give based on different LMBP neural network time-frequency input vectors combinations (△ t=1,2,3 ..., 10, △ f
=0,1,2) the precision of prediction figure obtained;Fig. 4 give based on no input vector number (N=1,2,3 ..., 14), when
Frequency two dimension LMBP neural networks (△ f=2) are the same as tradition LMBP neural networks and the precision of prediction comparison diagram of Markov algorithms.By
Fig. 2 is as it can be seen that ISM frequency spectrum time domains have highly relevant characteristic in short-term, as △ t increase relativity of time domain tends to 0.85;Frequency domain
Frequency point in f=± 1 △ and f=± 2 △, correlation are respectively 0.87 and 0.86, have higher spectrum prediction correlation.By
Fig. 3 as it can be seen that time domain LMBP neural networks precision of prediction before △ t=5 be incremented by, be basically stable at 0.92 after △ t=5
Below;In the time-frequency two-dimensional LMBP neural networks be made of △ f=1, △ f=2, △ f=3, as △ t < 5, precision of prediction compared with
Time domain LMBP improves a lot;Behind 5 > △ t, precision of prediction tends towards stability, and the precision of prediction of wherein △ f=2 is optimal, stablizes
95% or so.From fig. 4, it can be seen that the precision of prediction of time domain LMBP neural networks and Markov algorithms is stablized after N=5
86% and 91% or so, i.e., it cannot improve precision of prediction by increasing input vector;And time-frequency two-dimensional LMBP neural networks (△
F=2 first two method) is compared under conditions of increasing by 4 input vectors on year-on-year basis, and precision of prediction is carried compared to time domain LMBP algorithms
It is high by 4% or so, 9% or so are improved compared to Markov algorithms, and equally precision tends towards stability after N=9.By table 1 as it can be seen that
Under the same conditions, the required network training convergence time of time-frequency two-dimensional LMBP neural networks is shorter, with error for target error
Target 10-2For, 1.9 times faster than time domain LMBP of the training convergence time of time-frequency two-dimensional LMBP neural networks, than Markov algorithm
Fast 3.8 times.Institute's extracting method compares time domain LMBP prediction techniques and Markov prediction techniques same according to Fig. 2, Fig. 3, Fig. 4
Precision of prediction is more excellent under the conditions of input vector, and training convergence time is shorter, and time-frequency two-dimensional LMBP neural networks are with △ t=5 and △ f
=2 are used as input vector, and 95% spectrum prediction precision can be realized under conditions of input vector number N=9, is not increasing meter
Under conditions of calculating complexity, the precision of prediction of ISM band is effectively improved.Institute's extracting method can know frequency in advance by spectrum prediction
The occupied information of section effectively solves ISM band equipment room and is compatible with coexistence problems, obtains in ISM band more preferably spectrum prediction
Method.
Table 1
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited in above-mentioned
Particular implementation, those skilled in the art can make various deformations or amendments within the scope of the claims, this not shadow
Ring the substantive content of the present invention.
Claims (4)
1. a kind of spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band, which is characterized in that including step
Suddenly:
Step 1:ISM band measured data is collected, the correlation of the correlation and frequency domain of ISM band time domain is calculated;In step 1
In, after collecting ISM band frequency spectrum measured data, (Δ t), the correlation R of frequency domain be (Δ f's) by the correlation R of the frequency range time domain
It calculates as follows:
Wherein, the quantitative formula of CSI (t, c) is:
R(t,c)
Indicate the measured power value and channel state information of moment t lower channel c respectively with CSI (t, c);
The degree of correlation of two 0-1 sequencesIt is defined as follows:
The formula is usually used in assessing the correlation of two binary sequences, and wherein I (A) is discriminant function, if A values are true, then I (A)
=1, otherwise I (A)=0, by R, ((correlation curve of Δ f) obtains ISM band between time slot and neighbouring frequency point by Δ t), R
Correlation;
Step 2:The correlation of correlation and frequency domain based on ISM band time domain builds two-dimensional prediction matrix, in the first layer network
In realize the common training of time domain and frequency domain in a parallel fashion, obtain the output vector Y of the first layer network1t,Y2t, by first
The output vector Y of layer network1t,Y2tAs the input vector of the second layer network, final predicted value Y is obtainedt, that is, build time-frequency two
Tie up the spectrum prediction of LMBP neural fusion ISM bands;
Step 3:Using the measured data of the ISM band obtained in step 1 as training sequence, by adjusting formula, adjustment is public
Formula is:Adjust formulauk+1Indicate the obtained parameter of next step iteration to
Amount, ukIndicate the parameter vector of current iteration, J (uk) indicate e (uk) Jacobian matrixes,Indicate Tiny increment dt unit square
Battle array, e (uk) indicating error vector, for adjustment formula using error function F (u)≤ε as condition, ε indicates preset target error;It completes
The repetitive exercise of time-frequency two-dimensional LMBP neural networks, obtains the optimal solution of parameter vector u, with this obtain the weights of neural network to
Measure w and threshold vector b;Step 3 is specially:Using the measured data of ISM band as training sequence input time-frequency two-dimensional LMBP god
Through in network, obtaining output vector Y=[Y1,Y2,Y3,...,Ym]T, and constitute error function with object vector:
, ZiFor the object vector of neural network model, eiThe ginseng being made of neural network weight w and threshold values b for error vector, u
Number vector:
The Newton method learning rules of parameter vector u:WhereinFor Hessian inverse of a matrix matrixes;gkFor
The gradient of F (u), gk=▽ F (u)=2JT(u) e (u), J (uk) be e (u) Jacobian matrix:
Step 4:Build time domain and frequency domain input vector matrix Xt,Xf:By Xt、XfAs neural network input vector, pass through weights
Vectorial w and threshold vector b has reached the time-frequency two-dimensional LMBP neural networks of optimal value, obtains output vector Y=[Y1,Y2,
Y3,...,Ym]T, YmAs CSI (tm,cm) predicted value, tmIndicate m-th of moment, cmFor m-th of channel, spectrum prediction is completed.
2. the spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band according to claim 1, special
Sign is, the time domain X of the step 1 and step 21=[CSI (t-1, c), CSI (t-2, c) ..., CSI (t- Δs t, c)]TAnd frequency
Domain X2=[CSI (t, c ± 1), CSI (t, c ± 2) ..., CSI (t, c ± Δ f)]T, shown CSI (t- Δs t, c) expression moment (t-
The channel state information of Δ t) channels c, (t, c ± Δ f) indicate the moment t channel (channel state information of c ± Δs f) to CSI.
3. the spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band according to claim 1, special
Sign is, in step 2, frequency domain prediction point is added in neural network input vector, two-dimensional prediction matrix is built, in first layer
Time domain X is realized in network in a parallel fashion1=[CSI (t-1, c), CSI (t-2, c) ..., CSI (t- Δs t, c)]TAnd frequency domain
X2=[CSI (t, c ± 1), CSI (t, c ± 2) ..., CSI (t, c ± Δ f)]TCommon training, obtain output vector Y1t,Y2t,
By Y1t,Y2tAs the input vector of the second layer network, final predicted value Y is obtainedt, built in the method that time domain and frequency domain combine
Time-frequency two-dimensional LMBP neural networks realize the spectrum prediction of ISM band.
4. the spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band according to claim 1, special
Sign is, is specially in step 4:Time domain and frequency domain input vector matrix X are built with ISM band measured datat, Xf:
By Xt, XfAs neural network input vector, the time-frequency two-dimensional of optimal value is had reached by weight vector w and threshold vector b
LMBP neural networks obtain output vector Y=[Y1,Y2,Y3,...,Ym]T, YmAs CSI (tm,cm) predicted value.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610149087.8A CN105915299B (en) | 2016-03-16 | 2016-03-16 | Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610149087.8A CN105915299B (en) | 2016-03-16 | 2016-03-16 | Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band |
Publications (2)
Publication Number | Publication Date |
---|---|
CN105915299A CN105915299A (en) | 2016-08-31 |
CN105915299B true CN105915299B (en) | 2018-08-14 |
Family
ID=56745058
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610149087.8A Active CN105915299B (en) | 2016-03-16 | 2016-03-16 | Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN105915299B (en) |
Families Citing this family (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3635872A1 (en) | 2017-05-03 | 2020-04-15 | Virginia Tech Intellectual Properties, Inc. | Learning radio signals using radio signal transformers |
EP3825703B1 (en) * | 2019-11-20 | 2023-08-30 | Rohde & Schwarz GmbH & Co. KG | Method and system for detecting and/or classifying a wanted signal |
WO2021237423A1 (en) * | 2020-05-25 | 2021-12-02 | Oppo广东移动通信有限公司 | Channel state information transmission methods, electronic device, and storage medium |
CN116527177B (en) * | 2023-04-28 | 2024-03-12 | 哈尔滨工程大学 | Three-dimensional spectrum prediction method based on correlation analysis and graph convolution network |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103209417A (en) * | 2013-03-05 | 2013-07-17 | 北京邮电大学 | Method and device for predicting spectrum occupancy state based on neural network |
-
2016
- 2016-03-16 CN CN201610149087.8A patent/CN105915299B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103209417A (en) * | 2013-03-05 | 2013-07-17 | 北京邮电大学 | Method and device for predicting spectrum occupancy state based on neural network |
Non-Patent Citations (3)
Title |
---|
基于K-RBF神经网络的认知无线电频谱预测;吴建绒等;《电视技术》;20141231;第38卷(第5期);I136-72 * |
认知无线电系统中的频谱预测算法研究;陈斌华;《中国优秀硕士学位论文全文数据库 信息科技辑》;20111231;I136-610 * |
认知无线电网络中的频谱预测技术研究;邢晓双;《中国优秀博士论文全文数据库 信息科技辑》;20141231;第105-108页 * |
Also Published As
Publication number | Publication date |
---|---|
CN105915299A (en) | 2016-08-31 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN105915299B (en) | Spectrum prediction method based on time-frequency two-dimensional LMBP neural networks in ISM band | |
Angjelicinoski et al. | Comparative analysis of spatial interpolation methods for creating radio environment maps | |
CN105653502A (en) | Analysis method of electromagnetic radiation relevance of communication base station based on genetic algorithm | |
CN110445567B (en) | Electromagnetic spectrum map construction method | |
CN106231549A (en) | A kind of 60GHz pulse indoor orientation method based on restructing algorithm | |
JP2023535841A (en) | Energy-saving processing method and processing device for base station | |
CN110346654A (en) | Electromagnetic spectrum map construction method based on common kriging interpolation | |
CN115776724B (en) | Sensor node layout optimization method and system for electromagnetic spectrum map mapping | |
WO2023169589A1 (en) | Predictive channel modeling method based on adversarial network and long short-term memory network | |
CN105205114A (en) | Wi-Fi (wireless fidelity) positioning fingerprint database construction method based on image processing | |
CN103491551B (en) | A kind of weighting cooperative frequency spectrum sensing method of feature based vector | |
Deng et al. | A two-step learning and interpolation method for location-based channel database construction | |
CN105357683A (en) | Gibbs sampling-based ultra-dense heterogeneous network optimal cell range expansion bias adjustment method | |
CN101982953A (en) | Frequency domain multi-dimensional parameterized model of broadband wireless communication channel and modeling method | |
CN105188034B (en) | A kind of Cooperative Localization Method in wireless sensor network | |
Liang et al. | Mobile traffic prediction based on densely connected CNN for cellular networks in highway scenarios | |
CN104683050B (en) | Multi-antenna total blind spectrum sensing method capable of effectively resisting noise uncertainty | |
Mahbub | Comparative link-level analysis and performance estimation of channel models for IIoT (industrial-IoT) wireless communications | |
Laoudias et al. | Differential signal strength fingerprinting revisited | |
Singh et al. | A novel VL-based positioning model for obstacle location sensing and 3D shape detection in crowded indoor networks | |
Mallik et al. | EME-Net: A U-net-based indoor EMF exposure map reconstruction method | |
CN110912747B (en) | Random geometry-based power wireless private network performance analysis method | |
Lagum et al. | Quantifying the regularity of perturbed triangular lattices using CoV-based metrics for modeling the locations of base stations in HetNets | |
Nielsen et al. | On antenna design objectives and the channel capacity of MIMO handsets | |
Rathod et al. | Data-driven and GIS-based coverage estimation in a heterogeneous propagation environment |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |