CN112383369A - Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model - Google Patents
Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model Download PDFInfo
- Publication number
- CN112383369A CN112383369A CN202010718214.8A CN202010718214A CN112383369A CN 112383369 A CN112383369 A CN 112383369A CN 202010718214 A CN202010718214 A CN 202010718214A CN 112383369 A CN112383369 A CN 112383369A
- Authority
- CN
- China
- Prior art keywords
- cnn
- spectrum
- lstm
- layer
- lstm network
- 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.)
- Pending
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 82
- 230000001149 cognitive effect Effects 0.000 title claims abstract description 30
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000011176 pooling Methods 0.000 claims description 37
- 238000000605 extraction Methods 0.000 claims description 11
- 238000013461 design Methods 0.000 abstract description 4
- 238000012544 monitoring process Methods 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 30
- 230000006870 function Effects 0.000 description 24
- 238000004891 communication Methods 0.000 description 12
- 210000002569 neuron Anatomy 0.000 description 11
- 230000000694 effects Effects 0.000 description 9
- 238000012549 training Methods 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 7
- 230000008569 process Effects 0.000 description 6
- 238000004088 simulation Methods 0.000 description 6
- 238000013528 artificial neural network Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000004422 calculation algorithm Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 238000013468 resource allocation Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000002411 adverse Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012905 input function Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 238000012067 mathematical method Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04B—TRANSMISSION
- H04B17/00—Monitoring; Testing
- H04B17/30—Monitoring; Testing of propagation channels
- H04B17/382—Monitoring; Testing of propagation channels for resource allocation, admission control or handover
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W72/00—Local resource management
- H04W72/04—Wireless resource allocation
- H04W72/044—Wireless resource allocation based on the type of the allocated resource
- H04W72/0453—Resources in frequency domain, e.g. a carrier in FDMA
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
- H04W16/14—Spectrum sharing arrangements between different networks
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Electromagnetism (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
The invention discloses a cognitive radio multi-channel spectrum sensing method based on a CNN-LSTM network model, relates to the field of radio monitoring and spectrum management, and aims to solve the problem that the sensing accuracy of the existing single radio spectrum is low. The invention uses the CNN and LSTM prediction models as basic network structures, and performs combined design to obtain the CNN-LSTM prediction model for spectrum prediction under multiple channels, so that the sensing accuracy of radio spectrum is remarkably improved.
Description
Technical Field
The present invention relates to the field of radio monitoring and spectrum management.
Background
In recent years, the growth of wireless communication devices and the increase of demand for wireless communication have made radio frequency, an unrenewable resource, increasingly strained, and spectrum resources have been increasing as a strategic resource in each country due to scarcity. Currently, when managing and allocating wireless spectrum resources, a fixed spectrum resource allocation manner is generally adopted. Under the fixed spectrum resource allocation mode, the whole radio spectrum resource is divided into two parts of authorized spectrum and unauthorized spectrum, and the spectrum allocation system has the phenomenon of resource usage imbalance: on the one hand, the licensed spectrum occupies most of the spectrum resources, but in some cases, the licensed frequency band of the licensed user (primary user) is not frequently used by the authorized user, and most of the licensed spectrum is not used by the primary user in some time period or some place, but is in an unused state, and as a result, the utilization rate of the licensed spectrum is very low. Some research results on the measurement of spectrum resources by the Federal Communications Commission (FCC) in the united states have shown that the average usage of the measured spectrum is 15% -85% in most regions and for most of the measured time periods. On the other hand, the unlicensed spectrum reserved for unlicensed users occupies only a small portion of the total resources of the spectrum. With the rapid development of wireless communication technology, especially the rapid development of some new wireless service services, such as wireless local area network, wireless personal area network, wireless metropolitan area network, etc., more and more services and users depend on unlicensed spectrum, the unlicensed frequency band becomes saturated more and more, and the problem of spectrum resource shortage also becomes more and more serious. It is obvious that the low usage rate of the licensed spectrum is mainly caused by the inflexible fixed spectrum allocation strategy, which is a contradiction that the unlicensed users cannot use the corresponding frequency bands.
In order to solve the above-mentioned outstanding contradiction between the shortage of spectrum resources and the inflexible allocation strategy, which needs to be optimized and adjusted, the proposal of software radio makes it possible to dynamically configure communication devices. Based on the rapid development of machine learning technology and the emergence of software Radio, the concept of Cognitive Radio (CR) was creatively proposed by doctor Mitola. In a Cognitive Radio System (CRS), an unauthorized user (Cognitive user) can sense the surrounding Spectrum environment, monitor and analyze the Spectrum usage of an authorized frequency band, and intelligently adjust parameters such as the Spectrum range and access mode of the access of the unauthorized user (Cognitive user) without generating intolerable interference to the communication of the authorized user (primary user), so that the authorized user (primary user) can perform dynamic Spectrum access to realize the communication of the service by fully utilizing a Spectrum Hole (Spectrum Hole), i.e., a time slot or a frequency band which is not used by the primary user in a certain time range or region, and the primary user has a priority over the authorized user to access the frequency band in the authorized frequency band, and when the unauthorized user detects that the authorized user accesses the authorized frequency band, the unauthorized user must actively quit the use of the frequency band. A technology in which a CR can operate on an available frequency band by changing its own parameters is considered as an important technology for improving the utilization rate of current wireless spectrum resources and solving the problem of shortage of spectrum resources.
The four functions of spectrum sensing, spectrum decision, spectrum sharing and spectrum switching in the cognitive radio technology can enable a cognitive user to sense an authorized spectrum and perform opportunistic access to a spectrum cavity so as to perform service communication. However, the implementation of these four functions has some drawbacks, which severely limit the improvement of the communication performance of the cognitive radio system, and mainly include the following four aspects:
(1) when sensing broadband spectrum, generally, a cognitive user needs to scan and sense the whole frequency range, which causes huge sensing processing delay and energy loss.
(2) In an actual situation, due to the limitation of hardware equipment conditions and communication conditions of the cognitive user, a large amount of time is consumed in the spectrum sensing process, so that the time of a data transmission stage in a sensing period is reduced, the sensing accuracy is influenced, the spectrum decision accuracy is further reduced, and the whole spectrum resource utilization rate of a communication system is seriously influenced.
(3) In spectrum sharing, a cognitive user may need to perform communication Service requests with different Quality of Service (QoS) requirements at different times, and in order to ensure fairness and reasonableness of resource allocation, higher time delay may be caused when spectrum allocation is performed.
(4) In a passive spectrum switching mode, a cognitive user performs spectrum switching only after sensing that a master user initiates a service request, and in the passive spectrum switching mode, the cognitive user cannot avoid transmission conflicts with the master user.
Disclosure of Invention
The invention provides a cognitive radio multi-channel spectrum sensing method based on a CNN-LSTM network model, aiming at solving the problem that the sensing accuracy of the current single radio spectrum is low.
The cognitive radio multi-channel spectrum sensing method based on the CNN-LSTM network model is characterized by comprising the following steps: it comprises the following steps:
step one, constructing a CNN-LSTM network model, wherein the CNN-LSTM network model comprises a CNN network and an LSTM network; the CNN network is used for feature extraction of frequency spectrum data, and the LSTM network is used for prediction of frequency spectrum occupation states.
Using 2 layers of convolution kernels and 1 layer of pooling layers to perform feature extraction on a data sequence in the CNN network;
step two, optimizing the CNN-LSTM network model obtained in the step one to obtain an optimized CNN-LSTM network model;
and step three, performing multi-channel spectrum sensing on the cognitive radio by using the optimized CNN-LSTM network model obtained in the step two, and finishing one-time cognitive radio multi-channel spectrum sensing based on the CNN-LSTM network model.
The invention has the following beneficial effects: after a large number of simulation experiments, the accuracy of the single radio frequency spectrum is fully proved to be remarkably improved by referring to the simulation and results of FIG. 3.
Drawings
FIG. 1 is a structural illustration of a CNN-LSTM network model;
FIG. 2 is a simulation diagram of the LSTM model performance index varying with the length of the input sequence;
FIG. 3 is a simulation diagram of the CNN-LSTM model performance index varying with the length of the input sequence;
Detailed Description
The specific embodiment mode one, cognitive radio multichannel frequency spectrum perception method based on CNN-LSTM network model, its characteristic is: it comprises the following steps:
step one, constructing a CNN-LSTM network model, wherein the CNN-LSTM network model comprises a CNN network and an LSTM network; the CNN network is used for feature extraction of frequency spectrum data, and the LSTM network is used for prediction of frequency spectrum occupation states.
Using 2 layers of convolution kernels and 1 layer of pooling layers to perform feature extraction on a data sequence in the CNN network;
step two, optimizing the CNN-LSTM network model obtained in the step one to obtain an optimized CNN-LSTM network model;
and step three, performing multi-channel spectrum sensing on the cognitive radio by using the optimized CNN-LSTM network model obtained in the step two, and finishing one-time cognitive radio multi-channel spectrum sensing based on the CNN-LSTM network model.
The second embodiment is different from the first embodiment in that in the CNN network, the 1 st convolutional layer has 16 convolutional kernels, and the size of the convolutional kernels is 1 × 3; the 2 nd layer has 32 convolution kernels, the size is 1 multiplied by 4, and the moving step length is 1 multiplied by 1; the Pooling layer used was Average Pooling (Average Pooling), with a size of 1X 2.
The third specific embodiment, the difference between the second specific embodiment and the cognitive radio multi-channel spectrum sensing method based on the CNN-LSTM network model, is that the LSTM network includes a 1-layer hidden layer structure, and the number of LSTM units in the hidden layer is 32.
Convolutional neural network features
(1) Sparse connections
In the traditional BP network, the connection between the neurons of the front layer and the back layer is complete, namely the neurons of the n-1 layer are all connected with all the neurons of the n layer. The phenomenon that the training parameters are explosively increased along with the increase of the number of hidden layers can occur in the connection, so that the consumed resources during model training are too large, and an overfitting phenomenon can also occur, so that the model has poor generalization capability. In the CNN, however, neurons in each layer are partially connected, and by using local spatial correlation between layers, neurons in each layer are only partially connected with neurons in an upper layer, that is, sparsely connected or locally connected, so that the parameter scale of the CNN network architecture is greatly reduced.
Fig. 3-1 shows the difference between full-connection and sparse connection in structure, and from the schematic diagram of full-connection on the left, it can be seen that there are corresponding connections between the neuron in the previous layer and all the neurons in the next layer, and each connection has corresponding connection parameters, so there are many connection parameters in full-connection. The right graph is sparse connection, and it can be seen that connection exists between some neurons, and connection parameters are reduced accordingly.
(2) Weight sharing
In CNN, the convolutional layer performs a convolution operation on input data using a convolution filter, and the result of the convolution completes feature extraction on the input data. And the weight matrix and the bias item parameter of each convolution filter are the same.
When a conventional neural network extracts features from input data, complex calculation operations are usually performed on the input data, and then the extracted features are input into the neural network. The advantage of sharing weights is that feature extraction on the input data is done automatically. And the weight sharing also enables the scale of the training parameters of the CNN model to be greatly reduced.
(3) Pooling
Pooling is also an important element in CNN networks, often referred to as downsampling, to reduce network parameters. The advantage of pooling is that the entire network is not prone to overfitting. However, the larger the pooling ratio is, the better, and although the higher the pooling ratio can improve the generalization ability of the network and reduce the network scale, the larger the pooling ratio can lose much feature information and further has adverse effect on the model effect, so the pooling ratio is selected appropriately.
4.2.3 structure of convolutional neural network the structure of the convolutional neural network is generally composed of an input layer, a convolutional layer, a pooling layer, a fully-connected layer, and an output layer. The convolutional neural network firstly performs convolution operation on input data, and the size and the number of convolution kernels enable the input data to present different convolution effects. The output of the convolutional layer is the input of the pooling layer, and the operation of the pooling layer can realize the compression of data and reduce the characteristic scale so as to reduce overfitting.
And inputting the characteristic data processed by the convolution layer and the pooling layer into a full-connection layer for expansion, and outputting the result through an output layer. In the construction process of the CNN model, a convolutional layer and a pooling layer are very important in the CNN model, and a plurality of convolutional layers and pooling layers are usually used, and the two layers are alternately connected. The convolutional layer also acts as a neuron, which obtains an output value by performing weighted summation on input data, which is equivalent to the process of convolution, and is therefore named as convolutional neural network.
(1) Input layer
The input layer is used for inputting data. The input to the CNN may be one-dimensional data, two-dimensional data, or three-dimensional data. Two-dimensional data is common. The two-dimensional data is divided into single-channel input and multi-channel input.
(2) Convolutional layer
Convolutional layers are the core of the network and are the most important components of the model. Convolution cores in the convolutional layer locally weight the input signal, which is equivalent to the process of weighting the input by the weight connected with neurons in the artificial neural network. The main purpose of using convolutional layers is to perform feature extraction and feature mapping using convolutional kernels. The convolution layer performs convolution operation on the input data to obtain feature mapping output.
The convolution operation is a linear calculation method, has translation invariance, can reflect the property of a system, and the calculation result of the convolution depends on the input of the current time and also depends on the input of the previous time. Convolution is a very important mathematical method and has wide application. The mathematical description of the convolution is given below in terms of the definition of the convolution in the calculus.
Convolution of continuous function: two integrable functions f (t), g (t) are provided, and integration is performed as follows:
where denotes a convolution operation, the convolution operation can be specifically interpreted as: convolution is a mathematical operation on two functions, f (t) and g (t), and is defined as the integral of the product of one of the functions, which remains unchanged, and the other function, which is inverted and translated, and the other function.
Discrete form is shown in formula 4-2:
the matrix form of the formula (4-2) is:
s(n)＝(F*G)(n) (4-3)
the representation of the two-dimensional convolution can be written as:
in equations 4-13, F represents the input Function, G is the Kernel Function (Kernel Function) or weighting Function or Filter (Filter) of the convolutional layer, and F and G are matrices or tensors. As can be seen from the above equation, the output y (i, j) is a weighted sum of the inputs F, sometimes referred to as a feature map.
The convolution operation in CNN is usually performed on multidimensional input data, and the input data and elements in a convolution kernel are multiplied one by one, and then added to obtain convolved output data. The convolution operation has an input on the left and an output on the right. The size of the convolution kernel is usually much smaller than that of the input layer, and under the condition of the same output layer, the scale of the model parameters can be greatly reduced by means of sparse connection and parameter sharing, and meanwhile, the calculation efficiency of the model is improved.
(3) Pooling layer
The output of the convolutional layer is directly connected to the pooling layer, which is a form of downsampling, also known as the subsampling layer in the neural network. It is made up of a plurality of feature planes, each corresponding to a feature plane of the previous convolution layer, as shown in fig. 3-4.
The model in the neural network can cause the phenomenon of overfitting due to excessive parameters, the phenomenon can cause the prediction effect of the model to be unsatisfactory, and the parameters of the model can be reduced through pooling operation, so that the generalization capability of the model can be greatly improved. The pooling process reduces the parameter scale of the model and the complexity of calculation by pooling the output of the convolutional layer, and can further inhibit the phenomenon of overfitting.
Usually, the actual feature of a certain region in the CNN model is represented by the statistical feature of the region, and the pooling function is the method for calculating the statistical feature. A Max Pooling function or an Average Pooling function is often selected as the Pooling function of the Pooling layer, which uses the maximum or Average of all values in the neighboring areas of the size that we set as the output of the area, and other Pooling functions are the minimum, median, norm, weighted Average, etc. of the neighboring areas. The output layer is used for prediction and result output as the name implies. Meanwhile, the output layer can also carry out back propagation of errors, forward transfer of gradients is carried out in sequence, a loss function is calculated, and the weight of each layer of the model is updated. A commonly used function for the output layer in convolutional neural networks is typically the Softmax function. The Softmax function is often used in the context of category determination. This function maps its inputs to real numbers between 0-1 and the normalized sum is guaranteed to be equal to 1, i.e. the sum of the probabilities of the multi-classes is also exactly 1. The output categories and the number of categories can be set according to the prediction requirements in the prediction.
For known { (x)(i),y(i))；i∈1,2,...,N,y(i)E.g., 0,1, K-1, where x is(i)Denotes the ithAn input data, y(i)Is the predicted value whose tag value, the ith input belongs to the jth class
Can be obtained by calculation of the Softmax function, as shown in equations 4-5:
equations 4-5 yield the prediction probability. The class to which the probability value is the highest is the predicted classification value.
4.3 CNN-LSTM network model
The LSTM model controls the memory and forgetting of historical data through gate functions, and is suitable for processing problems with time series characteristics. The LSTM can better link the context, but due to the complex structure, the calculation amount is increased along with the increase of the input data, so that the link of the context is reduced, and the accuracy of the algorithm is reduced. The CNN can extract high-level features through convolution kernel operation, and has been successfully applied in the field of image processing. The network connection has obvious time sequence characteristics, the CNN model is utilized to extract the frequency spectrum data characteristics, the characteristics are used as the input of the LSTM model, and then the frequency spectrum occupation state is predicted.
4.3.1 CNN-LSTM model design
In the CNN-LSTM model constructed by the method, a CNN network is responsible for feature extraction of spectrum data, and an LSTM network is responsible for prediction of spectrum occupation states. The structure of the CNN-LSTM prediction model is shown in FIGS. 3-7, and the CNN network performs feature extraction on a data sequence by using 2 layers of convolution kernels and 1 layer of pooling layer. Wherein the 1 st convolution layer has 16 convolution kernels with the size of 1 multiplied by 3; the 2 nd layer has 32 convolution kernels, the size is 1 multiplied by 4, and the moving step length is 1 multiplied by 1; the Pooling layer used was Average Pooling (Average Pooling), with a size of 1X 2. The LSTM network comprises a 1-layer hidden layer structure, and the number of LSTM units in the hidden layer is 32. The loss function is "mae" and the optimizer is set to "RMSprop", i.e. (Root Mean Square Prop), which allows faster convergence of the training process and smaller amplitude of the fluctuations.
After the model performance evaluation index obtains the multi-channel spectrum prediction result in the cognitive radio, the prediction effect of the model is evaluated by adopting the index same as the index in section 3.4.3, and the detailed description is omitted here.
4.4 simulation analysis
The data used in this section of the experiment is the experimental data generated in ten channels in section 3.4.1, where the spectral occupancy of the four channels 1, 4, 7, 10 is low as can be seen from fig. 3-7. Because the quality of the prediction model depends on the training data of the model to a great extent, and different sliding window lengths generate different training data, the prediction accuracy of the trained prediction model may have greater difference. The sliding window lengths in this experiment were also set at 8, 16, 32, 64 and 128, respectively. In order to verify the effectiveness of the CNN-LSTM prediction model provided in this chapter on the time series data of the channel state, the LSTM prediction model with ideal prediction effect under a single channel is selected for comparison in the experiment. The structure of the LSTM model is the same as that of the single channel, except that the shape of the input data is different. The performance indexes of the two models are simulated along with the change of the length of the input sequence, so that simulation graphs shown in fig. 2 and 3 can be obtained.
The invention has the following beneficial effects: fig. 2 and fig. 3 can see that the prediction performances of the N-LSTM prediction algorithm and the LSTM prediction algorithm on the channels 4, 7, and 10 are substantially the same, which may be caused by low spectrum occupancy on the channels, that is, the data and the processed 0 information on the channel are more, which results in poor training effects of the two models. On the other channels, it can be seen that the prediction effect of LSTM prediction for each channel is not as good as that of the prediction method based on CNN-LSTM on the proposed performance index.
The invention
The CNN and LSTM prediction models are used as basic network structures in the chapter, and combined design is carried out to obtain the CNN-LSTM prediction model for spectrum prediction under multiple channels, which is the main body of the invention. The principle, structure and training method of CNN are mainly introduced; on the basis of two prediction models of CNN and LSTM, performing combined design to obtain a structure of the CNN-LSTM model for spectrum prediction under multiple channels; and finally, simulating and analyzing the performance of the spectrum prediction algorithm of the designed CNN-LSTM prediction model under multiple channels, verifying that the CNN-LSTM prediction model is expressed on three prediction indexes in a cognitive radio system, and comparing the CNN-LSTM prediction model with an ideal prediction effect under a single channel.
Claims (3)
1. The cognitive radio multi-channel spectrum sensing method based on the CNN-LSTM network model is characterized by comprising the following steps: it comprises the following steps:
step one, constructing a CNN-LSTM network model, wherein the CNN-LSTM network model comprises a CNN network and an LSTM network; the CNN network is used for feature extraction of frequency spectrum data, and the LSTM network is used for prediction of frequency spectrum occupation states.
Using 2 layers of convolution kernels and 1 layer of pooling layers to perform feature extraction on a data sequence in the CNN network;
step two, optimizing the CNN-LSTM network model obtained in the step one to obtain an optimized CNN-LSTM network model;
and step three, performing multi-channel spectrum sensing on the cognitive radio by using the optimized CNN-LSTM network model obtained in the step two, and finishing one-time cognitive radio multi-channel spectrum sensing based on the CNN-LSTM network model.
2. The cognitive radio multi-channel spectrum sensing method based on the CNN-LSTM network model as claimed in claim 1, wherein in the CNN network, the 1 st convolutional layer has 16 convolutional kernels with a size of 1 x 3; the 2 nd layer has 32 convolution kernels, the size is 1 multiplied by 4, and the moving step length is 1 multiplied by 1; the Pooling layer used was Average Pooling (Average Pooling), with a size of 1X 2.
3. The cognitive radio multi-channel spectrum sensing method based on the CNN-LSTM network model as claimed in claim 2, wherein the LSTM network comprises a 1-layer hidden layer structure, and the number of LSTM units in the hidden layer is 32.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010718214.8A CN112383369A (en) | 2020-07-23 | 2020-07-23 | Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010718214.8A CN112383369A (en) | 2020-07-23 | 2020-07-23 | Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112383369A true CN112383369A (en) | 2021-02-19 |
Family
ID=74586379
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010718214.8A Pending CN112383369A (en) | 2020-07-23 | 2020-07-23 | Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112383369A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112906640A (en) * | 2021-03-19 | 2021-06-04 | 电子科技大学 | Space-time situation prediction method and device based on deep learning and readable storage medium |
| CN115134024A (en) * | 2022-05-31 | 2022-09-30 | 南京邮电大学 | Frequency spectrum prediction method based on two-dimensional empirical mode decomposition |
| CN115209418A (en) * | 2022-06-13 | 2022-10-18 | 海南大学 | Intelligent broadband spectrum sensing technology based on pre-training basic model |
| CN115276853A (en) * | 2022-06-16 | 2022-11-01 | 宁波大学 | CNN-CBAM-based spectrum sensing method |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160099010A1 (en) * | 2014-10-03 | 2016-04-07 | Google Inc. | Convolutional, long short-term memory, fully connected deep neural networks |
| CN106782602A (en) * | 2016-12-01 | 2017-05-31 | 南京邮电大学 | Speech-emotion recognition method based on length time memory network and convolutional neural networks |
| CN107169566A (en) * | 2017-06-09 | 2017-09-15 | 山东师范大学 | Dynamic neural network model training method and device |
| CN109245840A (en) * | 2018-10-15 | 2019-01-18 | 哈尔滨工业大学 | Spectrum prediction method in cognitive radio system based on convolutional neural networks |
| CN110082841A (en) * | 2019-04-18 | 2019-08-02 | 东华大学 | A kind of short-term wind speed forecasting method |
| US10713493B1 (en) * | 2020-02-06 | 2020-07-14 | Shenzhen Malong Technologies Co., Ltd. | 4D convolutional neural networks for video recognition |
-
2020
- 2020-07-23 CN CN202010718214.8A patent/CN112383369A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160099010A1 (en) * | 2014-10-03 | 2016-04-07 | Google Inc. | Convolutional, long short-term memory, fully connected deep neural networks |
| CN106782602A (en) * | 2016-12-01 | 2017-05-31 | 南京邮电大学 | Speech-emotion recognition method based on length time memory network and convolutional neural networks |
| CN107169566A (en) * | 2017-06-09 | 2017-09-15 | 山东师范大学 | Dynamic neural network model training method and device |
| CN109245840A (en) * | 2018-10-15 | 2019-01-18 | 哈尔滨工业大学 | Spectrum prediction method in cognitive radio system based on convolutional neural networks |
| CN110082841A (en) * | 2019-04-18 | 2019-08-02 | 东华大学 | A kind of short-term wind speed forecasting method |
| US10713493B1 (en) * | 2020-02-06 | 2020-07-14 | Shenzhen Malong Technologies Co., Ltd. | 4D convolutional neural networks for video recognition |
Non-Patent Citations (4)
| Title |
|---|
| 何竞帆: "认知无线电频谱预测算法研究", 《中国优秀硕士学位论文全文数据库》 * |
| 卢光跃、施聪、吕少卿、周亮: "基于LSTM神经网络的频谱感知算法", 《信号处理》 * |
| 李含青、郭庆、李庆忠、贾敏: "基于迭代攻击检测的联合压缩频谱感知算法", 《四川大学学报( 工程科学版)》 * |
| 韦鹏程、邹杨、冉维: "《基于R语言数据挖掘的统计与分析》", 31 December 2017 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112906640A (en) * | 2021-03-19 | 2021-06-04 | 电子科技大学 | Space-time situation prediction method and device based on deep learning and readable storage medium |
| CN115134024A (en) * | 2022-05-31 | 2022-09-30 | 南京邮电大学 | Frequency spectrum prediction method based on two-dimensional empirical mode decomposition |
| CN115134024B (en) * | 2022-05-31 | 2023-07-11 | 南京邮电大学 | Spectrum prediction method based on two-dimensional empirical mode decomposition |
| CN115209418A (en) * | 2022-06-13 | 2022-10-18 | 海南大学 | Intelligent broadband spectrum sensing technology based on pre-training basic model |
| CN115276853A (en) * | 2022-06-16 | 2022-11-01 | 宁波大学 | CNN-CBAM-based spectrum sensing method |
| CN115276853B (en) * | 2022-06-16 | 2023-10-03 | 宁波大学 | Spectrum sensing method based on CNN-CBAM |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112383369A (en) | Cognitive radio multi-channel spectrum sensing method based on CNN-LSTM network model | |
| CN112003663A (en) | Cognitive radio frequency spectrum sensing method based on LSTM neural network | |
| US7657416B1 (en) | Hierarchical system design | |
| Zhang et al. | Spectrum allocation in cognitive radio networks using swarm intelligence | |
| CN112163601B (en) | Image classification method, system, computer device and storage medium | |
| CN106454856A (en) | Spectrum allocation method based on graph coloring and analytic hierarchy process in cognitive radio | |
| CN114698128B (en) | Anti-interference channel selection method and system for cognitive satellite-ground network | |
| Yu et al. | Spectrum availability prediction for cognitive radio communications: A DCG approach | |
| CN113596785A (en) | D2D-NOMA communication system resource allocation method based on deep Q network | |
| Liu et al. | A service-oriented spectrum allocation algorithm using enhanced PSO for cognitive wireless networks | |
| CN111200550B (en) | Virtual network mapping method and device | |
| CN109065176A (en) | A kind of blood glucose prediction method, device, terminal and storage medium | |
| CN112654063A (en) | Uplink capacity assessment method and device | |
| Liu et al. | Hierarchical agglomerative clustering and LSTM-based load prediction for dynamic spectrum allocation | |
| Zhao et al. | C-LSTM: CNN and LSTM Based Offloading Prediction Model in Mobile Edge Computing (MEC) | |
| Zhang et al. | Prediction of spectrum based on improved RBF neural network in cognitive radio | |
| CN114449536B (en) | 5G ultra-dense network multi-user access selection method based on deep reinforcement learning | |
| CN115811788A (en) | D2D network distributed resource allocation method combining deep reinforcement learning and unsupervised learning | |
| CN114928549A (en) | Communication resource allocation method and device of unauthorized frequency band based on reinforcement learning | |
| Ito et al. | Radio propagation estimation in a long-range environment using a deep neural network | |
| Jiang et al. | Dynamic spectrum access for femtocell networks: A graph neural network based learning approach | |
| CN108736991B (en) | Group intelligent frequency spectrum switching method based on classification | |
| CN108770010B (en) | Intelligent reconstruction method for service-oriented wireless network networking mode | |
| CN112383965B (en) | Cognitive radio power distribution method based on DRQN and multi-sensor model | |
| CN108183757B (en) | Segmented approximate convex decomposition method for perception duration and resource allocation joint optimization |
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 | ||
| CB03 | Change of inventor or designer information | ||
| CB03 | Change of inventor or designer information |
Inventor after: Jia Min Inventor after: Liu Yun Inventor after: Wang Linan Inventor after: Guo Qing Inventor after: Gu Xuemai Inventor before: Jia Min Inventor before: Liu Yun Inventor before: Wang Linan Inventor before: Guo Qing Inventor before: Gu Xuemai |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210219 |