CN114298166A - Spectrum availability prediction method and system based on wireless communication network - Google Patents
Spectrum availability prediction method and system based on wireless communication network Download PDFInfo
- Publication number
- CN114298166A CN114298166A CN202111507860.0A CN202111507860A CN114298166A CN 114298166 A CN114298166 A CN 114298166A CN 202111507860 A CN202111507860 A CN 202111507860A CN 114298166 A CN114298166 A CN 114298166A
- Authority
- CN
- China
- Prior art keywords
- deep learning
- learning model
- spectrum
- loss
- hyper
- 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 86
- 238000000034 method Methods 0.000 title claims abstract description 58
- 238000004891 communication Methods 0.000 title claims abstract description 43
- 238000013136 deep learning model Methods 0.000 claims abstract description 136
- 230000003595 spectral effect Effects 0.000 claims abstract description 36
- 230000001149 cognitive effect Effects 0.000 claims abstract description 28
- 238000005457 optimization Methods 0.000 claims abstract description 22
- 238000012549 training Methods 0.000 claims description 31
- 238000013528 artificial neural network Methods 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 17
- 238000012360 testing method Methods 0.000 claims description 10
- 238000012795 verification Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 8
- 230000006872 improvement Effects 0.000 claims description 5
- 238000010276 construction Methods 0.000 claims description 3
- 230000008569 process Effects 0.000 claims description 2
- 108091006146 Channels Proteins 0.000 description 11
- 238000010586 diagram Methods 0.000 description 5
- 239000011159 matrix material Substances 0.000 description 4
- 208000037004 Myoclonic-astatic epilepsy Diseases 0.000 description 3
- 238000000874 microwave-assisted extraction Methods 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000010801 machine learning Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000002360 explosive Substances 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Images
Landscapes
- Mobile Radio Communication Systems (AREA)
Abstract
The invention provides a frequency spectrum availability prediction method and a frequency spectrum availability prediction system based on a wireless communication network, wherein the method comprises the steps of obtaining a frequency spectrum data set; setting Bayesian optimization cycle times, a hyper-parameter search space and deep learning model iteration times; initializing the structure, weight, bias and hyper-parameters of the deep learning model; calculating a loss of the spectral data set between an input and an output of the deep learning model; updating the weight and the bias parameters of the deep learning model; judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not; if not, resetting the iteration times of the deep learning model; if so, fine tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization to obtain the optimal deep learning model. By adopting the scheme, the invention can accurately predict the available frequency spectrum gap of the future time slot, so that the cognitive user is prevented from being interfered by malicious users, and the normal communication between the cognitive user and the base station and between the cognitive users is ensured.
Description
Technical Field
The invention belongs to the technical field of wireless communication, and particularly relates to a spectrum availability prediction method and system based on a wireless communication network.
Background
With the popularization of 5G communication technology and the increasing progress of the technology of the Internet of things, frequency-using equipment is enabled to show explosive growth, which leads to the increasing shortage of frequency spectrum resources. In an actual cognitive wireless communication network, a cognitive user and a base station or other cognitive users carry out conventional communication, and a malicious user carries out interference on a communication link in the wireless communication network based on a certain attack task or purpose. Therefore, in order to avoid interference, it is necessary for users in the cognitive wireless communication network to be able to avoid interfered frequency bands in a complex and variable electromagnetic environment and quickly find available frequency spectrum. The frequency spectrum availability prediction method can predict future frequency spectrum gaps, cognitive users make judgment before interference does not come yet, and communication is carried out by using idle frequency bands, so that the anti-interference capability of a network is improved.
Currently, spectrum availability prediction is mainly focused on methods based on regression analysis, markov chain, moving average and machine learning. The first three methods have high computational complexity, such as updating of regression coefficients and increasing of markov orders, which lead to exponential increase of estimated parameters, and depend on prior information of frequency spectrum. In recent years, deep learning methods based on data driving are mainly used as methods based on machine learning, for example, methods based on MLP, ANN and LSTM are used, which can effectively complete prediction, but the accuracy of prediction needs to be improved.
Disclosure of Invention
Aiming at the defects in the prior art, the invention provides a spectrum availability prediction method and a spectrum availability prediction system based on a wireless communication network.
In a first aspect, the present invention provides a spectrum availability prediction method based on a wireless communication network, where the method is applied to a wireless communication network including a plurality of cognitive users and a malicious user, and includes:
acquiring a spectrum data set for deep learning model offline training, wherein the spectrum data set comprises a spectrum training set, a spectrum verification set and a spectrum test set;
setting Bayesian optimization cycle times, a hyper-parameter search space and deep learning model iteration times;
initializing the structure, weight, bias and hyper-parameters of the deep learning model;
calculating a loss of the spectral data set between an input and an output of the deep learning model;
updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished;
judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not;
if not, resetting the iteration times of the deep learning model until the loss curve and the accuracy rate fitting curve of the deep learning model are in a convergence state;
if yes, fine-tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization in the hyper-parameter search space until the circulation is finished to obtain an optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
Further, the deep learning model is a Bi-LSTM neural network.
Further, the calculating a loss of the spectral data set between an input and an output of the deep learning model comprises:
the loss of the spectral data set between the input and the output of the deep learning model is calculated according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;for predicted frequencySpectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
Further, the updating of the weight and the bias parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished comprises:
The algorithm for obtaining the optimal U is as follows:
wherein, W(s) and b(s) are respectively the weight and the bias of the deep learning model at the training time s; alpha is the deep learning model learning rate.
Further, the Bayesian optimization adopts a probability regression model of a Gaussian process and an expectedprovement acquisition function.
In a second aspect, the present invention provides a spectrum availability prediction system based on a wireless communication network, where the system is applied to a wireless communication network including a plurality of cognitive users and a malicious user, and the system includes:
the acquisition module is used for acquiring a spectrum data set for deep learning model offline training, wherein the spectrum data set comprises a spectrum training set, a spectrum verification set and a spectrum test set;
the first setting module is used for setting cycle times of Bayesian optimization, search space of hyper-parameters and iteration times of a deep learning model;
the initialization module is used for initializing the structure, the weight, the bias and the hyper-parameter of the deep learning model;
a calculation module for calculating a loss of the spectral data set between an input and an output of the deep learning model;
the updating module is used for updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished;
the judging module is used for judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not;
the second setting module is used for resetting the iteration times of the deep learning model under the condition that the judging module determines that the loss curve and the accuracy rate fitting curve of the deep learning model are not in the convergence state until the loss curve and the accuracy rate fitting curve of the deep learning model are in the convergence state;
the fine tuning module is used for fine tuning the super parameters of the deep learning model by adopting Bayesian optimization in the super parameter search space under the condition that the loss curve and the accuracy rate fitting curve of the deep learning model are determined to be in a convergence state by the judging module until the circulation is finished to obtain the optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
Further, the calculation module includes:
a calculation unit for calculating a loss of the spectral data set between an input and an output of the deep learning model according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;is the predicted spectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
Further, the update module includes:
The obtaining unit is used for obtaining the optimal U according to the following algorithm:
wherein, W(s) and b(s) are respectively used for training the weight and the bias of the initialized deep learning model for the s time; alpha is the deep learning model learning rate.
The invention provides a frequency spectrum availability prediction method and a frequency spectrum availability prediction system based on a wireless communication network, wherein the method comprises the steps of obtaining a frequency spectrum data set used for deep learning model off-line training, wherein the frequency spectrum data set comprises a frequency spectrum training set, a frequency spectrum verification set and a frequency spectrum test set; setting Bayesian optimization cycle times, a hyper-parameter search space and deep learning model iteration times; initializing the structure, weight, bias and hyper-parameters of the deep learning model; calculating a loss of the spectral data set between an input and an output of the deep learning model; updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished; judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not; if not, resetting the iteration times of the deep learning model until the loss curve and the accuracy rate fitting curve of the deep learning model are in a convergence state; if yes, fine-tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization in the hyper-parameter search space until the circulation is finished to obtain an optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model. By adopting the scheme, the invention can accurately predict the available frequency spectrum gap of the future time slot, so that the cognitive user is prevented from being interfered by malicious users, and the normal communication between the cognitive user and the base station and between the cognitive users is ensured.
Drawings
In order to more clearly illustrate the technical solution of the present invention, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious to those skilled in the art that other drawings can be obtained based on these drawings without creative efforts.
Fig. 1 is a flowchart illustrating a method for predicting spectrum availability based on a wireless communication network according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a spectrum availability prediction system based on a wireless communication network according to an embodiment of the present invention;
fig. 3 is a diagram illustrating a scenario in which a communication link of a wireless communication network is interfered according to an embodiment of the present invention;
fig. 4 is a schematic structural diagram of a spectrum availability prediction method based on a wireless communication network according to an embodiment of the present invention;
fig. 5 is a diagram illustrating a comparison of prediction errors between a spectrum availability prediction method based on a wireless communication network and other methods according to an embodiment of the present invention;
fig. 6 is a comparison graph of MAE and RMSE of a spectrum availability prediction method based on a wireless communication network according to an embodiment of the present invention with other methods;
fig. 7 is a graph comparing Rank correlations between a spectrum availability prediction method based on a wireless communication network and other methods according to an embodiment of the present invention;
fig. 8 is a graph comparing the prediction accuracy of a spectrum availability prediction method based on a wireless communication network with other methods under different snr according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
As shown in fig. 1, the present invention provides a spectrum availability prediction method based on a wireless communication network, where the method is applied to a wireless communication network including a plurality of cognitive users and a malicious user, and the method includes:
step S101, obtaining a spectrum data set for deep learning model off-line training, wherein the spectrum data set comprises a spectrum training set, a spectrum verification set and a spectrum test set.
In this step, the deep learning model is a Bi-LSTM neural network. We consider a wireless communication network consisting of a malicious user and a plurality of cognitive users. As shown in fig. 3, specifically, there areThe cognitive users communicate with the base station, all the cognitive users communicate in usable frequency bands, and the cognitive users do not interfere with each other. And one malicious user interferes the communication link between the j cognitive user and the base station, and the interference behavior of the malicious user is observed. The spectrum data measured at the nth time slot of the nth channel at the receiving end of the cognitive user can be represented as:
wherein,the signal energy (power level) of the j-th cognitive user,the signal energy (power level) of a malicious user,is gaussian white noise power.OrIn order to indicate the function,orIndicating that the signal is not present and,orIndicating that a signal is present. In the present embodiment, we consider a strong interference case. For the sake of theoretical analysis, we consider only the downlink of interfering communications. In addition, the cognitive user knows the communication frequency bands of other users. Based on the time-frequency diagram in fig. 3, the time-varying spectrum situation (available or unavailable) in different frequency bands can be used as a power matrix, wherein the rows of the matrix represent different channels (frequency bands) and the columns represent different time slots. Then the power vector for all slots of the nth channel is expressed as:
wherein,the power value of the T time slot of the nth channel is represented, wherein T is 1,2,3, …, T; n is 1,2,3, …, N. The power matrix for the N channel observations can be expressed as:
Γ=(p1,p2,p3,…,pn)T
thus, the spectrum availability prediction problem can be expressed as:
wherein,representing the predicted power values of N channels in t +1 time slot. M is the size of the observation time and p (-) is the conditional probability, i.e. the spectral data of the next time slot is predicted under the observation spectral data.
And simulating a scene of interference of a user communication link through a real experimental platform to generate a frequency spectrum data set for deep learning model training. The interference machine is realized by USRP 2943R, various interference modes (such as frequency sweep interference, fixed frequency interference and the like) are adopted, the interference-signal ratio is 10dB, and the interference signal strength (power) is greater than-50 dBm. The frequency band of the interference of the malicious user is between 1300MHz and 1700MHz, each interval of 20MHz is divided into one channel, the total number of channels is 20, and the SNR is 20 dB. The spectrum sensor (detector) collects power every 0.1 seconds and is considered to be disturbed (i.e., communication limited) when the power exceeds-50 dBm. A total of 4000 time slots of data are collected. The cognitive users are known to work in other cognitive users, and the power of the cognitive users is not more than-50 dBm.
And S102, setting cycle times of Bayesian optimization, search space of hyper-parameters and iteration times of a deep learning model.
Step S103, initializing the structure, weight, bias and hyper-parameters of the deep learning model.
In this step, the network configuration of the deep learning model is initialized: hidden layer is 1, LSTM unit number is 50, learning rate is 0.01, L2 regularization coefficient is 10-10The learning rate reduction factor is 0.5, MaxEpochs 200 times, and MiniBatch size 100.
As shown in FIG. 4, first, the observed interference spectral evolution (power value) matrix P passes through the forgetting gate of the Bi-LSTM neural network, which determines the last time output st-1And cell state ct-1Keeping some information to the current moment ct. Expressed by the forgetting gate formula:
ft=σ(Wf[st-1,pt]+bf)
where σ represents Sigmoid function, xtInput representing the current time, WfAnd bfRespectively representing the weight and bias of the forgetting gate. Then enters the input gate, which determines ptWhich information of (1) is reservedTo ctAnd implementing the state unit c by adopting Sigmoid and Tanh activation functionstAnd (4) updating. The formula is expressed as:
it=σ(Wi[st-1,pt]+bi)
wherein λ represents the Tanh function, WiAnd biRepresenting the weight and offset of the input gate, respectively. WcAnd bcRespectively representing the weight and the bias of the selected update unit. Updated ctComprises the following steps:
where denotes the dot product between elements. After selective memory and updating, the information finally enters the output gate. It accomplishes two tasks: filtering information and obtaining output; status unit ctIs selectively output to the next timing and output to the outside. The formula is expressed as:
ot=σ(Wo[st-1,pt]+bo)
st=ot*λ(ct)
wherein, WoAnd boRepresenting the weight and offset of the output gate, respectively. Then the data entering the full connection layer is expanded and predicted by the regression output layerFinally, a hard decision is made on the predicted power to determine the availability of the channel. Specifically, it can be expressed as:
where "0" represents that the channel is available and "1" represents that the channel is unavailable. λ is the decision threshold.
Step S104, calculating the loss of the spectrum data set between the input and the output of the deep learning model.
In this step, the loss of the spectral data set between the input and the output of the deep learning model is calculated according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;is the predicted spectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
And step S105, updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished.
In this step, let U be { W, b }, and the objective function be
Then, the Adam optimization algorithm for obtaining the optimal U is
Wherein, W(s) and b(s) are respectively the weight and the bias of the deep learning model at the training time s; alpha is the deep learning model learning rate.
And step S106, judging whether the loss curve and the accuracy rate fitting curve of the deep learning model are in a convergence state.
And step S107, if not, resetting the iteration times of the deep learning model until the loss curve and the accuracy rate fitting curve of the deep learning model are in a convergence state.
Step S108, if yes, fine-tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization in the hyper-parameter search space until the circulation is finished to obtain an optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
In this step, first, assume that there are m hyper-parameters to be optimized:
wherein,representing the value space of all hyper-parameters. R(m)Representing the value space of a certain hyper-parameter. From each hyper-parameter value space R(m)One value is selected to form a group of hyper-parameter combinations:
r=(r1,r2,...,rm)
wherein r ismA random value, r, representing the mth hyper-parameterm∈R(m). Then, an objective function f is defined:the prediction error of the deep learning verification set is used as an optimized objective function. The optimal set of hyper-parametric combinations can be expressed as:
therein, the domainIs a bounding box. Since the computational cost of the hyper-parameter adjustment is high, d is usually small. Here, depth science is usedLearning the error of the model verification set as the objective function f (r) to obtain the optimal hyper-parameter combination r*A known hyper-parameter combination data set is preset:
Θ={(r1,y1),(r2,y2),...,(ri,yi)},
wherein r is1Is a random set of hyper-parametric combinations, yi=f(ri). Then, the domain is usedA small set of samples in (1) initiates a probabilistic regression modelAfter the initialization phase, by optimizing an acquisition functionTo select new locations in the domain in turn, which uses a deep learning model as an inexpensive proxy for the expensive target f.
In practice, there is usually a quota on the total time available, then the number of loops g of the objective function f is set to 60. The specific implementation flow of the algorithm is shown in table 1. Here, ,by using the Gaussian processSuppose thatμ is the mean function and K is the covariance kernel. The kernel defines important parameters required for calculating the prediction distribution, including:
k(r)=(K(r,r1),...,K(r,ri)),Kj,k=K(rj,rk),
then, the user can use the device to perform the operation,
wherein,
y=(y1,...,yi)T
wherein,which represents the mean value of the estimates of the,representing the variance of the estimate.
TABLE 1 Bayesian optimization selection of optimal neural network hyper-parameters
Collection functionAn Expected Improvement is used. Assuming that f 'is the minimum observed so far, the Expected Improvement is such that the distance between the minimum found in the future and f' is the maximum, which can be expressed as:
u(r)=max(0,f'-f(r))
if f (r) is less than f', a prize is awarded, otherwise none. Then, solving r for the Expected Improvement acquisition function is expressed as
And using the trained target deep learning model (DL-BO model) for online prediction of spectrum availability. As can be seen from fig. 4, the power value of the current time slot is input into the trained DL-BO model to output the power value of the next time slot. Then, a hard decision is made on the power value to obtain the final spectral state 0 or 1.
To verify the effectiveness of the proposed DL-BO prediction method, the following experiments were performed: firstly, the proportion of the training set, the verification set and the test set in the data set is as follows: 4:1:1. The hyperparametric search space for bayesian optimization is shown in table 2. The optimal hyper-parameter combination obtained by Bayesian optimization is as follows: the number of hidden layers is 3, the number of Bi-LSTM units is 167, the learning rate is 0.046478, L2Coefficient of the regularization term is 3.2312-10. And inputting the test data into the trained optimal deep learning model to obtain a predicted value. The DL-BO method is shown in figure 5 in comparison to other neural network based prediction error pairs. As can be seen from FIG. 5, the prediction error of the proposed DL-BO method is significantly smaller than that of the MLP and LSTM methods in the selected 100 test data. From the prediction results, the trend of the prediction curve of the DL-BO is approximately the same as that of the observation curve, while the prediction curve based on the MLP method fluctuates greatly, the prediction value deviates from the observation value by a large distance, and the prediction curve based on the LSTM fluctuates to some extent but has a low overlapping rate. From the prediction error, the error between the predicted value and the observed value is limited to 20, 2 slots are provided for which the error of the DL-BO method exceeds 20, and 15 slots or more are provided for which the error exceeds 20 based on the MLP and LSTM methods.
FIG. 6 is a comparison of the DL-BO method with other MAEs and RMSE based neural network methods. It can be seen that the MAE of the DL-BO method is 11.2995 points lower than that based on MLP and 3.6266 points lower than that based on LSTM. The RMSE for the DL-BO method was 16.362 points lower than that based on MLP and 4.4838 points lower than that based on LSTM.
TABLE 2 hyper-parametric search space
FIG. 7 is a comparison of Rank correlation of DL-BO method with other neural network based methods. It can be seen that the Rank value for DL-BO is 0.4 higher than for MLP and 0.45 higher than for LSTM at different number of slots than for the other two methods. It is stated that Rank value and prediction accuracy are not sufficient conditions.
FIG. 8 is a comparison of the predicted RMSE for the DL-BO method versus other neural network based methods at different signal-to-noise ratios. It can be seen that the prediction accuracy of the DL-BO method is superior to that based on the MLP and LSTM methods. For example, the RMSE of the DL-BO method is 3.7dB lower than that based on MLP and 2dB lower than that based on LSTM, taking SNR as 20 dB.
As shown in fig. 2, an embodiment of the present invention further provides a spectrum availability prediction system based on a wireless communication network, where the system is applied to a wireless communication network including a plurality of cognitive users and a malicious user, and the system includes:
the acquisition module 10 is configured to acquire a spectrum data set used for deep learning model offline training, where the spectrum data set includes a spectrum training set, a spectrum verification set, and a spectrum test set;
the first setting module 20 is used for setting cycle times of Bayesian optimization, search space of hyper-parameters and iteration times of a deep learning model;
an initialization module 30 for initializing the structure, weights, biases, and hyper-parameters of the deep learning model;
a calculation module 40 for calculating a loss of the spectral data set between an input and an output of the deep learning model;
the updating module 50 is used for updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished;
a judging module 60, configured to judge whether a loss curve and an accuracy fitting curve of the deep learning model are in a convergence state;
a second setting module 70, configured to, when the determining module determines that the loss curve and the accuracy fit curve of the deep learning model are not in a convergence state, reset the iteration times of the deep learning model until the loss curve and the accuracy fit curve of the deep learning model are in the convergence state;
the fine tuning module 80 is used for fine tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization in the hyper-parameter search space under the condition that the loss curve and the accuracy rate fitting curve of the deep learning model are determined to be in a convergence state by the judging module until the circulation is finished to obtain the optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
Optionally, the calculation module includes:
a calculation unit for calculating a loss of the spectral data set between an input and an output of the deep learning model according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;is the predicted spectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
Optionally, the update module includes:
The obtaining unit is used for obtaining the optimal U according to the following algorithm:
wherein, W(s) and b(s) are respectively used for training the weight and the bias of the initialized deep learning model for the s time; alpha is the deep learning model learning rate.
The same and similar parts in the various embodiments in this specification may be referred to each other. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and the relevant points can be referred to the description in the method embodiment.
The invention has been described in detail with reference to specific embodiments and illustrative examples, but the description is not intended to be construed in a limiting sense. Those skilled in the art will appreciate that various equivalent substitutions, modifications or improvements may be made to the technical solution of the present invention and its embodiments without departing from the spirit and scope of the present invention, which fall within the scope of the present invention. The scope of the invention is defined by the appended claims.
Claims (8)
1. A spectrum availability prediction method based on a wireless communication network is applied to the wireless communication network comprising a plurality of cognitive users and a malicious user, and is characterized by comprising the following steps:
acquiring a spectrum data set for deep learning model offline training, wherein the spectrum data set comprises a spectrum training set, a spectrum verification set and a spectrum test set;
setting Bayesian optimization cycle times, a hyper-parameter search space and deep learning model iteration times;
initializing the structure, weight, bias and hyper-parameters of the deep learning model;
calculating a loss of the spectral data set between an input and an output of the deep learning model;
updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished;
judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not;
if not, resetting the iteration times of the deep learning model until the loss curve and the accuracy rate fitting curve of the deep learning model are in a convergence state;
if yes, fine-tuning the hyper-parameters of the deep learning model by adopting Bayesian optimization in the hyper-parameter search space until the circulation is finished to obtain an optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
2. The method for spectrum usability prediction according to claim 1 where the deep learning model is a Bi-LSTM neural network.
3. The method for spectrum availability prediction according to claim 2, wherein the calculating a loss of the spectral data set between an input and an output of a deep learning model comprises:
the loss of the spectral data set between the input and the output of the deep learning model is calculated according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;is the predicted spectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
4. The spectrum availability prediction method according to claim 3, wherein the updating of the weights and bias parameters of the deep learning model by gradient descent until the deep learning model training is finished comprises:
The algorithm for obtaining the optimal U is as follows:
wherein, W(s) and b(s) are respectively the weight and the bias of the deep learning model at the training time s; alpha is the deep learning model learning rate.
5. The method for spectrum availability prediction according to claim 1, wherein the bayesian optimization employs a probabilistic regression model of a gaussian process and an Expected Improvement acquisition function.
6. A spectrum availability prediction system based on a wireless communication network is applied to the wireless communication network comprising a plurality of cognitive users and a malicious user, and is characterized by comprising:
the acquisition module is used for acquiring a spectrum data set for deep learning model offline training, wherein the spectrum data set comprises a spectrum training set, a spectrum verification set and a spectrum test set;
the first setting module is used for setting cycle times of Bayesian optimization, search space of hyper-parameters and iteration times of a deep learning model;
the initialization module is used for initializing the structure, the weight, the bias and the hyper-parameter of the deep learning model;
a calculation module for calculating a loss of the spectral data set between an input and an output of the deep learning model;
the updating module is used for updating the weight and the offset parameter of the deep learning model by adopting gradient descent until the training of the deep learning model is finished;
the judging module is used for judging whether a loss curve and an accuracy rate fitting curve of the deep learning model are in a convergence state or not;
the second setting module is used for resetting the iteration times of the deep learning model under the condition that the judging module determines that the loss curve and the accuracy rate fitting curve of the deep learning model are not in the convergence state until the loss curve and the accuracy rate fitting curve of the deep learning model are in the convergence state;
the fine tuning module is used for fine tuning the super parameters of the deep learning model by adopting Bayesian optimization in the super parameter search space under the condition that the loss curve and the accuracy rate fitting curve of the deep learning model are determined to be in a convergence state by the judging module until the circulation is finished to obtain the optimal deep learning model; and predicting the available frequency spectrum on line by using the obtained optimal deep learning model.
7. The spectrum availability prediction system of claim 6, wherein the computation module comprises:
a calculation unit for calculating a loss of the spectral data set between an input and an output of the deep learning model according to the following formula:
wherein χ is the loss of the spectral data between the input and the output of the deep learning model; t is a time slot; t is the total time slot number; p is a radical oftIs the observed spectral power;is the predicted spectral power; eta is a regularization coefficient; l is the layer number of the Bi-LSTM neural network; w is alIs the weight vector of the l-th layer Bi-LSTM neural network.
8. The spectrum availability prediction system of claim 7, wherein the update module comprises:
The obtaining unit is used for obtaining the optimal U according to the following algorithm:
wherein, W(s) and b(s) are respectively used for training the weight and the bias of the initialized deep learning model for the s time; alpha is the deep learning model learning rate.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111507860.0A CN114298166A (en) | 2021-12-10 | 2021-12-10 | Spectrum availability prediction method and system based on wireless communication network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202111507860.0A CN114298166A (en) | 2021-12-10 | 2021-12-10 | Spectrum availability prediction method and system based on wireless communication network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN114298166A true CN114298166A (en) | 2022-04-08 |
Family
ID=80967522
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202111507860.0A Pending CN114298166A (en) | 2021-12-10 | 2021-12-10 | Spectrum availability prediction method and system based on wireless communication network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN114298166A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115208462A (en) * | 2022-07-14 | 2022-10-18 | 上海交通大学 | Optical module control parameter optimization method and system of optical communication system |
CN118052298A (en) * | 2024-02-27 | 2024-05-17 | 中国人民解放军海军特色医学中心 | Cognitive intervention system based on deep learning model |
-
2021
- 2021-12-10 CN CN202111507860.0A patent/CN114298166A/en active Pending
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN115208462A (en) * | 2022-07-14 | 2022-10-18 | 上海交通大学 | Optical module control parameter optimization method and system of optical communication system |
CN118052298A (en) * | 2024-02-27 | 2024-05-17 | 中国人民解放军海军特色医学中心 | Cognitive intervention system based on deep learning model |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US12021572B2 (en) | Systems and methods for wireless signal configuration by a neural network | |
CN114298166A (en) | Spectrum availability prediction method and system based on wireless communication network | |
Ak et al. | Avoiding jammers: A reinforcement learning approach | |
WO2021036414A1 (en) | Co-channel interference prediction method for satellite-to-ground downlink under low earth orbit satellite constellation | |
CN107180260B (en) | Short wave communication frequency selecting method based on Evolutionary Neural Network | |
CN114509732B (en) | Deep reinforcement learning anti-interference method of frequency agile radar | |
CN108924847B (en) | Cognitive radio frequency spectrum prediction method and device based on ANN | |
CN111786738B (en) | Anti-interference learning network structure based on long-term and short-term memory and learning method | |
Jacoby et al. | Short-term prediction of the attenuation in a commercial microwave link using LSTM-based RNN | |
Sun et al. | Learning to continuously optimize wireless resource in episodically dynamic environment | |
CN112307927A (en) | BP network-based identification research for MPSK signals in non-cooperative communication | |
CN113840297A (en) | Frequency spectrum prediction method based on radio frequency machine learning model drive | |
CN110290466B (en) | Floor distinguishing method, device, equipment and computer storage medium | |
Shachi et al. | Convolutional neural network for cooperative spectrum sensing with spatio-temporal dataset | |
CN113938897A (en) | Main user-friendly anti-interference dynamic spectrum access method | |
Zhang et al. | Machine learning based protocol classification in unlicensed 5 GHz bands | |
CN113115355A (en) | Power distribution method based on deep reinforcement learning in D2D system | |
CN114337881B (en) | Wireless spectrum intelligent sensing method based on multi-unmanned aerial vehicle distribution and LMS | |
CN114173421B (en) | LoRa logic channel based on deep reinforcement learning and power distribution method | |
Nikoukar et al. | Predictive interference management for wireless channels in the Internet of Things | |
Zhong et al. | Learning-Based Robust Anomaly Detection in the Presence of Adversarial Attacks | |
CN114912653A (en) | Short-term load combined prediction method based on self-adaptive chirp modal decomposition and SSA-BilSTM | |
CN113747447A (en) | Double-action reinforcement learning frequency spectrum access method and system based on priori knowledge | |
CN113747386A (en) | Intelligent power control method in cognitive radio network spectrum sharing | |
Wimalajeewa et al. | PSO for constrained optimization: Optimal power scheduling for correlated data fusion in wireless sensor networks |
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 |