CN118041471B - Spectrum sensing method and system based on machine learning logistic regression algorithm - Google Patents
Spectrum sensing method and system based on machine learning logistic regression algorithm Download PDFInfo
- Publication number
- CN118041471B CN118041471B CN202410432093.9A CN202410432093A CN118041471B CN 118041471 B CN118041471 B CN 118041471B CN 202410432093 A CN202410432093 A CN 202410432093A CN 118041471 B CN118041471 B CN 118041471B
- Authority
- CN
- China
- Prior art keywords
- signal
- value
- logistic regression
- activation function
- data
- 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
- 238000001228 spectrum Methods 0.000 title claims abstract description 72
- 238000007477 logistic regression Methods 0.000 title claims abstract description 60
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000010801 machine learning Methods 0.000 title claims abstract description 16
- 238000000605 extraction Methods 0.000 claims abstract description 21
- 238000007781 pre-processing Methods 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 15
- 230000009467 reduction Effects 0.000 claims abstract description 12
- 230000001149 cognitive effect Effects 0.000 claims abstract description 9
- 230000006870 function Effects 0.000 claims description 78
- 230000004913 activation Effects 0.000 claims description 41
- 238000012549 training Methods 0.000 claims description 24
- 238000012417 linear regression Methods 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000010606 normalization Methods 0.000 claims description 9
- 238000004364 calculation method Methods 0.000 claims description 6
- 238000013507 mapping Methods 0.000 claims description 6
- 238000012847 principal component analysis method Methods 0.000 claims description 4
- 238000002759 z-score normalization Methods 0.000 claims description 2
- 230000008054 signal transmission Effects 0.000 abstract description 2
- 238000001514 detection method Methods 0.000 description 9
- 238000012360 testing method Methods 0.000 description 8
- 230000008901 benefit Effects 0.000 description 3
- 230000008859 change Effects 0.000 description 3
- 238000009826 distribution Methods 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000005457 optimization Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 238000003860 storage Methods 0.000 description 2
- 102000002274 Matrix Metalloproteinases Human genes 0.000 description 1
- 108010000684 Matrix Metalloproteinases Proteins 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 238000011478 gradient descent method Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000000513 principal component analysis Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
- 239000013598 vector Substances 0.000 description 1
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Landscapes
- Complex Calculations (AREA)
Abstract
The invention relates to the technical field of signal transmission, in particular to a frequency spectrum sensing method and a frequency spectrum sensing system based on a machine learning logistic regression algorithm, wherein the method comprises the following steps: s1, collecting signals received by a cognitive node, and analyzing signal data from the signals; s2, preprocessing the analyzed signal data, wherein the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction; and S3, the pre-trained logistic regression model processes the pre-processed signal data, outputs a probability value for obtaining whether the frequency spectrum is idle or not, judges that the frequency spectrum is idle if the probability value is larger than a threshold value, and judges that the frequency spectrum is not idle if the probability value is smaller than or equal to the threshold value. The system comprises a data acquisition module, a preprocessing module and a frequency spectrum sensing module. The spectrum sensing method and the device are used for spectrum sensing, and are high in accuracy, small in calculated amount and high in sensing efficiency.
Description
Technical Field
The invention relates to the technical field of signal transmission, in particular to a frequency spectrum sensing method and a frequency spectrum sensing system based on a machine learning logistic regression algorithm.
Background
Spectrum sensing is one of the key technologies of cognitive radio, and its main function is to find spectrum holes. The cognitive radio dynamically and intelligently selects the frequency band and mode of transmission by sensing the surrounding electromagnetic spectrum environment in real time so as to maximally meet the requirements of users. Spectrum sensing techniques mainly involve a physical layer that mainly focuses on various specific local detection algorithms and a link layer that mainly focuses on collaboration among users and control and optimization of sensing mechanisms.
The main sensing algorithms include an energy detection algorithm, a matched filter detection algorithm, a cyclostationary feature detection algorithm, a covariance matrix detection algorithm, and collaborative sensing. In addition, in order to overcome the drawbacks of local detection and further improve the detection performance, collaborative awareness has been widely and deeply studied. The specific method of cooperative spectrum sensing mainly comprises the following steps: centralized cooperative spectrum sensing, distributed cooperative spectrum sensing, and cooperative spectrum sensing based on energy detection. In this type of centralized collaborative spectrum sensing, all secondary users (or secondary users) send their sensing results to a central node, which then makes decisions. This type has the advantage of a simple decision process, but has the disadvantage that a large amount of communication resources are required to transmit the perceived result, and that the central node may become a bottleneck. The distributed cooperative spectrum sensing is to directly exchange sensing results among secondary users and make decisions through a distributed algorithm. This type has the advantage that communication resources can be saved, but has the disadvantage that the decision process can be relatively complex. The majority of collaborative spectrum sensing based on energy detection is to perform weighted combination on hard/soft decision results to achieve fusion of sensing or decision results, and the weighted combination method comprises majority decision, equal Gain Combination (EGC), maximum Ratio Combination (MRC), user selection and the like.
The methods have respective advantages and disadvantages, and can be selected according to specific application scenes. However, these methods have a common disadvantage in that the calculation amount is relatively large, which in turn results in low processing efficiency.
Disclosure of Invention
The invention aims to provide a frequency spectrum sensing method and a frequency spectrum sensing system based on a machine learning logistic regression algorithm so as to improve the processing efficiency of frequency spectrum sensing.
In order to achieve the above object, the present invention provides the following technical solutions:
a frequency spectrum sensing method based on a machine learning logistic regression algorithm comprises the following steps:
s1, collecting signals received by a cognitive node, and analyzing signal data from the signals;
s2, preprocessing the analyzed signal data, wherein the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction;
And S3, the pre-trained logistic regression model processes the pre-processed signal data, outputs a probability value for obtaining whether the frequency spectrum is idle or not, judges that the frequency spectrum is idle if the probability value is larger than a threshold value, and judges that the frequency spectrum is not idle if the probability value is smaller than or equal to the threshold value.
In the scheme, the logistic regression model is adopted to process the signal data, the algorithm is simple, the calculated amount is small during classification, and therefore the processing speed is high, the efficiency is high, and the storage resource is low.
In the step S1, the analyzed signal data comprise application types, signal strength, required bandwidth, allocated bandwidth, signal energy and signal time delay; the logistic regression model comprises a linear regression function and an activation function, wherein the linear regression function is input, the activation function is output, and the linear regression function is:
；
The activation function is: ；
Wherein h (x) is a target value, x1, x2, x3, x4, x5 and x6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal time delay, w1, w2, w3, w4, w5 and w6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and weight of signal time delay, b is an adjustment coefficient, and θ T x is a matrix expression form of a signal data linear regression equation.
The scheme particularly selects the required bandwidth, the allocated bandwidth and the signal delay time of the signals as characteristics, and can be suitable for spectrum sensing of 5G and 6G signals.
In the training process of the logistic regression model, the output value of the activation function is mapped into (-1, 1) according to the following formula:
；
the loss calculation is then performed using the following function:
；
Wherein, Representing the output value of the mapped activation function, g being the output value of the activation function, g max and g min being the maximum and minimum values of the output value of the activation function, respectively, y representing the correct class, taking the value 1 or-1, beta being the mapping value of the threshold between (-1, 1), if/>Then the loss is/>; If/>The loss is 0.
In the scheme, the prediction value is mapped between (-1, 1) firstly, then the loss value is calculated through the optimized loss function, so that signals which are difficult to classify and finally cause false alarms or missing alarms are focused, the prediction capability of the logistic regression model can be improved, and more accurate results are output.
In the step S2, a dictionary feature extraction method is adopted to perform feature extraction processing on the signal data.
The application type of the signal data, the signal length, the required bandwidth and the distribution bandwidth of the signal, the energy of the signal and other data types are all of the type features, and after the dictionary features are extracted, the type non-numerical data are converted into numerical data, so that the logistic regression algorithm can better process the data.
In the step S2, the dimensionless normalization includes normalization and z-score normalization.
In the step S2, a principal component analysis method is adopted to perform characteristic dimension reduction treatment on the data after the z-score standardization.
A spectrum sensing system based on a machine learning logistic regression algorithm, comprising:
The data acquisition module is used for acquiring signals received by the cognitive node and analyzing signal data from the signals;
The preprocessing module is used for preprocessing the analyzed signal data, and the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction;
And the spectrum sensing module is used for processing the preprocessed signal data by utilizing a pre-trained logistic regression model, outputting a probability value for obtaining whether the spectrum is idle or not, judging that the spectrum is idle if the probability value is larger than a threshold value, and judging that the spectrum is not idle if the probability value is smaller than or equal to the threshold value.
The signal data analyzed by the data acquisition module comprises application type, signal strength, required bandwidth, allocated bandwidth, signal energy and signal time delay; the logistic regression model comprises a linear regression function and an activation function, wherein the linear regression function is input, the activation function is output, and the linear regression function is:
；
The activation function is: ；
Wherein h (x) is a target value, x1, x2, x3, x4, x5 and x6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal time delay, w1, w2, w3, w4, w5 and w6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and weight of signal time delay, b is an adjustment coefficient, and θ T x is a matrix expression form of a signal data linear regression equation.
The model training module is used for training to obtain the logistic regression model, and in the training process of the logistic regression model, the output value of the activation function is mapped into (-1, 1) according to the following formula:
；
the loss calculation is then performed using the following function:
；
Wherein, Representing the output value of the mapped activation function, g being the output value of the activation function, g max and g min being the maximum and minimum values of the output value of the activation function, respectively, y representing the correct class, taking the value 1 or-1, beta being the mapping value of the threshold between (-1, 1), if/>Then the loss is/>; If/>The loss is 0.
The preprocessing module adopts a dictionary feature extraction method to perform feature extraction processing on the signal data.
Compared with the prior art, the invention has the following beneficial effects:
Considering the problem of improving the system accuracy when the signal to noise ratio is low and the stability of the system performance when the malicious attack is caused by the development of the spectrum sensing technology, compared with the application of other algorithms in the spectrum sensing field, the logistic regression algorithm is simple to realize, the calculated amount is very small during classification, the speed is very high, the storage resource is low, the output result is not more than one classification result, the probability value can be output, the probability score of the observation sample can be conveniently observed, and the data can be more reasonably optimized. Training is performed on the basis of existing label data, so that the prediction result of the model is more accurate, the spectrum sensing result is actually a two-two classification problem, and the method is highly compatible with the action of a logistic regression algorithm.
And the spectrum sensing is performed based on the dictionary feature extraction and the logistic regression algorithm of the range loss function, so that the logistic regression model has good performance and high accuracy of sensing results.
Drawings
Fig. 1 is a flowchart of a spectrum sensing method based on a machine learning logistic regression algorithm provided in an embodiment.
FIG. 2 is a graph of ROC for testing a logistic regression model based on a test set.
FIG. 3 is a graph of evaluation results of testing a logistic regression model based on a test set.
Fig. 4 is a block diagram of a spectrum sensing system based on a machine learning logistic regression algorithm provided in an embodiment.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the invention, as presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be made by a person skilled in the art without making any inventive effort, are intended to be within the scope of the present invention.
Referring to fig. 1, the spectrum sensing method based on the machine learning logistic regression algorithm provided in the present embodiment includes the following steps:
s1, collecting signals received by a cognitive node, and analyzing signal data from the signals;
S2, preprocessing the analyzed signal data, wherein the preprocessing comprises feature extraction, dimensionless and feature dimension reduction;
And S3, the pre-trained logistic regression model processes the pre-processed signal data, outputs a probability value for obtaining whether the frequency spectrum is idle or not, judges that the frequency spectrum is idle if the probability value is larger than a threshold value, and judges that the frequency spectrum is not idle if the probability value is smaller than or equal to the threshold value.
The method can be used for sensing the 5G or 6G spectrum signals, and in S1, the analyzed signal data comprises application type, signal strength, required bandwidth and distribution bandwidth of signals, signal energy and signal delay time.
5G, 6G provides higher data rates, lower latency, and higher connection density. The bandwidth of a 5G network also depends on the spectral range it uses. It is expected that 6G will provide higher performance wireless connections and excellent user experience, peak rates can reach Tbps, and user experience rates can reach 10-100 gbps. The 6G will use the full band spectrum from millimeter wave to THz to visible light. The bandwidth and latency data of 5G and 6G will be more representative than 4G. Therefore, the method of the embodiment particularly selects the required bandwidth, the allocated bandwidth and the signal delay time of the signal as characteristic values.
In this embodiment, in S2, for the 5G or 6G signal, the feature extraction processing is performed on the signal data by using a dictionary feature extraction method, and the non-numeric signal data is generated into a sparse matrix or a one-hot encoding matrix, so that the sparse matrix or the one-hot encoding matrix is converted into numeric data. The application type of the signal data, the signal length, the required bandwidth of the signal, the allocation bandwidth, the signal energy and the data type of the signal delay time are all of the category type characteristics, and the category type non-numerical data is converted into numerical data after the dictionary characteristic is extracted in the embodiment, so that the logistic regression model can be facilitated to better process the data, and then the logistic regression model can be facilitated to output more reliable classification results.
The dimensionless process includes normalization and normalization. In this embodiment, because of the 5G signal data adopted, different dimensions and dimension units exist between different features, and the data orders of magnitude differ greatly. The bandwidth is defined according to the 3GPP protocol, and the maximum single carrier of 5G needs to support 100MHz, and millimeter waves even reach 400MHz per carrier. The protocol also specifies that 5G can support a maximum of 16 carrier aggregation, which means that the required bandwidth of 5G can reach 1.6GHz to 6.4GHz at maximum. And its signal strength is typically between-50 dBm (very good signal) and-120 dBm (very poor signal or no signal region). The delay is more in the order of milliseconds. As for 6G, at present, the signal strength, the required bandwidth and the allocated bandwidth are not clearly defined yet in the development stage, but the signal strength is smaller than that of 5G, the required bandwidth and the allocated bandwidth are larger than that of 5G, and the delay can reach microsecond level. It can be seen that the magnitude of the characteristic quantity values of the 5G, 6G signals differ very much. This can severely impact the results of the data analysis. Thus, such dimensional effects can be eliminated by dimensionless treatment so that the individual features are on the same order of magnitude, thereby improving the comparability of the data.
Normalization is a range transform method, also called a min-max method, in which numerical data after feature extraction is mapped between [0,1] by linear transformation.
The normalization method uses z-score, with the following formula:
Wherein mean is the mean of the features, σ is the standard deviation, X is the normalized data, and X' is the normalized data. After normalization, the data were converted to a mean value of 0 and standard deviation of 1.
In this embodiment, the feature dimension reduction processing is performed on the normalized data by using a principal component analysis method. Specifically, the 5G signal employed in this embodiment has a large data size, and includes a large amount of redundant information and noise. The principal component analysis method can effectively reduce the dimensionality of the data, simplify the complexity of the data, and eliminate noise and redundant information in the data, thereby improving the accuracy and reliability of the data. The principal component analysis mainly comprises the following steps:
1) The mean value is removed (i.e., the center is removed), i.e., each bit feature is subtracted from the respective mean value.
2) Calculating covariance matrixIn the present invention, n is the number of types of eigenvalues, and X is a matrix of eigenvalue data.
3) Covariance matrix solving by eigenvalue decomposition methodIs described.
4) And sorting the eigenvalues from large to small, selecting the largest k eigenvalues, and then respectively taking the k eigenvectors corresponding to the largest k eigenvalues as row vectors to form an eigenvector matrix P.
5) The data (data constituting the X matrix) is converted into a new space constructed of k eigenvectors, i.e., y=px.
After feature dimension reduction, the normalized high-dimensional data is kept with some of the most important features, and noise and unimportant features are removed. The data after feature dimension reduction can be input into a logistic regression model.
The logistic regression model applied in S3 is obtained through extensive data training. After a large number of signals are acquired, the same processing method as S1 and S2 is adopted for processing, namely, the acquired signals are firstly analyzed to obtain signal data, then the signal data are preprocessed, and the processed signal data are used as training samples for regression model training. The training samples need to be marked with characteristic values and target values, and the application type of signal data, signal strength, required bandwidth and distribution bandwidth of signals, energy of signals and signal time delay are used as characteristic values, and whether a frequency spectrum is idle or not is used as the target value.
The logistic regression model is to perform linear prediction on input data, input the result into an activation function, and judge the obtained result. The linear equation of logistic regression in this embodiment is:
Wherein h (x) is a target value, x1, x2, x3, x4, x5 and x6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal delay, w1, w2, w3, w4, w5 and w6 respectively represent weights of application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal delay, and b is an adjustment coefficient.
The loss function of linear regression is:
y i is the true value of the ith training sample, i.e. the target value corresponding to the sample, h w(xi) is the function value predicted by the combination of the characteristic values of the ith training sample, m represents the number of samples, i is the number of samples, i=1, 2, … m. The final calculated J (x) is expressed as a loss value.
In this embodiment, the initial logistic regression model is optimized by using a gradient descent method, and the weight W in the logistic regression model is solved, so that the loss is minimized. The gradient drop formula is:
Where w i is the linear equation weight coefficient, J (x) is the loss function, and α is the learning rate. In the application to signal processing, the learning rate alpha is set to be 0.7-1 in consideration of large information quantity of signal data so as to achieve a faster iteration rate, and specific numerical values can be adjusted according to actual conditions.
Inputting the optimized linear regression result into an activation function:
Wherein g (theta T x) is a probability value generated after inputting a regression result into an activation function, theta T x is a signal data linear regression equation matrix expression form, and theta T is a signal data linear equation weight matrix.
Inputting the regression result into the activation function, and outputting a probability value in the [0,1] interval. In order to achieve higher prediction rate and reduce false alarm and missing alarm probability, the threshold value is set to 0.8, if the probability value is larger than the threshold value, the spectrum is judged to be idle, and if the probability value is smaller than the threshold value, the spectrum is judged to be not idle. The penalty of a conventional logistic regression algorithm is called log likelihood penalty, and the formula is as follows:
Wherein y i is a true value corresponding to each sample, that is, a target value of each sample, and g (θ T x) is a probability value generated by inputting a regression result into the activation function.
Y is a true value, when y=1, the closer the predicted value g (θ T x) is to 1, the smaller the loss, and vice versa when y=0. And (3) optimizing by using a gradient descent algorithm, updating each weight in the logistic regression model, reducing the value of a loss function, improving the probability of originally belonging to the category 1, and reducing the probability of originally belonging to the category 0.
In the present invention, a finger Loss function is innovatively employed instead of log likelihood Loss. Theoretically, both the finger Loss and the logic Loss function can be used for optimization of the classification problem. The main difference between the finger Loss function and the logic Loss function is their way to handle misclassified samples. The finger Loss function focuses only on those samples that are difficult to classify or that are misclassified, with a Loss of 0 for samples that are correctly classified and have a sufficiently high confidence. However, in terms of spectrum sensing, signals that are difficult to classify and ultimately cause false alarms or false misses are of greater concern. Therefore, the use of the finger Loss function may be better when there is much noise in the data set or when there is more concern about the samples of the classification boundary (i.e., signal data that is not easily judged to be prone to false alarms or false misses).
Therefore, the invention adopts the Range Loss function to replace the log likelihood Loss, and optimizes the Range Loss function. The mathematical expression of the optimized finger Loss function is as follows:
Wherein, The output value representing the activation function (between (-1, 1) after mapping) is typically a soft result (meaning that the output is not-1 or 1, the result is any value between-1 and 1). y represents the correct class, generally indicated at-1 and 1. In this case, y=1 indicates that the spectrum is free, and y= -1 indicates that the spectrum is busy. Beta is the mapping value of the threshold between (-1, 1) of the activation function if/>Then the loss is/>; If/>The loss is 0. In the conventional range Loss function, if the input value is a probability value, the Loss can only approach 0 infinitely and cannot be 0. The optimized finger Loss function is obtained by adding a coefficient/>, which is set according to a threshold valueThe sample loss that is greater than the prediction threshold and that predicts correctly can be set directly to 0, making itself more focused on data that is not readily determinable.
Further processing of the output of the activation function is required before this loss function is used. And finally inputting a probability value of which the result is just 0 to 1 according to a logistic regression algorithm, wherein the predicted result meets the soft result requirement. But its value interval is (0, 1) and y represents the correct class-1 and 1, using this probability value no correct assessment can be made of the loss of y= -1, and therefore its value range needs to be mapped to (-1, 1).
The output result of the activation function is normalized, and the formula is as follows:
According to this formula, the result can be mapped between (-1, 1). g is the output value of the activation function, and g max and g min are the maximum and minimum values, respectively, of the output value of the activation function. At this time, the threshold value according to the preset activation function is 0.8, and the mapped beta value is 0.6, namely the coefficient 1.66. The processed result can be used for Loss evaluation of the optimized Range Loss function. And performing iterative optimization on the final model according to the loss result to reach the minimum loss value, and completing model training at the moment.
And after training, evaluating the logistic regression model so as to optimize parameters in the model. In logistic regression, model evaluations were performed by calculating their accuracy, recall, F1-score, and ROC curve and AUC index. The accuracy rate is the proportion of the predicted result to the true positive example in the positive example sample, the recall rate is the proportion of the predicted result to the positive example in the true positive example sample, and the F1-score formula is as follows:
Precision denotes Precision rate, recall denotes Recall rate. The nature of the F1-score mathematical model is that the product of the precision and recall divided by the sum of the precision and recall, i.e., the higher the precision and recall the more stable the model. The ROC curve is a graph showing the relationship between the true case rate (also called recall rate or sensitivity) and the false positive case rate of the model at different thresholds, the horizontal axis being the false positive case rate (FPR), representing the proportion of samples that are actually negative but are erroneously predicted as positive, and the vertical axis being the true case rate (TPR), representing the proportion of samples that are actually positive and are correctly predicted as positive. As the model threshold changes, the true and false positive rates change, and the ROC curve shows this change. AUC is the area under the ROC curve and represents the distinguishing capability of the model to positive and negative examples, and the value of AUC ranges from 0.5 to 1, and the closer to 1, the better the model performance.
The method comprises the steps of taking preprocessed signal data of 400 signals as a data set, taking signal data of 300 signals as a training set, taking signal data of 100 signals as a test set, firstly training by using the training set to obtain a logistic regression model, and then testing the trained logistic regression model by using the test set. The test results are shown in fig. 2 and 3, it can be seen that prediction is performed by a logistic regression algorithm, wherein the accuracy and recall rate of the idle state are both over 0.93, the F1-score coefficient can also be over 0.92, and the calculated AUC index also reaches 0.95, which is closer to 1, thus indicating that the model prediction performance is better. That is, the spectrum sensing method of the present embodiment has high reliability.
Referring to fig. 4, based on the same inventive concept, in this embodiment, a spectrum sensing system based on a machine learning logistic regression algorithm is provided at the same time, including:
The data acquisition module is used for acquiring signals received by the cognitive node and analyzing signal data from the signals;
The preprocessing module is used for preprocessing the analyzed signal data, and the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction;
The model training module is used for training to obtain a logistic regression model based on the preprocessed sample data;
And the spectrum sensing module is used for processing the preprocessed signal data by utilizing a pre-trained logistic regression model, outputting a probability value for obtaining whether the spectrum is idle or not, judging that the spectrum is idle if the probability value is larger than a threshold value, and judging that the spectrum is not idle if the probability value is smaller than or equal to the threshold value.
For the specific processing method or flow of each functional module, reference may be made to the related description in the foregoing method flow, for example, in the preprocessing module, a dictionary feature extraction method is adopted to perform feature extraction processing on the signal data; in another example, in the model training module, during the training process of the logistic regression model, the output value of the activation function is mapped into (-1, 1), and then the Loss calculation is performed by using the optimized change Loss function. The description is omitted for the sake of brevity.
The foregoing is merely illustrative of the present invention, and the present invention is not limited thereto, and any person skilled in the art will readily recognize that variations or substitutions are within the scope of the present invention.
Claims (6)
1. The frequency spectrum sensing method based on the machine learning logistic regression algorithm is characterized by comprising the following steps of:
s1, collecting signals received by a cognitive node, and analyzing signal data from the signals;
s2, preprocessing the analyzed signal data, wherein the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction;
S3, the pre-trained logistic regression model processes the pre-processed signal data and outputs a probability value for obtaining whether the frequency spectrum is idle or not, if the probability value is larger than a threshold value, the frequency spectrum is judged to be idle, and if the probability value is smaller than or equal to the threshold value, the frequency spectrum is judged to be idle;
in the step S1, the analyzed signal data comprise application types, signal strength, required bandwidth, allocated bandwidth, signal energy and signal time delay; the logistic regression model comprises a linear regression function and an activation function, wherein the linear regression function is input, the activation function is output, and the linear regression function is:
；
The activation function is: ；
Wherein h (x) is a target value, x1, x2, x3, x4, x5 and x6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal time delay, w1, w2, w3, w4, w5 and w6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and weight of signal time delay, b is an adjustment coefficient, and θ T x is a signal data linear regression equation matrix expression form;
in the training process of the logistic regression model, the output value of the activation function is mapped into (-1, 1) according to the following formula:
；
the loss calculation is then performed using the following function:
；
Wherein, Representing the output value of the mapped activation function, g being the output value of the activation function, g max and g min being the maximum and minimum values of the output value of the activation function, respectively, y representing the correct class, taking the value 1 or-1, beta being the mapping value of the threshold between (-1, 1), if/>Then the loss is/>; If/>The loss is 0.
2. The spectrum sensing method based on the machine learning logistic regression algorithm according to claim 1, wherein in S2, the feature extraction processing is performed on the signal data by using a dictionary feature extraction method.
3. The machine learning logistic regression algorithm based spectrum sensing method according to claim 1, wherein in S2, the dimensionless normalization includes normalization and z-score normalization.
4. The spectrum sensing method based on the machine learning logistic regression algorithm according to claim 2, wherein in S2, the feature dimension reduction processing is performed on the z-score normalized data by using a principal component analysis method.
5. A machine learning logistic regression algorithm-based spectrum sensing system, comprising:
The data acquisition module is used for acquiring signals received by the cognitive node and analyzing signal data from the signals;
The preprocessing module is used for preprocessing the analyzed signal data, and the preprocessing sequentially comprises feature extraction, dimensionless and feature dimension reduction;
The spectrum sensing module is used for processing the preprocessed signal data by utilizing a logistic regression model which is trained in advance, outputting a probability value for obtaining whether the frequency spectrum is idle, judging that the frequency spectrum is idle when the probability value is larger than a threshold value, and judging that the frequency spectrum is not idle when the probability value is smaller than or equal to the threshold value;
The signal data analyzed by the data acquisition module comprises application type, signal strength, required bandwidth, allocated bandwidth, signal energy and signal time delay; the logistic regression model comprises a linear regression function and an activation function, wherein the linear regression function is input, the activation function is output, and the linear regression function is:
；
The activation function is: ；
Wherein h (x) is a target value, x1, x2, x3, x4, x5 and x6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and signal time delay, w1, w2, w3, w4, w5 and w6 respectively represent application type, signal intensity, required bandwidth, allocated bandwidth, signal energy and weight of signal time delay, b is an adjustment coefficient, and θ T x is a signal data linear regression equation matrix expression form;
The model training module is used for training to obtain the logistic regression model, and in the training process of the logistic regression model, the output value of the activation function is mapped into (-1, 1) according to the following formula:
；
the loss calculation is then performed using the following function:
；
Wherein, Representing the output value of the mapped activation function, g being the output value of the activation function, g max and g min being the maximum and minimum values of the output value of the activation function, respectively, y representing the correct class, taking the value 1 or-1, beta being the mapping value of the threshold between (-1, 1), if/>Then the loss is/>; If/>The loss is 0.
6. The system of claim 5, wherein the preprocessing module performs feature extraction processing on the signal data by using a dictionary feature extraction method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410432093.9A CN118041471B (en) | 2024-04-11 | 2024-04-11 | Spectrum sensing method and system based on machine learning logistic regression algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202410432093.9A CN118041471B (en) | 2024-04-11 | 2024-04-11 | Spectrum sensing method and system based on machine learning logistic regression algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN118041471A CN118041471A (en) | 2024-05-14 |
| CN118041471B true CN118041471B (en) | 2024-06-11 |
Family
ID=90989751
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202410432093.9A Active CN118041471B (en) | 2024-04-11 | 2024-04-11 | Spectrum sensing method and system based on machine learning logistic regression algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118041471B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| AU2019101446A4 (en) * | 2019-11-23 | 2020-01-23 | Southwest University | A fast distributed strategy for large-scale machine learning |
| CN111342984A (en) * | 2018-12-18 | 2020-06-26 | 电信科学技术研究院有限公司 | Information processing method, system and device |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109245840B (en) * | 2018-10-15 | 2021-02-02 | 哈尔滨工业大学 | Frequency spectrum prediction method based on convolutional neural network in cognitive radio system |
-
2024
- 2024-04-11 CN CN202410432093.9A patent/CN118041471B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111342984A (en) * | 2018-12-18 | 2020-06-26 | 电信科学技术研究院有限公司 | Information processing method, system and device |
| AU2019101446A4 (en) * | 2019-11-23 | 2020-01-23 | Southwest University | A fast distributed strategy for large-scale machine learning |
Non-Patent Citations (3)
| Title |
|---|
| Sesham Srinu ; Manfred Jeremia ; Abisai. F.M.S.Development of Logistic regression based spectrum sensing algorithm using extreme eigenvalues.2023 International Conference in Advances in Power, Signal, and Information Technology (APSIT).2023,全文. * |
| 用户移动性下的毫米波自适应调制策略研究;李葛，杨博，文彦博，等;无线电工程;20240405;全文 * |
| 认知无线电中基于特征值的频谱感知增强技术研究;李贺;中国学位论文全文数据库;20230223;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN118041471A (en) | 2024-05-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109547133B (en) | SVM high-efficiency spectrum sensing method based on Cholesky decomposition sampling covariance matrix | |
| CN109379153B (en) | Spectrum sensing method | |
| CN111600667B (en) | CNN-LSTM-based spectrum sensing method | |
| CN109450573B (en) | frequency spectrum sensing method based on deep neural network | |
| CN109299698B (en) | Wireless channel scene recognition method based on support vector machine | |
| CN116866129A (en) | Wireless communication signal detection method | |
| CN112422213A (en) | Efficient spectrum sensing method based on support vector machine | |
| CN111817803A (en) | Frequency spectrum sensing method and system based on correlation coefficient and K-means clustering algorithm and computer readable storage medium | |
| CN116192307A (en) | Distributed cooperative multi-antenna cooperative spectrum intelligent sensing method, system, equipment and medium under non-Gaussian noise | |
| Tang et al. | Specific emitter identification for IoT devices based on deep residual shrinkage networks | |
| CN118041471B (en) | Spectrum sensing method and system based on machine learning logistic regression algorithm | |
| CN112350790B (en) | Deep learning-based spectrum sensing detection method, device and equipment | |
| Muzaffar et al. | A review of spectrum sensing in modern cognitive radio networks | |
| CN111934797B (en) | Collaborative spectrum sensing method based on covariance eigenvalue and mean shift clustering | |
| Sun et al. | A cooperative spectrum sensing algorithm based on principal component analysis and k-medoids clustering | |
| Gupta et al. | Cognitive radio spectrum classification using FLA-SVM | |
| CN114337883B (en) | CNN collaborative spectrum sensing method and system for covariance matrix Cholesky decomposition | |
| Cong et al. | Spectrum sensing using CNN with attention on switch of channel states | |
| Chen et al. | SVM and decision stumps based hybrid AdaBoost classification algorithm for cognitive radios | |
| Majumder | A variational Bayesian approach to multiantenna spectrum sensing under correlated noise | |
| Zhang et al. | CNN based wideband spectrum occupancy status identification for cognitive radios | |
| Shen et al. | Deep learning based source number estimation with single-channel mixtures | |
| CN116707682A (en) | Cooperative spectrum sensing method and system based on symmetric low-rank subspace clustering | |
| Zhao et al. | Multi-feature collaborative spectrum sensing method based on support vector machine | |
| CN114915526B (en) | Communication signal modulation identification method, device and system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |