Cooperative spectrum prediction method based on minimum Bayesian risk

Abstract

The invention discloses a cooperative spectrum prediction method based on minimum Bayesian risk, and relates to the technical field of cognitive radio and wireless communication. According to the method, each secondary user independently completes local spectrum sensing and power prediction; the fusion center performs system perception judgment by using a historical optimal fusion rule, calculates a channel state prior probability, and fits a predicted power condition probability density function of each user in two system states; then, according to a minimum Bayesian risk criterion, through alternative iteration optimization of each user judgment threshold and a system fusion rule, an optimal threshold and a fusion rule are obtained; and finally, obtaining a system-level cooperative spectrum prediction result through threshold comparison and fusion. The method does not depend on prior channel information, has good flexibility and expansibility when the number of users changes dynamically, improves prediction precision and robustness in a complex environment, and can balance main user protection and spectrum access efficiency through a cost factor.

Description

Technical Field

[0001] This invention relates to the field of cognitive radio technology, and more specifically to a cooperative spectrum prediction method. Background Technology

[0002] With the rapid development of mobile communication and information network technologies, the demand for wireless spectrum, as a fundamental resource supporting various services, continues to grow. In various wireless systems, including cellular base stations and satellite communications, the available licensed spectrum is becoming increasingly scarce. How to effectively improve the utilization efficiency of spectrum resources has become an important research direction in the field of wireless communication. Cognitive radio technology provides an effective way to solve this problem by sensing and dynamically utilizing idle spectrum. Among these technologies, spectrum prediction technology can predict the activity status of primary users (PUs) in future time slots based on historical spectrum occupancy information, thereby guiding secondary users (SUs) to opportunistically access the network. This is one of the key means to improve spectrum utilization.

[0003] Traditional collaborative spectrum prediction methods, such as those based on regression analysis or hidden Markov models, typically rely on strong assumptions about channel characteristics or spectrum occupancy statistics. These methods generally require each secondary user to perform predictions or calculate prediction statistics locally, then transmit the results to a fusion center for hard or soft decision fusion. When applied in complex real-world electromagnetic environments, the performance of such schemes is extremely sensitive to the accuracy of the pre-set model assumptions. If the actual environment deviates from the model assumptions, the prediction accuracy drops significantly, thus greatly limiting their practicality and robustness.

[0004] Neural network-based collaborative spectrum prediction methods alleviate the reliance on pre-defined models to some extent, and currently there are two main approaches. The first approach involves each node using a deep learning model for local prediction, followed by fusion at a fusion center. Essentially, this uses a neural network to generate local statistics instead of traditional methods, and then applies the traditional fusion algorithm. However, this approach often requires prior information such as the signal-to-noise ratio and relative distance of each node in the weight allocation stage of the fusion center, which is difficult to obtain stably and accurately in dynamically changing real-world scenarios. The second approach directly constructs a high-dimensional feature matrix from the observation data of multiple secondary users and inputs it into a unified neural network for end-to-end prediction. However, the network input dimension of this approach is directly tied to the number of participating secondary users. When the number of user nodes in the network changes, the dimension of the input features changes accordingly, usually requiring adjustments to the network structure or parameters, or even retraining. This results in poor flexibility and scalability, making it difficult to adapt to dynamically changing network topologies. Summary of the Invention

[0005] This invention proposes a cooperative spectrum prediction method based on minimum Bayesian risk. Its purpose is to eliminate the dependence on prior information (such as signal-to-noise ratio, relative distance, etc.) in complex wireless environments, while possessing good flexibility and scalability. It can achieve accurate spectrum prediction without adjusting the network structure or retraining the model when the number of secondary user nodes participating in the cooperation changes dynamically.

[0006] The technical solution of this invention is as follows:

[0007] A collaborative spectrum prediction method based on minimum Bayesian risk, for any _th_ element in the communication process... For each frame, spectrum prediction is performed according to steps S1 to S5:

[0008] Step S1, all in the collaborative spectrum prediction system Each secondary user independently performs spectrum sensing and power prediction, obtains local spectrum sensing results and predicted power values ​​for the next frame, and sends the obtained data to the fusion center through a dedicated control channel;

[0009] Step S2: The fusion center uses the spectrum sensing results of all secondary users in this frame and the optimal fusion rule obtained in the previous frame to complete the system-level spectrum sensing decision; then, using the system spectrum sensing results as prior guidance, it updates and maintains a sliding window set containing the system spectrum sensing results of recent multiple frames, as well as a set of predicted power values ​​for each secondary user, corresponding to the two system states of channel idle and occupied.

[0010] The fusion rule is a mapping relationship from a vector containing the binary perception or prediction results of all sub-users to the system-level binary decision result.

[0011] Step S3: Based on the set of predicted power values, calculate the prior probability of the channel state of the system in the next frame; at the same time, for each secondary user, use a Gaussian mixture model to fit the conditional probability density function of its predicted power under the two system state assumptions.

[0012] Step S4: Based on the minimum Bayes risk criterion, perform joint iterative optimization of the fusion rule and the user threshold to obtain the optimal threshold value for each secondary user to convert the predicted power value into a binary prediction result, and the optimal fusion rule for fusing the binary prediction results of all users.

[0013] Step S5: The fusion center compares the predicted power value of the next frame sent by each secondary user with the corresponding optimal threshold value obtained in step S4 to obtain the binary prediction result of each secondary user. Then, the vector composed of the binary prediction results of all secondary users is input into the optimal fusion rule obtained in step S4 to obtain the system-level cooperative spectrum prediction result of the current frame, and the system-level cooperative spectrum prediction result is sent to each secondary user.

[0014] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, the spectrum sensing process in step S1 is as follows:

[0015] For the For each sub-user, first count the number of sub-users in the [number]th [month / day]. The in-band power value of the frame is denoted as Then, solve the first... The second-level user in the first Adaptive thresholding of frames The calculation method is as follows:

[0016]

[0017] In the above formula, For the first The second-level user in the first Frame based on previous In-band power value sequence of frames The probability density function obtained through kernel density estimation; The probability density function The peak number; and They are respectively Time probability density function The minimum and maximum values ​​of the x-axis corresponding to all peaks. for trough values ​​within the interval;

[0018] After obtaining the threshold, the secondary user performs local spectrum sensing decision-making, assuming the first... The second-level user in the first The perceived result of the frame is The calculation formula is as follows:

[0019]

[0020] In the above formula, This indicates that the decision is due to channel occupancy. This indicates that the channel is idle.

[0021] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, the method for calculating the predicted power value of the next frame in step S1 is as follows: the secondary user uses an LSTM neural network for power prediction; the LSTM network previously... The historical power value of this secondary user in the frame. For input, output for the first... The predicted power value of the frame is denoted as ,Right now:

[0022]

[0023] In the above formula, This represents the forward computation process of an LSTM neural network. These are the parameters of the LSTM neural network.

[0024] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, the system-level spectrum sensing decision process in step S2 is as follows: the fusion center, based on the received... The spectrum sensing vector is composed of the spectrum sensing results of each secondary user. Combined with the first Optimal fusion rule obtained from frames Perform system-level spectrum sensing decisions to obtain system spectrum sensing results. ; For the first The local perception results obtained by each secondary user in step S1;

[0025] The method for updating and maintaining the sliding window set and the predicted power value set is as follows:

[0026] The system spectrum sensing results of the current frame Add to the system spectrum sensing results history collection In this context, the set uses a length of... The sliding window mechanism only saves the most recent... The system spectrum sensing results of the frame;

[0027] Based on system spectrum sensing results To guide this process, update the set of predicted power values ​​for each secondary user: for the If a secondary user Then the first The predicted power value of this frame obtained from the frame. Add to the set of idle predicted power values ​​corresponding to this secondary user Otherwise Add to the set of predicted power values ​​for this secondary user In the middle, the said and The sum of the number of elements equals the historical set of system spectrum sensing results. The number of elements; during the update, if the historical set of the current system spectrum sensing results is... The earliest element removed at that time Then synchronize from Remove the earliest element from the list; otherwise, start from... Remove the earliest element.

[0028] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, the specific process of step S3 is as follows:

[0029] First, randomly select one of the sets of predicted power values ​​corresponding to a secondary user, and calculate the prior probability of the channel state in the next frame of the system:

[0030]

[0031] Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; Indicates the number of elements in the set;

[0032] Secondly, for each secondary user, based on the set of predicted power values ​​maintained in step S2... and Fit the predicted power at respectively and The conditional probability density function under the following conditions; for the first... A secondary user, assuming ( When the condition is met, the predicted power is obtained by solving the Gaussian mixture model. The conditional probability density function is:

[0033]

[0034] In the above formula, let As The parameters of the Gaussian mixture model at the time of its establishment. It is the first The weights of the Gaussian components, and These are the mean and variance, respectively. This represents the probability density function of a Gaussian distribution. It is the number of Gaussian components;

[0035] parameter By maximizing the log-likelihood function To estimate:

[0036]

[0037] in It is a set The predicted power value in the middle, That is, set The number of elements; solved by the expectation-maximization algorithm. Maximize parameters That is, to obtain the fitted conditional probability density function. , .

[0038] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, step S4 specifically includes the following sub-steps:

[0039] Step S4-1: Initialize the threshold values ​​for each secondary user, and use the optimal fusion rule obtained in the previous frame as the initial fusion rule for this frame iteration; based on the current threshold values ​​and fusion rules, calculate the initial Bayesian risk of the system, and then let... ;

[0040] Step S4-2: Based on the current threshold values ​​of each secondary user, calculate and update the current threshold value according to the least Bayesian risk criterion. The rules for wheel fusion;

[0041] Step S4-3: According to the current fusion rules, calculate and update the individual threshold value of each secondary user in sequence, and use the calculated threshold value of each secondary user as the basis for calculating the threshold values ​​of other secondary users.

[0042] Step S4-4: Using the updated threshold values ​​and fusion rules for all secondary users, recalculate the current Bayesian risk of the system; if the difference between the current Bayesian risk and the previous Bayesian risk is less than a preset threshold, terminate the iteration and output the current threshold value and fusion rule as the optimal threshold value and optimal fusion rule for all secondary users in the current frame, respectively; otherwise, let... Then return to step S4-2 to continue the next iteration.

[0043] As a further improvement to the aforementioned collaborative spectrum prediction method based on minimum Bayesian risk: In step S4-1, let the first... The initial value for the threshold used by each secondary user is... The calculation formula is as follows:

[0044]

[0045] The above equation represents the conditional probability density function obtained in step S3 under the two system state assumptions corresponding to the secondary user. and The power value corresponding to the intersection point.

[0046] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk: In step S4, the Bayesian risk is calculated. The formula is:

[0047]

[0048] In the above formula, the variables are calculated. Calculate variables Calculate variables ; The cost factor represents the true state. The system predicts that The cost at that time is a preset value; Indicates the system spectrum sensing result for the next frame. , The result obtained in step S2 The prior probability of its validity; Indicates the system spectrum sensing result for the next frame. , The result obtained in step S2 The prior probability of its validity; For the system's collaborative prediction outcome variables, For the reason A vector consisting of binary variables of each secondary user. With the Each secondary user corresponds to one secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; Representing vectors For enumeration vectors At that time, the system collaborative prediction outcome variables The probability of being 1 is equal to the probability of the enumeration vector. The corresponding mapping result in the current fusion rule; In order to be in At the time of its establishment, the first The prediction results for each secondary user are as follows: The probability, Representing vectors The One element;

[0049] The calculation method is as follows:

[0050]

[0051] In the above formula, The first one obtained in step S3 Sub-users in both system states and The conditional probability density function under the assumptions, For the current number Threshold values ​​for each secondary user.

[0052] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, in step S4-2, the current number of... The way to merge the wheels is to iterate through all possible... Let the binary result combination of the secondary user be denoted as . The corresponding system binary decision result is:

[0053]

[0054] in, Indicates the first The fusion rules obtained in the next iteration , express At the time of establishment, all secondary users occupy the predicted result combination. The probability of occurrence is calculated using the following formula:

[0055]

[0056] In step S4-3, calculate the first... Threshold value for each sub-user The formula is as follows:

[0057]

[0058] In the above formula, To exclude the first Beyond individual users A vector consisting of binary variables of each secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; ,in for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule. for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule; express At the time of its establishment, the system, except for the first Beyond individual users Combination of prediction results for each secondary user The probability of occurrence is calculated using the following formula:

[0059]

[0060] In the formula, For vectors The One element, The calculation method is as follows:

[0061]

[0062] In the above formula, The first one obtained in step S3 Two conditional probability density functions for each secondary user. For the current number Threshold values ​​for each secondary user.

[0063] As a further improvement to the cooperative spectrum prediction method based on minimum Bayesian risk, in step S5, the first... The binary prediction results for each secondary user are obtained as follows:

[0064]

[0065] In the above formula, The first obtained in step S1 The predicted power value for the next frame for each secondary user; The first one obtained in step S4 The optimal threshold value for each secondary user;

[0066] The vector representation of the binary prediction results for all secondary users is as follows: The vector is input into the optimal fusion rule obtained in step S4 of the current frame. The system-level collaborative spectrum prediction results for the current frame are obtained:

[0067]

[0068] Indicates the system predicts the first The frame spectrum is idle, and secondary users are allowed to communicate in this frame; Indicates the system predicts the first The frame spectrum is occupied, and secondary users should avoid accessing it and wait for a later opportunity.

[0069] Compared with the prior art, the present invention has the following beneficial effects:

[0070] 1. This invention utilizes system spectrum sensing results as prior guidance at the fusion center to update and maintain a sliding window set containing recent multi-frame system spectrum sensing results and a set of predicted power values ​​for each secondary user. Then, based on the minimum Bayesian risk criterion, it jointly iteratively optimizes the fusion rules and user thresholds. Thus, in complex wireless environments, without relying on prior information such as the signal-to-noise ratio and relative distance of each secondary user, it can dynamically and adaptively obtain the optimal threshold value for each secondary user and the optimal fusion rules at the system level, ultimately yielding collaborative spectrum prediction results. This effectively overcomes the dependence of traditional collaborative spectrum prediction methods on strong assumptions or prior information, improving the practicality and robustness of the prediction method in real-world dynamic environments.

[0071] 2. In this invention, secondary users locally employ adaptive thresholds for spectrum sensing and utilize LSTM neural networks for power prediction. This approach does not rely on specific strong assumptions about the channel or data, enhancing the flexibility and environmental adaptability of local processing. Simultaneously, the fusion center fits the conditional probability density function of the predicted power of each secondary user under two system state assumptions using a Gaussian mixture model. This quantitatively and accurately characterizes the differences in prediction performance among secondary users, providing a reliable statistical basis for subsequent risk-based optimization.

[0072] 3. The fusion rule and user threshold joint iterative optimization mechanism based on minimum Bayesian risk proposed in this invention takes the overall Bayesian risk of the system as the optimization objective. By alternately fixing variables and sequentially optimizing the fusion rule and each user threshold, the system risk gradually converges to a local optimum. This process not only avoids the one-sidedness of traditional accuracy-based optimization, but also allows users to flexibly adjust the trade-off between primary user protection and spectrum access efficiency according to actual needs by introducing a configurable cost factor, thus achieving more comprehensive and controllable system performance optimization.

[0073] 4. In the collaborative prediction framework constructed in this invention, secondary users only need to send local binary sensing results and continuous power prediction values ​​to the fusion center. The optimization algorithm of the fusion center is entirely based on these data and historically maintained sets, and its input dimension is decoupled from the number of secondary users. When the number of secondary user nodes participating in the collaboration changes dynamically, only the dimension of the fusion rule mapping table and the number of users in the iterative calculation need to be adjusted accordingly. There is no need to change the neural network structure or retrain the model, which significantly improves the scalability and deployment flexibility of the system and can better adapt to dynamically changing network topologies. Attached Figure Description

[0074] Figure 1 This is a frame structure diagram that combines spectrum sensing and power prediction capabilities, used in the cooperative spectrum prediction method based on minimum Bayesian risk.

[0075] Figure 2 The flowchart shows the fusion algorithm based on minimum Bayesian risk for frame t.

[0076] Figure 3 The diagram shows the conditional probability density function and threshold variation of the three secondary users (SU) in the specific implementation.

[0077] Figure 4 This is a graph showing how the Bayesian risk of the system changes with iteration. Detailed Implementation

[0078] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0079] In this embodiment, a cooperative spectrum prediction system with N=3 secondary users (SUs) is considered, where all secondary users share a single channel. A simulation dataset is generated using the LTE-M uplink physical layer and queuing theory model. The system bandwidth is 10MHz, and the sampling rate is 15.36MHz. The simulation system is divided into 16 mutually orthogonal narrowband channels, and one channel is selected for modeling. Gaussian white noise of -11dB (secondary user 1), -6dB (secondary user 2), and -1dB (secondary user 3) is added to the signal to simulate the cooperative spectrum prediction process of the three nodes under different signal-to-noise ratio environments. A total of 12,000 consecutive time slots are simulated, with 2048 IQ samples per time slot.

[0080] For any _th _ in the communication process For each frame, spectral prediction is performed according to steps S1 to S5. This embodiment takes the prediction process of frame 500 to frame 501 as an example, combined with... Figure 1 The frame structure shown and Figure 2 The first one shown The flowchart of the fusion algorithm based on minimum Bayesian risk is shown in detail, illustrating the specific implementation steps of the present invention.

[0081] Step S1: Multiple secondary users in the collaborative spectrum prediction system independently perform spectrum sensing and power prediction, obtain local spectrum sensing results and predicted power values ​​for the next frame, and send the obtained data to the fusion center (FC) through a dedicated control channel.

[0082] This process does not rely on any specific strong assumptions about the channel or the data.

[0083] In this embodiment, the frame structure of each secondary user is as follows: Figure 1 As shown, it includes two parts: spectrum sensing and power prediction.

[0084] (1) The process of spectrum sensing is as follows:

[0085] For the For each sub-user, first count the number of sub-users in the [number]th [month / day]. Frame (in this embodiment) The in-band power value of ) is denoted as Then, solve for the... The second-level user in the first Adaptive thresholding of frames The calculation method is as follows:

[0086]

[0087] In the above formula, For the first The second-level user in the first Frame based on previous In-band power value sequence of frames The probability density function obtained through kernel density estimation; The probability density function The peak number; and They are respectively Time probability density function The minimum and maximum values ​​of the x-axis corresponding to all peaks. for The trough value within the interval.

[0088] When the signal-to-noise ratio is high, this probability density function has at least two peaks, where Corresponding noise peak, For the corresponding signal peak, the coordinates of the trough between the two peaks are taken as the threshold. If the function has only one peak, then the previous value is used. The average frame power value is used as the threshold.

[0089] In this embodiment, the sliding window length is set. .

[0090] After obtaining the threshold, the secondary user performs a local spectrum awareness decision. The second-level user in the first The perceived result of the frame is The calculation formula is:

[0091]

[0092] In the above formula, This indicates that the decision is due to channel occupancy. This indicates that the channel is idle.

[0093] (2) The method for calculating the predicted power value of the next frame is as follows:

[0094] Secondary users utilize LSTM neural networks for power prediction. LSTM networks were previously... Power values ​​within the history band of a frame For input, output for the first... The predicted power value of the frame is denoted as ,Right now:

[0095]

[0096] In the above formula, This represents the forward computation process of an LSTM neural network. These are the parameters of the LSTM neural network. The LSTM network learns the pattern of channel power changes through historical data to make predictions. Its structure and calculation process are existing technologies and will not be described in detail here.

[0097] After each secondary user independently completes the above operations, the local perception results will be... and predicted power value It is sent to the fusion center via a dedicated control channel.

[0098] In this embodiment, the calculation results of the three secondary users are shown in Table 1.

[0099] Table 1: Local calculation results for secondary users.

[0100]

[0101] Step S2: The fusion center uses the spectrum sensing results of all secondary users received in this frame and the optimal fusion rule obtained in the previous frame to complete the system-level spectrum sensing decision; then, using the system spectrum sensing results as prior guidance, it updates and maintains a sliding window set containing the system spectrum sensing results of recent multiple frames, as well as a set of predicted power values ​​for each secondary user, corresponding to the two system states of channel idle and occupied.

[0102] (1) The system-level spectrum sensing decision process is as follows:

[0103] The fusion center receives the data. The spectrum sensing vector is composed of the spectrum sensing results of each secondary user. Combined with the first Optimal fusion rule obtained from frames Perform system-level spectrum sensing decisions to obtain system spectrum sensing results. .

[0104] The aforementioned fusion rule is a vector containing the binary perception or prediction results of all sub-users. (in This relates the mapping relationship from the initial decision to the system-level binary decision result. The fusion rule can take the form of a logical function, decision tree, lookup table, etc. In this embodiment, the fusion rule is stored and executed in the form of a mapping table.

[0105] In this embodiment, This indicates that the system has determined that the channel is idle for the current frame.

[0106] (2) The method for updating and maintaining the sliding window set and the predicted power value set is as follows:

[0107] The system spectrum sensing results of the current frame Add to the system spectrum sensing results history collection In this context, the set uses a length of... The sliding window mechanism, which only saves the most recent... The system spectrum sensing results of the frame.

[0108] Based on system spectrum sensing results To guide this process, update the set of predicted power values ​​for each secondary user: for the If a secondary user Then the first The predicted power value of this frame obtained from the frame. Add to the set of idle predicted power values ​​corresponding to this secondary user Otherwise Add to the set of predicted power values ​​for this secondary user In the middle, the said and The sum of the number of elements equals the historical set of system spectrum sensing results. The number of elements. During the update, if the historical set of the current system spectrum sensing results... The earliest element removed at that time Then synchronize from Remove the earliest element from the list; otherwise, start from... Remove the earliest element.

[0109] In this way, the system uses the perception results of the current frame as more reliable prior information to classify and organize historical prediction data, providing a basis for subsequent probability calculations. Compared with directly using spectrum prediction results, spectrum perception results have higher mutual information with the actual channel occupancy status. Using this as a priori information can increase the amount of information in spectrum prediction, thereby improving prediction performance.

[0110] Step S3: Based on the set of predicted power values, calculate the prior probability of the channel state of the system in the next frame; at the same time, for each secondary user, use a Gaussian mixture model to fit the conditional probability density function of its predicted power under the two system state assumptions, so as to quantitatively characterize the prediction performance of the secondary user.

[0111] First, randomly select one of the sets of predicted power values ​​corresponding to a secondary user, and calculate the prior probability of the channel state in the next frame of the system:

[0112]

[0113] Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; This indicates the number of elements in the set.

[0114] In this embodiment, the calculation is obtained , .

[0115] Secondly, for each secondary user, based on the set of predicted power values ​​maintained in step S2... and Fit the predicted power at respectively and The conditional probability density function under the given condition. For the th A secondary user, assuming ( When the condition is met, the predicted power is obtained by solving the Gaussian mixture model. The conditional probability density function is:

[0116]

[0117] In the above formula, let As The parameters of the Gaussian mixture model at the time of its establishment. It is the first The weights of the Gaussian components, and These are the mean and variance, respectively. This represents the probability density function of a Gaussian distribution. This refers to the number of Gaussian components. To balance computational complexity and fitting performance, this embodiment uses [the specified value]. .

[0118] parameter By maximizing the log-likelihood function To estimate:

[0119]

[0120] in It is a set The predicted power value in the middle, That is, set The number of elements. Solving this using the expectation-maximization algorithm. Maximize parameters That is, to obtain the fitted conditional probability density function. , .

[0121] The conditional probability density function reflects the predictive performance of secondary users: the better the performance of a secondary user, the higher its predictive power. and The less overlap there is between the distributions under the two assumptions, the higher the discriminative power of the two conditional probability density functions.

[0122] In this embodiment, the conditional probability density function fitted by the three secondary users is as follows: Figure 3 As shown.

[0123] Step S4: Based on the minimum Bayesian risk criterion, perform joint iterative optimization of the fusion rule and the user threshold to obtain the optimal threshold value for each secondary user to convert the predicted power value into a binary prediction result, and the optimal fusion rule for fusing the binary prediction results of all users.

[0124] The specific process for this step is as follows: Figure 2 As shown, it includes the following sub-steps:

[0125] Step S4-1: Initialize the threshold values ​​for each secondary user, and use the optimal fusion rule obtained in the previous frame as the initial fusion rule for this frame iteration; based on the current threshold values ​​and fusion rules, calculate the initial Bayesian risk of the system, and then let... .

[0126] Let the first The initial value for the threshold used by each secondary user is... The calculation formula is as follows:

[0127]

[0128] The above formula represents the power value corresponding to the intersection of the two conditional probability density functions obtained in step S3.

[0129] In this embodiment, the initialization result is as follows: , , .

[0130] Calculate Bayesian risk The formula is:

[0131]

[0132] In the above formula, the variables are calculated. Calculate variables , ; The cost factor represents the true state. The system predicts that The cost of accurate prediction is usually zero. Users can configure according to their needs. and If priority needs to be given to protecting the primary user, then set If it is necessary to improve spectrum access efficiency, then set . For the system's collaborative prediction outcome variables, For the reason A vector consisting of binary variables of each secondary user. With the Each secondary user corresponds to one secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; Representing vectors For enumeration vectors At that time, the system collaborative prediction outcome variables The probability of being 1 is equal to the probability of the enumeration vector. The corresponding mapping result in the current fusion rule; In order to be in At the time of its establishment, the first The prediction results for each secondary user are as follows: The probability, Representing vectors The Each element.

[0133] The calculation method is as follows:

[0134]

[0135] In the above formula, The first one obtained in step S3 Two conditional probability density functions for each secondary user. For the current number Threshold values ​​for each secondary user.

[0136] In this embodiment, the initial Bayesian risk .

[0137] Step S4-2: Based on the current threshold values ​​of each secondary user, calculate and update the current threshold value according to the least Bayesian risk criterion. The rules for wheel fusion.

[0138] Calculate the current number The way to merge the wheels is to iterate through all possible... Let the binary result combination of the secondary user be denoted as . The corresponding system binary decision result is:

[0139]

[0140] in, express At the time of establishment, all secondary users occupy the predicted result combination. The probability of occurrence is calculated using the following formula:

[0141]

[0142] Step S4-3: According to the current fusion rules, calculate and update the individual threshold value of each secondary user in sequence, and use the calculated threshold value of each secondary user as the basis for calculating the threshold values ​​of other secondary users.

[0143] Let the current calculation be the first... Threshold value for each sub-user The calculation formula is as follows:

[0144]

[0145] In the above formula, To exclude the first Beyond individual users A vector consisting of binary variables of each secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; ,in for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule. for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule; express At the time of its establishment, the system, except for the first Beyond individual users Combination of prediction results for each secondary user The probability of occurrence is calculated using the following formula:

[0146]

[0147] In the formula, For vectors The One element, The calculation method is as follows:

[0148]

[0149] In the above formula, The first one obtained in step S3 Two conditional probability density functions for each secondary user. For the current number Threshold values ​​for each secondary user.

[0150] It should be noted that each time a threshold value is calculated, all threshold values, including the latest calculated threshold value, are used.

[0151] Step S4-4: Using the updated threshold values ​​and fusion rules for all secondary users, recalculate the current Bayesian risk of the system; if the difference between the current Bayesian risk and the previous Bayesian risk is less than a preset threshold, terminate the iteration and output the current threshold value and fusion rule as the optimal threshold value and optimal fusion rule for all secondary users in the current frame, respectively; otherwise, let... Then return to step S4-2 to continue the next iteration.

[0152] Set a small positive number As the iteration termination threshold, this embodiment takes Assess the amount of risk reduction. Is it less than If less than If the Bayesian risk has converged, the iteration terminates.

[0153] In this embodiment, after 3 iterations, the Bayesian risk converges to... The termination condition is met. The changes in the secondary user threshold values ​​during the iteration process are as follows: Figure 3 As shown, the changes in the Bayesian risk of the system are as follows: Figure 4 As shown in Table 2, the specific numerical changes and fusion rules are as follows.

[0154] Table 2: Threshold values ​​and fusion rule data for each secondary user during the optimization process.

[0155]

[0156] Step S5: The fusion center compares the predicted power value of the next frame sent by each secondary user with the corresponding optimal threshold value obtained in step S4 to obtain the binary prediction result of each secondary user. Then, the vector composed of the binary prediction results of all secondary users is input into the optimal fusion rule obtained in step S4 to obtain the system-level cooperative spectrum prediction result of the current frame, and the system-level cooperative spectrum prediction result is sent to each secondary user.

[0157] Calculate the first The binary prediction results for each secondary user are obtained as follows:

[0158]

[0159] In the above formula, The first obtained in step S1 The predicted power value for the next frame for each secondary user; The first one obtained in step S4 The optimal threshold value for each secondary user.

[0160] The vector representation of the binary prediction results for all secondary users is as follows: The vector is input into the optimal fusion rule obtained in step S4 of the current frame. The system-level collaborative spectrum prediction results for the current frame are obtained:

[0161]

[0162] Indicates the system predicts the first The frame spectrum is idle, and secondary users are allowed to communicate in this frame; Indicates the system predicts the first The frame spectrum is occupied, and secondary users should avoid accessing it and wait for a later opportunity.

[0163] Through this collaborative prediction mechanism, the system can more accurately perceive and predict spectrum status, thereby effectively improving the utilization rate of spectrum resources while protecting primary user communications.

[0164] It should be noted that, as will be apparent to those skilled in the art, the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics thereof. The scope of the present invention is defined by the claims rather than the foregoing description.

Claims

1. A collaborative spectrum prediction method based on minimum Bayesian risk, characterized in that, For any _th _ in the communication process For each frame, spectrum prediction is performed according to steps S1 to S5: Step S1, all in the collaborative spectrum prediction system Each secondary user independently performs spectrum sensing and power prediction, obtains local spectrum sensing results and predicted power values ​​for the next frame, and sends the obtained data to the fusion center through a dedicated control channel; Step S2: The fusion center uses the spectrum sensing results of all secondary users in this frame and the optimal fusion rule obtained in the previous frame to complete the system-level spectrum sensing decision; then, using the system spectrum sensing results as prior guidance, it updates and maintains a sliding window set containing the system spectrum sensing results of recent multiple frames, as well as a set of predicted power values ​​for each secondary user, corresponding to the two system states of channel idle and occupied. The fusion rule is a mapping relationship from a vector containing the binary perception or prediction results of all sub-users to the system-level binary decision result. Step S3: Based on the set of predicted power values, calculate the prior probability of the channel state of the system in the next frame; at the same time, for each secondary user, use a Gaussian mixture model to fit the conditional probability density function of its predicted power under the two system state assumptions. Step S4: Based on the minimum Bayes risk criterion, perform joint iterative optimization of the fusion rule and the user threshold to obtain the optimal threshold value for each secondary user to convert the predicted power value into a binary prediction result, and the optimal fusion rule for fusing the binary prediction results of all users. Step S5: The fusion center compares the predicted power value of the next frame sent by each secondary user with the corresponding optimal threshold value obtained in step S4 to obtain the binary prediction result of each secondary user. Then, the vector composed of the binary prediction results of all secondary users is input into the optimal fusion rule obtained in step S4 to obtain the system-level cooperative spectrum prediction result of the current frame, and the system-level cooperative spectrum prediction result is sent to each secondary user.

2. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 1, characterized in that, The spectrum sensing process in step S1 is as follows: For the For each sub-user, first count the number of sub-users in the [number]th [month / day]. The in-band power value of the frame is denoted as Then, solve the first... The second-level user in the first Adaptive thresholding of frames The calculation method is as follows: ； In the above formula, For the first The second-level user in the first Frame based on previous In-band power value sequence of frames The probability density function obtained through kernel density estimation; The probability density function The peak number; and They are respectively Time probability density function The minimum and maximum values ​​of the x-axis corresponding to all peaks. for trough values ​​within the interval; After obtaining the threshold, the secondary user performs local spectrum sensing decision-making, assuming the first... The second-level user in the first The perceived result of the frame is The calculation formula is as follows: ； In the above formula, This indicates that the decision is due to channel occupancy. This indicates that the channel is idle.

3. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 1, characterized in that, The method for calculating the predicted power value of the next frame in step S1 is as follows: the secondary user uses an LSTM neural network for power prediction; the LSTM network previously... The historical power value of this secondary user in the frame. For input, output for the first... The predicted power value of the frame is denoted as ,Right now: ； In the above formula, This represents the forward computation process of an LSTM neural network. These are the parameters of the LSTM neural network.

4. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 1, characterized in that, The system-level spectrum sensing decision process in step S2 is as follows: the fusion center determines the spectrum based on the received data... The spectrum sensing vector is composed of the spectrum sensing results of each secondary user. Combined with the first Optimal fusion rule obtained from frames Perform system-level spectrum sensing decisions to obtain system spectrum sensing results. ; For the first The local perception results obtained by each secondary user in step S1; The method for updating and maintaining the sliding window set and the predicted power value set is as follows: The system spectrum sensing results of the current frame Add to the system spectrum sensing results history collection In this context, the set uses a length of... The sliding window mechanism only saves the most recent... The system spectrum sensing results of the frame; Based on system spectrum sensing results To guide this process, update the set of predicted power values ​​for each secondary user: for the If a secondary user, Then the first The predicted power value of this frame obtained from the frame. Add to the set of idle predicted power values ​​corresponding to this secondary user Otherwise Add to the set of predicted power values ​​for this secondary user In the middle, the said and The sum of the number of elements equals the historical set of system spectrum sensing results. The number of elements; during the update, if the historical set of the current system spectrum sensing results is... The earliest element removed at that time Then synchronize from Remove the earliest element from the list; otherwise, start from... Remove the earliest element.

5. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 4, characterized in that, The specific process of step S3 is as follows: First, randomly select one of the sets of predicted power values ​​corresponding to a secondary user, and calculate the prior probability of the channel state in the next frame of the system: ； Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; Indicates the system spectrum sensing result for the next frame. , for The prior probability of its validity; Indicates the number of elements in the set; Secondly, for each secondary user, based on the set of predicted power values ​​maintained in step S2... and Fit the predicted power at respectively and The conditional probability density function under the following conditions; for the first... A secondary user, assuming ( When the condition is met, the predicted power is obtained by solving the Gaussian mixture model. The conditional probability density function is: ； In the above formula, let As The parameters of the Gaussian mixture model at the time of its establishment. It is the first The weights of the Gaussian components, and These are the mean and variance, respectively. This represents the probability density function of a Gaussian distribution. It is the number of Gaussian components; parameter By maximizing the log-likelihood function To estimate: ； in It is a set The predicted power value in the middle, That is, set The number of elements; solved by the expectation-maximization algorithm. Maximize parameters That is, to obtain the fitted conditional probability density function. , .

6. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 1, characterized in that, Step S4 specifically includes the following sub-steps: Step S4-1: Initialize the threshold values ​​for each secondary user, and use the optimal fusion rule obtained in the previous frame as the initial fusion rule for this frame iteration; based on the current threshold values ​​and fusion rules, calculate the initial Bayesian risk of the system, and then let... ; Step S4-2: Based on the current threshold values ​​of each secondary user, calculate and update the current threshold value according to the least Bayesian risk criterion. The rules for wheel fusion; Step S4-3: According to the current fusion rules, calculate and update the individual threshold value of each secondary user in sequence, and use the calculated threshold value of each secondary user as the basis for calculating the threshold values ​​of other secondary users. Step S4-4: Using the updated threshold values ​​and fusion rules for all secondary users, recalculate the current Bayesian risk of the system; if the difference between the current Bayesian risk and the previous Bayesian risk is less than a preset threshold, terminate the iteration and output the current threshold value and fusion rule as the optimal threshold value and optimal fusion rule for all secondary users in the current frame, respectively; otherwise, let... Then return to step S4-2 to continue the next iteration.

7. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 6, characterized in that: In step S4-1, let the first... The initial value for the threshold used by each secondary user is... The calculation formula is as follows: ； The above equation represents the conditional probability density function obtained in step S3 under the two system state assumptions corresponding to the secondary user. and The power value corresponding to the intersection point.

8. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 6, characterized in that: In step S4, the Bayesian risk is calculated. The formula is: ； In the above formula, the variables are calculated. Calculate variables Calculate variables ; The cost factor represents the true state. The system predicts that The cost at that time is a preset value; Indicates the system spectrum sensing result for the next frame. , The result obtained in step S2 The prior probability of its validity; Indicates the system spectrum sensing result for the next frame. , The result obtained in step S2 The prior probability of its validity; For the system's collaborative prediction outcome variables, For the reason A vector consisting of binary variables of each secondary user. With the Each secondary user corresponds to one secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; Representing vectors For enumeration vectors At that time, the system collaborative prediction outcome variables The probability of being 1 is equal to the probability of the enumeration vector. The corresponding mapping result in the current fusion rule; In order to be in At the time of its establishment, the first The prediction results for each secondary user are as follows: The probability, Representing vectors The One element; The calculation method is as follows: ； In the above formula, The first one obtained in step S3 Sub-users in both system states and The conditional probability density function under the assumptions, For the current number Threshold values ​​for each secondary user.

9. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 8, characterized in that, In step S4-2, calculate the current number of... The way to merge the wheels is to iterate through all possible... Let the binary result combination of the secondary user be denoted as . The corresponding system binary decision result is: ； in, Indicates the first The fusion rules obtained in the next iteration , express At the time of establishment, all secondary users occupy the predicted result combination. The probability of occurrence is calculated using the following formula: ； In step S4-3, calculate the first... Threshold value for each sub-user The formula is as follows: ； In the above formula, To exclude the first Beyond individual users A vector consisting of binary variables of each secondary user; For containing An enumeration vector of n elements, where each element has two values: 0 and 1, representing... All possible values; ,in for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule. for and Composed of sorting by secondary users The mapping result of the 3D binary vector in the current fusion rule; express At the time of its establishment, the system, except for the first Beyond individual users Combination of prediction results for each secondary user The probability of occurrence is calculated using the following formula: ； In the formula, For vectors The One element, The calculation method is as follows: ； In the above formula, The first one obtained in step S3 Two conditional probability density functions for each secondary user. For the current number Threshold values ​​for each secondary user.

10. The collaborative spectrum prediction method based on minimum Bayesian risk as described in claim 1, characterized in that, In step S5, calculate the first... The binary prediction results for each secondary user are obtained as follows: ； In the above formula, The first obtained in step S1 The predicted power value for the next frame for each secondary user; The first one obtained in step S4 The optimal threshold value for each secondary user; The vector representation of the binary prediction results for all secondary users is as follows: The vector is input into the optimal fusion rule obtained in step S4 of the current frame. The system-level collaborative spectrum prediction results for the current frame are obtained: ； Indicates the system predicts the first The frame spectrum is idle, and secondary users are allowed to communicate in this frame; Indicates the system predicts the first The frame spectrum is occupied, and secondary users should avoid accessing it and wait for a later opportunity.

Images