CN113014340A - Satellite spectrum resource dynamic allocation method based on neural network - Google Patents

Satellite spectrum resource dynamic allocation method based on neural network Download PDF

Info

Publication number
CN113014340A
CN113014340A CN202110197503.2A CN202110197503A CN113014340A CN 113014340 A CN113014340 A CN 113014340A CN 202110197503 A CN202110197503 A CN 202110197503A CN 113014340 A CN113014340 A CN 113014340A
Authority
CN
China
Prior art keywords
spectrum
channel
neural network
prediction
authorized user
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110197503.2A
Other languages
Chinese (zh)
Inventor
丁晓进
冯李杰
张更新
吴尘
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202110197503.2A priority Critical patent/CN113014340A/en
Publication of CN113014340A publication Critical patent/CN113014340A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B17/00—Monitoring; Testing
    • H04B17/30—Monitoring; Testing of propagation channels
    • H04B17/382—Monitoring; Testing of propagation channels for resource allocation, admission control or handover
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04B—TRANSMISSION
    • H04B7/00—Radio transmission systems, i.e. using radiation field
    • H04B7/14—Relay systems
    • H04B7/15—Active relay systems
    • H04B7/185—Space-based or airborne stations; Stations for satellite systems
    • H04B7/18578—Satellite systems for providing broadband data service to individual earth stations
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/02—Resource partitioning among network components, e.g. reuse partitioning
    • H04W16/10—Dynamic resource partitioning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/14—Spectrum sharing arrangements between different networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H04W72/0453—Resources in frequency domain, e.g. a carrier in FDMA
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/53—Allocation or scheduling criteria for wireless resources based on regulatory allocation policies
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/50—Allocation or scheduling criteria for wireless resources
    • H04W72/54—Allocation or scheduling criteria for wireless resources based on quality criteria
    • H04W72/541—Allocation or scheduling criteria for wireless resources based on quality criteria using the level of interference

Abstract

The invention discloses a satellite frequency spectrum resource dynamic allocation method based on a neural network, which comprises three parts of broadband signal data preprocessing, prediction model establishment and spectrum sharing; the broadband signal data preprocessing comprises the steps of carrying out probability density estimation on authorized user signals, setting corresponding threshold values for all authorized user signals, and carrying out quantization and division on input and output data sets on historical data according to the threshold values; the establishment of the prediction model is to respectively establish a prediction model and a fusion model based on a deep neural network, carry out multi-channel prediction on signals of each authorized user, input the prediction result of each channel into the fusion model, and output the future spectrum occupation state of the user in a fusion manner; the spectrum sharing is to output an allocation strategy according to the predicted future occupation state of the user. The method can carry out data preprocessing on the multi-authorization user signals on the frequency band and accurately predict the future spectrum occupation state of the multi-authorization user signals, and realizes dynamic allocation of spectrum resources, thereby achieving the aim of efficient utilization of the spectrum resources.

Description

Satellite spectrum resource dynamic allocation method based on neural network
Technical Field
The invention relates to a dynamic satellite spectrum resource allocation method based on a neural network, and belongs to the cognitive radio communication technology.
Background
With the rapid development of satellite communication and the continuous update of 5G technology, the number of end users has rapidly increased in the last years, and this trend will continue to continue in the future, with the continuous increase of communication traffic in wireless networks, and in order to meet the demand of wireless communication, the capacity of the network should be increased, so allocating spectrum resources becomes an effective method for increasing communication capacity. However, on the one hand, the irreproducibility and scarcity of spectrum resources are difficult to meet the requirement that each user has its own communication band. On the other hand, according to the conventional spectrum management strategy, a large part of the licensed spectrum is not fully utilized. Therefore, effectively improving the utilization rate of spectrum resources is a research hotspot of wireless communication nowadays.
The sharing of the frequency spectrum resources is the most effective method, under the condition that authorized users of the frequency spectrum allow, the unauthorized users can access the idle frequency spectrum resources opportunistically by dynamically sensing the frequency spectrum using state of the authorized users to achieve the aim of improving the overall utilization rate of the frequency spectrum, and the method can realize the full utilization of the frequency spectrum resources on the time frequency domain. It is worth mentioning that when the spectrum is shared, the unauthorized user does not generate any interference to the normal communication of the authorized user, and therefore, the spectrum monitoring and identifying signal bandwidth and the parameters thereof are the basis for the dynamic spectrum access of the unauthorized user. After the signal information is extracted, the use condition of the frequency spectrum is sensed, then the frequency spectrum hole can be mined and predicted, and finally, according to the communication requirement of an unauthorized user, the frequency spectrum is allocated according to the predicted frequency spectrum hole condition, so that the dynamic allocation of frequency spectrum resources is realized.
Disclosure of Invention
The purpose of the invention is as follows: in order to overcome the defects in the prior art, the invention provides a dynamic satellite spectrum resource allocation method based on a neural network, which comprises the steps of firstly identifying the signal bandwidth and related parameters of authorized users on a frequency band by a signal filtering and density clustering method, and achieving the purpose of monitoring the use condition of the frequency band; then, probability distribution estimation is carried out on each authorized user signal, a quantized threshold value is determined in a self-adaptive mode, and the signal is quantized into an idle state and an occupied state; and finally, inputting the result of multi-channel prediction of the authorized user into a trained prediction model to output, fusing the multi-channel prediction result by the fusion model to output the future spectrum use state of the user, and dynamically allocating spectrum resources by the spectrum allocation model according to the communication requirements of the unauthorized user on the basis of acquiring the future spectrum occupation state in advance, thereby improving the utilization rate of the spectrum resources.
The technical scheme is as follows: in order to achieve the purpose, the invention adopts the technical scheme that:
a satellite frequency spectrum resource dynamic allocation method based on a neural network comprises three parts of broadband signal data preprocessing, prediction model establishment and spectrum sharing, and specifically comprises the following steps:
(1) data preprocessing: firstly, carrying out probability distribution estimation on power spectral density time sequence values of all channels of an authorized user, then setting a self-adaptive quantization threshold for all channels of the authorized user, and finally dividing signals of the authorized user into a long-term and short-term memory and convolutional neural network prediction model training set and a fusion model training set according to a set length;
(2) establishing a hybrid prediction model: the hybrid prediction model comprises a long-term memory and short-term memory convolutional neural network prediction model and a fusion model; the long-short term memory and convolutional neural network prediction model is divided into a first half part and a second half part, the first half part comprises a layer of long-short term memory network layer, the second half part comprises a layer of convolutional neural network layer and a layer of full connection layer, the long-short term memory network layer is used for learning the time domain correlation of the authorized user signals, the convolutional neural network layer is used for learning the inter-channel correlation of the authorized user signals, and the full connection layer outputs a spectrum prediction result; the fusion model comprises three full-connection layers, and is used for fusing parameters learned by the long-term and short-term memory network layer, parameters learned by the convolutional neural network layer and spectrum prediction results to obtain the future spectrum occupation state of the authorized user, namely fused spectrum data;
(3) spectrum sharing: and inputting the power spectrum density time sequence value of the signal of the authorized user into a hybrid prediction model, predicting the future spectrum occupation state of the authorized user, selecting an appropriate idle channel of the authorized user through a distribution network for shared communication according to the prediction result and the required bandwidth of the unauthorized user, and outputting a spectrum distribution strategy.
Specifically, in the step (1), the data preprocessing includes the following steps:
(1.1) carrying out probability distribution estimation on each channel power spectral density time sequence value of an authorized user, and carrying out probability distribution estimation on a channel i, wherein the probability distribution estimation comprises the following steps:
Figure BDA0002946295840000021
wherein: x is a random value of the power spectral density of the channel i at time t, xi,tFor the power spectral density value of channel i at time t,
Figure BDA0002946295840000022
the probability distribution estimation value of the channel i is obtained, h is a probability distribution estimation parameter, T is the length of a power spectral density time sequence value, and K (·) is a Gaussian kernel function;
(1.2) mixing
Figure BDA0002946295840000023
Is taken as the noise power mean value N of the channel iiExpressed as:
Figure BDA0002946295840000024
wherein: x is the number of0Is composed of
Figure BDA0002946295840000025
The value of x when true;
(1.3) calculating an adaptive quantization threshold gamma of the channel ii＝Ni+ ω, ω is a preset positive offset;
(1.4) time-sequence value { x) of power spectral density of channel ii,1,xi,2,…,xi,TAt gammaiQuantizing to a channel occupancy state timing value { omega ] for a thresholdi,1,Ωi,2,…,Ωi,T}：
Figure BDA0002946295840000031
(1.5) if the authorized user signal occupies N channels and the backtracking window length is S, then: the input training set of the long-short term memory and convolutional neural network prediction model is { omega1:N,t-(S-1),Ω1:N,t-(S-2),…,Ω1:N,tH, the corresponding output training sample is omega1:N,t+1(ii) a The input training sample of the fusion model is omega1:N,t+1The corresponding output training sample is
Figure BDA0002946295840000032
And the central frequency point is in the occupied state at the moment (t + 1).
Specifically, in the step (2), the establishment of the hybrid prediction model includes the following steps:
(2.1) the front part of the long-short term memory and convolutional neural network prediction model consists of a long-short term memory network layer, and the rear part of the model consists of a convolutional neural network layer and a full-connection layer; model training is carried out on the prediction model based on an input training set and an output training sample of the long-short term memory and convolutional neural network prediction model, and a minimized model loss function is taken as a target:
Figure BDA0002946295840000033
wherein: omegapAs weight and deviation vector of the prediction model, NpThe number of parameters of the prediction model; beta is ai,tBeta is more than or equal to 0 as a frequency spectrum prediction result of the prediction model on the channel i at the time ti,t≤1，βi,tThe smaller the probability that channel i is idle at time t, βi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
(2.2) the fusion model comprises three full-connection layers, namely a parameter omega learned by the long-short term memory network layer and the convolutional neural network layerp、NpAnd spectral prediction result betai,tFusing to obtain the future spectrum occupation state of the authorized user, i.e. fusing the spectrum data mui,tFusing the spectral data mui,tThe prediction accuracy of (c) is characterized by Acc:
Figure BDA0002946295840000034
wherein: the larger the Acc value is, the higher the prediction accuracy is, and the smaller the Acc value is, the lower the prediction accuracy is; mu.si,tThe frequency spectrum prediction result is obtained by fusing the frequency spectrum prediction results of the long-term and short-term memory and convolutional neural network prediction models, and the frequency spectrum prediction result is more than or equal to mu and 0i,t≤1，μi,tThe smaller the probability that the channel i is idle at time t, the higher μi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
Figure BDA0002946295840000041
Ωcenter,tindicating the occupation state of the central frequency point at time t, i.e.
Figure BDA0002946295840000042
γcenterIndicating the adaptive quantization threshold of the central frequency point, i.e.
Figure BDA0002946295840000043
Specifically, in the step (3), the spectrum sharing includes the following steps:
to be provided with
Figure BDA0002946295840000044
Representing the signal bandwidth of the authorized user at the time t, representing the required bandwidth of the unauthorized user at the time t by H, and arranging the idle frequency band of the authorized user shared by the unauthorized user according to the following mode:
Figure BDA0002946295840000045
Figure BDA0002946295840000046
Figure BDA0002946295840000047
wherein: l represents the time gap length between time t and time (t + 1).
Has the advantages that: the satellite spectrum resource dynamic allocation method based on the neural network can monitor the signal use condition of the frequency band, identify the signal bandwidth and the parameters thereof, estimate the probability distribution of each authorized user, adaptively set the quantization threshold value, and output the final spectrum occupation prediction result after the constructed prediction model can predict the multi-channel use condition of the authorized users and is fused with the model.
Drawings
FIG. 1 is a flow chart of an embodiment of the present invention;
FIG. 2 is a block diagram of a neural network-based model for spectrum multi-channel fusion prediction in the present invention;
FIG. 3 is a schematic diagram of a training set interception method of the present invention;
FIG. 4 is a graph of the accuracy results of the spectrum fusion prediction of the method of the present invention;
fig. 5 is a graph comparing the spectrum utilization rate of the spectrum allocation method of the present invention with that of the conventional spectrum allocation method.
Detailed Description
The present invention will be further described with reference to the accompanying drawings.
The invention provides a satellite frequency spectrum resource dynamic allocation method based on a neural network, which carries out probability density estimation on each identified authorized signal through data preprocessing, sets corresponding threshold thresholds for different authorized user signals, and carries out quantization and division of input and output data sets on historical data; respectively establishing a prediction model and a fusion model based on a deep neural network, performing multi-channel prediction on each identified authorized user signal, inputting the prediction result of each channel into the fusion model, and fusing and outputting the future spectrum occupation state of the authorized user; a spectrum allocation model is established based on a neural network, a user future occupation state is obtained according to prediction, an allocation strategy is output, dynamic allocation of spectrum resources is achieved, and therefore the purpose of efficient utilization of the spectrum resources is achieved.
Fig. 1 is a flowchart of an implementation of a method for dynamically allocating satellite spectrum resources based on a neural network, which includes three parts, namely, broadband signal data preprocessing, prediction model building and spectrum sharing, and each part is specifically described below.
First, data preprocessing
Firstly, probability distribution estimation is carried out on power spectral density time sequence values of all channels of an authorized user, then, an adaptive quantization threshold is set for each channel of the authorized user, and finally, signals of the authorized user are divided into a long-term and short-term memory and convolutional neural network prediction model training set and a fusion model training set according to a set length. The method comprises the following steps:
(1.1) carrying out probability distribution estimation on each channel power spectral density time sequence value of an authorized user, and carrying out probability distribution estimation on a channel i, wherein the probability distribution estimation comprises the following steps:
Figure BDA0002946295840000051
wherein: x is a random value of the power spectral density of the channel i at time t, xi,tFor the power spectral density value of channel i at time t,
Figure BDA0002946295840000052
the probability distribution estimation value of the channel i is obtained, h is a probability distribution estimation parameter, T is the length of a power spectral density time sequence value, and K (·) is a Gaussian kernel function;
(1.2) mixing
Figure BDA0002946295840000053
Is taken as the noise power mean value N of the channel iiExpressed as:
Figure BDA0002946295840000054
wherein: x is the number of0Is composed of
Figure BDA0002946295840000055
The value of x when true;
(1.3) calculating an adaptive quantization threshold gamma of the channel ii＝Ni+ ω, ω is a preset positive offset;
(1.4) time-sequence value { x) of power spectral density of channel ii,1,xi,2,…,xi,TAt gammaiQuantizing to a channel occupancy state timing value { omega ] for a thresholdi,1,Ωi,2,…,Ωi,T}：
Figure BDA0002946295840000061
(1.5) if the authorized user signal occupies N channels and the backtracking window length is S, then: the input training set of the long-short term memory and convolutional neural network prediction model is { omega1:N,t-(S-1),Ω1:N,t-(S-2),…,Ω1:N,tH, the corresponding output training sample is omega1:N,t+1(ii) a The input training sample of the fusion model is omega1:N,t+1The corresponding output training sample is
Figure BDA0002946295840000062
And the central frequency point is in the occupied state at the moment (t + 1).
Second, mixed prediction model establishment
The hybrid prediction model comprises a long-term memory and short-term memory convolutional neural network prediction model and a fusion model; the long-short term memory and convolutional neural network prediction model is divided into a first half part and a second half part, the first half part comprises a layer of long-short term memory network layer, the second half part comprises a layer of convolutional neural network layer and a layer of full connection layer, the long-short term memory network layer is used for learning the time domain correlation of the authorized user signals, the convolutional neural network layer is used for learning the inter-channel correlation of the authorized user signals, and the full connection layer outputs a spectrum prediction result; the fusion model comprises three full-connection layers, and the parameters learned by the long-term and short-term memory network layer, the parameters learned by the convolutional neural network layer and the spectrum prediction result are fused to obtain the future spectrum occupation state of the authorized user, namely fused spectrum data. The method specifically comprises the following steps:
(2.1) the front part of the long-short term memory and convolutional neural network prediction model consists of a long-short term memory network layer, and the rear part of the model consists of a convolutional neural network layer and a full-connection layer; model training is carried out on the prediction model based on an input training set and an output training sample of the long-short term memory and convolutional neural network prediction model, and a minimized model loss function is taken as a target:
Figure BDA0002946295840000063
wherein: omegapAs weight and deviation vector of the prediction model, NpThe number of parameters of the prediction model; beta is ai,tBeta is more than or equal to 0 as a frequency spectrum prediction result of the prediction model on the channel i at the time ti,t≤1，βi,tThe smaller the probability that channel i is idle at time t, βi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
(2.2) the fusion model comprises three full-connection layers, namely a parameter omega learned by the long-short term memory network layer and the convolutional neural network layerp、NpAnd spectral prediction result betai,tFusing to obtain the future spectrum occupation state of the authorized user, i.e. fusing the spectrum data mui,tFusing the spectral data mui,tThe prediction accuracy of (c) is characterized by Acc:
Figure BDA0002946295840000071
wherein: the larger the Acc value is, the higher the prediction accuracy is, and the smaller the Acc value is, the lower the prediction accuracy is; mu.si,tThe frequency spectrum prediction result is obtained by fusing the frequency spectrum prediction results of the long-term and short-term memory and convolutional neural network prediction models, and the frequency spectrum prediction result is more than or equal to mu and 0i,t≤1，μi,tThe smaller the probability that the channel i is idle at time t, the higher μi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
Figure BDA0002946295840000072
Ωcenter,tindicating the occupation state of the central frequency point at time t, i.e.
Figure BDA0002946295840000073
γcenterIndicating the adaptive quantization threshold of the central frequency point, i.e.
Figure BDA0002946295840000074
Third, spectrum sharing
And inputting the power spectrum density time sequence value of the signal of the authorized user into a hybrid prediction model, predicting the future spectrum occupation state of the authorized user, selecting an appropriate idle channel of the authorized user through a distribution network for shared communication according to the prediction result and the required bandwidth of the unauthorized user, and outputting a spectrum distribution strategy.
To be provided with
Figure BDA0002946295840000075
Representing the signal bandwidth of the authorized user at the time t, representing the required bandwidth of the unauthorized user at the time t by H, and arranging the idle frequency band of the authorized user shared by the unauthorized user according to the following mode:
Figure BDA0002946295840000076
Figure BDA0002946295840000077
Figure BDA0002946295840000078
wherein: l represents the time gap length between time t and time (t + 1).
The evaluation index is the frequency spectrum utilization rate, namely the ratio of the number of the frequency spectrum utilization time slots of the user frequency band divided by the total number of the time slots after the frequency spectrum allocation in a period of time. Fig. 4 shows that the prediction accuracy of the prediction is calculated once every 10 seconds by using the trained prediction fusion model to predict the future spectrum utilization state of the real-time heaven-tong authorized user under the spectrum data acquired by the heaven-tong satellite 1, and the accuracy curve diagram predicts about 16 minutes, so that the prediction accuracy can also reach about 90% on the real-time heaven-tong data. And spectrum resource allocation is carried out on the basis of the predicted result, the existing 3 authorized users respectively have different communication bandwidths, and meanwhile, the other 3 unauthorized users need different authorized bandwidths for communication and meet the Poisson arrival service. The original utilization rate of the frequency spectrum of the 3 authorized channels is measured to be 37.7%, fig. 5 shows the change situation of the utilization rate of the frequency spectrum resources of the authorized channels after the unauthorized users perform frequency spectrum allocation with the traffic of different arrival rates, and the frequency spectrum utilization rate of the method of the present invention is obviously higher than that of the traditional frequency spectrum resource allocation method based on frequency spectrum sensing under the same arrival rate.
In summary, the method for dynamically allocating spectrum resources based on the neural network adaptively determines the quantized threshold value by performing probability distribution estimation on each authorized user signal, quantizes the signal into an idle state and an occupied state, and divides a corresponding training set of a prediction model and a fusion model; constructing a prediction and fusion model and training, wherein the trained prediction model can output a prediction result of multiple channels of an authorized user, and the fusion model fuses the prediction result of the multiple channels and outputs the future spectrum use state of the user to acquire the future spectrum occupation state in advance; a spectrum allocation model is established based on a neural network, and spectrum resources are dynamically allocated according to the communication requirements of unauthorized users, so that the utilization rate of the spectrum resources is improved.
The above description is only of the preferred embodiments of the present invention, and it should be noted that: it will be apparent to those skilled in the art that various modifications and adaptations can be made without departing from the principles of the invention and these are intended to be within the scope of the invention.

Claims (4)

1. A satellite spectrum resource dynamic allocation method based on a neural network is characterized in that: the method comprises three parts of broadband signal data preprocessing, prediction model establishment and spectrum sharing, and specifically comprises the following steps:
(1) data preprocessing: firstly, carrying out probability distribution estimation on power spectral density time sequence values of all channels of an authorized user, then setting a self-adaptive quantization threshold for all channels of the authorized user, and finally dividing signals of the authorized user into a long-term and short-term memory and convolutional neural network prediction model training set and a fusion model training set according to a set length;
(2) establishing a hybrid prediction model: the hybrid prediction model comprises a long-term memory and short-term memory convolutional neural network prediction model and a fusion model; the long-short term memory and convolutional neural network prediction model is divided into a first half part and a second half part, the first half part comprises a layer of long-short term memory network layer, the second half part comprises a layer of convolutional neural network layer and a layer of full connection layer, the long-short term memory network layer is used for learning the time domain correlation of the authorized user signals, the convolutional neural network layer is used for learning the inter-channel correlation of the authorized user signals, and the full connection layer outputs a spectrum prediction result; the fusion model comprises three full-connection layers, and is used for fusing parameters learned by the long-term and short-term memory network layer, parameters learned by the convolutional neural network layer and spectrum prediction results to obtain the future spectrum occupation state of the authorized user, namely fused spectrum data;
(3) spectrum sharing: and inputting the power spectrum density time sequence value of the signal of the authorized user into a hybrid prediction model, predicting the future spectrum occupation state of the authorized user, selecting an appropriate idle channel of the authorized user through a distribution network for shared communication according to the prediction result and the required bandwidth of the unauthorized user, and outputting a spectrum distribution strategy.
2. The method for dynamically allocating spectrum resources based on a neural network satellite according to claim 1, wherein: in the step (1), the data preprocessing comprises the following steps:
(1.1) carrying out probability distribution estimation on each channel power spectral density time sequence value of an authorized user, and carrying out probability distribution estimation on a channel i, wherein the probability distribution estimation comprises the following steps:
Figure FDA0002946295830000011
wherein: x is a random value of the power spectral density of the channel i at time t, xi,tFor the power spectral density value of channel i at time t,
Figure FDA0002946295830000012
the method comprises the following steps of (1) obtaining a probability distribution estimation value of a channel i, h is a probability distribution estimation parameter, T is the length of a power spectral density time sequence value, and K (·) is a Gaussian kernel function;
(1.2) mixing
Figure FDA0002946295830000013
Is taken as the noise power mean value N of the channel iiExpressed as:
Figure FDA0002946295830000014
wherein: x is the number of0Is composed of
Figure FDA0002946295830000021
The value of x when true;
(1.3) calculating an adaptive quantization threshold gamma of the channel ii＝Ni+ ω, ω is a preset positive offset;
(1.4) time-sequence value { x) of power spectral density of channel ii,1,xi,2,…,xi,TAt gammaiQuantizing to a channel occupancy state timing value { omega ] for a thresholdi,1,Ωi,2,…,Ωi,T}：
Figure FDA0002946295830000022
(1.5) if the authorized user signal occupies N channels and the backtracking window length is S, then: the input training set of the long-short term memory and convolutional neural network prediction model is { omega1:N,t-(S-1),Ω1:N,t-(S-2),…,Ω1:N,tH, the corresponding output training sample is omega1:N,t+1(ii) a The input training sample of the fusion model is omega1:N,t+1The corresponding output training sample is
Figure FDA0002946295830000023
And the central frequency point is in the occupied state at the moment (t + 1).
3. The method for dynamically allocating spectrum resources based on a neural network satellite according to claim 1, wherein: in the step (2), the establishment of the hybrid prediction model comprises the following steps:
(2.1) the front part of the long-short term memory and convolutional neural network prediction model consists of a long-short term memory network layer, and the rear part of the model consists of a convolutional neural network layer and a full-connection layer; model training is carried out on the prediction model based on an input training set and an output training sample of the long-short term memory and convolutional neural network prediction model, and a minimized model loss function is taken as a target:
Figure FDA0002946295830000024
wherein: omegapAs weight and deviation vector of the prediction model, NpThe number of parameters of the prediction model; beta is ai,tBeta is more than or equal to 0 as a frequency spectrum prediction result of the prediction model on the channel i at the time ti,t≤1，βi,tThe smaller the probability that channel i is idle at time t, βi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
(2.2) the fusion model comprises three full-connection layers, namely a parameter omega learned by the long-short term memory network layer and the convolutional neural network layerp、NpAnd spectral prediction result betai,tFusing to obtain the future spectrum occupation state of the authorized user, i.e. fusing the spectrum data mui,tFusing the spectral data mui,tThe prediction accuracy of (c) is characterized by Acc:
Figure FDA0002946295830000031
wherein: the larger the Acc value is, the higher the prediction accuracy is, and the smaller the Acc value is, the lower the prediction accuracy is; mu.si,tThe frequency spectrum prediction result is obtained by fusing the frequency spectrum prediction results of the long-term and short-term memory and convolutional neural network prediction models, and the frequency spectrum prediction result is more than or equal to mu and 0i,t≤1，μi,tThe smaller the probability that the channel i is idle at time t, the higher μi,tThe larger the channel i is, the higher the possibility that the channel i is in an occupied state at the moment t is;
Figure FDA0002946295830000032
Ωcenter,tindicating the occupation state of the central frequency point at time t, i.e.
Figure FDA0002946295830000033
γcenterIndicating the adaptive quantization threshold of the central frequency point, i.e.
Figure FDA0002946295830000034
4. The method for dynamically allocating spectrum resources based on a neural network satellite according to claim 1, wherein: in the step (3), the spectrum sharing includes the following steps:
to be provided with
Figure FDA0002946295830000035
Representing the signal bandwidth of the authorized user at the time t, representing the required bandwidth of the unauthorized user at the time t by H, and arranging the idle frequency band of the authorized user shared by the unauthorized user according to the following mode:
Figure FDA0002946295830000036
Figure FDA0002946295830000037
Figure FDA0002946295830000038
wherein: l represents the time gap length between time t and time (t + 1).
CN202110197503.2A 2021-02-22 2021-02-22 Satellite spectrum resource dynamic allocation method based on neural network Pending CN113014340A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110197503.2A CN113014340A (en) 2021-02-22 2021-02-22 Satellite spectrum resource dynamic allocation method based on neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110197503.2A CN113014340A (en) 2021-02-22 2021-02-22 Satellite spectrum resource dynamic allocation method based on neural network

Publications (1)

Publication Number Publication Date
CN113014340A true CN113014340A (en) 2021-06-22

Family

ID=76406009

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110197503.2A Pending CN113014340A (en) 2021-02-22 2021-02-22 Satellite spectrum resource dynamic allocation method based on neural network

Country Status (1)

Country Link
CN (1) CN113014340A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115276856A (en) * 2022-06-16 2022-11-01 宁波大学 Channel selection method based on deep learning
CN116318374A (en) * 2023-05-19 2023-06-23 红珊科技有限公司 Satellite communication data channel analysis system and method based on traffic analysis
CN116545556A (en) * 2023-04-28 2023-08-04 哈尔滨工程大学 Electromagnetic spectrum occupancy rate two-dimensional prediction method based on dynamic threshold and residual convolution network

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109245840A (en) * 2018-10-15 2019-01-18 哈尔滨工业大学 Spectrum prediction method in cognitive radio system based on convolutional neural networks
CN110138475A (en) * 2019-05-08 2019-08-16 南京邮电大学 A kind of adaptive threshold channel occupation status prediction technique based on LSTM neural network
CN110730046A (en) * 2019-10-18 2020-01-24 中国人民解放军陆军工程大学 Cross-frequency-band spectrum prediction method based on deep migration learning
CN111224726A (en) * 2019-11-21 2020-06-02 南京邮电大学 Video live broadcast system based on long-term and short-term memory network and implementation method thereof

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109245840A (en) * 2018-10-15 2019-01-18 哈尔滨工业大学 Spectrum prediction method in cognitive radio system based on convolutional neural networks
CN110138475A (en) * 2019-05-08 2019-08-16 南京邮电大学 A kind of adaptive threshold channel occupation status prediction technique based on LSTM neural network
CN110730046A (en) * 2019-10-18 2020-01-24 中国人民解放军陆军工程大学 Cross-frequency-band spectrum prediction method based on deep migration learning
CN111224726A (en) * 2019-11-21 2020-06-02 南京邮电大学 Video live broadcast system based on long-term and short-term memory network and implementation method thereof

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
HAOYU LI等: "Online Spectrum Prediction With Adaptive Threshold Quantization", 《IEEE ACCESS》 *
XIAOJIN DING等: "Deep Learning Aided Spectrum Prediction for Satellite Communication Systems", 《IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY》 *
刘稳等: "卫星S频段下行链路频谱占用建模与预测", 《太赫兹科学与电子信息学报》 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115276856A (en) * 2022-06-16 2022-11-01 宁波大学 Channel selection method based on deep learning
CN115276856B (en) * 2022-06-16 2023-09-29 宁波大学 Channel selection method based on deep learning
CN116545556A (en) * 2023-04-28 2023-08-04 哈尔滨工程大学 Electromagnetic spectrum occupancy rate two-dimensional prediction method based on dynamic threshold and residual convolution network
CN116545556B (en) * 2023-04-28 2024-03-29 哈尔滨工程大学 Electromagnetic spectrum occupancy rate two-dimensional prediction method based on dynamic threshold and residual convolution network
CN116318374A (en) * 2023-05-19 2023-06-23 红珊科技有限公司 Satellite communication data channel analysis system and method based on traffic analysis
CN116318374B (en) * 2023-05-19 2023-07-21 红珊科技有限公司 Satellite communication data channel analysis system and method based on traffic analysis

Similar Documents

Publication Publication Date Title
CN113014340A (en) Satellite spectrum resource dynamic allocation method based on neural network
CN108123828B (en) Ultra-dense network resource allocation method based on access user mobility prediction
CN106534333A (en) Bidirectional selection computing unloading method based on MEC and MCC
CN108901075B (en) GS algorithm-based resource allocation method
CN111711666B (en) Internet of vehicles cloud computing resource optimization method based on reinforcement learning
CN101640570A (en) Frequency spectrum cognitive method and energy detection method and device
CN111294812B (en) Resource capacity-expansion planning method and system
CN111491312B (en) Method and equipment for predicting allocation, acquisition and training of wireless resources and neural network
CN113504999A (en) Scheduling and resource allocation method for high-performance hierarchical federated edge learning
CN106656374A (en) Cooperative broadband spectrum sensing method based on double-threshold energy detection
CN112003663A (en) Cognitive radio frequency spectrum sensing method based on LSTM neural network
CN108029138B (en) Method for opportunistic access to spectrum
CN109347584B (en) Spectrum sensing method based on exponential smoothing prediction
Mosavat-Jahromi et al. Prediction and modeling of spectrum occupancy for dynamic spectrum access systems
Sexton et al. On provisioning slices and overbooking resources in service tailored networks of the future
CN115190489A (en) Cognitive wireless network dynamic spectrum access method based on deep reinforcement learning
CN111741450A (en) Network flow prediction method and device and electronic equipment
CN109219055B (en) Main user duty ratio estimation method
CN112508408B (en) Mapping model construction method of radio resource management index under edge calculation
CN114222371A (en) Flow scheduling method for coexistence of eMBB (enhanced multimedia broadcast/multicast service) and uRLLC (unified radio link control) equipment
CN108012341B (en) Access method for joint node clustering and data caching in narrowband Internet of things
WO2013000240A1 (en) Method and system for sensing frequency spectrum for combined multi-node
CN116567843A (en) Wireless resource allocation optimization device and method
CN108173610B (en) Second-order statistic-based cooperative spectrum sensing method for heterogeneous wireless network
CN112073138B (en) Double-threshold cooperative spectrum sensing method based on quantization

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
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20210622

WD01 Invention patent application deemed withdrawn after publication