CN108616471A - A kind of signal modulate method and apparatus based on convolutional neural networks - Google Patents
A kind of signal modulate method and apparatus based on convolutional neural networks Download PDFInfo
- Publication number
- CN108616471A CN108616471A CN201810427110.4A CN201810427110A CN108616471A CN 108616471 A CN108616471 A CN 108616471A CN 201810427110 A CN201810427110 A CN 201810427110A CN 108616471 A CN108616471 A CN 108616471A
- Authority
- CN
- China
- Prior art keywords
- convolutional neural
- neural networks
- signal
- input
- modulation
- 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.)
- Granted
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0012—Modulated-carrier systems arrangements for identifying the type of modulation
Landscapes
- Engineering & Computer Science (AREA)
- Computer Networks & Wireless Communication (AREA)
- Signal Processing (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The signal modulate method and apparatus based on convolutional neural networks that the invention discloses a kind of, method include:Whether the length N and the input signal length L that convolutional neural networks are supported for comparing the signal received are equal;When N is equal to L, directly the signal received is input in the convolutional neural networks, to obtain the signal modulation style of the convolutional neural networks output;It when N is not equal to L, is input in the convolutional neural networks after the signal received is carried out corresponding polishing or segment processing, to obtain the signal modulation style of the convolutional neural networks output.To obtain recognition accuracy more higher than the method for traditional feature based extraction under low signal-to-noise ratio.And different signal lengths to be identified have been adapted to, the complete information of signal to be identified is can make full use of, information waste is avoided.
Description
Technical field
The present invention relates to signal modulate technical fields, and in particular to a kind of signal modulation based on convolutional neural networks
Recognition methods and device.
Background technology
Traditional signal modulate is mainly realized by extracting signal characteristic method.There is scholar to propose a kind of modulation
Kind identification method, this method avoid special feature extraction step, directly using radio signal sample data as convolution
Identification of the input and completion of neural network to signal modulation mode, obtains under low signal-to-noise ratio and is carried than traditional feature based
The higher recognition accuracy of method taken.
However, the length of the signal to be identified in this Modulation Identification method be it is fixed (such as length be 128 sampling
Point), in practical application, signal length to be identified is not known simultaneously, may be than signal input length that convolutional neural networks design more
It is long or shorter.When bigger than the signal of design input length for signal length to be identified, a kind of simple way is from waiting knowing
One section of signal segment for meeting convolutional neural networks input is intercepted in level signal, and the new number section of interception is identified, but in this way
The signal message for not inputting convolutional neural networks can be not only slatterned, but also recognition accuracy can be influenced, therefore, there is an urgent need for improve.
Invention content
The signal modulate method and apparatus based on convolutional neural networks that the present invention provides a kind of, to solve existing letter
Number Modulation identification technology fails the problem for making full use of signal message and recognition accuracy to be identified not high.
According to the one side of the application, a kind of signal modulate method based on convolutional neural networks is provided, is wrapped
It includes:
Whether the length N and the input signal length L that convolutional neural networks are supported for comparing the signal received are equal;
When N is equal to L, directly the signal received is input in the convolutional neural networks, it is described to obtain
The signal modulation style of convolutional neural networks output;
When N is not equal to L, the volume is input to after the signal received is carried out corresponding polishing or segment processing
In product neural network, to obtain the signal modulation style of the convolutional neural networks output.
Optionally, when N is not equal to L, institute is input to after the signal received is carried out corresponding polishing or segment processing
Stating convolutional neural networks includes:
If N < L, according to the input signal length L that the convolutional neural networks are supported, the signal received is mended
After zero, input a signal into the convolutional neural networks;
If N > L, the signal received is divided into each signal segment that length is L, and each signal segment is input to described
In convolutional neural networks.
Optionally, after carrying out zero padding to the signal received, inputting a signal into the convolutional neural networks includes:
After N-L 0 is filled in the tail portion of the sample sequence x (n) of the signal received, by sample sequence x (n) (n=0,1,
2 ..., L-1) it is input in convolutional neural networks according to the input format of the convolutional neural networks to be modulated type knowledge
Not;
The signal received is divided into length and is each signal segment of L, and each signal segment is input to the convolutional Neural
Network includes:
Sample sequence x (n) to the signal received chooses each signal segment y that length is L by interval P slidingsi(m), will
Each signal segment yi(m) it is input in convolutional neural networks according to the input format of convolutional neural networks to be modulated type knowledge
Not;
Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+m), m=0,1,2 ..., L-1, i=1,2 ...,
I,Indicate the maximum integer no more than (N-L+1)/P;1≤P≤(N-L+1).
Optionally, by each signal segment yi(m) convolutional neural networks are input to simultaneously according to the input format of convolutional neural networks
After being modulated type identification, amalgamation judging is carried out according to recognition result, or amalgamation judging is carried out according to confidence level.
Optionally, carrying out amalgamation judging according to recognition result includes:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ...,
M, M indicate that the convolutional neural networks support the modulation type number of identification;
Step S2, by yi(m) convolutional neural networks are inputted and is modulated type identification, obtain original recognition result oi,
Wherein oi∈[1,2,...,M];
Step S3, if oi=j, then by the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi
(m) identification finishes;
Step S4 calculates vjMaximum value, work as vjOnly there are one when maximum value, then the corresponding subscript j of maximum value is represented
Modulation type is exported as final modulation identification result;Work as vjMaximum value when being more than two, then randomly select one
The corresponding subscript j of the maximum value modulation types represented are exported as final modulation identification result after a maximum value.
Optionally, carrying out amalgamation judging according to confidence level includes:
Step A1, by signal segment yi(m) convolutional neural networks are input to and are modulated type identification, and obtain confidence
Spend vector pi=[pi,1,pi,2,...,pi,M],
Wherein, pi,jIt indicates signal segment yi(m) confidence level of jth kind modulation type, j=1,2 ..., M, M tables are identified as
Show that the convolutional neural networks support the modulation type number of identification;
Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M;
Step A3 determines the average value wjMaximum value, work as wjIt is there are one only when maximum value, then maximum value is corresponding
The modulation type that subscript j is represented as final modulation identification result and exports;Work as wjMaximum value when being more than two,
After then randomly selecting a maximum value, the corresponding subscript j of the maximum value modulation types represented are known as final modulation type
Other result output.
According to another aspect of the present invention, a kind of signal modulate device based on convolutional neural networks is provided,
Including:
Comparison module, the input signal length L that the length N for comparing the signal received is supported with convolutional neural networks
It is whether equal;
Identification module, for when N is equal to L, the signal received to be directly input to the convolutional neural networks
In, to obtain the signal modulation style of the convolutional neural networks output;When N be not equal to L when, by the signal received into
It is input in the convolutional neural networks after the corresponding polishing of row or segment processing, to obtain the convolutional neural networks output
Signal modulation style.
Optionally, the identification module is specifically used for, if N < L, believes according to the input that the convolutional neural networks are supported
Number length L is input a signal into after carrying out zero padding to the signal received in the convolutional neural networks;If N > L, will connect
The signal received is divided into length and is each signal segment of L, and each signal segment is input in the convolutional neural networks.
Optionally, the identification module is used for, and N-L 0 is filled in the tail portion of the sample sequence x (n) of the signal received
Afterwards, sample sequence x (n) (n=0,1,2 ..., L-1) is input to convolution god according to the input format of the convolutional neural networks
Through in network to be modulated type identification;And the sample sequence x (n) for the signal to receiving, by interval P slidings
Choose each signal segment y that length is Li(m), by each signal segment yi(m) it is input to convolution according to the input format of convolutional neural networks
To be modulated type identification in neural network;Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+m), m=0,1,
2 ..., L-1, i=1,2 ..., I,Indicate the maximum integer no more than (N-L+1)/P;1≤P≤
(N-L+1)。
Optionally, the identification module is additionally operable to each signal segment yi(m) defeated according to the input format of convolutional neural networks
Enter to convolutional neural networks and after carrying out modulation identification, according to recognition result carry out amalgamation judging, or according to confidence level into
Row amalgamation judging;
Specifically, including according to recognition result progress amalgamation judging:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ...,
M, M indicate that the convolutional neural networks support the modulation type number of identification;
Step S2, by yi(m) convolutional neural networks are inputted and is modulated type identification, obtain original recognition result oi,
Wherein oi∈[1,2,...,M];
Step S3, if oi=j, then by the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi
(m) identification finishes;
Step S4 calculates vjMaximum value, work as vjOnly there are one when maximum value, then the corresponding subscript j of maximum value is represented
Modulation type is exported as final modulation identification result;Work as vjMaximum value when being more than two, then randomly select one
The corresponding subscript j of the maximum value modulation types represented are exported as final modulation identification result after a maximum value;
Carrying out amalgamation judging according to confidence level includes:
Step A1, by signal segment yi(m) convolutional neural networks are input to and are modulated type identification, and obtain confidence
Spend vector pi=[pi,1,pi,2,...,pi,M],
Wherein, pi,jIt indicates signal segment yi(m) confidence level of jth kind modulation type, j=1,2 ..., M, M tables are identified as
Show that the convolutional neural networks support the modulation type number of identification;
Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M;
Step A3 determines the average value wjMaximum value, work as wjIt is there are one only when maximum value, then maximum value is corresponding
The modulation type that subscript j is represented as final modulation identification result and exports;Work as wjMaximum value when being more than two,
After then randomly selecting a maximum value, the corresponding subscript j of the maximum value modulation types represented are known as final modulation type
Other result output.
The advantageous effect of the embodiment of the present invention is:The signal modulate based on convolutional neural networks of the embodiment of the present invention
Scheme, whether the length N and the input signal length L that convolutional neural networks are supported for first comparing the signal received are equal, if N
Equal to L, directly the signal received is input in convolutional neural networks, if N is not equal to L, the signal received is carried out
It is input in convolutional neural networks after corresponding polishing or segment processing, to obtain the signal modulation class of convolutional neural networks output
Type.The technical solution of the present embodiment has adapted to different signal lengths to be identified as a result, can make full use of the complete of signal to be identified
Whole information avoids information waste while improving the recognition accuracy of signal modulation style.
Description of the drawings
Fig. 1 is a kind of convolutional neural networks structural schematic diagram of one embodiment of the invention;
Fig. 2 is the flow chart of the signal modulate method based on convolutional neural networks of one embodiment of the invention;
Fig. 3 is the flow signal of the signal modulate method based on convolutional neural networks of another embodiment of the present invention
Figure;
Fig. 4 is the signal modulate device block diagram based on convolutional neural networks of one embodiment of the invention.
Specific implementation mode
In dynamic spectrum access network, unauthorized user can use authorized user (also referred to as primary user) is current to be not used
Spectrum interposition communicated, the utilization rate of radio spectrum resources is improved with this.One key technology of dynamic spectrum access is
It needs to be detected authorized user and causes harmful interference to avoid to authorized user.Modulation identification is received by judging
Radio signal which kind of modulation type primary user is assisted in identifying using, for judgement primary user's type for have important meaning
Justice.Based on this, the signal modulate method and apparatus based on convolutional neural networks that the present embodiment provides a kind of improving identification letter
The accuracy rate of number modulation system.
For ease of understanding, signal modulation style and convolutional neural networks are briefly described here.
In order to ensure communication efficiency, the problems in distant signal transmission is overcome, improve the availability of frequency spectrum and communication quality,
Signal spectrum to be moved in high frequency channel by modulation and be transmitted.This signal loading that will be sent is to high-frequency signal
Process just cry modulation.Three kinds of digital signal most basic modulator approach amplitude modulation, frequency modulation and phase modulation, other various modulator approaches are all
It is the improvement or combination of above method.Convolutional neural networks (Convolutional Neural Network, CNN) because its
The protrusion effect of image domains, there is important position in deep learning, is with a kind of one of neural network the most extensive.
Compared with traditional neural network, there are two apparent features.First, convolutional neural networks using weights are shared, make to input
The output of corresponding channel is obtained with identical filter.Secondly, convolutional neural networks provide pondization operation, to have certain journey
The translation invariant shape of degree, and pass through the down-sampled computation complexity for reducing deeper.Most of convolutional neural networks structures
Usually contain four Primary layers:Convolutional layer, normalization layer, nonlinear activation layer and pond layer.
The roads I (real part for receiving complex baseband signal) for receiving signal and the roads Q (are received the void of complex baseband signal by the present invention
Portion) it is combined into input of two row of matrix as convolutional neural networks respectively, a kind of convolutional neural networks structure is as shown in Figure 1, branch
The input signal length held is L, therefore the input format supported is the matrix form that L rows 2 arrange (i.e. Lx2).In Fig. 1, conv is represented
Convolutional layer, the number (i.e. 21x1) before conv indicate the size of convolution kernel, and number (128) later indicates of convolution kernel
Number;ReLU indicates that rectification linearly activates;Dropout indicates Dropout layers, and bracket inner digital (0.5) indicates Dropout probability;
Fc indicates that full articulamentum, digital (that is, 256) represent neuron number;SoftMax indicates SoftMax layers, this layer of neuron number
For M, i.e. Modulation Types classification sum;Finally output is classification, and class label is encoded using One-Hot.In convolutional layer and non-thread
Property active coating between also include batch normalization layer do not draw in Fig. 1 for simplicity.In one embodiment, network instruction
Practice object function and cross entropy loss function may be used, stochastic gradient descent method may be used in training method.
Fig. 2 is the flow chart of the signal modulate method based on convolutional neural networks of one embodiment of the invention, ginseng
See that Fig. 2, the signal modulate method based on convolutional neural networks of the present embodiment include the following steps:
Whether step S201 compares input signal length L that the length N of the signal received is supported with convolutional neural networks
It is equal;
The signal received is directly input in the convolutional neural networks by step S202 when N is equal to L, with
Obtain the signal modulation style of the convolutional neural networks output;
Step S203 will be defeated after the corresponding polishing of signal progress received or segment processing when N is not equal to L
Enter into the convolutional neural networks, to obtain the signal modulation style of the convolutional neural networks output.
As shown in Figure 2 it is found that this signal modulate method of the present embodiment first compares the length of the signal received
The relationship for the input signal length supported with convolutional neural networks, and different disposal is carried out according to the different of comparison result, to
It the signal to be identified of different length is realized, and makes full use of the information of signal to be identified, while raising signal modulation style
Recognition accuracy.
Fig. 3 is the flow signal of the signal modulate method based on convolutional neural networks of another embodiment of the present invention
Figure carries out weight with reference to Fig. 3 to the realization step of the signal modulate method based on convolutional neural networks of the present embodiment
Point explanation.
Referring to Fig. 3, flow starts, and first carries out step S301, judges the length N and convolutional neural networks of the signal received
The relationship of the input signal length L of support;
It is appreciated that the signal received is signal to be identified.Signal length to be identified is supported with convolutional neural networks
Input signal length between three kinds nothing but of relationship, that is, be less than relationship, be more than relationship or be equal to relationship.The present embodiment is directed to three
Kind different relationships take different processing steps, and it is relatively easy to be equal to relationship, directly input a signal into convolutional neural networks into
Row identification, and when signal length N is not equal to the length L that convolutional neural networks are supported, letter that the present embodiment will receive
It is input in convolutional neural networks again after number carrying out corresponding polishing or segment processing, if N < L are specifically included, according to convolution
The input signal length L that neural network is supported inputs a signal into convolutional neural networks after carrying out zero padding to the signal received
In;If N > L, the signal received is divided into length and is each signal segment of L, and each signal segment is input to convolutional Neural
In network.
Referring to Fig. 3, step S302 and step S305 is executed when for the first aforementioned relationship.
Specifically, step S302, if N is less than L, by signal zero padding, zero padding number is N-L.
In the present embodiment, after the tail portion of the sample sequence x (n) of the signal received fills N-L 0, by sample sequence x
(n) (n=0,1,2 ..., L-1) it is input in convolutional neural networks according to the input format of the convolutional neural networks to carry out
Modulation identification.Namely N-L 0, i.e. x (n)=0, n=N ..., L- are mended below in signal sample sequence x (n) to be identified
1。
Then step S305 is executed, input convolutional neural networks are identified, and recognition result is the highest modulation of confidence level
Type.Here it is that x (n) (n=0,1,2 ..., L-1) is inputted into convolutional Neural according to the input format that convolutional neural networks are supported
Network is modulated pattern identification, and recognition result is that highest modulation type of confidence level.
Step S303 and step S306 is executed when second of relationship.
Signal is segmented by step S303 if N is more than L, is L per segment length.
Here, the signal received is divided into length and is each signal segment of L, and each signal segment is input to the volume
Accumulating neural network includes:Sample sequence x (n) to the signal received chooses each signal that length is L by interval P slidings
Section yi(m), by each signal segment yi(m) it is input in convolutional neural networks according to the input format of convolutional neural networks to be adjusted
Type identification processed;Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+m), m=0,1,2 ..., L-1, i=1,
2 ..., I,Indicate the maximum integer no more than (N-L+1)/P;1≤P≤(N-L+1).
That is, treating identification signal sample sequence x (n), n=0,1,2 ..., N-1, wherein N are signal length, are pressed
Interval P carries out each signal segment y that length L is chosen in slidingi(m), wherein yi(m)=x ((i-1) P+m), m=0,1,2 ..., L-
1, i=1,2 ..., I. Indicate the maximum integer no more than (N-L+1)/P.
1≤P≤N-L+1。
Step S306 input convolutional neural networks will be identified per segment signal, melt by recognition result fusion or confidence level
It closes and determines final recognition result.
Each signal segment y that will divide in step S303i(m) input format supported according to convolutional neural networks is input to volume
Product neural network is modulated pattern identification, and amalgamation judging is carried out according to recognition result or confidence level.
Carrying out amalgamation judging according to recognition result in the present embodiment includes:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ...,
M, M indicate that the convolutional neural networks support the modulation type number of identification;
Step S2, by yi(m) convolutional neural networks are inputted and is modulated type identification, obtain original recognition result oi,
Wherein oi∈[1,2,...,M];
Step S3, if oi=j, then by the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi
(m) identification finishes;
Step S4 calculates vjMaximum value, work as vjOnly there are one when maximum value, then the corresponding subscript j of maximum value is represented
Modulation type is exported as final modulation identification result;Work as vjMaximum value when being more than two, then randomly select one
The corresponding subscript j of the maximum value modulation types represented are exported as final modulation identification result after a maximum value.
That is, enabling identification signal section y firsti(m) it is the number v of jth kind modulation typej=0, j=1,2 ..., M,
M indicates that designed convolutional neural networks support the modulation type number of identification.By yi(m) according to the defeated of convolutional neural networks
Entry format is input to convolutional neural networks and is modulated identification, recognition result oi, oi∈ [1,2 ..., M], if oi=j then will
Recognition result is that the number of jth kind modulation type adds 1, i.e. vj=vj+ 1, the process is repeated until all yi(m) identification finishes.
Finally seek vjThe maximum value of (j=1,2 ..., M), if only there are one maximum values, corresponding subscript j is i.e. as final
Modulation identification result (recognition result is jth kind modulation type), if there is multiple maximum values, then randomly selects a maximum
Value, corresponding subscript j is i.e. as final modulation identification result (recognition result is jth kind modulation type).
Carrying out amalgamation judging according to confidence level includes:
Step A1, by signal segment yi(m) convolutional neural networks are input to and are modulated type identification, and obtain confidence
Spend vector pi=[pi,1,pi,2,...,pi,M],
Wherein, pi,jIt indicates signal segment yi(m) confidence level of jth kind modulation type, j=1,2 ..., M, M tables are identified as
Show that the convolutional neural networks support the modulation type number of identification;
Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M;
Step A3 determines the average value wjMaximum value, work as wjIt is there are one only when maximum value, then maximum value is corresponding
The modulation type that subscript j is represented as final modulation identification result and exports;Work as wjMaximum value when being more than two,
After then randomly selecting a maximum value, the corresponding subscript j of the maximum value modulation types represented are known as final modulation type
Other result output.
That is, first by yi(m) convolutional neural networks are input to according to the input format of convolutional neural networks to carry out
Modulation Identification obtains confidence level vector (i.e. convolutional neural networks SoftMax layers of outputs) pi=[pi,1,pi,2,...,pi,M],
Wherein pi,jIt indicates yi(m) it is identified as the confidence level of jth kind modulation type, j=1,2 ..., M, M indicate the convolution god of design
The modulation type number of identification is supported through network.Then the average value for calculating the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M.Finally seek wjThe maximum value of (j=1,2 ..., M), if maximum value there are one only,
Its corresponding subscript j i.e. as final modulation identification result (recognition result be jth kind modulation type), if there is it is multiple most
Big value, then randomly select a maximum value, and i.e. as final modulation identification result, (recognition result is corresponding subscript j
Jth kind modulation type).
Step S304 is executed when the third relationship, if N is equal to L, convolutional neural networks is input a signal into and is identified, and is known
Other result is the highest modulation type of confidence level.
From the foregoing, it will be observed that present embodiment discloses a kind of convolutional neural networks Modulation Identification sides adapting to unlike signal length
Method is adapted it to when signal sample sequence length to be identified is less than the input length that convolutional neural networks are supported by zero padding
Convolutional neural networks input format;When signal sample sequence length to be identified is more than the input length that convolutional neural networks are supported
When, it is modulated pattern identification by carrying out segmentation input convolutional neural networks to signal sample sequence, then pass through recognition result
Fusion or confidence level fusion provide final Modulation Types recognition result.The information for making full use of complete signal to be identified, is improved
Modulation identification accuracy rate.
Fig. 4 is the signal modulate device block diagram based on convolutional neural networks of one embodiment of the invention, referring to figure
4, the signal modulate device 400 based on convolutional neural networks of the present embodiment includes:
Comparison module 401, it is long for comparing the input signal that the length N of the signal received and convolutional neural networks are supported
Whether equal spend L;
Identification module 402, for when N is equal to L, the signal received to be directly input to the convolutional Neural net
In network, to obtain the signal modulation style of the convolutional neural networks output;When N is not equal to L, by the signal received
It is input in the convolutional neural networks after carrying out corresponding polishing or segment processing, is exported with obtaining the convolutional neural networks
Signal modulation style.
In one embodiment of the invention, identification module 402 is specifically used for, if N < L, according to the convolutional Neural
The input signal length L of network support inputs a signal into the convolutional neural networks after carrying out zero padding to the signal received
In;If N > L, the signal received is divided into length and is each signal segment of L, and each signal segment is input to the convolution
In neural network.
In one embodiment of the invention, identification module 402 is in the tail portion of the sample sequence x (n) of the signal received
It is after filling N-L 0, sample sequence x (n) (n=0,1,2 ..., L-1) is defeated according to the input format of the convolutional neural networks
Enter into convolutional neural networks to be modulated type identification;And the sample sequence x (n) for the signal to receiving, it presses
It is spaced P slidings and chooses each signal segment y that length is Li(m), by each signal segment yi(m) according to the input format of convolutional neural networks
It is input in convolutional neural networks to be modulated type identification;Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+
M), m=0,1,2 ..., L-1, i=1,2 ..., I,Indicate whole no more than the maximum of (N-L+1)/P
Number;1≤P≤(N-L+1).
In one embodiment of the invention, identification module 402 is additionally operable to each signal segment yi(m) according to convolutional Neural net
After the input format of network is input to convolutional neural networks and carries out modulation identification, amalgamation judging is carried out according to recognition result,
Or amalgamation judging is carried out according to confidence level;Specifically, including according to recognition result progress amalgamation judging:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ...,
M, M indicate that the convolutional neural networks support the modulation type number of identification;Step S2, by yi(m) convolutional Neural is inputted
Network is modulated type identification, obtains original recognition result oi, wherein oi∈[1,2,...,M];Step S3, if oi=j, then
By the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi(m) identification finishes;Step S4 is calculated
vjMaximum value, work as vjOnly there are one the modulation types that when maximum value, then represent the corresponding subscript j of maximum value as finally
Modulation identification result exports;Work as vjMaximum value when being more than two, then randomly select maximum value after a maximum value
The modulation type that corresponding subscript j is represented is exported as final modulation identification result;
Carrying out amalgamation judging according to confidence level includes:Step A1, by signal segment yi(m) convolutional neural networks are input to
It is modulated type identification, and obtains confidence level vector pi=[pi,1,pi,2,...,pi,M], wherein pi,jIt indicates signal segment yi
(m) it is identified as the confidence level of jth kind modulation type, j=1,2 ..., M, M indicate that the convolutional neural networks support the tune of identification
Number of types processed;Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,
2,...,M;Step A3 determines the average value wjMaximum value, work as wjIt is there are one only when maximum value, then maximum value is corresponding
The modulation type that subscript j is represented as final modulation identification result and exports;Work as wjMaximum value when being more than two,
After then randomly selecting a maximum value, the corresponding subscript j of the maximum value modulation types represented are known as final modulation type
Other result output.
It should be noted that the signal modulate device based on convolutional neural networks of the present embodiment is based on aforementioned
The signal modulate method of convolutional neural networks is corresponding, thus to the signal based on convolutional neural networks in the present embodiment
The content that Modulation Identification device does not describe can be found in the explanation in preceding method embodiment, and which is not described herein again.
One embodiment of the invention provides a kind of electronic equipment, and electronic equipment includes:Processor, and for storing
State the memory of processor-executable instruction.Wherein, the processor, for execute stored in memory correspond to it is aforementioned
Instruction in embodiment the step of signal modulate method based on convolutional neural networks.
In addition, the logical order in above-mentioned memory can be realized by the form of SFU software functional unit and be used as independence
Product sale or in use, can be stored in a computer read/write memory medium.
It should be understood by those skilled in the art that, the embodiment of the present invention can be provided as method, system or computer program
Product.Therefore, complete hardware embodiment, complete software embodiment or reality combining software and hardware aspects can be used in the present invention
Apply the form of example.Moreover, the present invention can be used in one or more wherein include computer usable program code computer
The form for the computer program product that usable storage medium is implemented.
It should be noted that the terms "include", "comprise" or its any other variant are intended to the packet of nonexcludability
Contain, so that the process, method, article or equipment including a series of elements includes not only those elements, but also includes
Other elements that are not explicitly listed, or further include for elements inherent to such a process, method, article, or device.
In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including the element
Process, method, article or equipment in there is also other identical elements.
In the specification of the present invention, numerous specific details are set forth.Although it is understood that the embodiment of the present invention can
To put into practice without these specific details.In some instances, well known method, structure and skill is not been shown in detail
Art, so as not to obscure the understanding of this description.Similarly, it should be understood that disclose in order to simplify the present invention and helps to understand respectively
One or more of a inventive aspect, in the above description of the exemplary embodiment of the present invention, each spy of the invention
Sign is grouped together into sometimes in single embodiment, figure or descriptions thereof.However, should not be by the method solution of the disclosure
It releases and is intended in reflection is following:The feature that i.e. the claimed invention requirement ratio is expressly recited in each claim is more
More features.More precisely, as the following claims reflect, inventive aspect is to be less than single reality disclosed above
Apply all features of example.Therefore, it then follows thus claims of specific implementation mode are expressly incorporated in the specific implementation mode,
Wherein each claim itself is as a separate embodiment of the present invention.
The above description is merely a specific embodiment, under the above-mentioned introduction of the present invention, those skilled in the art
Other improvement or deformation can be carried out on the basis of the above embodiments.It will be understood by those skilled in the art that above-mentioned tool
Body description only preferably explains that the purpose of the present invention, protection scope of the present invention are subject to the protection scope in claims.
Claims (10)
1. a kind of signal modulate method based on convolutional neural networks, which is characterized in that including:
Whether the length N and the input signal length L that convolutional neural networks are supported for comparing the signal received are equal;
When N is equal to L, directly the signal received is input in the convolutional neural networks, to obtain the convolution
The signal modulation style of neural network output;
When N is not equal to L, the convolution god is input to after the signal received is carried out corresponding polishing or segment processing
Through in network, to obtain the signal modulation style of the convolutional neural networks output.
2. according to the method described in claim 1, it is characterized in that, when N is not equal to L, the signal received is carried out corresponding
Polishing or segment processing after be input to the convolutional neural networks and include:
If N < L, according to the input signal length L that the convolutional neural networks are supported, zero padding is carried out to the signal received
Afterwards, it inputs a signal into the convolutional neural networks;
If N > L, the signal received is divided into length and is each signal segment of L, and each signal segment is input to the convolution
In neural network.
3. according to the method described in claim 2, it is characterized in that, after carrying out zero padding to the signal received, input a signal into
Include to the convolutional neural networks:
After N-L 0 is filled in the tail portion of the sample sequence x (n) of the signal received, by sample sequence x (n) (n=0,1,
2 ..., L-1) it is input in convolutional neural networks according to the input format of the convolutional neural networks to be modulated type knowledge
Not;
The signal received is divided into length and is each signal segment of L, and each signal segment is input to the convolutional neural networks
Include:
Sample sequence x (n) to the signal received chooses each signal segment y that length is L by interval P slidingsi(m), Jiang Gexin
Number section yi(m) it is input in convolutional neural networks according to the input format of convolutional neural networks to be modulated type identification;
Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+m), m=0,1,2 ..., L-1, i=1,2 ..., I,Indicate the maximum integer no more than (N-L+1)/P;1≤P≤(N-L+1).
4. according to the method described in claim 3, it is characterized in that, by each signal segment yi(m) according to the input of convolutional neural networks
After format is input to convolutional neural networks and carries out modulation identification, amalgamation judging is carried out according to recognition result, or according to setting
Reliability carries out amalgamation judging.
5. according to the method described in claim 4, it is characterized in that, including according to recognition result progress amalgamation judging:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ..., M, M
Indicate that the convolutional neural networks support the modulation type number of identification;
Step S2, by yi(m) convolutional neural networks are inputted and is modulated type identification, obtain original recognition result oi, wherein
oi∈[1,2,...,M];
Step S3, if oi=j, then by the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi(m)
Identification finishes;
Step S4 calculates vjMaximum value, work as vjOnly there are one modulation when maximum value, then represented the corresponding subscript j of maximum value
Type is exported as final modulation identification result;Work as vjMaximum value when being more than two, then randomly select one most
The corresponding subscript j of the maximum value modulation types represented are exported as final modulation identification result after big value.
6. according to the method described in claim 4, it is characterized in that, including according to confidence level progress amalgamation judging:
Step A1, by signal segment yi(m) convolutional neural networks are input to and are modulated type identification, and obtain confidence level to
Measure pi=[pi,1,pi,2,...,pi,M],
Wherein, pi,jIt indicates signal segment yi(m) it is identified as the confidence level of jth kind modulation type, j=1,2 ..., M, M indicate institute
State the modulation type number that convolutional neural networks support identification;
Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M;
Step A3 determines the average value wjMaximum value, work as wjThere are one only when maximum value, then by the corresponding subscript j of maximum value
The modulation type of representative is as final modulation identification result and exports;Work as wjMaximum value when being more than two, then with
After machine chooses a maximum value, the modulation type that the corresponding subscript j of maximum value is represented is as final modulation identification knot
Fruit exports.
7. a kind of signal modulate device based on convolutional neural networks, which is characterized in that including:
Comparison module, whether the input signal length L that the length N for comparing the signal received is supported with convolutional neural networks
It is equal;
Identification module, for when N is equal to L, directly the signal received to be input in the convolutional neural networks, with
Obtain the signal modulation style of the convolutional neural networks output;When N is not equal to L, the signal received is subjected to phase
It is input in the convolutional neural networks after the polishing or segment processing answered, to obtain the signal of the convolutional neural networks output
Modulation type.
8. device according to claim 7, which is characterized in that the identification module is specifically used for,
If N < L, according to the input signal length L that the convolutional neural networks are supported, zero padding is carried out to the signal received
Afterwards, it inputs a signal into the convolutional neural networks;If N > L, the signal received is divided into each letter that length is L
Number section, and each signal segment is input in the convolutional neural networks.
9. device according to claim 8, which is characterized in that the identification module is used for,
After N-L 0 is filled in the tail portion of the sample sequence x (n) of the signal received, by sample sequence x (n) (n=0,1,
2 ..., L-1) it is input in convolutional neural networks according to the input format of the convolutional neural networks to be modulated type knowledge
Not;
And the sample sequence x (n) for the signal to receiving, choose each signal segment y that length is L by interval P slidingsi
(m), by each signal segment yi(m) it is input in convolutional neural networks according to the input format of convolutional neural networks to be modulated class
Type identifies;Wherein, n=0,1,2 ..., N-1;yi(m)=x ((i-1) P+m), m=0,1,2 ..., L-1, i=1,2 ...,
I,Indicate the maximum integer no more than (N-L+1)/P;1≤P≤(N-L+1).
10. device according to claim 9, which is characterized in that the identification module is additionally operable to each signal segment yi(m) it presses
After being input to convolutional neural networks according to the input format of convolutional neural networks and carry out modulation identification, according to recognition result into
Row amalgamation judging, or amalgamation judging is carried out according to confidence level;
Specifically, including according to recognition result progress amalgamation judging:
Step S1, initializing signal section yi(m) it is identified as the number v of jth kind modulation typej=0, wherein j=1,2 ..., M, M
Indicate that the convolutional neural networks support the modulation type number of identification;
Step S2, by yi(m) convolutional neural networks are inputted and is modulated type identification, obtain original recognition result oi, wherein
oi∈[1,2,...,M];
Step S3, if oi=j, then by the number v of jth kind modulation typejAdd 1, repeats step S2 to S3 until signal segment yi(m)
Identification finishes;
Step S4 calculates vjMaximum value, work as vjOnly there are one modulation when maximum value, then represented the corresponding subscript j of maximum value
Type is exported as final modulation identification result;Work as vjMaximum value when being more than two, then randomly select one most
The corresponding subscript j of the maximum value modulation types represented are exported as final modulation identification result after big value;
Carrying out amalgamation judging according to confidence level includes:
Step A1, by signal segment yi(m) convolutional neural networks are input to and are modulated type identification, and obtain confidence level to
Measure pi=[pi,1,pi,2,...,pi,M],
Wherein, pi,jIt indicates signal segment yi(m) it is identified as the confidence level of jth kind modulation type, j=1,2 ..., M, M indicate institute
State the modulation type number that convolutional neural networks support identification;
Step A2 calculates the average value of the confidence level of jth kind modulation type, obtainsJ=1,2 ..., M;
Step A3 determines the average value wjMaximum value, work as wjThere are one only when maximum value, then by the corresponding subscript j of maximum value
The modulation type of representative is as final modulation identification result and exports;Work as wjMaximum value when being more than two, then with
After machine chooses a maximum value, the modulation type that the corresponding subscript j of maximum value is represented is as final modulation identification knot
Fruit exports.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810427110.4A CN108616471B (en) | 2018-05-07 | 2018-05-07 | Signal modulation identification method and device based on convolutional neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810427110.4A CN108616471B (en) | 2018-05-07 | 2018-05-07 | Signal modulation identification method and device based on convolutional neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108616471A true CN108616471A (en) | 2018-10-02 |
CN108616471B CN108616471B (en) | 2020-11-20 |
Family
ID=63662061
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810427110.4A Active CN108616471B (en) | 2018-05-07 | 2018-05-07 | Signal modulation identification method and device based on convolutional neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108616471B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109274624A (en) * | 2018-11-07 | 2019-01-25 | 中国电子科技集团公司第三十六研究所 | A kind of carrier frequency bias estimation based on convolutional neural networks |
CN109787929A (en) * | 2019-02-20 | 2019-05-21 | 深圳市宝链人工智能科技有限公司 | Signal modulate method, electronic device and computer readable storage medium |
CN110048980A (en) * | 2019-04-19 | 2019-07-23 | 中国电子科技集团公司第三十六研究所 | A kind of blind demodulation method of digital communication and device |
CN110266620A (en) * | 2019-07-08 | 2019-09-20 | 电子科技大学 | 3D MIMO-OFDM system channel estimation method based on convolutional neural networks |
CN112039820A (en) * | 2020-08-14 | 2020-12-04 | 哈尔滨工程大学 | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103067325A (en) * | 2013-01-31 | 2013-04-24 | 南京邮电大学 | Cooperative modulation identification method based on multi-class characteristic parameters and evidence theory |
CN103220241A (en) * | 2013-03-29 | 2013-07-24 | 南京信息职业技术学院 | Signal box dimension feature extraction method under low signal-to-noise ratio condition |
US9530400B2 (en) * | 2014-09-29 | 2016-12-27 | Nuance Communications, Inc. | System and method for compressed domain language identification |
CN106778594A (en) * | 2016-12-12 | 2017-05-31 | 燕山大学 | Mental imagery EEG signal identification method based on LMD entropys feature and LVQ neutral nets |
CN107808138A (en) * | 2017-10-31 | 2018-03-16 | 电子科技大学 | A kind of communication signal recognition method based on FasterR CNN |
-
2018
- 2018-05-07 CN CN201810427110.4A patent/CN108616471B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103067325A (en) * | 2013-01-31 | 2013-04-24 | 南京邮电大学 | Cooperative modulation identification method based on multi-class characteristic parameters and evidence theory |
CN103220241A (en) * | 2013-03-29 | 2013-07-24 | 南京信息职业技术学院 | Signal box dimension feature extraction method under low signal-to-noise ratio condition |
US9530400B2 (en) * | 2014-09-29 | 2016-12-27 | Nuance Communications, Inc. | System and method for compressed domain language identification |
CN106778594A (en) * | 2016-12-12 | 2017-05-31 | 燕山大学 | Mental imagery EEG signal identification method based on LMD entropys feature and LVQ neutral nets |
CN107808138A (en) * | 2017-10-31 | 2018-03-16 | 电子科技大学 | A kind of communication signal recognition method based on FasterR CNN |
Non-Patent Citations (2)
Title |
---|
NATHAN WEST: "Deep architectures for modulation recognition", 《IEEE INTERNATIONAL SYMPOSIUM ON DYNAMIC SPECTRUM ACCESS NETWORKS》 * |
杨发权: "无线通信信号调制识别关键技术与理论研究", 《中国优秀博士学位论文数据库》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109274624A (en) * | 2018-11-07 | 2019-01-25 | 中国电子科技集团公司第三十六研究所 | A kind of carrier frequency bias estimation based on convolutional neural networks |
CN109274624B (en) * | 2018-11-07 | 2021-04-27 | 中国电子科技集团公司第三十六研究所 | Carrier frequency offset estimation method based on convolutional neural network |
CN109787929A (en) * | 2019-02-20 | 2019-05-21 | 深圳市宝链人工智能科技有限公司 | Signal modulate method, electronic device and computer readable storage medium |
CN110048980A (en) * | 2019-04-19 | 2019-07-23 | 中国电子科技集团公司第三十六研究所 | A kind of blind demodulation method of digital communication and device |
CN110266620A (en) * | 2019-07-08 | 2019-09-20 | 电子科技大学 | 3D MIMO-OFDM system channel estimation method based on convolutional neural networks |
CN112039820A (en) * | 2020-08-14 | 2020-12-04 | 哈尔滨工程大学 | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network |
CN112039820B (en) * | 2020-08-14 | 2022-06-21 | 哈尔滨工程大学 | Communication signal modulation and identification method for quantum image group mechanism evolution BP neural network |
Also Published As
Publication number | Publication date |
---|---|
CN108616471B (en) | 2020-11-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108616471A (en) | A kind of signal modulate method and apparatus based on convolutional neural networks | |
CN109389078B (en) | Image segmentation method, corresponding device and electronic equipment | |
CN103679185B (en) | Convolutional neural networks classifier system, its training method, sorting technique and purposes | |
CN109086722B (en) | Hybrid license plate recognition method and device and electronic equipment | |
US9152888B2 (en) | System and method for automated object detection in an image | |
US20200279358A1 (en) | Method, device, and system for testing an image | |
US9239948B2 (en) | Feature descriptor for robust facial expression recognition | |
CN110222760B (en) | Quick image processing method based on winograd algorithm | |
CN109902723A (en) | Image processing method and device | |
CN110705585A (en) | Network fraud identification method and device, computer device and storage medium | |
CN107958247A (en) | Method and apparatus for facial image identification | |
CN107491674A (en) | Feature based information carries out the method and device of user's checking | |
CN111260220B (en) | Group control equipment identification method and device, electronic equipment and storage medium | |
CN113240127B (en) | Training method and device based on federal learning, electronic equipment and storage medium | |
CN110222330A (en) | Method for recognizing semantics and device, storage medium, computer equipment | |
US8571255B2 (en) | Scalable media fingerprint extraction | |
CN116978011B (en) | Image semantic communication method and system for intelligent target recognition | |
CN115941112B (en) | Portable hidden communication method, computer equipment and storage medium | |
CN117616424A (en) | Systems and methods for balancing sparsity in weights for accelerating deep neural networks | |
CN108230269A (en) | Grid method, device, equipment and storage medium are gone based on depth residual error network | |
Yu et al. | A multi-task learning CNN for image steganalysis | |
Qiu et al. | Deepsig: A hybrid heterogeneous deep learning framework for radio signal classification | |
CN110751490A (en) | Fraud identification method and device, electronic equipment and computer-readable storage medium | |
CN112102200B (en) | Image complement model initialization method, training method and image complement method | |
CN112733932A (en) | Model accelerated training method and device based on training data similarity aggregation |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |