CN113395225B - Universal intelligent processing method and device for directly receiving communication signal waveform to bit - Google Patents
Universal intelligent processing method and device for directly receiving communication signal waveform to bit Download PDFInfo
- Publication number
- CN113395225B CN113395225B CN202110648777.9A CN202110648777A CN113395225B CN 113395225 B CN113395225 B CN 113395225B CN 202110648777 A CN202110648777 A CN 202110648777A CN 113395225 B CN113395225 B CN 113395225B
- Authority
- CN
- China
- Prior art keywords
- signal
- sample data
- neural network
- demodulated
- bit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L27/00—Modulated-carrier systems
- H04L27/0008—Modulated-carrier systems arrangements for allowing a transmitter or receiver to use more than one type of modulation
-
- 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
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Digital Transmission Methods That Use Modulated Carrier Waves (AREA)
Abstract
The invention provides a general intelligent processing method and a general intelligent processing device for directly receiving a communication signal from a waveform to a bit, and relates to the field of communication signal demodulation processing. The method comprises the following steps: setting a specific signal modulation mode type to be demodulated, and acquiring data samples of digital signal waveforms of different modulation modes under high-power sampling; determining the maximum modulation order of a signal output by the neural network, and constructing a single neural network demodulation model for directly receiving a digital signal waveform; and performing hybrid training on the neural network demodulation model by using the acquired data samples, and performing independent testing on each modulation mode signal. And inputting the data sample of the signal to be demodulated into the trained demodulation neural network to realize the direct intelligent reception of digital signals of various different modulation modes. The method provided by the invention can realize the demodulation processing of digital signals with different modulation modes by only constructing a single neural network, directly recover corresponding bit information, and has better demodulation performance and stronger universality.
Description
Technical Field
The invention belongs to the technical field of digital signal demodulation, and particularly relates to a general intelligent processing method and a general intelligent processing device for directly receiving a communication signal from a waveform to a bit.
Background
Bit information recovery is a critical step in the reception of digital signals. In the demodulation process at the receiving end, in order to reduce the complexity of the design of the transceiver, the classical communication receiving method usually designs the modules corresponding to each parameter independently. In the conventional digital signal receiving method, a large number of algorithms are used to optimize each module in a receiving system such as modulation identification, timing synchronization, carrier synchronization, channel equalization, detection decision, or the like, or perform joint optimization on part of the modules, as shown in fig. 1. Meanwhile, in order to ensure the performance, a specific demodulation processing algorithm needs to be adopted for different modulation signals. The local optimal mode of the classical receiving system loses part of the coding modulation gain, and the globally optimal demodulation performance of a receiving end cannot be obtained. The application scene of the communication system in practical application is very complex, most of the existing receiving systems work in a mode of pre-optimization design, new optimization is difficult to automatically carry out along with the change of requirements or environments, and the performance is also lost. In addition, a large number of nonlinear factors exist in an actual communication system, the influence of the nonlinearity is difficult to be fully described by classical linearization processing or high-order Taylor expansion, and reasonable utilization of the nonlinear factors is lost.
The neural network has typical advantages in solving the non-linear problem due to the non-linear characteristic of the neural network. After certain data training, the neural network can accurately simulate any nonlinear function theoretically. At present, neural networks have been intensively studied in signal detection, signal modulation identification, interference suppression, target individual identification, channel estimation, signal demodulation, and the like.
In the aspect of signal demodulation by using a neural network, the former uses the neural network to demodulate digital signals and has the following problems: (1) In the prior art, a neural network is used for extracting features or replacing part of traditional functional modules and the like, but a single neural network is not used for carrying out integrated intelligent processing on the functional modules in demodulation, such as timing synchronization, carrier synchronization, channel equalization, judgment and the like; (2) Most of the existing methods assume that there are exactly integer sampling points in one symbol period, which implies that the timing synchronization of signals needs to be completed in advance. (3) The existing method only demodulates a specific signal of a certain modulation mode and cannot adapt to various different modulation modes.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a general intelligent processing method for directly receiving the waveform of a communication signal from a bit, and the demodulation processing of digital signals of various modulation modes can be realized without the aid of modules such as traditional signal synchronization, signal equalization, detection judgment and the like.
The invention also provides a general intelligent processing device for directly receiving the communication signal from the waveform to the bit, which is used for ensuring the realization and the application of the method in practice.
A general intelligent processing method for direct waveform-to-bit reception of communication signals, comprising the following steps:
acquiring a sample data set, wherein the sample data set comprises signal sample data of each preset modulation mode;
setting a bit label sequence for each signal sample data based on a preset signal maximum modulation order;
performing hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model;
and under the condition of acquiring a signal to be demodulated, demodulating the signal to be demodulated by applying the target neural network demodulation model to acquire bit information of the signal to be demodulated so as to complete the reception of the signal to be demodulated, wherein the signal to be demodulated is a signal obtained by applying any modulation mode.
Optionally, the above method, performing hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit tag sequence to obtain a target neural network demodulation model, includes:
inputting the signal sample data of each set bit label sequence into the initial neural network demodulation model to perform hybrid training on the initial neural network demodulation model;
testing the initial neural network demodulation model which is trained by applying a preset test data set, wherein the test data set comprises test sample data of each modulation mode;
and if the trained initial neural network demodulation models meet the test conditions corresponding to each modulation mode, taking the trained initial neural network demodulation models as target neural network demodulation models.
The method, optionally, may obtain a to-be-demodulated signal, including:
sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain the signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
Optionally, the above method, where each modulation scheme includes any multiple of the following: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, multilevel quadrature amplitude modulation MQAM, multilevel frequency shift keying MFSK, minimum frequency shift keying MSK, gaussian minimum frequency shift keying GMSK, offset quadrature phase shift keying OQPSK, orthogonal frequency division multiplexing OFDM.
Optionally, the above method, where the setting a bit label sequence for each signal sample data based on a preset maximum modulation order of the signal includes:
judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not, and obtaining a judgment result of each signal sample data;
for each signal sample data, if the judgment result of the signal sample data is negative, performing a complement operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; and if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as the bit label sequence of the signal sample data.
A general purpose intelligent processing apparatus for waveform-to-bit direct reception of a communication signal, comprising:
the system comprises a sample acquisition unit, a signal acquisition unit and a signal processing unit, wherein the sample acquisition unit is used for acquiring a sample data set which comprises signal sample data of each preset modulation mode;
the label setting unit is used for setting a bit label sequence for each signal sample data based on a preset signal maximum modulation order;
the model training unit is used for carrying out hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model;
and the signal demodulation unit is used for applying the target neural network demodulation model to demodulate the signal to be demodulated under the condition of acquiring the signal to be demodulated, so as to acquire bit information of the signal to be demodulated, and complete the reception of the signal to be demodulated, wherein the signal to be demodulated is a signal obtained by applying any modulation mode.
The above apparatus, optionally, the model training unit includes:
a training subunit, configured to input the signal sample data of each set bit label sequence to the initial neural network demodulation model, so as to perform hybrid training on the initial neural network demodulation model;
a testing subunit, configured to apply a preset test data set to test the initial neural network demodulation model that has been trained, where the test data set includes test sample data of each modulation mode;
and the execution subunit is used for taking the trained initial neural network demodulation model as a target neural network demodulation model if the trained initial neural network demodulation model meets the test condition corresponding to each modulation mode.
The above apparatus, optionally, the signal demodulation unit includes:
the demodulation method comprises the steps that a subunit is adopted, and the subunit is used for sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain a signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
Optionally, the above apparatus, where each modulation scheme includes any multiple of the following: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, multilevel quadrature amplitude modulation MQAM, multilevel frequency shift keying MFSK, minimum frequency shift keying MSK, gaussian minimum frequency shift keying GMSK, offset quadrature phase shift keying OQPSK, orthogonal frequency division multiplexing OFDM.
The above apparatus, optionally, the label setting unit includes:
the judging subunit is used for judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not to obtain a judgment result of each signal sample data;
a tag setting subunit, configured to, for each signal sample data, if the determination result of the signal sample data is negative, perform a padding operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; and if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as the bit label sequence of the signal sample data.
Compared with the prior art, the invention has the following advantages:
the method is different from the traditional receiver modularization processing method and various condition limiting modes existing in the demodulation of the existing neural network, the signal modulation mode and other related signal parameters are not needed to be known, and the received high-power sampling digital signal sample data is directly processed by using a single neural network, so that various functions of timing synchronization, carrier synchronization, channel equalization, judgment and the like in the digital signal receiving of different modulation modes can be realized, and the direct receiving of the communication signal waveform to bits is completed. The invention supports the demodulation processing of different digital modulation signals with transmission time delay, carrier frequency, phase offset, multipath fading and sampling frequency deviation, and has better demodulation performance and stronger universality.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
Fig. 1 is a diagram illustrating a signal demodulation process based on a conventional multi-module demodulation model according to an embodiment of the present invention;
FIG. 2 is a schematic flow chart of a general intelligent processing method for directly receiving a communication signal waveform to a bit according to the present invention;
FIG. 3 is a schematic flow chart of a training process of an initial neural network demodulation model according to the present invention;
FIG. 4 is a flow chart of a signal demodulation process by a single neural network demodulation model according to the present invention;
fig. 5 is a schematic structural diagram of an LSTM network provided in the present invention;
FIG. 6 is a schematic structural diagram of a bidirectional LSTM network provided by the present invention;
fig. 7 is a schematic structural diagram of a general intelligent processing apparatus for directly receiving a communication signal waveform to a bit according to the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In this application, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
The embodiment of the present invention provides a general intelligent processing method for directly receiving a communication signal waveform to a bit, which is applied to an electronic device, where the electronic device may include a processor, and a flowchart of the method is shown in fig. 2, and specifically includes:
s201: and acquiring a sample data set, wherein the sample data set comprises signal sample data of each preset modulation mode.
In the embodiment provided by the present invention, the sample data set includes sample data of each signal, where each preset modulation mode includes any of the following multiple types: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, multilevel quadrature amplitude modulation MQAM, multilevel frequency shift keying MFSK, minimum frequency shift keying MSK, gaussian minimum frequency shift keying GMSK, offset quadrature phase shift keying OQPSK, orthogonal frequency division multiplexing OFDM, etc.
In this embodiment of the present invention, the digital signals with different modulation schemes may be represented as:
wherein f is c Representing the carrier frequency, s T (t) represents a transmitted band-pass signal, s (t) = ∑ α k g T (t-kT B ) Indicating the base band signal of the transmitting end, alpha k Denotes the k-th transmitted channel symbol, g T (T) represents a shaping pulse waveform, T B Indicating the symbol period.
During transmission, the digital signal may be affected by a transmission channel and transmission equipment, such as gaussian noise pollution, non-gaussian noise pollution, transmission delay, carrier frequency deviation, sampling frequency deviation, phase offset, multipath fading, time jitter, and device nonlinear distortion. Thus, the received signal can be expressed as:
r(t)＝f[s T (t)]+z(t)
f [ ] represents the transmission channel and transmission equipment effects, and z (t) represents the superimposed noise pollution.
At a receiving end, sampling is carried out on the signal by utilizing analog-to-digital conversion equipment according to the requirement of a sampling theorem, and the sampling rate is greater than the modulation rate of the signal. Here, r (t) may be directly sampled, or first, down-conversion equipment is used to perform down-conversion processing on r (t), and then, a low intermediate frequency signal or a zero intermediate frequency signal output by a frequency converter is sampled, so as to obtain a sampled signal (signal sample data, test sample data, and a signal to be demodulated).
S202: and setting a bit label sequence for each signal sample data based on a preset maximum modulation order of the signal.
In the method provided by the embodiment of the invention, when a single neural network demodulation model is constructed, the maximum modulation order of a signal to be demodulated and processed by the network is firstly determined, and according to signals with different modulation orders, when a corresponding bit label sequence is constructed, a padding operation is carried out according to the form of the maximum modulation order.
S203: and performing hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model.
In the method provided by the embodiment of the present invention, the initial neural network demodulation model may be pre-constructed, wherein the initial neural network demodulation model may be a demodulation model of a single neural network.
In the embodiment of the invention, the target neural network demodulation model does not need the assistance of a traditional demodulation module, is a multifunctional system, has the functions of timing synchronization, carrier synchronization, channel equalization, detection judgment and the like in digital signal demodulation processing, and can be provided with no explicit independent functional module.
The input data format of the target neural network demodulation model can be a real number tensor of a digital signal waveform sample or a complex number tensor, and the output data format is a bit tensor corresponding to the signal sample.
Referring to fig. 3, a schematic flow chart of a training process of an initial neural network demodulation model according to an embodiment of the present invention is shown; after an initial neural network demodulation model is constructed, training the initial neural network demodulation by using a sample data set.
Inputting the signal sample data to the constructed initial neural network demodulation model for training until the loss function corresponding to the neural network demodulation model is converged; and testing by applying test sample data, storing the initial neural network model meeting the preset requirements, and determining the initial neural network model as a target neural network demodulation model which is tested.
Specifically, in the training process of the initial neural network demodulation model, network weight parameters can be initialized first, training samples are input into the initial neural network demodulation model for forward propagation calculation, a cost function is calculated based on an output result, the network weights are updated layer by layer in a backward propagation mode, specifically, a gradient is calculated, and the weight parameters are updated by a gradient descent method; and under the condition that the preset training stopping condition is not met, inputting a new training sample into the initial neural network demodulation model again until the initial neural network demodulation model meets the preset training stopping condition.
Specifically, the hybrid training refers to training a neural network by putting signal sample data of acquired digital signal waveforms of multiple different modulation modes together, and during training of the initial neural network demodulation model, data of training samples need to be randomly scrambled in sequence and sent into the initial neural network demodulation model in batch for training, so that it is ensured that training data used in updating weight parameters each time contains samples of different modulation modes.
Specifically, when testing the initial neural network demodulation model, it is necessary to test each modulation mode signal, and when each modulation mode signal reaches a set demodulation performance, it can be determined that the initial neural network demodulation model converges.
S204: and under the condition of acquiring a signal to be demodulated, demodulating the signal to be demodulated by applying the target neural network demodulation model to acquire bit information of the signal to be demodulated so as to complete the reception of the signal to be demodulated, wherein the signal to be demodulated is a signal obtained by applying any modulation mode.
In the method provided by the embodiment of the invention, the data of the signal to be demodulated is input into the neural network demodulation model, so that the bit information corresponding to the signal to be demodulated can be obtained, and the number of the bit information is determined by the length of the signal to be demodulated and the modulation mode.
As shown in fig. 4, a signal to be demodulated may be input to the target neural network demodulation model, the signal to be demodulated may be a high-power sampled digital signal waveform sample, and the model output is bit information corresponding to the digital signal waveform sample. Compared with the prior art, the general intelligent processing method for directly receiving the communication signal from the waveform to the bit can realize the demodulation processing of digital signals of various modulation modes only by applying a single neural network without using modules such as traditional signal synchronization, signal equalization, detection judgment and the like, and has better demodulation performance and stronger universality.
In an embodiment provided by the present invention, based on the foregoing implementation process, specifically, the performing hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit tag sequence to obtain the target neural network demodulation model includes:
inputting the signal sample data of each set bit label sequence into the initial neural network demodulation model to perform hybrid training on the initial neural network demodulation model;
testing the initial neural network demodulation model which is trained by applying a preset test data set, wherein the test data set comprises test sample data of each modulation mode;
and if the trained initial neural network demodulation models meet the test conditions corresponding to each modulation mode, taking the trained initial neural network demodulation models as target neural network demodulation models.
In an embodiment of the present invention, the test condition may be that an accuracy of the test result is greater than a preset accuracy threshold.
In the embodiment provided by the present invention, based on the implementation process, specifically, the process of acquiring the signal to be demodulated includes:
sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain the signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
In an embodiment provided by the present invention, based on the foregoing implementation process, specifically, the setting a bit label sequence for each signal sample data based on a preset maximum modulation order of a signal includes:
judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not, and obtaining a judgment result of each signal sample data;
for each signal sample data, if the judgment result of the signal sample data is negative, performing a complement operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; and if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as the bit label sequence of the signal sample data.
In an embodiment provided by the invention, the constructed neural network demodulation model can realize demodulation of digital signals of different modulation modes without prior information such as signal modulation modes and other signal parameters, and directly output bit information corresponding to digital signal samples, so that the neural network demodulation model has better demodulation performance and stronger universality.
In the practical application process, the embodiment of the invention can set the specific signal modulation mode type to be demodulated and obtain the data samples of the digital signal waveforms of different modulation modes under high-power sampling. Determining the maximum modulation order of a signal output by the neural network, and constructing a single neural network demodulation model for directly receiving a digital signal waveform; and performing hybrid training on the neural network demodulation model by using the acquired data samples, and performing independent test on each modulation mode signal to obtain a target neural network demodulation model. And inputting the data sample of the signal to be demodulated into the trained target neural network demodulation model to realize the direct intelligent reception of digital signals in various different modulation modes.
In the method provided in the embodiment of the present invention, based on the implementation process, specifically, the feasible method for obtaining the target neural network demodulation model specifically includes:
setting a specific signal modulation mode type to be demodulated, acquiring data samples of digital signal waveforms of different modulation modes under high-power sampling, constructing a bit label sequence corresponding to the data samples according to the maximum modulation order of the signal, and dividing the data samples into training samples and testing samples.
Constructing an initial neural network demodulation model, wherein the initial neural network demodulation model comprises an output layer and outputs bit information corresponding to input different modulation digital signal waveforms; the output layer has a corresponding loss function, and the training sample data contains a label value corresponding to the output layer;
randomly disordering the training sample data, and sending the training sample data into the initial neural network demodulation model in batch to train the initial neural network demodulation model until the loss function corresponding to the output layer of the initial neural network demodulation model is converged;
testing the initial neural network demodulation with the converged loss function based on the test sample data to obtain an evaluation index of an output layer of the initial neural network demodulation model;
and under the condition that the evaluation indexes all meet the corresponding evaluation conditions, determining the converged initial neural network demodulation model as the tested neural network demodulation model.
The neural network demodulation model can be constructed by using models such as a feed-Forward Neural Network (FNN), a stacked self-coding SAE network, a Convolutional Neural Network (CNN), a Recurrent Neural Network (RNN), an LSTM, a GRU, a transform, and improved models of these networks. The training data used for training the initial neural network demodulation model has the influence of carrier frequency deviation, timing deviation, multipath channel and other factors.
The construction of the demodulation neural network is described below by taking the LSTM network as an example. The schematic structural diagram of the LSTM network is shown in fig. 5, and the fitting performance of the network to the data in the form of a sequence is improved by introducing a cell state, a forgetting gate, an input gate, an output gate and the like.
Optionally, the initial neural network demodulation model may also be in a Bidirectional LSTM (Bidirectional LSTM) network form, as shown in fig. 6, which is a schematic structural diagram of a Bidirectional LSTM network, and the performance of the final output of the network is improved by performing joint processing on two output hidden states, namely forward and backward in the time dimension, and is suitable for the problem that the sequence has front and back correlation, so that the performance of the whole network can be improved in a basic network unit structure of the Bidirectional LSTM and by using a multi-layer network form, where a is a i And A' i Is a basic unit of LSTM networks.
For an output layer of the neural network demodulation model, a sigmoid activation function form can be adopted, a corresponding loss function adopts a binary classification cross entropy (BCE) form, and an evaluation index is detected bit accuracy, wherein a label of a training sample corresponding to the output layer can be a bit sequence, the bit sequence label is a two-dimensional matrix form, a first dimension is the number of symbols to be demodulated, a second dimension is the number of bits corresponding to a single symbol, and considering the condition that received signals are different modulation orders, when constructing a corresponding bit label sequence, all bit sequences in a training set need to be complemented according to the form of the maximum modulation order, and simultaneously, in the training process, an input mask sequence (valid information corresponds to 1, and invalid information corresponds to 0) is multiplied by corresponding output bits, so that the loss function corresponding to valid bits can be propagated backwards.
Wherein, assuming that x is a network input sample, y is a network output prediction value, a is a label value, and N is a sample scale, the loss function is as follows:
sigmoid activation function:
LSTM network calculation formula:
c k ＝f k c k-1 +i k z k
h k ＝o k tanh(c k )
in the training phase, a classical Adam optimizer can be adopted, and the learning rate method used is an update mode of the warmup. In the process of the super-parameter optimization, the fitting performance of the network can be improved by improving the scale of the LSTM layer unit, increasing the number of LSTM layers, adopting a weight initialization algorithm such as gloot and the like, reasonably setting the learning rate and the like. In order to reduce the degree of network overfitting, the performance of the test set can be improved in the modes of reasonably setting a dropout layer, introducing a batch normalization layer, introducing a regularization item and the like.
Corresponding to the method shown in fig. 1, an embodiment of the present invention further provides a general intelligent processing apparatus for directly receiving a communication signal waveform from a bit, which is used to implement the method shown in fig. 1 specifically, and the general intelligent processing apparatus provided in the embodiment of the present invention may be applied to an electronic device, and a schematic structural diagram of the general intelligent processing apparatus is shown in fig. 7, and specifically includes:
a sample obtaining unit 701, configured to obtain a sample data set, where the sample data set includes signal sample data of each preset modulation mode;
a tag setting unit 702, configured to set a bit tag sequence for each signal sample data based on a preset maximum modulation order of the signal;
a model training unit 703, configured to perform hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit tag sequence to obtain a target neural network demodulation model;
a signal demodulation unit 704, configured to, in a case that a signal to be demodulated is obtained, demodulate the signal to be demodulated by using the target neural network demodulation model to obtain bit information of the signal to be demodulated, so as to complete receiving of the signal to be demodulated, where the signal to be demodulated is a signal obtained by applying any one of the modulation manners.
In an embodiment provided by the present invention, based on the above scheme, optionally, the model training unit 703 includes:
a training subunit, configured to input the signal sample data of each set bit label sequence to the initial neural network demodulation model, so as to perform hybrid training on the initial neural network demodulation model;
a testing subunit, configured to apply a preset test data set to test the initial neural network demodulation model that has been trained, where the test data set includes test sample data of each modulation mode;
and the execution subunit is used for taking the trained initial neural network demodulation model as a target neural network demodulation model if the trained initial neural network demodulation model meets the test condition corresponding to each modulation mode.
In an embodiment provided in the present invention, based on the above scheme, optionally, the signal demodulating unit 704 includes:
the demodulation method comprises the steps that a subunit is adopted and used for sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain a signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
In an embodiment provided by the present invention, based on the above scheme, optionally, each modulation mode includes any of the following multiple modulation modes: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, multilevel quadrature amplitude modulation MQAM, multilevel frequency shift keying MFSK, minimum frequency shift keying MSK, gaussian minimum frequency shift keying GMSK, offset quadrature phase shift keying OQPSK, orthogonal frequency division multiplexing OFDM.
In an embodiment provided by the present invention, based on the above scheme, optionally, the tag setting unit 702 includes:
the judging subunit is used for judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not to obtain a judgment result of each signal sample data;
a tag setting subunit, configured to, for each signal sample data, if the determination result of the signal sample data is negative, perform a padding operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; and if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as the bit label sequence of the signal sample data.
The specific principle and the execution process of each module in the general intelligent processing device for directly receiving a communication signal waveform to a bit disclosed in the embodiment of the present invention are the same as the general intelligent processing method for directly receiving a communication signal waveform to a bit disclosed in the embodiment of the present invention, and reference may be made to the corresponding parts in the method provided in the embodiment of the present invention, which are not described herein again.
For the device-like embodiment, since it is basically similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising a … …" does not exclude the presence of another identical element in a process, method, article, or apparatus that comprises the element.
For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the units may be implemented in the same software and/or hardware or in a plurality of software and/or hardware when implementing the invention.
From the above description of the embodiments, it is clear to those skilled in the art that the present invention can be implemented by software plus necessary general hardware platform. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which may be stored in a storage medium, such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method according to the embodiments or some parts of the embodiments.
The general intelligent processing method for directly receiving a communication signal waveform to a bit provided by the invention is described in detail above, a specific example is applied in the text to explain the principle and the implementation of the invention, and the description of the above embodiment is only used to help understand the method of the invention and the core idea thereof; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (6)
1. A method for universal intelligent processing of communications signal waveform-to-bit direct reception, comprising:
acquiring a sample data set, wherein the sample data set comprises signal sample data of each preset modulation mode;
setting a bit label sequence for each signal sample data based on a preset signal maximum modulation order;
performing hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model;
under the condition of acquiring a signal to be demodulated, demodulating the signal to be demodulated by applying the target neural network demodulation model to acquire bit information of the signal to be demodulated so as to complete the reception of the signal to be demodulated, wherein the signal to be demodulated is a signal obtained by applying any modulation mode;
the setting a bit label sequence for each signal sample data based on a preset maximum modulation order of the signal comprises:
judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not, and obtaining a judgment result of each signal sample data;
for each signal sample data, if the judgment result of the signal sample data is negative, performing a complement operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as a bit label sequence of the signal sample data; the hybrid training of the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model comprises the following steps:
inputting the signal sample data of each set bit label sequence into the initial neural network demodulation model so as to perform hybrid training on the initial neural network demodulation model;
testing the trained initial neural network demodulation model by using a preset test data set, wherein the test data set contains test sample data of each modulation mode;
and if the trained initial neural network demodulation models meet the test conditions corresponding to each modulation mode, taking the trained initial neural network demodulation models as target neural network demodulation models.
2. The method of claim 1, wherein the process of acquiring the signal to be demodulated comprises:
sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain the signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
3. The method of claim 1, wherein each modulation scheme comprises any of: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, multilevel quadrature amplitude modulation MQAM, multilevel frequency shift keying MFSK, minimum frequency shift keying MSK, gaussian minimum frequency shift keying GMSK, offset quadrature phase shift keying OQPSK, orthogonal frequency division multiplexing OFDM.
4. A general-purpose intelligent processing apparatus for waveform-to-bit direct reception of communication signals, comprising:
the system comprises a sample acquisition unit, a signal acquisition unit and a signal processing unit, wherein the sample acquisition unit is used for acquiring a sample data set which comprises signal sample data of each preset modulation mode;
the label setting unit is used for setting a bit label sequence for each signal sample data based on a preset signal maximum modulation order;
the model training unit is used for carrying out hybrid training on the initial neural network demodulation model based on the signal sample data of each set bit label sequence to obtain a target neural network demodulation model;
the signal demodulation unit is used for applying the target neural network demodulation model to demodulate the signal to be demodulated under the condition of acquiring the signal to be demodulated to acquire bit information of the signal to be demodulated so as to complete the reception of the signal to be demodulated, wherein the signal to be demodulated is a signal obtained by applying any modulation mode;
the label setting unit includes:
the judging subunit is used for judging whether the modulation order corresponding to the modulation mode to which each signal sample data belongs is consistent with the maximum modulation order or not to obtain a judgment result of each signal sample data;
a tag setting subunit, configured to, for each signal sample data, if the determination result of the signal sample data is negative, perform a padding operation on bit information of the signal sample data to obtain a bit tag sequence of the signal sample data; if the judgment result of the signal sample data is yes, using the bit information of the signal sample data as a bit label sequence of the signal sample data;
the model training unit comprises:
a training subunit, configured to input the signal sample data of each set bit label sequence to the initial neural network demodulation model, so as to perform hybrid training on the initial neural network demodulation model;
a testing subunit, configured to apply a preset test data set to test the initial neural network demodulation model that has been trained, where the test data set includes test sample data of each modulation mode;
and the execution subunit is used for taking the trained initial neural network demodulation model as a target neural network demodulation model if the trained initial neural network demodulation model meets the test condition corresponding to each modulation mode.
5. The apparatus of claim 4, wherein the signal demodulation unit comprises:
the demodulation method comprises the steps that a subunit is adopted, and the subunit is used for sampling a received initial signal by using analog-to-digital conversion equipment based on a preset sampling mode to obtain a signal to be demodulated, wherein the sampling rate of the sampling mode is greater than the modulation rate of the initial signal; the signal to be demodulated is an intermediate frequency signal, a low intermediate frequency signal or a zero intermediate frequency signal.
6. The apparatus of claim 4, wherein each modulation scheme comprises any of: amplitude keying MASK, frequency shift keying MFSK, phase shift keying MPSK, M QAM, MFSK, MSK, GMSK, offset quadrature phase shift keying OQPSK, OFDM.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648777.9A CN113395225B (en) | 2021-06-10 | 2021-06-10 | Universal intelligent processing method and device for directly receiving communication signal waveform to bit |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110648777.9A CN113395225B (en) | 2021-06-10 | 2021-06-10 | Universal intelligent processing method and device for directly receiving communication signal waveform to bit |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113395225A CN113395225A (en) | 2021-09-14 |
| CN113395225B true CN113395225B (en) | 2023-01-31 |
Family
ID=77620337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110648777.9A Active CN113395225B (en) | 2021-06-10 | 2021-06-10 | Universal intelligent processing method and device for directly receiving communication signal waveform to bit |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113395225B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114222291B (en) * | 2021-12-01 | 2024-01-30 | 重庆邮电大学 | Wireless body area network data encryption method |
| CN115348139B (en) * | 2022-07-18 | 2024-01-05 | 中国人民解放军国防科技大学 | Modulation identification method based on cascade feature fusion and multi-stage classification |
| CN115834310B (en) * | 2023-02-15 | 2023-05-09 | 四川轻化工大学 | LGTransformer-based communication signal modulation identification method |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107147600A (en) * | 2017-04-28 | 2017-09-08 | 电子科技大学 | A kind of digital modulation signals demodulator and its demodulation method based on neutral net |
| CN109672639A (en) * | 2019-01-22 | 2019-04-23 | 哈尔滨工业大学 | A kind of signal demodulating method based on machine learning |
| CN110120926A (en) * | 2019-05-10 | 2019-08-13 | 哈尔滨工程大学 | Modulation mode of communication signal recognition methods based on evolution BP neural network |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150249554A1 (en) * | 2013-06-21 | 2015-09-03 | Dhadesugoor Vaman | Adaptive demodulation method and apparatus using an artificial neural network to improve data recovery in high speed channels |
| CN107124381B (en) * | 2017-03-27 | 2020-05-22 | 华南理工大学 | Automatic identification method for digital communication signal modulation mode |
| CN108234370B (en) * | 2017-12-22 | 2020-12-15 | 西安电子科技大学 | Communication signal modulation mode identification method based on convolutional neural network |
| US10944440B2 (en) * | 2018-04-11 | 2021-03-09 | Booz Allen Hamilton Inc. | System and method of processing a radio frequency signal with a neural network |
| CN109657604B (en) * | 2018-12-17 | 2021-02-12 | 中国人民解放军战略支援部队信息工程大学 | Satellite amplitude-phase signal identification and demodulation method and device based on cyclic neural network |
| 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 |
| CN110445741A (en) * | 2019-08-14 | 2019-11-12 | 北京中科晶上科技股份有限公司 | A kind of modulation mapping method of low storage overhead |
-
2021
- 2021-06-10 CN CN202110648777.9A patent/CN113395225B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107147600A (en) * | 2017-04-28 | 2017-09-08 | 电子科技大学 | A kind of digital modulation signals demodulator and its demodulation method based on neutral net |
| CN109672639A (en) * | 2019-01-22 | 2019-04-23 | 哈尔滨工业大学 | A kind of signal demodulating method based on machine learning |
| CN110120926A (en) * | 2019-05-10 | 2019-08-13 | 哈尔滨工程大学 | Modulation mode of communication signal recognition methods based on evolution BP neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113395225A (en) | 2021-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113395225B (en) | Universal intelligent processing method and device for directly receiving communication signal waveform to bit | |
| CN111404849B (en) | OFDM channel estimation and signal detection method based on deep learning | |
| CN112600772B (en) | OFDM channel estimation and signal detection method based on data-driven neural network | |
| JP2001522197A (en) | Composite equalization and detection method for multi-user signals | |
| CN113708855B (en) | OTFS data driving and receiving method, system and medium based on deep learning | |
| CN110690931B (en) | Digital signal adaptive code rate estimation method and device based on multi-wavelet-base combination | |
| CN112688772B (en) | Machine learning superimposed training sequence frame synchronization method | |
| CN115250216A (en) | Underwater sound OFDM combined channel estimation and signal detection method based on deep learning | |
| Park et al. | Deep learning-based automatic modulation classification with blind OFDM parameter estimation | |
| CN107566307A (en) | Blind equalizing apparatus and method, data modulation system and method | |
| Park et al. | Learning how to demodulate from few pilots via meta-learning | |
| Li et al. | IQ symbols processing schemes with LSTMs in OFDM system | |
| Zhao et al. | Multi-task learning based underwater acoustic OFDM communications | |
| CN112422208B (en) | Signal detection method based on antagonistic learning under unknown channel model | |
| CN110233808B (en) | FTN system signal detection method | |
| CN109818891B (en) | Lattice reduction assisted low-complexity greedy sphere decoding detection method | |
| CN112636855A (en) | OFDM signal detection method | |
| CN111865849B (en) | Signal modulation method and device and server | |
| CN113595680B (en) | Sparsity and step length adaptive matching tracking method based on generalized Jaccard coefficient | |
| Wang et al. | Modulation type recognition algorithm based on modulation instantaneous structure difference and deep learning | |
| CN114679363A (en) | Artificial intelligence assisted OFDM receiver offline learning method | |
| CN106302279A (en) | FBMC system equalization method based on interference variance statistics | |
| CN113472713A (en) | High-order modulation signal demodulation method based on neural network and receiver | |
| Chen et al. | Deep learning based single-channel blind separation of co-frequency modulated signals | |
| CN111917674A (en) | Modulation identification method based on deep learning |
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 |