A signal processing method and system based on deep learning

Abstract

The application discloses a kind of based on deep learning's signal processing method and system, belong to signal processing technical field, this method includes: by obtaining signal to be handled, and extract the signal feature of the signal to be handled, obtain the signal feature corresponding to the signal to be handled of the signal to be handled;Pre-trained deep learning model is dispatched to the signal feature to be handled and is identified, determines the modulation mode corresponding to the signal to be handled;With the modulation mode corresponding to the signal to be handled as basis, the signal to be handled is demodulated, obtains the processing result corresponding to the signal to be handled, signal is processed in conjunction with deep learning technology, can effectively cope with environmental interference, improve the performance of signal blind identification, to quickly determine signal modulation mode, finally improve the processing efficiency and accuracy to signal.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of signal processing, and particularly relates to a signal processing method and system based on deep learning. BACKGROUND

[0002] In a complex communication environment, especially in a scenario lacking prior information, blind signal identification is a challenging task. Traditional signal identification methods, such as methods based on matched filtering, feature parameter extraction (such as modulation index, bandwidth, power spectral density, etc.), usually rely on prior knowledge of signal types, and when the signal is affected by noise, multipath fading, frequency offset, etc., the identification performance will decrease significantly. In addition, with the rapid development of communication technology, new modulation methods, coding schemes and waveform designs emerge in an endless stream, making the robustness and adaptability of traditional methods face severe challenges. SUMMARY

[0003] The application provides a signal processing method and system based on deep learning, to solve the problem that the signal processing effect is poor due to the large influence of the environment on the modulation mode identification performance in the process of processing the communication signal in the prior art.

[0004] The first aspect of the application provides a signal processing method based on deep learning, comprising:

[0005] Obtaining a to-be-processed signal, and extracting the signal features of the to-be-processed signal to obtain the to-be-processed signal features corresponding to the to-be-processed signal;

[0006] Scheduling a pre-trained deep learning model to identify the to-be-processed signal features, and determining the modulation mode corresponding to the to-be-processed signal;

[0007] Based on the modulation mode corresponding to the to-be-processed signal, demodulating the to-be-processed signal to obtain the processing result corresponding to the to-be-processed signal.

[0008] In a possible implementation, the training method of the pre-trained deep learning model comprises:

[0009] Initializing the parameters of the deep learning model, and encoding the initialized parameters into a vector to determine the parameter code;

[0010] Obtaining a plurality of different parameter codes;

[0011] Performing neighborhood exploration on the parameter code to determine the parameter code after neighborhood exploration;

[0012] perform segmented information fusion exploration on the parameter coding after the neighborhood exploration to determine parameter coding after the segmented information fusion exploration;

[0013] perform fuzzy information learning exploration on the parameter coding after the segmented information fusion exploration to determine parameter coding after the fuzzy information learning exploration;

[0014] perform chaotic mapping exploration on the parameter coding after the fuzzy information learning exploration to determine parameter coding after the chaotic mapping exploration;

[0015] repeat the neighborhood exploration, the segmented information fusion exploration, the fuzzy information learning exploration and the chaotic mapping exploration until a maximum number of training times is reached, and obtain the target parameter coding according to the parameter coding after the chaotic mapping exploration in the last training process;

[0016] use the parameters of the target parameter coding as final parameters of a deep learning model to complete pre-training of the deep learning model.

[0017] In a possible implementation, the neighborhood exploration on the parameter coding to determine the parameter coding after the neighborhood exploration includes:

[0018] adjust the search step based on a preset step adjustment coefficient and the current training number to determine a current training step;

[0019] generate search values in multiple different directions based on the current training step;

[0020] perform neighborhood exploration on the parameter coding according to the search values in the multiple different directions to determine neighborhood exploration values in the multiple different directions;

[0021] determine the parameter coding after the neighborhood exploration based on the neighborhood exploration values in the multiple different directions.

[0022] In a possible implementation, the segmented information fusion exploration on the parameter coding after the neighborhood exploration to determine the parameter coding after the segmented information fusion exploration includes:

[0023] obtain a difference between the parameter coding after the neighborhood exploration and other parameter codings, and obtain an influence parameter corresponding to the parameter coding after the neighborhood exploration according to an absolute value of the difference;

[0024] perform segmented information fusion exploration on the parameter coding after the neighborhood exploration according to a Euclidean distance between the parameter coding after the neighborhood exploration and other parameter codings and the influence parameter corresponding to the parameter coding after the neighborhood exploration to obtain parameter coding after the segmented information fusion exploration.

[0025] In a possible implementation, the parameter coding of the segmented information fusion exploration is subjected to fuzzy information learning exploration, and the parameter coding after the fuzzy information learning exploration is determined, including:

[0026] An adaptability corresponding to the parameter coding of the segmented information fusion exploration is obtained, and a first adaptability is obtained.

[0027] According to the first adaptability, a membership degree of the parameter coding of the segmented information fusion exploration with respect to an optimal parameter coding is determined.

[0028] According to the membership degree, a fuzzy parameter factor is determined.

[0029] According to the fuzzy parameter factor, a first adaptive search control factor and a second adaptive search control factor are obtained.

[0030] According to the first adaptive search control factor and the second adaptive search control factor, the parameter coding of the segmented information fusion exploration is subjected to fuzzy information learning exploration, and the parameter coding after the fuzzy information learning exploration is obtained.

[0031] In a possible implementation, the parameter coding after the fuzzy information learning exploration is subjected to chaotic mapping exploration, and the parameter coding after the chaotic mapping exploration is determined, including:

[0032] According to the parameter coding after the fuzzy information learning exploration, a chaotic decision factor is determined.

[0033] According to the chaotic decision factor, the parameter coding after the fuzzy information learning exploration is subjected to cross mutation processing, and the parameter coding after the chaotic mapping exploration is obtained.

[0034] In a possible implementation, the target parameter coding is obtained according to the parameter coding after the chaotic mapping exploration in the last training process, including:

[0035] An adaptability corresponding to the parameter coding after the chaotic mapping exploration in the last training process is obtained, a second adaptability is obtained, and the parameter coding with the maximum second adaptability is determined as the target parameter coding.

[0036] In a possible implementation, the to-be-processed signal is one of an AM signal, an ASK signal, an FSK signal, a BPSK signal, a QPSK signal, or an MSK signal.

[0037] In a possible implementation, the signal feature of the to-be-processed signal is extracted, and the to-be-processed signal feature corresponding to the to-be-processed signal is obtained, including: a constellation feature of the to-be-processed signal is extracted, and the to-be-processed signal feature corresponding to the to-be-processed signal is obtained.

[0038] The second aspect of the present application provides a signal processing system based on deep learning, comprising: a feature extraction module, a feature recognition model, and a signal processing module.

[0039] The feature extraction module is configured to obtain a to-be-processed signal and extract a signal feature of the to-be-processed signal to obtain a to-be-processed signal feature corresponding to the to-be-processed signal.

[0040] The feature recognition model is configured to schedule a pre-trained deep learning model to recognize the to-be-processed signal feature and determine a modulation mode corresponding to the to-be-processed signal.

[0041] The signal processing module is configured to demodulate the to-be-processed signal based on the modulation mode corresponding to the to-be-processed signal to obtain a processing result corresponding to the to-be-processed signal.

[0042] The signal processing method and system based on deep learning provided by the present application can effectively cope with environmental interference and improve the performance of signal blind recognition by processing signals in combination with deep learning technology, so as to quickly determine the signal modulation mode and ultimately improve the processing efficiency and accuracy of signals. BRIEF DESCRIPTION OF DRAWINGS

[0043] The accompanying drawings, which are incorporated into and form part of the specification, illustrate embodiments consistent with the present application and, together with the specification, serve to explain the principles of the present application.

[0044] Figure 1 The flowchart of the signal processing method based on deep learning provided by the embodiment of the present application.

[0045] Figure 2 The structural schematic diagram of the signal processing system based on deep learning provided by the embodiment of the present application.

[0046] Among them, 201-feature extraction module, 202-feature recognition model, 203-signal processing module.

[0047] The above drawings have shown the specific embodiments of the present application, and will be described in more detail in the following. These drawings and textual descriptions are not intended to limit the scope of the concept of the present application by any means, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0048] The exemplary embodiments will be described in detail herein with reference to the attached drawings. The description herein refers to the accompanying drawings, which show by way of example specific embodiments. In the following description, same numbers refer to same or similar elements throughout the drawings. The embodiments described in the following exemplary embodiments are not meant to represent all embodiments consistent with the present application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the present application as detailed in the appended claims.

[0049] Embodiments of the present application will be described in detail herein with reference to the attached drawings.

[0050] As Figure 1 shown, an embodiment of the present application provides a signal processing method based on deep learning, comprising:

[0051] S101, obtaining a to-be-processed signal, and extracting a signal feature of the to-be-processed signal to obtain a to-be-processed signal feature corresponding to the to-be-processed signal;

[0052] S102, scheduling a pre-trained deep learning model to identify the to-be-processed signal feature, and determining a modulation mode corresponding to the to-be-processed signal;

[0053] S103, based on the modulation mode corresponding to the to-be-processed signal, demodulating the to-be-processed signal to obtain a processing result corresponding to the to-be-processed signal.

[0054] In a possible implementation, a training method corresponding to the pre-trained deep learning model comprises:

[0055] A1, initializing parameters of the deep learning model, and encoding the initialized parameters into a vector to determine a parameter code;

[0056] For example, the deep learning model can be set as a convolutional neural network. For the parameters of the convolutional neural network, generally weights and thresholds, and these weights and thresholds are often between (0, 1), so random initialization can be performed between the upper limit of the parameters and the lower limit of the parameters, and then the initialized parameters are encoded into a vector to obtain the parameter code.

[0057] A2, obtaining a plurality of different parameter codes;

[0058] A3, performing neighborhood exploration on the parameter codes to determine parameter codes after neighborhood exploration;

[0059] A4, performing segmented information fusion exploration on the parameter codes after neighborhood exploration to determine parameter codes after segmented information fusion exploration;

[0060] A5. Explore fuzzy information learning for the parameter coding of segmented information fusion exploration, and determine the parameter coding after fuzzy information learning exploration;

[0061] A6. Explore chaotic mapping of the parameter encoding after learning and exploring fuzzy information, and determine the parameter encoding after chaotic mapping exploration;

[0062] A7. Repeat the neighborhood exploration, segmented information fusion exploration, fuzzy information learning exploration, and chaotic mapping exploration until the maximum number of training iterations is reached. Obtain the target parameter code based on the parameter code after the chaotic mapping exploration in the last training process.

[0063] A8. Use the parameters encoded by the target parameters as the final parameters of the deep learning model to complete the pre-training of the deep learning model.

[0064] Optionally, after each exploration, out-of-bounds handling can be performed on the parameter encoding to ensure parameter validity.

[0065] In one possible implementation, the step of performing neighborhood exploration on the parameter encoding to determine the parameter encoding after neighborhood exploration includes:

[0066] A3.1. Based on the preset step size adjustment coefficient and combined with the current number of training iterations, adjust the search step size to determine the current training step size as follows: ;in, This represents the preset step size factor. This represents the current training step size in the t-th training process. This indicates the preset maximum number of training iterations. This represents the step size adjustment coefficient. Indicates the current number of training iterations;

[0067] A3.2 Based on the current training step size, generate search values ​​in multiple different directions: ;in, Represents the sine function. Represents pi (π). =1,2,…, h , h This indicates the total number of search directions.

[0068] A3.3. Based on the search values ​​in multiple different directions, perform neighborhood exploration on the parameter encoding to determine the neighborhood exploration values ​​in multiple different directions: ;in, Indicates the first t During the training process, the first i The first parameter encoding d Dimensional parameters, i= 1, 2, …, NP, NP represents total number of parameter encodings, d = 1, 2, …, D, D represents total dimension of parameters in parameter encoding, represents the i dimensional parameter corresponding to the p neighborhood search value of the d parameter encoding;

[0069] A3.4, determine the parameter encoding after neighborhood exploration based on the neighborhood search values in the plurality of different directions.

[0070] For example, the neighborhood search value with the maximum fitness can be determined as the parameter encoding after neighborhood exploration. The fitness can be obtained by taking the reciprocal of the loss function value, and in order to avoid a zero denominator, the loss function value corresponding to the parameter encoding can be added to a minimum constant term (such as 0.0001) before the reciprocal operation is performed. The method for obtaining the loss function value can include: obtaining a cross-entropy loss function value using historical signal features and corresponding modulation mode labels. In order to ensure the recognition rate, historical signal features and corresponding modulation mode labels under different environmental disturbances can be made to improve the environmental disturbance resistance.

[0071] By performing multi-directional search within the neighborhood range as described above, the neighborhood search efficiency can be effectively improved, and better solutions within the neighborhood range can be avoided.

[0072] In a possible implementation, the parameter encoding after neighborhood exploration is subjected to segmented information fusion exploration, and a parameter encoding after segmented information fusion exploration is determined, including:

[0073] A4.1, obtain the difference between the parameter encoding after neighborhood exploration and other parameter encodings, and obtain the influence parameter corresponding to the parameter encoding after neighborhood exploration according to the absolute value of the difference as follows: ; wherein, represents the j influence parameter of the parameter encoding after the k neighborhood exploration on the parameter encoding after the j = 1, 2, …, NP, represents the influence intensity parameter, and is set to 0.5; represents the influence scale parameter, and is set to 1.5; represents a natural constant, represents the t dimensional parameter of the parameter encoding after the j neighborhood exploration in the d training process, represents the t training process, kthe parameter coding after the neighborhood exploration d dimensional parameter, k =1,2,…,NP, and j is not equal to k | | represents the absolute value;

[0074] A4.2, according to the Euclidean distance between the parameter coding after the neighborhood exploration and other parameter codings and the influence parameter corresponding to the parameter coding after the neighborhood exploration, the parameter coding after the neighborhood exploration is segmented information fusion exploration, and the parameter coding after the segmented information fusion exploration is obtained as: ; wherein, represents the Euclidean distance between the parameter coding after the neighborhood exploration and the parameter coding after the neighborhood exploration, j represents the Euclidean distance between the parameter coding after the neighborhood exploration and the parameter coding after the neighborhood exploration, k represents the upper limit of the dimensional parameter, represents the lower limit of the dimensional parameter, d represents the dimensional parameter of the optimal parameter coding, represents the dimensional parameter of the parameter coding after the neighborhood exploration, d represents the dimensional parameter of the parameter coding after the neighborhood exploration, represents the dimensional parameter of the parameter coding after the neighborhood exploration, d represents the dimensional parameter of the optimal parameter coding, represents the dimensional parameter of the parameter coding after the neighborhood exploration, j represents the dimensional parameter of the parameter coding after the neighborhood exploration, d represents the dimensional parameter of the parameter coding after the neighborhood exploration, represents the segmented speed control parameter, and ; represents the upper limit of the segmented speed control parameter, and is set to 1; represents the lower limit of the segmented speed control parameter, and is set to 0.00001.

[0075] The early algorithm should be large-scale exploration for the entire search domain, and the later local exploration based on the optimal individual can avoid falling into the local optimal value. Through information fusion of parameter coding, the exploration of unknown domain can be effectively improved, and the global search ability can be improved.

[0076] In one possible implementation, the fuzzy information learning exploration of the parameter coding after the segmented information fusion exploration is determined, including:

[0077] A5.1, obtaining the fitness corresponding to the parameter coding after the segmented information fusion exploration, to obtain the first fitness;

[0078] A5.2, according to the first fitness, the membership degree of the parameter coding after the segmented information fusion exploration with respect to the optimal parameter coding is determined as: ; wherein, represents the Euclidean distance between the parameter coding after the neighborhood exploration and the parameter coding after the neighborhood exploration, ma membership of the parameter coding of the segmented information fusion exploration about the optimal parameter coding, denotes the maximum first fitness, denotes the minimum first fitness, denotes the first fitness of the parameter coding of the segmented information fusion exploration; m

[0079] A5.3, according to the membership, determining a fuzzification parameter factor as: ; wherein, denotes the fuzzification parameter factor, denotes an intermediate parameter, and , denotes an adjustable parameter factor, and is located between [0.8, 1);

[0080] A5.4, according to the fuzzification parameter factor, obtaining a first adaptive search control factor and a second adaptive search control factor as:

[0081]

[0082]

[0083] wherein, denotes the first adaptive search control factor, denotes the second adaptive search control factor, denotes a first random number which is randomly 0 or 1, denotes a second random number which is randomly 0 or 1;

[0084] A5.5, according to the first adaptive search control factor and the second adaptive search control factor, performing a fuzzification information learning exploration on the parameter coding of the segmented information fusion exploration, obtaining the parameter coding after the fuzzification information learning exploration as: ; wherein, denotes the first fitness of the parameter coding of the segmented information fusion exploration; t m denotes the first fitness of the parameter coding of the segmented information fusion exploration; d denotes the first fitness of the parameter coding of the segmented information fusion exploration; m d denotes the first fitness of the parameter coding of the segmented information fusion exploration; denotes the first fitness of the parameter coding of the segmented information fusion exploration. d

[0085] ​​​​​Through the exploration of the above fuzzy parameter factor control algorithm, with the increase of the iteration number of the algorithm, the absolute values of the first adaptive search control factor and the second adaptive search control factor gradually decrease. In the initial stage of the algorithm operation, by setting a larger particle step, extensive and comprehensive exploration can be carried out. Such a large-scale search strategy helps the algorithm to quickly identify the potential optimal solution region, avoid falling into local optimum, and significantly improve the global surveying ability. As the algorithm enters the later stage, the requirement for solution accuracy gradually increases. By reducing the particle step, the algorithm can carry out more detailed and accurate search near the identified potential optimal solution region. Such a small-scale search strategy helps the algorithm to further optimize the solution quality in detail, improve the solution accuracy, and ensure that the final solution is as close as possible to or reaches the global optimum.

[0086] In a possible implementation, the parameter coding after the fuzzy information learning exploration is chaotically mapped for exploration, and the parameter coding after chaotic mapping exploration is determined, including:

[0087] A6.1, according to the parameter coding after the fuzzy information learning exploration, the chaotic decision factor is determined as: ; wherein, represents the first parameter of the parameter coding after the fuzzy information learning exploration in the i th training process, t represents the i th parameter of the parameter coding after the fuzzy information learning exploration in the j th training process, n represents the i th parameter of the parameter coding after the fuzzy information learning exploration in the j th training process, d represents the i th parameter of the parameter coding after the fuzzy information learning exploration in the j th training process. represents the corresponding chaotic decision factor, represents the corresponding chaotic decision factor, represents the inverse tangent function;

[0088] A6.2, according to the chaotic decision factor, the parameter coding after the fuzzy information learning exploration is cross-variation processed, and the parameter coding after chaotic mapping exploration is obtained as:

[0089]

[0090] wherein, represents the first random number between (0, 1), represents the i th parameter of the parameter coding after the chaotic mapping exploration in the j th training process, n represents the i th parameter of the parameter coding after the chaotic mapping exploration in the j th training process, d represents the i th parameter of the parameter coding after the chaotic mapping exploration in the j th training process.

[0091] Through the above cross-variation processing, the global search ability of the algorithm can be effectively improved, and the algorithm can be prevented from falling into local optimum. In addition, the cross-variation processing can be controlled by a greedy strategy to ensure the training speed of the algorithm.

[0092] In a possible implementation, the parameter code after the chaos mapping exploration in the last training process is obtained, including: obtaining the fitness corresponding to the parameter code after the chaos mapping exploration in the last training process, obtaining a second fitness, and determining the parameter code with the maximum second fitness as the target parameter code.

[0093] Through the mutual cooperation of the above-mentioned several exploration modes, the global search ability and the local search ability of the algorithm can be effectively improved, the training effect of the deep learning model can be improved, so that the deep learning model after training can more accurately identify the modulation mode of the signal to be processed.

[0094] In a possible implementation, the signal to be processed is one of an AM signal (amplitude modulation signal), an ASK signal (amplitude shift keying signal), an FSK signal (frequency shift keying signal), a BPSK signal (binary phase shift keying signal), a QPSK signal (quadrature phase shift keying signal), or an MSK signal (minimum shift keying signal).

[0095] It should be noted that the above-mentioned signal types of the signal to be processed are only examples of embodiments of the present application, and other signals can also be identified and processed.

[0096] In a possible implementation, the signal feature of the signal to be processed is extracted to obtain the signal feature corresponding to the signal to be processed, including: extracting the constellation feature of the signal to be processed to obtain the signal feature corresponding to the signal to be processed.

[0097] It should be noted that the constellation feature is only a preferred embodiment provided by the present application, and other time domain features or frequency domain features of the signal to be processed can also be used as the signal feature of the signal to be processed.

[0098] The present application provides a signal processing method based on deep learning, which acquires a signal to be processed, extracts a signal feature of the signal to be processed, and obtains a signal feature corresponding to the signal to be processed; a pre-trained deep learning model is dispatched to identify the signal feature, and a modulation mode corresponding to the signal to be processed is determined; the signal to be processed is demodulated based on the modulation mode corresponding to the signal to be processed, and a processing result corresponding to the signal to be processed is obtained. The signal is processed by combining the deep learning technology, which can effectively cope with environmental interference, improve the performance of signal blind identification, quickly determine the signal modulation mode, and ultimately improve the processing efficiency and accuracy of the signal.

[0099] Based on the same inventive concept, the embodiments of the present application also provide a signal processing system based on deep learning, which comprises: a feature extraction module 201, a feature recognition model 202, and a signal processing module 203.

[0100] The feature extraction module 201 is configured to acquire a to-be-processed signal and extract a signal feature of the to-be-processed signal, to obtain a to-be-processed signal feature corresponding to the to-be-processed signal.

[0101] The feature recognition model 202 is configured to schedule a pre-trained deep learning model to recognize the to-be-processed signal feature, to determine a modulation mode corresponding to the to-be-processed signal.

[0102] The signal processing module 203 is configured to demodulate the to-be-processed signal based on the modulation mode corresponding to the to-be-processed signal, to obtain a processing result corresponding to the to-be-processed signal.

[0103] The above-described signal processing system based on deep learning can execute the above-described method technical solution, and has similar principles and beneficial effects, which will not be described herein again.

[0104] Those skilled in the art should understand that embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can adopt a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can adopt a computer program product implemented on one or more computer usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer usable program codes.

[0105] The present application is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate a means for implementing the functions specified in the flowcharts and / or block diagrams. Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The function specified in one flow or multiple flows and / or blocks

[0106] These computer program instructions can also be stored in a computer readable storage medium capable of guiding a computer or other programmable data processing devices to work in a specific manner, so that the instructions stored in the computer readable storage medium produce a product including instruction means, which implements the functions specified in the flowcharts and / or block diagrams. Figure 1 The function specified in one flow or multiple flows and / or blocks Figure 1 The function specified in one flow or multiple flows and / or blocks

[0107] These computer program instructions can also be loaded into a computer or other programmable data processing device, so that a series of operational steps are performed on the computer or other programmable data processing device to generate a computer implemented process, so that the instructions executed on the computer or other programmable data processing device provide a process for implementing the functions specified in the flowchart Figure 1 one flow or multiple flows and / or the functions specified in the block Figure 1 one block or multiple blocks.

[0108] Those of ordinary skill in the art can understand that all or part of the steps of the above-mentioned facts and methods can be completed by programs instructing relevant hardware, and the programs involved or the programs mentioned can be stored in a computer-readable storage medium. When the program is executed, the following steps are included: at this time, the corresponding method steps are derived, and the storage medium can be ROM / RAM, a magnetic disc, an optical disc, etc.

[0109] The above specific embodiments further illustrate the purposes, technical solutions and beneficial effects of the present application. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A deep learning-based signal processing method, characterized by, The method comprises the following steps: acquiring a to-be-processed signal and extracting a signal feature of the to-be-processed signal to obtain a to-be-processed signal feature corresponding to the to-be-processed signal; scheduling a pre-trained deep learning model to recognize the to-be-processed signal feature to determine a modulation mode corresponding to the to-be-processed signal; based on the modulation mode corresponding to the to-be-processed signal, demodulating the to-be-processed signal to obtain a processing result corresponding to the to-be-processed signal; a training method corresponding to the pre-trained deep learning model comprises: initializing parameters of the deep learning model and encoding the initialized parameters into a vector to determine parameter encoding; acquiring a plurality of different parameter encodings; performing neighborhood exploration on the parameter encoding to determine parameter encoding after neighborhood exploration; performing segmented information fusion exploration on the parameter encoding after neighborhood exploration to determine parameter encoding after segmented information fusion exploration; performing fuzzy information learning exploration on the parameter encoding of segmented information fusion exploration to determine parameter encoding after fuzzy information learning exploration; performing chaotic mapping exploration on the parameter encoding after fuzzy information learning exploration to determine parameter encoding after chaotic mapping exploration; repeating the neighborhood exploration, segmented information fusion exploration, fuzzy information learning exploration and chaotic mapping exploration until the maximum number of training is reached, and acquiring target parameter encoding according to the parameter encoding after chaotic mapping exploration in the last training process; using the parameters of the target parameter encoding as the final parameters of the deep learning model to complete the pre-training of the deep learning model; the neighborhood exploration on the parameter encoding to determine the parameter encoding after neighborhood exploration comprises: based on a preset step length adjustment coefficient, adjusting the search step length according to the current training number to determine the current training step length; based on the current training step length, generating search values in multiple different directions; based on the search values in multiple different directions, performing neighborhood exploration on the parameter encoding to determine neighborhood exploration values in multiple different directions; based on the neighborhood exploration values in multiple different directions, determining the parameter encoding after neighborhood exploration; the segmented information fusion exploration on the parameter encoding after neighborhood exploration to determine the parameter encoding after segmented information fusion exploration comprises: acquiring the difference between the parameter encoding after neighborhood exploration and other parameter encodings, and acquiring the influence parameter corresponding to the parameter encoding after neighborhood exploration according to the absolute value of the difference; based on the Euclidean distance between the parameter encoding after neighborhood exploration and other parameter encodings and the influence parameter corresponding to the parameter encoding after neighborhood exploration, performing segmented information fusion exploration on the parameter encoding after neighborhood exploration to obtain the parameter encoding after segmented information fusion exploration; the fuzzy information learning exploration on the parameter encoding of segmented information fusion exploration to determine the parameter encoding after fuzzy information learning exploration comprises: acquiring the fitness corresponding to the parameter encoding of segmented information fusion exploration to obtain a first fitness; According to the first fitness, membership of the parameter coding of the segmented information fusion exploration with respect to optimal parameter coding is determined; According to the membership, a fuzzification parameter factor is determined; According to the fuzzification parameter factor, a first adaptive search control factor and a second adaptive search control factor are obtained; According to the first adaptive search control factor and the second adaptive search control factor, the parameter coding of the segmented information fusion exploration is subjected to fuzzy information learning exploration, and the parameter coding after the fuzzy information learning exploration is obtained; The chaotic mapping exploration of the parameter coding after the fuzzy information learning exploration is performed to determine the parameter coding after the chaotic mapping exploration, including: According to the parameter coding after the fuzzy information learning exploration, a chaotic decision factor is determined; According to the chaotic decision factor, the parameter coding after the fuzzy information learning exploration is subjected to cross mutation processing to obtain the parameter coding after the chaotic mapping exploration; The target parameter coding is obtained according to the parameter coding after the chaotic mapping exploration in the last training process, including: The fitness corresponding to the parameter coding after the chaotic mapping exploration in the last training process is obtained to obtain a second fitness, and the parameter coding with the maximum second fitness is determined as the target parameter coding. 2.The deep learning based signal processing method of claim 1, wherein, The to-be-processed signal is one of an AM signal, an ASK signal, an FSK signal, a BPSK signal, a QPSK signal or an MSK signal.

3. The deep learning based signal processing method of claim 2, wherein, The signal feature of the to-be-processed signal is extracted to obtain the to-be-processed signal feature corresponding to the to-be-processed signal, including: the constellation feature of the to-be-processed signal is extracted to obtain the to-be-processed signal feature corresponding to the to-be-processed signal.

4. A deep learning-based signal processing system capable of performing the deep learning-based signal processing method according to any one of claims 1 to 3, characterized by Including: a feature extraction module, a feature recognition model and a signal processing module; The feature extraction module is configured to obtain a to-be-processed signal and extract a signal feature of the to-be-processed signal to obtain a to-be-processed signal feature corresponding to the to-be-processed signal; The feature recognition model is configured to schedule a pre-trained deep learning model to recognize the to-be-processed signal feature to determine a modulation mode corresponding to the to-be-processed signal; The signal processing module is configured to demodulate the to-be-processed signal based on the modulation mode corresponding to the to-be-processed signal to obtain a processing result corresponding to the to-be-processed signal.

Images