Pulse convolutional neural network-based modulation identification method, device and equipment

Abstract

The invention discloses a modulation identification method, device and equipment based on a pulse convolutional neural network, and relates to the technical field of wireless communication, and the method comprises the steps: connecting and combining a plurality of pulse residual blocks fusing an attention mechanism, and constructing a convolutional layer and a full connection layer of the pulse residual blocks through LIF neurons; based on the pulse residual blocks of the connection combination, creating a deep network model which takes a pulse residual neural network as a backbone network framework and is fused with an attention mechanism; and obtaining an IQ signal data set to train the created deep network model, thereby realizing modulation identification classification of the IQ signal data to be identified based on the trained deep network model. The wireless signals are modulated and identified through the improved deep network model, the effects of improving the identification speed and reducing the operation power consumption are achieved, and improvement of the performance of a wireless communication system is facilitated.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to a modulation recognition method, device and equipment based on a pulse convolutional neural network. Background Art

[0002] Rapid and accurate modulation recognition of electromagnetic signals in complex electromagnetic space is an important research topic in the field of wireless communication, such as information reconnaissance, electronic countermeasures, and electromagnetic spectrum monitoring. Among them, the common method of modulation recognition is to perform feature analysis and extraction on the IQ (In-phase / Quadrature) data of the target signal, which is used as the basis for modulation recognition, and a specially designed classifier is used to give the processing result. With the rapid development of various communication technologies, the number of wireless communication devices and the types of communication signals have increased rapidly, and the density of communication signals in the electromagnetic spectrum has continued to increase, resulting in the existing modulation recognition technology being unable to cope with the increasingly complex electromagnetic environment. It is more urgent to study modulation recognition methods with better performance.

[0003] With the rise of deep learning technology, many studies have applied various deep learning algorithms and models to the field of modulation recognition and have achieved a series of outstanding results. However, the nonlinear operations such as sigmoid and atan, which are widely used and necessary in deep learning, have the problems of complex implementation, high computing resources and high energy consumption in the actual wireless communication system embedded devices, which is not conducive to the further improvement of wireless communication system performance and the reduction of operating energy consumption. Therefore, how to effectively implement modulation recognition based on deep learning has become a problem that needs to be solved urgently. Summary of the invention

[0004] The present application provides a modulation recognition method, device and equipment based on a pulse convolutional neural network, which performs modulation recognition on wireless signals through an improved deep network model, thereby achieving the effect of increasing recognition speed and reducing computing power consumption, and helping to improve the performance of wireless communication systems.

[0005] In a first aspect, an embodiment of the present application provides a modulation recognition method based on a pulse convolutional neural network, and the modulation recognition method based on a pulse convolutional neural network includes:

[0006] Combine multiple pulse residual blocks with fused attention mechanisms, and construct convolutional layers and fully connected layers of the pulse residual blocks through LIF neurons;

[0007] Based on the connected combined pulse residual block, a deep network model with a pulse residual neural network as the backbone network framework and an attention mechanism is created;

[0008] An IQ signal data set is obtained to train the created deep network model, so as to realize modulation recognition and classification of the IQ signal data to be recognized based on the trained deep network model.

[0009] In combination with the first aspect, in one implementation, multiple pulse residual blocks in the deep network model are connected in sequence.

[0010] In combination with the first aspect, in one implementation, the deep network model further includes a normalization layer and a fully connected layer, and both the normalization layer and the fully connected layer of the deep network model are composed of LIF neurons.

[0011] In combination with the first aspect, in one embodiment, for a LIF neuron, the LIF model is used to describe the relationship between the neuron membrane potential and the neuron input current, and the change process of the neuron membrane potential is specifically expressed as follows:

[0012]

[0013] Among them, τ m is the membrane time constant, R m represents resistance, V represents neuron membrane potential, I represents the sum of synaptic currents generated by the discharge of each presynaptic neuron, and V rest Represents the resting membrane potential.

[0014] In combination with the first aspect, in one embodiment, the change process of the neuron membrane potential can also be achieved by a differential method, specifically:

[0015]

[0016] Among them, V(t) represents the membrane potential voltage of the LIF neuron, X(t) represents the sum of the pulses received by the LIF neuron at time t, and when the membrane potential voltage of the LIF neuron exceeds the set threshold, the LIF neuron emits a pulse, and then the neuron membrane potential returns to the zero potential resting state.

[0017] In combination with the first aspect, in one implementation,

[0018] The pulse residual block includes a channel attention mechanism for learning the relationship between features;

[0019] The channel attention mechanism is specifically expressed as:

[0020]

[0021] h c =α c ×h c

[0022] Among them, αc represents the attention weight of channel c, f c represents the attention function of channel c, C represents the total number of channels, and h c represents the input vector of channel c, and exp represents the natural exponential function.

[0023] In combination with the first aspect, in one implementation, for the channel attention mechanism, specific implementation steps include:

[0024] For each channel, global average pooling is performed, and each channel obtains 1 scalar, and C channels obtain C scalars;

[0025] After passing through the fully connected layer, ReLU activation function, fully connected layer, and Sigmoid activation function, C scalars between 0 and 1 are obtained as the channel weights;

[0026] Each channel is weighted by using the corresponding weight to obtain a new weighted feature.

[0027] In a second aspect, an embodiment of the present application provides a modulation recognition device based on a pulse convolutional neural network, and the modulation recognition device based on a pulse convolutional neural network includes:

[0028] A building module, which is used to connect and combine multiple pulse residual blocks fused with attention mechanisms, and construct the convolutional layer and fully connected layer of the pulse residual block through LIF neurons;

[0029] A creation module is used to create a deep network model based on the pulse residual block of the connection combination, with the pulse residual neural network as the backbone network framework and integrating the attention mechanism;

[0030] An execution module is used to obtain an IQ signal data set to train the created deep network model, thereby realizing modulation recognition and classification of the IQ signal data to be recognized based on the trained deep network model.

[0031] In conjunction with the second aspect, in one implementation,

[0032] Multiple impulse residual blocks in the deep network model are connected in sequence;

[0033] The deep network model also includes a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer of the deep network model are both composed of LIF neurons.

[0034] In the third aspect, the embodiment of the present application provides that the modulation recognition device based on the pulse convolutional neural network includes a processor, a memory, and a modulation recognition program based on the pulse convolutional neural network stored in the memory and executable by the processor, wherein when the modulation recognition program based on the pulse convolutional neural network is executed by the processor, the steps of the modulation recognition method based on the pulse convolutional neural network described above are implemented.

[0035] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0036] By connecting and combining multiple pulse residual blocks that integrate attention mechanisms, and constructing convolutional layers and fully connected layers of pulse residual blocks through LIF neurons, then based on the connected and combined pulse residual blocks, a deep network model with a pulse residual neural network as the backbone network framework and integrating an attention mechanism is created, and finally an IQ signal data set is obtained to train the created deep network model, so as to realize modulation recognition and classification of the IQ signal data to be identified based on the trained deep network model, SNN is applied to modulation recognition, combined with deep learning, and trained on the IQ signal data set to establish a network model with a pulse residual neural network as the backbone network framework and integrating an attention mechanism. Compared with traditional deep learning models, the application of SNN in this application will reduce complex nonlinear operations such as sigmoid and atan, thereby reducing power consumption while improving the computing performance of wireless communication systems, providing new ideas and methods for the application of deep learning in modulation recognition, achieving the effect of improving recognition speed and reducing computing power consumption, and contributing to improving the performance of wireless communication systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a flow chart of a modulation recognition method based on a pulse convolutional neural network in this application;

[0038] Figure 2 This is a schematic diagram of the structure of the deep network model of this application;

[0039] Figure 3 Schematic diagram of the implementation of the channel attention mechanism;

[0040] Figure 4 This is a schematic diagram of the improved pulse residual block of this application;

[0041] Figure 5 This is a schematic diagram of the functional modules of the modulation recognition device based on the pulse convolutional neural network of the present application;

[0042] Figure 6 This is a schematic diagram of the hardware structure of the modulation recognition device based on the pulse convolutional neural network in this application. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0044] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0045] On the first aspect, the embodiments of the present application provide a modulation recognition method based on a pulse convolutional neural network, which performs modulation recognition on wireless signals through an improved deep network model, thereby achieving the effect of increasing recognition speed and reducing computing power consumption, thereby helping to improve the performance of wireless communication systems.

[0046] In one embodiment, referring to Figure 1 , Figure 1 Schematic diagram of the flow of the modulation recognition method based on pulse convolutional neural network in this application. Figure 1 As shown, the modulation recognition method based on pulse convolutional neural network includes:

[0047] S1: Combine multiple pulse residual blocks with fused attention mechanisms, and construct the convolutional layer and fully connected layer of the pulse residual block through LIF neurons;

[0048] S2: Based on the connected combined pulse residual block, a deep network model is created with the pulse residual neural network as the backbone network framework and integrated with the attention mechanism;

[0049] This application combines SNN (Spiking Neural Network) with the attention mechanism to construct a deep network model with a deep pulse residual network as the backbone. The constructed deep network model is as follows Figure 2 As shown, based on Figure 2As shown in the figure, the deep network model is realized by connecting and combining multiple pulse residual blocks with fusion attention mechanism. Unlike the common residual network, the convolutional layer and fully connected layer in the pulse residual block are composed of LIF (Leakin Integrated-and-fire neuron). The pulse residual block with fusion attention mechanism further integrates the advantages of CNN (Convolutional Neural Networks) and SNN. The calculation process is more in line with the information processing method of practical applications to achieve modulation recognition under high noise background. The pulse neural network is the third generation of neural network after the artificial neural network. Its information processing method is similar to that of biological neurons. It uses discrete action potentials (i.e. pulses) to represent and transmit information. It has the advantages of high biological credibility and low energy consumption, and has broad application prospects.

[0050] For further information, see Figure 2 As shown in the figure, multiple pulse residual blocks in the deep network model are connected in sequence; the deep network model also includes a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer of the deep network model are both composed of LIF neurons. LIF neurons are widely used in SNNs and are currently generally recognized as the pulse neuron model that achieves the best balance between biological interpretability and implementation complexity. During the "integration" process, neurons will accumulate input pulses over time, and the threshold judgment will determine whether to emit pulses. If the threshold is not reached and there is no new pulse accumulation, the neuron will gradually "leak" the existing accumulation to the resting state. After the threshold is reached, the neuron will enter a short refractory period after emitting pulses. During the refractory period, the neuron cannot emit pulses, and the "integration" will be restarted after the resting state is restored. Figure 2 In the figure, BN represents the normalization layer and FC represents the fully connected layer.

[0051] Furthermore, for LIF neurons, the LIF model is used to describe the relationship between neuron membrane potential and neuron input current, and the change process of neuron membrane potential is specifically expressed as:

[0052]

[0053] Among them, τ m is the membrane time constant, R m represents resistance, V represents neuron membrane potential, I represents the sum of synaptic currents generated by the discharge of each presynaptic neuron, dt represents the integral over time t, V rest Represents the resting membrane potential, usually 0.

[0054] That is, the LIF model intuitively describes the relationship between neuronal membrane potential and neuronal input current, and the change process of neuronal membrane potential is described by the above first-order differential equation.

[0055] Furthermore, the change process of neuronal membrane potential can also be realized by differential method, specifically:

[0056]

[0057] Among them, V(t) represents the membrane potential voltage of the LIF neuron, X(t) represents the sum of the pulses received by the LIF neuron at time t, and when the membrane potential voltage of the LIF neuron exceeds the set threshold, the LIF neuron emits a pulse, and then the neuron membrane potential returns to the zero potential resting state.

[0058] S3: Acquire an IQ signal data set to train the created deep network model, thereby realizing modulation recognition and classification of the IQ signal data to be recognized based on the trained deep network model.

[0059] It should be noted that the pulse residual block includes a channel attention mechanism for learning the relationship between features; the channel attention mechanism is specifically expressed as:

[0060]

[0061] h c =α c ×h c

[0062] Among them, α c represents the attention weight of channel c, f c represents the attention function of channel c, C represents the total number of channels, and h c Represents the input vector of channel c, which is calculated by calculating the attention weights of each channel and then multiplying it by the input vector to obtain the output of the network model. exp represents the natural exponential function.

[0063] Channel Attention Mechanism (CAM) is an effective attention mechanism that can help network models better understand multi-dimensional data. Its basic idea is to introduce a channel attention mechanism into the network model to learn the relationship between features so that the network model can better understand multi-dimensional data.

[0064] Since deep learning generally uses deep neural networks, in order to more efficiently calculate channel attention features, it is often necessary to compress the spatial dimensions of the feature map. The commonly used dimension compression method is to use the global average pooling method. Figure 3 As shown in the figure, it is a schematic diagram of the implementation of the channel attention mechanism.

[0065] Furthermore, in one embodiment, for the channel attention mechanism, specific implementation steps include:

[0066] S11: For each channel, global average pooling is performed, and each channel obtains 1 scalar, and C channels obtain C scalars;

[0067] S12: After passing through the fully connected layer, ReLU activation function, fully connected layer, and Sigmoid activation function in sequence, C scalars between 0 and 1 are obtained as the channel weights;

[0068] S13: Perform weighted operation on each channel using corresponding weights to obtain new weighted features.

[0069] Specifically, for each output channel, global average pooling is performed first to obtain 1 scalar for each channel and C scalars for C channels. Then, FC-ReLU-FC-Sigmoid is used to obtain C scalars between 0 and 1 as the channel weight. Then, each channel of the original output channel is weighted with the corresponding weight to obtain a new weighted feature.

[0070] The improved pulse residual block is as follows Figure 4 As shown, from Figure 4 It can be seen that it is basically similar to the ordinary residual block, and the biggest difference is that the improved pulse residual block has an additional channel attention mechanism module. Since the output of the pulse neural network is a binary value (0,1), if the result of each run is used for classification, it is easy to be interfered. However, when the T value is set to a large value, it will have the disadvantages of high power consumption and high latency. Therefore, after a reasonable value of T is selected, the average emission frequency of the output layer after T time steps is used as the basis for classification. Figure 4 The Conv in represents the convolutional layer.

[0071] The modulation recognition method based on the pulse convolutional neural network in the embodiment of the present application is achieved by connecting and combining multiple pulse residual blocks that integrate the attention mechanism, and constructing the convolution layer and the fully connected layer of the pulse residual block through the LIF neuron, and then creating a deep network model with the pulse residual neural network as the backbone network framework and integrating the attention mechanism based on the connected and combined pulse residual blocks, and finally obtaining an IQ signal data set to train the created deep network model, so as to realize the modulation recognition and classification of the IQ signal data to be identified based on the trained deep network model, and applying SNN to modulation recognition, combining deep learning, and training on the IQ signal data set to establish a network model with the pulse residual neural network as the backbone network framework and integrating the attention mechanism. Compared with the traditional deep learning model, the application of SNN in the present application will reduce complex nonlinear operations such as sigmoid and atan, thereby reducing power consumption while improving the computing performance of the wireless communication system, providing new ideas and methods for the application of deep learning in modulation recognition, achieving the effect of improving recognition speed and reducing computing power consumption, and contributing to the improvement of the performance of the wireless communication system.

[0072] In a second aspect, an embodiment of the present application also provides a modulation recognition device based on a pulse convolutional neural network.

[0073] In one embodiment, referring to Figure 5 , Figure 5 This is a functional module diagram of the modulation recognition device based on the pulse convolutional neural network of this application. Figure 5 As shown, the modulation recognition device based on the pulse convolutional neural network includes: a building module, a creation module, and an execution module.

[0074] The construction module is used to connect and combine multiple pulse residual blocks that integrate the attention mechanism, and to construct the convolution layer and fully connected layer of the pulse residual block through the LIF neuron; the creation module is used to create a deep network model with the pulse residual neural network as the backbone network framework and the attention mechanism based on the connected and combined pulse residual blocks; the execution module is used to obtain the IQ signal data set to train the created deep network model, so as to realize the modulation recognition and classification of the IQ signal data to be identified based on the trained deep network model.

[0075] In the present application, multiple pulse residual blocks in a deep network model are connected in sequence; the deep network model also includes a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer of the deep network model are both composed of LIF neurons.

[0076] Among them, the functional implementation of each module in the above-mentioned modulation identification device based on pulse convolutional neural network corresponds to the various steps in the above-mentioned modulation identification method embodiment based on pulse convolutional neural network, and its functions and implementation process will not be repeated here one by one.

[0077] In the third aspect, an embodiment of the present application provides a modulation recognition device based on a pulse convolutional neural network. The modulation recognition device based on a pulse convolutional neural network can be a personal computer (PC), a laptop computer, a server, or other device with data processing capabilities.

[0078] Reference Figure 6 , Figure 6 The hardware structure diagram of the modulation recognition device based on the pulse convolutional neural network involved in the embodiment of the present application is shown in FIG. In the embodiment of the present application, the modulation recognition device based on the pulse convolutional neural network may include a processor, a memory, a communication interface and a communication bus.

[0079] The communication bus may be of any type and is used to interconnect the processor, the memory, and the communication interface.

[0080] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, etc., which are used to realize the interconnection of devices inside the modulation recognition device based on the pulse convolutional neural network, and an interface used to realize the interconnection between the modulation recognition device based on the pulse convolutional neural network and other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, an optical fiber interface, an ATM interface, etc.; the user device can be a display (Display), a keyboard (Keyboard), etc.

[0081] The memory can be various types of storage media, such as random access memory (RAM), read-only memory (ROM), non-volatile RAM (NVRAM), flash memory, optical storage, hard disk, programmable ROM (PROM), erasable PROM (EPROM), electrically erasable PROM (EEPROM), etc.

[0082] The processor may be a general-purpose processor, which may call the modulation recognition program based on the pulse convolutional neural network stored in the memory and execute the modulation recognition method based on the pulse convolutional neural network provided in the embodiment of the present application. For example, the general-purpose processor may be a central processing unit (CPU). The method executed when the modulation recognition program based on the pulse convolutional neural network is called may refer to the various embodiments of the modulation recognition method based on the pulse convolutional neural network of the present application, which will not be repeated here.

[0083] Those skilled in the art will understand that Figure 6 The hardware structure shown in the figure does not constitute a limitation on the present application, and may include more or less components than shown in the figure, or combine certain components, or arrange the components differently.

[0084] The terms "including" and "having" and any variations thereof in the specification and claims of this application and the above-mentioned drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, products or devices. The terms "first", "second" and "third" are used to distinguish different objects, etc., and do not represent a sequence, nor do they limit "first", "second" and "third" to different types.

[0085] In the description of the embodiments of the present application, "exemplary", "for example" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary", "for example" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary", "for example" or "for example" is intended to present related concepts in a specific way.

[0086] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0087] In some processes described in the embodiments of the present application, multiple operations or steps that appear in a specific order are included, but it should be understood that these operations or steps may not be executed in the order in which they appear in the embodiments of the present application or in parallel, and the sequence number of the operation is only used to distinguish the different operations, and the sequence number itself does not represent any execution order. In addition, these processes may include more or fewer operations, and these operations or steps may be executed in sequence or in parallel, and these operations or steps may be combined.

[0088] Through the description of the above implementation methods, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, disk, CD) as described above, and includes a number of instructions for a terminal device to execute the methods described in each embodiment of the present application.

[0089] The above are only preferred embodiments of the present application, and are not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A modulation recognition method based on pulse convolutional neural network, characterized in that: The modulation recognition method based on pulse convolutional neural network includes: Combine multiple pulse residual blocks with fused attention mechanisms, and construct the convolutional layer and fully connected layer of the pulse residual block through LIF neurons; Based on the connected combined pulse residual block, a deep network model with a pulse residual neural network as the backbone network framework and an attention mechanism is created; An IQ signal data set is obtained to train the created deep network model, so as to realize modulation recognition and classification of the IQ signal data to be recognized based on the trained deep network model.

2. The modulation recognition method based on pulse convolutional neural network according to claim 1, characterized in that: The multiple pulse residual blocks in the deep network model are connected in sequence.

3. The modulation recognition method based on pulse convolutional neural network according to claim 1, characterized in that: The deep network model also includes a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer of the deep network model are both composed of LIF neurons.

4. A modulation recognition method based on a pulse convolutional neural network as claimed in claim 2 or 3, characterized in that: For LIF neurons, the LIF model is used to describe the relationship between neuronal membrane potential and neuronal input current, and the change process of neuronal membrane potential is specifically expressed as: Among them, τ m is the membrane time constant, R m represents resistance, V represents neuron membrane potential, I represents the sum of synaptic currents generated by the discharge of each presynaptic neuron, and V rest Represents the resting membrane potential.

5. A modulation recognition method based on pulse convolutional neural network as claimed in claim 4, characterized in that: The change process of neuronal membrane potential can also be realized by differential method, specifically: Among them, V(t) represents the membrane potential voltage of the LIF neuron, X(t) represents the sum of the pulses received by the LIF neuron at time t, and when the membrane potential voltage of the LIF neuron exceeds the set threshold, the LIF neuron emits a pulse, and then the neuron membrane potential returns to the zero potential resting state.

6. The modulation recognition method based on pulse convolutional neural network according to claim 1, characterized in that: The pulse residual block includes a channel attention mechanism for learning the relationship between features; The channel attention mechanism is specifically expressed as: h c =a c ×h c Among them, α c represents the attention weight of channel c, f c represents the attention function of channel c, C represents the total number of channels, and h c represents the input vector of channel c, and exp represents the natural exponential function.

7. A modulation recognition method based on pulse convolutional neural network as claimed in claim 6, characterized in that: For the channel attention mechanism, the specific implementation steps include: For each channel, global average pooling is performed, and each channel obtains 1 scalar, and C channels obtain C scalars; After passing through the fully connected layer, ReLU activation function, fully connected layer, and Sigmoid activation function, C scalars between 0 and 1 are obtained as the channel weights; Each channel is weighted by using the corresponding weight to obtain a new weighted feature.

8. A modulation recognition device based on pulse convolutional neural network, characterized in that: The modulation recognition device based on pulse convolutional neural network includes: A building module, which is used to connect and combine multiple pulse residual blocks fused with attention mechanisms, and construct the convolutional layer and fully connected layer of the pulse residual block through LIF neurons; A creation module is used to create a deep network model based on the pulse residual block of the connection combination, with the pulse residual neural network as the backbone network framework and integrating the attention mechanism; An execution module is used to obtain an IQ signal data set to train the created deep network model, thereby realizing modulation recognition and classification of the IQ signal data to be recognized based on the trained deep network model.

9. A modulation recognition device based on pulse convolutional neural network as claimed in claim 8, characterized in that: Multiple impulse residual blocks in the deep network model are connected in sequence; The deep network model also includes a normalization layer and a fully connected layer, and the normalization layer and the fully connected layer of the deep network model are both composed of LIF neurons.

10. A modulation recognition device based on a pulse convolutional neural network, characterized in that: The modulation recognition device based on pulse convolutional neural network includes a processor, a memory, and a modulation recognition program based on pulse convolutional neural network stored in the memory and executable by the processor, wherein when the modulation recognition program based on pulse convolutional neural network is executed by the processor, the steps of the modulation recognition method based on pulse convolutional neural network as described in any one of claims 1 to 7 are implemented.

Images