PAM4 signal recognition method, device, equipment and readable storage medium

Abstract

A PAM4 signal recognition method, device and equipment and readable storage medium. The PAM4 signal recognition method comprises: copying a length L of received end sampling values corresponding to a PAM4 symbol to form a sequence with a data format of [T, L] after T time steps; inputting the sequence with the data format of [T, L] into a pulse attention mechanism neural network to obtain T time steps; and identifying the PAM4 symbol according to an average pulse firing frequency of the T time steps. By copying the received end sampling values for T time steps to form a sequence with a data format of [T, L], inputting the sequence into a pulse attention mechanism neural network to obtain T time steps, and identifying the PAM4 symbol according to an average pulse firing frequency of the T time steps, the low energy consumption and low calculation amount pulse neural network is combined with the high sensitivity attention mechanism, the operation amount for identifying the PAM4 symbol is reduced, the energy consumption is reduced, and the technical problem of occupying many operation resources and high time delay and energy consumption in the related art is solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of signal recognition, in particular to a PAM4 signal recognition method, device and equipment and readable storage medium. BACKGROUND

[0002] At present, the optical fiber communication system commonly uses a four-pulse amplitude modulation method (referred to as PAM4 method) to modulate optical signals, so it is necessary to identify the electrical signal sampling value at the receiving end of the optical fiber communication system to recover the original bit stream. Therefore, it is extremely important to improve the identification accuracy of the PAM4 electrical signal of the optical fiber communication system.

[0003] In related technologies, as a device for transmitting and receiving optical signals in optical fiber communication, the optical fiber receiving end mainly processes the electrical signal converted from the optical signal through digital signal processing technology. With the development of artificial intelligence technology, the model algorithm related to artificial intelligence highlights the stronger performance compared with the traditional digital signal processing technology. Among them, the pulse neural network, which is called the third generation of neural network, highlights its prospects due to its low power consumption, event-driven, and biological interpretability. At the same time, the introduction of the attention mechanism enables the neural network to focus on a subset of its input (or features), which has been widely used in natural language processing and communication fields. The three most important matrices in self-attention mechanism calculation are: Quary matrix (Q), Key matrix (K) and Value matrix (V). In the traditional self-attention mechanism calculation, the processed original data is multiplied by the floating-point number matrix Q and V, and then subjected to the softmax operation containing the exponential power operation and division. The above operation has the problems of complex implementation, high occupation of operation resources, high time delay and high energy consumption in the embedded device of the optical fiber communication system, which is not conducive to the further improvement of the performance of the optical fiber communication system and the reduction of the operation energy consumption.

[0004] Therefore, it is necessary to design a new PAM4 signal recognition method based on the pulse attention mechanism to overcome the above problems. SUMMARY

[0005] The present application provides a PAM4 signal recognition method, device, equipment and readable storage medium, which can solve the technical problems of high occupation of operation resources, high time delay and high energy consumption in related technologies.

[0006] In a first aspect, the present application provides a PAM4 signal recognition method based on a pulse attention mechanism, which comprises:

[0007] copying the length L receiving end sampling value corresponding to the PAM4 symbol for T time steps to form a sequence with a data format of [T, L];

[0008] inputting a sequence in a data format of [T, L] into a pulse attention mechanism neural network to obtain T time steps;

[0009] recognize the PAM4 symbol according to the average pulse firing frequency of the T time steps.

[0010] In combination with the first aspect, in an implementation manner, the inputting the sequence in the data format of [T, L] into the pulse attention mechanism neural network to obtain the T time steps comprises:

[0011] obtaining pulse features after processing the sequence in the data format of [T, L];

[0012] inputting the pulse features into a pulse self-attention module to obtain pulse self-attention encoding E;

[0013] inputting the pulse self-attention encoding E into a full connection module to obtain the T time steps.

[0014] In combination with the first aspect, in an implementation manner, the obtaining the pulse features after processing the sequence in the data format of [T, L] comprises:

[0015] obtaining a preprocessed sequence after preprocessing the sequence in the data format of [T, L];

[0016] obtaining the pulse features after adding the preprocessed sequence and a position vector obtained by training the neural network.

[0017] In combination with the first aspect, in an implementation manner, the inputting the pulse features into the pulse self-attention module to obtain the pulse self-attention encoding E comprises:

[0018] obtaining Q, K and V matrices in a pulse form after the pulse features pass through a three-way linear full connection layer, a BN layer and a LIF neuron layer;

[0019] inputting the Q, K and V matrices in the pulse form into a linear full connection layer, a BN layer and a LIF neuron layer in sequence after matrix point multiplication and passing through the LIF neuron layer, to obtain the pulse self-attention encoding E.

[0020] In combination with the first aspect, in an implementation manner, the inputting the pulse self-attention encoding E into the full connection module to obtain the T time steps comprises:

[0021] obtaining the T time steps after the pulse self-attention encoding E passes through a plurality of full connection modules comprising a linear full connection layer, a BN layer and a LIF neuron layer.

[0022] In combination with the first aspect, in an implementation manner, the recognizing the PAM4 symbol according to the average pulse firing frequency of the T time steps comprises:

[0023] determine the target output neuron sequence number corresponding to the average pulse firing frequency of the T time steps according to the average pulse firing frequency of the T time steps and the correspondence between the average pulse firing frequency and the target output neuron sequence number;

[0024] identify the PAM4 symbol based on the target output neuron sequence number and the correspondence between the target output neuron sequence number and the PAM4 symbol.

[0025] In combination with the first aspect, in an implementation manner, the received end sampling value corresponding to the PAM4 symbol with the length of L includes 2 sampling values of the received end corresponding to the PAM4 symbol of the sending end and 2*tap sampling values of the received end corresponding to tap symbols before and after the PAM4 symbol of the sending end.

[0026] The second aspect, the embodiment of the application provides a PAM4 signal identification device, the PAM4 signal identification device includes:

[0027] The sequence generation module is used to copy the received end sampling value corresponding to the PAM4 symbol with the length of L to form a sequence with the data format of [T, L] after T time steps.

[0028] The processing module is used to input the sequence with the data format of [T, L] into the pulse attention mechanism neural network to obtain T time steps.

[0029] The signal identification module is used to identify the PAM4 symbol according to the average pulse firing frequency of the T time steps.

[0030] The third aspect, the embodiment of the application provides a PAM4 signal identification device, the PAM4 signal identification device includes a processor, a memory, and a PAM4 signal identification program stored in the memory and executable by the processor, wherein when the PAM4 signal identification program is executed by the processor, the steps of the PAM4 signal identification method described above are implemented.

[0031] The fourth aspect, the embodiment of the application provides a computer readable storage medium, the computer readable storage medium stores a PAM4 signal identification program, wherein when the PAM4 signal identification program is executed by a processor, the steps of the PAM4 signal identification method described above are implemented.

[0032] The technical scheme provided by the embodiment of the application has the beneficial effects that:

[0033] By replicating the receiving-end sampling values ​​for T time steps, a sequence with a data format of [T, L] can be formed. Inputting the sequence into a pulse attention mechanism neural network can obtain T time steps, and PAM4 symbols can be identified based on the average pulse emission frequency of T time steps. This embodiment combines a low-energy and low-computation pulse neural network with a high-sensitivity attention mechanism, which can reduce the amount of computation required to identify PAM4 symbols and reduce energy consumption, thereby solving the technical problems of high computational resource consumption and high latency and energy consumption in related technologies. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 This is a flow chart of an embodiment of a PAM4 signal recognition method of the present application;

[0035] Figure 2 This is the correspondence between each PAM4 symbol at the transmitter and two samples at the receiver;

[0036] Figure 3 The 2*tap samples at the receiving end correspond to each tap symbol before and after the PAM4 symbol of this application;

[0037] Figure 4 This is a schematic diagram of the structure of an embodiment of a pulse attention mechanism neural network of the present application;

[0038] Figure 5 This is a schematic diagram of the structure of another embodiment of the pulse attention mechanism neural network of the present application;

[0039] Figure 6 This is a schematic diagram of the matrix dot multiplication of this application;

[0040] Figure 7 This is a schematic diagram of the hardware structure of the PAM4 signal recognition device involved in the embodiment of the present application. DETAILED DESCRIPTION

[0041] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0042] First, some technical terms in this application are explained to facilitate those skilled in the art to understand this application.

[0043] SNN: Spiking Neural Networks, pulse neural network.

[0044] PAM4: 4-level pulse amplitude modulation.

[0045] LIF neuron: Leaky integrated-and-fire neuron, leaky integrated-and-fire neuron.

[0046] In order to make the objectives, technical solutions and advantages of this application clearer, the implementation methods of this application will be further described in detail below with reference to the accompanying drawings.

[0047] In a first aspect, an embodiment of the present application provides a PAM4 signal recognition method.

[0048] In one embodiment, referring to Figure 1 , Figure 1 This is a flow chart of the first embodiment of the PAM4 signal recognition method based on the pulse attention mechanism of this application. Figure 1 As shown, the PAM4 signal recognition method includes the following steps:

[0049] S1: The receiving-end sampling value of length L corresponding to the PAM4 symbol is copied for T time steps to form a sequence with the data format of [T, L].

[0050] S2: Input the sequence of data in the format of [T, L] into the pulse attention mechanism neural network to obtain T time steps.

[0051] S3: Identify PAM4 symbols based on the average pulse emission frequency over T time steps.

[0052] In this embodiment, see Figure 2 As shown in Figure 1, each PAM4 symbol (ie, electrical signal) at the transmitter corresponds to two sampled values ​​at the receiver. Figure 3 As shown, when identifying each PAM4 symbol sent by the transmitter of the optical fiber communication system, the identification of each PAM4 symbol at the transmitter is performed by combining the two sampling values ​​of the PAM4 symbol at the receiver and the 2*tap sampling values ​​at the receiver corresponding to each tap symbol before and after the PAM4 symbol at the transmitter. In other words, the identification of each PAM4 symbol at the transmitter requires combining L=2*(2*tap+1) sampling values ​​at the receiver. That is, the receiving end sampling values ​​of length L corresponding to the above PAM4 symbol include the two sampling values ​​of the PAM4 symbol at the transmitter corresponding to the receiver and the 2*tap sampling values ​​at the receiver corresponding to each tap symbol before and after the PAM4 symbol at the transmitter.

[0053] See also Figure 4 As shown in the figure, the 2*(2*tap+1) sampling values ​​at the receiving end are fed into the pulse attention mechanism neural network, and the PAM4 symbol is determined based on the output of the 4 output layer neurons.

[0054] The embodiment can form a sequence with a data format of [T, L] by copying the receiving end sampling value for T time steps, input the sequence into the pulse attention mechanism neural network, obtain T time steps, and identify the PAM4 symbol according to the average pulse firing frequency of the T time steps; the embodiment combines the SNN and the artificial intelligence attention mechanism, combines the low-energy and low-computing pulse neural network (SNN) and the high-sensitivity attention mechanism, can reduce the operation amount of identifying the PAM4 symbol, reduce the occupation of the operation resource, reduce the energy consumption, and solve the technical problems of occupying many operation resources and high time delay and energy consumption in the related art.

[0055] Referring to Figure 5 The pulse neuron shown in the figure is composed of a leaky integrated-and-fire neuron (LIF) neuron, the LIF neuron is widely used in SNN, is a pulse neuron model that best balances biological interpretability and implementation complexity, the neuron will accumulate input pulses over time in the “integration” process, threshold decision will determine whether to fire a pulse, if the threshold is not reached and no new pulse is accumulated, the neuron will gradually “leak” the existing accumulation to the resting state, and the neuron will fire a pulse after reaching the threshold, and then enter a short refractory period, the neuron cannot fire a pulse during the refractory period, and the neuron will re-integrate after returning to the resting state.

[0056] The LIF model can intuitively describe the relationship between the membrane potential V of the neuron and the input current I, and the change process of the membrane potential V is described by the following first-order differential equation:

[0057]

[0058] In the formula, τ m =C m R m is called a membrane time constant, I is the sum of synaptic currents generated by the firing behavior of each presynaptic neuron, V rest is the resting voltage of the membrane potential, and is usually 0. In practice, the following difference equation is usually used to implement it:

[0059]

[0060] In the formula, V(t) is the membrane potential voltage of the LIF neuron, and X(t) is the total number of pulses received by the LIF neuron at time t. When the membrane potential voltage of the LIF neuron exceeds the threshold value, the LIF neuron will fire a pulse, and then the membrane potential of the neuron will recover to the zero potential resting state.

[0061] Further, in an embodiment, the inputting the sequence in the data format of [T, L] into the pulse attention mechanism neural network to obtain T time steps can include:

[0062] S21: obtaining pulse features after processing the sequence in the data format of [T, L].

[0063] S22: inputting the pulse features into a pulse self-attention module to obtain pulse self-attention encoding E.

[0064] S23: inputting the pulse self-attention encoding E into a full connection module to obtain T time steps.

[0065] Referring to FIG. 2, Figure 5 In an embodiment, processing the sequence in the data format of [T, L] can obtain pulse features that can be processed by the neural network. In an alternative implementation, the sequence in the data format of [T, L] can be preprocessed to obtain a preprocessed sequence. For example, the sequence [T, L] can be subjected to one-dimensional convolution with N convolution kernels, a batch normalization (BN) layer, and a LIF neuron layer, and maximum pooling to form a sequence in the format of [T, N, D] (where D is the length of the sequence after convolution with the convolution kernel). Then, the preprocessed sequence and a position vector obtained by training the neural network are added to obtain pulse features in the format of [T, N, D]. Of course, in other embodiments, other processing methods can be used to obtain pulse features, which are not limited herein. In this embodiment, the pulse self-attention encoding E is also in the format of [T, N, D].

[0066] Further, in some embodiments, the inputting the pulse features into the pulse self-attention module to obtain the pulse self-attention encoding E can include:

[0067] S221: obtaining Q, K, and V matrices in pulse form after the pulse features pass through a three-way linear fully connected layer, a BN layer, and a LIF neuron layer;

[0068] S222: performing matrix point multiplication on the Q, K, and V matrices in pulse form, passing through a LIF neuron layer, and then sequentially inputting into a linear fully connected layer, a BN layer, and a LIF neuron layer to obtain the pulse self-attention encoding E.

[0069] Referring to FIG. 2, Figure 5As shown, after obtaining the pulse feature, the pulse feature then enters the pulse self-attention module, and at the same time, the pulse self-attention module has a residual connection. After the pulse feature enters the pulse self-attention module, it passes through three linear fully connected layers, BN layers and LIF neuron layers, each of which has a linear fully connected layer (Linear), a BN layer and a LIF neuron layer. After passing through the three linear fully connected layers, the BN layers and the LIF neuron layers, the Q, K and V matrices in the form of pulses are obtained, specifically:

[0070]

[0071] wherein the format of W Q , W K and W V is [D,D].

[0072] Then, the obtained Q, K and V matrices are subjected to matrix point multiplication (Q Figure 6 As shown in the example, Figure 6 wherein N is 3 and D is 2) QK T V, and then sequentially enter the linear fully connected layer (Linear), the BN layer and the LIF neuron layer after the LIF neuron layer. The output data format of the pulse self-attention encoding E is [T,N,D]. Specifically:

[0073] E = LIF (BN (Linear (LIF (QK T V)))).

[0074] Further, in an embodiment, referring to Figure 5 As shown, the input of the pulse self-attention encoding E into the full connection module obtains T time steps, which includes: the pulse self-attention encoding E passes through several full connection modules containing linear fully connected layers, BN layers and LIF neuron layers to obtain T time steps. In this embodiment, the full connection module contains a residual connection from input to output. The pulse self-attention encoding E passes through several full connection modules to obtain an output, which outputs T time steps.

[0075] In some embodiments, the identification of the PAM4 symbol according to the average pulse firing frequency of the T time steps can include:

[0076] S31: determining the target output neuron sequence number corresponding to the average pulse firing frequency of the T time steps according to the average pulse firing frequency of the T time steps and the corresponding relationship between the average pulse firing frequency and the target output neuron sequence number;

[0077] S32: identifying the PAM4 symbol based on the target output neuron sequence number and the corresponding relationship between the target output neuron sequence number and the PAM4 symbol.

[0078] In this embodiment, as shown in Table 1 below, there is a corresponding relationship between the serial number of the target output neuron and the PAM4 symbol; at the same time, there is also a corresponding relationship between the average pulse firing frequency and the serial number of the target output neuron (not listed); after T time steps are obtained, the PAM4 symbol can be identified according to the average pulse firing frequency of the T time steps in combination with the corresponding relationship between the average pulse firing frequency and the serial number of the target output neuron and the corresponding relationship between the serial number of the target output neuron and the PAM4 symbol.

[0079] For example, the four neurons of the output layer correspond to PAM4 electrical signals '0' '0', '0' '1', '1' '1' and '1' '0' in turn. After T time steps, if the pulse firing frequency of the first neuron of the output layer within the T time steps is the highest, the identified PAM4 signal is '0' '0'; if the pulse firing frequency of the second neuron of the output layer within the T time steps is the highest, the identified PAM4 signal is '0' '1'; if the pulse firing frequency of the third neuron of the output layer within the T time steps is the highest, the identified PAM4 signal is '1' '1'; and if the pulse firing frequency of the fourth neuron of the output layer within the T time steps is the highest, the identified PAM4 signal is '1' '0'.

[0080] Table 1

[0081] Output neuron number PAM4 electrical signal 1 ‘0’‘0’ 2 ‘0’‘1’ 3 ‘1’‘1’ 4 ‘1’‘0’

[0082] In a second aspect, the embodiments of the present application also provide a PAM4 signal identification device.

[0083] In an embodiment, the PAM4 signal identification device comprises: a sequence generation module configured to copy the PAM4 symbol corresponding received end sampling value with a length of L for T time steps to form a sequence with a data format of [T, L]; a processing module configured to input the sequence with the data format of [T, L] into a pulse attention mechanism neural network to obtain T time steps; and a signal identification module configured to identify the PAM4 symbol according to the average pulse firing frequency of the T time steps.

[0084] Further, in an embodiment, the processing module is configured to obtain pulse features by processing the sequence with the data format of [T, L]; input the pulse features into a pulse self-attention module to obtain pulse self-attention encoding E; and input the pulse self-attention encoding E into a full connection module to obtain the T time steps.

[0085] Further, in an embodiment, the processing of the sequence in the data format of [T, L] to obtain the pulse feature includes: pre-processing the sequence in the data format of [T, L] to obtain a pre-processed sequence; and adding the pre-processed sequence and a position vector obtained by training a neural network to obtain the pulse feature.

[0086] Further, in an embodiment, the inputting of the pulse feature into the pulse self-attention module to obtain pulse self-attention encoding E includes: obtaining Q, K, and V matrices in a pulse form by processing the pulse feature through a three-way linear fully connected layer, a BN layer, and a LIF neuron layer; performing matrix point multiplication on the Q, K, and V matrices in the pulse form, and then processing the result through a LIF neuron layer, and then sequentially inputting the result into a linear fully connected layer, a BN layer, and a LIF neuron layer to obtain the pulse self-attention encoding E.

[0087] Further, in an embodiment, the inputting of the pulse self-attention encoding E into the full connection module to obtain T time steps includes: obtaining the T time steps by processing the pulse self-attention encoding E through a plurality of full connection modules each including a linear fully connected layer, a BN layer, and a LIF neuron layer.

[0088] Further, in an embodiment, the signal recognition module is configured to determine a target output neuron sequence number corresponding to an average pulse firing frequency of the T time steps according to the average pulse firing frequency of the T time steps and a corresponding relationship between the average pulse firing frequency and the target output neuron sequence number; and recognize a PAM4 symbol based on the target output neuron sequence number and a corresponding relationship between the target output neuron sequence number and the PAM4 symbol.

[0089] Further, in an embodiment, the PAM4 symbol corresponds to a length-L received end sampling value including 2 sampling values corresponding to the PAM4 symbol at a transmitting end and 2*tap sampling values corresponding to tap symbols before and after the PAM4 symbol at the transmitting end at a received end.

[0090] The functions of each module in the PAM4 signal recognition device correspond to the steps in the PAM4 signal recognition method embodiments, and the functions and implementation processes will not be described here.

[0091] In a third aspect, the embodiments of the present application provide a PAM4 signal recognition device. The PAM4 signal recognition device can be a personal computer (PC), a notebook computer, a server, or other devices with data processing functions.

[0092] Reference Figure 7 , Figure 7A hardware structure diagram of a PAM4 signal recognition device involved in an embodiment of the present application is shown in FIG. 1. In the embodiment of the present application, the PAM4 signal recognition device can include a processor, a memory, a communication interface, and a communication bus.

[0093] The communication bus can be of any type, used to interconnect the processor, the memory, and the communication interface.

[0094] The communication interface includes an input / output (I / O) interface, a physical interface, and a logical interface, etc. used to implement the interconnection of devices inside the PAM4 signal recognition device, and an interface used to implement the interconnection of the PAM4 signal recognition device and other devices (such as other computing devices or user devices). The physical interface can be an Ethernet interface, a fiber interface, an ATM interface, etc.; the user device can be a display (Display), a keyboard (Keyboard), etc.

[0095] 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.

[0096] The processor can be a general-purpose processor, which can invoke a PAM4 signal recognition program stored in the memory and execute the PAM4 signal recognition method provided by the embodiment of the present application. For example, the general-purpose processor can be a central processing unit (CPU). The method executed by the PAM4 signal recognition program when invoked can refer to various embodiments of the PAM4 signal recognition method of the present application, which will not be described here.

[0097] Those skilled in the art can understand that the hardware structure shown in FIG. 1 does not constitute a limitation on the present application, and can include more or fewer components than shown, or combine certain components, or different component arrangements. Figure 7

[0098] In a fourth aspect, the embodiment of the present application further provides a readable storage medium.

[0099] ​The PAM4 signal identification program is stored on the readable storage medium of the application, and when the PAM4 signal identification program is executed by the processor, the steps of the PAM4 signal identification method are realized.

[0100] The method realized when the PAM4 signal identification program is executed can refer to the embodiments of the PAM4 signal identification method of the application, which will not be described here.

[0101] It should be noted that the above-mentioned serial numbers of the embodiments of the application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0102] The terms "comprising" and "having" and any variations thereof in the specification and claims of the application and the above-mentioned drawings are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but can optionally include steps or units not listed, or can optionally include other steps or units inherent to the process, method, product or device. The terms "first", "second" and "third" and the like are used to distinguish different objects, and do not represent the order of sequence, nor limit the "first", "second" and "third" to be different types.

[0103] In the description of the embodiments of the application, "exemplary", "for example" or "for instance" is used to mean as an example, illustration or description. Any embodiment or design scheme described as "exemplary", "for example" or "for instance" in the embodiments of the application should not be interpreted as more preferred or more advantageous than other embodiments or design schemes. On the contrary, the use of "exemplary", "for example" or "for instance" is intended to present the relevant concept in a specific manner.

[0104] In the description of the embodiments of the application, unless otherwise specified, " / " means or, for example, A / B can mean A or B; "and / or" in the text only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, in the description of the embodiments of the application, "multiple" means two or more than two.

[0105] In some of the processes described in the embodiments of the application, a plurality of operations or steps are included in a specific order, but it should be understood that these operations or steps can be executed or executed in parallel without the order in which they appear in the embodiments of the application. The serial number of the operation is only used to distinguish different operations, and the serial number itself does not represent any execution order. In addition, these processes can include more or fewer operations, and these operations or steps can be executed in sequence or in parallel, and these operations or steps can be combined.

[0106] Those skilled in the art can clearly understand the above-mentioned embodiment method can be realized by means of software and the necessary general hardware platform, of course, can also be through hardware, but in many cases the former is a better implementation. Based on such understanding, the technical solutions of the present application essentially or say the part of the prior art contribution can be embodied in the form of software products, the computer software product is stored in a storage medium (such as ROM / RAM, magnetic disc, optical disc) as described above, including a number of instructions to make a terminal device execute the method described in various embodiments of the present application.

[0107] The above is only the preferred embodiment of the present application, not therefore limit the patent scope of the present application, all use the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, including the equivalent structure or equivalent process transformation, or are included in the patent protection scope of the present application.

Claims

1. A PAM4 signal recognition method based on a pulse attention mechanism, characterized in that: The PAM4 signal recognition method includes: The receiving end sampling value of length L corresponding to the PAM4 symbol is copied for T time steps to form a sequence with the data format of [T, L]; Input the sequence of data in the format of [T, L] into the pulse attention mechanism neural network to obtain T time steps; Determine the target output neuron number corresponding to the average pulse firing frequency of T time steps based on the average pulse firing frequency of T time steps and the corresponding relationship between the average pulse firing frequency and the target output neuron number; The PAM4 symbol is identified based on the target output neuron number and the correspondence between the target output neuron number and the PAM4 symbol.

2. The PAM4 signal recognition method according to claim 1, wherein: The sequence of data in the format of [T, L] is input into the pulse attention mechanism neural network to obtain T time steps, including: The pulse characteristics are obtained after processing the sequence with the data format of [T, L]; Input the pulse feature into the pulse self-attention module to obtain the pulse self-attention code E; The spike self-attention code E is fed into the fully connected module to obtain T time steps.

3. The PAM4 signal recognition method according to claim 2, wherein: The pulse characteristics are obtained by processing the sequence in the data format of [T, L], including: Preprocess the sequence in the data format of [T, L] to obtain the preprocessed sequence; The pulse feature is obtained by adding the preprocessed sequence to the position vector obtained by training the neural network.

4. The PAM4 signal recognition method according to claim 2, wherein: The step of inputting the pulse feature into the pulse self-attention module to obtain the pulse self-attention code E includes: After the pulse features pass through the three-way linear fully connected layer, BN layer and LIF neuron layer, the Q, K, and V matrices in the pulse form are obtained; The Q, K, and V matrices in pulse form are matrix-multiplied and passed through the LIF neuron layer, and then sequentially input into the linear fully connected layer, BN layer, and LIF neuron layer to obtain the pulse self-attention code E.

5. The PAM4 signal recognition method according to claim 2, wherein: The pulse self-attention code E is input into the fully connected module to obtain T time steps, including: The pulse self-attention code E is passed through several fully connected modules including linear fully connected layers, BN layers and LIF neuron layers to obtain T time steps.

6. The PAM4 signal recognition method according to claim 1, wherein: The receiving end sampling value of length L corresponding to the PAM4 symbol includes 2 sampling values ​​of the receiving end corresponding to the PAM4 symbol of the transmitting end and 2*tap sampling values ​​of the receiving end corresponding to each tap symbol before and after the PAM4 symbol at the transmitting end.

7. A PAM4 signal recognition device, characterized in that: The PAM4 signal recognition device includes: A sequence generation module is used to copy the receiving end sampling value of length L corresponding to the PAM4 symbol for T time steps to form a sequence with a data format of [T, L]; A processing module, which is used to input a sequence of data in the format of [T, L] into the pulse attention mechanism neural network to obtain T time steps; A signal recognition module is used to determine the target output neuron number corresponding to the average pulse emission frequency of T time steps based on the average pulse emission frequency of T time steps and the correspondence between the average pulse emission frequency and the target output neuron number; and identify the PAM4 symbol based on the target output neuron number and the correspondence between the target output neuron number and the PAM4 symbol.

8. A PAM4 signal recognition device, characterized in that: The PAM4 signal identification device includes a processor, a memory, and a PAM4 signal identification program stored in the memory and executable by the processor, wherein when the PAM4 signal identification program is executed by the processor, the steps of the PAM4 signal identification method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a PAM4 signal recognition program, wherein when the PAM4 signal recognition program is executed by a processor, the steps of the PAM4 signal recognition method according to any one of claims 1 to 6 are implemented.

Images