Real signal OFDM underwater acoustic communication method and system

Abstract

The invention discloses a real signal OFDM underwater acoustic communication method and system, and relates to the technical field of underwater acoustic communication, and the method comprises the steps: deploying a unitary neural network model at a transmitting end, and deploying a deep neural network model at a receiving end; performing joint training on the unitary neural network model and the deep neural network model to obtain a trained unitary neural network model and a trained deep neural network model; inputting an information bit stream to the transmitting end, generating pre-coded data according to the information bit stream by using the trained unitary neural network model, and transmitting the pre-coded data to the receiving end in the form of a real OFDM (Orthogonal Frequency Division Multiplexing) signal; and receiving a real OFDM signal by using a receiving end, and performing signal detection by using the trained deep neural network model to obtain an information bit stream. According to the method, the calculation complexity can be reduced, and the data detection performance is improved.

Description

Technical Field

[0001] The present application relates to the technical field of underwater acoustic communication, and in particular to a real signal OFDM underwater acoustic communication method and system. Background Art

[0002] At present, in OFDM (Orthogonal Frequency Division Multiplexing) underwater acoustic communication (UWAC) systems, due to the complex characteristics of underwater acoustic channels, such as severe multipath effects and Doppler effects, reliable underwater acoustic communication depends on accurate channel estimation and carrier frequency offset (CFO) estimation. However, in order to track these channel effects, a considerable number of subcarriers in OFDM underwater acoustic communication systems are used to carry overhead packets, which occupy limited bandwidth resources, resulting in a serious reduction in bandwidth efficiency.

[0003] In order to overcome the problem of reduced bandwidth efficiency of OFDM underwater acoustic communication systems caused by additional overhead packets, researchers have proposed a variety of methods. Some studies have used guard intervals to track underwater acoustic channels, but these methods assume that the channel is sparse, which is difficult to guarantee in practical applications, and the performance is often affected by inter-block interference. Other studies have proposed the use of Discrete Hartley Transform (DHT) instead of Discrete Fourier Transform (DFT) to reduce the required overhead packets. However, these methods still require the transmission of pilot sequences and may require additional empty subcarriers to track the CFO effect. In addition, the Superimposed Training (ST) method has been introduced as an alternative to pilot symbol-assisted modulation to overcome the problem of bandwidth reduction by superimposing overhead packets on data symbols, but this method is limited by interference between pilot and data signals. Therefore, some researchers proposed the method of Data-nulling Superimposed Pilots, but this method can only provide acceptable data detection performance when the pilot sequence is short. In actual underwater acoustic communication systems, 25% of OFDM symbols are usually used to carry pilot sequences, which leads to a sharp drop in the data detection performance of this method. In recent years, methods based on deep learning (DL) have been introduced into underwater acoustic communication systems to solve the problems of channel estimation and data detection. However, due to the unpredictability of underwater acoustic channels and the long length of OFDM symbols in underwater acoustic communications, the computational complexity required to train and test deep learning models is very large, which limits the practical application of this method.

[0004] In summary, the real-signal OFDM underwater acoustic communication methods in the existing technology generally have the problems of complex calculations and low data detection performance. Therefore, how to provide a real-signal OFDM underwater acoustic communication method with simple calculations and high data detection performance has become a technical problem to be solved in this field. Summary of the invention

[0005] The purpose of this application is to provide a real-signal OFDM underwater acoustic communication method and system, which can reduce computational complexity and improve data detection performance.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a real-signal OFDM underwater acoustic communication method, which is used for underwater acoustic communication between a transmitter and a receiver, wherein the transmitter is used to input an information bit stream and then send a real OFDM signal to the receiver, and the receiver is used to obtain the information bit stream according to the real OFDM signal; the real-signal OFDM underwater acoustic communication method comprises the following steps:

[0008] A unitary neural network model is deployed at the transmitting end, and a deep neural network model is deployed at the receiving end; the input of the unitary neural network model is the modulation data corresponding to the information bit stream, and the output is the precoded data, and the unitary neural network model is used to generate the precoded data according to the input information bit stream; the input of the deep neural network model is the detection data corresponding to the information bit stream, and the output is the detection result corresponding to the information bit stream, and the deep neural network model is used to detect the information bit stream.

[0009] The unitary neural network model and the deep neural network model are jointly trained to obtain a trained unitary neural network model and a trained deep neural network model.

[0010] The information bit stream is input to the transmitting end, and the precoded data is generated according to the information bit stream using the trained unitary neural network model, and is sent to the receiving end in the form of the real OFDM signal.

[0011] The real OFDM signal is received by the receiving end, and the trained deep neural network model is used to perform signal detection on the real OFDM signal to obtain the information bit stream.

[0012] In the second aspect, the present application provides a real-signal OFDM underwater acoustic communication system, which includes a transmitter and a receiver. The transmitter is used to input an information bit stream and then send a real OFDM signal to the receiver, and the receiver is used to obtain the information bit stream based on the real OFDM signal. When underwater acoustic communication is performed between the transmitter and the receiver, the real-signal OFDM underwater acoustic communication method described in the first aspect is executed.

[0013] According to the specific embodiments provided in this application, this application has the following technical effects:

[0014] The present application provides a real-signal OFDM underwater acoustic communication method and system, which combines a unitary neural network model and a deep neural network model, respectively deploys the unitary neural network model and the deep neural network model at the transmitting end and the receiving end and performs joint training, and uses the unitary neural network model at the transmitting end to generate precoded data, so that the method does not need to include an underwater acoustic channel during the training process, thereby reducing the computational complexity and solving the problem of overly complex computations in the training process of traditional methods. At the same time, the present application also deploys a deep neural network model at the receiving end for signal detection, which can quickly and effectively detect the information bit stream contained in the real OFDM signal, thereby improving the performance of data detection and solving the problem of decreased data detection performance of traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 A flowchart of a real signal OFDM underwater acoustic communication method provided in one embodiment of the present application;

[0017] Figure 2 A schematic diagram of a signal processing flow provided in an embodiment of the present application;

[0018] Figure 3 A bit error rate curve diagram of each method provided in an embodiment of the present application in an additive white Gaussian noise channel;

[0019] Figure 4 A bit error rate curve diagram of each method provided in an embodiment of the present application under a perfect known underwater acoustic channel;

[0020] Figure 5 A bit error rate curve diagram of each method provided in an embodiment of the present application under different CFO values;

[0021] Figure 6 This is a bit error rate curve diagram of each method provided in an embodiment of the present application with or without a CFO compensation process. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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.

[0023] In order to make the above-mentioned objects, features and advantages of the present application more obvious and easy to understand, the present application is further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0024] like Figure 1 As shown, this embodiment proposes a real-signal OFDM underwater acoustic communication method, which is mainly used in a real-signal OFDM underwater acoustic communication system. The real-signal OFDM underwater acoustic communication system includes a transmitter and a receiver, and underwater acoustic communication is performed between the transmitter and the receiver. The transmitter is used to input an information bit stream and then send a real OFDM signal to the receiver, and the receiver is used to obtain the information bit stream according to the real OFDM signal. The real-signal OFDM underwater acoustic communication method specifically includes the following steps:

[0025] Step S1: deploy a unitary neural network model (UNN) at the transmitting end, and deploy a deep neural network model (DNN) at the receiving end.

[0026] In this embodiment, the unitary neural network model and the deep neural network model are combined, and the unitary neural network model and the deep neural network model are deployed at the transmitting end and the receiving end respectively. Among them, the input of the unitary neural network model is the modulation data corresponding to the information bit stream, and the output is the precoded data. The unitary neural network model is used to generate precoded data according to the input information bit stream. The input of the deep neural network model is the detection data corresponding to the information bit stream, and the output is the detection result corresponding to the information bit stream. The deep neural network model is used for signal detection to detect the information bit stream. In this embodiment, the final input of the deep neural network model is the overhead packet removal signal, and the final output is the model detection signal.

[0027] Step S2: jointly train the unitary neural network model and the deep neural network model to obtain a trained unitary neural network model and a trained deep neural network model.

[0028] In this embodiment, step S2 jointly trains the unitary neural network model and the deep neural network model to obtain a trained unitary neural network model and a trained deep neural network model, which specifically includes the following steps:

[0029] Step S21, obtain the modulated data s, and divide the modulated data s into G groups, the length of the modulated data s is N, and the length of each group is Get the modulation data set s g (g=0, 1,…, G-1).

[0030] Step S22: according to the modulation data set s g The g-th group of modulated data in generates the g-th unitary precoding matrix W g .

[0031] Step S23: Based on the g-th unitary precoding matrix W g , add constraint W g W g T =I g , to initialize the weights of the unitary neural network model in the transmitter as a unitary matrix, where (.) T represents transpose, I g represents the identity matrix of size g×g.

[0032] Step S24: Perform singular value decomposition on the weights of the unitary neural network model to force the g-th unitary precoding matrix W g Satisfying the unitarity constraint, we obtain the unitary matrix U and the unitary matrix V.

[0033] In this embodiment, the following formula is used to represent the singular value decomposition of the weights of the unitary neural network model:

[0034] W g =UV H .

[0035] Among them, W g represents the g-th unitary precoding matrix, U and V are unitary matrices, (.) H represents the conjugate transpose.

[0036] Step S25: reconstruct the g-th unitary precoding matrix W according to the unitary matrix U and the unitary matrix V. g , and obtain the reconstructed precoding matrix W g '.

[0037] In this embodiment, the following formula is used to represent the reconstruction of the g-th unitary precoding matrix W g :

[0038] W g '=U∑V H .

[0039] Among them, W g' represents the reconstructed precoding matrix, U and V are unitary matrices, ∑ is a rectangular diagonal matrix with non-negative real diagonals, (.) H represents the conjugate transpose.

[0040] Step S26: Based on the reconstructed precoding matrix W g ', with the goal of minimizing the bit error rate, the deep neural network model of the receiving end is trained to obtain a deep neural network model

[0041] In this embodiment, the loss function aims to minimize the bit error rate. The loss function used in the deep neural network model training can be expressed as follows:

[0042]

[0043] Where ζ(θ) represents the loss function, θ is the g-th unitary precoding matrix W g and deep neural network models Parameters, s g Represents the reconstructed precoding matrix W g 'output, Represents a deep neural network model Output.

[0044] Step S27, jump to step S22, until all groups of modulation data are trained to obtain the corresponding g-th unitary precoding matrix W g And the deep neural network model The g-th unitary precoding matrix W obtained by training all groups of modulation data g And the neural network model Splice in sequence to generate the unitary precoding matrix W and the deep neural network model Among them, the unitary neural network model corresponding to the unitary precoding matrix W is used as the trained unitary neural network model, and the deep neural network model As a trained deep neural network model.

[0045] Step S3: input the information bit stream to the transmitting end, and use the trained unitary neural network model to generate the precoded data according to the information bit stream, and send it to the receiving end in the form of the real OFDM signal.

[0046] In this embodiment, step S3 inputs the information bit stream to the transmitting end, generates the precoded data according to the information bit stream using the trained unitary neural network model, and sends it to the receiving end in the form of the real OFDM signal, specifically including the following steps:

[0047] Step S31, performing source coding and interleaving on the input information bit stream to obtain a source coded and interleaved information bit stream.

[0048] Step S32: Generate modulated data using a channel encoder and M-ary pulse amplitude modulation according to the information bit stream after source coding and interleaving.

[0049] Step S33: input the modulated data into the trained unitary neural network model to obtain precoded data.

[0050] Step S34: performing equal-spaced zeroing and overhead packet superposition processing on the precoded data to obtain an overhead packet signal.

[0051] In this embodiment, step S34 performs equal-spaced zeroing and superimposed overhead packet processing on the precoded data to obtain a superimposed overhead packet signal, which specifically includes the following steps:

[0052] Step S341: Based on the precoded data, two diagonal matrices are determined according to the length of the pilot sequence and the number of empty subcarriers; the two diagonal matrices are used to specify the position of the pilot sequence and the position of the empty subcarriers, respectively.

[0053] Step S342: According to the two diagonal matrices, the positions of the designated subcarriers in the precoded data are set to zero to obtain the precoded data after the setting to zero.

[0054] Step S343: Perform superposition overhead packet processing on the precoded data after the zeroing processing to obtain the superposition overhead packet signal.

[0055] Step S35: Perform an IDHT (Inverse Discrete Hartley) transform on the superimposed overhead packet signal to obtain a time domain signal.

[0056] Step S36, perform cyclic prefix processing and up-conversion processing on the time domain signal to obtain the real OFDM signal, and send the real OFDM signal to the receiving end.

[0057] Step S4: using the receiving end to receive the real OFDM signal, and using the trained deep neural network model to perform signal detection on the real OFDM signal to obtain the information bit stream.

[0058] In this embodiment, step S4 uses the receiving end to receive the real OFDM signal, and uses the trained deep neural network model to perform signal detection on the real OFDM signal to obtain the information bit stream, which specifically includes the following steps:

[0059] Step S41: Determine the actual received signal of the receiving end according to the real OFDM signal.

[0060] Step S42: performing synchronization processing, Doppler compensation processing, down-conversion processing and cyclic prefix removal processing on the actual received signal to obtain a prefix-removed signal.

[0061] Step S43: perform phase offset estimation and compensation processing on the de-prefixed signal to obtain an estimated compensated signal.

[0062] Step S44: Perform DFT transformation on the estimated compensation signal to obtain a frequency domain received signal.

[0063] Step S45: perform channel estimation on the frequency domain received signal, determine the time domain impulse response of the estimated channel, and further obtain the frequency response of the estimated channel.

[0064] Step S46: Perform channel equalization and X-transform on the frequency domain received signal according to the frequency response of the estimated channel to obtain a received signal that has undergone channel equalization and X-transform.

[0065] Step S47: remove overhead packets from the received signal that has undergone channel equalization and X-transformation to obtain an overhead packet-removed signal.

[0066] Step S48: input the overhead packet removal signal into the trained deep neural network model for signal detection to obtain a model detection signal.

[0067] Step S49: demodulate and decode the model detection signal to obtain the information bit stream.

[0068] In order to make the technical solution of this embodiment clearer, the specific implementation process of this embodiment is described in detail below in the form of examples.

[0069] like Figure 2 As shown, a real signal OFDM underwater acoustic communication method proposed in this embodiment is described by taking N=2048, P1=512 / 256, and P2=64 as an example. The signal processing flow of the method specifically includes the following steps:

[0070] S1: Jointly train the unitary neural network model at the transmitter and the deep neural network model at the receiver, including the following steps:

[0071] (1) Divide the modulated data s of length N into G groups, each group has a length of Get the modulation data set s g (g = 0, 1, ..., G-1). For the g-th group of modulated data, training is performed to generate the g-th unitary precoding matrix and deep neural network models

[0072] (2) Add constraint W g W g T =I g , to initialize the weights of the unitary neural network model as a unitary matrix, where (.) T represents transpose, I g represents the identity matrix of size g×g. Perform singular value decomposition on the weights of the unitary neural network model to force W g Satisfying the unitarity constraint, the unitarity matrix U and the unitarity matrix V obtained from the singular value decomposition are then used to reconstruct the g-th unitarity precoding matrix W g , and obtain the reconstructed precoding matrix W g ', the singular value decomposition process is: W g =UV H , the reconstruction process is: W g '=U∑V H , where ∑ is a rectangular diagonal matrix with a nonnegative real diagonal, (.) H represents the conjugate transpose.

[0073] (3) Train the deep neural network model to minimize the bit error rate loss function ζ(θ). The loss function ζ(θ) is expressed as follows:

[0074]

[0075] Where θ is the g-th unitary precoding matrix W g and deep neural network models Parameters, s g Represents the reconstructed precoding matrix W g 'output, Represents a deep neural network model The output, s g and The relationship is as follows:

[0076]

[0077] Where D is the DHT matrix, v represents the baseband noise, Λ 1,g and Λ 2,g for The diagonal matrix has the same structure as Λ1 and Λ2. The advanced stochastic gradient descent method with adaptive momentum estimation is used to update θ. The update process is: Among them, η and Respectively represent the learning rate and the gradient of the loss function ζ(θ) with respect to θ, and := represents the update sign. In this embodiment, Ψ is set to 1000 and η is set to 0.001.

[0078] (4) The unitary precoding matrix W obtained by training all groups g and deep neural network models Splice them together in sequence to generate the unitary precoding matrix W and the deep neural network model Among them, the deep neural network model It represents the trained deep neural network model.

[0079] S2: The transmitter inputs the information bit stream and sends the real OFDM signal z(t). Specifically, it includes the following steps:

[0080] S201: Source coding and interleaving of the input information bit stream, followed by channel coding Output symbol of length N. Use M-ary pulse amplitude modulation to generate modulated data s ∈R N×1 .

[0081] S202: Modulate the data s =[s0, s1, …, s N-1 ] Insert the unitary neural network model to generate precoded data s∈R N×1 .

[0082] S203: Set the equally spaced positions of the precoded data s to zero, and superimpose the overhead packet to obtain a superimposed overhead packet signal x , specifically including the following steps:

[0083] (1) Determine two N×N diagonal matrices Λ1 and Λ2 to specify the positions of the pilot sequence and the null subcarriers, respectively. The diagonal elements are expressed as:

[0084]

[0085] Among them, τ1 and τ2 are arbitrary fixed integers in the range of [0, K1] and [0, K2] respectively.

[0086] (2) Set the specified subcarrier position in the precoded data s to zero and superimpose the overhead packet to obtain the superimposed overhead packet signal x , expressed as the following formula:

[0087] x =(I-Λ1-Λ2)W s +Λ1p.

[0088] Wherein, I is an N×N identity matrix, and p is an N×1 training pilot vector.

[0089] S204: Perform IDHT transformation on the superimposed overhead packet signal x to obtain a time domain signal x(t).

[0090] S205: Add a length of T to the time domain signal x(t). CP The cyclic prefix is ​​added and up-converted to obtain a real OFDM signal z(t). In this embodiment, the roll-off coefficient is 0.65.

[0091] Although the transmitter sends a real OFDM signal z(t) to the receiver, the real OFDM signal z(t) is usually subject to various interferences such as channel noise and signal attenuation during transmission. Therefore, the signal actually received by the receiver is not the real OFDM signal z(t) transmitted by the transmitter, but also includes channel noise, etc. In this embodiment, the signal finally received by the receiver is the actual received signal r(t), and the expression of the actual received signal r(t) is:

[0092]

[0093] Among them, r(t) represents the actual received signal, α l represents the time-varying attenuation factor on path l, z(t) represents the real OFDM signal, δ l represents the multipath delay component, ρ represents the Doppler scale factor, v(t) represents the channel noise, t represents the time, L represents the channel delay cutoff, and f c represents the carrier frequency, and p represents the signal power.

[0094] S3: The receiving end obtains the actual received signal r(t), and performs the following signal processing on the actual received signal r(t).

[0095] S301: Perform synchronization, Doppler compensation and down-conversion processing on the actual received signal r(t), and remove the cyclic prefix to obtain a de-prefixed signal r.

[0096] S302: Estimate and compensate for the phase offset ∈ of the de-prefixed signal r to obtain an estimated compensated signal r.

[0097] S303: Perform DFT transformation on the estimated compensation signal r to obtain a frequency domain received signal R, and perform channel estimation on the frequency domain received signal R to obtain a time domain impulse response of the estimated channel The estimation process is expressed as follows:

[0098]

[0099] Where F represents the DFT matrix, is the P1×L submatrix of the DFT matrix, R=Fr is the received signal in the frequency domain, is the receiving pilot allocated at position τ1+K1c1.

[0100] S304: The frequency domain received signal R passes through the frequency domain equalizer q and performs X-transformation to obtain a received signal after channel equalization and X-transformation. The minimum mean square error frequency domain equalizer is used, and the equalizer coefficient is in It is through The channel frequency response obtained by performing N-point fast Fourier transform, σ 2 Represents the variance of the channel noise. The received signal after channel equalization and X-transformation It is expressed as the following formula:

[0101]

[0102] Where M represents the X transformation matrix, which combines the DHT transform with the DFT transform.

[0103] S305: receiving the signal after channel equalization and X-transformation Remove the overhead packet to get the overhead packet removal signal Expressed as Then use the deep neural network model Detect the removal overhead packet signal Get model detection signal

[0104] S306: Detecting signal on the model Demodulate and decode to obtain the information bit stream contained in the real OFDM signal z(t) sent by the transmitter.

[0105] The real signal OFDM underwater acoustic communication method proposed in this embodiment is specifically applied in an underwater acoustic communication scenario. In this underwater acoustic communication scenario, the real signal OFDM underwater acoustic communication system is a single-input single-output system. In this single-input single-output system, the transmitting end is the transmitter, the receiving end is the receiver, the transmitter and the receiver are placed at a water depth of 4m, and the distance between the transmitter and the receiver is 1km. The channel delay is truncated to L=128, and the channel coding rate is 0.5.

[0106] In this embodiment, the number of symbols transmitted is 104, the number of subcarriers is N=2048, wherein the number of pilot subcarriers is set to P1=512 and P2=256 respectively, the number of empty subcarriers is P2=64, and the number of guard interval subcarriers is N cp=512, using M-PAM (M-ary pulse amplitude modulation) modulation. The method proposed in this embodiment is compared with the traditional UWAC-OFDM (orthogonal frequency division multiplexing water acoustic communication method), real signal UWAC-OFDM and traditional ST-based OFDM to verify the effectiveness of the method in this embodiment. Under ideal CFO conditions, the bit error rates of the above methods in the additive white Gaussian noise channel are as follows: Figure 3 As shown in Figure 2, the bit error rate in a perfectly known underwater acoustic channel is Figure 4 As shown. In general, compared with the traditional ST-based method, the method proposed in this embodiment overcomes the problem of deterioration of data detection performance caused by precoding data loss in the traditional ST-based method, so the bit error rate performance is better than the traditional ST-based method. In the case where the channel is a perfectly known UWAC channel, when K1=8, since the method proposed in this embodiment uses a unitary neural network model at the transmitter to achieve diversity, the bit error rate performance is better than the traditional OFDM. Under the condition of estimating the UWAC channel, the bit error rates of each method at different CFO values ​​are shown as follows Figure 5 When the CFO value is 0.05, the traditional DFT-OFDM is more sensitive to the CFO value, but the method proposed in this embodiment can still provide acceptable performance. When the CFO value is 0.15, since the CFO compensation process is required in this case, the bit error rate performance of all methods deteriorates. The bit error rates of each method with and without the CFO compensation process are shown in Figure 6 As shown. It can be seen that the method proposed in this embodiment not only avoids the spectrum efficiency loss caused by the additional transmission of empty subcarriers in the traditional real signal OFDM underwater acoustic communication system, but also shows a stronger bit error rate performance.

[0107] This embodiment uses a unitary neural network model at the transmitting end to generate precoded data, so that the method does not need to include an underwater acoustic channel in the training process, which solves the problem of high computational complexity in the training process of previous methods, and the trained model can be used in any simulation and experimental environment, with universal applicability and a wider range of applications. After eliminating 25% of the symbol subcarriers, the transmitting end replaces these subcarriers with pilot sequences, overcoming the interference caused by the superimposed pilot sequence. The CFO effect is tracked by estimating and compensating for the carrier frequency offset by the empty subcarriers superimposed on the transmitted data, without sacrificing additional subcarriers, thereby improving the transmission rate of the real signal OFDM underwater acoustic communication system. By deploying a deep neural network model for data detection at the receiving end, the degradation of data detection performance is avoided.

[0108] This embodiment proposes a real-signal OFDM underwater acoustic communication method based on deep learning and superposition training. The method uses discrete Hartley transform instead of discrete Fourier transform, generates precoded data by using a unitary neural network model at the transmitter, and replaces these subcarriers with pilot sequences after eliminating 25% of the symbol subcarriers, thereby completely overcoming the interference caused by the additional pilot sequence. Another deep neural network model is deployed at the receiving end to detect the transmitted information bit stream. The two models are jointly trained to avoid the degradation of data detection performance, thereby achieving the purpose of enhancing data detection performance. At the same time, since the unitary neural network model is used to generate precoded data, the maximum diversity gain can be obtained in the multipath channel, which makes it unnecessary to include the underwater acoustic channel in the training process. Therefore, the trained model can be used in any simulation and experimental environment, which can meet the application requirements of various simulation scenarios and experimental environments.

[0109] Based on the same inventive concept, the embodiment of the present application also provides a real signal OFDM underwater acoustic communication system for implementing the real signal OFDM underwater acoustic communication method involved above. The implementation scheme for solving the problem provided by the system is similar to the implementation scheme recorded in the above method, so the specific limitations in the real signal OFDM underwater acoustic communication system embodiment provided below can refer to the limitations of the real signal OFDM underwater acoustic communication method above, and will not be repeated here.

[0110] In an implementable manner, a real signal OFDM underwater acoustic communication system is provided, the real signal OFDM underwater acoustic communication system comprising a transmitter and a receiver, the transmitter is used to input an information bit stream and then send a real OFDM signal to the receiver, the receiver is used to obtain the information bit stream according to the real OFDM signal. When underwater acoustic communication is performed between the transmitter and the receiver, the real signal OFDM underwater acoustic communication method is executed.

[0111] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0112] This article uses specific examples to illustrate the principles and implementation methods of this application. The description of the above embodiments is only used to help understand the method and core ideas of this application. At the same time, for those skilled in the art, according to the ideas of this application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A real signal OFDM underwater acoustic communication method, characterized in that: The real signal OFDM underwater acoustic communication method is used for underwater acoustic communication between a transmitting end and a receiving end, wherein the transmitting end is used to input an information bit stream and then send a real OFDM signal to the receiving end, and the receiving end is used to obtain the information bit stream according to the real OFDM signal; The real signal OFDM underwater acoustic communication method comprises: Deploy a unitary neural network model at the transmitting end, and deploy a deep neural network model at the receiving end; The input of the unitary neural network model is the modulation data corresponding to the information bit stream, and the output is the precoded data. The unitary neural network model is used to generate the precoded data according to the input information bit stream; the input of the deep neural network model is the detection data corresponding to the information bit stream, and the output is the detection result corresponding to the information bit stream. The deep neural network model is used to detect the information bit stream; Jointly training the unitary neural network model and the deep neural network model to obtain a trained unitary neural network model and a trained deep neural network model; Input the information bit stream to the transmitting end, and generate the precoded data according to the information bit stream using the trained unitary neural network model, and send the precoded data to the receiving end in the form of the real OFDM signal; The real OFDM signal is received by the receiving end, and the trained deep neural network model is used to perform signal detection on the real OFDM signal to obtain the information bit stream.

2. The real signal OFDM underwater acoustic communication method according to claim 1, characterized in that: Jointly training the unitary neural network model and the deep neural network model to obtain a trained unitary neural network model and a trained deep neural network model specifically includes: Get the modulated data s and divide it into G groups. The length of the modulated data s is N and the length of each group is Get the modulation data set s g (g=0, 1, ..., G-1); According to the modulation data set s g The g-th group of modulated data in generates the g-th unitary precoding matrix W g ; Based on the g-th unitary precoding matrix W g , add constraint W g W g T =I g , to initialize the weights of the unitary neural network model in the transmitter as a unitary matrix, where (.) T represents transpose, I g represents the identity matrix of size g×g; The weights of the unitary neural network model are subjected to singular value decomposition to force the g-th unitary precoding matrix W g Satisfy the unitary constraint and obtain the unitary matrix U and the unitary matrix V; Reconstruct the g-th unitary precoding matrix W according to the unitary matrix U and the unitary matrix V g , get the reconstructed precoding matrix W g '; Based on the reconstructed precoding matrix W g ', with the goal of minimizing the bit error rate, the deep neural network model of the receiving end is trained to obtain a deep neural network model Jump to step "According to the modulation data set s g The g-th group of modulated data in generates the g-th unitary precoding matrix W g ", until all groups of modulated data are trained to obtain the corresponding g-th unitary precoding matrix W g And the deep neural network model The g-th unitary precoding matrix W obtained by training all groups of modulation data g And the neural network model Splice in sequence to generate the unitary precoding matrix W and the deep neural network model The unitary neural network model corresponding to the unitary precoding matrix W is used as the trained unitary neural network model, and the deep neural network model As the trained deep neural network model.

3. The real signal OFDM underwater acoustic communication method according to claim 2, characterized in that: The following formula is used to express the singular value decomposition of the weights of the unitary neural network model: W g =UV H ; Among them, W g represents the g-th unitary precoding matrix, U and V are unitary matrices, (.) H represents the conjugate transpose.

4. The real signal OFDM underwater acoustic communication method according to claim 2, characterized in that: The g-th unitary precoding matrix W is reconstructed using the following formula: g : IN g '=U∑V H ; Among them, W g ' represents the reconstructed precoding matrix, U and V are unitary matrices, ∑ is a rectangular diagonal matrix with a non-negative real diagonal, (.) H represents the conjugate transpose.

5. The real signal OFDM underwater acoustic communication method according to claim 2, characterized in that: The loss function used in the training of the deep neural network model is expressed as: Where ζ(θ) represents the loss function, θ is the g-th unitary precoding matrix W g and deep neural network models Parameters, s g Represents the reconstructed precoding matrix W g 'output, Represents a deep neural network model Output.

6. The real signal OFDM underwater acoustic communication method according to claim 1, characterized in that: Inputting the information bit stream to the transmitting end, and using the trained unitary neural network model to generate the precoded data according to the information bit stream, and sending the precoded data to the receiving end in the form of the real OFDM signal, specifically includes: Performing source coding and interleaving on the input information bit stream to obtain a source coded and interleaved information bit stream; Generate modulated data using a channel encoder and M-ary pulse amplitude modulation according to the information bit stream after source coding and interleaving; Inputting the modulated data into the trained unitary neural network model to obtain precoded data; Performing equal-spaced zeroing and superimposed overhead packet processing on the precoded data to obtain a superimposed overhead packet signal; Performing IDHT transformation on the superimposed overhead packet signal to obtain a time domain signal; The time domain signal is subjected to cyclic prefix addition and frequency up-conversion processing to obtain the real OFDM signal, and the real OFDM signal is sent to the receiving end.

7. The real signal OFDM underwater acoustic communication method according to claim 6, characterized in that: The precoded data is subjected to equal-spaced position zeroing and overlay overhead packet processing to obtain an overlay overhead packet signal, specifically comprising: Based on the precoded data, two diagonal matrices are determined according to the length of the pilot sequence and the number of empty subcarriers; the two diagonal matrices are used to specify the position of the pilot sequence and the position of the empty subcarriers respectively; According to the two diagonal matrices, positions of designated subcarriers in the precoded data are set to zero to obtain precoded data after the setting to zero; The precoded data after the zeroing process is subjected to superimposed overhead packet processing to obtain the superimposed overhead packet signal.

8. The real signal OFDM underwater acoustic communication method according to claim 1, characterized in that: Receiving the real OFDM signal by the receiving end, and performing signal detection on the real OFDM signal by using the trained deep neural network model to obtain the information bit stream, specifically includes: Determining an actual received signal at the receiving end according to the real OFDM signal; Performing synchronization processing, Doppler compensation processing, down-conversion processing and cyclic prefix removal processing on the actual received signal to obtain a prefix-removed signal; performing phase offset estimation and compensation processing on the de-prefixed signal to obtain an estimated compensated signal; Performing DFT transformation on the estimated compensation signal to obtain a frequency domain received signal; Performing channel estimation on the frequency domain received signal, determining a time domain impulse response of the estimated channel, and obtaining a frequency response of the estimated channel; According to the frequency response of the estimated channel, performing channel equalization and X-transformation on the frequency domain received signal to obtain a received signal after channel equalization and X-transformation; Performing overhead packet removal processing on the received signal that has undergone channel equalization and X-transformation to obtain an overhead packet removed signal; Inputting the overhead packet removal signal into the trained deep neural network model for signal detection to obtain a model detection signal; The model detection signal is demodulated and decoded to obtain the information bit stream.

9. The real signal OFDM underwater acoustic communication method according to claim 8, characterized in that: The actual received signal is expressed by the following formula: Among them, r(t) represents the actual received signal, α l represents the time-varying attenuation factor on path l, z(t) represents the real OFDM signal, δ l represents the multipath delay component, ρ represents the Doppler scale factor, v(t) represents the channel noise, t represents the time, L represents the channel delay cutoff, and f c represents the carrier frequency, and p represents the signal power.

10. A real signal OFDM underwater acoustic communication system, characterized in that: The real-signal OFDM underwater acoustic communication system includes a transmitter and a receiver. The transmitter is used to send a real OFDM signal to the receiver after inputting an information bit stream, and the receiver is used to obtain the information bit stream based on the real OFDM signal. When underwater acoustic communication is performed between the transmitter and the receiver, the real-signal OFDM underwater acoustic communication method described in any one of claims 1 to 9 is executed.

Images