Time-varying underwater acoustic communication method based on superposition training sequence

Abstract

The invention provides an underwater acoustic communication method for a time-varying underwater acoustic channel based on a superposition training sequence. Specifically, the invention relates to a technology for realizing underwater-acoustic communication in a time-varying channel by combining a periodic superposition training sequence with a frequency domain iterative equalization technology. The invention provides an underwater acoustic communication technology under a time-varying channel based on a superposition training sequence and frequency domain soft equalization. An equalizer mainly comprises three parts: channel coarse estimation based on the superposition training sequence; estimating a time-varying channel based on Gaussian mixture distribution and a hidden Markov chain; and carrying out soft iteration equalization based on a time-varying channel estimation result. On the basis, aiming at a complex Doppler effect and a time-varying channel generated in a UUV mobile communication process, a frequency domain equalization and channel method based on a molecular block is provided, so that the channel and equalization can be tracked more effectively, and the data of a pool test and an underwater communication test of the Songhua Lake are processed and analyzed, so that the reliability of the UUV mobile communication is improved. The result shows that the method can realize the underwater acoustic communication under the time-varying channel under the condition of high signal-to-noise ratio.

Description

Technical Field

[0001] The invention belongs to the technical field of underwater mobile communications, and in particular relates to a time-varying underwater acoustic communication method based on superimposed training sequences. Background Art

[0002] Although satellite remote sensing can obtain a large amount of ocean observation data, it cannot penetrate deep below the sea surface. In order to perceive the temporal and spatial changes in the ocean, various ocean observation technologies such as unmanned underwater vehicles (UUVs) and underwater sensor networks have developed rapidly. UUVs can carry a variety of sensors to obtain underwater environmental data and transmit the data back to the surface mother ship through underwater acoustic communication technology. On the one hand, the Doppler effect and channel time variation will occur during the UUV mobile communication process, causing the received signal to be distorted, so it is necessary to estimate and compensate for the Doppler; on the other hand, the underwater acoustic channel is considered to be the most difficult wireless communication channel. In order to overcome the serious multipath expansion and rapid fading of the underwater acoustic channel, iterative signal processing is usually required, which is computationally intensive and inadequate for UUV platforms with limited resources. To this end, this method explores the design of point-to-point reception using a single transducer transmitter and a single hydrophone receiver, combined with the method of channel estimation using superimposed training sequences and frequency domain soft equalization technology, to achieve a method for robust underwater acoustic communication of UUVs, and for channel conditions with fast time variation, a molecular block processing method is used to overcome the problem of mismatch between channel estimation results and data to be equalized. Summary of the invention

[0003] The present invention proposes a time-varying underwater acoustic communication method based on superimposed training sequences. The method uses a method of superimposed training sequences on the sending signal to transmit the signal, and uses the superimposed training sequences to perform continuous channel estimation at the receiving end, which can overcome the problem of channel variation in mobile underwater acoustic communication. In view of the problem that decoding interference exists in the presence of superimposed training sequence interference, resulting in poor decoding effect, an iterative algorithm based on frequency domain soft equalization and superimposed training sequence interference elimination is proposed.

[0004] The technical solution of the present invention is implemented as follows: a time-varying underwater acoustic communication method based on superimposed training sequences, the method comprising the following steps:

[0005] A signal is generated at a transmitting end of a communication device and a signal is decoded at a receiving end, and a periodic training sequence is linearly superimposed on the signal;

[0006] Iterative channel estimation is performed based on the combination of coarse channel estimation based on superimposed training sequences and iterative equalization in frequency domain;

[0007] According to the iterative channel estimation results, the time-varying channel information is updated based on the hidden Markov chain;

[0008] And based on the time-varying channel update results, soft equalization based on approximate message passing is implemented.

[0009] The object of the present invention is achieved in that a communication machine is divided into a transmitting end and a receiving end.

[0010] At the transmitting end: generate transmission data, which can be a bit stream or encoded audio or picture information; channel coding, interleaving, and code mapping. The current experimental channel coding uses convolutional codes, and the code mapping uses QPSK modulation. Higher-order or lower-order modulation methods can be used according to actual conditions; superimpose the training sequence, and the training sequence and data are linearly added in a certain ratio; upsample the signal and set the appropriate code element width; pulse shaping to reduce out-of-band leakage; modulate the signal onto a high-frequency carrier, add a synchronization signal, and transmit the signal.

[0011] At the receiving end: firstly, the hydrophone received signal is preprocessed, including filtering and amplification, removing out-of-band noise, and amplifying the signal amplitude; matching filters are used for signal synchronization and Doppler coarse compensation; the initial channel estimation is performed by using the characteristics that the first-order statistics of the information sequence is zero and the first-order statistics of the training sequence is not zero; the received signal is converted to the frequency domain, and the estimated channel is used to cancel the interference of the training sequence; the received signal and channel after interference cancellation enter the equalizer. Iterative channel estimation is performed using the equalizer result and the reconstructed signal, where the channel estimation method is a channel estimation method based on Gaussian mixture distribution, and the block time-varying channel information is updated based on the hidden Markov chain; soft iterative equalization based on approximate message passing is performed based on the time-varying channel estimation result; the decoding result is output.

[0012] As a preferred implementation, in the periodic training sequence, the time division multiplexing training sequence is added and replaced by direct linear superposition in the time domain to add the training sequence, and the communication rate is improved by direct linear superposition in the time domain, so as to ensure the traversal of the training sequence in time and avoid the mismatch between the channel information and the observation sequence caused by time variation.

[0013] As a preferred implementation, when performing rough channel estimation, a first-order statistic based on a superimposed training sequence is used for rough channel estimation, followed by iterative equalization in the frequency domain. The signal is reconstructed after equalization, and iterative channel estimation can be performed using the reconstructed signal. The EM-GMM-AMP algorithm based on Gaussian mixture distribution is used to improve the channel estimation accuracy. After providing the channel estimation accuracy, the block algorithm and the AR-HMM model are combined to update the time-varying channel information.

[0014] As a preferred implementation, the equalizer used in the soft equalization adopts frequency domain iterative soft equalization, and the frequency domain iterative soft equalization is an approximate message passing iterative soft-in-soft-out frequency domain iterative equalization based on channel estimation, and the interference of the superimposed training sequence and the influence of the channel are resisted through soft equalization.

[0015] As a preferred embodiment, the communication device includes a transmitting end and a receiving end, wherein the transmitting end generates transmission data, and the transmission data is any one of a bit stream, encoded audio or picture information. When sending data, any one of channel coding, interleaving, and coding mapping is selected for modulation, and then a training sequence is superimposed. The training sequence and data are linearly added in proportion, the signal is upsampled, a matching code element width is set, and the signal is modulated onto a high-frequency carrier and a synchronization signal is added before the signal is transmitted.

[0016] As a preferred implementation, the receiving end first preprocesses the hydrophone received signal, and the preprocessing includes filtering and amplification, removing out-of-band noise and amplifying the signal amplitude; after completing the preprocessing, a matched filter is used to perform signal synchronization and Doppler coarse compensation; the initial channel estimation is performed by utilizing the characteristics that the first-order statistic of the information sequence is zero while the first-order statistic of the training sequence is not zero; the received signal is converted to the frequency domain, and the estimated channel is used to cancel the interference of the training sequence. The received signal and the channel after the interference cancellation enter the equalizer, and the equalizer result and the reconstructed signal are used to perform iterative channel estimation and then the decoding result is output.

[0017] As a preferred embodiment, the iterative channel estimation method adopted is a channel estimation method based on Gaussian mixture distribution, and the block time-varying channel information is updated based on the hidden Markov chain, and the soft iterative equalization of approximate message passing is performed based on the time-varying channel estimation result to output the decoding result.

[0018] After adopting the above technical solution, the beneficial effects of the present invention are as follows: the communication technology adopts the method of superimposing training sequences, which has higher spectral efficiency and faster communication rate than the traditional method, and can estimate the channel in real time, overcoming the problem of mismatch between the estimated channel and the equalizer channel information in time-varying underwater acoustic communication. The frequency domain iterative soft equalization adopted by the equalizer has a small amount of frequency domain equalization calculation compared to the time domain equalization, and the soft equalization method has a good anti-noise effect compared to the hard decision equalization method, and can better overcome the interference of the superimposed training sequence. In addition, the method can be implemented without an array, which is convenient for UUV equipment.

[0019] Specifically: Different from the conventional time-division or frequency-division multiplexing training sequence addition method, this method adopts the superimposed training sequence addition method, which can effectively improve the spectrum utilization rate compared with the conventional method. By superimposing the training sequence, the problem of mismatch between the channel estimation result and the channel used for equalization during block equalization can be solved, and the anti-interference ability of the communication system for time-varying channels can be enhanced. The robustness of the system can be greatly improved through frequency domain iterative soft equalization technology. In addition, the core of this communication technology lies in the estimation of time-varying channels, updating according to time-varying channel information, and equalization using corresponding channel information. The present invention first uses Gaussian mixture distribution to accurately estimate the time-varying channel, and then updates the channel information based on the autoregressive hidden Markov chain (AR-HMM), which is more in line with the distribution characteristics of the time-varying channel.

[0020] In summary, this communication technology has relatively simple equipment requirements and only requires a single-transmit and single-receive form. It is easier to implement than conventional methods and has good anti-noise effect. It is a robust underwater acoustic communication technology that can be used for interconnection and communication between UUVs under mobile conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0022] Figure 1 It is a frequency domain soft equalization communication flow chart based on superimposed training sequence;

[0023] Figure 2 It is a schematic diagram of training sequence superposition;

[0024] Figure 3 is a linear system represented by a factor graph;

[0025] Figure 4 It is a time-varying linear system represented based on factor graph;

[0026] Figure 5 is the FFG factor graph represented based on AR-HMM;

[0027] Figure 6 It is the parameter table of simulation and field test;

[0028] Figure 7 Simulation results for communication system;

[0029] Figure 8 This is a schematic diagram of the water tank test layout;

[0030] Fig. 9 This is the channel estimation result of the water tank test;

[0031] Fig.10 This is a table showing the percentage of decoding errors of different methods under horizontal movement conditions;

[0032] Fig.11 This is a table showing the percentage of decoding errors of different methods under vertical movement conditions;

[0033] Fig.12 This is a table showing the percentage of decoding bit errors for signals with different sub-block lengths at Songhua Lake. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 the present invention.

[0035] Example:

[0036] like Figures 1 to 12 As shown, Figure 1 This is a description of the overall implementation of the system. This technology can achieve robust underwater acoustic communication between UUVs under mobile conditions. The key technology of the present invention lies in the receiver part, so the receiver part is elaborated in detail.

[0037] Channel estimation by superimposing training sequences

[0038] according to Figure 2 , the received signal can be expressed as:

[0039]

[0040] where y = [y T ,y 2T ,…,y kT ] T , H is a circulant channel matrix, w is Gaussian white noise, x=[x1,x2,…,x N ] T is the signal to be transmitted, x n Depend on Mapped, and p=[p1,p2,…,p N ] T is the training sequence. H can be expressed as:

[0041] H=F H DF (2)

[0042] where F is the normalized Fourier transform matrix.

[0043] D=Diag{d1,d2,…,d N} (3)

[0044] in h is the first column in H. According to equations (1) and (2), we can get:

[0045] z1=DFx+(DFp+w′) (4)

[0046] Where w′=Fw, z1=Fy is the frequency domain representation of the received signal.

[0047] Using the least squares algorithm for channel estimation, we first construct a T×L order Toeplitz matrix A, where T is the length of the training sequence and L is the length of the channel to be estimated:

[0048]

[0049] In this way, according to the least squares algorithm, the initial estimation value of the channel can be obtained according to the superimposed training sequence:

[0050]

[0051] 2. Channel equalization based on approximate message passing

[0052] When the channel information is known, the model can be simplified as:

[0053] y=Hx+w (7)

[0054] Among them, y is the received signal, H is the channel matrix, x is the signal to be detected, and w is Gaussian white noise. On this basis, the problem can be solved as follows:

[0055]

[0056] According to the Bayesian formula, the above problem is converted into a MAP problem, and we can get:

[0057] p(x|y,H)∝p(x|H)p(y|x,H)(9)

[0058] Obviously, the distribution of x is independent of H. According to the above model, we can get the factor graph as shown in the following figure. Figure 3 .

[0059] According to the factor graph, the sum-product algorithm expression based on the factor graph can be obtained as:

[0060]

[0061]

[0062] According to the sum-product algorithm, the iterative formula can be obtained:

[0063]

[0064] in, Since the channel is time-varying, the channel can be written as H = [h1,…,h n ], where σ is the estimated noise power. It is assumed that the noise power is Gaussian white noise, so it does not change over time. The definitions of η(·,·) and κ(·,·) functions are:

[0065]

[0066] For MPSK signals, according to the definition of η(·,·) and κ(·,·) functions, it is assumed that the MPSK signal mapping set is You can and Function writing:

[0067]

[0068]

[0069] If M=4, that is, the transmitted signal is a QPSK signal, the above function can be further simplified to:

[0070]

[0071] 3. Accurate channel estimation based on channel distribution and time correlation model

[0072] The previous article adopted a channel estimation method based on superposition training sequence, which is actually LS channel estimation in essence. It just uses the characteristics that the superposition of cyclic training sequence itself has power gain, while the communication signal and white noise cannot be superimposed, so that it is still effective when the signal-to-noise ratio is low. Since this method is very poor in the use of prior information, it is not effective in many practical uses. In order to improve the performance of this method, the channel estimation method is further improved on the basis of rough channel estimation and equalization based on message passing, and an approximate message passing channel estimation method based on time-varying channel model is proposed. Moreover, the superposition training sequence method based on cyclic superposition proposed in the previous article is invalid when the channel changes very quickly. Due to the influence of the channel, the power of each cyclic training sequence cannot be superimposed synchronously, or the situation after superposition is the superposition of each channel, so that the obtained channel is not a single channel, but the accumulation of multiple cyclic sub-block channels. In this way, if the channel matrix is ​​used during equalization, the error caused is very large. In order to overcome this problem, it is assumed that there is a certain relationship between the changes before and after the channel, that is, the change of the signal conforms to the AR-HMM model, that is:

[0073]

[0074] Where β∈(-1,1) is the AR coefficient, which can be expressed in matrix form as:

[0075]

[0076] Furthermore, it is assumed that the channel distribution at each moment obeys the Gaussian mixture distribution, that is:

[0077]

[0078] Because it is generally believed that the underwater acoustic channel is a multipath channel with obvious sparse characteristics, the channel is represented by h t =[h k,t ], so that it can be represented by a Gaussian mixture distribution as follows:

[0079]

[0080] Among them, N C represents a complex Gaussian distribution. Assume h t If there are K independent and identically distributed elements, we can get:

[0081]

[0082] With h t The probability distribution of h can be estimated using the Bayesian optimal solution. t , that is, calculate h t The conditional expectation of:

[0083]

[0084] in:

[0085]

[0086] However, it is very difficult to obtain the solutions of the above two equations, and the amount of calculation is extremely large. In addition, in the process of solving the problems, some prior information ψ is required. t Therefore, after initialization, the AMP algorithm is used for estimation each time, and the EM algorithm is used to estimate the parameter ψ after each iteration. t of learning.

[0087] The AMP-based iterative algorithm is still used, but this time the estimation object is the channel instead of the sequence to be detected. The channel estimation method based on approximate message passing can be updated as follows:

[0088]

[0089] Since the distribution of channels is different from that of signals, the η(·,·) and κ(·,·) functions are updated, and the modified functions under GMM are as follows:

[0090]

[0091] After AMP iteration stops, you can get h k,t The posterior probability density function can be obtained by using Bayes' theorem. h k,t The posterior probability of h k,t The probability of can be written as:

[0092]

[0093] in,

[0094]

[0095]

[0096] However, as can be seen from the previous steps, the hyperparameter ψ in the iteration process is not known, so the EM algorithm is introduced in each iteration to learn the hyperparameter. The learning steps can be written as:

[0097]

[0098] According to the EM algorithm, the iterative formula can be obtained:

[0099]

[0100] And while learning the GMM parameters, the noise parameters can also be obtained by learning:

[0101]

[0102] in,

[0103] 4. Bidirectional channel estimation update based on time-correlation model

[0104] The iterative algorithm only obtains the channel estimation result of the current block. According to the AR-HMM model hypothesis mentioned above, although the channel changes at the previous moment, there is a strong correlation in a short time, especially for the adjacent previous and next channels. In order to better utilize this correlation, the factor graph can be obtained as shown in the attached figure. Figure 4 .

[0105] From the attached Figure 4 It can be seen that although the description based on the AR-HMM model does not have a direct impact on the channel estimation of each time period, it still has a very close impact on the previous and next channels. According to the principle of factor graph message passing, when the estimate of h1 is obtained, the h1 factor graph will no longer have an impact on h2. Only h1 will have an impact on the h2 node. Therefore, removing the influence of other nodes, we can get the following: Figure 5 Factor graph of the AR-HMM network.

[0106] In order to better describe the estimated channel and the relationship between the implicit channel and the final estimated channel, the FFG factor graph is used. Obviously, this description of the channel delay position change is not large, so before performing this step, the impact of channel Doppler must be reduced through Doppler compensation.

[0107] According to the HMM algorithm, combined with the idea of ​​message passing based on factor graph, the forward update coefficient can be obtained as:

[0108]

[0109] In the above formula, h is the estimated channel vector, and V is a transformation of the measurement matrix. Due to the AR-HMM assumption, the following relationship can be obtained:

[0110]

[0111] where α 2 +β 2 =1, obviously, for the first channel, there is no forward channel transmission information, we can get:

[0112]

[0113] Similarly, the backward message update coefficient is obtained:

[0114]

[0115] And the two satisfy the relationship:

[0116]

[0117] For the rightmost channel, there is also no message from the backward direction, so for the last signal:

[0118]

[0119] Finally, by fusing the forward and backward messages at time t, we get the forward-backward algorithm based on the AR-HMM model:

[0120]

[0121] In this way, based on the channel estimated by the EM-GMM-AMP channel in the previous text, the channel estimation result for each sub-block can be obtained, and then the channel estimation result is updated through the HMM-based forward-backward algorithm to update the channel coefficient, and then fed back to the decoding end for iterative decoding and a new round of channel estimation until the iteration is completed. After the channel is updated, according to the system block diagram, jump to the equalizer based on the approximate message passing algorithm to iterate until the specified number of iterations is reached.

[0122] 5. Numerical simulation

[0123] In order to verify the algorithm proposed in this paper, the system was simulated according to the time-varying channel matrix proposed in this paper. The experiment used Milica's time-varying channel model, which is the channel shown in the figure below. It can be seen that the delay of the channel changes with time, so some channels are extracted from the series of time-varying channels as simulation channels. The channel addition method uses the form of a time-varying channel matrix, so after extracting the channel, a Toeplitz-like matrix is ​​constructed based on the extracted channel, and then the channel is passed through the baseband and Gaussian white noise is added.

[0124] The simulation parameters are as follows Figure 6 The simulation and field mobile communication test parameter table in the figure. The simulation results are shown in the attached figure. The algorithm proposed in this paper has obvious performance improvement compared with the reference algorithm based on frequency domain LMMSE equalization and Gaussian distribution channel estimation under time-varying channels, especially in the case of low signal-to-noise ratio, the algorithm proposed in this paper has better effect, and with the increase of signal-to-noise ratio and number of iterations, the performance of the two gradually approaches.

[0125] 6. Experimental data processing

[0126] In order to study the communication effect of the soft iterative system under time-varying channels, a mobile underwater acoustic communication experiment was carried out in a water pool. In order to verify the impact of different time-varying channels on communication performance, two experiments were carried out: horizontal movement and vertical movement. The experimental arrangement is shown in the attached figure. Figure 8 As shown. Under horizontal movement conditions, the maximum distance between the transmitter and the receiver is set to 9m, and then the distance that the signal moves in one cycle is 5m, and the length of one frame of signal is about 5.5s. Roughly estimated, the speed of horizontal movement fluctuates between 0.6m / s and 1m / s. Under vertical movement conditions, because the pool is shallow, it does not move in one direction, but moves up and down, simulating the up and down surges in the case of large waves. Move up and down back and forth at a depth of 2m to 3m. Other parameters in the experiment are as shown in the attached Figure 6 Test parameter table in .

[0127] For underwater acoustic communication, the signal-to-noise ratio is a very important parameter. The blank signal around the synchronization signal is intercepted as noise, and then the signal-to-noise ratio in the experiment is estimated to be:

[0128]

[0129] Based on the water pool test data, the channel estimation results are shown in the attached figure. Fig. 9 As can be seen from the figure, the channel changes very dramatically regardless of horizontal or vertical movement. The difference is that the channel moving horizontally has a very obvious tilt because the transducer moves relative to each other. The tilt in the waterfall chart reflects the influence of Doppler. In the pool, because there are no waves and almost no noise, the channel condition in a static state is as follows Fig. 9 (b), its channel correlation coefficient is as follows Fig. 9 (e) is a highly correlated time-invariant channel. As for the vertically moving channel, we can see that its change is also quite dramatic, but what is interesting is that its correlation coefficient has a significant increase at 1.5s. Because it is moving up and down in the pool, it is inferred that the height at a certain moment is consistent with that at the previous moment, so this relatively high correlation occurs.

[0130] Based on the test data, the following Fig.10 and attached Fig.11 The decoding result of Fig.10 and Fig.11 It can be seen that for both horizontal and vertical movement conditions, after a few iterations, the decoding can be close to zero error under high signal-to-noise ratio conditions such as the water pool, and the number of iterations required for zero error in the vertical movement experiment is even smaller. Moreover, the algorithm proposed in this paper performs better than the reference algorithm. Fig.10In the experiment, the reference algorithm still has bit errors after multiple iterations in the third block, while the channel estimation based on Gaussian mixture model and approximate message passing iterative equalization proposed in this method can achieve zero bit errors. The reason for this is that in addition to the better performance of the algorithm proposed in this paper, it may also be due to the mismatch of correlation coefficients. In the Songhua Lake field test, the problem of block division and correlation coefficient was focused on.

[0131] In the Songhua Lake test, both the transmitter and receiver were on ships, and the horizontal distance between the two ships was about 400m to 1400m. The receiving ship was anchored and basically remained stationary, while the transmitter drifted with the waves. Based on the time and drift distance, it is believed that the transmitter floated at an average speed of about 0.6m / s, and the distance from the receiving ship gradually increased. The parameter configuration in the test is the same as in the attached Figure 6 Medium table.

[0132] The Songhua Lake experiment verifies the effect of the proposed algorithm on time-varying channels when the correlation coefficient and the length of the block are different. The correlation coefficient is related to the speed of channel change. Generally speaking, the smaller the correlation coefficient, the faster the channel changes, and the shorter the block should be. Fig.12 The data processing results of the Songhua Lake experiment are compared, and the decoding effects of different sub-block lengths and the values ​​of the correlation coefficients when the decoding effect is the best are compared. In addition, in order to have a certain comparison effect, each set of data is compared at close distances and long distances. Fig.12 The second line is the signal-to-noise ratio value obtained from the received data test. As can be seen from the table, when the correlation coefficient is relatively low, the short block strategy is more effective, and when the correlation coefficient approaches 1, the system basically degenerates into a system similar to Turbo equalization. Overall, the time-varying channel underwater acoustic communication system proposed in this paper has achieved good results in the time-varying environment experiment.

[0133] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A time-varying underwater acoustic communication method based on superimposed training sequences, characterized in that: The method comprises the following steps: A signal is generated at a transmitting end of a communication device and a signal is decoded at a receiving end, and a periodic training sequence is linearly superimposed on the signal; Iterative channel estimation is performed based on the combination of coarse channel estimation based on superimposed training sequences and iterative equalization in frequency domain; According to the iterative channel estimation results, the time-varying channel information is updated based on the hidden Markov chain; And based on the time-varying channel update results, soft equalization based on approximate message passing is implemented.

2. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 1, characterized in that: In the periodic training sequence, the time-division multiplexed training sequence is added and replaced by direct linear superposition in the time domain to add the training sequence. The communication rate is improved by direct linear superposition in the time domain, and the traversal of the training sequence in time is guaranteed to avoid the mismatch between the channel information and the observation sequence caused by time variation.

3. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 1, characterized in that: When performing rough channel estimation, the first-order statistics based on the superimposed training sequence are used for rough channel estimation, followed by iterative equalization in the frequency domain. The signal after equalization is reconstructed, and the reconstructed signal can be used for iterative channel estimation. The EM-GMM-AMP algorithm based on Gaussian mixture distribution is used to improve the channel estimation accuracy. After providing the channel estimation accuracy, the block algorithm and AR-HMM model are combined to update the time-varying channel information.

4. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 1, characterized in that: The equalizer used in the soft equalization adopts frequency domain iterative soft equalization, which is an approximate message passing iterative soft-in-soft-out frequency domain iterative equalization based on channel estimation, and resists the interference of the superimposed training sequence and the influence of the channel through soft equalization.

5. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 1, characterized in that: The communication device includes a transmitting end and a receiving end, wherein the transmitting end generates transmission data, and the transmission data is any one of a bit stream, encoded audio or picture information. When sending data, any one of channel coding, interleaving, and coding mapping is selected for modulation, and then a training sequence is superimposed. The training sequence and data are linearly added in proportion, the signal is upsampled, a matching code element width is set, and the signal is modulated onto a high-frequency carrier and a synchronization signal is added before transmitting the signal.

6. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 1, characterized in that: The receiving end first pre-processes the hydrophone received signal, and the pre-processing includes filtering and amplification, removing out-of-band noise and amplifying the signal amplitude; After preprocessing, a matched filter is used to perform signal synchronization and Doppler coarse compensation. The initial channel estimation is performed by taking advantage of the fact that the first-order statistic of the information sequence is zero while the first-order statistic of the training sequence is not zero. The received signal is converted to the frequency domain, and the estimated channel is used to cancel the interference of the training sequence. After the interference is canceled, the received signal and channel enter the equalizer, and the equalizer result and the reconstructed signal are used to iteratively estimate the channel and then the decoding result is output.

7. A time-varying underwater acoustic communication method based on superimposed training sequences as claimed in claim 6, characterized in that: The iterative channel estimation method adopted is a channel estimation method based on Gaussian mixture distribution, and the block time-varying channel information is updated based on the hidden Markov chain. The soft iterative equalization of approximate message passing is performed based on the time-varying channel estimation result, and the decoding result is output.

Images