Single-carrier multi-user underwater acoustic communication method

Abstract

The invention discloses a single-carrier multi-user underwater acoustic communication method, which belongs to the technical field of underwater acoustic communication, and is characterized in that after information source bit information is subjected to channel coding and interleaving, the information of each user is modulated to a carrier phase, and a signal is subjected to block processing. In the receiver, interference elimination signals of all users are iterated last time, channels of all users are updated through a time correlation method, interference elimination is carried out, the signals are input into a passive time reversal mirror, multi-channel signals are combined into single-channel signals, and the single-channel signals and the updated channels are input into a direct adaptive equalizer. A bit sequence is obtained after de-interleaving and channel decoding, the sequence enters next iteration after interleaving and mapping, and final decoding is achieved through multiple iterations. According to the invention, the precision of channel estimation in a time-varying environment is improved, and strong multiple access interference suppression and elimination capability is provided; according to the channel estimation method, high precision is realized by using time correlation, and the calculation complexity of the system is reduced.

Description

Technical Field

[0001] The present invention provides a single-carrier multi-user underwater acoustic communication method, a multi-user underwater acoustic communication method based on continuous weight update sparse Bayesian learning channel estimation of approximate message passing and channel update decision for improving time-varying environment adaptation, and belongs to the field of underwater acoustic communication. Background Art

[0002] While underwater acoustic communication using multiple-input multiple-output and multi-user schemes greatly improves the communication rate in various scenarios, it faces the following major technical challenges: large Doppler shift, extremely long propagation delay, fast channel time variation, limited bandwidth and multiple access interference. Since the receiver receives multiple data streams at the same time, multi-user systems will generate not only inter-symbol interference but also multiple access interference, and as the number of users increases, multiple access interference gradually becomes the main interference in multi-user systems.

[0003] As a promising solution to combat harsh multi-access interference underwater acoustic channels, Turbo equalization has received widespread attention in recent years. It greatly reduces the bit error rate of underwater acoustic communication by transmitting soft information between the equalizer and the decoder. When processing multi-channel received signals, the computational complexity of multi-channel Turbo equalization used in multi-user communication becomes increasingly unacceptable. The passive time reversal technology can convert multi-channel signals into single-channel signals and compress multipath delay channels, so that the equalizer only needs very short taps, which greatly reduces the computational complexity of the channel system. Combined with continuous interference cancellation, it achieves excellent multi-access interference suppression capabilities. The performance of continuous interference cancellation and passive time reversal depends on the accuracy of channel estimation. Sparse Bayesian learning algorithms have excellent performance, but also have high complexity. In order to further improve performance and reduce complexity, an approximate message passing algorithm based on belief propagation and compressed sensing, as well as a generalized approximate message passing algorithm applicable to a wider range of scenarios, are proposed.

[0004] Traditional time domain receivers mainly consider using only training sequences for channel estimation, which cannot adapt to underwater acoustic time-varying channels. In order to improve the estimation accuracy of time-varying channels, sub-block partitioning-based channel estimation methods have been widely used in underwater acoustic channel estimation. Given the limited length of the training sequence, sub-block partitioning-based channel estimation usually requires the use of soft symbol feedback from the equalizer or decoder for channel estimation. This method is called a decision-oriented channel estimation algorithm and has been widely used in underwater acoustic communications. In the channel estimation problem based on sub-block partitioning, multi-task sparse Bayesian learning can use the statistical correlation of the channel for joint estimation, and has derived a variety of channel estimation methods. Summary of the invention

[0005] In view of the above background technology, the problem to be solved by the present invention is to provide a multi-user underwater acoustic communication method based on sparse Bayesian learning channel estimation with continuous weight update based on approximate message passing and channel update decision with improved adaptation to time-varying environment, which can achieve high-precision decoding of multi-user underwater acoustic communication with low computational complexity.

[0006] To solve the above problems, the present invention proposes a single-carrier multi-user underwater acoustic communication method. Considering a single-carrier transmission system based on phase shift keying modulation, the source bit information is channel encoded and interleaved, and then the information of each user is modulated onto the carrier phase. Considering the time-varying environment, the signal is processed in blocks. When the receiver detects the symbols of all users in the current block, the next block is processed, and this process is continued until all data processing is completed.

[0007] The receiver uses the interference cancellation signal of all users in the previous iteration to update the channels of all users using the sparse Bayesian learning channel estimation method with continuous weight update based on approximate message passing. Then, the updated channels and the original received signals are used to cancel interference. The new interference cancellation signal of all users is input into the next iteration and the passive time reversal mirror. The passive time reversal mirror merges the multi-channel signals into a single-channel signal. The single-channel signal and the updated channel are then input into the direct adaptive equalizer. The equalized soft symbols are deinterleaved and channel decoded to obtain a bit sequence. The sequence enters the next iteration after interleaving and mapping. The receiver achieves final decoding through multiple iterations.

[0008] A multi-user underwater acoustic communication method based on a single carrier comprises the following steps:

[0009] Step 1: Establish an initial block model based on the interference cancellation signals of each hydrophone m∈[1,M], M is the number of receiving array elements, the index of the initialized target data block i=1, i∈[1,I], I is the number of blocks, the index of the target user n=1, n∈[1,N], N is the number of transmitting users, the index of the iteration number iter=1, the maximum number of iterations ITER; interference elimination signal Interference removal from the previous iteration;

[0010] Step 2: Perform sparse Bayesian learning channel estimation based on continuous weight update of approximate message passing on the interference cancellation signal, and use factor graphs for message passing between different blocks to utilize the time correlation between channels of different blocks, improve the algorithm convergence speed, reduce complexity, and improve estimation accuracy.

[0011] Enter the number of iterations K max , number of sub-blocks I, mutual information Interference Cancellation Signal Prior mean The number of inner iterations is T; initialization P = P (0) ,σ=σ (0) , k = 1;

[0012] Step 3: Update μ after T iterations by following the following steps i , V i and initialization

[0013] 1)

[0014] 2)

[0015] 3)

[0016] 4)

[0017] 5)

[0018] 6)

[0019] in, Then the estimated mean of I sub-blocks is calculated by the following steps: With variance

[0020] 7)

[0021] 8)

[0022] 9)

[0023]

[0024] Finally, the shared hyperparameters P and variance σ are calculated. 2 :

[0025]

[0026] The channel update decision can only be used after obtaining the interference cancellation signal and prior mean of each user. Instead of performing continuous interference cancellation in each iteration, an accurate channel estimate is obtained first. The estimated channel can accurately reflect the channel state of the current sub-block, while also improving the performance of subsequent continuous interference cancellation and passive time reversal processing.

[0027] Adaptive weight ρ ii′ Update, run the factor graph serially, only the weight factor of the first sub-block is based on the initial estimate μ at the beginning of the algorithm iterationi and V i Calculation, the weight factor update of subsequent sub-blocks needs to take into account the estimated value of the previous sub-block factor graph update and Weight for the current sub-block and The estimated value of is passed to the factor graph of the next sub-block, which has higher accuracy than the initial estimate, and then replaces μ i and V i It is used for message transmission and weight update, and this cycle is repeated until all sub-blocks are updated. This weight update method speeds up the convergence of the algorithm.

[0028] Step 4: k = k + 1, execute step 3 until k = K max , output

[0029] Step 5: n=n+1, execute steps 2 to 4 until the channels of all users are estimated;

[0030] Step 6: Initialize the interference cancellation signal Feedback prior mean n=1, i=1;

[0031] Step 7: Perform continuous interference cancellation, and the interference of other users to the target user n is reconstructed as:

[0032]

[0033] Where n represents the target user, l represents the other interfering users, and X represents the a priori symbol matrix fed back from the previous iteration. The reconstructed interference is subtracted from the original received signal to obtain the multi-channel data corresponding to the target user;

[0034] Step 8: Perform inter-block interference elimination using The prior mean of the previous data block is used to reconstruct the inter-block interference. The interference of the i-1th data block to the i-th data block can be expressed as:

[0035]

[0036] Among them, N b Denotes the length of the data block, and L denotes the length of the channel. Subtract the inter-block interference from the multi-channel data obtained in step 5, and only subtract it from the first L-1 symbols of the current data block, keeping the other symbols unchanged, and obtain the data containing only the i-th block of the target antenna and the noise

[0037] Step 9: Multi-channel data merging based on passive time reversal mirror. The merged data is

[0038] Step 10: Use a decision feedback equalizer based on the normalized least squares adaptive algorithm to equalize the combined data and obtain the symbol estimate of the i-th sub-block of user n

[0039] Step 11: n=n+1, execute steps 6 to 10 until all users of the target data block are detected;

[0040] Step 12: i=i+1, execute step 6-step 11 until all users in all blocks are detected;

[0041] Step 13: Then combine the estimated values ​​of all sub-blocks to get the symbol estimate of any user n

[0042]

[0043] Step 14: Demap the equalizer output to obtain log-likelihood ratio information Again Deinterleave to get decoder prior information Will Input decoder for decoding, output decoder external information and bit sequence Use interleaving sequence to get equalizer prior information right By mapping, we can get the symbolic prior mean of user n and used in the next receiver iteration;

[0044] Step 15: Continue to execute step 2 to step 14 until the number of iterations iter reaches the maximum number of iterations ITER.

[0045] The present invention proposes a single-carrier time-domain multi-user underwater acoustic communication method, which improves the estimation accuracy of the time-varying channel by utilizing the time correlation between sub-blocks, while reducing the computational complexity of the system; through the improved channel update decision, the channel of each sub-block is first updated, and then the channel estimation value is used to perform continuous interference elimination and passive time reversal, and the system performance is improved by adapting to the time-varying channel; by serially running the factor graph, the message of the sub-block that has run the factor graph is transmitted in the subsequent sub-block, and the message and weight factor are updated in turn, which speeds up the convergence speed; and the approximate message passing algorithm is used in the sparse Bayesian learning algorithm to reduce the complexity. Its advantages are summarized as follows: (1) It has a strong ability to eliminate and suppress multiple access interference under time-varying channels; (2) It has a low computational complexity; (3) It can achieve high-precision time-varying channel estimation.

[0046] Compared with the existing technology, the present invention has the following improvements: by estimating the channel first and then eliminating the interference, the channel update decision can better match the time-varying nature of the channel, improve the accuracy of channel estimation, and further improve the ability to suppress and eliminate multiple access interference; the continuous weight update sparse Bayesian learning channel estimation method based on approximate message passing can utilize the time correlation of the channel, and pass the information of other blocks to the current block through the factor graph to accurately estimate the time-varying channel, and the use of the approximate message passing algorithm greatly reduces the complexity of the algorithm; the continuous update adaptive weight scheme utilizes the updated message, which can accelerate the convergence speed of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a flow chart of multi-user underwater acoustic communication technology based on continuous weight update sparse Bayesian learning channel estimation and new channel update decision based on approximate message passing;

[0048] Figure 2 It is the test deployment location diagram;

[0049] Figure 3 It is the bit error rate statistics corresponding to the 2×4 multi-user system;

[0050] Figure 4 is the bit error rate statistics corresponding to the 3×10 multi-user system;

[0051] Figure 5 is the bit error rate statistics corresponding to the 3×10 multi-user system;

[0052] Figure 6 It is the bit error rate statistics corresponding to the 4×14 multi-user system;

[0053] Figure 7 It is the bit error rate statistics corresponding to 8PSK modulation;

[0054] Figure 8 It is the bit error rate statistics corresponding to 16QAM modulation. DETAILED DESCRIPTION

[0055] The present invention is described in more detail below with reference to the accompanying drawings and examples.

[0056] The basic implementation scheme of the present invention is: based on a single carrier system of phase shift keying modulation, the information of each user is modulated onto the carrier phase, and the source bit information is channel encoded and interleaved respectively through a convolution encoder and an interleaver. In the first iteration, since there is no prior symbol information, the receiver can only work in a block-by-block iterative manner, using the estimated channel and equalized symbols of the previous block to perform continuous interference elimination and inter-block interference elimination, and then using the interference elimination signal to update the channel of the current block. When the receiver detects the symbols of all antennas in the current block, it processes the next block and continues this process until all data processing is completed. Starting from the second iteration, with the prior symbols and the interference cancellation signal provided by the previous iteration, we can first use the sparse Bayesian learning channel estimation based on approximate message passing to estimate the channel of each block of all users, and then reconstruct and eliminate the interfering user signal and inter-block interference signal in the received signal. After that, the interference cancellation signal of the target user and the estimated channel are input into the passive time reversal mirror. The passive time reversal mirror merges the multi-channel signals into a single-channel signal, and then the single-channel signal is input into the adaptive equalizer. The symbols output by the equalizer are deinterleaved and channel decoded to output a bit sequence. After interleaving and mapping, the bit sequence enters the next iteration again. The receiver achieves final decoding through multiple iterations.

[0057] Combined with the accompanying drawings, Figure 1 As shown, the steps of the present invention are as follows:

[0058] Step 1: Establish a multi-user underwater acoustic communication system. The multi-user system can be regarded as a linear superposition of multiple single-user systems. Assuming that the preprocessing can completely eliminate the influence of Doppler frequency shift and the inter-block interference has been eliminated, the received signal model can be described as:

[0059]

[0060] in, i is the data block index, N b The length of each data block, x n,i is the i-th block of data of the n-th user, is the noise vector, H m,n,i is the channel h m,n,i =[h m,n,i (0),h m,n,i (1),…,h m,n,i (L-1)], the corresponding channel matrix can be expressed as:

[0061]

[0062] Where L is the length of the channel, the index of the initialized target data block i=1, i∈[1,I], I is the number of blocks, the index of the target antenna n=1, n∈[1,N], N is the number of transmitting antennas, the index of the iteration number iter=1, the maximum number of iterations ITER, and the interference cancellation signal from the previous iteration is

[0063] Step 2, channel estimation: the interference cancellation signal is estimated by continuous weight update sparse Bayesian learning based on approximate message passing, and the time correlation between channels of different data blocks is used to improve the estimation accuracy; the continuous update of adaptive weight factor method makes the operation of factor graphs of each sub-block change from parallel to serial, and the update of a sub-block is completed. and Then, assign it to μ i and V i , which is used in the factor graph of the next sub-block. The green variable node represents the updated and The variable node h of the sub-block i′ Passed to factor node f i,i′ The mean and variance of the Gaussian message are the reassigned μ i′ and V i′ . Factor node f i,i′ Passed to variable node h i The news is also updated by μ i′ and V i′ Calculated. and The variable node h of the sub-block i′ When the message is updated, the weight factor ρ needs to be updated accordingly ii′ , to measure the importance of these messages, and finally update the current sub-block and According to the serial operation principle of the factor graph, only the weight factor of the first sub-block is based on the initial estimate μ at the beginning of the algorithm iteration. i and V i Calculation, the weight factor update of subsequent sub-blocks needs to take into account the estimated value of the previous sub-block factor graph update and

[0064] Step 3, n=n+1, execute step 2 until the channels of all users are estimated, that is, n=N;

[0065] Step 4. Initialization: Interference Cancellation Signal Feedback prior mean n=1, i=1;

[0066] Step 5: Perform continuous interference elimination, and reconstruct the interference of other users to the target user n as follows:

[0067]

[0068] Where n represents the target user, l represents the other interfering users, and X represents the a priori symbol matrix fed back from the previous iteration. The reconstructed interference is subtracted from the original received signal to obtain the multi-channel data corresponding to the target user;

[0069] Step 6: Perform inter-block interference elimination using The prior mean of the previous data block is used to reconstruct the inter-block interference. The interference of the i-1th data block to the i-th data block is reconstructed as:

[0070]

[0071] Among them, N b Denotes the length of the data block, and L denotes the length of the channel. Subtract the inter-block interference from the multi-channel data obtained in step 5, and only subtract it from the first L-1 symbols of the current data block, keeping the other symbols unchanged, and obtain the data containing only the i-th block of the target antenna and the noise

[0072] Step 7, passive time reversal processing: multi-channel data merging based on passive time reversal mirror, the merged single-channel signal data It is expressed as:

[0073]

[0074] Among them, q m,nn,i and q m,nl,i are the channels h of antenna n and antenna l respectively m,n,i and h m,l,i and The result of convolution, Z l,i is the residual part of the information sequence of user l; W n,i is the noise on each hydrophone and The result of convolution and summation;

[0075] Step 8: Use a decision feedback equalizer based on the normalized least squares adaptive algorithm to equalize the combined data to obtain the symbol estimate of the i-th sub-block of user n. It is expressed as:

[0076]

[0077] in is a feedforward filter, is the feedback filter, is the output of passive time reversal in the time window [k-K2, k+K1],

[0078] Step 9, n=n+1, execute steps 4 to 8 until all users of the target data block are detected, that is, n=N;

[0079] Step 10, i=i+1, execute step 4-step 9 until all users of all blocks are detected, that is, i=I;

[0080] Step 11: Then combine the estimated values ​​of all sub-blocks to obtain the symbol estimate of any user n

[0081]

[0082] Step 12: Demapping: Demap the result of the equalizer output to obtain log-likelihood ratio information. Deinterleaving, channel decoding: Deinterleave to get decoder prior information Will Input decoder for decoding, output decoder external information and bit sequence

[0083] Use interleaving sequence to get equalizer prior information right By mapping, we can get the symbolic prior mean of user n and used in the next receiver iteration;

[0084] Step 13, iter=iter+1, execute steps 1 to 12 until the number of iterations iter reaches the maximum number of iterations ITER.

[0085] In a specific embodiment, the test data processing of the present invention is:

[0086] (1) Test conditions and parameters:

[0087] In December 2023, a communication experiment was conducted in the South China Sea. The experiment used a 1×32 SIMO system. The data collected in the experiment verified the performance of the receiver. The system parameter settings are shown in Table 1. The bottom of the 32-element receiving array is fixed to the seabed by weights and vertically suspended in the sea by a floating platform. The unit spacing is 2m, and the array element M1 is 72m from the water surface. The distance between the transmitting platform and the receiving array is 1km and 3km respectively. The water depth of the transmitting transducer is 30m and 56m at a distance of 1km, and 36m and 71m at a distance of 3km. The deployment is as follows: Figure 2 shown.

[0088] Table 1

[0089]

[0090] (2) Experimental data processing

[0091] Table 2 shows the decoding performance corresponding to the quadrature phase shift keying (QPSK) modulation of the 2×4, 3×10 and 4×14 multi-user systems, including the bit error rate performance and the output signal-to-noise ratio performance. The results given in the table are the average of the results of all users of multiple groups of data, and all achieve error-free decoding, indicating the effectiveness of the receiver of the present invention.

[0092] Table 2

[0093]

[0094] Figure 3 and Figure 4 The error performance comparison of receivers using different channel update decisions in a 2×4MU system and a 3×10MU system using QPSK modulation is described, where I represents a receiver that uses a linear minimum mean square error channel estimation and then continuously cancels interference and then updates the channel, II represents a receiver that uses a sparse Bayesian learning algorithm to estimate the channel from the second iteration, and III represents a receiver that uses a channel update decision of first estimating the channel and then canceling interference from the second iteration. Figure 3 and Figure 4 The percentage of users reaching the specified bit error rate range after each iteration is shown. It can be seen that the receiver III using the channel update decision of the present invention has the best performance, and the proportion of users with 0 bit error rate is always the highest.

[0095] Figure 5 and Figure 6 The present invention describes a comparison of the error performance of receivers using different channel estimation methods in 3×10 and 4×14 multi-user systems using QPSK modulation, wherein receiver IV represents a receiver using a factor graph-based multi-task sparse Bayesian learning channel estimation algorithm from the second iteration of III, and receiver V represents a receiver using the continuous weight update sparse Bayesian learning channel estimation based on approximate message passing of the present invention from the second iteration of III, Figure 5 and Figure 6 The percentage of users reaching the specified bit error rate range after each iteration is shown. It can be seen that the receiver V using the channel estimation method of the present invention has the best performance, the fastest convergence, and the proportion of users with zero bit error rate is always the highest.

[0096] Figure 7 and Figure 8The error performance comparison of different receivers in a 2×14 multi-user system using 8PSK modulation and a 2×16 multi-user system using 16QAM modulation under different modulation schemes is described. It can be seen that regardless of the modulation system, the performance of the receiving V of the present invention is superior and robust.

Claims

1. A single-carrier multi-user underwater acoustic communication method, characterized in that: Considering a single-carrier transmission system based on phase-shift keying modulation, after the source bit information is channel-coded and interleaved, the information of each user is modulated onto the carrier phase. Considering the time-varying environment, the signal is processed in blocks. When the receiver detects the symbols of all users in the current block, it processes the next block and continues this process until all data processing is completed; The receiver uses the interference cancellation signal of all users in the previous iteration to update the channels of all users using the sparse Bayesian learning channel estimation method with continuous weight update based on approximate message passing. Then, the updated channels and the original received signals are used to cancel interference. The new interference cancellation signal of all users is input into the next iteration and the passive time reversal mirror. The passive time reversal mirror merges the multi-channel signals into a single-channel signal. The single-channel signal and the updated channel are then input into the direct adaptive equalizer. The equalized soft symbols are deinterleaved and channel decoded to obtain a bit sequence. The sequence enters the next iteration after interleaving and mapping. The receiver achieves final decoding through multiple iterations.

2. A single-carrier multi-user underwater acoustic communication method according to claim 1, characterized in that: The specific steps include: Step 1: Establish an initial block model based on the interference cancellation signals of each hydrophone m∈[1,M], M is the number of receiving array elements, the index of the initialized target data block i=1, i∈[1,I], I is the number of blocks, the index of the target user n=1, n∈[1,N], N is the number of transmitting users, the index of the iteration number iter=1, the maximum number of iterations ITER; interference elimination signal Interference removal from the previous iteration; Step 2: Perform sparse Bayesian learning channel estimation based on continuous weight update of approximate message passing on the interference cancellation signal, using factor graph for message passing between different blocks; Enter the number of iterations K max , number of sub-blocks I, mutual information Interference Cancellation Signal Prior mean The number of inner iterations is T; initialization P = P (0) ,σ=σ (0) , Step 3: Update μ after T iterations by following the following steps i , V i and initialization in, Then the estimated mean of I sub-blocks is calculated by the following steps: With variance Finally, the shared hyperparameters P and variance σ are calculated. 2 : Adaptive weight ρ ii′ Update, run the factor graph serially, only the weight factor of the first sub-block is based on the initial estimate μ at the beginning of the algorithm iteration i and V i Calculation, the weight factor update of subsequent sub-blocks needs to take into account the estimated value of the previous sub-block factor graph update and Weight for the current sub-block and The estimated value of is passed to the factor graph of the next sub-block, which has higher accuracy than the initial estimate, and then replaces μ i and V i Used for message transmission and weight update, and repeat this cycle until all sub-blocks are updated; Step 4: k = k + 1, execute step 3 until k = K max , output Step 5: n=n+1, execute steps 2 to 4 until the channels of all users are estimated; Step 6: Initialize the interference cancellation signal Feedback prior mean Step 7: Perform continuous interference cancellation, and the interference of other users to the target user n is reconstructed as: Where n represents the target user, l represents the other interfering users, and X represents the a priori symbol matrix fed back from the previous iteration. The reconstructed interference is subtracted from the original received signal to obtain the multi-channel data corresponding to the target user; Step 8: Perform inter-block interference elimination using The prior mean of the previous data block is used to reconstruct the inter-block interference. The interference of the i-1th data block to the i-th data block is expressed as: Among them, N b represents the length of the data block, L represents the length of the channel; subtract the inter-block interference from the multi-channel data obtained in step 5, and only subtract it from the first L-1 symbols of the current data block, keeping other symbols unchanged, and obtain the data containing only the i-th block of the target antenna and the noise Step 9: Multi-channel data merging based on passive time reversal mirror. The merged data is Step 10: Use a decision feedback equalizer based on the normalized least squares adaptive algorithm to equalize the combined data and obtain the symbol estimate of the i-th sub-block of user n Step 11: n=n+1, execute steps 6 to 10 until all users of the target data block are detected; Step 12: i=i+1, execute step 6-step 11 until all users in all blocks are detected; Step 13: Then combine the estimated values ​​of all sub-blocks to get the symbol estimate of any user n Step 14: Demap the equalizer output to obtain log-likelihood ratio information Again Deinterleave to get decoder prior information Will Input decoder for decoding, output decoder external information and bit sequence Use interleaving sequence to get equalizer prior information right By mapping, we can get the symbolic prior mean of user n and used in the next receiver iteration; Step 15: Continue to execute step 2 to step 14 until the number of iterations iter reaches the maximum number of iterations ITER.

Images