Underwater acoustic channel estimation method under deep sea reliable sound path

Abstract

The invention discloses an underwater acoustic channel estimation method under a deep sea reliable sound path, and belongs to the technical field of underwater acoustic communication, in a single carrier system based on phase shift keying modulation, a channel estimation method based on variational Bayesian inference is adopted under a turbo iterative equilibrium framework, a reliable sound path channel is modeled into series connection of each cluster of sub-channels, and the reliable sound path channel is modeled into a reliable sound path channel. In the first iteration, a first-order autoregression model is utilized to capture prior information of a previous direct phonon channel so as to support a current subblock, so that joint estimation of symbols and channels is carried out, and for a reflection phonon channel, a successive interference elimination mode is utilized under a VBI framework to carry out detection. In the subsequent iteration, the prior mean value of the current block symbol can be obtained and is used for improving the estimation precision of each sub-channel. According to the method, the estimation precision of the clustering time-varying underwater acoustic channel under the deep sea reliable acoustic path can be remarkably improved.

Description

Technical Field

[0001] The invention relates to a method for estimating an underwater acoustic channel under a deep-sea reliable acoustic path, in particular to a method for estimating an underwater acoustic channel under a deep-sea reliable acoustic path based on variational Bayesian inference, and belongs to the field of underwater acoustic communication. Background Art

[0002] In recent years, more and more researchers have placed hydrophones below the critical depth (the speed of sound at this depth is equal to the maximum speed near the surface) to use the reliable acoustic path (RAP) for underwater positioning and detection, and the effect is much better than that in other areas. As one of the important pathways for deep-sea sound propagation, RAP has the following advantages: RAP is mainly direct sound with low transmission loss. At the same distance, the transmission loss of RAP is 10-20dB lower than that of shallow water channels; RAP has low environmental noise. Experimental observations show that the environmental noise below the critical depth is 5-20dB lower than that above the critical depth. However, the current research on RAP mainly focuses on target detection and positioning, and there are few reports on research on communication technology under RAP.

[0003] In the deep sea reliable acoustic path area, the underwater acoustic channel presents an obvious cluster distribution. Affected by the interface reflection, the sub-channel characteristics of different clusters have significant differences. The direct phonon channel has the characteristics of strong energy, time-invariant and strong sparsity, while the reflected phonon channel has weak energy, strong time-varying and weak sparsity. The existing channel estimation methods are not suitable for the above characteristics of deep sea RAP channels. First, the implementation of accurate underwater acoustic channel estimation under long multipath extension requires more training sequences, which will reduce the rate of the communication system. At the same time, the long multipath extension channel requires larger data blocks, which puts a greater test on the computing power, cache space size, node delay and time-varying channel robustness of the hardware platform. Secondly, the obvious time-varying differences between different sub-channels reduce the performance of the TC-based channel estimation algorithm, because in order to avoid overfitting, the existing methods all assume that the different multipaths of the channel have the same correlation coefficient; in addition, the obvious sparsity differences between different sub-channels will also reduce the performance of the existing sparse reconstruction methods. Therefore, studying high-performance underwater acoustic channel estimation algorithms under deep sea RAP is a direction that needs to be broken through. Summary of the invention

[0004] In view of the above background technology, the main purpose of the present invention is to provide a method for underwater acoustic channel estimation based on variational Bayesian inference under a deep-sea reliable acoustic path, which can achieve high-precision underwater acoustic channel estimation under a deep-sea reliable acoustic path with lower complexity.

[0005] To solve the above problems, the present invention proposes a method for estimating an underwater acoustic channel under a deep-sea reliable acoustic path, comprising the following steps:

[0006] Step 1: Establish an underwater acoustic signal receiving model And establish the corresponding cluster channel model,

[0007]

[0008] Where D represents the number of channel clusters, h d =[h d (1),h d (2),...,h d (L d )] T represents the d-th cluster subchannel, L d represents the length of the d-th cluster subchannel. The corresponding receiving signal model is:

[0009] y=X(1:L1)h1+...+X(LL D +1:L)h D +ω

[0010] =X1h1+...+X D h D +ω,

[0011] Here, X1=X(1:L1) represents the 1st to L1th columns of X.

[0012] Initialize the maximum number of algorithm iterations K, and initialize the algorithm iteration index k = 1;

[0013] Step 2: Initialize the number of subchannels D, subchannel index d, data block index i=1, and turbo iteration number iter=1;

[0014] Step 3: Check the condition. If iter=1, continue to step 4. Otherwise, go to step 7.

[0015] Step 4: Estimate subchannel 1 of the previous data block based on the clustered channel model in step 1;

[0016]

[0017] Among them, <·> represents the expectation of the variable;

[0018] Step 5: Based on the subchannel 1 of the previous data block estimated in step 4, the subchannel 1 of the current data block is solved using a first-order autoregressive model;

[0019]

[0020] in, is the subchannel corresponding to the i-th block of data The hyper prior variance of .

[0021] Step 6: Perform initial channel equalization using subchannel 1 of the current data block obtained in step 5;

[0022] Step 7: Using the result of channel equalization, perform channel estimation for all sub-channels of the current data block, specifically including:

[0023]

[0024] in, is the subchannel corresponding to the i-th block of data The hyper prior variance of .

[0025] The channel estimation algorithm is shown in Table 1;

[0026] Table 1 Channel estimation algorithm

[0027]

[0028] Step 8: Perform channel equalization of the current data block using the estimation results of all sub-channels in step 7;

[0029] Step 9: k=k+1, execute steps 7-8 until k>K;

[0030] Step 10: i=i+1, execute steps 4-9 until i reaches the maximum number of iterations;

[0031] Step 11: Turbo decoding;

[0032] Step 12: iter=iter+1, execute step 3 until iter reaches the maximum number of iterations;

[0033] Step 13: Output channel estimation results and channel equalization results;

[0034] Step 14: End the algorithm.

[0035] Compared with the existing technology, the present invention has the following improvements: the deep-sea RAP channel is modeled as a series of cluster sub-channels, and each sub-channel can be iteratively estimated through variational Bayesian learning, and its complexity is much lower than the traditional sparse Bayesian learning channel estimation method; in addition, the present invention separates the time-varying nature of each sub-channel, and uses the time correlation of the direct phonon channel between consecutive data blocks to jointly estimate the channel and symbol of the current data block, effectively improving the estimation accuracy of each sub-channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a technical flow chart of the underwater acoustic channel estimation method based on variational Bayesian inference under deep-sea reliable acoustic path;

[0037] Figure 2 is the test scene diagram;

[0038] Figure 3 It is the transmission signal frame structure;

[0039] Figure 4 is the bit error rate statistics corresponding to the decoding of the first packet of data;

[0040] Figure 5 is the bit error rate statistics corresponding to the decoding of the second packet of data; DETAILED DESCRIPTION

[0041] The present invention is described in more detail below with reference to the accompanying drawings and examples.

[0042] Specific implementation: In conjunction with the attached drawings, a method for estimating underwater acoustic channels under a deep-sea reliable acoustic path of the present invention is as follows: Figure 3 , the steps are as follows:

[0043] (1) An underwater acoustic communication system based on single-carrier time-domain equalization is established. The matrix form of the received signal model can be described as:

[0044] y i =X i h i +ω i

[0045] in,

[0046]

[0047]

[0048] Where i represents the block index, N b Represents the length of the block, X i is the data matrix after inter-block interference elimination;

[0049] (2) Establish a clustered channel model;

[0050]

[0051] And establish the corresponding receiving signal model

[0052] y=X(1:L1)h1+...+X(LL D +1:L)h D +ω

[0053] =X1h1+...+X D h D +ω,

[0054] Initialize the maximum number of algorithm iterations K, and initialize the algorithm iteration index k = 1;

[0055] (3) Initialization: number of subchannels D, subchannel index d, data block index i = 1, number of turbo iterations iter = 1;

[0056] (4) Judgement: Judge the condition. If iter = 1, continue to step 4; otherwise, go to step 7.

[0057] (5) Estimate subchannel 1 of the previous data block based on the clustered channel model of (2);

[0058]

[0059] Where <·> represents the expectation of the variable;

[0060] (6) Based on the subchannel 1 of the previous data block estimated in (5), the subchannel 1 of the current data block is solved using a first-order autoregressive model;

[0061]

[0062] in, is the subchannel corresponding to the i-th block of data The hyper-prior variance of ;

[0063] (7) Perform initial channel equalization on subchannel 1 of the current data block using the solution obtained in (6):

[0064] (8) Using the result of channel equalization, perform channel estimation of all subchannels of the current data block. The channel estimation algorithm is shown in Table 1.

[0065] (9) performing channel equalization of the current data block using the estimation results of all subchannels in (8);

[0066] (10) k = k + 1, execute (8) to (9) until k > K;

[0067] (11) i = i + 1, execute (5) to (10) until i reaches the maximum number of iterations;

[0068] (12) Turbo decoding;

[0069] (13) iter = iter + 1, execute (4) until iter reaches the maximum number of iterations;

[0070] (14) Output channel estimation results and channel equalization results;

[0071] (15)End the algorithm.

[0072] Experimental data processing of this embodiment:

[0073] (1) Test conditions and parameters:

[0074] The RAP underwater acoustic communication experiment was conducted in the South China Sea in May 2023. Our experimental system setup includes a transmitting transducer for transmitting acoustic communication signals and a receiving hydrophone for receiving acoustic communication signals. Figure 2 As shown in the figure, the receiver is fixed and the transmitter is deployed from the ship at a depth of 100m. The seabed in the experimental area is relatively flat, with an average sea depth of about 4325m. At the receiving end, 8 hydrophones are fixed on a buoy near the seabed. The depth of the lowest hydrophone is 4279m, and the distance between the hydrophones is 60m. Table 2 gives the parameters of the communication system in the experiment.

[0075] Table 2

[0076]

[0077] (2) Experimental data processing

[0078] Table 3

[0079]

[0080] Table 4

[0081]

[0082] The transmission signal frame structure in the experiment is as follows Figure 3 As shown in Figure 1, the LFM signals before and after the signal are used for coarse synchronization and Doppler estimation, thereby performing data extraction and Doppler compensation. The Doppler bandwidth is estimated by the peak value of the matched filter output and compensated by resampling. The data block size is set to N b =300 and N b =400, Tables 3 and 4 respectively give the bit error rate (BER) and frame error rate (FER) performance of various algorithms in the second iteration, where the original BER represents the bit error rate before channel decoding, the coded BER represents the bit error rate after channel decoding, and FER represents the percentage of data frames whose coded bit error rate is not 0.

[0083] When the data is divided into N blocks b=300, the traditional SBL-based receiver achieved a coding bit error rate of 2.61% and a frame error rate of 66.7%; by utilizing the correlation between two consecutive data blocks, the TSBL algorithm has improved performance compared with the SBL algorithm, and achieved a bit error rate of 1.93% and a frame error rate of 44.4% after decoding; compared with the above two algorithms, the SBL-joint algorithm alleviates the impact of channel time variation to a certain extent through iterative channel estimation, and obtains a lower bit error rate. However, when the reflected sound subchannel is severely time-varying, the initial channel estimation will bring a large error, so it still has error propagation; on the contrary, the proposed method processes different subchannels separately, separates the time variation and sparsity between different subchannels, thereby effectively improving the performance of the system. After 2 iterations, the proposed method achieves a coding bit error rate of 0.04% and a frame error rate of 8.9%, which is an order of magnitude better than the existing method. When the data block becomes longer, better channel estimation results will be obtained in theory. As can be seen from Table 5, when the data block is N b =400, all methods produce lower frame error rates. However, relative to N b = 300, the SBL algorithm and the TSBL algorithm obtain higher BER results because the increase in block length exacerbates the time-variability between data blocks. On the contrary, the SBL-joint algorithm and the proposed algorithm take the time-variability of the channel into account and obtain better channel estimation performance when the data block length becomes longer.

[0084] Figure 4 and Figure 5 The BER performance of the different algorithms is presented, showing the percentage of data frames that reach the BER range after the second iteration. Figure 4 In the example, the BER of TSBL and SBL-joint algorithms is <10 -3 The percentage of blocks is basically the same, while Figure 5 In the example, the BER corresponding to the SBL-joint is <10 -3 The percentage of blocks with a length of 2 is significantly better than that of the TSBL algorithm. This is because the error of the initial SBL channel estimation is larger when the block length is shorter, which leads to a larger error in the second channel estimation. This situation does not exist for the proposed method, which achieves the best performance in both block lengths. Figure 4 ,exist Figure 5 The encoding BER of all methods is <10 -3 However, for SBL and TSBL, the coding BER>10 -1 This is because as the length of the blocks increases, the time-varying nature of the corresponding channels between different blocks also becomes stronger. At this time, directly using the channel of the previous data block for the current block will result in a larger error.

Claims

1. A method for estimating underwater acoustic channels under a deep sea reliable acoustic path, characterized in that: Here are the steps: Step 1: Establish an underwater acoustic signal receiving model And establish the corresponding cluster channel model, initialize the maximum number of algorithm iterations K, and initialize the algorithm iteration index k=1; Step 2: Initialize the number of subchannels D, subchannel index d, data block index i=1, and turbo iteration number iter=1; Step 3: Check the condition. If iter=1, continue to step 4. Otherwise, go to step 7. Step 4: Estimate subchannel 1 of the previous data block based on the clustered channel model in step 1: Among them, <·> represents the expectation of the variable; Step 5: Based on the subchannel 1 of the previous data block estimated in step 4, the subchannel 1 of the current data block is solved using a first-order autoregressive model; Step 6: Perform initial channel equalization using subchannel 1 of the current data block obtained in step 5; Step 7: Use the result of channel equalization to perform channel estimation for all subchannels of the current data block; Step 8: Perform channel equalization of the current data block using the estimation results of all sub-channels in step 7; Step 9: k=k+1, execute steps 7-8 until k>K; Step 10: i=i+1, execute steps 4-9 until i reaches the maximum number of iterations; Step 11: Turbo decoding; Step 12: iter=iter+1, execute step 3 until iter reaches the maximum number of iterations; Step 13: Output channel estimation results and channel equalization results; Step 14: End the algorithm.

2. The method for estimating underwater acoustic channels under a deep sea reliable acoustic path according to claim 1, characterized in that: Clustered channel model in step 1: Where D represents the number of channel clusters, h d =[h d (1),h d (2),...,h d (L d )] T represents the d-th cluster subchannel, L d represents the length of the d-th cluster subchannel; the corresponding receiving signal model is: y=X(1:L1)h1+...+X(LL D +1:L)h D +ω =X1h1+...+X D h D +oh, Here, X1=X(1:L1) represents the 1st to L1th columns of X.

3. The method for estimating underwater acoustic channels under a deep sea reliable acoustic path according to claim 1, characterized in that: In step 5, the channel result of the current data block is solved according to the channel result of the previous data block, which specifically includes: in, is the subchannel corresponding to the i-th block of data The hyper prior variance of .

4. The method for estimating underwater acoustic channels under a deep sea reliable acoustic path according to claim 1, characterized in that: In step 7, channel estimation of all subchannels of the current data block is performed according to the channel equalization result, and the estimation result of each subchannel is output.

Images