OFDM signal receiving and processing system and method suitable for NAVDAT system

Abstract

The invention provides an OFDM signal receiving and processing system and method suitable for an NAVDAT system. According to the method, the characteristics that NAVDAT system signals have synchronous headers, pilot frequencies and the like are fully considered, firstly, the synchronous headers are used for carrying out correlation operation on received data to search a frame starting point, and a signal frame to be demodulated and decoded is obtained in combination with a constant false alarm detection algorithm; and subsequently, carrying out symbol synchronization to determine the FFT window position. Secondly, in combination with a discrete pilot signal, estimating an integral multiple frequency offset error, namely coarse synchronization, and further realizing estimation of a decimal multiple frequency offset error, namely fine synchronization, so as to realize frequency synchronization. And then, channel estimation is carried out by using the pilot signal and adopting a least square method to obtain the channel frequency response of the pilot, the frequency response of the complete signal is obtained through interpolation, and the rough estimation result is further processed through the designed RCAFCN network to obtain accurate channel estimation, so that the original OFDM signal is recovered. And finally, extracting a data part in the OFDM signal according to a frame structure, realizing soft demodulation through a log-likelihood ratio mode, decoding data of a signaling part by using a polarization code, and decoding data of a service part by using a low-density parity check code, thereby realizing complete demodulation and decoding. The method provided by the invention can effectively reduce inter-symbol interference and inter-carrier interference existing in OFDM signals, compensate channel distortion, reduce noise influence, improve decoding precision, improve communication quality, realize optimization of NAVDAT system baseband processing, and guarantee sailing safety of ships on the sea by combining with an existing marine broadcast system.

Description

Technical Field

[0001] The invention belongs to the field of communication technology, and relates to an OFDM signal receiving and processing system and method suitable for a NAVDAT system, and is suitable for a mid-frequency band marine signal propagation environment. Background Art

[0002] In recent years, with the rapid development of the marine economy, maritime activities around the world have become increasingly frequent. Establishing an efficient and reliable maritime information security service system to provide real-time and accurate safety information services for ships sailing at sea is of great significance to protecting the lives and property of maritime workers. With the review and modernization of the Global Maritime Distress and Safety System (GMDSS), traditional communication technologies and equipment are facing innovation. Among them, the most typical navigation warning system, NAVTEX, uses narrow band direct printing (NBDP) and still occupies a certain market due to its narrow frequency band, wide coverage and low communication costs. However, due to its shortcomings such as low data transmission rate, complex operation and single function, it is imperative to upgrade the NAVTEX system with new technologies. In view of the limited capacity of the existing maritime safety information broadcasting system, the IMO electronic navigation system can guarantee the data transmission speed requirements in the shore-to-ship direction, and the 500kHz channel can provide a good coverage for digital transmission. On this basis, the Radio Rescue Subcommittee proposed the maritime digital broadcasting system NAVDAT (Navigational Data). As a new generation of digital maritime safety information broadcasting system, NAVDAT has a data transmission rate of 15 to 25 kb per second and a near-shore coverage range of 400 nautical miles. It can transmit text, pictures, charts, images, electronic chart update packages and other file formats at high speed, and can meet the growing data volume requirements of coastal safety information.

[0003] Deep learning (DL) is a branch of machine learning. Its main goal is to design and implement a neural network model with multi-layer nonlinear feature extraction and adaptive learning capabilities. Due to its good learning effect on input-output mapping relationships, deep learning is widely used in many fields, such as image processing, computer vision, speech recognition, target detection, and unmanned driving. At the same time, deep learning has also brought major changes to wireless communication systems. There have been a lot of studies based on deep learning in the physical layer and network layer of wireless communication. For example, in channel estimation, CSI feedback, signal detection, beamforming, signal modulation and other directions, deep learning methods can be used to improve system performance. In channel estimation, traditional methods either have low accuracy or high accuracy but require a lot of computing resources and time, and have high computational complexity. Channel estimation methods based on deep learning can improve estimation accuracy and reduce computational complexity by training neural networks to predict channel state information, so they have attracted much attention and research.

[0004] The OFDM system is one of the key technologies of wireless communication systems. It is widely used in many wireless communication networks (such as the NAVDAT system) due to its strong ability to resist multipath fading and high spectrum utilization. In the OFDM system, accurate acquisition of channel state information is crucial for correct data transmission. An excellent channel estimation solution can help understand the channel transmission characteristics, restore the received signal, and achieve channel equalization, thereby improving communication quality and data transmission rate, and enhancing system performance and user experience.

[0005] With the development and application of deep learning technology in the field of machine learning, OFDM channel estimation algorithm based on deep learning has become a research hotspot. Deep learning can learn more accurate mapping relationships from data, effectively solve channel estimation problems, improve the accuracy and efficiency of estimation, and provide reliable and effective technical support for the optimization and development of OFDM systems. Therefore, the present invention designs a neural network based on attention mechanism and residual learning.

[0006] The present invention proposes an OFDM signal receiving and processing system and method suitable for the NAVDAT system. The method fully considers the signal characteristics of the NAVDAT system, effectively detects the starting position of the signal frame, reduces the error through synchronization, adopts cutting-edge deep learning technology, realizes channel estimation through the designed network RCAFCN, and combines polarization code and low-density parity check code to realize signal demodulation and decoding. Summary of the invention

[0007] The present invention proposes an OFDM signal receiving and processing system and method suitable for a NAVDAT system. The method fully considers the characteristics of NAVDAT system signals such as synchronization headers and pilots. First, the synchronization header is used to perform correlation operations on the received data to find the frame starting point, and the signal frame to be demodulated and decoded is obtained in combination with a constant false alarm detection algorithm. Subsequently, symbol synchronization is performed to determine the FFT window position. Secondly, in combination with a discrete pilot signal, the integer frequency offset error is first estimated, i.e., coarse synchronization, and then the estimation of the fractional frequency offset error is further realized, i.e., fine synchronization, thereby realizing frequency synchronization. Then, the pilot signal is also used to perform channel estimation using the least squares method to obtain the channel frequency response of the pilot, and the frequency response of the complete signal is obtained by interpolation. The rough estimation result is further processed by the designed RCAFCN network to obtain an accurate channel estimation, thereby restoring the original OFDM signal. Finally, the data part of the OFDM signal is extracted according to the frame structure, and soft demodulation is realized by means of a log-likelihood ratio. The signaling part data is decoded using a polarization code, and the service part data uses a low-density parity check code to realize complete demodulation and decoding. The method proposed in the present invention can effectively reduce the inter-symbol interference and inter-carrier interference in OFDM signals, compensate for channel distortion, reduce the influence of noise, improve decoding accuracy, enhance communication quality, optimize the baseband processing of the NAVDAT system, and combine with the existing marine broadcasting system to ensure the navigation safety of ships at sea.

[0008] The present invention designs a full convolution neural network RCAFCN based on attention assistance and residual learning for channel estimation. First, the complex form of the channel estimation matrix of the pilot position is obtained by pilot and least square method, the real part and the imaginary part are separated to obtain two matrices, and the channel estimation of the complete OFDM signal is obtained by the DFT interpolation algorithm. The result is regarded as a low-resolution image, and the result after the real channel response is converted from a complex number to a real number matrix is ​​regarded as a high-resolution image, and the set of data is used as the training of the network. Among them, RCAFCN is based on a full convolution neural network FCN. Compared with the traditional neural network, the input feature map will be expanded into a one-dimensional vector in the last few fully connected layers, and the spatial information is lost. FCN can process the spatial features in the data more carefully, and the number of parameters is usually much less than CNN. In the middle layer, a convolution and attention fusion module CAFM is introduced. Because the convolution operation is limited by the local properties and the perception domain, it is insufficient in modeling global features. On the contrary, the attention mechanism is good at extracting global features and capturing long-range dependencies. Convolution and attention complement each other and can model global and local features. In this way, the attention mechanism can be used in the global branch to capture a wider range of data information, while the local branch is more focused on extracting local features to achieve comprehensive denoising. At the same time, the CAFM and FCN parts are improved by introducing residual learning, which can effectively avoid the problem of gradient disappearance or explosion caused by multiple nonlinear operations, and can also accelerate the convergence speed and improve the model performance and generalization ability.

[0009] The OFDM signal receiving and processing system and method applicable to the NAVDAT system of the present invention comprises the following steps:

[0010] Step 1: The onshore transmitter and the ship receiver continue to work, perform normal communication, and obtain frame OFDM signals.

[0011] Step 2: perform correlation calculation based on the locally known synchronization sequence to estimate the data starting point position, and select the most appropriate frame for subsequent demodulation and decoding based on the correlation peak.

[0012] Step 3: Perform symbol timing synchronization based on the discrete pilot phase difference in the frequency domain.

[0013] Step 4: Carrier synchronization is performed on the signal frame to be processed, specifically including the following steps:

[0014] Step 4 (a) selects the OFDM symbols in the signal frame except the synchronization header used to determine the starting point. Due to the characteristics of the NAVDAT system, the positions of the discrete pilots are known. Therefore, it is only necessary to perform sliding correlation and calculation with the received OFDM symbols and the corresponding discrete pilots, and then derive the integer multiple carrier frequency deviation through the offset of the maximum correlation peak, and compensate the signal frame.

[0015] Step 4 (ii): still select the scattered pilot. The scattered pilot of each symbol is known to the receiver. The absolute phase deviation of the symbol is obtained by the correlation between each received symbol and the corresponding local pilot, and then the frequency deviation is obtained by the difference between the absolute phase deviations of two adjacent symbols.

[0016] Step 5: In order to further reduce the bit error rate and improve the communication quality, the data of the pilot position in the received signal is divided by the locally known pilot signal in the frequency domain to obtain the channel frequency response of the pilot position. Then, the channel time-frequency response of a frame of signal is obtained through DFT interpolation. The rough estimation result is used as the input of the designed RCAFCN network, and finally a more accurate channel estimation result is output, and subsequent compensation is performed.

[0017] Step 6: During demodulation, the received signal is not directly judged to be 0 or 1, but the probability information of each symbol being 0 or 1 is output, which is usually expressed as a log-likelihood ratio, that is, the soft demodulation result is calculated using the log-likelihood ratio based on the maximum a posteriori probability criterion.

[0018] Step 7: Decode the soft-demodulated data to restore the original signal, which specifically includes the following steps:

[0019] Step 7 (a) uses a belief propagation polarization decoding algorithm with 50 iterations for the modulation information stream MIS and the transmitter information stream TIS in the data part.

[0020] Step 7 (ii): Decode the data stream DS in the data part using the logarithmic domain belief propagation algorithm of the low-density parity check code.

[0021] Beneficial Effects

[0022] The present invention proposes an OFDM signal receiving and processing system and method suitable for a NAVDAT system. The method fully considers the characteristics of NAVDAT system signals such as synchronization headers and pilots. First, the synchronization header is used to perform correlation operations on the received data to find the frame starting point, and the signal frame to be demodulated and decoded is obtained in combination with a constant false alarm detection algorithm. Subsequently, symbol synchronization is performed to determine the FFT window position. Secondly, in combination with a discrete pilot signal, the integer frequency offset error is first estimated, that is, coarse synchronization, and then the estimation of the fractional frequency offset error is further realized, that is, fine synchronization, so as to realize frequency synchronization. Then, the pilot signal is also used to perform channel estimation using the least squares method to obtain the channel frequency response of the pilot, and the frequency response of the complete signal is obtained by interpolation. Then, the rough estimation result is further processed by the designed RCAFCN network to obtain an accurate channel estimation, thereby restoring the original OFDM signal. Finally, the data part of the OFDM signal is extracted according to the frame structure, and soft demodulation is realized by means of a log-likelihood ratio. A part of the data is decoded using a polarization code, and the other part of the data is decoded using a low-density parity check code to realize complete demodulation and decoding. The method proposed in the present invention can effectively reduce the inter-symbol interference and inter-carrier interference in OFDM signals, compensate for channel distortion, reduce the influence of noise, improve decoding accuracy, enhance communication quality, optimize the baseband processing of the NAVDAT system, and combine with the existing marine broadcasting system to ensure the navigation safety of ships at sea.

[0023] The present invention designs a full convolution neural network RCAFCN based on attention assistance and residual learning for channel estimation. First, the complex form of the channel estimation matrix of the pilot position is obtained by pilot and least square method, the real part and the imaginary part are separated to obtain two matrices, and the channel estimation of the complete OFDM signal is obtained by the DFT interpolation algorithm. The result is regarded as a low-resolution image, and the result after the real channel response is converted from a complex number to a real number matrix is ​​regarded as a high-resolution image, and the set of data is used as the training of the network. Among them, RCAFCN is based on a full convolution neural network FCN. Compared with the traditional neural network, the input feature map will be expanded into a one-dimensional vector in the last few fully connected layers, and the spatial information is lost. FCN can process the spatial features in the data more carefully, and the number of parameters is usually much less than CNN. In the middle layer, a convolution and attention fusion module CAFM is introduced. Because the convolution operation is limited by the local properties and the perception domain, it is insufficient in modeling global features. On the contrary, the attention mechanism is good at extracting global features and capturing long-range dependencies. Convolution and attention complement each other and can model global and local features. In this way, the attention mechanism can be used in the global branch to capture a wider range of data information, while the local branch is more focused on extracting local features to achieve comprehensive denoising. At the same time, the CAFM and FCN parts are improved by introducing residual learning, which can effectively avoid the problem of gradient disappearance or explosion caused by multiple nonlinear operations, and can also accelerate the convergence speed and improve the model performance and generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a frame structure diagram;

[0025] Figure 2 It is the OFDM symbol structure diagram;

[0026] Figure 3 It is a comparison diagram between data and threshold value;

[0027] Figure 4 It is the CFAR detection map;

[0028] Figure 5 is a frequency offset estimation flow chart;

[0029] Figure 6 The overall architecture diagram of the deep learning network for channel estimation;

[0030] Figure 7 is a 4-QAM mapping diagram;

[0031] Figure 8 For the processing of the module;

[0032] Fig. 9 It is the flow chart of the MS decoding algorithm of low density parity check code; Specific implementation plan

[0033] The technical solution of the present invention is further described in detail below with reference to the accompanying drawings:

[0034] like Figures 1 to 8 The method for optimizing baseband processing of OFDM signals in a NAVDAT system shown in the figure specifically comprises the following steps:

[0035] Step 1: The onshore transmitter and the ship receiver continue to work, communicate normally, and obtain the p-frame OFDM signal. The transmission frame structure is as follows: Figure 1 As shown, one frame of signal has 15 OFDM symbols. The first symbol is a locally known synchronization data header sequence, and the subsequent symbols contain the data part. The OFDM symbol structure is as follows Figure 2 As shown, the number of subcarriers for each symbol is 229, a guard interval of length 25 is added, and equally spaced discrete pilots are inserted into other symbols. The discrete pilot interval of the same symbol is 6, and the discrete pilot position of each symbol is shifted right by 2 subcarriers within the valid subcarrier range from the discrete pilot position of the previous symbol. The discrete pilots of 2 symbols separated by 3 symbol intervals are exactly the same.

[0036] Step 2: Perform correlation operations on the entire received data based on the locally known synchronization sequence, and regard the result of the operation as a series of units. Use the constant false alarm detection algorithm in a unit averaging mode to determine whether there is a target, that is, whether an ideal correlation peak appears, and obtain the index value, that is, the frame starting point. The specific method is to first set a protection unit around the test unit, that is, the received signal subcarrier, to eliminate the excess energy of the signal around the test unit, and then select a suitable reference unit around the protection unit. At the same time, since there are several units on the left side of the unit to be tested, in order to prevent the sequence from becoming shorter after the algorithm runs, it is compensated by zero padding. Take the average of the reference units on the left and right sides of the test unit, then take the average value, and then multiply it by the threshold factor α, and then compare the result with the threshold, such as Figure 3 As shown in the figure, the dotted line is the calculated threshold value, and the solid line is the received data. If it is greater than the threshold, it means there is a target, otherwise it is considered that there is no target, so the following is obtained: Figure 4 The result of the threshold factor is 1, which means that the point is detected as the frame start point. The calculation formula of the threshold factor is as follows:

[0037]

[0038] Where N is the number of reference units selected, which is also the length of the OFDM baseband symbol. represents the expected false alarm probability.

[0039] Step 3: Since the pilot symbol phase is known, the timing offset can be estimated by comparing the phase of the received pilot data with the known phase of the transmitted pilot data. Considering the discrete pilots used in the NAVDAT system, they are equally spaced in the frequency domain, and the interval between adjacent pilots is 6 subcarriers, i.e. Δk = 6. Consider the kth and k+Δkth pilot subcarriers of the received signal, and conjugate and multiply the two adjacent pilot subcarriers of the received OFDM symbol to obtain:

[0040]

[0041] where X′ k and X′ k+Δk is the pilot data after demodulating the pilot phase, k represents the subcarrier index, Y k It is the data of the kth subcarrier in the frequency domain at the receiving end, and k is the pilot position index value. τ d is the timing position deviation normalized by the signal sampling time interval. Since the pilot sequence is the same except for the phase being 0 or π, X′ k =X′ k+Δk , so the timing deviation is:

[0042]

[0043] Step 4: Carrier synchronization is performed on the signal frame to be processed. The flow chart is as follows: Figure 5 As shown, the specific steps include:

[0044] Step 4 (a), first realize the integer multiple carrier frequency error estimation, first cyclically shift the frequency domain subcarriers of the received OFDM symbol as a whole, take out the pilot data according to the discrete pilot position, conjugate multiply the adjacent pilot data, and conjugate multiply the locally generated pilot data with the adjacent pilot data, and then perform correlation operation on the two results. The algorithm expression is as follows:

[0045]

[0046] where Y m,k+l and Y m,k+Δk+l They represent the pilot data on the kth and k+Δkth subcarriers of the mth OFDM symbol at the receiving end after all subcarriers are cyclically shifted by l subcarriers and then taken out according to the pilot position, respectively. m,k and X m,k+Δk They respectively represent the pilot data on the kth and k+Δkth subcarriers of the mth OFDM symbol generated locally, L is the window length of the cyclic shift, and P is the total number of pilots.

[0047] Step 4 (ii) uses the scattered pilot fractional frequency offset estimation and compensation, and obtains the absolute phase offset of each received symbol through the correlation between each received symbol and the corresponding local pilot, and then obtains the frequency offset from the difference between the absolute phase offsets of two adjacent symbols, which is expressed as:

[0048]

[0049] in represents the absolute phase deviation of the mth symbol, N s Indicates the sum of the number of subcarriers and guard intervals.

[0050] Step 5: In order to further reduce the bit error rate and improve the communication quality, firstly, the pilot and least square method are used to obtain the complex channel estimation matrix of the pilot position, and the real part and the imaginary part are separated to obtain two matrices. The DFT interpolation algorithm is used to obtain the channel estimation of the complete OFDM signal. The result is regarded as a low-resolution image, and a more refined channel estimation value is obtained through the designed RCAFCN network output for channel compensation. The network structure is as follows: Figure 6 As shown above, the overall structure introduces residual learning, adding the initial input to the output after the middle layer. The first layer is the convolution layer, which is equivalent to extracting and representing the image. The convolution kernel size is 2x2x64 and the step size is 1. The convolution kernel size of the second layer is 5x5x32 and the step size is 1, which is equivalent to nonlinear mapping. The convolution and attention fusion module CAFM and the residual learning-based design are combined to obtain a new module RCAFM, whose structure is as follows: Figure 6 As shown below, it is added to the second layer. RCAFM has two branches, one is the local branch. In order to enhance cross-channel interaction and promote information integration, 1*1 convolution is first used to adjust the channel dimension, and then the channel information is further mixed through channel shuffling operation. Specifically, the input tensor along the channel dimension is divided into groups, where depth-separable convolution is used in each group. Then the resulting output tensors from each group are connected along the channel dimension to generate a new output tensor, and then the features are extracted through 3*3*3 convolution. The local branch can be expressed as:

[0051]

[0052] In the global branch, query, key and value are first generated through 1x1 convolution and 3x3 depth convolution, resulting in three shapes: Next, Q is reshaped into K is reshaped as By calculating these two interactions, we get the attention map The global branch can be expressed as:

[0053]

[0054] where α is a learnable scaling parameter used to control and The final CAFM module output is:

[0055] Y=Y local +Y global (8)

[0056] The output result is then passed through the dropconnect regularization technique and the leaky Relu activation function, and X in The final RCAFM output is obtained by adding them together. Dropconnect prevents network overfitting by introducing randomness, and the resulting sparse connection structure can be used for model compression. The leaky Relu activation function can effectively alleviate the problem of neuron death and accelerate the convergence speed. The network can achieve good estimation results using a small amount of data set.

[0057] Step 6: During demodulation, the probability information of each symbol being 0 or 1 is output, which is usually expressed as a log-likelihood ratio. That is, the soft demodulation result is calculated using the log-likelihood ratio based on the maximum a posteriori probability criterion. The formula is as follows:

[0058]

[0059] Where LLR k represents the LLR output of the kth bit in the corresponding symbol, N0 is the noise power on each symbol, and Represents the symbol set where the kth bit in the constellation point is equal to 0 and 1 respectively. The signal adopts 4QAM modulation, and its constellation mapping is as follows Figure 6 shown.

[0060] Step 7: The NAVDAT system frame data part includes the modulation information stream, the transmitter information stream and the data stream. Different data adopts different encoding and decoding methods, which specifically includes the following steps:

[0061] Step 7 (a) uses belief propagation BP polar decoding for the modulated information stream and the transmitter information stream. Initialize the frozen bit position, save the information subchannel position, calculate the index matrix according to the modulation order, decode the original information bit (before encoding) according to the LLR, input the information position and the MIS and TIS bit position sequence in the frame structure, and output the original information bit (the original information bit before encoding).

[0062] like Figure 7 As shown, Where i,j represents the i-th minimum decoding unit module in the j-th column, mod(a,b) represents the remainder of a divided by b, Indicates rounding down, L p,q and Rp,q Indicates the elements at the corresponding positions in the L and R matrices. The dotted box indicates reading the data, and the solid box indicates assigning / overwriting the data, that is, the solid box data is calculated based on the dotted box data. The calculation rules in the above figure are as follows:

[0063]

[0064] Among them, the f(x,y) function is approximated by f(x,y)=αsign(x)sign(y)min{|x|,|y|}.

[0065] The received LLR is used as the rightmost input data in the factor graph and filled into the n+1th column of L; the prior LLR of the source bit is used as the leftmost input data in the factor graph and filled into the first column of R. If i∈A, A is the information subchannel position, then u i is the information bit, P(u i )=0.5 is always true, so u i The prior LLR of is always 0; if i∈A C , A C is the frozen bit position, then u i = 0 is a frozen bit, P(u i =0)=1, so u i The prior LLR of is positive infinity. Start all 2×2 modules on the nth column, and according to the above operation rules, calculate the data in the two Ls to the left, and do not calculate the data in R to the right. The completion of all 2×2 modules on the nth column means that the nth column of the L matrix is ​​filled with data, and then, by analogy, complete the calculation of all 2×2 modules in the n-1, n-2, ..., 1th columns. Start all 2×2 modules on the 1st column, and according to the above operation rules, calculate the data in the two Rs to the right, and do not calculate the data in L to the left. The completion of all 2×2 modules on the 1st column means that the 2nd column of the R matrix is ​​filled with data, and then, by analogy, complete the calculation of all 2×2 modules in the 2nd, 3rd, 4th, ..., nth columns.

[0066] The above left and right calculations can be regarded as half an iteration respectively, and together they are one iteration. The effect of one iteration is relatively weak. In the case of a wireless channel, it is basically impossible to obtain a correct decoding result. In this case, you only need to repeat the iteration process until the maximum number of iterations T set in advance is reached. The typical value of T is between 20-50. After reaching the maximum number of iterations, bit judgment is required. For the source sequence (u1, u2, ..., u N ) and the codeword sequence (x1,x2,...,x N ), their decision formula is:

[0067]

[0068] The reason why frozen bits and information bits are not distinguished is that for frozen bits u i , the corresponding R i,1 =+∞, according to the above calculation rules, it will be judged as 0.

[0069] Finally, the decoding result can be checked by CRC. The original information bits (generating CRC check bits) and CRC check bits are selected, and the information bits and check bits are combined for checking again. If the remainder is 0, it is output correctly. Alternatively, only the information bits can be selected, and CRC check bits are generated again. They are compared with the original CRC check bits. If they are the same, the decoding result is output.

[0070] Step 7 (ii): The data stream DS in the data part adopts the minimum sum (Min-Sum) decoding algorithm, the algorithm principle is the improved LDPC logarithmic domain BP decoding algorithm, and the node update formula is modified. Set the number of iterations to 50 times, and the algorithm flow is as follows Figure 8 As shown, the likelihood information is first initialized, where the initial likelihood information has been obtained in step six. Then the likelihood information of the check node is processed. Each variable node passes its own probability information to all the check nodes it participates in, and calculates the information obtained by the check node from the variable node. Next is the likelihood information processing of the variable node. The check node passes the likelihood information to the variable node and calculates it. The likelihood information obtained by each variable node from the check node in this iteration is added to the initial likelihood information obtained from the channel output data to calculate the new likelihood information value. If the bit likelihood value is greater than 0, it is judged as 1, otherwise it is judged as 0. In addition, if the encoded codeword c is multiplied by the transpose of the check matrix H T If it is 0 or the maximum number of decoding iterations is reached, the decoding is terminated; otherwise, the iterative process continues from the initial likelihood information.

Claims

1. A method for optimizing baseband processing of OFDM signals in a NAVDAT system, characterized in that: The specific steps include: Step 1: The onshore transmitter and the ship receiver continue to work, perform normal communication, and obtain p-frame OFDM signals. Step 2: Implement frame synchronization based on a locally known synchronization sequence. Step 3: Since the pilot symbol phase is known, the timing offset can be estimated by comparing the phase of the received pilot data with the known phase of the transmitted pilot data. Considering the discrete pilots used in the NAVDAT system, they are equally spaced in the frequency domain, and the interval between adjacent pilots is 6 subcarriers, i.e. Δk = 6. Consider the kth and k+Δkth pilot subcarriers of the received signal, and conjugate and multiply the two adjacent pilot subcarriers of the received OFDM symbol to obtain: where X′ k and X′ k+Δk is the pilot data after demodulating the pilot phase, k represents the subcarrier index, Y k It is the data of the kth subcarrier in the frequency domain at the receiving end, and k is the pilot position index value. τ d is the timing position deviation normalized by the signal sampling time interval. Since the pilot sequence is the same except for the phase being 0 or π, X′ k =X′ k+Δk , so the timing deviation is: Step 4: Carrier synchronization is performed on the signal frame to be processed. Step 5: Realize channel estimation and compensation through the designed RCAFCN network. Step 6: During demodulation, the probability information of each symbol being 0 or 1 is output, which is usually expressed as a log-likelihood ratio. That is, the soft demodulation result is calculated using the log-likelihood ratio based on the maximum a posteriori probability criterion. The formula is as follows: Where LLR k represents the LLR output of the kth bit in the corresponding symbol, N0 is the noise power on each symbol, and They represent the symbol sets where the kth bit in the constellation point is equal to 0 and 1. The signal is modulated using 4QAM. Step 7: The NAVDAT system frame data part includes the modulation information stream, the transmitter information stream and the data stream. Different data adopts different encoding and decoding methods, which specifically includes the following steps: Step 7 (a) uses belief propagation BP polar decoding for the modulated information stream and the transmitter information stream. Initialize the frozen bit position, save the information subchannel position, calculate the index matrix according to the modulation order, decode the original information bit (before encoding) according to the LLR, input the information position and the MIS and TIS bit position sequence in the frame structure, and output the original information bit (the original information bit before encoding). in Where i,j represents the i-th minimum decoding unit module in the j-th column, mod(a,b) represents the remainder of a divided by b, Indicates rounding down, L p,q and R p,q Indicates the elements at the corresponding positions in the L and R matrices. The dotted box indicates reading the data, and the solid box indicates assigning / overwriting the data, that is, the solid box data is calculated based on the dotted box data. The calculation rules in the above figure are as follows: Among them, the f(x,y) function is approximated by f(x,y)=αsign(x)sign(y)min{|x|,|y|}. The received LLR is used as the rightmost input data in the factor graph and filled into the n+1th column of L; the prior LLR of the source bit is used as the leftmost input data in the factor graph and filled into the first column of R. If i∈A, A is the information subchannel position, then u i is the information bit, P(u i )=0.5 is always true, so u i The prior LLR of is always 0; if i∈A C , A C is the frozen bit position, then u i = 0 is a frozen bit, P(u i =0)=1, so u i The prior LLR of is positive infinity. Start all 2×2 modules on the nth column, and according to the above operation rules, calculate the data in the two Ls to the left, and do not calculate the data in R to the right. The completion of all 2×2 modules on the nth column means that the nth column of the L matrix is ​​filled with data, and then, by analogy, complete the calculation of all 2×2 modules in the n-1, n-2, ..., 1th columns. Start all 2×2 modules on the 1st column, and according to the above operation rules, calculate the data in the two Rs to the right, and do not calculate the data in L to the left. The completion of all 2×2 modules on the 1st column means that the 2nd column of the R matrix is ​​filled with data, and then, by analogy, complete the calculation of all 2×2 modules in the 2nd, 3rd, 4th, ..., nth columns. The above left and right calculations can be regarded as half an iteration respectively, and together they are one iteration. The effect of one iteration is relatively weak. In the case of a wireless channel, it is basically impossible to obtain a correct decoding result. In this case, you only need to repeat the iteration process until the maximum number of iterations T set in advance is reached. The typical value of T is between 20-50. After reaching the maximum number of iterations, bit judgment is required. For the source sequence (u1, u2, ..., u N ) and the codeword sequence (x1,x2,...,x N ), their decision formula is: The reason why frozen bits and information bits are not distinguished is that for frozen bits u i , the corresponding R i,1 =+∞, according to the above calculation rules, it will be judged as 0. Finally, the decoding result can be checked by CRC. The original information bits (generating CRC check bits) and CRC check bits are selected, and the information bits and check bits are combined for checking again. If the remainder is 0, it is output correctly. Alternatively, only the information bits can be selected, and CRC check bits are generated again. They are compared with the original CRC check bits. If they are the same, the decoding result is output. Step 7 (ii), the minimum sum (Min-Sum) decoding algorithm is used for the data stream DS in the data part. The algorithm principle is the improved LDPC logarithmic domain BP decoding algorithm, and the node update formula is modified. Set the number of iterations to 50 times. First, initialize the likelihood information, where the initial likelihood information has been obtained in step 6. Then the likelihood information of the check node is processed. Each variable node passes its own probability information to all the check nodes it participates in, and calculates the information obtained by the check node from the variable node. Next is the likelihood information processing of the variable node. The check node passes the likelihood information to the variable node again and calculates it. The likelihood information obtained by each variable node from the check node during this iteration is added to the initial likelihood information obtained from the channel output data to calculate the new likelihood information value. If the bit likelihood value is greater than 0, it is judged as 1, otherwise it is judged as 0. In addition, if the encoded codeword c is multiplied by the transpose H of the check matrix T If it is 0 or the maximum number of decoding iterations is reached, the decoding is terminated; otherwise, the iterative process continues from the initial likelihood information.

2. The method for optimizing baseband processing of OFDM signals in a NAVDAT system according to claim 1, characterized in that: As shown in step 2, a correlation operation is performed on the entire received data according to the locally known synchronization sequence, and the result after the operation is regarded as a series of units. The constant false alarm detection algorithm of the unit average method is used to determine whether there is a target, that is, whether an ideal correlation peak appears and obtain the index value, that is, the frame starting point position. The specific method is to first set a protection unit around the test unit, that is, the received signal subcarrier, to eliminate the excess energy of the signal around the test unit, and then select a suitable reference unit around the protection unit. At the same time, since there are several units on the left side of the unit to be tested, in order to prevent the sequence from becoming shorter after the algorithm runs, it is compensated by filling with zeros. Take the mean of the reference units on the left and right sides of the test unit, then take the average value, and then multiply it with the threshold factor α, and then compare the result with the threshold. If it is greater than the threshold, it means there is a target, otherwise it is considered that there is no target. It is judged as 1, which means that the point is detected as the frame starting point. The calculation formula of the threshold factor is as follows: Where N is the number of reference units selected, which is also the length of the OFDM baseband symbol. represents the expected false alarm probability.

3. The method for optimizing baseband processing of OFDM signals in a NAVDAT system according to claim 1, characterized in that: Carrier synchronization of the signal frame to be processed as shown in step 4 specifically includes the following steps: Step 4 (a), first realize the integer multiple carrier frequency error estimation, first cyclically shift the frequency domain subcarriers of the received OFDM symbol as a whole, take out the pilot data according to the discrete pilot position, conjugate multiply the adjacent pilot data, and conjugate multiply the locally generated pilot data with the adjacent pilot data, and then perform correlation operation on the two results. The algorithm expression is as follows: where Y m,k+l and Y m,k+Δk+l They represent the pilot data on the kth and k+Δkth subcarriers of the mth OFDM symbol at the receiving end after all subcarriers are cyclically shifted by l subcarriers and then taken out according to the pilot position, respectively. m,k and X m,k+Δk They respectively represent the pilot data on the kth and k+Δkth subcarriers of the mth OFDM symbol generated locally, L is the window length of the cyclic shift, and P is the total number of pilots. Step 4 (ii) uses the scattered pilot fractional frequency offset estimation and compensation, and obtains the absolute phase offset of each received symbol through the correlation between each received symbol and the corresponding local pilot, and then obtains the frequency offset from the difference between the absolute phase offsets of two adjacent symbols, which is expressed as: in represents the absolute phase deviation of the mth symbol, N s Indicates the sum of the number of subcarriers and guard intervals.

4. The method for optimizing baseband processing of OFDM signals in a NAVDAT system according to claim 1, characterized in that: In order to further reduce the bit error rate and improve the communication quality, as shown in step 5, the pilot and least squares method are used to obtain the complex channel estimation matrix of the pilot position, and the real and imaginary parts are separated to obtain two matrices. The channel estimation of the complete OFDM signal is obtained by the DFT interpolation algorithm. The result is regarded as a low-resolution image, and a more refined channel estimation value is obtained by the designed RCAFCN network output for channel compensation. The overall structure introduces residual learning, and the initial input is added to the output after the intermediate layer. The first layer is a convolution layer, which is equivalent to extracting and representing the image. The convolution kernel size is 2x2x64 and the step size is 1. The convolution kernel size of the second layer is 5x5x32, with a step size of 1, which is equivalent to nonlinear mapping, and the convolution and attention fusion module CAFM and the new module RCAFM based on residual learning design are obtained and added to the second layer. RCAFM has two branches, one is a local branch. In order to enhance cross-channel interaction and promote information integration, 1*1 convolution is used to adjust the channel dimension, and then the channel information is further mixed and stirred through channel shuffling operation. Specifically, the input tensor along the channel dimension is divided into groups, where a depth-wise separable convolution is used within each group. The resulting output tensors from each group are then concatenated along the channel dimension to generate a new output tensor, and features are extracted through a 3*3*3 convolution. The local branch can be expressed as: Y local =W 3×3×3 (CS(W 1×1 (X in ))) (9) In the global branch, query, key and value are first generated through 1x1 convolution and 3x3 depth convolution, resulting in three shapes: Next, Q is reshaped into K is reshaped as By calculating these two interactions, we get the attention map The global branch can be expressed as: where α is a learnable scaling parameter used to control and The final CAFM module output is: Y=Y local +Y global (11) The output result is then passed through the dropconnect regularization technique and the leaky Relu activation function, and X in The final RCAFM output is obtained by adding them together. Dropconnect prevents network overfitting by introducing randomness, and the resulting sparse connection structure can be used for model compression. The leaky Relu activation function can effectively alleviate the problem of neuron death and accelerate the convergence speed. The network can achieve good estimation results using a small amount of data set.

Images