An OCDM assisted physical sensing joint channel estimation method for a through anchor block

Abstract

The application provides an OCDM pass-through anchor block auxiliary physical perception joint channel estimation method, relates to the technical field of communication navigation integration and wireless channel estimation, and comprises the following steps: performing LS-DFT truncation processing and time delay domain energy aggregation on an anchor block of a received target signal, combining normalized energy threshold detection and maximum path number constraint to obtain an anchor block support set; performing sparse LS processing on a communication block under the constraint of the anchor block support set to obtain an estimated vector of a physical baseline; performing LS-DFT truncation processing on the communication block to obtain an original time delay domain track corresponding to the communication block; generating a network input tensor based on the original time delay domain track and the physical baseline, and inputting the network input tensor into a residual compensation network of physical perception to perform bounded residual repair on the physical baseline, so that a channel estimation result is obtained. Through the above method, the performance degradation problem caused by the traditional LS-DFT unified truncation method in the maritime scene can be effectively alleviated, and the estimation accuracy, system stability and implementation complexity are considered under complex channel conditions.

Description

Technical Field

[0001] This invention relates to the field of integrated communication and navigation technology and wireless channel estimation technology, and in particular to a joint channel estimation method for orthogonal linear frequency modulation multiplexing communication and navigation anchor blocks assisted by physical sensing. Background Technology

[0002] In recent years, unmanned surface vessels (USVs) have shown significant application potential in missions such as maritime patrol, environmental monitoring, fleet coordination, and maritime search and rescue. Their high maneuverability and mission autonomy place more stringent demands on reliable communication and precise navigation. To ensure the quality of service of both data links and navigation links within limited spectrum and power budgets, integrated communication and navigation (ICN) has gradually become an important supporting technology for USV platforms.

[0003] In maritime scenarios, compared to typical terrestrial cellular environments, maritime broadband channels typically exhibit stronger frequency selectivity and time-varying characteristics. Influenced by factors such as sea state, waves, and platform motion, they often possess significant delay spread, a Rician factor that varies considerably with the environment, and Doppler dispersion that increases with speed. Meanwhile, due to spectral efficiency and mission payload constraints, pilot overhead in engineering systems is strictly limited, making it difficult to employ high-density uniform pilots to support complex estimation algorithms.

[0004] For Orthogonal Chirp Division Multiplexing (OCDM) communication waveforms, a common engineering practice is to alternately transmit anchor blocks containing high-density pilots and navigation symbols, and communication blocks containing only sparse pilots and carrying data, within the coherent time interval. However, the channel recovery accuracy and robustness of existing receivers under low pilot and complex multipath conditions are still limited, and the channel estimation performance is poor. Summary of the Invention

[0005] To address the poor channel estimation accuracy and robustness caused by the sparsity of communication block pilots in maritime broadband time-varying multipath channels, this invention proposes an OCDM communication-guided anchor block-assisted physical sensing joint channel estimation method, comprising:

[0006] Step S1: Receive the target signal, wherein the target signal has an alternating frame structure of OCDM anchor blocks and communication blocks;

[0007] Step S2: Perform a coarse time-delay domain estimation based on LS-DFT truncation on the anchor block portion of the target signal, then perform time-delay domain energy aggregation, and combine normalized energy threshold detection and maximum path number constraint to obtain the anchor block support set, where the anchor block support set represents the set of tap positions of the multipath channel based on the anchor block.

[0008] Step S3: Perform sparse LS processing based on anchor block support set constraints on the communication block part of the target signal to obtain the estimated vector of the physical baseline.

[0009] Step S4: Perform LS-DFT truncation processing on the communication block portion of the target signal to obtain the original time-delay domain trajectory corresponding to the communication block, and generate a network input tensor based on the original time-delay domain trajectory and the physical baseline;

[0010] Step S5: Input the network input tensor into the physical sensing residual compensation network to perform gated bounded residual repair on the physical baseline to obtain the channel estimation result.

[0011] Optionally, step S2 includes:

[0012] The pilot subcarriers in the anchor block portion of the target signal are estimated using the following formula:

[0013] ,

[0014] in, This represents the LS estimation result of the k-th pilot subcarrier in the m-th anchor block; This represents the received component of the k-th pilot subcarrier in the m-th anchor block. This represents the transmitted component of the k-th pilot subcarrier in the m-th anchor block. This represents the set of pilot subcarrier indices on the anchor block;

[0015] Based on the results of LS estimation according to the following formula... IFFT and truncation are performed to obtain a coarse estimation result of the time delay domain for a single anchor block.

[0016] ,

[0017] in, This represents the l-th tap obtained after the m-th anchor block undergoes IFFT and truncation processing; The number of subcarriers in the anchor block; L is the preset length; It is the symbol for imaginary numbers;

[0018] Based on the coarse estimation results in the time delay domain of the single anchor block, energy aggregation across anchor blocks is performed according to the following formula:

[0019] ,

[0020] in, This represents the cross-anchor block energy aggregation result corresponding to the l-th tap; This indicates the number of anchor blocks within a coherent time window;

[0021] Based on the cross-anchor block energy aggregation results, threshold detection and screening are performed according to the following formula to obtain the anchor block support set:

[0022] ,

[0023] in, Indicates the anchor block support set, Indicates the number of centralized taps supported by the anchor block; For the normalization threshold, For the maximum number of paths constraint, This represents the maximum value of the energy aggregation result across the anchor block among the L taps. This represents the tap position variable.

[0024] Optionally, step S3 includes:

[0025] For the nth communication block, based on the anchor block support set and the communication block, the observation model is constructed according to the following formula:

[0026] ,

[0027] in, Represents the physical baseline vector as a variable. This represents the pilot receive vector in the nth communication block. This represents the pilot symbol vector in the nth communication block; This indicates that the pilot subcarriers in the nth communication block are based on the anchor block support set and the nth communication block. A submatrix extracted from a point DFT matrix; Represents the frequency domain noise vector; Used to construct diagonal matrices;

[0028] Solve the sparse LS problem based on the observation model to obtain the closed-form solution of the physical baseline;

[0029] The closed-form demapping is then performed back to a time-delay domain vector with a uniform length L to obtain the estimated vector of the physical baseline corresponding to the communication block.

[0030] Optionally, step S4 includes:

[0031] Step S41: Perform LS-DFT truncation processing on the communication block portion of the target signal to obtain the original time delay domain trajectory corresponding to the communication block;

[0032] Step S42: For inference time n, based on the physical baseline and the original time-delay domain trajectory, obtain a continuous communication block sequence of length T according to the following formula:

[0033] ,

[0034] in, This represents the sequence of consecutive communication blocks corresponding to inference time n; Let represent the estimated vector of the physical baseline corresponding to the t-th communication block. This represents the original time delay domain trajectory corresponding to the t-th communication block;

[0035] Step S43: Normalize the continuous communication block sequence, split the complex number into real and imaginary parts, and stack them by channel to obtain the network input tensor.

[0036] Optionally, step S41 includes:

[0037] The pilot subcarriers in the communication block portion of the target signal are estimated using the following formula:

[0038] ,

[0039] in, This represents the LS estimation result of the k-th pilot subcarrier in the m-th communication block; This represents the received component of the k-th pilot subcarrier in the m-th communication block. This represents the transmitted component of the k-th pilot subcarrier in the m-th communication block. This represents the set of pilot subcarrier indices on the communication block;

[0040] Based on the results of LS estimation according to the following formula... Perform IFFT and truncation processing to obtain the original time-delay domain trajectory corresponding to the communication block;

[0041] ,

[0042] ,

[0043] in, This represents the original time delay domain trajectory corresponding to the m-th communication block. This represents the l-th tap obtained after the m-th communication block undergoes IFFT and truncation processing; The number of subcarriers in the communication block; L is the preset length.

[0044] Optionally, step S5 includes:

[0045] The network input tensor is subjected to channel projection and Bi-GRU timing coding to obtain the output code:

[0046] The output code is input into the residual head based on the fully connected layer to obtain the output residual; and the output code is input into the gating head to obtain the gating factor.

[0047] Based on the output residual and the gating factor, the physical baseline is repaired with bounded residuals according to the following formula to obtain the channel estimation result:

[0048]

[0049] in, This represents the channel estimation result in the normalized domain corresponding to the nth communication block; This represents the estimated vector of the physical baseline in the normalized domain corresponding to the nth communication block. This represents the gating factor corresponding to the nth communication block. This represents the output residual corresponding to the nth communication block.

[0050] Optionally, the training loss function of the physically-aware residual compensation network is expressed according to the following formula:

[0051] ,

[0052] in, Indicates joint loss; Indicates the time delay domain loss. Indicates frequency domain loss, Indicates sample-level weights. Represents the balance coefficient. This represents the expected function.

[0053] The embodiments described in this invention have the following advantages:

[0054] In summary, this invention provides a joint channel estimation method for OCDM communication and communication block-assisted physical sensing. The method involves: performing a coarse time-delay estimation based on LS-DFT truncation on the anchor block portion of the received target signal; then performing time-delay domain energy aggregation; and combining normalized energy threshold detection and maximum path number constraints to obtain the anchor block support set. The method further involves performing sparse LS processing based on the anchor block support set constraints on the communication block portion of the target signal to obtain the estimated vector of the physical baseline. The method then performs LS-DFT truncation on the communication block portion of the target signal to obtain the original time-delay domain trajectory corresponding to the communication block, and generates a network input tensor based on the original time-delay domain trajectory and the physical baseline. Finally, the network input tensor is input into the physical sensing residual compensation network to perform gated bounded residual repair on the physical baseline, yielding the channel estimation result.

[0055] In the above method, by fully utilizing the multipath delay information provided by the high-density pilots in the anchor block, a stable acquisition of the channel structure prior is achieved at the receiving end, thereby effectively improving the physical consistency of channel modeling under complex maritime broadband time-varying channel conditions. By extracting the multipath support set based on the anchor block pilots and introducing this structural prior into the channel estimation process of the communication block, low-dimensional estimation is achieved only within the effective delay support range when the communication block pilots are sparse. This avoids the noise amplification and model mismatch problems caused by traditional unified truncation or full-dimensional estimation methods, and obtains robust physical baseline channel estimation results without increasing the communication block pilot overhead, providing a reliable foundation for subsequent frequency domain equalization and data demodulation.

[0056] Meanwhile, this invention retains the original LS delay domain trajectory constructed based on sparse pilots in the communication block, ensuring the complete preservation of error information such as noise, aliasing, and incomplete characterization by the physical model. This information, along with the physical baseline estimation results, serves as the input features for the subsequent residual compensation network. By introducing a physically-aware residual compensation structure and adaptively constraining the residual correction strength using a gating factor, the system can effectively improve estimation accuracy in complex scenarios with significant model mismatch. Conversely, when predictions are unreliable or the environment changes rapidly, the system can automatically degenerate to the physical baseline results, thereby avoiding error amplification and ensuring the stability and controllability of the system operation.

[0057] Therefore, through the above technical solution, the present invention can effectively alleviate the performance degradation problem caused by the traditional LS-DFT unified truncation method in typical maritime scenarios such as port rich scattering and significant weak path with long time delay. It can simultaneously take into account estimation accuracy, system stability and implementation complexity under complex channel conditions, and improve the engineering reliability of the integrated communication and navigation system.

[0058] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0059] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0060] Figure 1 A flowchart illustrating an embodiment of the OCDM communication and navigation anchor block-assisted physical sensing joint channel estimation method provided by the present invention;

[0061] Figure 2 This is a schematic diagram of an embodiment of an alternating frame structure and pilot configuration for an OCDM anchor block / communication block provided by the present invention;

[0062] Figure 3This is a flowchart of an embodiment of an anchor block support set extraction and communication block sparsity estimation method based on anchor block support set provided by the present invention;

[0063] Figure 4 A schematic diagram of the structure of a physical sensing residual compensation network embodiment provided by the present invention;

[0064] Figure 5 A comparison diagram of the 16-QAM constellation distributions corresponding to the three channel estimation algorithms provided in this invention;

[0065] Figure 6 Comparison chart of cumulative distribution functions provided for this invention;

[0066] Figure 7 This is a performance comparison chart of average bit error rate based on the same set of test data provided by the present invention. Detailed Implementation

[0067] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0068] To address the issues of poor channel estimation accuracy and robustness caused by the sparsity of pilot signals in communication blocks under maritime broadband time-varying multipath channels, referencing Figure 1 The present invention provides a flowchart of an embodiment of an OCDM communication and navigation anchor block-assisted physical sensing joint channel estimation method, which specifically includes:

[0069] Step S1: Receive the target signal, the target signal having an alternating frame structure of OCDM anchor blocks and communication blocks.

[0070] Regarding the acquisition of the target signal, the transmitter uses Discrete Fresnel Transform (DFnT) to perform OCDM modulation and inserts a Cyclic Prefix (CP). Let the... The frequency domain transmission vector of each OCDM symbol is , Let represent the number of subcarriers in the signal block. Then, the time-domain transmitted signal after CP insertion is:

[0071]

[0072] in Let DFnT be the matrix. The matrix is ​​IDFnT. Insert the CP operator.

[0073] When the CP satisfies the condition for eliminating inter-symbol interference, after CP removal and OCDM demodulation, an equivalent frequency domain input-output relationship consistent with OFDM (Orthogonal Frequency Division Multiplexing) is obtained:

[0074]

[0075] in , The first The transmit and receive components of each subcarrier. For the equivalent frequency domain channel, For noise; This indicates the number of subcarriers in the signal block.

[0076] Define the length as Delay domain tap vector:

[0077]

[0078] and frequency domain channel vector .when At that time, there were:

[0079]

[0080] in, For normalization The point discrete Fourier transform (DFT) matrix is ​​used to diagonalize the equivalent circular convolution in the frequency domain under the cyclic prefix condition, mapping the time-delay domain channel to the frequency domain channel response; it is also used to construct the pilot observation submatrix in the communication block sparse estimation based on the anchor block support set.

[0081] Regarding the structure of the target signal, refer to... Figure 2 The present invention provides a schematic diagram of an embodiment of an alternating frame structure and pilot configuration of OCDM anchor block / communication block. The target signal adopts an alternating frame structure of "anchor block + communication block": the anchor block is configured with high-density pilots for multipath structure detection, and the communication block is configured with sparse pilots to improve spectral efficiency.

[0082] Anchor block pilot subcarrier index set Defined as:

[0083]

[0084] Anchor block data subcarrier index set for:

[0085]

[0086] Pilot subcarrier index set of communication block Defined as:

[0087]

[0088] Data subcarrier index set of communication blocks for:

[0089]

[0090] in, Integer index value representing a subcarrier.

[0091] In one implementation, the pilot signals can be arranged in an equally spaced comb-like structure:

[0092]

[0093] in, and These represent the interval step size of the pilot subcarriers in the anchor block and the communication block respectively in the frequency domain index, and , As the starting index, To satisfy index falling in Integers.

[0094] Unified definition of the first The set of pilot subcarrier indices for each signal block is The data subcarrier index set is Its frequency domain transmission symbols satisfy:

[0095]

[0096] in For pilot symbols; anchor blocks Carrying navigation data and navigation symbols, communication blocks It carries business data.

[0097] Regarding the channel model under maritime time-varying multipath channels, this invention adopts a time-varying multipath Rician channel model and a continuous-time baseband equivalent impulse response. for:

[0098]

[0099] in and The first The time delay and complex gain of each physical path, The number of physical paths. Represents the impulse function; the average power of each path satisfies:

[0100]

[0101] This application supplements the above-mentioned target signal and channel model to help better understand and explain the technical solution described in this application. This part of the content belongs to the prior art known to those skilled in the art.

[0102] After discretization, let the sampling period be... If the block fading approximation is used within a single symbol, then the th The equivalent discrete tap corresponding to each symbol can be represented as:

[0103]

[0104] in An equivalent projection kernel is introduced for shaping and matched filtering and sampling. Indicates the physical symbol period of OCDM. and The first The time delay and complex gain of the physical path, where L is the preset length.

[0105] Step S2: Perform a coarse time-delay domain estimation based on LS-DFT truncation on the anchor block portion of the target signal, then perform time-delay domain energy aggregation, and combine normalized energy threshold detection and maximum path number constraint to obtain the anchor block support set, where the anchor block support set represents the set of tap positions of the multipath channel based on the anchor block.

[0106] Among them, LS-DFT (Least Squares-Discrete Fourier Transform) estimation refers to least squares-discrete Fourier estimation.

[0107] Extracting the anchor block support set allows full utilization of the multipath delay information provided by the high-density pilots in the anchor block, enabling stable acquisition of the channel structure prior at the receiver. Introducing this structural prior into the channel estimation process of the communication block allows for robust physical baseline channel estimation results without increasing the communication block pilot overhead, even with sparse communication block pilots.

[0108] Optionally, refer to Figure 3 A flowchart of an embodiment of an anchor block support set extraction method provided by the present invention is given, wherein, This represents the received component of the k-th pilot subcarrier in the m-th anchor block. To distinguish it from the communication block, it can also be represented as... ; This represents the l-th tap obtained after the m-th anchor block undergoes IFFT and truncation processing, and can also be represented as... Step S2 includes:

[0109] The pilot subcarriers in the anchor block portion of the target signal are estimated using the following formula:

[0110]

[0111] in, This represents the LS estimation result of the k-th pilot subcarrier in the m-th anchor block; This represents the received component of the k-th pilot subcarrier in the m-th anchor block. This represents the transmitted component of the k-th pilot subcarrier in the m-th anchor block. This represents the set of pilot subcarrier indices on the anchor block;

[0112] Based on the results of LS estimation according to the following formula... IFFT and truncation are performed to obtain a coarse estimation result of the time delay domain for a single anchor block.

[0113]

[0114] in, This represents the l-th tap obtained after the m-th anchor block undergoes IFFT and truncation processing; The number of subcarriers in the anchor block; L is the preset length; It is the symbol for imaginary numbers;

[0115] Based on the coarse estimation results in the time delay domain of the single anchor block, energy aggregation across anchor blocks is performed according to the following formula:

[0116]

[0117] in, This represents the cross-anchor block energy aggregation result corresponding to the l-th tap; This indicates the number of anchor blocks within a coherent time window;

[0118] Based on the cross-anchor block energy aggregation results, threshold detection and screening are performed according to the following formula to obtain the anchor block support set:

[0119]

[0120] in, Indicates the anchor block support set, Indicates the number of centralized taps supported by the anchor block; For the normalization threshold, For the maximum number of paths constraint, This represents the maximum value of the energy aggregation result across the anchor block among the L taps. This represents the tap position variable.

[0121] Within the same channel coherence time window The time-delay domain energy of each consecutive anchor block is aggregated and averaged to obtain... And introduce a normalization threshold Estimated anchor block support set Simultaneously set a maximum multipath constraint. When the number of taps that meet the threshold exceeds At that time, only the one with the highest energy is retained. The number of tap positions helps to suppress noise disturbances in the coarse estimation of single anchor blocks and enhances the robustness of support set extraction.

[0122] Step S3: Perform sparse LS processing based on anchor block support set constraints on the communication block portion of the target signal to obtain the estimated vector of the physical baseline.

[0123] By performing sparse LS processing based on anchor block support set constraints on the communication block portion of the target signal, prior information about the channel structure is introduced into the channel estimation process of the communication block, realizing the transformation from "full-dimensional unconstrained LS" to "structure-constrained low-dimensional LS". In the case of sparse pilots in the communication block, low-dimensional estimation is performed only within the effective time delay support range, avoiding the noise amplification and model mismatch problems caused by traditional unified truncation or full-dimensional estimation methods. Robust physical baseline channel estimation results are obtained without increasing the pilot overhead of the communication block, providing a reliable foundation for subsequent frequency domain equalization and data demodulation.

[0124] Optionally, refer to Figure 3 The present invention provides a flowchart of an embodiment of a communication block sparse estimation method based on anchor block support sets, wherein step S3 includes:

[0125] For the nth communication block, based on the anchor block support set and the communication block, the observation model is constructed according to the following formula:

[0126]

[0127] in, Represents the physical baseline vector as a variable. This represents the pilot receive vector in the nth communication block. This represents the pilot symbol vector in the nth communication block; This indicates that the pilot subcarriers in the nth communication block are based on the anchor block support set and the nth communication block. A submatrix extracted from a point DFT matrix; Represents the frequency domain noise vector; Used to construct diagonal matrices;

[0128] Solve the sparse LS problem based on the observation model to obtain the closed-form solution of the physical baseline;

[0129] The closed-form demapping is then performed back to a time-delay domain vector with a uniform length L to obtain the estimated vector of the physical baseline corresponding to the communication block.

[0130] in, , From The submatrix is ​​obtained by extracting the rows corresponding to the pilot subcarriers and the columns corresponding to the anchor block support set from the point DFT matrix. In this way, The complete observation matrix is ​​cropped into The low-dimensional submatrix enables structure-constrained low-dimensional LS, which helps improve the robustness of channel estimation in the case of sparse pilots in the communication block.

[0131] According to the observation model represented by formula (18), the corresponding sparse LS problem is:

[0132]

[0133] Closed-form solution to the sparse LS problem Represented as:

[0134]

[0135] Will Map back to a uniform length time delay domain vector That is, padding with zeros at positions not supported by the support set:

[0136]

[0137] in, Indicates anchor block support set The position of the m-th tap. Indicates the anchor block support set The low-dimensional vector obtained by solving the above The m-th component.

[0138] This yields the time-delay domain baseline estimation vector for Nav-Sparse. That is, the estimated vector of the physical baseline corresponding to the nth communication block:

[0139]

[0140] Optionally, after obtaining the channel estimation results, the data subcarrier index set of the nth signal block can be used. Perform single-shot equalization:

[0141]

[0142] in, This represents the received vector of the k-th subcarrier in the n-th signal block. To be Zero padding to length Then proceed The result obtained from point DFT is:

[0143]

[0144] for The k-th component, For normalized Point Discrete Fourier Transform (DFT) matrix.

[0145] Single-tap equalization refers to subcarrier-by-subcarrier single-tap frequency domain equalization, as shown in formula (2). Because after the CP meets the condition and is demodulated by OCDM, the system can be written as an equivalent frequency domain input-output relationship consistent with OFDM, so each subcarrier only needs one complex coefficient. Equalization is performed. Based on channel estimation and single-cycle equalization to recover data symbols, it can compensate for frequency-selective amplitude and phase distortion caused by multipath; moreover, it has lower complexity compared to time-domain or full-matrix equalization.

[0146] Step S4: Perform LS-DFT truncation processing on the communication block portion of the target signal to obtain the original time-delay domain trajectory corresponding to the communication block, and generate a network input tensor based on the original time-delay domain trajectory and the physical baseline.

[0147] The communication block portion of the target signal is truncated using LS-DFT, which preserves the original LS time-delay domain trajectory constructed based on sparse pilots in the communication block. The trajectory retains the original observation characteristics such as aliasing, truncation, and noise, and can be used as a low-resolution approximate observation of the real channel. This provides non-sparse residual information clues for subsequent residual compensation, which helps to improve the accuracy of channel estimation.

[0148] Optionally, step S4 includes:

[0149] Step S41: Perform LS-DFT truncation processing on the communication block portion of the target signal to obtain the original time delay domain trajectory corresponding to the communication block;

[0150] Step S42: For inference time n, based on the physical baseline and the original time-delay domain trajectory, obtain a continuous communication block sequence of length T according to the following formula:

[0151]

[0152] in, This represents the sequence of consecutive communication blocks corresponding to inference time n; Let represent the estimated vector of the physical baseline corresponding to the t-th communication block. This represents the original time delay domain trajectory corresponding to the t-th communication block;

[0153] Step S43: Normalize the continuous communication block sequence, split the complex number into real and imaginary parts, and stack them by channel to obtain the network input tensor.

[0154] Specifically, step S43 may include:

[0155] based on Define the normalization coefficient at inference time n. :

[0156]

[0157] in, Let L represent the l-th tap in the estimation vector of the physical baseline corresponding to the t-th communication block, where L represents the preset length. It is a numerically stable term; the position of signal blocks such as communication blocks and anchor blocks can be used as a metric during inference.

[0158] Normalization process:

[0159]

[0160] The complex number is split into real and imaginary parts and stacked according to channels to form the network input tensor:

[0161]

[0162] The above normalization process enhances cross-sample scale consistency and helps accelerate network convergence.

[0163] Optionally, step S41 includes:

[0164] The pilot subcarriers in the communication block portion of the target signal are estimated using the following formula:

[0165]

[0166] in, This represents the LS estimation result of the k-th pilot subcarrier in the m-th communication block; This represents the received component of the k-th pilot subcarrier in the m-th communication block. This represents the transmitted component of the k-th pilot subcarrier in the m-th communication block. This represents the set of pilot subcarrier indices on the communication block;

[0167] Based on the results of LS estimation according to the following formula... Perform IFFT and truncation processing to obtain the original time-delay domain trajectory corresponding to the communication block;

[0168]

[0169]

[0170] in, This represents the original time delay domain trajectory corresponding to the m-th communication block. This represents the l-th tap obtained after the m-th communication block undergoes IFFT and truncation processing; The number of subcarriers in the communication block is the same as the number of subcarriers in the anchor block; L is the preset length.

[0171] Step S5: Input the network input tensor into the physical sensing residual compensation network to perform gated bounded residual repair on the physical baseline to obtain the channel estimation result.

[0172] The original LS delay domain trajectory, constructed based on sparse pilots, is retained in the communication block. This ensures the complete preservation of error information, including noise, aliasing, and aspects not fully characterized by the physical model. This information, along with the physical baseline estimation result, serves as the input feature for the subsequent residual compensation network. By introducing a physically-aware residual compensation structure and adaptively constraining the residual correction strength using a gating factor, the corrected physical baseline, i.e., the channel estimation result, is obtained. This achieves a joint channel estimation mechanism led by the physical model and enhanced by artificial intelligence, enabling the system to effectively improve estimation accuracy in complex scenarios with significant model mismatch.

[0173] In addition, after obtaining the channel estimation results corresponding to the communication block, the channel can be comprehensively estimated based on the channel estimation results corresponding to multiple communication blocks, or the channel estimation of the communication block at the next inference time can be continued based on the channel estimation results corresponding to the communication block at the previous inference time. This invention does not limit this.

[0174] Optionally, step S5 includes:

[0175] The network input tensor is subjected to channel projection and Bi-GRU timing coding to obtain the output code:

[0176] The output code is input into the residual head based on the fully connected layer to obtain the output residual; and the output code is input into the gating head to obtain the gating factor.

[0177] Based on the output residual and the gating factor, the physical baseline is repaired with bounded residuals according to the following formula to obtain the channel estimation result:

[0178]

[0179] in, This represents the channel estimation result in the normalized domain corresponding to the nth communication block; This represents the estimated vector of the physical baseline in the normalized domain corresponding to the nth communication block. This represents the gating factor corresponding to the nth communication block. This represents the output residual corresponding to the nth communication block.

[0180] Reference Figure 4 The present invention provides a schematic diagram of a physical sensing residual compensation network embodiment, wherein ACACE-Net refers to a physical sensing residual compensation network in the Anchor-Communication Alternating Channel Estimation (ACACE) framework.

[0181] ACACE-Net employs a "channel projection + Bi-GRU timing coding + dual-head output (residual head / gating head)" structure: channel projection uses two layers. Convolution and ReLU functions on network input tensors Perform channel blending and project the number of channels from 4 to , The hidden layer has the number of channels. The features at each time step are then flattened and input into a two-layer Bi-GRU to characterize the coupling features between cross-block temporal evolution and delay distribution. The residual head uses a fully connected layer structure and outputs... Real-valued vectors are reduced to complex residuals. The gate controller uses a "fully connected layer + Sigmoid" architecture, outputting a scalar through the Sigmoid function. It is used to adaptively adjust the estimated vector of the physical baseline and correct the intensity.

[0182] Since the result is a fusion result in the normalization domain, inverse normalization is also required to obtain the final time delay domain estimate, i.e., the channel estimate result.

[0183]

[0184] in, This represents the normalization coefficient corresponding to the nth communication block.

[0185] The overall mapping of the ACACE-Net network is denoted as:

[0186]

[0187] in, For network parameters. When When the prediction is unreliable or the environment changes rapidly, the output degenerates to a pure physical baseline, automatically reverting to the physical baseline result to avoid error amplification and ensure the stability and controllability of the system operation; when the residual prediction is reliable, It enables auxiliary enhancement of the baseline, thereby possessing the engineering characteristics of "degradable, controllable strength, and bounded repair".

[0188] Optionally, the training loss function of the physically-aware residual compensation network is expressed according to the following formula:

[0189] ,

[0190] in, Indicates joint loss; Indicates the time delay domain loss. Indicates frequency domain loss, Indicates sample-level weights. Represents the balance coefficient. This represents the expected function.

[0191] The specific steps are as follows, and the meaning of each symbolic variable can be referred to in the aforementioned formula:

[0192] Real time delay domain tap vector Normalization :

[0193]

[0194] in, This represents the normalization coefficient corresponding to the nth inference time. The inference time can be measured by the position of signal blocks such as communication blocks and anchor blocks.

[0195] Calculate the tap mean squared error loss (TSE) in the time delay domain:

[0196]

[0197] in, This represents the estimated value of the l-th tap in the normalized channel estimation vector corresponding to the n-th communication block. This represents the value of the l-th tap in the normalized real delay domain tap vector corresponding to the n-th communication block.

[0198] The channel estimation vector corresponding to the nth communication block is represented by... Generate full-band frequency domain channel :

[0199]

[0200] Balanced processing:

[0201]

[0202] in, express The k-th frequency domain component This represents the received component of the k-th subcarrier in the n-th signal block of the training set signal; equalization processing can be used to generate equalized symbols and calculate the relative EVM loss during the training process, which helps to reduce the computational overhead of receiving and constructing the training target.

[0203] Frequency domain relative error vector magnitude (EVM) loss:

[0204]

[0205] This represents the transmitted component of the k-th subcarrier in the n-th signal block of the training set signal. This represents a numerically stable term.

[0206] Joint losses:

[0207]

[0208] in, For sample-level weights, These are the balancing coefficients. Sample-level weights. It can be used for high SNR samples in port rich scattering scenarios to guide the network to focus on learning hard samples of "model mismatch caused by unified truncation and continuous delay leakage", thereby enhancing the compensation ability for long-delay weak path long tail.

[0209] Optionally, the gating head can set the final layer bias to a positive value during the training initialization phase, so that the initial gating factor... To facilitate residual branch gradient propagation, the loss is then adaptively adjusted via joint loss. This enables automatic adjustment and constraint of the correction intensity.

[0210] In summary, this invention provides a joint channel estimation method for OCDM communication and communication block-assisted physical sensing. The method involves: performing a coarse time-delay estimation based on LS-DFT truncation on the anchor block portion of the received target signal; then performing time-delay domain energy aggregation; and combining normalized energy threshold detection and maximum path number constraints to obtain the anchor block support set. The method further involves performing sparse LS processing based on the anchor block support set constraints on the communication block portion of the target signal to obtain the estimated vector of the physical baseline. The method then performs LS-DFT truncation on the communication block portion of the target signal to obtain the original time-delay domain trajectory corresponding to the communication block, and generates a network input tensor based on the original time-delay domain trajectory and the physical baseline. Finally, the network input tensor is input into the physical sensing residual compensation network to perform gated bounded residual repair on the physical baseline, yielding the channel estimation result.

[0211] In the above method, by fully utilizing the multipath delay information provided by the high-density pilots in the anchor block, a stable acquisition of the channel structure prior is achieved at the receiving end, thereby effectively improving the physical consistency of channel modeling under complex maritime broadband time-varying channel conditions. By extracting the multipath support set based on the anchor block pilots and introducing this structural prior into the channel estimation process of the communication block, low-dimensional estimation is achieved only within the effective delay support range when the communication block pilots are sparse. This avoids the noise amplification and model mismatch problems caused by traditional unified truncation or full-dimensional estimation methods, and obtains robust physical baseline channel estimation results without increasing the communication block pilot overhead, providing a reliable foundation for subsequent frequency domain equalization and data demodulation.

[0212] Meanwhile, this invention retains the original LS delay domain trajectory constructed based on sparse pilots in the communication block, ensuring the complete preservation of error information such as noise, aliasing, and incomplete characterization by the physical model. This information, along with the physical baseline estimation results, serves as the input features for the subsequent residual compensation network. By introducing a physically-aware residual compensation structure and adaptively constraining the residual correction strength using a gating factor, the system can effectively improve estimation accuracy in complex scenarios with significant model mismatch. Conversely, when predictions are unreliable or the environment changes rapidly, the system can automatically degenerate to the physical baseline results, thereby avoiding error amplification and ensuring the stability and controllability of the system operation.

[0213] Therefore, through the above technical solution, the present invention can effectively alleviate the performance degradation problem caused by the traditional LS-DFT unified truncation method in typical maritime scenarios such as port rich scattering and significant weak path with long time delay. It can simultaneously take into account estimation accuracy, system stability and implementation complexity under complex channel conditions, and improve the engineering reliability of the integrated communication and navigation system.

[0214] To better explain the effectiveness of the OCDM communication-anchor block-assisted physical sensing joint channel estimation method provided by this invention from the perspectives of modulation quality and statistical indicators, this invention, based on I (In-phase) Q (Quadrature) data collected in a lake surface experimental environment, provides the following... Figure 5 The comparison of the 16-QAM constellation diagram shown. Figure 6 The comparison chart showing the root mean square (RMS) - error vector magnitude (EVM) - cumulative distribution function (CDF) is as follows: Figure 7 The comparison results of the average bit error rate (BER) are shown. The test environment was an open artificial lake, where an unmanned surface vessel (USV) equipped with a communication and navigation module sailed along a preset route, during which typical multipath factors such as shoreline reflection and partial obstruction existed.

[0215] The waveforms and system parameters related to the test are shown in Table 1.

[0216] Table 1: Waveform and System Parameter Table

[0217]

[0218] The test experiment details are as follows:

[0219] Within each coherence time interval, a navigation block containing high-density comb pilots and QPSK navigation symbols is transmitted first, followed by several communication blocks. These communication blocks employ high-order QAM modulation and retain only 16 sparse pilots. The software-defined radio platform (SDR platform) on the USV operates according to the sampling rate. The received baseband signal is continuously sampled and transmitted to the host for storage in real time. During the test, this invention collected dozens of raw IQ data segments, each several seconds long, at multiple times and under different USV tracks, providing samples for subsequent offline channel estimation and performance evaluation.

[0220] During the offline processing phase, all acquired OCDM frames are first synchronized and coarsely offset compensated. Then, high-density LS-DFT estimation and delay truncation are performed on the navigation block to obtain a relatively high-precision "reference channel." Based on this, this invention applies three channel estimation algorithms to the communication block data of the same frame:

[0221] LS–DFT: LS–DFT estimation is performed using only sparse pilots on the communication block;

[0222] Nav-Sparse: Only steps S1-S3 are performed, i.e., using anchor block support sets. Sparse LS processing based on anchor block support set constraints is performed on the communication block pilot;

[0223] ACACE: Based on the output of Nav-Sparse, further input ACACE-Net is used for residual repair.

[0224] Reference Figure 5 The present invention provides a comparison of the 16-QAM constellation distributions corresponding to the three channel estimation algorithms provided in this invention. Among them, (a) is the constellation distribution result after LS-DFT interpolation using only sparse pilots on the communication block, (b) is the constellation distribution result after the anchor block-assisted sparse estimation algorithm Nav-Sparse introduces anchor block support set information on this basis, and (c) is the constellation distribution result after ACACE-Net performs physical sensing residual compensation on the Nav-Sparse output. As can be observed from the figure: In the LS-DFT case, the constellation point cloud appears as an inflated "square cloud" with blurred boundaries between symbol clusters. Some points are close to or even cross the decision boundary, reflecting that under sparse pilot conditions, frequency domain interpolation alone is insufficient to accurately compensate for long-delay weak paths and time-varying mismatches. After using anchor block-assisted sparse estimation, the 16 constellation clusters have clearly taken shape, and the cloud has significantly shrunk and gathered near the ideal constellation points, effectively reducing the channel estimation error. However, slight elliptical stretching and eccentricity are still visible, indicating that the unified truncation and threshold strategy still has residual errors for some weak paths. In the ACACE-Net case, the constellation clusters are further tightened, and the cloud shape is closer to a circle. The stretching in the diagonal direction is alleviated, and the number of outliers far from the decision boundary is significantly reduced, demonstrating the role of residual compensation in refining the physical estimation.

[0225] Reference Figure 6This paper presents a comparison chart of the cumulative distribution function provided by this invention. This invention calculates the RMSEVM for each frame and plots CDF curves on 2500 frames of test data. Statistical results show that the RMS EVM distribution of LS-DFT is generally large, with an average of approximately 57%, a median of approximately 53%, and 80% and 90% quantiles of approximately 73% and 89%, respectively. Anchor block-assisted sparse estimation (Nav-Sparse) significantly improves modulation quality, reducing the average RMS EVM to approximately 29%, the median to approximately 25%, and the 80% and 90% quantiles to approximately 37% and 47%, respectively, which is about half the average EVM of LS-DFT. Based on this, ACACE-Net further reduces the average RMS EVM to approximately 28%, the median to approximately 24%, and the high quantiles also decrease slightly. The three curves show an ordered relationship across the entire range: "LS-DFT is relatively weak, Nav-Sparse shows significant improvement, and ACACE shows a slight improvement above Nav-Sparse." It can be noted that ACACE's CDF curve is slightly higher than Nav-Sparse in most probability intervals, while the two remain close in the long tail region. This indicates that ACACE improves the average EVM performance without introducing significant worst-case degradation, thus ensuring the stability and controllability of channel estimation.

[0226] Reference Figure 7 The paper presents a performance comparison chart of the average bit error rate (BER) based on the same set of test data provided by this invention. Statistical results show that the average BER of LS-DFT is approximately... Anchor block-assisted sparse estimation Nav-Sparse reduces it to approximately The relative bit error rate was reduced by approximately 75%; based on this, the average BER of ACACE-Net was further reduced to approximately Compared to Nav-Sparse, it achieves an additional 8% reduction in BER. Combined with... Figure 5 The convergence effect of constellations and Figure 6 The EVM distribution shows that Nav-Sparse is the main source of performance improvement in the actual test environment, while ACACE provides stable small gain by bounded correction of the residuals on this basis, without compromising the reliability of the physical baseline in terms of both modulation quality and bit error rate performance.

[0227] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0228] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0229] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention without departing from the principles and spirit of the present invention.

Claims

1. A joint channel estimation method for OCDM communication and navigation anchor blocks assisted by physical sensing, characterized in that, The method includes: Step S1: Receive the target signal, wherein the target signal has an alternating frame structure of OCDM anchor blocks and communication blocks; Step S2: Perform a coarse time-delay domain estimation based on LS-DFT truncation on the anchor block portion of the target signal, then perform time-delay domain energy aggregation, and combine normalized energy threshold detection and maximum path number constraint to obtain the anchor block support set, where the anchor block support set represents the set of tap positions of the multipath channel based on the anchor block. Specifically, the pilot subcarriers in the anchor block portion of the target signal are estimated using the following formula: ， In the formula, This represents the LS estimation result of the k-th pilot subcarrier in the m-th anchor block; This represents the received component of the k-th pilot subcarrier in the m-th anchor block. This represents the transmitted component of the k-th pilot subcarrier in the m-th anchor block. This represents the set of pilot subcarrier indices on the anchor block; Based on the results of LS estimation according to the following formula... IFFT and truncation are performed to obtain a coarse estimation result of the time delay domain for a single anchor block. ， in, This represents the l-th tap obtained after the m-th anchor block undergoes IFFT and truncation processing; The number of subcarriers in the anchor block; L is the preset length; j is the imaginary sign; Based on the coarse estimation results in the time delay domain of the single anchor block, energy aggregation across anchor blocks is performed according to the following formula: ， in, This represents the cross-anchor block energy aggregation result corresponding to the l-th tap; This indicates the number of anchor blocks within a coherent time window; Based on the cross-anchor block energy aggregation results, threshold detection and screening are performed according to the following formula to obtain the anchor block support set: ， in, Indicates the anchor block support set, Indicates the number of centralized taps supported by the anchor block; For the normalization threshold, For the maximum number of paths constraint, This represents the maximum value of the energy aggregation result across the anchor block among the L taps. Indicates the tap position variable; Step S3 involves performing sparse LS processing based on anchor block support set constraints on the communication block portion of the target signal to obtain the estimated vector of the physical baseline; specifically including: For the nth communication block, based on the anchor block support set and the communication block, the observation model is constructed according to the following formula: ， in, Represents the physical baseline vector as a variable. This represents the pilot receive vector in the nth communication block. This represents the pilot symbol vector in the nth communication block; This indicates that the pilot subcarriers in the nth communication block are based on the anchor block support set and the nth communication block. A submatrix extracted from a point DFT matrix; Represents the frequency domain noise vector; Used to construct diagonal matrices; Solve the sparse LS problem based on the observation model to obtain the closed-form solution of the physical baseline; The closed-form demapping is then performed back to a time-delay domain vector with a uniform length L to obtain the estimated vector of the physical baseline corresponding to the communication block. Step S4 involves performing LS-DFT truncation on the communication block portion of the target signal to obtain the original time-delay domain trajectory corresponding to the communication block, and generating a network input tensor based on the original time-delay domain trajectory and the physical baseline; specifically including: Step S41: Perform LS-DFT truncation processing on the communication block portion of the target signal to obtain the original time delay domain trajectory corresponding to the communication block; Step S42: For inference time n, based on the physical baseline and the original time-delay domain trajectory, obtain a continuous communication block sequence of length T according to the following formula: ， in, This represents the sequence of consecutive communication blocks corresponding to inference time n; Let represent the estimated vector of the physical baseline corresponding to the t-th communication block. This represents the original time delay domain trajectory corresponding to the t-th communication block; Step S43: Normalize the continuous communication block sequence, split the complex number into real and imaginary parts, and stack them by channel to obtain the network input tensor; Step S5: Input the network input tensor into the physical sensing residual compensation network to perform gated bounded residual repair on the physical baseline to obtain the channel estimation result.

2. The method according to claim 1, characterized in that, Step S41 includes: The pilot subcarriers in the communication block portion of the target signal are estimated using the following formula: ， in, This represents the LS estimation result of the k-th pilot subcarrier in the m-th communication block; This represents the received component of the k-th pilot subcarrier in the m-th communication block. This represents the transmitted component of the k-th pilot subcarrier in the m-th communication block. This represents the set of pilot subcarrier indices on the communication block; Based on the results of LS estimation according to the following formula... Perform IFFT and truncation processing to obtain the original time-delay domain trajectory corresponding to the communication block; ， ， in, This represents the original time delay domain trajectory corresponding to the m-th communication block. This represents the l-th tap obtained after the m-th communication block undergoes IFFT and truncation processing; The number of subcarriers in the communication block; L is the preset length.

3. The method according to claim 1, characterized in that, Step S5 includes: The network input tensor is subjected to channel projection and Bi-GRU temporal coding to obtain the output code: The output code is input into the residual head based on the fully connected layer to obtain the output residual; and the output code is input into the gating head to obtain the gating factor. Based on the output residual and the gating factor, the physical baseline is repaired with bounded residuals according to the following formula to obtain the channel estimation result: in, This represents the channel estimation result in the normalized domain corresponding to the nth communication block; This represents the estimated vector of the physical baseline in the normalized domain corresponding to the nth communication block. This represents the gating factor corresponding to the nth communication block. This represents the output residual corresponding to the nth communication block.

4. The method according to claim 1, characterized in that, The training loss function of the physical perception residual compensation network is expressed by the following formula: ， in, Joint losses; Indicates the time delay domain loss. Indicates frequency domain loss, Indicates sample-level weights. Indicates the balance coefficient; This represents the expected function.

Images