GNSS pseudo code estimation method and device based on TurboUNet

Abstract

The application relates to a GNSS code estimation method and device based on a Turbo UNet, which comprises splicing a received coherent adaptive subcarrier modulation complex baseband GNSS signal and a locally generated C 1 code sequence to generate input features; splicing the input features of the current iteration round and soft probabilities output by a UNet module of the previous iteration round, inputting the splicing result into the UNet module to obtain soft probability output of the current iteration round, until a preset number of iterations, outputting the soft probability as soft probability estimation of GNSS code, and obtaining C 2 code sequences and C 3 code sequences in GNSS code estimation values according to the soft probability estimation. The application improves the anti-noise performance and decoding accuracy under the condition of not requiring high-power hardware by simulating a Turbo code iterative optimization process, continuously optimizing the splicing result of the input UNet module through iteration, and improving the accuracy of subsequent prediction through feedback of the previous stage.

Description

Technical Field

[0001] This application relates to the field of satellite navigation pseudocode estimation technology, and in particular to a GNSS pseudocode estimation method and apparatus based on TurboUNet. Background Technology

[0002] Blind estimation of pseudo-random noise (PRN) codes plays a crucial role in ensuring the integrity and reliability of Global Navigation Satellite System (GNSS) signals. Currently, many civilian GNSS signals employ complex encryption and long codes to ensure signal integrity and resist unauthorized access. These signal characteristics make them indispensable in high-value scenarios such as secure civilian applications. However, with the widespread adoption of such GNSS signals, their development has exhibited a degree of disorder. Therefore, monitoring these signal systems is particularly necessary, and code estimation is a key component in this process.

[0003] With the modernization of GNSS signals, various constant envelope multiplexing schemes, such as CASM, Interplex, and (weighted) majority voting, have been researched and applied to synthesize multiple signals without significantly sacrificing power amplifier efficiency, in order to accommodate more components and new signal systems in parallel. Complex modulation schemes significantly increase the difficulty of code estimation. Conventional methods typically rely on known pseudocodes and estimate the code sequence to be estimated through the polarity of I / Q branch data. However, the estimation accuracy of these methods is limited by the signal-to-noise ratio (SNR), thus often requiring high-gain antenna arrays to achieve an acceptable bit error rate (BER), leading to excessive hardware complexity and cost. Furthermore, the requirement for antenna aperture to increase exponentially with SNR gain also poses a practical constraint on system scalability. Summary of the Invention

[0004] Therefore, it is necessary to provide a TurboUNet-based GNSS pseudocode estimation method and apparatus that can improve noise immunity and decoding accuracy without the need for high-power hardware, in order to address the above-mentioned technical problems.

[0005] A GNSS pseudocode estimation method based on TurboUNet includes the following steps: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with the locally generated... C One-code sequence concatenation generates input features; wherein, the complex baseband GNSS signal is obtained through IQ modulation, and the IQ modulation employs... C 1 code C 2 codes and C 3 codes; The input features of the current iteration are concatenated with the soft probabilities output by the UNet module from the previous iteration, and the concatenated result is input into the UNet module to obtain the soft probability output of the current iteration. This process is repeated until a preset number of iterations are reached. The output soft probabilities are then used as the soft probability estimate of the GNSS pseudocode, and the GNSS pseudocode estimate is obtained based on this soft probability estimate. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

[0006] In one embodiment, the received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with a locally generated... C One-code sequence concatenation generates input features, including: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is decomposed into in-phase and quadrature components, and then compared with locally generated... C One-code sequence concatenation forms a three-channel one-dimensional input feature: ; in, For input features; This refers to the in-phase component of a complex baseband GNSS signal; These are the orthogonal components of the complex baseband GNSS signal; for C 1-code sequence; This refers to the batch size processed in a single training iteration. For feature dimensions.

[0007] In one embodiment, the concatenation result is input into the UNet module to obtain the soft probability output for the current iteration round, including: The stitching result is input into the UNet module; the UNet module includes: multiple encoders connected sequentially for downsampling, wherein the first... i The input terminal of the encoder is connected to the first... i -1 encoder output terminal, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The input data for the decoder; i ∈[0,…, I ]; The input of the bottleneck layer is connected to the first I The output data of the bottleneck layer is input to the output of the encoder. IOne gated jump connection unit; The splicing result is input into the 0th encoder, processed by the UNet module, and then the soft probability of the current iteration is output by the 0th decoder.

[0008] In one embodiment, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased.

[0009] In one embodiment, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased, including: ; in, For the first i Advanced temporal features obtained through subsampling; To correct the linear unit; For batch normalization processing; It is a one-dimensional convolution; Indicates the fill size; Indicates the first i The step size of the next downsampling.

[0010] In one embodiment, the input features of the current iteration and the soft probabilities output by the UNet module of the previous iteration are concatenated, including: ; in, For the first n Input features for the next iteration; For input features; For the first n The soft probability of -1 iteration.

[0011] In one embodiment, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The decoder's input data includes: No. i The first gated jump connection unit controls the input of the first... i The encoded data from each encoder is transposed and convolved to restore temporal resolution; No. i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i Decoder input data: ; ; in, For the first i The input data for the decoder; For learnable gating; For the first i +1 decoder's decoded data; For the first i Encoded data from one encoder; It is the inner product; It is the sigmoid function; It is a one-dimensional convolution.

[0012] In one embodiment, the concatenation result is input into the 0th encoder, processed by the UNet module, and then the 0th decoder outputs the soft probability of the current iteration, including: The splicing result is input into the 0th encoder. After being processed by the UNet module, the 0th decoder outputs the decoded data of the current iteration round. The decoded data is then converted into soft probabilities through a 1×1 convolution before being output.

[0013] In one embodiment, the method further includes: Training is performed end-to-end using cross-entropy loss, and the difference between the estimated GNSS pseudocode value and the true GNSS pseudocode value is measured by the loss function: ; in, The loss function; For the first Each bit in category d The unique hot tag below; For the final network iteration Code position belongs to The predicted probability of a class; L This is the total length of the PRN code sequence; D This represents the number of classes corresponding to binary codes 0 and 1.

[0014] A GNSS pseudocode estimation device based on TurboUNet includes: The input feature generation unit is used to combine the received coherent adaptive subcarrier modulated complex baseband GNSS signal with locally generated features. C One-code sequence concatenation generates input features; wherein, the complex baseband GNSS signal is obtained through IQ modulation, and the IQ modulation employs... C 1 code C 2 codes and C 3 codes; The pseudocode estimation unit is used to concatenate the input features of the current iteration round and the soft probabilities output by the UNet module in the previous iteration round, and input the concatenation result back into the UNet module to obtain the soft probability output of the current iteration round. This process continues until a preset number of iterations, outputting the soft probabilities as the soft probability estimate of the GNSS pseudocode. Based on this soft probability estimate, the GNSS pseudocode estimation value is obtained. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

[0015] The aforementioned GNSS pseudocode estimation method and apparatus based on TurboUNet concatenates the input features of the current iteration round with the soft probabilities output by the UNet module in the previous iteration round, and inputs the concatenation result into the UNet module to obtain the soft probability output of the current iteration round. By simulating the Turbo code iterative optimization process, the concatenation result input to the UNet module is continuously optimized through iteration, and the feedback of the previous stage is used to improve the accuracy of subsequent predictions, thereby improving noise resistance and decoding accuracy without the need for high-power hardware. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a GNSS pseudocode estimation method based on TurboUNet in one embodiment. Figure 2 This is a schematic diagram of the UNet module in one embodiment; Figure 3 The image shows GNSS pseudocode estimation results for different methods in one embodiment. Figure 3 (a) is C The estimation results of the 2-code sequence are shown in the figure. Figure 3 (b) is C The estimation results of the 3-code sequence are shown in the figure; Figure 4 This is a block diagram of a GNSS pseudocode estimation device based on TurboUNet in one embodiment. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] In one embodiment, such as Figure 1 As shown, a GNSS pseudocode estimation method based on TurboUNet is provided, including the following steps: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with the locally generated... C One-code sequence concatenation generates input features; the complex baseband GNSS signal is obtained through IQ modulation, which employs... C 1 code C 2 codes and C 3 digits.

[0019] The input features of the current iteration are concatenated with the soft probabilities output by the UNet module from the previous iteration, and the concatenated result is input back into the UNet module to obtain the soft probability output for the current iteration. This process is repeated for a preset number of iterations, and the output soft probabilities are used as the soft probability estimate of the GNSS pseudocode. Based on this soft probability estimate, the GNSS pseudocode estimate is obtained. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

[0020] Understandable. C 1 is a known pseudo-code. C 2-code sequence and C The 3-code sequence is the pseudocode sequence to be estimated.

[0021] In the aforementioned GNSS pseudocode estimation method based on TurboUNet, the input features of the current iteration round and the soft probabilities output by the UNet module in the previous iteration round are concatenated, and the concatenation result is input into the UNet module to obtain the soft probability output of the current iteration round. By simulating the Turbo code iterative optimization process, the concatenation result input to the UNet module is continuously optimized through iteration. The feedback from the previous stage is used to improve the accuracy of subsequent predictions, thereby improving noise resistance and decoding accuracy without the need for high-power hardware.

[0022] In one embodiment, the received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with a locally generated... C One-code sequence concatenation generates input features, including: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is decomposed into in-phase and quadrature components, and then compared with locally generated... COne-code sequence concatenation forms a three-channel one-dimensional input feature: ; in, For input features; This refers to the in-phase component of a complex baseband GNSS signal; These are the orthogonal components of the complex baseband GNSS signal; for C 1-code sequence; This refers to the batch size processed in a single training iteration. For feature dimensions.

[0023] In one embodiment, the concatenation result is input into the UNet module to obtain the soft probability output for the current iteration round, including: The stitching result is input into the UNet module; the UNet module includes multiple encoders connected sequentially for downsampling, wherein the first... i The input terminal of the encoder is connected to the first... i -1 encoder output terminal, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The input data for the decoder; i ∈[0,…, I The input of the bottleneck layer is connected to the first... I The output data of the bottleneck layer is input to the output of the encoder. I A gated jump connection unit. The splicing result is input into the 0th encoder, processed by the UNet module, and the soft probability of the current iteration is output by the 0th decoder.

[0024] In this embodiment, by introducing gated jump connection units to selectively fuse multi-scale temporal features, the generalization ability and robustness of the model can be improved, as well as the noise resistance and decoding accuracy can be enhanced.

[0025] Reference Figure 2 In one embodiment, I =3, with a total of 4 encoders, 4 decoders, 4 gated skip connection units, and 1 bottleneck layer. The bottleneck layer is connected through two consecutive... Layers, combined with the SiLU activation function and batch normalization, enable nonlinear transformation capabilities.

[0026] In one embodiment, i When =0, the first iWhen an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased.

[0027] In one embodiment, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased, including: ; in, For the first i Advanced temporal features obtained through subsampling; To correct the linear unit; For batch normalization processing; It is a one-dimensional convolution; Indicates the fill size; Indicates the first i The step size of the next downsampling, , .

[0028] In this embodiment, i When the value is greater than 0, the implicit encoder features are downsampled by 2 times and the channel dimension is increased to extract a gradually abstract time-spectral representation of the input signal, while preserving the low-level details in the early feature map for subsequent skip connections.

[0029] In one embodiment, the input features of the current iteration and the soft probabilities output by the UNet module of the previous iteration are concatenated, including: ; in, For the first n Input features for the next iteration; For input features; For the first n The soft probability of the -1st iteration. For the 1st iteration, .

[0030] In one embodiment, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first...i The decoder's input data includes: No. i The first gated jump connection unit controls the input of the first... i The encoded data from the first encoder is used to recover temporal resolution through transposed convolution. i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i Decoder input data: ; ; in, For the first i The input data for the decoder; For learnable gating; For the first i +1 decoder's decoded data; For the first i Encoded data from one encoder; It is the inner product; The sigmoid function represents element-wise multiplication; It is a one-dimensional convolution.

[0031] In this embodiment, learnable gating is used to adaptively balance the contributions of decoder features and corresponding encoder jump features, thereby improving noise resistance.

[0032] In one embodiment, the concatenation result is input into the 0th encoder, processed by the UNet module, and then the 0th decoder outputs the soft probability of the current iteration, including: The concatenation result is input into the 0th encoder. After being processed by the UNet module, the 0th decoder outputs the decoded data of the current iteration round. The decoded data is then converted into soft probabilities through a 1×1 convolution before being output.

[0033] In one embodiment, the soft probability outputs from each iteration are stacked to form a three-dimensional tensor: ; in, Indicates channel c In position k Belongs to symbols b The probability, It is the batch size for each iteration.

[0034] Taking argmax from the symbolic dimension of the three-dimensional tensor, we obtain the hard decision matrix. : ; In the hard decision matrix, the first row is... C Two-digit judgment, second act C 3-bit decision. Since argmax is a non-differentiable operation, soft probabilities are retained during training for loss calculation and uncertainty analysis.

[0035] exist After rounds of iteration, the last UNet module is passed The output is used as a GNSS pseudocode estimate. : ; This tensor pairs each encoded channel ( C 2-digit code and C The soft probability estimate for each chip location (3-bit) is encoded. There is no additional fusion layer, consistent with the network implementation.

[0036] In one embodiment, the method further includes: Training is performed end-to-end using cross-entropy loss, and the difference between the estimated GNSS pseudocode value and the true GNSS pseudocode value is measured by the loss function: ; in, The loss function; For the first Bits in category d The unique hot tag below; For the final network iteration Code position belongs to The predicted probability of a class; L This is the total length of the PRN code sequence; D This represents the number of classes corresponding to binary codes 0 and 1.

[0037] In this embodiment, the GNSS pseudocode estimate is measured using a loss function. The model is trained by comparing the difference between the pseudocode value and the actual GNSS code value.

[0038] It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0039] In one embodiment, such as Figure 4 As shown, a GNSS pseudocode estimation device based on TurboUNet is provided, comprising: The input feature generation unit is used to combine the received coherent adaptive subcarrier modulated complex baseband GNSS signal with locally generated features. C One-code sequence concatenation generates input features; the complex baseband GNSS signal is obtained through IQ modulation, which employs... C 1 code C 2 codes and C 3 digits.

[0040] The pseudocode estimation unit is used to concatenate the input features of the current iteration round and the soft probabilities output by the UNet module in the previous iteration round, and input the concatenation result back into the UNet module to obtain the soft probability output of the current iteration round. This process continues until a preset number of iterations, outputting the soft probabilities as the soft probability estimate of the GNSS pseudocode. Based on this soft probability estimate, the GNSS pseudocode estimation value is obtained. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

[0041] In one embodiment, the received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with a locally generated... C One-code sequence concatenation generates input features, including: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is decomposed into in-phase and quadrature components, and then compared with locally generated... C One-code sequence concatenation forms a three-channel one-dimensional input feature: ; in, For input features; This refers to the in-phase component of a complex baseband GNSS signal; These are the orthogonal components of the complex baseband GNSS signal; for C 1-code sequence; This refers to the batch size processed in a single training iteration. For feature dimensions.

[0042] In one embodiment, the concatenation result is input into the UNet module to obtain the soft probability output for the current iteration round, including: The stitching result is input into the UNet module; the UNet module includes multiple encoders connected sequentially for downsampling, wherein the first... i The input terminal of the encoder is connected to the first... i -1 encoder output terminal, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The input data for the decoder; i ∈[0,…, I The input of the bottleneck layer is connected to the first... I The output data of the bottleneck layer is input to the output of the encoder. I A gated jump connection unit. The splicing result is input into the 0th encoder, processed by the UNet module, and the soft probability of the current iteration is output by the 0th decoder.

[0043] In one embodiment, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased.

[0044] In one embodiment, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased, including: ; in, For the first i Advanced temporal features obtained through subsampling; To correct the linear unit; For batch normalization processing; It is a one-dimensional convolution; Indicates the fill size; Indicates the first i The step size of the next downsampling, , .

[0045] In one embodiment, the input features of the current iteration and the soft probabilities output by the UNet module of the previous iteration are concatenated, including: ; in, For the first n Input features for the next iteration; For input features; For the first n The soft probability of the -1st iteration. For the 1st iteration, .

[0046] In one embodiment, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The decoder's input data includes: No. i The first gated jump connection unit controls the input of the first... i The encoded data from the first encoder is used to recover temporal resolution through transposed convolution. i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i Decoder input data: ; ; in, For the first i The input data for the decoder; For learnable gating; For the first i +1 decoder's decoded data; For the first i Encoded data from one encoder; It is the inner product; The sigmoid function represents element-wise multiplication; It is a one-dimensional convolution.

[0047] In one embodiment, the concatenation result is input into the 0th encoder, processed by the UNet module, and then the 0th decoder outputs the soft probability of the current iteration, including: The concatenation result is input into the 0th encoder. After being processed by the UNet module, the 0th decoder outputs the decoded data of the current iteration round. The decoded data is then converted into soft probabilities through a 1×1 convolution before being output.

[0048] In one embodiment, the soft probability outputs from each iteration are stacked to form a three-dimensional tensor: ; in, Indicates channel c In position k Belongs to symbols b The probability, It is the batch size for each iteration.

[0049] Taking argmax from the symbolic dimension of the three-dimensional tensor, we obtain the hard decision matrix. : ; In the hard decision matrix, the first row is... C Two-digit judgment, second act C 3-bit decision. Since argmax is a non-differentiable operation, soft probabilities are retained during training for loss calculation and uncertainty analysis.

[0050] exist After rounds of iteration, the last UNet module is passed The output is used as a GNSS pseudocode estimate. : ; This tensor pairs each encoded channel ( C 2-digit code and C The soft probability estimate for each chip location (3-bit) is encoded. There is no additional fusion layer, consistent with the network implementation.

[0051] In one embodiment, the device further includes: The training module is used for end-to-end training using cross-entropy loss, and measures the difference between the estimated GNSS pseudocode value and the true GNSS pseudocode value using a loss function. ; in, The loss function; For the first Each bit in category d The unique hot tag below; For the final network iteration Code position belongs to The predicted probability of a class; L This is the total length of the PRN code sequence;D This represents the number of classes corresponding to binary codes 0 and 1.

[0052] Specific limitations regarding the TurboUNet-based GNSS pseudocode estimation device can be found in the limitations of the TurboUNet-based GNSS pseudocode estimation method described above, and will not be repeated here. Each module in the aforementioned TurboUNet-based GNSS pseudocode estimation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in hardware or independent of the processor in a computer device, or stored in software in the memory of a computer device, so that the processor can call and execute the corresponding operations of each module.

[0053] Reference Figure 3 This paper presents GNSS pseudocode estimation results using different methods. The bit error rate (BER) results for the code estimation cover SNR conditions from 0 dB to -20 dB. Figure 1 As can be seen, compared with the UNet method, ResNet method, CNN method, and Traditional method, the method described in this invention has advantages in... C 2-digit code and C The 3-code consistently achieves the lowest BER (Bit Error Rate) value under both conditions.

[0054] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0055] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A GNSS pseudocode estimation method based on TurboUNet, characterized in that, Includes the following steps: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with the locally generated... C One-code sequence concatenation generates input features; wherein, the complex baseband GNSS signal is obtained through IQ modulation, and the IQ modulation employs... C 1 code C 2 codes and C 3 codes; The input features of the current iteration are concatenated with the soft probabilities output by the UNet module from the previous iteration, and the concatenated result is input into the UNet module to obtain the soft probability output of the current iteration. This process is repeated until a preset number of iterations are reached. The output soft probabilities are then used as the soft probability estimate of the GNSS pseudocode, and the GNSS pseudocode estimate is obtained based on this soft probability estimate. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

2. The method according to claim 1, characterized in that, The received coherent adaptive subcarrier modulated complex baseband GNSS signal is combined with the locally generated... C One-code sequence concatenation generates input features, including: The received coherent adaptive subcarrier modulated complex baseband GNSS signal is decomposed into in-phase and quadrature components, and then compared with locally generated... C One-code sequence concatenation forms a three-channel one-dimensional input feature: ； in, Input features; This refers to the in-phase component of a complex baseband GNSS signal; These are the orthogonal components of the complex baseband GNSS signal; for C 1-code sequence; This refers to the batch size processed in a single training iteration. For feature dimensions.

3. The method according to claim 2, characterized in that, The concatenated result is input into the UNet module to obtain the soft probability output for the current iteration, including: The stitching result is input into the UNet module; the UNet module includes: multiple encoders connected sequentially for downsampling, wherein the first... i The input terminal of the encoder is connected to the first... i -1 encoder output terminal, the first i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The input data for the decoder; ; The input of the bottleneck layer is connected to the first I The output data of the bottleneck layer is input to the output of the encoder. I One gated jump connection unit; The splicing result is input into the 0th encoder, processed by the UNet module, and then the soft probability of the current iteration is output by the 0th decoder.

4. The method according to claim 3, characterized in that, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased.

5. The method according to claim 4, characterized in that, i When =0, the first i When an encoder downsamples the stitching result, it maintains the original temporal resolution. i When >0, the first i When downsampling an encoder, the implicit encoder features are downsampled by a factor of 2, and the channel dimension is increased, including: ； in, For the first i Advanced temporal features obtained through subsampling; To correct the linear unit; For batch normalization processing; It is a one-dimensional convolution; Indicates the fill size; Indicates the first i The step size of the next downsampling.

6. The method according to claim 1, characterized in that, The input features of the current iteration are concatenated with the soft probabilities output by the UNet module of the previous iteration, including: ； in, For the first n Input features for the next iteration; Input features; For the first n The soft probability of -1 iteration.

7. The method according to claim 3, characterized in that, No. i The encoded data of the encoder is input to the first encoder. i The first gated jump connection unit, the first i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i The decoder's input data includes: No. i The first gated jump connection unit controls the input of the first... i The encoded data from each encoder is transposed and convolved to restore temporal resolution; No. i The first gated jump connection unit according to the first i The encoded data of the first encoder and the first i The decoded data of +1 decoder yields the first... i Decoder input data: ； ； in, For the first i The input data for the decoder; For learnable gating; For the first i +1 decoder's decoded data; For the first i Encoded data from one encoder; It is the inner product; It is the sigmoid function; It is a one-dimensional convolution.

8. The method according to claim 3, characterized in that, The concatenation result is input into the 0th encoder, processed by the UNet module, and then the soft probability of the current iteration is output by the 0th decoder. This also includes: The splicing result is input into the 0th encoder. After being processed by the UNet module, the 0th decoder outputs the decoded data of the current iteration round. The decoded data is then converted into soft probabilities through a 1×1 convolution before being output.

9. The method according to claim 1, characterized in that, The method further includes: Training is performed end-to-end using cross-entropy loss, and the difference between the estimated GNSS pseudocode value and the true GNSS pseudocode value is measured by the loss function: ； in, The loss function; For the first Each bit in category d The unique hot tag below; For the final network iteration Code position belongs to The predicted probability of a class; L This is the total length of the PRN code sequence; D This represents the number of classes corresponding to binary codes 0 and 1.

10. A GNSS pseudocode estimation device based on TurboUNet, characterized in that, include: The input feature generation unit is used to combine the received coherent adaptive subcarrier modulated complex baseband GNSS signal with locally generated features. C One-code sequence concatenation generates input features; wherein, the complex baseband GNSS signal is obtained through IQ modulation, and the IQ modulation employs... C 1 code C 2 codes and C 3 codes; The pseudocode estimation unit is used to concatenate the input features of the current iteration round and the soft probabilities output by the UNet module in the previous iteration round, and input the concatenation result back into the UNet module to obtain the soft probability output of the current iteration round. This process continues until a preset number of iterations, outputting the soft probabilities as the soft probability estimate of the GNSS pseudocode. Based on this soft probability estimate, the GNSS pseudocode estimation value is obtained. C 2-code sequence and C A 3-code sequence; wherein the soft probability is the soft posterior probability of each bit within each chip interval.

Images