High-dimensional lossy information source decoding method and device for automatic driving

Abstract

The invention provides a high-dimensional lossy information source decoding method and device for automatic driving, and the method comprises the steps: inputting n-dimensional information source data from a vehicle-mounted sensor into a lossy information source encoder network, and obtaining a k-dimensional quaternary sequence; on the basis of a pre-stored check matrix of protograph low-density parity check codes, algebraic reconstruction is carried out on the k-dimensional quaternary sequence on an integer ring, and n-dimensional code words meeting check constraints are generated; performing centralized coding on the n-dimensional code word, and mapping a discrete quaternary symbol in the n-dimensional code word into a continuous numerical sequence taking zero as a center; and inputting the continuous numerical sequence into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating point sequence for automatic driving to perform environment perception. Therefore, by utilizing the scheme provided by the embodiment of the invention, the decoding efficiency and the data reconstruction quality in the automatic driving system are effectively improved.

Description

Technical Field

[0001] This invention relates to the field of information processing, to communication physical layer decoding technology, and particularly to a method and apparatus for high-dimensional lossy source decoding for autonomous driving. Background Technology

[0002] High-dimensional lossy source encoding and decoding technology has demonstrated significant value in multiple fields. In autonomous driving systems, efficient compression, transmission, and accurate reconstruction of high-dimensional point cloud data generated by sensors such as LiDAR are core components for achieving safe and reliable environmental perception. This technology aims to leverage the inherent sparsity and correlation of high-dimensional data to significantly reduce the data load on in-vehicle networks and the storage pressure on computing units while ensuring that key features are not distorted.

[0003] Currently, D. Song et al., in their paper "Gaussian Source Coding Based on P-LDPC Code," proposed a lossy source coding and decoding system based on a cascaded confidence propagation-inverse confidence propagation algorithm, optimizing for efficient compression and reconstruction of Gaussian sources. The lossy source decoding network employs a fully connected layer structure, capable of reconstructing compressed binary sequences into Gaussian floating-point sources. Compared to traditional methods, this decoding scheme offers advantages such as simple implementation, low iterative complexity, and ease of parallelization, providing a new technical path for efficient information representation and reconstruction.

[0004] However, in real-time applications where safety is critical, such as autonomous driving, the compressed data stream received from the vehicle network must be recovered into the original sensor readings extremely quickly and accurately. Current lossy source decoding methods mainly target binary information for reconstruction, and there are bottlenecks in information density and decoding efficiency.

[0005] Therefore, how to solve the problem of efficient and accurate reconstruction of compressed sensor data streams in autonomous driving systems is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] This invention provides a method and apparatus for high-dimensional lossy source decoding in autonomous driving systems, which can effectively improve decoding efficiency and data reconstruction quality. Compared to the original binary representation, the high-dimensional data compression provided by this invention can represent multi-dimensional information of autonomous driving, achieving efficient multi-target information representation and providing a new technical path for autonomous driving recognition schemes.

[0007] The first aspect of this invention provides a method for decoding high-dimensional lossy sources for autonomous driving, comprising: The n-dimensional source data from the vehicle sensor is input into the lossy source encoder network to obtain the k-dimensional quaternary sequence; Based on the parity-check matrix of the pre-stored prototype graph low-density parity-check code, in the integer ring The above algebraic reconstruction of the k-dimensional quaternary sequence generates an n-dimensional codeword that satisfies the check constraint; The n-dimensional codeword is centrally encoded, and its discrete quaternary symbols are mapped to a continuous numerical sequence centered at zero. A continuous numerical sequence is input into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for autonomous driving to perceive the environment.

[0008] Optionally, based on the parity-check matrix of the pre-stored prototype low-density parity-check code, in the integer ring... The above algebraic reconstruction of a k-dimensional quaternary sequence generates an n-dimensional codeword that satisfies the check constraint, including: The information bit index is determined based on the configuration information of the low-density parity check code in the original model diagram; The received quaternary sequence is filled into the codeword vector of length n at the position specified by the information bit index; Based on the row echelon form of the parity check matrix, an inverse substitution method involving modulo 4 operations is used to calculate and fill all the parity bits of the codeword vector, generating an n-dimensional codeword that satisfies the parity check constraints.

[0009] Optionally, the process of calculating the check bit value is implemented using the following formula:

[0010] in, This represents the value of the check bit that needs to be solved. The element in the i-th row and j-th column of the row echelon check matrix. The codeword vector excluding the currently determined check bit Other determined bits besides; This indicates a modulo 4 operation.

[0011] Optionally, before centrally encoding the n-dimensional codeword, the method further includes: Verify whether the n-dimensional codeword satisfies the following check equation:

[0012] This represents converting an n-dimensional codeword from a row vector to a column vector. Represents a row echelon form parity check matrix; This indicates a modulo 4 operation.

[0013] Optionally, the deep decoder network includes an input layer, a hidden layer, and an output layer, wherein: The input layer maps continuous numerical sequences to a 4n-dimensional hidden space through a linear transformation; The hidden layer uses the GELU activation function; The output layer maps the 4n-dimensional representation of the hidden layer back to the n-dimensional reconstructed information source.

[0014] Optionally, the weights of the input, hidden, and output layers of the deep decoder network are initialized using a uniform Xavier distribution.

[0015] Optionally, the centralized encoding maps the quaternary symbols {0, 1, 2, 3} to {-1.5, -0.5, 0.5, 1.5} respectively.

[0016] A second aspect of the present invention provides a high-dimensional lossy source decoding device for autonomous driving, installed in an in-vehicle communication system, for reconstructing point cloud data collected by sensors, comprising: The quantization unit is used to quantize the n-dimensional source data from the vehicle sensor into a k-dimensional quaternary sequence by inputting it into the lossy source encoder network; The algebraic reconstruction unit is used to perform algebraic reconstruction on the k-dimensional quaternary sequence on the integer ring based on the parity check matrix of the pre-stored original model low-density parity check code, generating an n-dimensional codeword that satisfies the parity check constraint. The centralization unit is used to centrally encode n-dimensional codewords, mapping the discrete quaternary symbols in them into a continuous numerical sequence centered at zero. The decoding unit is used to input continuous numerical sequences into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for autonomous driving to perform environmental perception.

[0017] A third aspect of the present invention provides a high-dimensional lossy source decoding device for autonomous driving, comprising: One or more processors; A memory on which one or more programs are stored; When one or more programs are executed by one or more processors, the one or more processors implement a method for high-dimensional lossy source decoding for autonomous driving, as described above.

[0018] A fourth aspect of the present invention provides a computer storage medium for storing a program, which, when executed, is used to implement a method for high-dimensional lossy source decoding for autonomous driving as described in any of the above.

[0019] Compared with the prior art, the beneficial effects of the present invention include: Improving Information Density and Decoding Efficiency: This invention employs quaternary encoding, where each symbol carries 2 bits of information, resulting in twice the information density of binary encoding. When processing the same amount of data, the number of symbols required is halved, significantly reducing the complexity of the decoding algorithm and improving processing speed and efficiency, thus meeting the stringent requirements of real-time applications such as autonomous driving.

[0020] Improving the quality of reconstructed signals: Quaternary encoding provides a finer level of quantization, which can effectively reduce quantization noise compared to binary schemes, improve the accuracy of reconstructed data, and provide a more reliable data foundation for subsequent environmental perception algorithms.

[0021] Combining the advantages of algebra and deep learning: This invention innovatively combines the algebraic reconstruction characteristics of protograph low-density parity-check (P-LDPC) codes with the powerful nonlinear mapping capabilities of deep learning. First, the validity of codewords is guaranteed using algebraic methods. Then, a neural network is used to learn the complex mapping from codewords to the information source, fully leveraging the advantages of both to achieve superior reconstruction performance with the same network capacity.

[0022] High engineering practicality: The high-dimensional lossy source decoding method and device for autonomous driving of the present invention can be directly integrated into existing vehicle communication units or edge computing devices without modifying the hardware infrastructure, and has good feasibility and application prospects. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 A flowchart illustrating a high-dimensional lossy source decoding method for autonomous driving provided in an embodiment of the present invention; Figure 2 A schematic diagram of the lossy source decoding process provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a high-dimensional lossy source decoding device for autonomous driving provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a high-dimensional lossy source decoding device for autonomous driving, provided as an embodiment of the present invention. Detailed Implementation

[0025] This invention provides a method and apparatus for high-dimensional lossy source decoding in autonomous driving. It accurately reconstructs a compressed quaternary data point into the original floating-point number, providing a data reconstruction scheme with higher decoding efficiency. It reconstructs the compressed quaternary data sequence received from the vehicle network into a Gaussian distributed floating-point number sequence, which can effectively improve the decoding efficiency and data reconstruction quality in autonomous driving systems.

[0026] Compared to existing binary decoding schemes, this technique proposes a high-dimensional lossy source decoding method based on quaternion. Since each quaternary symbol carries twice the information of a binary symbol, the decoder needs to process half the number of symbols to handle the same information content, significantly improving the convergence speed and processing efficiency of the decoding algorithm. Especially in scenarios sensitive to quantization noise, such as LiDAR intensity values, quaternion-based reconstruction provides more refined posterior information, improving the peak signal-to-noise ratio of the reconstructed signal by 2-4 dB, which is crucial for subsequent target detection and recognition algorithms.

[0027] See Figure 1 This figure is a flowchart illustrating a high-dimensional lossy source decoding method for autonomous driving provided by an embodiment of the present invention. The high-dimensional lossy source decoding method for autonomous driving provided by this embodiment of the present invention can be implemented, for example, through the following steps S101-104.

[0028] Combination Figure 2 To explain, Figure 2 The schematic diagram of the lossy source decoding process provided in this embodiment of the invention illustrates the module logic and data flow from compressed data to reconstructed source: First, the input is a compressed quaternary sequence; then, the codeword is reconstructed algebraically, and a complete codeword satisfying the check constraint is generated based on the P-LDPC code parity-check matrix; next, the quaternary codeword is mapped to a continuous value centered at zero through centralized encoding (e.g., the quaternary symbol {0,1,2,3} is mapped to {-1.5,-0.5,0.5,1.5}); the area within the dashed box is the "deep decoder," which contains a multi-layer network structure of input layer → hidden layer → output layer, and performs nonlinear transformation on the centralized value; finally, the output is a reconstructed Gaussian floating-point sequence (restored to the original high-dimensional floating-point source of LiDAR, used for environmental perception in autonomous driving).

[0029] S101: Input n-dimensional source data from vehicle-mounted sensors into a lossy source encoder network to obtain a k-dimensional quaternary sequence.

[0030] In this embodiment of the invention, n-dimensional floating-point data from an onboard LiDAR sensor, which approximately follows a standard normal distribution, is input into a lossy source encoder network and quantized into a k-dimensional quaternary compressed sequence u, where the dimension satisfies n≥k.

[0031] Specifically, the lossy source encoder network adopts a deep fully connected structure, comprising an input layer, a feature extraction layer, and a quantization output layer, used to compress and encode n-dimensional floating-point source data into a k-dimensional quaternary sequence. The input layer receives n-dimensional floating-point data from an onboard LiDAR sensor, which approximately follows a standard normal distribution. The feature extraction layer contains three fully connected layers, with each layer using a GELU activation function for non-linear transformation. The first fully connected layer maps the n-dimensional input to a 2n-dimensional feature space, the second fully connected layer maps the 2n-dimensional features to an n-dimensional feature space, and the third fully connected layer maps the n-dimensional features to a k-dimensional feature space. The quantization output layer performs quaternary quantization on continuous features, using a learnable threshold to quantize continuous values ​​into a quaternary compressed sequence u. The quantization function employs a uniform quantization strategy, and the quantization step size is adaptively adjusted according to the statistical characteristics of the source.

[0032] S102: Parity check matrix based on pre-stored prototype graph low-density parity check code, in integer ring. The k-dimensional quaternary sequence is algebraically reconstructed to generate an n-dimensional codeword that satisfies the check constraint.

[0033] In this embodiment of the invention, the information bit index is determined according to the configuration information of the original modulus low-density parity check code; the received quaternary sequence is filled into the codeword vector of length n at the position specified by the information bit index; based on the row ladder form of the parity check matrix, all parity bits of the codeword vector are calculated and filled by the inverse substitution method including modulo 4 operation to generate an n-dimensional codeword that satisfies the parity check constraint.

[0034] Specifically, an empty codeword vector of length n is created, and the compressed sequence is filled into the positions in this vector determined by the information bit indices of the P-LDPC code, cfg.kept_indices. At this point, the information bits are determined, and the check bits need to be calculated. Based on the given P-LDPC code, a row echelon check matrix is ​​implemented, and all check bits are calculated and filled using the inverse substitution method.

[0035] Process each check bit sequentially in reverse order of its index. For the current check bit to be calculated, its corresponding check equation contains only one unknown quantity, while all other terms are determined. The value of the current check bit is obtained by calculating the dot product of this check equation and the codeword vector currently partially filled, and then performing a modulo-4 operation on the result. The calculation formula is as follows:

[0036] in, This represents the value of the check bit that needs to be solved. The element in the i-th row and j-th column of the row echelon check matrix. The codeword vector excluding the currently determined check bit Other determined bits besides; This indicates a modulo 4 operation.

[0037] Fill the calculated parity bit values ​​into the corresponding positions in the codeword vector, and repeat the above calculation process until all parity bits have been calculated.

[0038] In one implementation of this invention, after the algebraic reconstruction is completed, the generated codeword is represented as follows: Simultaneously, the validity is verified based on the parity check matrix of the P-LDPC code to ensure that it satisfies the parity check equation, as shown below:

[0039] in, This represents converting an n-dimensional codeword from a row vector to a column vector. Represents a row echelon form parity check matrix; This indicates a modulo-4 operation. After successful verification, proceed to step S103.

[0040] S103: Center the n-dimensional codeword and map the discrete quaternary symbols in it to a continuous numerical sequence centered at zero.

[0041] In this embodiment of the invention, the codeword Implement centralized coding, that is The quaternary symbols {0, 1, 2, 3} are mapped to {-1.5, -0.5, 0.5, 1.5} respectively. This centralized coding transforms the discrete quaternary sequence into a continuous numerical sequence representation centered at zero, enabling the decoder network to learn the mapping relationship from codewords to information sources more effectively.

[0042] S104: Input the continuous numerical sequence into the pre-trained deep decoder network to obtain the reconstructed Gaussian floating-point sequence for autonomous driving to perform environmental perception.

[0043] In this embodiment of the invention, the deep decoder network includes an input layer, a hidden layer, and an output layer, wherein: the input layer maps a continuous numerical sequence to a 4n-dimensional hidden space through a linear transformation; the hidden layer uses the GELU activation function; and the output layer maps the 4n-dimensional representation of the hidden layer back to an n-dimensional reconstructed information source. The weights of the input, hidden, and output layers of the deep decoder network are all initialized using a Xavier uniform distribution.

[0044] For details, see Figure 2 The centralized codeword will be reconstructed. Input depth decoder network The network structure mainly consists of an input layer, hidden layers, and an output layer. First, the input layer reconstructs the codewords in an n-dimensional centralized manner. The hidden layer is mapped to a 4n-dimensional hidden space through a linear transformation; secondly, the hidden layer uses the GELU activation function to provide nonlinear transformation capability; subsequently, the output layer maps the 4n-dimensional representation of the hidden layer back to the n-dimensional reconstructed source. The weights of each layer are initialized using a uniform Xavier distribution.

[0045] This invention provides a method and apparatus for high-dimensional lossy source decoding in autonomous driving. In this method, n-dimensional source data from onboard sensors is input into a lossy source encoder network to obtain a k-dimensional quaternary sequence. Based on the parity-check matrix of a pre-stored original model low-density parity-check code, the k-dimensional quaternary sequence is algebraically reconstructed on an integer ring to generate an n-dimensional codeword that satisfies the parity-check constraint. The n-dimensional codeword is then centered and encoded, mapping the discrete quaternary symbols to a continuous numerical sequence centered at zero. This continuous numerical sequence is input into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for environmental perception in autonomous driving. Therefore, by utilizing the scheme provided in this invention, the compressed quaternary sequence is reconstructed into a floating-point sequence following a standard normal distribution. While learning source features, the algebraic properties of the codeword are fully utilized, achieving better reconstruction performance with the same network capacity, effectively improving decoding efficiency and data reconstruction quality in autonomous driving systems.

[0046] Based on the methods provided in the above embodiments, this invention also provides a high-dimensional lossy source decoding device for autonomous driving. The high-dimensional lossy source decoding device for autonomous driving is described below with reference to the accompanying drawings.

[0047] See Figure 3 The figure is a schematic diagram of a high-dimensional lossy source decoding device for autonomous driving provided by an embodiment of the present invention.

[0048] The high-dimensional lossy source decoding device 300 for autonomous driving provided in this embodiment of the invention includes: a quantization unit 301, an algebraic reconstruction unit 302, a centralization unit 303, and a decoding unit 304.

[0049] The quantization unit 301 is used to input n-dimensional source data from the vehicle sensor into the lossy source encoder network and quantize it into a k-dimensional quaternary sequence. Algebraic reconstruction unit 302 is used to perform algebraic reconstruction on a k-dimensional quaternary sequence on an integer ring based on the parity check matrix of a pre-stored original model low-density parity check code, generating an n-dimensional codeword that satisfies the parity check constraint. The centralization unit 303 is used to centrally encode the n-dimensional codeword, mapping the discrete quaternary symbols therein to a continuous numerical sequence centered at zero; The decoding unit 304 is used to input a continuous numerical sequence into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for autonomous driving to perform environmental perception.

[0050] In one possible implementation, the algebraic reconstruction unit 302 is specifically used for: The information bit index is determined based on the configuration information of the low-density parity check code in the original model diagram; The received quaternary sequence is filled into the codeword vector of length n at the position specified by the information bit index; Based on the row echelon form of the parity check matrix, an inverse substitution method involving modulo 4 operations is used to calculate and fill all the parity bits of the codeword vector, generating an n-dimensional codeword that satisfies the parity check constraints.

[0051] In one possible implementation, the algebraic reconstruction unit 302 is specifically used for:

[0052] in, This represents the value of the check bit that needs to be solved. The element in the i-th row and j-th column of the row echelon check matrix. The codeword vector excluding the currently determined check bit Other determined bits besides; This indicates a modulo 4 operation.

[0053] In one possible implementation, the high-dimensional lossy source decoding device 300 for autonomous driving further includes a verification unit for: Verify whether the n-dimensional codeword satisfies the following check equation:

[0054] in, This represents converting an n-dimensional codeword from a row vector to a column vector. Represents a row echelon form parity check matrix; This indicates a modulo 4 operation.

[0055] In one possible implementation, the deep decoder network comprises an input layer, a hidden layer, and an output layer, wherein: The input layer maps continuous numerical sequences to a 4n-dimensional hidden space through a linear transformation; The hidden layer uses the GELU activation function; The output layer maps the 4n-dimensional representation of the hidden layer back to the n-dimensional reconstructed information source.

[0056] In one possible implementation, the weights of the input, hidden, and output layers of the deep decoder network are initialized using a uniform Xavier distribution.

[0057] In one possible implementation, centralized encoding maps the quaternary symbols {0, 1, 2, 3} to {-1.5, -0.5, 0.5, 1.5} respectively.

[0058] Since the high-dimensional lossy source decoding device 300 for autonomous driving is a device corresponding to the high-dimensional lossy source decoding method for autonomous driving provided in the above method embodiments, the specific implementation of each unit of the high-dimensional lossy source decoding device 300 for autonomous driving is based on the same concept as in the above method embodiments. Therefore, for the specific implementation of each unit of the high-dimensional lossy source decoding device 300 for autonomous driving, please refer to the description of the high-dimensional lossy source decoding method for autonomous driving in the above method embodiments, and it will not be repeated here.

[0059] This invention also provides a high-dimensional lossy source decoding device for autonomous driving, the device comprising: a processor and a memory; The memory is used to store instructions; The processor is configured to execute the instructions in the memory to perform the high-dimensional lossy source decoding method for autonomous driving, which is executed by the analysis device as mentioned in the above embodiments.

[0060] It should be noted that the hardware structure of the high-dimensional lossy source decoding device for autonomous driving provided in the embodiments of the present invention can be as follows: Figure 4 The structure shown, Figure 4 This is a schematic diagram of the structure of a device provided in an embodiment of the present invention.

[0061] Please see Figure 4 As shown, device 400 includes: processor 410, communication interface 420, and memory 430. The number of processors 410 in device 400 can be one or more. Figure 4 Taking a processor as an example, in this embodiment of the invention, the processor 410, communication interface 420, and memory 430 can be connected via a bus system or other means. Figure 4 Taking the connection between China and Israel via the bus system 440 as an example.

[0062] Processor 410 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP. Processor 410 may further include hardware chips. These hardware chips may be application-specific integrated circuits (ASICs), programmable logic devices (PLDs), or combinations thereof. The PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0063] The memory 430 may include volatile memory, such as random-access memory (RAM); the memory 430 may also include non-volatile memory, such as flash memory, hard disk drive (HDD) or solid-state drive (SSD); the memory 430 may also include a combination of the above types of memory.

[0064] Optionally, the memory 430 stores an operating system and programs, executable modules, or data structures, or subsets thereof, or extended sets thereof. The programs may include various operation instructions for implementing various operations. The operating system may include various system programs for implementing various basic services and handling hardware-based tasks. The processor 410 can read the programs from the memory 430 to implement the high-dimensional lossy source decoding method for autonomous driving provided in this embodiment of the invention.

[0065] The bus system 440 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The bus system 440 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0066] This invention also provides a computer-readable storage medium including instructions that, when executed on a computer, cause the computer to perform the high-dimensional lossy source decoding method for autonomous driving mentioned in the above embodiments.

[0067] This invention also provides a computer program product containing instructions that, when run on a computer, cause the computer to execute the high-dimensional lossy source decoding method for autonomous driving mentioned in the above embodiments.

[0068] Although the invention has been specifically shown and described in conjunction with preferred embodiments, those skilled in the art should understand that various changes in form and detail may be made to the invention without departing from the spirit and scope of the invention as defined in the appended claims, all of which shall be within the scope of protection of the invention.

Claims

1. A method for decoding high-dimensional lossy sources for autonomous driving, characterized in that, The method is executed by an onboard communication system and is used to reconstruct point cloud data acquired by sensors, including: The n-dimensional source data from the vehicle sensor is input into the lossy source encoder network to obtain the k-dimensional quaternary sequence; Based on the parity-check matrix of the pre-stored prototype graph low-density parity-check code, in the integer ring The k-dimensional quaternary sequence is algebraically reconstructed to generate an n-dimensional codeword that satisfies the check constraint. The n-dimensional codeword is centered and encoded, and the discrete quaternary symbols therein are mapped to a continuous numerical sequence centered at zero. The continuous numerical sequence is input into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for autonomous driving to perform environmental perception.

2. The high-dimensional lossy source decoding method according to claim 1, characterized in that, The parity check matrix based on the pre-stored original model low-density parity check code is in the integer ring. The above algebraic reconstruction of the k-dimensional quaternary sequence to generate an n-dimensional codeword that satisfies the check constraint includes: The information bit index is determined based on the configuration information of the low-density parity check code in the original model diagram; The received quaternary sequence is filled into the codeword vector of length n at the position specified by the information bit index; Based on the row echelon form of the parity check matrix, all parity bits of the codeword vector are calculated and filled using the inverse substitution method involving modulo 4 operations to generate an n-dimensional codeword that satisfies the parity check constraints.

3. The high-dimensional lossy source decoding method according to claim 2, characterized in that, The process of calculating the check digit value is achieved through the following formula: in, This represents the value of the check bit that needs to be solved. The element in the i-th row and j-th column of the row echelon check matrix. The codeword vector excluding the currently determined check bit Other determined bits besides; This indicates a modulo 4 operation.

4. The high-dimensional lossy source decoding method according to claim 1, characterized in that, Before performing centralized encoding on the n-dimensional codeword, the method further includes: Verify whether the n-dimensional codeword satisfies the following check equation: in, This represents converting an n-dimensional codeword from a row vector to a column vector. Represents a row echelon form parity check matrix; This indicates a modulo 4 operation.

5. The high-dimensional lossy source decoding method according to claim 1, characterized in that, The deep decoder network comprises an input layer, a hidden layer, and an output layer, wherein: The input layer maps the continuous numerical sequence to a 4n-dimensional hidden space through a linear transformation. The hidden layer uses the GELU activation function; The output layer maps the 4n-dimensional representation of the hidden layer back to the n-dimensional reconstructed information source.

6. The high-dimensional lossy source decoding method according to claim 5, characterized in that, The weights of the input layer, the hidden layer, and the output layer of the deep decoder network are all initialized using a Xavier uniform distribution.

7. The high-dimensional lossy source decoding method according to claim 1, characterized in that, The centralized encoding maps the quaternary symbols {0, 1, 2, 3} to {-1.5, -0.5, 0.5, 1.5} respectively.

8. A high-dimensional lossy source decoding device for autonomous driving, characterized in that, The device is installed in the vehicle communication system and is used to reconstruct point cloud data collected by sensors, including: The quantization unit is used to quantize the n-dimensional source data from the vehicle sensor into a k-dimensional quaternary sequence by inputting it into the lossy source encoder network; The algebraic reconstruction unit is used to perform algebraic reconstruction on the k-dimensional quaternary sequence on the integer ring based on the parity check matrix of the pre-stored original model low-density parity check code, to generate an n-dimensional codeword that satisfies the parity check constraint. A centralization unit is used to centrally encode the n-dimensional codeword, mapping the discrete quaternary symbols therein to a continuous numerical sequence centered at zero; The decoding unit is used to input the continuous numerical sequence into a pre-trained deep decoder network to obtain a reconstructed Gaussian floating-point sequence for autonomous driving to perform environmental perception.

9. A high-dimensional lossy source decoding device for autonomous driving, characterized in that, The device includes: a processor and a memory; The memory is used to store instructions; The processor is configured to execute the instructions in the memory to perform the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Including instructions that, when run on a computer, cause the computer to perform the method described in any one of claims 1-7 above.

Images