Non-cooperative communication system channel coding blind identification method based on multi-feature fusion

Abstract

The invention belongs to the technical field of communication signal processing and code identification, and solves the problem of channel code identification when a non-cooperative communication system lacks coding parameter prior information and training data with labels. According to the technical scheme, the method is characterized in that a unified recognition framework is constructed based on five features, namely, a rank fluctuation feature is used for distinguishing convolutional codes and linear block codes, a continuous zero spectrum feature is used for recognizing BCH codes, a multi-dimensional structure feature is used for detecting Turbo product codes, a bit distribution deviation feature is used for recognizing polarization codes, and an average absolute column correlation feature is used for distinguishing LDPC codes and non-LDPC codes; blind recognition of five types of channel codes including BCH codes, LDPC codes, convolutional codes, Turbo product codes and polarization codes is achieved. The method does not depend on coding parameter prior information, does not need training, is high in recognition rate under a low signal-to-noise ratio, is high in generalization capability, and is suitable for real-time and resource-limited scenes.

Description

Technical Field

[0001] This invention belongs to the field of communication signal processing and coding recognition technology, specifically relating to a channel coding blind recognition method that does not require prior parameters and training data in non-cooperative communication scenarios. Background Technology

[0002] In non-cooperative communication systems, the receiver often cannot know the channel coding type and its parameters used by the transmitter, such as code length, code rate, and generator polynomial. Traditional channel coding identification methods mainly rely on the algebraic properties of specific codes or deep neural networks based on supervised learning. The former, such as utilizing the rank defect of the parity check matrix or the frozen bit property of polar codes, is effective under ideal conditions, but has poor generalization ability and is difficult to apply to multiple coding families; the latter, such as CNN and BiLSTM, can automatically learn features, but heavily rely on a large amount of labeled data, and the model has weak generalization ability, with performance deteriorating sharply at low signal-to-noise ratios or when parameters change. Therefore, existing technologies have the following drawbacks:

[0003] (1) It depends on a specific encoding structure and cannot uniformly handle multiple encodings;

[0004] (2) It has a high data dependency and requires a large amount of labeled data to train the model;

[0005] (3) Performance degrades significantly at low signal-to-noise ratios;

[0006] (4) It has high computational complexity and is difficult to run in real time in resource-constrained environments. Summary of the Invention

[0007] The present invention aims to provide a channel coding blind identification method that is parameter-independent, training-independent, and applicable to multiple coding families, overcoming the dependence of existing technologies on specific coding structures and labeled data, and improving the robustness and real-time performance of identification in low signal-to-noise ratio environments.

[0008] To achieve the above objectives, this invention proposes a hierarchical recognition framework based on five types of robust features, which include:

[0009] (1) Rank fluctuation characteristics: By analyzing the rank change pattern of the observation matrix under the sliding window, we can distinguish between convolutional codes with aperiodic rank characteristics and block codes with deterministic rank periodicity.

[0010] (2) Continuous zero spectrum feature: Identification is performed by utilizing the algebraic construction feature of the continuous spectrum zeros of BCH code under the generalized finite field Fourier transform;

[0011] (3) Multidimensional structural features: TPC is identified by detecting whether the parity check matrix has a block structure composed of Kronecker products;

[0012] (4) Bit distribution deviation characteristics: Utilizing the statistical characteristic that the proportion of bit '1' in the encoded sequence of polar codes decreases significantly under specific segments due to the fixed frozen bits;

[0013] (5) Average absolute column correlation: taking advantage of the fact that the inter-column correlation of LDPC codes is weaker than that of non-LDPC codes with global constraints due to the sparse parity-check matrix.

[0014] Based on the above characteristics, the blind identification method performs the following steps in sequence:

[0015] (1) Signal preprocessing and observation matrix construction: The received signal sequence is segmented according to the candidate code length to construct the observation matrix.

[0016] (2) Convolutional code identification: Based on the rank fluctuation characteristics, determine whether the rank of the observation matrix exhibits an indeterminate periodic change. If so, it is determined to be a convolutional code, and the process ends; otherwise, proceed to the next step.

[0017] (3) LDPC code recognition: Calculate the average absolute column correlation value of the observation matrix and compare it with the adaptive threshold. If it is lower than the threshold, it is determined to be an LDPC code and the process ends; otherwise, proceed to the non-LDPC code fine-grained recognition process.

[0018] (4) Fine-grained identification of non-LDPC codes: This stage distinguishes between BCH codes, polar codes, and TPC codes.

[0019] 1. BCH code recognition: Perform a generalized finite-field Fourier transform on the signal and detect whether there are continuous zeros in its spectrum.

[0020] 2. Polar code recognition: The signal sequence is segmented, and the proportion of '1' bits in each segment is calculated. The system then detects whether this proportion decreases significantly at a specific segment length.

[0021] 3. TPC Identification: Detect its multidimensional coding structure by estimating the length of row and column subcodes and verifying the two-dimensional structure.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] (1) Training is data-independent: No labeled data is required for model training, avoiding the cost of data collection and labeling, and having the ability to encode unknown new parameters.

[0024] (2) Parameter independence: The proposed features capture the intrinsic structural characteristics of the coding family rather than specific parameters, so they are applicable to various code length, code rate and other parameter configurations.

[0025] (3) Strong robustness under low signal-to-noise ratio: The mathematical features based on the coding nature still maintain good distinguishability under noise interference, and can still maintain a high recognition rate in low signal-to-noise ratio environments of 0~8 dB.

[0026] (4) Strong generalization ability: The framework is unified and can effectively identify five major categories of mainstream channel coding, including BCH, LDPC, convolutional codes, TPC, and polar codes. Attached Figure Description

[0027] (1) Figure 1 : Flowchart of the hierarchical recognition algorithm proposed in this invention;

[0028] (2) Figure 2 : Recognition accuracy of five types of codes under different signal-to-noise ratios;

[0029] (3) Figure 3 Comparison results with ResNet and BiLSTM in terms of recognition accuracy;

[0030] (4) Figure 4 Comparison results with ResNet and BiLSTM in terms of generalization ability. Detailed Implementation

[0031] The invention will now be described in detail with reference to the accompanying drawings and mathematical formulas.

[0032] 1. Signal preprocessing

[0033] Received signal sequence By window length Divided into observation matrices:

[0034]

[0035] Where is the total length of the sequence.

[0036] 2. Convolutional code recognition

[0037] Calculate under different window sizes The rank of a convolutional code has no deterministic periodicity, meaning that for any integer... ,have:

[0038]

[0039] If such non-periodic fluctuations are detected, they are identified as convolutional codes.

[0040] 3. LDPC code recognition

[0041] Calculate AACC:

[0042]

[0043] like If it is, then it is determined to be an LDPC code, where: .

[0044] 4. Fine-grained recognition of non-LDPC codes

[0045] (4.1) BCH code identification: Perform a generalized finite-field Fourier transform on the received sequence and calculate its spectrum. Detect whether there is a continuous spectrum zero points ( (For error correction capability), this feature is a necessary and sufficient condition for BCH codes.

[0046] (4.2) Polar code identification will receive the sequence According to different segment lengths Reassemble into a matrix, and apply the polar code-specific recovery transform to each row. Then calculate the proportion of '1' bits in the recovered sequence. Polar codes, due to the frozen bits being fixed at 0, have a segment length of [missing information]. Equal to the actual code length When it is an integer multiple of itself, It will drop sharply from about 0.5 to about ( (For bit rate). When the decrease Δ When the time comes, it is determined to be a polar code.

[0047] (4.3) TPC code recognition

[0048] TPC identification is a three-level process:

[0049] Line code length estimation: Traversing candidate lengths Reorganize the sequence into a matrix. Calculate the proportion of each row that has an even weight as a codeword. Correct line code length Will make A significant peak appeared:

[0050]

[0051] Subcode length estimation: in determining Then, the sequence is rearranged as follows: The matrix. Repeat the above even-number weight statistics process for each column to estimate the column subcode length. .

[0052] Structural verification: Verify whether the total sequence length meets the requirements. If the conditions are met, it is determined to be a TPC.

[0053] 5. Experimental Verification

[0054] Under additive white Gaussian noise (AWG) channel conditions, five types of coding systems—LDPC (IEEE 802.11n), polar codes, convolutional codes, BCH codes, and TPC—were tested. Experimental results show that the proposed method achieves an overall recognition accuracy exceeding 90% for all five types of coding systems at a signal-to-noise ratio (SNR) of 4 dB, and reaches 97.8% at an SNR of 8 dB. Compared to deep learning methods such as ResNet and BiLSTM, this invention demonstrates significant advantages in low SNR performance and generalization ability when dealing with coding systems with unknown parameters.

[0055] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A blind identification method for channel codes in a non-cooperative communication system, characterized in that, Includes the following steps: (1) Receive communication signals that have been channel-coded; (2) Construct the observation matrix and perform rank analysis on the observation matrix to identify convolutional codes; (3) If it is not a convolutional code, calculate the average absolute column correlation value and compare it with the preset threshold to identify low-density parity check codes. (4) If it is not a low-density parity-check code, further identify BCH code, polar code or Turbo product code through spectrum analysis, bit distribution analysis or multidimensional structure analysis.

2. The method according to claim 1, characterized in that, The rank analysis includes: (1) Divide the received sequence into multiple windows and construct an observation matrix; (2) Analyze the variation of the rank of the observation matrix with the window size; (3) If the rank change has no deterministic periodicity, it is determined to be a convolutional code; (4) If the rank change has deterministic periodicity and the period is equal to the code length, then it is determined to be a block code.

3. The method according to claim 1, characterized in that, The average absolute column correlation value is calculated as follows: in, For the observation matrix, the first Column and number Pearson correlation coefficient between columns The candidate code length.

4. The method according to claim 1, characterized in that, The spectral analysis is used to identify BCH codes, including: (1) Perform a generalized finite field transformation on the received sequence; (2) Detect whether there are continuous zero-value spectral lines in the spectrum; (3) If there are continuous zero-value spectral lines, it is determined to be a BCH code.

5. The method according to claim 1, characterized in that, The bit distribution analysis is used to identify polar codes, including: (1) Perform polar code decoding structure transformation on the received sequence; (2) Analyze whether the proportion of bit "1" decreases significantly at the actual code length or its integer multiples; (3) If the decrease is greater than the preset threshold, it is determined to be a polar code.

6. The method according to claim 1, characterized in that, The multidimensional structural analysis is used to identify Turbo product codes, including: (1) Estimate the code length of the row and column component codes; (2) Verify whether the total sequence length is equal to the product of the row and column code lengths; (3) If satisfied, it is determined to be a Turbo product code.

7. The method according to claim 6, characterized in that, The estimation of the row and column component code lengths is based on the statistical characteristics of even weights, and is achieved by maximizing the deviation of the proportion of even weight codewords.

8. A channel code blind identification system in a non-cooperative communication system, characterized in that, include: (1) Preprocessing module, used to receive signals and construct observation matrix; (2) A convolutional code recognition module, used to perform the rank analysis as described in claim 2; (3) LDPC identification module, used to perform the AACC calculation and comparison as described in claim 3; (4) Fine-grained recognition module, used to perform BCH, polarization, and TPC recognition as described in claims 4, 5, and 6.

9. A computer-readable storage medium having a computer program stored thereon that, when executed by a processor, implements the method as described in any one of claims 1 to 7.

10. A communication receiving device, comprising a memory and a processor, the memory storing a computer program, the processor executing the program to implement the method as described in any one of claims 1 to 7.

Images