Adaptive equalization and phase tracking method and system based on multi-branch parallel decision
Patent Information
- Application Number
- CN202611147195.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-30
- Publication Date
- 2026-10-13
AI Technical Summary
[0009]本发明的目的在于克服现有技术的缺点，提供一种基于多分支并行判决的自适应均衡与相位跟踪方法及系统，旨在通过为每个符号生成包含最近星座点和至少一个次近星座点的多个候选判决，构建多个并行处理分支，各分支独立执行联合均衡与相位跟踪更新，并以累积误差作为路径度量进行幸存分支选择和窗口回溯，从而在恶劣信道条件下提高接收机的误码性能和抗噪声能力，同时提供一种能够有效识别出错符号位置并进行纠错的技术手段；解决现有自适应均衡与载波相位恢复技术中，因以最近星座点作为单一硬判决依据而导致判决边界附近易发生误判、且缺乏纠错机制的问题
1、本发明通过为每个符号生成包含最近星座点和至少一个次近星座点的多个候选判决，构建多个处理分支进行并行探索，当某一处理分支因噪声干扰做出错误判决时，其他处理分支仍可能保持正确路径。配合基于累积误差的路径度量机制和幸存处理分支选择策略，系统能够在多个候选路径中动态选择最优路径，显著降低了判决边界附近的误判概率，提高了接收机在低信噪比条件下的误码性能。
Smart Images
Figure CN122909081A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of digital communication technology, and in particular to an adaptive equalization and phase tracking based on multi-branch parallel decision. Background Technology
[0002] In digital communication systems, the receiver needs to perform equalization and carrier phase recovery on the signal transmitted through the channel to compensate for the impact of channel distortion and phase noise on signal quality. Adaptive equalization and carrier phase recovery are key technologies in coherent optical and wireless communication receivers.
[0003] In traditional receiver architectures, adaptive equalizers and carrier phase recovery modules are typically processed serially and independently. That is, the signal first undergoes channel compensation via the adaptive equalizer, and then the equalized signal is sent to the carrier phase recovery module for phase noise compensation. In this architecture, the equalizer and phase recovery module operate independently, updating their parameters based on their respective cost functions.
[0004] In adaptive equalizers, conventional adaptive algorithms (such as Least Mean Square (LMS) or Recursive Least Squares (RLS) typically use the constellation point closest to the equalized output signal as the desired decision symbol for error calculation when calculating the cost function. For Constant Modulus (CMA) algorithms, the magnitude of the constellation point is used as a reference. An inherent characteristic of this approach is that when the received signal is subject to strong noise interference, if the equalized signal happens to fall near the constellation decision boundary, the equalizer is prone to incorrectly assigning the signal to an adjacent constellation point, resulting in misclassification.
[0005] When the equalizer misjudges a signal, the incorrect decision not only leads to demodulation errors in the current symbol but is also used as a reference for error calculation, resulting in erroneous updates to the equalizer weight coefficients. Furthermore, in a serial architecture, this erroneous decision can propagate to subsequent carrier phase recovery modules, causing incorrect phase estimates to be updated as well. This cascading propagation effect of local errors degrades the overall receiver performance under adverse channel conditions.
[0006] Some existing studies have proposed a joint equalization and carrier recovery processing scheme, which uses a unified error function to update the equalizer coefficients and carrier phase estimates simultaneously, thus improving the error propagation problem to some extent compared to the serial, independent processing method. However, when calculating errors, these schemes still rely on a hard decision mechanism that uses the nearest constellation point as the expected decision symbol, and do not solve the fundamental problem of misjudgment risk near the decision boundary.
[0007] In addition, some studies have improved receiver performance through different algorithms at the module level and multi-stage processing. For example, two different equalization algorithms (such as multi-mode and decision-guided algorithms) are run simultaneously and mixed between the two algorithms using a dynamic weighting function; or a two-layer adaptive equalizer structure is used, where the signal quality is evaluated by comparing the signal with multiple reference voltages after the first-stage equalization, and then the decision to start the second-stage equalization process is made based on the evaluation results. These schemes have attempted to address algorithm switching or hierarchical processing, but the cost function of each processing path still uses the nearest reference constellation point as the expected output when calculating errors.
[0008] In summary, existing technologies use the nearest constellation point as the hard decision criterion to calculate the error cost function during adaptive equalization and carrier phase recovery. In scenarios with low signal-to-noise ratio or poor channel conditions, signals near the decision boundary are prone to misjudgment, and there is a lack of effective error correction mechanisms to identify and correct such misjudgments. Summary of the Invention
[0009] The purpose of this invention is to overcome the shortcomings of existing technologies and provide an adaptive equalization and phase tracking method and system based on multi-branch parallel decision-making. This method aims to generate multiple candidate decisions for each symbol, including the nearest constellation point and at least one secondary nearest constellation point, constructing multiple parallel processing branches. Each branch independently performs joint equalization and phase tracking updates, and uses accumulated error as a path metric for surviving branch selection and window backtracking. This improves the receiver's bit error rate performance and noise immunity under adverse channel conditions. Simultaneously, it provides a technical means to effectively identify the location of erroneous symbols and perform error correction. This addresses the problems in existing adaptive equalization and carrier phase recovery technologies, where using the nearest constellation point as a single hard decision criterion leads to easy misjudgment near the decision boundary and a lack of error correction mechanisms.
[0010] To achieve the above objectives, this application proposes an adaptive equalization and phase tracking method based on multi-branch parallel decision, comprising: Preprocess the received signal; Generate at least two candidate constellation points for the current symbol, wherein the at least two candidate constellation points include a nearest constellation point and at least one second nearest constellation point; Multiple processing branches are generated based on each candidate constellation point, wherein the state of each processing branch inherits the equalizer weight coefficient and phase estimate of the corresponding processing branch that survived the previous symbol processing; for the first symbol, the equalizer weight coefficient and phase estimate of the processing branch are preset initial values by the system. For each processing branch, joint equalization and phase update are performed independently to obtain the compensated symbol of the processing branch, as well as the updated equalizer weight coefficients and phase estimates. The cumulative error is calculated based on the error of each processing branch at the current symbol, and is used as the path metric for the processing branch. Based on the path metric, a surviving processing branch is selected from the plurality of processing branches; When the number of processed symbols meets the preset window triggering condition, window reprocessing is performed. The processing branch with the smallest surviving path metric is backtracked as the final decision path for all symbols in the window, and the equalization and phase compensation results of each symbol on the final decision path are output.
[0011] As a further solution, the independent execution of joint equalization and phase update for each processing branch includes: The received signal of the current symbol is subjected to equalization filtering using the equalizer weighting coefficients of the processing branch to obtain an equalized output; Phase compensation is performed on the equalization output using the phase estimate from the processing branch to obtain the compensated symbol; Based on the candidate constellation points corresponding to the processing branch and the compensated symbol calculation error; The equalizer weight coefficients of the error update processing branch are used as follows; The phase error is calculated based on the compensated symbol and the corresponding candidate constellation point, and the phase estimate of the processing branch is updated based on the phase error.
[0012] As a further solution, the calculation of cumulative error based on the error of each processing branch at the current symbol, as a path metric for the processing branch, includes: The sum of the squares of the errors calculated at each symbol from the starting symbol to the current symbol in the processing branch is used as the path metric of the processing branch.
[0013] As a further solution, the step of using the sum of squares of the errors of the processing branch from the starting symbol to the current symbol as the path metric of the processing branch further includes: Under nonlinear channel conditions, the sum of squares is weighted or modified by utilizing the correlation between the decisions of preceding and following symbols to generate an enhanced path metric.
[0014] As a further solution, the selection of the surviving processing branch from the plurality of processing branches based on the path metric includes: The processing branch generated based on the nearest constellation point is fully retained; The path metrics of other processing branches generated based on the second nearest constellation point are compared with the path metrics of the processing branch generated based on the nearest constellation point. The K processing branches with the smallest path metrics are retained as surviving processing branches, where K is a preset positive integer.
[0015] As a further solution, the selection of the survivor processing branch further includes: During the selection of surviving processing branches, if there are two processing branches with identical candidate constellation point sequences from the starting symbol to the current symbol, only the processing branch with the smaller path metric is retained, while the other processing branch is deleted.
[0016] As a further solution, the processing branch state of the final decision path includes equalizer weight coefficients and phase estimates, which are used as the initial state for generating processing branches in the next window processing.
[0017] As a further solution, the preset window triggering conditions include a preset fixed window length or a variable window parameter that is dynamically adjusted according to the current error level; wherein, when the accumulated error increases, the window length is increased; when the accumulated error remains below a preset threshold, the window length is decreased.
[0018] As a further solution, it also includes: The first processing and the second processing are run in parallel. The first processing only uses the nearest constellation point as a candidate for single processing branch decision, while the second processing adopts the multi-processing branch parallel decision. By comparing the outputs of the first processing path and the second processing path, the location of the erroneous symbol in the demodulated signal is identified and error correction is performed.
[0019] On the other hand, the present invention also provides an adaptive equalization and phase tracking system based on multi-branch parallel decision, for implementing the adaptive equalization and phase tracking method based on multi-branch parallel decision as described in any of the preceding claims, comprising: The input preprocessing module is used to preprocess the received signal; The candidate decision processing module is used to generate at least two candidate constellation points for the current symbol, wherein the at least two candidate constellation points include a nearest constellation point and at least one second nearest constellation point; Multiple processing branches are generated based on each candidate constellation point, wherein the state of each processing branch inherits the equalizer weight coefficient and phase estimate of the corresponding processing branch that survived the previous symbol processing; for the first symbol, the equalizer weight coefficient and phase estimate of the processing branch are preset initial values by the system. The joint equalization and phase tracking module is used to independently perform joint equalization and phase update for each processing branch to obtain the compensated symbol of the processing branch and the updated equalizer weight coefficients and phase estimates. The path metric calculation module is used to calculate the cumulative error based on the error of each processing branch at the current symbol, and use it as the path metric of the processing branch. The surviving branch selection module is used to select a surviving processing branch from the plurality of processing branches based on the path metric. The window reprocessing module is used to perform window reprocessing when the number of processed symbols meets the preset window triggering conditions. It backtracks to the processing branch with the smallest surviving path metric as the final decision path for all symbols in the window. The output decision module is used to output the equalization and phase compensation results of each symbol on the final decision path.
[0020] Compared with related technologies, the adaptive equalization and phase tracking based on multi-branch parallel decision provided by this invention has the following advantages: 1. This invention generates multiple candidate decisions for each symbol, including the nearest constellation point and at least one second-nearest constellation point, and constructs multiple processing branches for parallel exploration. When one processing branch makes an incorrect decision due to noise interference, other processing branches may still maintain the correct path. Combined with a path metric mechanism based on accumulated error and a surviving processing branch selection strategy, the system can dynamically select the optimal path from multiple candidate paths, significantly reducing the probability of misjudgment near the decision boundary and improving the receiver's bit error rate performance under low signal-to-noise ratio conditions.
[0021] 2. This invention adopts a joint equalization and phase tracking mechanism. Each processing branch coordinates and updates the equalizer weight coefficients and phase estimates based on a common error signal, avoiding the problem of repeated transmission and mutual amplification of errors between the two modules in the separate architecture, thereby obtaining more stable tracking and convergence performance under harsh channel conditions.
[0022] 3. This invention introduces a window reprocessing mechanism. After processing a certain number of symbols, it backtracks to the processing branch with the smallest path metric among the currently surviving processing branches as the final decision path for all symbols in the window. The decisions in the window are treated as a whole for global optimization, which can effectively correct suboptimal decisions that may occur in the early symbol-by-symbol selection and further improve the accuracy of the decision.
[0023] 4. This invention employs a dynamic window length adjustment mechanism, enabling the system to adaptively adjust computational complexity based on the current error level. When accumulated error increases or channel conditions deteriorate, the window length is automatically increased to enhance error correction capabilities; conversely, when the system is stable and channel conditions are favorable, the window length is automatically decreased to reduce computational load. Simultaneously, a branch merging optimization mechanism merges redundant processing branches with identical candidate constellation point sequences, effectively controlling the growth of the total number of processing branches. The synergistic operation of these two mechanisms allows this invention to achieve a dynamic balance between processing performance and computational complexity while ensuring sufficient parallel exploration of multiple paths, demonstrating good engineering feasibility.
[0024] 5. This invention can achieve parallel processing of two paths—the first path uses only the nearest constellation point as a candidate for single-branch decision-making, while the second path employs multi-branch parallel decision-making—by comparing the output results of the two paths, the location of the erroneous symbol in the demodulated signal can be effectively identified, and the decision result of the second path can replace the first path for error correction. Simultaneously, by statistically analyzing the number and distribution characteristics of inconsistencies between the two paths, reference information can be provided for system performance monitoring and channel quality assessment, offering a "built-in self-checking" capability that traditional solutions lack. Attached Figure Description
[0025] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.
[0026] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.
[0027] Figure 1 A schematic diagram of the process of an adaptive equalization and phase tracking method based on multi-branch parallel decision provided by the present invention; Figure 2 The structural block diagram of multi-branch parallel processing provided by the present invention; Figure 3 This is a schematic diagram of the candidate branch encoding provided by the present invention; The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation
[0028] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0029] Please see Figure 1 This embodiment provides Figure 1 This invention illustrates an adaptive equalization and phase tracking method based on multi-branch parallel decision-making, comprising: Preprocess the received signal; Generate at least two candidate constellation points for the current symbol, wherein the at least two candidate constellation points include a nearest constellation point and at least one second nearest constellation point; Multiple processing branches are generated based on each candidate constellation point, wherein the state of each processing branch inherits the equalizer weight coefficient and phase estimate of the corresponding processing branch that survived the previous symbol processing; for the first symbol, the equalizer weight coefficient and phase estimate of the processing branch are preset initial values by the system. For each processing branch, joint equalization and phase update are performed independently to obtain the compensated symbol of the processing branch, as well as the updated equalizer weight coefficients and phase estimates. The cumulative error is calculated based on the error of each processing branch at the current symbol, and is used as the path metric for the processing branch. Based on the path metric, a surviving processing branch is selected from the plurality of processing branches; When the number of processed symbols meets the preset window triggering condition, window reprocessing is performed, and the processing branch with the smallest surviving path metric is backtracked as the final decision path for all symbols in the window. Output the equalization and phase compensation results of each symbol on the final decision path.
[0030] The present invention will now be described in detail with reference to various embodiments.
[0031] Example 1 This embodiment details the basic implementation process of the adaptive equalization and phase tracking processing method based on multi-branch parallel decision, which includes the following steps: Step S1: Receive the input signal and perform preprocessing.
[0032] Frequency offset estimation and compensation. In coherent optical and wireless communication systems, the received signal is often affected by frequency offset. If it is processed directly without compensation, the frequency offset will cause the constellation diagram to rotate continuously, making the equalizer and phase tracker unable to function properly. After coarse frequency offset compensation, the residual frequency offset is small enough that the subsequent phase tracking module only needs to deal with the residual phase noise and the phase drift introduced by the laser linewidth.
[0033] Receive complex baseband signals transmitted through the channel First, the received signal is synchronized in terms of timing and frames, then coarse frequency offset estimation and compensation are performed. The preprocessed signal is denoted as... , where n represents the current symbol number being processed, and n is an integer greater than or equal to 1.
[0034] Step S2: Generate multiple candidate constellation points for the current symbol.
[0035] Before describing the specific steps in this section, we will first analyze the existing technical methods and their problems.
[0036] In conventional adaptive equalization processing, the equalizer output is denoted as... Its expression is: Where w is the equalizer weight coefficient vector. The input signal vector, This indicates the conjugate transpose.
[0037] When calculating the cost function for decision-making, classical methods typically employ one of the following cost functions. For the Least Mean Squares (LMS) or Recursive Least Squares (RLS) algorithm, the error cost function is: , where d is the expected decision symbol for the current symbol. For the Constant Modulus Algorithm (CMA), the cost function is: , where R is the magnitude corresponding to the constellation point; for QPSK modulation, R is a constant value; for QAM modulation, the magnitude is different for different constellation points.
[0038] In conventional methods, when calculating the aforementioned cost function, d or R is typically chosen as the constellation point closest to the output signal after equalization and phase compensation. This method, using the nearest constellation point as the sole hard decision criterion, works effectively under good channel conditions and high signal-to-noise ratios. However, when the received signal is subject to strong noise interference, if the symbol after equalization and phase compensation happens to fall near the decision boundary between two constellation points, using the nearest constellation point as the decision criterion can easily lead to misjudgment. More seriously, once a misjudgment occurs, the incorrect decision symbol will be fed back to the update process of the equalizer and phase tracker through the cost function, causing the equalizer weight coefficients and phase estimates to be updated incorrectly simultaneously, forming a cascading propagation of local errors, further deteriorating the decision accuracy of subsequent symbols. Symbol error rate analysis of actual coherent communication systems shows that in most cases of misjudgments occurring near the decision boundary, the correct decision result is not the nearest constellation point, but the second nearest constellation point.
[0039] To address the aforementioned issues, this step abandons the traditional approach of using a single nearest constellation point as the decision criterion, instead generating multiple candidate constellation points for each symbol. Specifically, for the received signal of the preprocessed nth symbol, preliminary equalization and phase compensation are performed to obtain the output signal. The Euclidean distance between this output signal and all constellation points on the constellation diagram is calculated. These distances are sorted in ascending order, and the point with the smallest distance is selected as the nearest constellation point, followed by the next smallest, forming the candidate decision set for the current symbol, denoted as . .
[0040] This embodiment uses the retention of two candidate constellation points, namely the nearest and the second nearest constellation point, as an example for illustration. For QPSK modulation, the nearest and second nearest constellation points correspond to the constellation symbols with the smallest and second smallest Euclidean distances on the constellation diagram, respectively; for higher-order QAM modulation, the second nearest constellation points can be selected in order of Euclidean distance. By retaining multiple candidate decisions for each symbol, this method preserves the possibility of selecting the correct constellation point near the decision boundary, thereby reducing the risk of misjudgment caused by a single hard decision from the source.
[0041] Step S3: Generate multiple processing branches based on each candidate constellation point.
[0042] Based on the two candidate constellation points retained in step S2, two new processing branches are generated. The state of each processing branch inherits the equalizer weight coefficients and phase estimates of the corresponding processing branch that survived the previous symbol processing, as follows: Here, the superscript b represents the index of the parent processing branch (i.e., the number of the processing branch that survived the previous symbol processing). Indicates the candidate decision type used by the current symbol. Let N be the equalizer weight coefficient vector of the b-th parent processing branch at the end of the previous symbol processing, with a length of N (i.e., the number of equalizer taps), denoted as... . This is the phase estimate of the b-th parent processing branch at the end of the previous symbol processing.
[0043] For the first symbol ( If no surviving processing branch exists from the previous symbol processing, the equalizer weight coefficients and phase estimates of that processing branch are set to the system's preset initial values. The initial values of the equalizer weight coefficients can be set by setting the center tap to 1 and the remaining taps to 0. The initial value of the phase estimate can be set to... .
[0044] Figure 2 A block diagram of the multi-branch parallel processing structure is shown. Figure 3 The diagram illustrates the candidate branch encoding. Taking the nearest and second nearest candidate target symbols as examples, parallel processing of processing branches with a binary tree structure is generated within four symbol periods.
[0045] It's important to note that this step utilizes a state inheritance mechanism, allowing each processing branch to continue the tracing history of its parent branch, rather than starting from scratch. The advantage of this design is that each processing branch, while inheriting existing tracing results, only explores different candidate decisions for the current symbol, maintaining tracing continuity while enabling parallel exploration along multiple paths. Compared to simply making multiple independent decisions, the state inheritance mechanism focuses the differences between processing branches on the symbol positions where ambiguity truly exists, avoiding unnecessary redundant calculations and providing a complete historical decision history for subsequent window backtracking.
[0046] Step S4: Each processing branch independently performs joint equalization and phase update.
[0047] For each newly generated processing branch (denoted as processing branch) ), and execute the following sub-steps independently.
[0048] In the existing serial independent processing architecture, the equalizer and phase recovery module operate independently. When the equalizer is updated, it is assumed that the phase has been correctly compensated, and when the phase is updated, it is assumed that the equalizer has converged. This separate update method ignores the coupling relationship between equalization error and phase error, which can easily lead to repeated propagation and mutual amplification of errors between the two modules under low signal-to-noise ratio conditions.
[0049] To address the aforementioned issues, this step employs a joint equalization and phase tracking mechanism, treating equalization and phase compensation as a unified process and coordinating their updates based on a shared error signal. The specific sub-steps are as follows: Sub-step S4.1: Equalizer output calculation.
[0050] The equalizer weighting coefficients of this processing branch are used to perform equalization filtering on the received signal of the current symbol, resulting in an equalized output: in, The input signal vector contains the received sample values of the current symbol and its neighboring symbols. This indicates the conjugate transpose.
[0051] Sub-step S4.2: Phase compensation.
[0052] The phase estimate from this processing branch is used to perform phase compensation on the equalization output, resulting in the compensated symbol: Sub-step S4.3: Error calculation.
[0053] Based on the candidate constellation points corresponding to this processing branch and the compensated sign calculation error: in, The candidate constellation point corresponding to the current processing branch (i.e. or This error signal will be used simultaneously for equalizer weight updates and phase tracking updates, achieving joint optimization of equalization and phase recovery.
[0054] Sub-step S4.4: Equalizer weight update.
[0055] Update the equalizer weight coefficients of this processing branch based on the error: Where μ is the weight refresh step size factor, which is usually a positive real number less than 1; This indicates taking the conjugate.
[0056] Sub-step S4.5: Phase tracking update.
[0057] First, calculate the phase error: Then update the phase estimate: Where α is the phase refresh weighting factor, satisfying ; This indicates the argument operation.
[0058] The sub-steps S4.1 to S4.5 described above constitute the joint equalization and phase update process for a single processing branch. The update processes of each processing branch are independent of each other and can be executed in parallel. Through joint equalization and phase tracking, this method enables the equalizer and phase tracker to share the same error signal for coordinated updates, avoiding the problem of repeated error propagation between the two modules in a separate architecture, thereby achieving more stable tracking performance under low signal-to-noise ratio conditions.
[0059] Step S5: Calculate the path metric for each processing branch.
[0060] For each processing branch, its path metric is calculated. In existing multi-branch processing methods, effectively evaluating and comparing the merits of different branches is a key technical issue. If branch selection is based solely on the instantaneous error of the current symbol, it is easy to erroneously discard the correct branch due to a single noise fluctuation.
[0061] To address this issue, this step employs cumulative error as a path metric to evaluate the quality of each processing branch from the overall perspective of the historical trajectory. The path metric is defined as the sum of squares of the errors calculated at each symbol from the starting symbol to the current symbol for that processing branch: in, This represents the modulo value. The smaller the path metric, the better the decision sequence of the processing branch matches the actual received signal, indicating that the processing branch is superior.
[0062] The advantages of using cumulative error as a path metric are twofold: First, the accumulation mechanism smooths out instantaneous noise, avoiding the randomness of branch selection based on single errors. Second, by accumulating the errors across the entire historical sequence, it effectively reflects the tracking quality differences of each processing branch over long periods, providing a more reliable evaluation criterion for subsequent surviving processing branch selection. Furthermore, when a processing branch makes an erroneous decision early on, its subsequent errors tend to remain large; cumulative error can amplify this error effect, making erroneous branches easier to identify and eliminate.
[0063] Step S6: Selecting a survivor processing branch.
[0064] As the number of symbols processed increases, the total number of branches will grow exponentially after multiple rounds of candidate decision derivation. To control computational complexity, it is necessary to filter the processing branches, retaining only the most promising ones for further processing.
[0065] The specific operation for selecting the surviving processing branch is as follows: First, the processing branch generated based on the nearest constellation point is completely retained, that is, regardless of its path metric size, this processing branch is unconditionally retained as a surviving processing branch, and is used as a basic reference. This design is based on the following consideration: at most symbol positions, the nearest constellation point is the correct decision result, and completely retaining this branch can ensure that the correct path has at least one branch that continues, avoiding the premature elimination of the correct path due to accidental noise fluctuations.
[0066] Then, the path metrics of other processing branches generated based on the second nearest constellation point are compared one by one with the path metrics of the processing branch generated based on the nearest constellation point. The K processing branches with the smallest path metrics are selected as surviving processing branches, where K is a preset positive integer. In a typical implementation, K can take values from 2 to 4 to achieve a balance between processing performance and computational complexity.
[0067] Figure 3 The process of selecting surviving processing branches is illustrated using a branch coding diagram spanning four symbol cycles. In each symbol cycle, by comparing the path metrics of each processing branch, the K optimal processing branches are retained for processing the next symbol. Figure 2 The diagram also shows the functional location of the surviving branch selection module in the entire processing flow.
[0068] By employing the aforementioned survival processing branch selection mechanism based on path metrics, this method maintains the advantages of multi-path parallel exploration while keeping the number of branches within an acceptable range, thus achieving an effective balance between performance and complexity.
[0069] Step S7: Window reprocessing and output of optimal results.
[0070] In the symbol-by-symbol survival branch selection process of step S6, the branch selection decision is made on an instantaneous basis, that is, at each symbol, the K branches with the best path metric at that time are retained. However, in some cases, a branch that seems optimal at the current symbol may be proven not to be the globally optimal choice at subsequent symbols. Without a backtracking mechanism, the correct decision paths that may have been contained in the branches that were discarded earlier cannot be recovered.
[0071] To address the aforementioned issues, this step introduces a window reprocessing mechanism. Window reprocessing is executed when the number of processed symbols meets a preset window trigger condition. The processing branch with the smallest surviving path metric is backtracked and used as the final decision path for all symbols within that window. The equalization and phase compensation results for each symbol on this final decision path are then output.
[0072] Specifically, in the above processing, each processing branch records its generation source information, including which parent processing branch it inherits from and the type of candidate constellation point selected (most recent or second most recent). During window reprocessing, the processing branch with the smallest path metric among all surviving processing branches is first determined. Then, the generation source chain of this processing branch is traced backward: from the current symbol back to the starting symbol of the window, the parent processing branch source and the corresponding candidate constellation point selection at each symbol are determined sequentially. This reconstructs the decision result (i.e., the selected candidate constellation point) for each symbol within the window, and outputs the symbol value of each symbol after equalization and phase compensation. .
[0073] Meanwhile, the processing branch state of the final decision path (including equalizer weight coefficients and phase estimates) is used as the initial state for generating processing branches in the next window processing, thereby achieving a smooth transition between windows.
[0074] The core value of the window reprocessing mechanism lies in its global optimization of decisions made within a window period, rather than local optimization on a symbol-by-symbol basis. By backtracking to select the globally optimal path at the end of the window, it can correct suboptimal decisions that may occur in the early symbol-by-symbol selection, significantly improving the accuracy of the decisions.
[0075] Example 2 This embodiment, based on Embodiment 1, further introduces an optimization mechanism for merging processing branches. In the processing of Embodiment 1, although the surviving processing branch selection mechanism has controlled the number of branches to within K, in some cases, different processing branches may converge to the same decision path because they have selected the same candidate constellation point sequence. In this case, continuing to maintain multiple branches with completely identical or highly similar states will cause unnecessary waste of computing resources.
[0076] To address this issue, this embodiment introduces a branch merging mechanism during the surviving processing branch selection process in step S6: when two processing branches have identical candidate constellation point sequences from the starting symbol to the current symbol, it indicates that these two processing branches have converged to the same decision path. In this case, only the processing branch with the smaller path metric is retained, while the other processing branch is deleted.
[0077] For example, within a certain symbol period, processing branch A (a branch of the nearest constellation point candidate path) and processing branch B (a branch of the second nearest constellation point candidate path) may have different sources of generation for the current symbol, but their historical decision sequences from the first symbol to the current symbol are completely consistent. According to the optimization strategy of this embodiment, these two processing branches are merged, retaining the processing branch with the smaller path metric to continue participating in subsequent processing, while terminating the update of the other processing branch.
[0078] The branch merging mechanism and the surviving branch selection mechanism work together: the surviving branch selection retains the optimal K branches from the perspective of path metric, while branch merging eliminates redundant branches from the perspective of decision sequence consistency. Working together, they effectively control the growth of the total number of processing branches, reducing computational complexity while ensuring sufficient parallel exploration of multiple paths.
[0079] Example 3 This embodiment, based on Embodiment 1 or Embodiment 2, further introduces a dynamic adjustment mechanism for the window triggering conditions. In the fixed window length processing method, the window length needs to be preset during system design and cannot be adaptively adjusted according to changes in actual channel conditions. If the window length is set too short, the error correction capability is insufficient; if the window length is set too long, the computational complexity and processing delay are too large.
[0080] To resolve this contradiction, the preset window triggering conditions in this embodiment include two scenarios: one is a preset fixed window length L; the other is a variable window parameter that is dynamically adjusted according to the current error level.
[0081] When employing a dynamic window adjustment strategy, the system evaluates the current error level at the end of each window. Specifically, when the accumulated error increases, it indicates that the current channel conditions have deteriorated and the probability of decision errors has risen. In this case, the window length L is increased to allow for path metric comparison and selection of surviving processing branches over a longer time span, thereby enhancing error correction capabilities. When the accumulated error remains below a preset threshold, it indicates that the channel conditions are good and the system is in a stable operating state. In this case, the window length L is decreased to reduce computational load and processing latency.
[0082] The preset threshold can be determined in advance through simulation or experiment based on the system's operating signal-to-noise ratio range and target bit error rate requirements. For example, it can be set to 1.5 times the average cumulative error value when the system is operating stably under the target signal-to-noise ratio.
[0083] The dynamic window adjustment mechanism enables the system to automatically enhance error correction capabilities under poor channel conditions and automatically reduce computational overhead under good channel conditions, achieving a dynamic balance between receiver performance and power consumption. This is of great significance in practical deployments. For example, in wireless mobile communication systems, channel quality fluctuates with terminal movement speed and environmental changes; the dynamic window adjustment mechanism allows the receiver to adaptively match the current channel conditions.
[0084] Example 4 Based on Example 1, this embodiment provides an enhanced path metric calculation method for application scenarios under nonlinear channel conditions.
[0085] In the basic path metric calculation of Example 1, the path metric is defined as the sum of squares of the errors at each symbol. This method can effectively evaluate the performance of each processing branch under linear Gaussian channel conditions. However, under nonlinear channel conditions, due to the existence of channel memory effects, there is a correlation between the decisions of consecutive symbols. For example, in the nonlinear effect scenario of optical fiber communication, the waveform distortion of the current symbol is affected by the decision results of the previous few symbols. In this case, relying solely on the sum of squares of the independent errors of each symbol for path metric calculation cannot fully utilize the correlation information between symbols, which may lead to inaccurate path metric evaluation.
[0086] To address this issue, this embodiment uses the correlation between preceding and following symbol decisions to weight or modify the sum of squares when calculating the path metric for each processing branch, generating a strengthened path metric. One specific implementation involves introducing a weighting factor into the path metric calculation, which penalizes processing branches with inconsistent decisions on adjacent symbols. in, The decision transition penalty function takes a positive value when a candidate constellation point of an adjacent symbol jumps and the jump does not conform to the expected channel memory characteristics, otherwise it is zero; β is a weighting coefficient used to control the weight of the penalty term.
[0087] By introducing a decision transition penalty term, the path metric is enhanced to reflect not only the instantaneous error of each processing branch but also the rationality of its decision sequence under the channel memory characteristics. This makes the evaluation of the path metric more accurate and the selection of surviving processing branches more reliable under nonlinear channel conditions.
[0088] Example 5 Based on any of the above embodiments, this embodiment further provides an implementation method that can identify the location of erroneous symbols and perform error correction.
[0089] In the deployment and maintenance of practical communication systems, obtaining only the final demodulation result is often insufficient. Operators and system maintenance personnel typically also need to understand the transmission quality and identify the specific location of problematic symbols in order to perform targeted system optimization and troubleshooting. However, existing receiver schemes based on a single decision path lack comparable reference paths, making it difficult to provide such error location information.
[0090] To address this issue, this embodiment runs two processing paths in parallel within the system. The first processing path uses only the nearest constellation point as a candidate for single-branch decision-making, employing the traditional hard-decision method for equalization and phase recovery, without multi-branch parallel processing. The decision result of this path represents the output behavior of a conventional receiver. The second processing path employs the multi-branch parallel decision-making method described in any of Embodiments 1 to 4. This path achieves higher decision accuracy near the decision boundary through multiple candidate decisions and multi-branch parallel exploration.
[0091] By comparing the outputs of the first and second processing paths, the location of the erroneous symbol in the demodulated signal can be identified. Specifically, when the two processing paths make inconsistent decisions at the same symbol location, that symbol location is marked as a suspected error location. This discrepancy arises because the first processing path misjudged near the decision boundary, while the second processing path, through multi-branch parallel exploration and window backtracking, makes the correct decision choice with a high probability.
[0092] After identifying the symbol position where the two decision paths are inconsistent, the second path processing has better error correction performance through multi-branch parallel decision and window backtracking. Therefore, the decision result of the second path processing is used instead of the decision result of the first path processing as the final output of the symbol position, thereby achieving error correction.
[0093] Meanwhile, by statistically analyzing the number and distribution characteristics of locations where decisions differ between the two processing paths, reference information can be provided for system performance monitoring and channel quality assessment. For example, when the proportion of inconsistent symbols exceeds a preset threshold, a system alarm can be triggered or a request to adjust transmission parameters can be made. This mechanism provides communication systems with a "built-in self-checking" capability that traditional solutions do not possess.
[0094] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.
Claims
1. An adaptive equalization and phase tracking method based on multi-branch parallel decision, characterized in that, include: Preprocess the received signal; Generate at least two candidate constellation points for the current symbol, wherein the at least two candidate constellation points include a nearest constellation point and at least one second nearest constellation point; Multiple processing branches are generated based on each candidate constellation point, wherein the state of each processing branch inherits the equalizer weight coefficient and phase estimate of the corresponding processing branch that survived the previous symbol processing; for the first symbol, the equalizer weight coefficient and phase estimate of the processing branch are preset initial values by the system. For each processing branch, joint equalization and phase update are performed independently to obtain the compensated symbol of the processing branch, as well as the updated equalizer weight coefficients and phase estimates. The cumulative error is calculated based on the error of each processing branch at the current symbol, and is used as the path metric for the processing branch. Based on the path metric, a surviving processing branch is selected from the plurality of processing branches; When the number of processed symbols meets the preset window triggering condition, window reprocessing is performed. The processing branch with the smallest surviving path metric is backtracked as the final decision path for all symbols in the window, and the equalization and phase compensation results of each symbol on the final decision path are output.
2. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 1, characterized in that, The independent execution of joint equalization and phase update for each processing branch includes: The received signal of the current symbol is subjected to equalization filtering using the equalizer weighting coefficients of the processing branch to obtain an equalized output; Phase compensation is performed on the equalization output using the phase estimate from the processing branch to obtain the compensated symbol; Based on the candidate constellation points corresponding to the processing branch and the compensated symbol calculation error; The equalizer weight coefficients of the error update processing branch are used as follows; The phase error is calculated based on the compensated symbol and the corresponding candidate constellation point, and the phase estimate of the processing branch is updated based on the phase error.
3. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 2, characterized in that, The step of calculating the cumulative error based on the error of each processing branch at the current symbol, and using it as the path metric for the processing branch, includes: The sum of the squares of the errors calculated at each symbol from the starting symbol to the current symbol in the processing branch is used as the path metric of the processing branch.
4. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 3, characterized in that, The step of using the sum of squares of the errors of the processing branch from the starting symbol to the current symbol as the path metric of the processing branch further includes: Under nonlinear channel conditions, the sum of squares is weighted or modified by utilizing the correlation between the decisions of preceding and following symbols to generate an enhanced path metric.
5. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 1, characterized in that, The step of selecting a surviving processing branch from the plurality of processing branches based on the path metric includes: The processing branch generated based on the nearest constellation point is fully retained; The path metrics of other processing branches generated based on the second nearest constellation point are compared with the path metrics of the processing branch generated based on the nearest constellation point. The K processing branches with the smallest path metrics are retained as surviving processing branches, where K is a preset positive integer.
6. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 5, characterized in that, The selection of the survivor processing branch further includes: During the selection of surviving processing branches, if there are two processing branches with identical candidate constellation point sequences from the starting symbol to the current symbol, only the processing branch with the smaller path metric is retained, while the other processing branch is deleted.
7. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 1, characterized in that, The processing branch state of the final decision path includes the equalizer weight coefficient and the phase estimate, which are used as the initial state for generating the processing branch in the next window processing.
8. The adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to claim 1, characterized in that, The preset window triggering conditions include a preset fixed window length or a variable window parameter that is dynamically adjusted according to the current error level; wherein, when the cumulative error increases, the window length is increased; when the cumulative error continues to be lower than a preset threshold, the window length is decreased.
9. An adaptive equalization and phase tracking method based on multi-branch parallel decision-making according to any one of claims 1 to 8, characterized in that, Also includes: The first processing and the second processing are run in parallel. The first processing only uses the nearest constellation point as a candidate for single processing branch decision, while the second processing adopts the multi-processing branch parallel decision. By comparing the outputs of the first processing path and the second processing path, the location of the erroneous symbol in the demodulated signal is identified and error correction is performed.
10. An adaptive equalization and phase tracking system based on multi-branch parallel decision, used to implement the adaptive equalization and phase tracking method based on multi-branch parallel decision as described in any one of claims 1 to 9, characterized in that, include: The input preprocessing module is used to preprocess the received signal; The candidate decision processing module is used to generate at least two candidate constellation points for the current symbol, wherein the at least two candidate constellation points include a nearest constellation point and at least one second nearest constellation point; Multiple processing branches are generated based on each candidate constellation point, wherein the state of each processing branch inherits the equalizer weight coefficient and phase estimate of the corresponding processing branch that survived the previous symbol processing; for the first symbol, the equalizer weight coefficient and phase estimate of the processing branch are preset initial values by the system. The joint equalization and phase tracking module is used to independently perform joint equalization and phase update for each processing branch to obtain the compensated symbol of the processing branch and the updated equalizer weight coefficients and phase estimates. The path metric calculation module is used to calculate the cumulative error based on the error of each processing branch at the current symbol, and use it as the path metric of the processing branch. The surviving branch selection module is used to select a surviving processing branch from the plurality of processing branches based on the path metric. The window reprocessing module is used to perform window reprocessing when the number of processed symbols meets the preset window triggering conditions. It backtracks to the processing branch with the smallest surviving path metric as the final decision path for all symbols in the window. The output decision module is used to output the equalization and phase compensation results of each symbol on the final decision path.