Method and device for optimizing adaptive equalizer for coherent optical communication system

Abstract

The invention provides a coherent optical communication system-oriented adaptive equalizer optimization method and device, and the method comprises the steps: obtaining a current input signal and a current output signal of a current signal equalizer when a tap coefficient of the signal equalizer is updated, determining a data loss function of each tap point based on an error term corresponding to a current output signal, the current input signal and the current output signal; determining an update weight of each tap point according to the current weight of each tap point, the data loss function and a weight sparsification function so as to optimize the adaptive equalizer; wherein the weight sparse function of each tap point is determined through a set L1 regularization parameter and the current weight of each tap point. According to the invention, the calculation complexity of the adaptive equalizer can be reduced while the performance of the adaptive equalizer is maintained.

Description

Technical Field

[0001] The present invention relates to the field of optical communication technology, and in particular to an optimization method and device for an adaptive equalizer of a coherent optical communication system. Background Art

[0002] With the continuous explosive growth of global network traffic, traditional intensity modulation direct detection (IM / DD) systems require wider electro-optical bandwidth, higher complexity and more advanced digital signal processing (DSP) algorithms to support higher transmission rates, which leads to a significant increase in system cost and power consumption. In contrast, digital coherent optical transmission systems are highly competitive in the next generation of high-speed and large-capacity optical communication solutions due to their excellent spectral efficiency and receiver sensitivity. However, coherent optical transmission systems still face the challenges of high computational complexity and power consumption, especially the power consumption of the DSP at the receiving end, which can account for more than 50% of the total power consumption of the optical module, severely limiting its actual deployment in industrial applications. Specifically, the complexity of the DSP of coherent optical communication is mainly concentrated in the adaptive equalizer (AEQ), such as the traditional 4x4 MIMO AEQ and 2x2 MIMO AEQ, but the traditional AEQ calculation is still relatively complex. Therefore, in order to reduce the computational burden of the DSP at the receiving end, low-complexity AEQ has become the focus of extensive research.

[0003] Existing research on reducing the computational complexity of AEQ mainly focuses on decomposing the traditional multiple-input multiple-output (MIMO) equalizer architecture into two stages: polarization demultiplexing and signal equalization. In the two-stage AEQ design, the M-tap butterfly filter (which can be composed of four interrelated finite impulse response filters) can be used for polarization demultiplexing, and the N-tap finite impulse response (FIR) filter can perform adaptive equalization on optical signals of each polarization state respectively (where M < N). Existing two-stage AEQ architectures can include MN CV AEQ, M-NRV (1x1) AEQ, and MN RV (2x2) AEQ. Compared with traditional AEQ, the number of N-tap FIR filters in this type of design will be halved, which can greatly reduce the computational complexity without significantly reducing system performance, so as to achieve low-complexity adaptive equalization.

[0004] However, research has found that even with the simplified AEQ architecture of the above two stages, there is still a certain degree of coefficient redundancy in the tap design of the FIR filter. There is still a lot of room for optimization in the research of further reducing redundant tap coefficients. Therefore, how to further reduce the computational complexity of AEQ while maintaining AEQ performance is an urgent problem to be solved. Summary of the invention

[0005] In view of this, an embodiment of the present invention provides an optimization method and device for an adaptive equalizer for a coherent optical communication system, which can significantly reduce the computational complexity of the AEQ while maintaining the performance of the AEQ.

[0006] One aspect of the present invention provides an optimization method for an adaptive equalizer (AEQ) for a coherent optical communication system, wherein the adaptive equalizer includes a polarization demultiplexer and a signal equalizer receiving an output signal of the polarization demultiplexer, and the method includes the following steps:

[0007] Obtaining a current input signal and a current output signal of the signal equalizer, and determining a data loss function of each tap point based on an error term corresponding to the current output signal, the current input signal, and the current output signal;

[0008] The new weight of each tap point is determined according to the current weight of each tap point, the data loss function and the weight sparsification function to optimize the adaptive equalizer; wherein the weight sparsification function of each tap point is determined based on the set L1 regularization parameter and the current weight of each tap point.

[0009] In some embodiments of the present invention, the weight sparsification function of each tap point is determined based on the set L1 regularization parameter and the current weight of each tap point, including:

[0010] The positive or negative value of the weight sparsification function of each tap point is determined by the positive or negative value of the current weight of each tap point, and the absolute value of the weight sparsification function of each tap point is determined by the absolute value of the product of the L1 regularization parameter and the convergence step size.

[0011] In some embodiments of the present invention, after determining the new weights of the tap points, the method further comprises: setting the new weights below the decision threshold to zero.

[0012] In some embodiments of the present invention, the set L1 regularization term parameter is determined by the following method:

[0013] In the training phase, the training input signal in the training set is input into the adaptive equalizer with the initial L1 regularization parameter, the error term of the signal equalizer in the adaptive equalizer is updated based on the current training output signal, and the tap coefficient of the signal equalizer in the adaptive equalizer is updated based on the current training input signal and the current training output signal, and the next training input signal is input into the updated adaptive equalizer; the iterative tap coefficient corresponding to the initial L1 regularization parameter is obtained by iteratively updating multiple training input signals, thereby obtaining the iterative adaptive equalizer corresponding to the initial L1 regularization parameter;

[0014] In the test phase, a test input signal in the test set is input into the iterative adaptive equalizer, and based on the test input signal and the corresponding test output signal, it is determined whether the iterative adaptive equalizer satisfies the set regularization condition;

[0015] If the set regularization condition is not met, the initial L1 regularization parameter is corrected and the training phase and the testing phase are repeated. If the set regularization condition is met, the initial L1 regularization parameter is used as the set L1 regularization parameter.

[0016] In some embodiments of the present invention, the regularization condition is set as a bit error rate less than a set bit error value and a sparsity degree greater than a set sparsity number; wherein the bit error rate is determined based on the output signal of the signal equalizer in the iterative adaptive equalizer during the test phase and the standard reference signal, and the sparsity degree is measured by the number of tap points in the iterative adaptive equalizer whose weights are greater than the set weight value.

[0017] In some embodiments of the present invention, the signal equalizer is constructed using a finite impulse response filter, and the architecture of the adaptive equalizer is MN CV AEQ, MN RV (1x1) AEQ, or MN RV (2x2) AEQ;

[0018] The data loss function of each tap point is determined based on the error term corresponding to the current output signal, the current input signal and the current output signal, including:

[0019] In the MN CV AEQ architecture, for optical signals in each polarization state, the product of the conjugate operation result of the current input signal, the current output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point.

[0020] In the MN RV (1x1) AEQ architecture, for optical signals of each polarization state, the real part of the current input signal, the real part of the current output signal, the convergence step size, and the product of the error term corresponding to the current output signal are calculated to obtain the data loss function of each tap point;

[0021] In the MN RV (2x2) AEQ architecture, for optical signals in each polarization state, if the current input signal and the current output signal are both the real part of the signal or the imaginary part of the signal, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point; if the current input signal and the current output signal are the real part of the signal and the imaginary part of the signal, respectively, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point.

[0022] In some embodiments of the present invention, the initial weights of the tap points in the signal equalizer are determined by:

[0023] The initial weight of the tap point located in the middle of the signal equalizer is set to 1, and the sliding window convolution method is used for adaptive iteration to determine the initial weight of each tap point in the signal equalizer.

[0024] In some embodiments of the present invention, the error term corresponding to the current output signal is obtained by calculating the difference between the square of the modulus of the current output signal and the square of the reference modulus value using a cascaded multi-mode algorithm.

[0025] In some embodiments of the present invention, the data loss function of each tap point in the signal equalizer is the same.

[0026] Another aspect of the present invention provides an optimization device for an adaptive equalizer of a coherent optical communication system, comprising a processor, a memory, and a computer program / instructions stored in the memory, wherein the processor is used to execute the computer program / instructions, and when the computer program / instructions are executed, the device implements the steps of the method described in any of the above embodiments.

[0027] The optimization method and device of the adaptive equalizer for coherent optical communication systems proposed in the present invention can suppress the redundant taps of the FIR filter in the signal equalizer based on the two-stage designed AEQ by introducing the L1 regularization parameter in the tap coefficient update process, thereby achieving low power consumption and high efficiency of the coherent optical communication system.

[0028] Additional advantages, purposes, and features of the present invention will be described in part in the following description, and will become apparent to those skilled in the art after studying the following, or may be learned from the practice of the present invention. The purposes and other advantages of the present invention may be achieved and obtained by the structures specifically indicated in the specification and the accompanying drawings.

[0029] Those skilled in the art will appreciate that the objectives and advantages that can be achieved with the present invention are not limited to the above specific description, and the above and other objectives that can be achieved by the present invention will be more clearly understood from the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of the present application, and do not constitute a limitation of the present invention. In the drawings:

[0031] Figure 1 It is a flowchart of an optimization method of AEQ for a coherent optical communication system in one embodiment of the present invention.

[0032] Figure 2 FIG. 1 is a schematic diagram of an AEQ architecture with a two-stage MN tap configuration according to an embodiment of the present invention.

[0033] Figure 3 It is a schematic diagram of updating the tap coefficients of the filter in AEQ according to an embodiment of the present invention.

[0034] Figure 4 Schematic diagram of a signal light back-to-back transmission system in one embodiment of the present invention.

[0035] Figure 5 Schematic diagram of the performance verification results of optimizing AEQ in one embodiment of the present invention. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0037] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, only structures and / or processing steps closely related to the solutions according to the present invention are shown in the accompanying drawings, while other details that are not closely related to the present invention are omitted.

[0038] It should be emphasized that the term “include / comprises” when used herein refers to the presence of features, elements, steps or components, but does not exclude the presence or addition of one or more other features, elements, steps or components.

[0039] Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. In the accompanying drawings, the same reference numerals represent the same or similar components, or the same or similar steps.

[0040] The present application can be applied to coherent optical communication systems, and the adaptive equalizer in the present application is an AEQ with a two-stage architecture, that is, the adaptive equalizer in the present application can be divided into a polarization demultiplexer and a signal equalizer, and the output signal of the polarization demultiplexer is input into the signal equalizer. Considering that the polarization demultiplexer in the AEQ is used to eliminate the interference and damage caused by factors such as birefringence effect and polarization mode dispersion in the optical communication system to each polarization state signal (including X polarization state and Y polarization state), the optimization pruning method of the adaptive equalizer proposed in the present application does not include the update process of the polarization demultiplexer, but only involves optimizing the signal equalizer in the AEQ.

[0041] Furthermore, in digital filters, especially FIR filters and recursive filters (such as Infinite Impulse Response, IIR, infinite impulse response filters), a tap point (tap) can represent a position for adjusting filter parameters (i.e., a specific position on the delay line of the filter), and the signal at this position can be sampled or weighted. Considering that most existing AEQs are constructed using FIR filters, the following description takes the AEQ constructed using FIR filters as an example. When constructing AEQ, other types of filters can also be used to construct AEQ, and the present invention is not limited to this.

[0042] The tap coefficient of the FIR filter refers to the specific weight value of each tap of the filter in the transfer function of the filter (that is, the tap coefficient can be regarded as a general term for the weights of each tap point in the filter). The input signal can be weighted and added through these tap points to generate an output signal. That is, the FIR filter contains many convolution operations consisting of multiplication and accumulation. The number of effective FIR tap points can indirectly indicate the number of calculations required.

[0043] Compared with the traditional AEQ, although the computational complexity of the AEQ designed by combining the polarization demultiplexer and the signal equalizer has been simplified, there is a certain redundancy in its tap coefficients, making it difficult to reduce the computational complexity of the AEQ. The present application introduces an L1 regularization parameter in the tap coefficient update process of the FIR filter in the signal equalizer to achieve optimized pruning of the adaptive equalizer to adaptively sparse the tap coefficients of the filter, and can be combined with a threshold value to further control the number of effective taps in the AEQ (i.e., reduce the number of non-important taps in the AEQ), thereby reducing the computational complexity while maintaining the performance of the AEQ.

[0044] Figure 1 This is a schematic diagram of the optimization process of AEQ for coherent optical communication systems proposed in this application. Figure 1 As shown, the method may include steps S110 to S120.

[0045] Step S110: obtaining a current input signal and a current output signal of the signal equalizer, and determining a data loss function of each tap point based on an error term corresponding to the current output signal, the current input signal and the current output signal.

[0046] In some embodiments of the present invention, the calculation methods of the data loss function of different two-stage AEQ architectures are different. For example, Figure 2 The architecture of the two-stage MN tap configuration adaptive equalizer can be MN CV AEQ, MN RV (1x1) AEQ or MN RV (2x2) AEQ. Figure 2 (a) MN CV AEQ means using an M-tap 4x4 MIMO filter to implement polarization demultiplexing, and two independent N-tap CV 2x2 FIR filters to achieve signal equalization; Figure 2 (b) MN RV(1x1)AEQ means using an M-tap 4x4 MIMO filter to implement polarization demultiplexing, and two independent N-tap RV 1x1 FIR filters to implement equalization; Figure 2 (c) MN RV(2x2)AEQ in FIG. 4 shows that polarization demultiplexing is realized by using M-tap 4x4 MIMO, and two independent N-tap RV 2x2 FIR filters are used for equalization. For optical signals in each polarization state, Figure 2 The calculation method of the data loss function of each AEQ architecture in can be as follows:

[0047] In the MN CV AEQ architecture, the product of the conjugate operation result of the current input signal, the current output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point;

[0048] In the MN RV (1x1) AEQ architecture, the real part of the current input signal, the real part of the current output signal, the convergence step, and the product of the error term corresponding to the current output signal are calculated to obtain the data loss function of each tap point;

[0049] In the MN RV (2x2) AEQ architecture, if the current input signal and the current output signal are both the real part of the signal or the imaginary part of the signal, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point; if the current input signal and the current output signal are the real part of the signal and the imaginary part of the signal, respectively, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point.

[0050] Specifically, the optical signal transmitted in the coherent optical communication system is a complex signal that can be expressed as i+jq. Taking the optical signal in the X polarization state as an example, the calculation formula of the data loss function of the signal equalizer in the AEQ with the above three MN tap configurations can be expressed as:

[0051]

[0052] Among them, μ is the convergence step length, is the empirical value, and ε x Represents the error term corresponding to the current output signal, X mid Indicates the current input signal of the signal equalizer, X out Indicates the current output signal of the signal equalizer, X mid * Represents the conjugate operation result of the current input signal, Re(X out ) indicates X out The real part, X mid,a ' represents part X mid,a , The data loss function for a signal equalizer whose input and output are both the real part of the signal or the imaginary part of the signal, It represents the data loss function of a signal equalizer whose input and output are the real part of the signal and the imaginary part of the signal respectively, and a and b represent the i-path or q-path.

[0053] From the above formula, we can see that in MN RV (2x2) AEQ architecture, if the current input signal and the current output signal are both the real part of the signal (a is i), then the data loss function of each tap point is the product of the current input signal, the real part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal; if the current input signal and the current output signal are both the imaginary part of the signal (a is q), then the data loss function of each tap point is the product of the current input signal, the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal; if the current input signal is the real part of the signal and the current output signal is the imaginary part of the signal (a is i and b is q), then the data loss function of each tap point is the product of the current input signal, the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal; if the current input signal is the imaginary part of the signal and the current output signal is the real part of the signal (a is q and b is i), then the data loss function of each tap point is the product of the current input signal, the real part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal. The real part of the signal and the imaginary part of the signal mentioned above are only used for the signal type (i or q) of the input signal equalizer and have nothing to do with the specific transmitted optical signal.

[0054] As an example, Figure 2As shown, optical signals of different polarization states need to be input into different signal equalizers, so the tap coefficient update process of the signal equalizer proposed in this application is described for a single signal equalizer. In addition, the data loss function of each tap point in a single signal equalizer is the same (but the data loss function of different signal equalizers may be different), that is, the error terms corresponding to the current output signal of each tap point in the signal equalizer of each AEQ are the same.

[0055] In some embodiments of the present invention, considering that the length of the filter is determined by the number of tap points, the number of taps of AEQ in the signal equalization stage can be customized according to the needs, and the initial weight of each tap point can be determined by the following method:

[0056] The initial weight of the tap point located in the middle of the signal equalizer is set to 1, and the sliding window convolution method is used for adaptive iteration, so as to determine the initial weight of each tap point in the signal equalizer. That is, assuming that there are N tap points in the signal equalizer, the initial weight of the tap point at (N+1) / 2 (or N / 2+1 or N / 2-1) is set to 1, and the adaptive iterative algorithm is used to determine the initial weight of each tap point in the signal equalizer. The adaptive iterative algorithm mentioned in the present application may be a sliding window convolution algorithm or other iterative algorithms, and the present invention does not specifically limit it.

[0057] Step S120: Determine the updated weight of each tap point according to the current weight of each tap point, the data loss function and the weight sparsification function to optimize the AEQ.

[0058] More specifically, the present application can increase the weight of important positions and reduce the weight of unimportant positions through sparseness. In theory, some weights can be sparsely reduced to zero, and the tap coefficients and number of taps of the FIR filter can be optimized to improve the computational efficiency of the AEQ model while retaining the performance of the model as much as possible. For example, sparse regularization (such as L1 regularization) or pruning can be used. The following uses L1 regularization as an example to sparsely process the tap coefficients of the filter in the signal equalizer.

[0059] The present application introduces an L1 regularization parameter in the process of updating the tap coefficients of the FIR filter, determines the weight sparsification function based on the set L1 regularization parameter λ and the current weight of each tap point, and determines the sparsification target using the data loss function and the weight sparsification function, thereby determining the update weight of each tap point of the filter according to the current weight of each tap point and the sparsification target.

[0060] In some embodiments of the present invention, the weight sparsification function can be used to represent the gradient of weight change, and the weight sparsification function of each tap point may be different. The weight sparsification function of each tap point is determined based on the set L1 regularization parameter and the current weight of each tap point, including: the positive and negative values ​​of the weight sparsification function of each tap point are determined by the positive and negative values ​​of the current weights of each tap point, and the absolute value of the weight sparsification function of each tap point is determined by the absolute value of the product of the L1 regularization parameter and the convergence step size. That is, the weight sparsification function of each tap point can be expressed as R = μ·λ·sign(W), where sign(·) is a sign function and W can be the current weight of the tap point.

[0061] Sparsification may affect both signal quality and computational complexity, so this application requires setting a sparsification target when setting the weight sparsification function. The sparsification target in this application is to minimize the data loss function and maximize sparsity, so the sparsification target can be expressed as That is, the sparsification target = data loss function - weight sparsification function.

[0062] In some embodiments of the present invention, after determining the update weights of each tap point, the method further includes: step S130, setting the update weights of the updated tap coefficients of the signal equalizer that are lower than the decision threshold value to zero. That is, after the tap coefficients are sparse, the threshold decision mechanism can be used to further optimize the tap coefficients and reduce the number of redundant taps. This process ensures the sparseness of the FIR filter and greatly reduces the computational complexity without significantly affecting the system performance. For example, a threshold value can be set, and any weight less than the threshold value will be set to zero, further improving the sparsity of the filter.

[0063] In some embodiments of the present invention, Figure 3 As shown, the set L1 regularization parameter is determined before updating the tap coefficients of the FIR filter in the signal equalizer, and the specific determination process is as follows:

[0064] In the training phase, the training input signal in the training set is input into the adaptive equalizer with the initial L1 regularization parameter, the error term of the signal equalizer in the adaptive equalizer is updated based on the current training output signal, and then the tap coefficient of the signal equalizer in the adaptive equalizer is updated using the current training input signal, the current training output signal, the error term corresponding to the current training output signal and the initial L1 regularization parameter, and the next training input signal is input into the updated adaptive equalizer; the iterative tap coefficient corresponding to the initial L1 regularization parameter is obtained by iteratively updating multiple training input signals, thereby obtaining the iterative adaptive equalizer corresponding to the initial L1 regularization parameter;

[0065] In the test phase, a test input signal in the test set is input into the iterative adaptive equalizer, and based on the test input signal and the corresponding test output signal, it is determined whether the iterative adaptive equalizer satisfies the set regularization condition;

[0066] If the set regularization condition is not met, the initial L1 regularization parameter is corrected and the training phase and the testing phase are repeated. If the set regularization condition is met, the initial L1 regularization parameter is used as the set L1 regularization parameter.

[0067] As an example, during the training phase and the update phase, a gradient descent algorithm (such as Cascaded Multi-modulus Algorithm, CMMA) can be used to obtain the error term corresponding to the output signal to minimize the data loss function. Specifically, during the training phase, the cascaded multi-modulus algorithm is used to calculate the difference between the square of the modulus of the current training output signal and the square of the reference modulus value to obtain the error term corresponding to the current training output signal; during the update phase, the cascaded multi-modulus algorithm is used to calculate the difference between the square of the modulus of the current output signal and the square of the reference modulus value to obtain the error term corresponding to the current output signal. In addition, during the training phase, the initial L1 regularization parameter and decision threshold can be introduced in the signal equalizer update phase for iterative update (such as Figure 3 As shown), if the AEQ obtained by iterative update is determined to meet the set regularization condition in the test phase, the initial L1 regularization parameter is used as the set L1 regularization parameter, and the decision threshold is used as the threshold value. The initial regularization parameter and the decision threshold can be set according to the sparsification target, and the present invention does not make specific limitations.

[0068] In some embodiments of the present invention, the regularization condition is set as a bit error rate less than a set bit error value and a sparsity degree greater than a set sparsity number; wherein the bit error rate is determined based on the output signal of the signal equalizer in the iterative adaptive equalizer during the test phase and the standard reference signal, and the sparsity degree is measured by the number of tap points in the iterative adaptive equalizer whose weights are greater than the set weight value.

[0069] As an example, a weighted method can be used to comprehensively quantify and measure the bit error rate and sparsity of AEQ, or other methods can be used, and the present invention does not specifically limit them. The bit error value and the set weight value can be set by yourself. For example, the set bit error value can be determined according to the current bit error rate, and the set weight value can be set to zero. In addition, the above-mentioned setting of regularization conditions is only an example. For example, only the bit error rate can be used as an indicator to determine whether the set regularization conditions are met, and the present invention is not limited to this.

[0070] As an example, multiple initial L1 regularization parameters can also be set in the present application. The bit error rate and sparsity degree of each initial L1 regularization parameter are determined through a training phase and a testing phase. The initial L1 regularization parameter with the smallest bit error rate change and the largest sparsity degree is selected from the multiple set initial L1 regularization parameters and is used as the set L1 regularization parameter λ.

[0071] Further, according to the AEQ optimization method proposed in this application, taking the optical signal in the X polarization state as an example, Figure 2 The tap coefficient update formulas for the three AEQs shown can be shown as follows:

[0072] W x =W x-1 +μ·(ε x ·X out ·X mid * -λ·sign(W x-1 )); (4)

[0073] W x =W x-1 +μ·(ε x ·Re(X out )·Re(X mid )-λ·sign(W x-1 )); (5)

[0074]

[0075] Among them, W x-1 and W x are the current tap coefficient and the updated tap coefficient respectively, W x,aa represents the tap coefficients of the real-real and imaginary-imaginary filters of the N1 tap number, W x,ab represents the tap coefficient of the real-imaginary filter with N2 taps, X mid,a ′ represents the input signal vector X mid,a part of, for example, X mid,a Represents the input signal of 21 tap points, X mid,a ' can represent the signals of the first 13 tap points.

[0076] As an example, the process of optimizing AEQ in the present application may be as follows: input a coherent optical signal; determine an error term based on the output optical signal, and use the input optical signal, the output optical signal, the determined error term and the set L1 regularization parameter to update the tap coefficients of the adaptive equalizer; after the tap coefficients are updated, use the threshold decision mechanism to further sparse the tap coefficients of the AEQ, and finally obtain an optimized AEQ that can be used to equalize the signal. The optimization method proposed in the present application can be divided into three stages. In the training stage, the initial L1 regularization parameter and decision threshold are set and an iterative adaptive equalizer is obtained. In the testing stage, the initial L1 regularization parameter and decision threshold corresponding to the iterative adaptive equalizer that meets the set regularization conditions are determined, and they are used as the set L1 regularization parameter and threshold value respectively. In the update stage, steps S110, S120 and S130 are executed to update the weights of each tap point based on the set L1 regularization parameter and threshold value.

[0077] In a specific embodiment of the present invention, a CV filter refers to a filter that performs complex operations on an input signal, and a RV filter refers to a filter that performs real operations on an input signal. In addition, the signal input to the AEQ can be converted between complex numbers and real numbers based on the operation method applicable to the filter (for example, using Matlab for conversion). If the filter performs one complex multiplication, it is equivalent to four real multiplications. Taking into account the poor equalization effect of the traditional 2x2 MIMO AEQ, this application uses the traditional 4x4 RV MIMOAEQ as a reference to compare the computational complexity and performance of multiple two-stage AEQs. The comparison results of the computational complexity of different AEQs can be as follows. Compared with the traditional 4x4 MIMO RV AEQ, the computational complexity of MN CV AEQ, MN RV (1x1) AEQ and MN RV (2x2) AEQ are reduced by 40.5%, 65.5% and 50%, respectively:

[0078] (1) Assuming the number of taps N is 21, the conventional 4×4 MIMO RV AEQ requires 4×4×21=336 real number multiplications to equalize a signal.

[0079] (2) Assuming that M is 2 and N is 21 for the two-stage AEQ, then Figure 2 (a) MN CV AEQ in the figure performs multiplication on the real and imaginary parts of the input complex signal in the polarization demultiplexing stage, and performs complex multiplication on the complex signals of the X and Y polarization states in the signal equalization stage. Therefore, MN CV AEQ requires 16×M+2×4×N=200 real multiplications to equalize one signal.

[0080] (3) Assuming that M is 2 and N is 21 for the two-stage AEQ, then Figure 2(b)MN RV(1x1)AEQ in the figure performs multiplication on the real and imaginary parts of the input complex signal in the polarization demultiplexing stage, and performs real number operations on the complex signals of the two polarization states x and y in the signal equalization stage (multiplying the real and imaginary parts of the complex signals respectively). Therefore, 16×M+2×2×N=116 real number multiplications are required to equalize a signal.

[0081] (4) Assuming that M of the two-stage AEQ is 2, N1 is 21, and N2 is 13, then Figure 2 In (c)MN RV(2x2)AEQ, the M-tap filter performs multiplication on the real and imaginary parts of the input complex signal in the polarization demultiplexing stage, the N1-tap filter performs real multiplication on the real parts of the complex signals in the two polarization states of x and y in the signal equalization stage, and the N2-tap filter performs real multiplication on the imaginary parts of the complex signals in the two polarization states of X and Y in the signal equalization stage. Therefore, 16×M+2×2×N1+2×2×N2=168 real multiplications are required to equalize a signal.

[0082] Further, using the AEQ optimization method proposed in this application, based on the L1 regularization parameter λ and threshold value set in Table 1, the per-symbol computational complexity (i.e., the complexity of equalizing a signal) of 4x4 MIMO RV AEQ, MN CV AEQ, MN RV (1x1) AEQ, and MN RV (2x2) AEQ is reduced to 146, 160, 96, and 101, respectively, after optimization, and the computational complexity is reduced by 56.5%, 52.4%, 71.5%, and 70%, respectively. That is, compared with the traditional 4x4 MIMO RV AEQ, the method of this application can effectively reduce the computational complexity of AEQ.

[0083] Table 1 Example of setting L1 regularization parameter λ and threshold value of AEQ of various architectures

[0084] AEQ Architecture The L1 regularization parameter λ is set Threshold 4x4 MIMO RV AEQ 0.01 0.02 MN CV AEQ 0.02 0.02 MN RV(1x1)AEQ 0.005 0.01 MN RV(2x2)AEQ 0.02 0.02

[0085] In some embodiments of the present invention, the Figure 4 The 50GBaud polarization multiplexed 16-quadrature amplitude modulation (PDM-16QAM) signal optical back-to-back transmission system shown in the figure verifies the optimized AEQ performance. The transmission process of the optical signal is as follows:

[0086] In the transmitter DSP, the generated 50GBaud 16QAM signal is first pulse shaped by a root raised cosine filter with a roll-off factor of 0.1. After resampling, the two polarization independent signals are loaded into an arbitrary waveform generator running at 120GSa / s, and then the modulator is used to modulate the electrical signal onto a 1550nm optical carrier. A variable optical attenuator is used to control the received optical power (ROP). Subsequently, a digital sampling oscilloscope running at 256GSa / s is used to capture the output signal of the integrated coherent receiver. The offline receiver DSP includes resampling of 2 samples per symbol (2 samples per signal) and IQ imbalance compensation. After matched filtering, AEQ is used for polarization demultiplexing and equalization. Then, a carrier recovery algorithm is performed to remove frequency offset and phase noise, followed by decision, symbol demapping and bit error rate calculation.

[0087] Based on the above verification system, Figure 2 The performance of the adaptive equalizer constructed by the three AEQ architectures is tested. The verification results are as follows: Figure 3 As shown. Based on the back-to-back transmission of light, Figure 5 (a) shows the transmission performance under different received optical powers. Figure 5 From (a) in Figure 1, it can be seen that when the ROP (receiver, integrated coherent optical receiver) is -20.2dBm, the bit error rates of the various AEQs are similar, and the receiver sensitivity is 2×10 -2 , almost at the same position. That is, the performance of the sparse AEQ optimized by the pruning scheme proposed in this patent is not only comparable to that of the traditional 4x4 MIMO RV AEQ, but also maintains the receiver sensitivity comparable to that of the existing AEQ, proving that the AEQ after optimization and pruning in this application can strike a balance between reduced complexity and robust system performance. Figure 5 (b) and (c) in the figure show the tolerance of the AEQ to IQ imbalance after optimized pruning in this application. The experimental results show that the AEQ after using the optimized pruning scheme can still maintain the robustness to IQ imbalance (amplitude and phase imbalance of I and Q signals) comparable to that of the non-pruned AEQ.

[0088] This application optimizes AEQ based on the idea of ​​lightweight model. The core idea is to introduce L1 regularization term and decision threshold to sparse the tap coefficients of the filter during the tap coefficient update process of the FIR filter of AEQ, thereby effectively reducing the computational complexity. In this application, the introduction of L1 regularization parameter to achieve tap coefficient sparseness is only an example. Other sparseness methods (such as Elastic Net regularization) can also be used to update the FIR filter of the signal equalizer in AEQ, and its update process can be customized according to the sparseness target.

[0089] The optimization method of AEQ for coherent optical communication systems proposed in this application can adaptively sparse the number of filter taps during the weight update process through the effect of L1 regularization, reduce the invalid or less influential tap coefficients in the system, and thus reduce the computational complexity. The updated AEQ can adaptively suppress redundant taps for the received signal, while significantly reducing the computational complexity of the equalizer, effectively maintaining the performance of AEQ, and maintaining a high tolerance for IQ amplitude and phase imbalance, thereby providing a new technical path for realizing low-power, high-efficiency coherent optical communication systems.

[0090] Corresponding to the above method, the present invention also provides an optimization device for an adaptive equalizer of a coherent optical communication system, the device comprising a computer device, the computer device comprising a processor and a memory, the memory storing a computer program / instructions, the processor being used to execute the computer program / instructions stored in the memory, and when the computer program / instructions are executed by the processor, the device implements the steps of the method described above.

[0091] It should be understood by those skilled in the art that the exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0092] It should be clear that the present invention is not limited to the specific configuration and processing described above and shown in the figures. For the sake of simplicity, a detailed description of the known method is omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of the present invention is not limited to the specific steps described and shown, and those skilled in the art can make various changes, modifications and additions, or change the order between the steps after understanding the spirit of the present invention.

[0093] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with features of other embodiments or replace features of other embodiments.

[0094] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the embodiments of the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An optimization method for an adaptive equalizer AEQ for a coherent optical communication system, wherein the adaptive equalizer comprises a polarization demultiplexer and a signal equalizer receiving an output signal of the polarization demultiplexer, characterized in that: The method comprises the following steps: Acquire a current input signal and a current output signal of a signal equalizer, and determine a data loss function of each tap point based on an error term corresponding to the current output signal, the current input signal, and the current output signal; The new weights of each tap point are determined according to the current weights of each tap point, the data loss function and the weight sparsification function to optimize the adaptive equalizer; wherein the weight sparsification function of each tap point is determined based on the set L1 regularization parameter and the current weights of each tap point.

2. The method according to claim 1, characterized in that The weight sparsification function of each tap point is determined based on the set L1 regularization parameter and the current weight of each tap point, including: The positive or negative value of the weight sparsification function of each tap point is determined by the positive or negative value of the current weight of each tap point, and the absolute value of the weight sparsification function of each tap point is determined by the absolute value of the product of the L1 regularization parameter and the convergence step size.

3. The method according to claim 1, characterized in that After determining the new weights of the tap points, the method further includes: setting the new weights below the decision threshold to zero.

4. The method according to claim 1, characterized in that: The L1 regularization term parameters are determined in the following way: In the training phase, the training input signal in the training set is input into the adaptive equalizer with the initial L1 regularization parameter, the error term of the signal equalizer in the adaptive equalizer is updated based on the current training output signal, and the tap coefficient of the signal equalizer in the adaptive equalizer is updated based on the current training input signal and the current training output signal, and the next training input signal is input into the updated adaptive equalizer; the iterative tap coefficient corresponding to the initial L1 regularization parameter is obtained by iteratively updating multiple training input signals, thereby obtaining the iterative adaptive equalizer corresponding to the initial L1 regularization parameter; In the test phase, a test input signal in a test set is input into the iterative adaptive equalizer, and based on the test input signal and a corresponding test output signal, it is determined whether the iterative adaptive equalizer satisfies a set regularization condition; If the set regularization condition is not met, the initial L1 regularization parameter is corrected and the training phase and the testing phase are repeated. If the set regularization condition is met, the initial L1 regularization parameter is used as the set L1 regularization parameter.

5. The method according to claim 4, characterized in that The set regularization condition is that the bit error rate is less than the set bit error value and the sparsification degree is greater than the set sparsification number; wherein the bit error rate is determined based on the output signal of the signal equalizer in the iterative adaptive equalizer during the test phase and the standard reference signal, and the sparsification degree is measured by the number of tap points in the iterative adaptive equalizer whose weights are greater than the set weight value.

6. The method according to claim 1, characterized in that The signal equalizer is constructed using a finite impulse response (FIR) filter, and the architecture of the adaptive equalizer is MN CV AEQ, MN RV (1x1) AEQ or MN RV (2x2) AEQ; The determining of the data loss function of each tap point based on the error term corresponding to the current output signal, the current input signal and the current output signal comprises: In the MN CV AEQ architecture, for optical signals in each polarization state, the product of the conjugate operation result of the current input signal, the current output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point. In the MN RV (1x1) AEQ architecture, for optical signals of each polarization state, the real part of the current input signal, the real part of the current output signal, the convergence step size, and the product of the error term corresponding to the current output signal are calculated to obtain the data loss function of each tap point; In the MN RV (2x2) AEQ architecture, for optical signals in each polarization state, if the current input signal and the current output signal are both the real part of the signal or the imaginary part of the signal, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point; if the current input signal and the current output signal are the real part of the signal and the imaginary part of the signal, respectively, the product of the current input signal, the real part or the imaginary part of the AEQ output signal, the convergence step, and the error term corresponding to the current output signal is calculated to obtain the data loss function of each tap point.

7. The method according to claim 1, characterized in that The initial weights of the tap points in the signal equalizer are determined in the following manner: The initial weight of the tap point located in the middle of the signal equalizer is set to 1, and the sliding window convolution method is used for adaptive iteration to determine the initial weight of each tap point in the signal equalizer.

8. The method according to claim 1, characterized in that The error term corresponding to the current output signal is obtained by calculating the difference between the square of the modulus of the current output signal and the square of the reference modulus value using a cascaded multi-mode algorithm.

9. The method according to claim 1, characterized in that: The data loss function is the same for each tap point in the signal equalizer.

10. An optimization device for an adaptive equalizer of a coherent optical communication system, comprising a processor, a memory, and a computer program / instruction stored in the memory, characterized in that: The processor is used to execute the computer program / instructions. When the computer program / instructions are executed, the device implements the steps of the method according to any one of claims 1 to 9.

Images