CN-115776341-B - Volterra nonlinear equalization method and system

CN115776341BCN 115776341 BCN115776341 BCN 115776341BCN-115776341-B

Abstract

The invention relates to a Volterra nonlinear equalization method and a Volterra nonlinear equalization system, wherein the method comprises the following steps of obtaining an input signal; when the improved Volterra equalizer is designed, the traditional algorithm of the prior art that all training sequences are directly used for training are improved into a new algorithm of intercepting and disturbing the groups, dividing the training set and the testing set and training in two stages. Compared with the prior art, the invention effectively solves the invisible and uncontrollable problems of the overfitting phenomenon during the training of the equalizer while improving the training efficiency of the equalizer, thereby improving the compensation effect, and simultaneously has good universality.

Inventors

YU JIANJUN
Sang Bohan
KONG MIAO

Assignees

复旦大学

Dates

Publication Date: 20260512
Application Date: 20221130

Claims (8)

1. A Volterra nonlinear equalization method, comprising the steps of: S1, acquiring an input signal; S2, inputting the input signal into an improved Volterra equalizer after a pre-DSP step, S3, outputting an equalization result by the improved Volterra equalizer; wherein the acquisition of the improved Volterra equalizer comprises the steps of: 1) Acquiring a signal sequence of a transmitting end ; 2) Generating training sequence, transmitting the training sequence through channel to obtain receiving end signal sequence, and pre-DSP step to obtain Volterra equalizer input signal sequence ; 3) Inputting the Volterra equalizer into a signal sequence Grouping interception is carried out to obtain an input signal sequence fragment set G; 4) The input signal sequence fragment set and the signal sequence of the transmitting end are combined Obtaining a fragment pair set P in one-to-one correspondence; 5) The fragments are subjected to disorder treatment on elements in the collection, so that a disorder collection is obtained ; 6) Aggregating the out-of-order set Segmentation into training sets And test set ; 7) Training the training set The signal sequence segments and the corresponding signal sequence segments of the transmitting end are input into a Volterra equalizer, a training loss value is calculated, error feedback is carried out, and a weight coefficient is updated; 8) Integrating the test set Inputting the signal sequence segments in (1) and the corresponding signal sequence segments of the transmitting end into a Volterra equalizer, calculating a test loss value without returning errors, comparing the test loss value with the training loss value obtained in the step 7), and calculating the error code performance of the system; 9) If the error code performance of the system does not meet the preset requirement, adjusting each parameter of the second-order Volterra equalizer, and repeating the step 7) and the step 8), otherwise, executing the next step; 10 Inputting the signal sequence segments in the test set and the corresponding signal sequence segments of the transmitting end into a Volterra equalizer, calculating a loss value and carrying out error feedback, and updating a weight coefficient to obtain a weight coefficient after training is completed; 11 Initializing the Volterra equalizer by using the weight coefficient after training to obtain an improved Volterra equalizer; step 3), the Volterra equalizer input signal sequence is processed The packet interception, get the segment set G of the input signal sequence, including the following steps: defining the number of taps for packet interception as The expression is: Wherein, the For said Volterra equalizer The size of the order kernel; The expression of the obtained input signal sequence fragment set is: Wherein, the For inputting signal sequences The first of (3) The number of elements to be added to the composition, Total number of segments of the input signal sequence; in the step 7) and the step 8), the calculation formula of the test loss value and the training loss value is as follows: Wherein, the As a function of the error, Is a transmitting end signal sequence.
2. The Volterra nonlinear equalization method of claim 1, wherein in step 6), the training set-based ratio coefficients are specified Coefficient of ratio to test set The out-of-order set Segmentation into training sets And test set 。
3. The Volterra nonlinear equalization method of claim 2, wherein the training set ratio coefficients Coefficient of ratio to test set The following relationship is provided: and the data length of the training set is: ; the data length of the test set is: , Is the total number of segments of the input signal sequence.
4. The method of claim 1, wherein the Volterra equalizer comprises a first order portion and a second order portion, the first order portion storing a length of Is a sequence of signals of (2) : The second-order part stores a length of Is a sequence of signals of (2) : And Respectively by And In a set of signal sequence fragments And Intercepting to obtain the final product.
5. The method of claim 4, wherein the first-order part comprises first-order kernels consisting of a single storage signal for compensating for linear impairments, the number of first-order kernels being The first order kernel vector expression is: The second-order part comprises a second-order core, the second-order core is formed by the product of any two stored signals and is used for compensating nonlinear damage, and the number of the second-order cores is that The second order kernel vector expression is: 。
6. the Volterra nonlinear equalization method of claim 5, wherein said first order kernel and second order kernel each correspond to a set of weight coefficients, said first order kernel weight coefficients Expressed as: the weight coefficient of the second order kernel Expressed as: The weight coefficient And Is 0.
7. The Volterra nonlinear equalization method of claim 6, wherein the method is based on a first order kernel vector, a second order kernel vector, and weight coefficients And Obtaining the output of the Volterra equalizer The expression is: 。
8. A Volterra nonlinear equalization system, which is characterized by being used for realizing the Volterra nonlinear equalization method according to any one of claims 1-7, and comprising a packet interception module, a signal alignment module, an out-of-order module, a data segmentation module and a Volterra equalizer module; a packet interception module for inputting the equalizer into the signal sequence According to the number of taps Intercepting to obtain an input signal sequence fragment set ; A signal alignment module for collecting the input signal sequence fragments Corresponding to the signal sequences of the transmitting end one by one, and packaging and storing to obtain fragment pair sets ; The disorder module is used for carrying out disorder processing on the fragment pair set to obtain a disorder set ; Data segmentation module based on training set ratio coefficient Coefficient of ratio to test set The out-of-order set Segmentation into training sets And test set ; The Volterra equalizer module is used for equalizing the input signal according to the weight coefficient of each step, and the Volterra equalizer module can calculate the error of the equalizing result by using the reference signal and return the error to optimize the weight coefficient.

Description

Volterra nonlinear equalization method and system Technical Field The invention relates to the field of optical fiber communication, in particular to a Volterra nonlinear equalization method and system. Background Along with the continuous development and development progress of technology, new technologies such as big data, cloud computing, artificial intelligence and the like are continuously emerging. These developments have actually improved people's lives, creating tremendous commercial value, with the consequent tremendous bandwidth demands worldwide. High capacity and high speed have become the primary targets for the development of next generation fiber optic transmission systems. In an optical fiber communication system, nonlinear damage has a very significant effect on system performance, and great obstruction is brought to capacity and rate improvement. In terms of hardware, various optical/electrical devices in the existing optical fiber communication system, such as a Mach-Zehnder modulator (MZM), a Photodiode (PD), an amplifier, a laser and the like, bring nonlinear damage, and the nonlinear damage brought by the optical fiber also becomes not neglected in long-distance transmission. These hardware are difficult to make breakthrough progress in reducing nonlinear damage in a short time, so nonlinear compensation algorithm at the receiving end is particularly important. The Volterra equalization algorithm is an excellent nonlinear compensation method and has an excellent compensation effect on the linear and nonlinear damage of the system. The algorithm is based on a multi-order Volterra series, and a training sequence is used for iterative training to obtain tap coefficients. In literature 【M.Kong,K.Wang,J.Ding,J.Zhang,W.Li,J.Shi,F.Wang,L.Zhao,C.Liu,Y.Wang,W.Zhou,and J.Yu,"640-Gbps/carrier WDM Transmission over6,400km Based on PS-16QAM at 106Gbaud Employing Advanced DSP,"Journal of Lightwave Technology,pp.1-1,2020】, the authors used 4 independent Volterra equalizers for 4 PS-PAM4 signals (i.e. the in-phase/quadrature components of the X/Y polarization) in the PS-16QAM signal, training the linear and nonlinear tap coefficients of the 4 PAM signals according to the PS-16QAM training sequence, respectively, and using the tap coefficients obtained by training for the actual transmission. However, training a set of training sequences that are much shorter than the actual transmission sequence in multiple iterations will not only allow the Volterra equalizer to learn the linear and nonlinear characteristics inherent in the channel, but will also fit some random processes. The Volterra equalizer obtained only according to the error rate result of the equalization training sequence cannot determine whether over fitting and the degree of over fitting occur in the training process, and the situation that the over fitting phenomenon cannot be effectively avoided during parameter adjustment training can lead to a gap between the equalizer effect and the theoretical effect obtained in training when the tap coefficient obtained through training is used for actual transmission. Disclosure of Invention The invention aims to overcome the defects of the prior art and provide the Volterra nonlinear equalizer which is applied to an optical fiber communication system, and the Volterra nonlinear equalizer not only has the compensation function of the traditional Volterra algorithm on linear and nonlinear damages, but also improves the training efficiency, and effectively solves the problems of invisible and uncontrollable overfitting when the Volterra equalizer is trained. The aim of the invention can be achieved by the following technical scheme: a Volterra nonlinear equalization method comprising the steps of: S1, acquiring an input signal; S2, inputting the input signal into an improved Volterra equalizer after a pre-DSP step, S3, outputting an equalization result by the improved Volterra equalizer; wherein the acquisition of the improved Volterra equalizer comprises the steps of: (1) Acquiring a signal sequence x of a transmitting end; (2) Generating a training sequence, and transmitting the training sequence through a channel to obtain a receiving end signal sequence; (3) Intercepting the input signal sequence y of the Volterra equalizer in a grouping way to obtain an input signal sequence fragment set G; (4) The input signal sequence fragment set is in one-to-one correspondence with the signal sequence x of the transmitting end, and a fragment pair set P is obtained; (5) Carrying out disorder treatment on elements in the set by the fragments to obtain a disorder set P shuffled; (6) Partitioning the out-of-order set P shuffled into a training set P train and a test set P test; (7) Inputting the signal sequence segments in the training set P train and the corresponding signal sequence segments of the transmitting end into a Volterra equalizer, calculating a training loss value, carrying out error feedback, and