A DSP method and system for QAM coherent optical communication

By unifying the pilot form in QAM coherent optical transmission communication, the overhead and cost problems caused by inconsistent pilot symbols are solved, and higher pilot utilization and transmission performance are achieved, ensuring the effective transmission of high-order QAM signals.

CN117353827BActive Publication Date: 2024-11-22SUZHOU UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202311252420.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-26
Publication Date
2024-11-22
Estimated Expiration
2043-09-26

AI Technical Summary

Technical Problem

Inconsistency of pilot symbols in the prior art leads to additional pilot overhead and hardware costs, affecting the efficiency of QAM coherent optical transmission communication.

Method used

The pilot symbol is constructed by receiving QAM signals and applied to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization at the same pilot ratio during the DSP process, unified pilot form to improve Utilization.

Benefits of technology

It reduces hardware costs, improves pilot utilization, and realizes effective transmission of PS-PDM-1024QAM signals in 80 kilometers standard single-mode fiber, improving transmission performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117353827B_ABST
    Figure CN117353827B_ABST
Patent Text Reader

Abstract

The present invention relates to a DSP method and system for QAM coherent optical communication, the method comprising: step S1: receiving a QAM signal, constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution; step S2: applying the same pilot symbol to the polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the DSP process with the same pilot ratio for the QAM signal in the DSP process. The present invention unifies the pilot symbols of the five major processing modules in the DSP, greatly improving the utilization rate of the pilot. Experiments have shown that the present invention can greatly improve the transmission performance when applied to the DSP of QAM coherent optical communication.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of digital signal processing, and in particular to a DSP method and system for QAM coherent optical communication. Background Art

[0002] With the continuous development of services such as the Internet of Things, big data, and cloud computing, data center interconnect (DCI) is required to achieve higher speed and larger capacity in medium and long-distance optical fiber transmission. In order to improve the transmission capacity of a single carrier, increasing the modulation order and increasing the spectrum efficiency (SE) is a necessary step. In wireless access networks, in order to obtain higher data transmission rates, the number of modulation formats has increased from 64-QAM in IEEE 802.11 Wi-Fi 4 to 256QAM in Wi-Fi 5, 1024QAM in Wi-Fi 6, and 4096QAM in Wi-Fi 7.

[0003] Meanwhile, in optical fiber communications, 256QAM, 1024QAM, and 4096QAM signals have been able to be transmitted over SSMF through probability shaping (PS) technology. However, with the increase in modulation formats, the amplitude and phase information carried by the signal also increases, making it more susceptible to component defects and various transmission degradations. Therefore, recovering the original information from the distorted signal is a challenge, and effective digital signal processing (DSP) algorithms become indispensable. In traditional QAM coherent optical communications, the DSP architecture includes pilot-assisted algorithms and blind algorithms. Compared with blind algorithms, pilot-assisted algorithms sacrifice the net bit rate due to the insertion of pilot symbols while achieving fast convergence. Therefore, effective use of these pilot symbols is attractive to pilot-assisted DSP algorithms.

[0004] In recent years, different pilot forms have been applied to different equalization algorithms. One uses some pre-QAM symbols as training sequences to initialize the equalization filter, and the remaining QAM symbols are interleaved with QPSK pilots to improve the elasticity of the constant modulus algorithm (CMA) equalization. At the achievable information rate (AIR), a 1600km transmission experiment of 256 / 1024QAM at 510GBaud was completed. Another pilot symbol-based carrier recovery algorithm compares the phase noise tolerance of PS-1024QAM and US-256QAM in scenarios with different laser linewidths. The conclusion is that the laser linewidth and CPR algorithm should be considered when selecting the signal shaping form for system design. In addition, Miao Kong et al. proposed a Volterra equalizer (VE) consisting of a first-order 2×2 multiple-input multiple-output (MIMO) replica generalized linear (WL) equalizer and a second-order real-valued nonlinear equalizer. Using certain pilot training taps, single-carrier 800Gb transmission was achieved with the modulation format of PS-64QAM. Nokia Bell Labs used pilot-assisted pre-convergence for polarization demultiplexing and used extremely narrow linewidth lasers to improve tolerance to phase noise at low baud rates, ultimately achieving 200km fiber transmission of PS-PDM-4096QAM with a spectrum efficiency of 19.77bit / s / Hz. In fact, in the above-mentioned pilot-assisted equalization, carrier phase recovery and polarization demultiplexing technologies, the structure of pilot symbols is varied. Inconsistency in pilot symbols will bring additional pilot overhead and hardware costs. Summary of the invention

[0005] Therefore, the technical problem to be solved by the present invention is to overcome the problem of additional pilot overhead and hardware cost caused by the inconsistency of pilot symbols in the prior art.

[0006] In order to solve the above technical problems, the present invention provides a DSP method for QAM coherent optical communication, comprising:

[0007] Step S1: receiving a QAM signal, and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution;

[0008] Step S2: In the DSP process, the same pilot symbol is applied to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the QAM signal with the same pilot ratio.

[0009] In one embodiment of the present invention, in step S1, the pilot symbol is constructed according to the received QAM signal, and the method includes:

[0010] The QAM signal is taken as a code block to be transmitted, and in one of the code blocks, a preset number of symbols are taken as pilot symbols.

[0011] In one embodiment of the present invention, the pilot ratio formula is:

[0012] R=P / L

[0013] Wherein, L is the length of the code block and P is the number of pilot symbols.

[0014] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to polarization demultiplexing of a DSP process with the same pilot ratio, and the method includes:

[0015] By taking a preset number of symbols in front of a code block as a known symbol sequence, a radius blind decision is avoided when the polarization mode multiplexing of the QAM signal is demultiplexed by the MMA algorithm, the MMA algorithm is tapped and trained by using the known symbol sequence, and then the MMA algorithm is applied to the entire code block;

[0016] Among them, the MMA algorithm uses one tap and the tap iteration formula of the MMA algorithm is:

[0017]

[0018] Among them, p and q represent the X and Y polarizations, n is the current iteration number, μ is the step size, and e p (n) is the error of polarization p at the nth iteration, v p (n) is the output signal of polarization p at the nth iteration, u q (n) is the input signal of polarization q at the nth iteration, the expression of the error e is e=d-|v|, d is the amplitude that the output signal should be judged to be at the current iteration number, and v is the output signal.

[0019] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to polarization demultiplexing of a DSP process with the same pilot ratio, and the method further includes:

[0020] Use 33 tap coefficients between nonlinear equalization and linear equalization, and change H pq Separate the real and imaginary parts of the signal, and the iteration formula is:

[0021] H pq-ab (n+1)=H pq-ab (n)+μ·e p-a (n) ·u q-b (n)

[0022] Among them, p and q represent the X and Y polarizations, a and b represent the real and imaginary parts of the QAM signal, n is the current iteration number, μ is the step size, and e is the real part. p-a (n) is the error of the a part of polarization p at the nth iteration, u q-b (n) is the input signal of part b of polarization q at the nth iteration.

[0023] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to the frequency offset estimation of the DSP process with the same pilot ratio, and the method includes:

[0024] The frequency offset is estimated by combining the fourth power FFT to obtain the frequency offset and the pilot-assisted FFT to obtain the frequency offset. Specifically, the input signal y after the depolarization mode multiplexing is n Extract pilot signal p′ n and the complex conjugate of the transmitted signal After multiplication, calculate the FFT and find the index of the maximum value I p , the frequency offset to which the pilot symbol is located is:

[0025] Δf p =R S I p / N FFT

[0026] Among them, R S is the symbol rate of the signal, N FFT is the number of transformation points, Δf p ±Δf i is the range for searching for precise frequency deviation, which corresponds to y n The index range of the fourth power of FFT is [(Δf p -Δf i )·4N FFT / R S ,(Δf p +Δf i )·4N FFT / R S ], the index of the maximum value found in the index range is I, and the final estimated frequency offset is:

[0027] Δf=R S I / 4N FFT

[0028] Apply frequency offset compensation to y n For each QAM symbol, the formula is:

[0029]

[0030] Among them, T s is the symbol period, yn,i and z n,i are the i-th symbol in a code block in the input and output respectively.

[0031] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to the carrier phase recovery of the DSP process with the same pilot ratio, and the method includes:

[0032] Take m code blocks, the average offset phase of the pilot symbol of each code block is θ ri , i = 1, 2, .., m, and zero the remaining phases of the input signal after frequency offset estimation, and perform smoothing and decomposition processing at the same time, then connect each one to obtain the phase offset trend, and finally perform three interpolations to obtain the phase offset of the entire signal, and compensate the obtained phase offset of the entire signal to the entire signal.

[0033] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to the nonlinear equalization of the DSP process with the same pilot ratio, and the method includes:

[0034] The nonlinear equalization is achieved using the Volterra nonlinear equalization algorithm based on the second-order Volterra series. The formula is:

[0035]

[0036] The above formula is the non-recursive series expansion expression of Volterra equalizer, where f 0i is the tap coefficient of the i-th order filter, l i is the memory length of the i-th order filter, r(k) is the k-th symbol input to the Volterra nonlinear equalization algorithm;

[0037] The tap coefficients of the Volterra nonlinear equalization algorithm are trained by the pilot symbols, and the tap coefficients of the Volterra nonlinear equalization algorithm are updated based on a recursive least squares scheme.

[0038] In one embodiment of the present invention, in step S2, the same pilot symbol is applied to the linear equalization of the DSP process with the same pilot ratio, and the method includes:

[0039] Linear equalization is achieved by a constellation clustering algorithm based on minimum mean square error. The tap iteration formula of the constellation clustering algorithm based on minimum mean square error is:

[0040] ω(n+1)=ω(n)+μ LMS ·e LMS (n) ·u LMS * (n)

[0041] Among them, n is the current iteration number, μ LMS is the step length, e LMS (n) is the error of the nth iteration, u LMS (n) is the input signal of the nth iteration.

[0042] In order to solve the above technical problems, the present invention provides a DSP system for QAM coherent optical communication, comprising:

[0043] A construction module: used for receiving a QAM signal and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution;

[0044] Application module: used for applying the same pilot symbol with the same pilot ratio to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the DSP process during the DSP process of the QAM signal.

[0045] The above technical solution of the present invention has the following advantages compared with the prior art:

[0046] The present invention unifies the pilot forms of five equalization modules, including polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization, and linear equalization, and applies the same pilot to each module at the same ratio, which greatly improves the utilization rate of the pilot and the utilization rate of the pilot symbol while improving the transmission performance;

[0047] The present invention greatly reduces the hardware cost of the high-order QAM coherent optical communication system, and experiments have confirmed that the proposed DSP method can ensure the effective transmission of PS-PDM-1024QAM signals on 80 km standard single-mode optical fiber (SSMF). BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to make the contents of the present invention more clearly understood, the present invention is further described in detail below based on specific embodiments of the present invention in conjunction with the accompanying drawings.

[0049] Figure 1 is a flow chart of the method of the present invention;

[0050] Figure 2 It is a flowchart of an off-line DSP program of a receiver in an embodiment of the present invention;

[0051] Figure 3 1 is a schematic diagram comparing the bit error rate and shaping parameter β of pre-QAM and pre-QPSK at a pilot ratio of 10% in an embodiment of the present invention;

[0052] Figure 4 (a) is a principle block diagram and an extended real-virtual block diagram of the MMA algorithm in an embodiment of the present invention;

[0053] Figure 4 (b) is a schematic diagram of Rician distribution density at each radius under 1024QAM in an embodiment of the present invention;

[0054] Figure 4 (c) is a schematic diagram showing the comparison of convergence of using pre-pilot assistance and using only blind radius decision for different entropy values ​​in an embodiment of the present invention;

[0055] Figure 4 (d) is a schematic diagram of bit error rate and bit error imbalance before and after using N-tap MMA compensation at a pilot ratio of 5% in an embodiment of the present invention;

[0056] Figure 5 It is a block diagram of an algorithm combining quartic FFT frequency offset calculation and pilot-assisted FFT frequency offset calculation in an embodiment of the present invention;

[0057] Figure 6 is a block diagram of a pilot-assisted phase offset algorithm in an embodiment of the present invention;

[0058] Figure 7 is a block diagram of a VNLE in an embodiment of the present invention;

[0059] Figure 8 Schematic diagram of the relationship between OSNR and EVM of PS-PDM-1024QAM signal transmission with a shaping parameter of 3.25 under different forms of linear equalization in an embodiment of the present invention;

[0060] Fig. 9 Schematic diagram of an experimental device for ultra-high-order PS-PDM-QAM transmission in an embodiment of the present invention;

[0061] Fig.10 is a schematic diagram of the relationship between shaping parameters and NGMI in BTB transmission of PS-PDM signals in an embodiment of the present invention;

[0062] Fig.11 Schematic diagram of the relationship between OSNR and AIR under different pilot ratios R when the PS-PDM signal is transmitted over 80km SSMF in an embodiment of the present invention;

[0063] Fig.12 Schematic diagram of the relationship between OSNR and NGMI of PS-PDM-1024QAM at a pilot ratio of 10% under different line width conditions in an embodiment of the present invention. DETAILED DESCRIPTION

[0064] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments so that those skilled in the art can better understand the present invention and implement it, but the embodiments are not intended to limit the present invention.

[0065] Embodiment 1

[0066] Reference Figure 1 As shown, the present invention relates to a DSP method for QAM coherent optical communication, comprising:

[0067] Step S1: receiving a QAM signal, and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution;

[0068] Step S2: In the DSP process, the same pilot symbol is applied to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the QAM signal with the same pilot ratio.

[0069] This embodiment unifies the pilot forms of the five equalization modules, including polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization, and linear equalization, and applies the same pilot to each module at the same ratio, which greatly improves the utilization rate of the pilot while improving the transmission performance. This embodiment compares the shaping gains that can be obtained by 1024QAM and 4096QAM under different shaping parameters, and adjusts the pilot ratio to about 5% based on this parameter to obtain the maximum achievable information rate (AIR). In addition, appropriately adjusting the form of some equalization modules under different OSNR conditions is conducive to obtaining higher gains. At the same time, this embodiment also studies the influence of laser linewidth to find out whether it is necessary to use expensive narrow linewidth lasers.

[0070] The following is a detailed introduction to this embodiment:

[0071] The structure of the present invention is as follows: in Section 2, this embodiment introduces the definition of a universal pre-pilot and how it is applied to each equalization module; in Section 3, this embodiment introduces an experimental device for transmitting PS-PDM-1024 / 4096QAM signals; in Section 4, this embodiment evaluates and discusses the experimental results; in Section 5, this embodiment gives a conclusion.

[0072] Section 2: Definition and use of pre-placed pilot symbols

[0073] In this section, this embodiment introduces the definition of the universal pilot header and its use in various DSP algorithms, and compares the performance with the algorithm that only uses unknown (blind) data. Algorithms based on pre-pilot include: depolarization mode multiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization, and linear equalization. The pilot symbols used by the above algorithms can be unified, that is, in a code block with a specific code length, a certain number of symbols are prepended as pilot symbols, which can be used in the above five algorithms, which greatly improves the utilization rate of pilot symbols. The offline DSP program in the receiver is as follows: Figure 2 As shown, it includes resampling, matched filtering, IQ orthogonalization, chromatic dispersion compensation (CDC), clock recovery, single-tap MMA for polarization demultiplexing, FOE, CPR, Volterra nonlinear equalization (VNLE), multi-tap MMA for linear equalization, LMS / direct decision least mean square (DD-LMS) and bit error rate calculation. The received signal consists of a frame alignment sequence (FAS) and a series of QAM symbols that conform to the Maxwell-Boltzmann (mb) distribution. The preceding QAM symbol is used as a general pilot symbol. After clock recovery is completed, these pilot symbols will be used by the MMA, FOE, CPR and equalization modules.

[0074] It should be noted that when optimizing polarization demultiplexing, nonlinear equalization and linear equalization, only one code block is needed. When optimizing frequency offset estimation and carrier phase recovery, several code blocks are needed.

[0075] 2.1 Probability Shaping and Preamble Definition

[0076] Probability amplitude shaping technology has been widely used in optical communication transmission, improving transmission performance by changing the probability distribution of each level. Among them, probability amplitude shaping (PAS) using constant component distribution matching (CCDM) is the most commonly used. Given a string of random binary sequences, it is sent to CCDM and encoded according to the set amplitude probability, and finally a column of level symbols that meet a certain probability distribution will come out. When applied to QAM coherent optical communication, two independent probability shaping PAM signals are output, which are used as I-channel signals and Q-channel signals respectively, and then sent to the modulator for modulation. The Maxwell-Boltzmann (mb) distribution of the constellation points is optimized for linear additive white Gaussian noise (AWGN) channels, so in medium and short distance optical fiber transmission, that is, when the nonlinear effect is not strong, a better shaping gain can be obtained. For one of the PAM signals, the mb distribution expression is:

[0077]

[0078] Where P(x i ) is a single symbol x iThe probability of occurrence in the entire sequence, v is the shaping factor. The shaping parameter β is defined as β = H(P)-1, where H(·) represents the entropy of the PAM symbol. The shaping parameter of the overall signal can be changed by modifying v, and the transmission rate of the entire system is affected.

[0079] A group of symbols that satisfy the mb distribution are symmetrically mapped into PAM symbols through a symbol sequence, which becomes a code block to be transmitted (since two PAM signals (Q signals of I signals) enter the modulator for modulation, the modulation results in a QAM signal, so the Q signal of I signals is split into two and becomes a PAM signal, and combined into a QAM signal. For ease of description, this embodiment uses the QAM signal as a code block to be transmitted). During mapping, the first part of the symbols of a code block is taken as a pilot symbol to assist in equalization and carrier recovery. For a code block of length L, the pilot ratio is defined as R=P / L, where P is the number of pilot symbols. In addition, a more commonly used pilot is to replace part of the original QAM signal with a QPSK signal to obtain better phase tracking performance, but this pilot setting method will destroy the probability distribution of the overall code block, and the information ratio will decrease. Figure 3 The relationship between the shaping parameter β and the bit error rate (BER) of two different pilot setting methods under a 10% pilot ratio of the modulation format QAM1024 is shown. It is not difficult to find that when the bit error rate and shaping parameter β of pre-QAM and pre-QPSK are compared under a pilot ratio of 10%, the transmission performance of the pilot form of the pre-original QAM is better than that of the pre-QPSK without considering the additional overhead.

[0080] 2.2MMA Balance

[0081] The constant modulus algorithm (CMA) is a popular algorithm for demultiplexing polarization mode multiplexing, but it is only applicable to the case where the modulation signal has only one radius, such as QPSK, 8PSK, etc. When the modulation format is upgraded to a high-order QAM format (16QAM, 64QAM, 256QAM), or even an ultra-high-order QAM format (1024QAM, 4096QAM), CMA becomes invalid and the multi-mode algorithm (MMA) must be used. The algorithm principle block diagram is shown below. Figure 4 (a) As shown in the left half, the tap iteration formula is:

[0082]

[0083] Among them, p and q represent the X and Y polarizations, n is the current iteration number, μ is the step size, and e p (n) is the error of polarization p at the nth iteration, v p (n) is the output signal of polarization p at the nth iteration, u q(n) is the input signal of polarization q at the nth iteration, and the expression of error e is e=d-|v|, d is the amplitude that the output signal should be judged to at the current iteration number, and v is the output signal. Therefore, the MMA algorithm is often also called radius-directed decision (RDE).

[0084] One of the criteria for judging whether the MMA algorithm can effectively perform polarization mode demultiplexing is to judge the degree of contraction of the output signal and its corresponding standard radius. The radius judgment can be made using the geometric radius judgment method, that is, R d =(R i +R i+1 ) / 2, with R d As a benchmark, the decision is to R i or R i+1 The radius generated by the Rician distribution can also be used. The advantage is that the decision radius can be flexibly adjusted according to the additive white Gaussian noise (AWGN) of the current channel. Figure 4 (b) shows the Rician distribution density of each radius under 1024QAM. For the specific distribution formula, please refer to the literature. The advantage of the two blind decisions is that they can iteratively converge all the data of a whole block of codes, but the correctness of the decision radius cannot be guaranteed, especially in high-order modulation, the number of radii increases sharply (9 for 64QAM, 32 for 256QAM, and 109 for 1024QAM), and the distance between adjacent radii is extremely small, which will cause extremely serious misjudgment. For an entire code block that meets the mb distribution, this embodiment uses a small block as a known symbol sequence to avoid radius blind decision, and only uses the known pilot symbol sequence of the previous small block for tap training, and then applies it to the entire code block.

[0085] The convergence degree α can be used to reflect the degree of depolarization mode multiplexing, which is defined as α = (e x +e y ) / (e xp +e yp ), where e xp and e yp They represent the radius error before equalization, Figure 4 (c) shows the convergence comparison of different entropy values ​​using pre-pilot assistance and blind radius decision alone. It can be seen that the convergence can be reduced to below 0.8 by using pre-pilot assistance for polarization mode demultiplexing, and the best bit error effect can be obtained when the shaping parameter is 3.25, that is, the single polarization information entropy is 8.5. Both blind decision methods cannot achieve effective convergence, and the use of Rician radius can achieve certain convergence performance when the shaping parameter is small.

[0086] It is worth mentioning that, in order to avoid introducing new equalization enhanced phase noise (EEPN) as much as possible, this embodiment uses only one-order tap in the previous MMA equalization, so it is only a rough demultiplexing of polarization mode. Therefore, this embodiment introduces multi-tap MMA equalization between nonlinear equalization and linear equalization to solve the problem of imbalance between X polarization and Y polarization. For multi-tap MMA equalization, this embodiment uses 33-order tap coefficients and sets H pq Separate the real and imaginary parts of the signal, each H pq Disassemble into H pq-ab (a and b can both represent r and i), the original 2×2 MIMO is converted to 16×4 MIMO for iterative equalization. The iterative formula is:

[0087] H pq-ab (n+1)=H pq-ab (n)+μ·e p-a (n) ·u q-b (n) (2-3)

[0088] Among them, p and q represent the X and Y polarizations, a and b represent the real and imaginary parts of the QAM signal, n is the current iteration number, μ is the step size, and e is the real part. p-a (n) is the error of the a part of polarization p at the nth iteration, u q-b (n) is the input signal of part b of polarization q at the nth iteration.

[0089] Figure 4 (a) The right half shows the block diagram converted into real and imaginary form, which will be used to compensate for the imbalance of signal X, Y polarization and the imbalance of I, Q in each polarization. Figure 4 (d) shows the comparison of bit error rate and bit error imbalance before and after compensation at a pilot ratio of 5%. The XY imbalance is expressed as the bit error difference between the two polarizations and can be expressed by the formula. The 33-tap MMA equalization with pilot assistance has better bit error performance than that without it, and as the shaping parameters increase, the error between the two polarizations will decrease, but the use of a multi-tap equalizer can significantly reduce the XY deviation.

[0090] 2.3 Frequency Offset Estimation

[0091] The traditional blind FOE algorithm applied to QAM is derived from the FO estimation of the M-ary phase shift keying (M-PSK) modulation format. The discrete spectrum of the symbols of the M-PSK constellation produces a peak at MΔf after the M-power operation. For the square M-QAM signal, the two diagonals have the most points with the same phase, so the discrete spectrum of the M-QAM signal after the fourth power produces a peak at 4Δf. However, probability shaping will disrupt the original square of the constellation points, and the constellation points on the diagonals will be sparse, which will eventually lead to the smaller the shaping parameter β, the more the constellation points will appear to be the center of the circle, and the spectrum after the fourth power will have multiple peaks, which seriously affects the frequency offset estimation. FOE can also be performed with the assistance of pilots. When receiving the signal y n (Polarization mode multiplexing demultiplexed signal) Extract pilot signal p′ n , (due to the pilot signal p′ n After the transmission channel response, it and the original pilot symbol p n is not the same), and then with the transmitted pilot signal p n The complex conjugate Multiply and remove the modulation phase information, so that the discrete spectrum of the signal can be easily found in Δf p However, the frequency offset of the signal will change with time, and the frequency offset estimated in this way will produce errors when applied to the entire code block, so this embodiment combines the two methods to estimate the optimal frequency offset.

[0092] Figure 5 The block diagram of the algorithm combining the fourth power FFT and pilot assisted FFT to find the frequency offset is shown. n Extract pilot signal p′ n and the transmitted signal p n The complex conjugate After multiplication, calculate the FFT and find the index of the maximum value I p The pilot-assisted positioning frequency offset is

[0093] Δf p =R S I p / N FFT (2-4)

[0094] Where R S is the symbol rate of the signal, N FFT is the number of transformation points. p ±Δf i is the range for searching for precise frequency deviation, which corresponds to y n The index range of the fourth power of FFT is [(Δf p -Δf i )·4N FFT / R S ,(Δfp +Δf i )·4N FFT / R S ]. The index of the maximum value found within this range is I, and the final estimated frequency offset is

[0095] Δf=R S I / 4N FFT (2-5)

[0096] Then frequency offset compensation is applied n For each QAM symbol, the formula is

[0097]

[0098] Where T s is the symbol period, y n,i and z n,i are the i-th symbol in a code block in the input and output respectively.

[0099] 2.4 Carrier Phase Recovery

[0100] In traditional coherent optical systems, the carrier phase recovery algorithm for M-QAM mainly adopts the blind phase search (BPS) algorithm and the improved blind phase search (BPS / ML) algorithm based on maximum likelihood (ML). Although BPS has a good phase recovery effect for uniform low-order QAM, even if two-stage BPS is used, the complexity of the algorithm is quite high, and it also faces the dual influence of high-order QAM and probability shaping. The correctness of the search phase cannot be guaranteed at all. Since the change rate of the laser phase noise is much slower than other additive noises, such as spontaneous radiation noise, the phase noise of the laser can be estimated through the pilot information of continuous short code blocks. High baud rate signals can improve the phase noise tolerance of wide linewidth lasers, but after using pilot symbols, there is good phase tracking continuity even at low baud rates.

[0101] This embodiment again uses the leading pilot symbol p′ of each code block. n The complex conjugate of the transmitted signal Multiply and remove the modulation phase to obtain the offset phase θ of the pilot symbol r . Figure 6 The algorithm flow of using pilot-assisted phase offset is shown. Take m code blocks, and the average phase offset of the pilot symbol of each code block is θ ri (i=1,2,..,m), and zero the remaining phases of the input signal after frequency offset estimation, and perform smoothing (unwrap) processing. Then connect each θ riThe approximate phase shift trend is obtained, and finally three interpolations are performed to obtain the phase shift of the entire signal, and the obtained phase shift of the entire signal is compensated to the entire signal.

[0102] 2.5 Nonlinear Equalization and Linear Equalization

[0103] The nonlinear distortion caused by optoelectronic devices during transmission, such as the nonlinear response of the I / Q modulator, the square law detection of the photodetector (PD), and the nonlinearity of the amplifier, needs to be compensated by nonlinear equalization. This embodiment uses the Volterra nonlinear equalization algorithm (VNLE) based on the second-order Volterra series. The principle is as follows:

[0104]

[0105] The above formula is the non-recursive series expansion expression of Volterra equalizer, where f 0i is the tap coefficient of the i-th order filter, l i is the memory length of the i-th order filter, and r(k) is the k-th symbol input to the VNLE. The tap coefficients of the VNLE are updated based on the recursive least squares (RLS) scheme, and the training sequence also uses a universal pre-pilot. The block diagram of the VNLE is shown in Figure 7 In this embodiment, the pilot symbols are used to train the tap coefficients, and their updating is based on the recursive least squares method (RLS).

[0106] Although there is a first-order linear equalizer based on RLS in VNLE, since it is to equalize the I and Q signals of each polarization state separately in the real domain, it will cause the combined constellation points to diverge regularly to a certain extent. Therefore, it is necessary to connect another level of linear equalization to achieve constellation point aggregation. The usual linear equalization algorithm is a constellation aggregation algorithm based on the minimum mean square error (LMS), and its tap iteration formula is:

[0107] ω(n+1)=ω(n)+μ LMS ·e LMS (n) ·u LMS * (n) (2-8)

[0108] Among them, n is the current iteration number, μ LMS is the step length, e LMS (n) is the error of the nth iteration, u LMS (n) is the input signal of the nth iteration.

[0109] Compared with MMA, the iterative formula here cancels the joint equalization of the two polarizations and changes the error from a scalar to a vector. In this embodiment, the taps trained with the aid of pilots are first applied to the input signal, and then the direct decision-minimum mean square error algorithm (DD-LMS) is used for blind equalization. The error performance of the signal can be further improved while the pilot is effectively used and the direction of the constellation point aggregation is correct. The error vector magnitude (EVM) can be used to measure the degree of constellation point aggregation:

[0110]

[0111] Among them, E k,i is the ideal constellation point vector, E k,r is the received signal vector after complete demodulation. Figure 8 The EVM curves of pilot-assisted LMS equalization, DD-LMS equalization, and the cascaded use of the two under different OSNR conditions are shown in the 1024QAM modulation format with a shaping parameter β of 3.25. It can be seen that the effect of using only DD-LMS equalization is very poor. The reason is that when the constellation points are not yet clustered, using the minimum Euclidean distance to judge the corresponding points will produce most of the misjudgments, while the effect of using only pilot-assisted LMS equalization is steadily decreasing with the increase of OSNR. When the OSNR is higher than 29dB, the equalization effect of the cascaded combination of the two is better than that of using only pilot-assisted LMS, but when the OSNR is lower than this threshold, DD-LMS begins to produce misjudgments and the equalization effect drops sharply. Therefore, when dealing with different OSNR conditions, this embodiment will use the most appropriate equalization method.

[0112] 3. Experimental Setup

[0113] The experimental setup for ultra-high-order PS-PDM-QAM transmission over 80 km of standard single-mode fiber (SSMF) is shown in Figure 1. Fig. 9 As shown. The center frequency of the micro-integrable tunable laser assembly (μITLA) used in the experiment is 1550.112nm, and the line width is less than 40kHz. The signal generated by the 8GSa / s two-channel 14-bit arbitrary waveform generator (AWG) passes through a root raised cosine filter (RRC) with a roll-off factor of 0.05, and the upsampling multiple is 2. The output voltage of the two 4GBd positive channel signals is 350mV, which are respectively sent to the XI and XQ of the in-phase / quadrature (I / Q) modulator through a 30cm cable. In order to ensure that the signal power is absolutely stable during the link transmission process, this embodiment abandons the use of a polarization multiplexing simulator (PDME), and instead the negative channel is sent to the YI and YQ of the I / Q modulator through two 2m cables (the number of delay symbols is 30 at 4GBd) to realize polarization mode multiplexing.

[0114] The signal from the modulator is amplified by an erbium-doped fiber amplifier (EDFA) and then enters a variable optical attenuator (VOA) to adjust the transmission power entering the transmission link. The link uses a G652D standard single-mode fiber (SSMF) with a length of 80km, an average loss of 0.18dB / km, and a dispersion of 16.09ps / nm / km. Under a noise reference bandwidth of 0.1nm, the received optical signal-to-noise ratio (OSNR) after transmission of 0km and 80km is 40.57dB and 36.03dB respectively. The received signal is amplified by the EDFA and filtered by a 1nm optical bandpass filter (OBPF). The local oscillator (LO) light source uses the same model as the transmitter for local detection. Then, the signal is received by an optical coherent receiver consisting of a polarization and phase distributed fiber hybrid and four balanced broadcast diodes (BPDs) with a bandwidth of 33GHz. The received signal is digitized by a 33GHz bandwidth digital storage oscilloscope (DSO) at a sampling rate of 50GSa / s and finally sent to an offline computer for offline digital signal processing (DSP).

[0115] 4. Experimental results and discussion

[0116] This embodiment uses ideal rate adaptive forward error correction coding (FEC) to achieve system performance evaluation. First, the generalized mutual information (GMI) in two polarization states can be estimated based on the received symbols as

[0117]

[0118] Where m is the number of bits corresponding to a single PAM symbol (1024QAM corresponds to m is 5, 4096QAM corresponds to m is 6), b k,i is the i-th bit of the k-th transmitted symbol, y k is the kth received symbol, N is the number of received symbols, Indicates that in the case of 2 The probability density function of the received symbol y given the transmitted symbol x in a memoryless additive white Gaussian noise (AWGN) auxiliary channel with a one-dimensional noise variance channel. GMI is used to express the maximum achievable information rate (AIR) under the AWGN channel, usually indicating the number of information bits per symbol. The pilot overhead with a pilot ratio of R is R / (1-R), and AIR can be described as

[0119]

[0120] For a given output, a higher R improves the quality of the signal, but it also reduces the information entropy of the output, which requires a certain trade-off.

[0121] For probabilistically shaped signals of different entropy values, the normalized generalized mutual information (NGMI) can be used to perform channel metrics, and a given soft-decision forward error correction coding (SD-FEC) scheme can be evaluated regardless of the modulation format. The expression of NGMI is as follows

[0122]

[0123] This embodiment selects 25% LDPC overhead, and the corresponding NGMI threshold is 0.828. No matter what modulation format the signal is in, if the NGMI exceeds this threshold after being transmitted through the optical fiber, error-free transmission can be achieved.

[0124] By changing the shaping factor v in formula 2-1, the shaping parameter β for transmitting 1024QAM signal is adjusted from 3.05 to 3.95, and the shaping parameter β for transmitting 4096QAM signal is adjusted from 4.05 to 4.95. Fig.10 The relationship between the shaping parameter β and the measured NGMI when the pilot ratio is 5% for 4GBaud PS-PDM-1024 / 4096QAM signals are transmitted back-to-back is given. Fig.10 (a) Regarding 1024QAM, smaller β can meet the 0.828-NGMI threshold. When β exceeds 3.25, the system transmission performance begins to decline rapidly. Fig.10 (b) About 4096QAM. When β does not exceed 4.35, the 0.828-NGMI threshold can be met. This embodiment will use a 1024QAM signal with a β of 3.25 and a 4096QAM signal with a β of 4.35 to evaluate the performance of the pilot ratio R during transmission. Taking into account the roll-off factor of the frequency band 0.05, the spectrum efficiency SE is 16.191bit / s / Hz when β is 3.25, and the spectrum efficiency SE is 20.381bit / s / Hz when β is 4.35. Considering that 1024QAM is still some distance away from the NGMI threshold of 0.828 when β is 3.25, FEC coding with lower overhead can be used to continue to improve the spectrum efficiency.

[0125] exist Fig.11 In the paper, the comparison of AIR and OSNR under different pilot ratios R when a single-carrier high-order QAM modulation format is transmitted over 80km of standard single-mode fiber (SSMF) using PS-PDM technology is shown. Pilot symbols are used in various equalization modules. The AIR shows different performance results with the change of pilot ratio R when transmitting 1024QAM with β of 3.25 and 4096QAM with β of 4.35. For both modulation formats, no matter how much R is, when OSNR exceeds a certain value, the system performance reaches saturation, but the saturation point of 1024QAM is more advanced. Fig.11(a) For 1024QAM, at low OSNR points, 10% pilots have better equalization performance, but have no advantage over 5% pilots in terms of total AIR. After the saturation point, 10% pilots require higher pilot overhead, so 5% pilots have an improvement of 0.9 bit / symbol / 2-pol in AIR compared to 10% pilots. Fig.11 (b) For 1024QAM, as OSNR increases, AIR rises more smoothly. The best pilot ratio for the two different modulation formats is around 5%, and the transmission performance improves by 2-3 dB when the ratio increases from 2% to 5%.

[0126] Since the tolerance of 4GBand signal to the laser line width is relatively poor, this embodiment designs a comparison test of 100 Hz narrow line width and 40 kHz wide line width. Fig.12 The relationship between OSNR and NGMI of PS-PDM-1024QAM under 10% pilot ratio transmission over 80km SSMF under two linewidth conditions is shown. Before OSNR enters the saturation region, the 100Hz linewidth laser has only 1dB more gain than the 100kHz linewidth, and can exceed the 0.828NGMI threshold when OSNR exceeds 22dB. Therefore, when the pilot ratio R reaches the equilibrium saturation point, the use of pilot-assisted equalization algorithm for transmitting ultra-high-order PS-PDN-QAM signals is not necessary for narrow linewidth lasers.

[0127] It can be seen that this embodiment proposes a general pilot-assisted DSP structure for PS-PDM high-order QAM coherent optical communication, which has been experimentally verified in PS-PDM-1024 / 4096QAM signals transmitted at 80 km SSMF. The general pilot symbol comes from the partial probability shaped QAM symbol instead of inserting additional QPSK symbols at the transmitter. Since the mb distribution of the entire symbol sequence remains unchanged, the proposed method can improve the spectral efficiency according to the higher shaping parameters under the expected bit error rate performance. In the entire DSP structure, the general pilot symbol is used for polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization. For PS-PDM-1024QAM and 4096QAM signals, the optimal pilot ratio of 5% can make the original SE reach 16.191 and 20.381bit / s / Hz respectively. Based on the proposed fast convergence DSP module, the narrow linewidth laser can be replaced by the traditional ECL, thereby reducing the cost of the ultra-high-order QAM coherent optical communication system.

[0128] Embodiment 2

[0129] This embodiment provides a DSP system for a QAM coherent optical transmission system, including:

[0130] A construction module: used for receiving a QAM signal and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution;

[0131] Application module: used for applying the same pilot symbol with the same pilot ratio to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the DSP process during the DSP process of the QAM signal.

[0132] Although the preferred embodiments of the present application have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.

[0133] Obviously, the above embodiments are merely examples for clear explanation and are not intended to limit the implementation methods. For those skilled in the art, other different forms of changes or modifications can be made based on the above description. It is not necessary and impossible to list all the implementation methods here. The obvious changes or modifications derived from these are still within the protection scope of the invention.

Claims

1. A DSP method for QAM coherent optical communication, Features: include: Step S1: receiving a QAM signal, and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution; Step S2: in the DSP process, the same pilot symbol is applied to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the DSP process with the same pilot ratio for the QAM signal; In step S1, a pilot symbol is constructed according to the received QAM signal, and the method includes: Taking the QAM signal as a code block to be transmitted, taking a preset number of symbols in front of the code block as pilot symbols; The pilot ratio formula is: R=P / L Where L is the length of the code block and P is the number of pilot symbols; In step S2, the same pilot symbol is applied to polarization demultiplexing of a DSP process with the same pilot ratio, and the method includes: By taking a preset number of symbols in front of a code block as a known symbol sequence, a radius blind decision is avoided when the polarization mode multiplexing of the QAM signal is demultiplexed by the MMA algorithm, the MMA algorithm is tapped and trained by using the known symbol sequence, and then the MMA algorithm is applied to the entire code block; Among them, the MMA algorithm uses one tap and the tap iteration formula of the MMA algorithm is: Among them, p and q represent the X and Y polarizations, n is the current iteration number, μ is the step size, and e p (n) is the error of polarization p at the nth iteration, v p (n) is the output signal of polarization p at the nth iteration, u q (n) is the input signal of polarization q at the nth iteration, the * sign indicates the conjugation operation, the expression of the error e is e=d-|v|, d is the amplitude that the output signal should be judged to at the current iteration number, and v is the output signal; In step S2, the same pilot symbol is applied to polarization demultiplexing of the DSP process with the same pilot ratio, and the method further comprises: Use 33 tap coefficients between nonlinear equalization and linear equalization, and change H pq Separate the real and imaginary parts of the signal, and the iteration formula is: H pq-ab (n+1)=H pq-ab (n)+μ·e p-a (n)·u q-b (n) Among them, p and q represent the X and Y polarizations, a and b represent the real and imaginary parts of the QAM signal, n is the current iteration number, μ is the step size, and e is the real part. p-a (n) is the error of the a part of polarization p at the nth iteration, u q-b (n) is the input signal of part b of polarization q at the nth iteration; In step S2, the same pilot symbol is applied to the frequency offset estimation of the DSP process with the same pilot ratio, and the method includes: The frequency offset is estimated by combining the fourth power FFT and the pilot assisted FFT. Specifically, the input signal y after depolarization mode multiplexing is n Extract pilot signal p′ n and the complex conjugate of the transmitted signal p n * After multiplication, calculate the FFT and find the index of the maximum value I p , the frequency offset to which the pilot symbol is located is: Δf p =R S ·I p / N FFT Among them, R S is the symbol rate of the signal, N FFT is the number of transformation points, Δf p ±Δf i is the range for searching for precise frequency deviation, which corresponds to y n The index range of the fourth power of FFT is [(Δf p -Δf i )·4N FFT / R S ,(Δf p +Δf i )·4N FFT / R S ], the index of the maximum value found in the index range is I, and the final estimated frequency offset is: Δf=R S ·I / 4N FFT Apply frequency offset compensation to y n For each QAM symbol, the formula is: Among them, T s is the symbol period, y n,i and z n,i are the i-th symbol in a code block in the input and output respectively; In step S2, the same pilot symbol is applied to the carrier phase recovery of the DSP process with the same pilot ratio, and the method includes: Take m code blocks, the average offset phase of the pilot symbol of each code block is θ ri , i=1,2,..,m, and zero the remaining phases of the input signal after frequency offset estimation, and perform smoothing and decomposition processing at the same time, then connect each to obtain the phase offset trend, and finally perform three interpolations to obtain the phase offset of the entire signal, and compensate the obtained phase offset of the entire signal to the entire signal; In step S2, the same pilot symbol is applied to the nonlinear equalization of the DSP process with the same pilot ratio, and the method includes: The nonlinear equalization is achieved using the Volterra nonlinear equalization algorithm based on the second-order Volterra series. The formula is: The above formula is the non-recursive series expansion expression of Volterra equalizer, where f 0i is the tap coefficient of the i-th order filter, l i is the memory length of the i-th order filter, r(k) is the k-th symbol input to the Volterra nonlinear equalization algorithm; Training the tap coefficients of the Volterra nonlinear equalization algorithm through the pilot symbols, wherein the tap coefficients of the Volterra nonlinear equalization algorithm are updated based on a recursive least squares scheme; In step S2, the same pilot symbol is applied to the linear equalization of the DSP process with the same pilot ratio, and the method includes: Linear equalization is achieved by a constellation clustering algorithm based on minimum mean square error. The tap iteration formula of the constellation clustering algorithm based on minimum mean square error is: ω(n+1)=ω(n)+μ LMS ·e LMS (n)·u LMS * (n) Among them, n is the current iteration number, μ LMS is the step length, e LMS (n) is the error of the nth iteration, u LMS (n) is the input signal of the nth iteration, and the * sign indicates the conjugate operation.

2. A DSP system for QAM coherent optical communication, Features: include: A construction module: used for receiving a QAM signal and constructing a pilot symbol according to the received QAM signal, wherein the QAM signal is probability shaped and conforms to the Maxwell-Boltzmann distribution; Application module: used for applying the same pilot symbol with the same pilot ratio to polarization demultiplexing, frequency offset estimation, carrier phase recovery, nonlinear equalization and linear equalization of the DSP process during the DSP process of the QAM signal; The construction module constructs a pilot symbol according to the received QAM signal, including: Taking the QAM signal as a code block to be transmitted, taking a preset number of symbols in front of the code block as pilot symbols; The pilot ratio formula is: R=P / L Where L is the length of the code block and P is the number of pilot symbols; The application module applies the same pilot symbol to polarization demultiplexing of the DSP process with the same pilot ratio, including: By taking a preset number of symbols in front of a code block as a known symbol sequence, a radius blind decision is avoided when the polarization mode multiplexing of the QAM signal is demultiplexed by the MMA algorithm, the MMA algorithm is tapped and trained by using the known symbol sequence, and then the MMA algorithm is applied to the entire code block; Among them, the MMA algorithm uses one tap and the tap iteration formula of the MMA algorithm is: Among them, p and q represent the X and Y polarizations, n is the current iteration number, μ is the step size, and e p (n) is the error of polarization p at the nth iteration, v p (n) is the output signal of polarization p at the nth iteration, u q (n) is the input signal of polarization q at the nth iteration, the * sign indicates the conjugation operation, the expression of the error e is e=d-|v|, d is the amplitude that the output signal should be judged to at the current iteration number, and v is the output signal; The application module applies the same pilot symbol to polarization demultiplexing of the DSP process with the same pilot ratio, and further includes: Use 33 tap coefficients between nonlinear equalization and linear equalization, and change H pq Separate the real and imaginary parts of the signal, and the iteration formula is: H pq-ab (n+1)=H pq-ab (n)+μ·e p-a (n)·u q-b (n) Among them, p and q represent the X and Y polarizations, a and b represent the real and imaginary parts of the QAM signal, n is the current iteration number, μ is the step size, and e is the real part. p-a (n) is the error of the a part of polarization p at the nth iteration, u q-b (n) is the input signal of part b of polarization q at the nth iteration; The application module applies the same pilot symbol with the same pilot ratio to the frequency offset estimation of the DSP process, including: The frequency offset is estimated by combining the fourth power FFT and the pilot assisted FFT. Specifically, the input signal y after depolarization mode multiplexing is n Extract pilot signal p′ n and the complex conjugate of the transmitted signal p n * After multiplication, calculate the FFT and find the index of the maximum value I p , the frequency offset to which the pilot symbol is located is: Δf p =R S ·I p / N FFT Among them, R S is the symbol rate of the signal, N FFT is the number of transformation points, Δf p ±Δf i is the range for searching for precise frequency deviation, which corresponds to y n The index range of the fourth power of FFT is [(Δf p -Δf i )·4N FFT / R S ,(Δf p +Δf i )·4N FFT / R S ], the index of the maximum value found in the index range is I, and the final estimated frequency offset is: Δf=R S ·I / 4N FFT Apply frequency offset compensation to y n For each QAM symbol, the formula is: Among them, T s is the symbol period, y n,i and z n,i are the i-th symbol in a code block in the input and output respectively; The application module applies the same pilot symbol to carrier phase recovery of the DSP process with the same pilot ratio, including: Take m code blocks, the average offset phase of the pilot symbol of each code block is θ ri , i=1,2,..,m, and zero the remaining phases of the input signal after frequency offset estimation, and perform smoothing and decomposition processing at the same time, then connect each to obtain the phase offset trend, and finally perform three interpolations to obtain the phase offset of the entire signal, and compensate the obtained phase offset of the entire signal to the entire signal; The application module applies the same pilot symbol to the nonlinear equalization of the DSP process with the same pilot ratio, including: The nonlinear equalization is achieved using the Volterra nonlinear equalization algorithm based on the second-order Volterra series. The formula is: The above formula is the non-recursive series expansion expression of Volterra equalizer, where f 0i is the tap coefficient of the i-th order filter, l i is the memory length of the i-th order filter, r(k) is the k-th symbol input to the Volterra nonlinear equalization algorithm; Training the tap coefficients of the Volterra nonlinear equalization algorithm through the pilot symbols, wherein the tap coefficients of the Volterra nonlinear equalization algorithm are updated based on a recursive least squares scheme; The application module applies the same pilot symbol to the linear equalization of the DSP process with the same pilot ratio, including: Linear equalization is achieved by a constellation clustering algorithm based on minimum mean square error. The tap iteration formula of the constellation clustering algorithm based on minimum mean square error is: ω(n+1)=ω(n)+μ LMS ·e LMS (n)·u LMS * (n) Among them, n is the current iteration number, μ LMS is the step length, e LMS (n) is the error of the nth iteration, u LMS (n) is the input signal of the nth iteration, and the * sign indicates the conjugate operation.