CN103780519B - A Joint Parallel Method for Channel Equalization and Frequency Offset Estimation Based on LMS - Google Patents

Abstract

The invention discloses a combined parallel method of channel equalization and frequency offset estimation based on LMS. In the initialization stage of the optical receiver, the training sequence signal is converted into a parallel signal and then sent to a parallel signal processing branch for equalization and frequency offset estimation respectively. Calculate the average value of the estimated frequency offset of all branches and the error signal of each branch. Each group of parallel signals adopts a unified equalizer tap coefficient, and the mean value of the branch error signal is used when the equalizer tap coefficient is updated; it is sent in the data transmission stage The end inserts training symbols into the data symbols, and the optical receiver converts the data signals into parallel signals. The training signal uses the equalized signal and known training symbols for frequency offset estimation, and the data signal uses the accumulated phase error of the corresponding training signal and the obtained frequency The average value of the offset estimation is compensated and then judged, and then the equalized signal and the judgment signal are used to estimate the frequency offset. The invention adopts signal parallel processing to reduce the limited influence of data signal processing hardware on system performance.

Description

A Joint Parallel Method for Channel Equalization and Frequency Offset Estimation Based on LMS

technical field

The invention belongs to the technical field of optical burst receivers, and more specifically relates to a joint parallel method of channel equalization and frequency offset estimation based on LMS (Least Mean Square, least mean square algorithm).

Background technique

In the current high-speed coherent optical communication system, PDM-QPSK (Polarization Multiplexing-Quadrature Absolute Phase Shift Keying) coherent optical transmission system is one of the most potential technical solutions. In the PDM-QPSK coherent optical transmission system, the transmission signal is mainly affected by the linear damage of the optical fiber's chromatic dispersion (CD) and polarization mode dispersion (Polarization Mode Dispersion, PMD) and the frequency offset generated by the laser at the transceiver end , these two problems seriously affect the performance of the optical receiver. The adaptive equalization technology can basically eliminate the intersymbol interference caused by dispersion, and the frequency offset estimation technology can be used to solve the impact of frequency offset. Since the equalizer and frequency offset estimator will influence each other, a joint algorithm of time domain equalization and frequency offset estimation can be used.

For the optical burst transmission system, the characteristics of the optical burst require the equalizer in the optical burst receiver to be able to achieve rapid convergence. Figure 1 is a system block diagram of an optical burst receiver. As shown in Fig. 1, the input signal r(t) of the receiver is the signal of two optical PDM-QPSK signals whose polarization directions are perpendicular to each other through polarization coupling and transmitted through a certain distance of optical fiber channel. During fiber channel transmission, optical signals will be affected by factors such as dispersion, polarization mode dispersion, and optical amplifier noise, resulting in a decrease in the quality of the transmitted signal. The PDM-QPSK signal r(t) enters the 90-degree mixer together with the FTLO (fast tunable laser) light wave for coherent demodulation. The four signals after coherent demodulation are subjected to AD sampling and quantization. The 4-way signals Ix, Qx, Iy, and Qy output after sampling and quantization respectively represent the in-phase and quadrature modulation signals of the two polarization states x, y, and these 4-way signals enter the digital signal processing module for channel equalization (Channel Equalization) and Frequency Offset Estimation (Frequency OffsetEstimation, FOE) and compensation, and finally the phase judgment restores the sent data.

The joint algorithm of channel equalization and frequency offset estimation based on LMS is an effective method to improve the performance of coherent optical receiver. However, in the design of optical burst receivers, when using FPGA (Field-Programmable Gate Array, Field Programmable Gate Array) or application-specific integrated circuits to implement digital signal processing algorithms, calculation speed and chip area are two major issues that restrict each other. . Therefore, it is necessary to choose between performance and implementation complexity. Due to the high-speed characteristics of optical fiber communication, taking the 112Gb/s PDM-QPSK optical fiber transmission system as an example, the symbol rate of each of the four signals after coherent demodulation is 28G/s. Double-rate AD sampling and quantization, the signal rate of each branch is as high as 56G/s, so the symbols entering the equalizer are discrete signals with a symbol rate of 56G/s. The subsequent digital signal processing unit (DSPU) cannot process this rate on the hardware, so parallel processing must be used. According to the rate of input data and the processing speed of the chip, the number of parallel branches may use a larger value , which requires that the algorithm must meet the requirements of parallel processing in real-time applications. At the same time, the update of the tap coefficients of the equalizer and the frequency offset estimation algorithm based on pre-decision both require signal feedback, and the time delay caused by feedback in parallel implementation also has a great impact on the performance of the system. Therefore, in the specific implementation design of the optical burst receiver, the impact of parallelism and feedback delay on receiver performance must also be considered.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies of the prior art, provide a joint parallel method of channel equalization and frequency offset estimation based on LMS, and reduce the restriction influence of data signal processing hardware on system performance.

In order to achieve the above-mentioned purpose of the invention, the joint parallel method of channel equalization and frequency offset estimation based on LMS of the present invention comprises the following steps:

S1: Use the training sequence for initialization, including steps:

S1.1: Send the training sequence to the optical burst receiver, perform serial-to-parallel conversion on the training sequence signal after coherent demodulation and sampling and quantization to obtain N parallel signals; set the equalizer tap coefficient corresponding to the n=1 group of parallel signals

S1.2: The nth group of parallel signals enters N parallel signal processing branches, each parallel signal processing branch includes an equalizer and a frequency offset estimation module, and the equalizer of the i-th branch obtains an equalized signal Where k=(n-1)×N+i, 1≤i≤N;

S1.3: Frequency offset estimation module based on known training symbols For balanced signal Perform frequency offset estimation to obtain the cumulative phase error and frequency offset estimates

S1.4: Calculate the estimated frequency offset of N branches Perform averaging to obtain the average value of the frequency offset estimation of the nth group of parallel signals

S1.5: N branches respectively calculate their error signals ε _n,i :

S1.6: Update the equalizer tap coefficients used by the n+1th group of parallel signals:

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}& \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; D.</span>}\\ {\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}<span\; class="notranslate">\; \&Center\; Dot;</span>\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}<span\; class="notranslate">\; \&Center\; Dot;</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&rsqb;</span>& \mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; D.</span>}\end{array}\right.$

in, Respectively represent the equalizer tap coefficients used by the n+1th group and the nth group of parallel signals; D represents the delay of the error signal; λ is the set iteration step size, which is a positive number; N _c represents the selection from N branches The number of error signals involved in the calculation of tap coefficients, 1≤N _c ≤N, 1≤i _c ≤N _c ; express The corresponding observation vector, express the conjugate;

S1.7: Determine whether the training sequence has been processed. If not, return to step S1.2 to continue processing the next group of parallel signals. If processed, proceed to step S2;

S2: Enter the data sending stage to process the data, including steps:

S2.1: The data sending end inserts training symbols into the data symbols. The insertion method is: take N sending symbols as a group, and then divide the N sending symbols into R groups, and in each group N/R sending symbols Contains a training symbol, and the sequence number of R training symbols in parallel symbols is denoted as i _r , 1≤r≤R;

S2.2: Send the data signal to the optical burst receiver, perform serial-to-parallel conversion on the data signal after coherent demodulation and sampling and quantization to obtain N parallel signals, the nth group of parallel signals enters N parallel signal processing branches, and the data The parallel signal processing branch in the sending stage includes an equalizer, a frequency offset estimation module and a decision module, and the equalizer processing of the i-th branch obtains an equalized signal

S2.3: Perform frequency offset estimation on N branches respectively:

When the branch is the training signal, directly based on the known training symbols For balanced signal Perform frequency offset estimation to obtain the cumulative phase error and frequency offset estimates

When the branch is a data signal, first balance the signal Perform phase compensation, the signal after phase compensation for:

where d represents the delay of the frequency offset estimation mean value, Indicates rounding up; the judgment module Make a decision to get a decision signal According to the decision signal For balanced signal Perform frequency offset estimation to obtain frequency offset estimation value and cumulative phase error

S2.4: Calculate the frequency offset estimates of N branches Perform averaging to obtain the average value of the frequency offset estimation of the nth group of parallel signals

S2.5: N branches respectively calculate their error signals ε _n,i :

When the branch is the training signal, i=i _r , the error signal ε _n,i is:

When the branch is a data signal, the error signal ε _n,i is:

S2.6: Update the equalizer tap coefficients used by the n+1th group of parallel symbols:

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}^{<span\; class="notranslate">\; *</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&rsqb;</span>$

Among them, λ ^* is the iteration step size set in the data sending stage, Indicates the number of error signals selected from the N branches in the data transmission stage to participate in the calculation of the tap coefficients,

S2.7: Determine whether the data has been processed, if not, return to step S2.2 to continue processing, and end if processed.

Further, the specific method of frequency offset estimation includes the following steps:

S3.1: Calculate the equalized signal The cumulative phase error of :

When the branch is a training symbol, the accumulated phase error where θ _k denotes the phase of the known training symbols and Φ _k denotes the equalized signal the phase of

When the branch is a data signal, the accumulated phase error in Indicates data decision signal the phase of

S3.2: Calculate the estimated value of the frequency offset of this branch in Indicates the cumulative phase error of the k-1th signal.

The combined parallel method of channel equalization and frequency offset estimation based on LMS in the present invention converts the sampling signal of the training sequence signal into a parallel signal through serial-to-parallel conversion in the initialization stage of the optical receiver, and then sends it to the parallel signal processing branch. Equalization and frequency offset estimation are performed on each branch, and the frequency offset estimates obtained by all branches are averaged to obtain the average frequency offset estimate. Each branch calculates its error signal according to the average frequency offset estimate. The equalizer tap coefficient is updated by the mean value of the branch error signal when the equalizer tap coefficient is updated; in the data sending stage, the sending end inserts training symbols into the data symbols, and passes the coherently demodulated and sampled and quantized data signals through serial And transform to obtain parallel signals, the training signal uses the equalized signal and known training symbols for frequency offset estimation, the data signal uses the accumulated phase error of the corresponding training signal and the obtained average value of frequency offset estimation to compensate and then judge, and then uses the equalized signal The frequency offset is estimated with the decision signal, and then the equalizer tap coefficients are updated in the same way as in the initialization stage.

The present invention has the following beneficial effects:

(1) The present invention reduces the signal rate through parallelization, thereby reducing the limitation impact of data signal processing hardware on system performance;

(2) In the initialization stage, the error signal is calculated by using the estimated average value of the frequency offset, which can improve the reliability of the equalizer initialization;

(3) In the data transmission stage, inserting a certain number of training symbols in parallel symbols can improve the accuracy of judgment, frequency offset estimation and error signal feedback;

(4) The simulation shows that the present invention has good tolerance to the delay of the error signal.

Description of drawings

Fig. 1 is a system block diagram of an optical burst receiver;

Fig. 2 is a schematic diagram of parallel signal branch algorithm in the initialization stage;

Fig. 3 is a schematic diagram of a parallel signal branch algorithm in the data transmission stage;

Fig. 4 is the comparison diagram of the equalizer convergence speed of parallel method and serial method of the present invention;

Fig. 5 is a comparison diagram of the convergence speed with and without calculation delay in the parallel method of the present invention;

Fig. 6 is a comparison chart of bit error rates between the serial method and the parallel method under different delays of the present invention.

detailed description

Specific embodiments of the present invention will be described below in conjunction with the accompanying drawings, so that those skilled in the art can better understand the present invention. It should be noted that in the following description, when detailed descriptions of known functions and designs may dilute the main content of the present invention, these descriptions will be omitted here.

Example

This embodiment still takes the 112Gb/s PDM-QPSK optical fiber transmission system as an example, and converts the 56G/s rate signal into multiple parallel signals through serial parallel conversion, here is 256 channels, so that the rate of each channel can be effectively reduced , which can reduce the difficulty of its physical realization. Since the sampling is doubled, each group of 256 signals will be determined to obtain 128 symbols. It can be seen that the number of parallel processing branches to be configured is N=128.

The work of the optical burst receiver is divided into two stages: initialization stage and data transmission stage. After detecting the arrival of the optical burst signal, the optical burst receiver first enters the initialization stage, and uses the training sequence to iteratively update the tap coefficients of the equalizer until convergence, and completes the initialization of the equalizer and the optical burst receiver. After the initialization of the optical burst receiver is completed, it enters the data sending stage. The algorithm of the two stages in the present invention will be described in detail below.

1. Initialization phase

Fig. 2 is a schematic diagram of the parallel signal branch algorithm in the initialization stage. As shown in Figure 2, compared with the serial algorithm, the difference of the parallel algorithm of the present invention is that one group of 128 parallel symbols corresponds to 128 equalizers (EQ) based on the LMS algorithm, and each equalizer adopts the same The tap coefficient of . The update of the tap coefficient is the key to the initialization of the equalizer. The update of the tap coefficient needs to use the error signal, so the error signal of each branch needs to be obtained first. The training sequence used in the initialization phase should be long enough to ensure that the equalizer tap coefficients obtained by initialization can converge. The initialization phase includes the following specific steps:

S101: Send the training sequence to the optical burst receiver, perform serial-parallel conversion on the training sequence signal after coherent demodulation and sampling and quantization to obtain N parallel signals; set the equalizer tap coefficient corresponding to the n=1 group of parallel signals In this embodiment, there are 128 channels.

S102: The nth group of parallel signals enters N parallel signal processing branches, each parallel signal processing branch includes an equalizer and a frequency offset estimation module, and the equalizer of the i-th branch obtains an equalized signal Where k=(n-1)×N+i, 1≤i≤N.

S103: The frequency offset estimation module uses known training symbols For balanced signal Perform frequency offset estimation to obtain the cumulative phase error and frequency offset estimates

The frequency offset estimation method used in this embodiment is a phase estimation method based on pre-decision, and the algorithm idea is: equalize the signal The phase Φ _k of can be expressed as:

Among them, θ _k is the phase information carried by the symbol, is the cumulative phase error. phase error Can be expressed as:

Among them, φ _0,k is caused by laser phase noise, which changes slowly for high-speed signals, so it can be considered as a constant for a group of parallel symbols, φ _n is the phase caused by ASE (spontaneous emission) noise The fluctuation, kΔω _n,i T is caused by the frequency offset. It can be seen that after subtracting the symbol phase θ _k from the phase of each signal, the remaining Perform a differential operation on the cumulative phase error of adjacent symbols to obtain the estimated frequency offset of each branch Then perform the average operation to obtain the average value of frequency offset estimation It can suppress the influence of φ _n to a certain extent. The frequency offset changes slowly for high-rate signals, so in the present invention, it can be considered that the magnitude of the frequency offset is the same for a group of parallel symbols. The specific steps of frequency offset estimation include:

S3.1: Calculate the equalized signal The cumulative phase error of where Φ _k represents the balanced signal phase.

S3.2: Calculate the estimated value of the frequency offset of this branch in Indicates the cumulative phase error of the k-1th symbol of the training sequence. Obviously, when the first branch in the initialization phase calculates the estimated value of frequency offset, the accumulated phase error initial value of

In this embodiment, as shown in Figure 2, step S3.1 and step S3.2 use the equalized signal with training sequence symbols After multiplying the conjugate (conj(·)) of and Multiplying the conjugates of right Take the angle (arg( )) to get the frequency offset estimate

S104: Calculate the frequency offset estimation results of N branches Perform averaging to obtain the average value of the frequency offset estimation of the nth group of parallel signals which is

S105: N branches respectively calculate their error signals ε _n,i :

Since a known training sequence is used in the initialization phase, it is not necessary to use a decision signal when calculating an error signal, but to directly use known training symbols.

S106: Update the equalizer tap coefficients used by the n+1th group of parallel symbols:

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}& \mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; D.</span>}\\ {\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\frac{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}<span\; class="notranslate">\; \&CenterDot;</span>\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}}<span\; class="notranslate">\; \&CenterDot;</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&rsqb;</span>& \mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; D.</span>}\end{array}\right.$

in, respectively represent the equalizer tap coefficients used by the n+1th group and the nth group of parallel signals. D represents the delay of the error signal, that is, the time from the parallel symbol input to the error signal feedback to the equalizer. Due to the delay, when 1≤n≤D, the tap coefficient cannot be updated, and the tap coefficient always uses the initial value λ is the set iteration step size, which is a positive number, and its selection needs to be small enough to ensure that the iterative process can converge.

The equalizer adopted in the present invention is an equalizer based on the LMS algorithm. When updating the tap coefficient, it is not the error signal of a single branch, but the mean value of the branch error signal, i.e. N _c represents the number of error signals selected from N branches and involved in the calculation of tap coefficients, 1≤N _c ≤N, 1≤ic _{≤N c .} When the number of branches is large, calculating a set of error signals will cause a large delay. Therefore, under the condition of meeting the convergence, the number of error signals involved in the average calculation can be reduced, that is, N _c <N. express The corresponding observation vector, which is the signal input to the equalizer, express the conjugate.

S107: Determine whether the training sequence signal has been received and processed. If not, return to step S102 to continue processing the next group of parallel signals. If the processing is completed, enter the data sending stage.

2. Data sending stage

The main difference between the data transmission phase and the initialization phase is that the phase of the data symbol is unknown, which needs to be recovered through judgment. At the same time, because the phase judgment process of the data symbols may be wrong, it is necessary to insert a certain number of training symbols into the transmitted data symbols to obtain the reference phase required for phase compensation. Fig. 3 is a schematic diagram of a parallel signal branch algorithm in the data sending stage. As shown in Figure 3, the data sending phase includes the following steps:

S201: At the data sending end, insert training symbols into the data symbols, the insertion method is as follows: take N sending symbols as a group, and then divide the N sending symbols into R groups, and in each group N/R sending symbols Contains one training symbol, and the sequence number of R training symbols in parallel symbols is denoted as i _r , 1≤r≤R.

In the data transmission stage, since each group of N symbols needs to be pre-judged on N branches at the same time, if there is an error in the judgment, the feedback error signal is likely to be wrong, which will have a bad impact on the performance of the system. Therefore, in order to obtain a relatively accurate phase compensation when making a decision, the present invention inserts a certain number of training symbols into the transmitted symbols when the transmitting end sends signals. The calculation method of the branch corresponding to the training symbol is the same as that of each branch in the training sequence stage. The training signal branch first calculates the accurate cumulative phase as the reference phase used in the judgment of several data signal branches before and after the branch, so as to carry out More accurate frequency offset estimation.

In order to minimize the amplified noise of the frequency offset estimation value during the judgment, the caret symbol should be inserted in the middle of this small group, taking a group of 128 transmission symbols as an example, the position of the inserted training symbol at this time is: 128/(R×2)+x×128/R, x=0, 1, . . . , R-1. For example, when 4 symbols are inserted in each group, the 32 symbols in each group are judged based on the same cumulative phase error, so the 4 training symbols are respectively inserted at the i ₁ =16, i ₂ =48, i ₃ =80, i ₄ =112 branches, so that the error of phase compensation can be reduced when making a decision, and a maximum of 16 times the estimated value of the frequency offset of the training signal branch can be used.

S202: Send the data signal to the optical burst receiver, perform serial-to-parallel conversion on the data signal after coherent demodulation and sampling and quantization to obtain N parallel signals, the nth group of parallel signals enters N parallel signal processing branches, and the data transmission stage The parallel signal processing branch of includes an equalizer, a frequency offset estimation module and a decision module, and the equalizer processing of the i-th branch obtains an equalized signal

The serial number n of the parallel symbol group in the data transmission stage is continuously arranged from the serial number of the last parallel symbol group in the initialization stage. The equalizer tap coefficient when the first group of parallel symbols in the data transmission stage is processed by the equalizer is the final equalization obtained in the initialization stage The device tap coefficient.

S203: Perform frequency offset estimation on the N branches respectively. The processing flow of the training signal and the data signal are different.

When the branch is a training signal, as shown in Figure 3 x _{n, 16} signals, use the same algorithm as the initialization phase to obtain the error signal, that is: directly based on the known training symbols For balanced signal Perform frequency offset estimation to obtain frequency offset estimation value and cumulative phase error which is and

When the branch is a data signal, as shown in Figure 3 x _{n, 15} and x _{n, 17} signals, first balance the signal Perform phase compensation, the signal after phase compensation for:

where d represents the delay of the frequency offset estimation mean value, Indicates rounding up. It can be seen that the phase compensation of the data signal in the nth group of parallel signals in the data transmission stage uses the estimated average value of the frequency offset obtained from the nd group of parallel signals and the cumulative phase error obtained from the training signals in the group to which it belongs.

Taking x _n,15 as an example, since Therefore its phase compensated signal for:

Judgment module for signal Make a decision to get a decision signal According to the decision signal For balanced signal Perform frequency offset estimation to obtain frequency offset estimation value and cumulative phase error As shown in Figure 3 and

S204: Calculate the frequency offset estimation results of N branches Perform averaging to obtain the average value of the frequency offset estimation of the nth group of parallel signals which is

S205: N branches respectively calculate their error signals ε _n,i . Similarly, the processing methods of the training signal and the data signal are different.

When the branch is the training signal, i=i _r , the error signal ε _n,i is:

When the branch is a data signal, the error signal ε _n,i is:

S206: Update the equalizer tap coefficients used by the n+1th group of parallel symbols:

${\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; =</span>{\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; C</span>}}}_{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; \&lambda;</span>}}^{<span\; class="notranslate">\; *</span>}}{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; \&Center\; Dot;</span>\underset{{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{{\mathrm{<span\; class="notranslate">\; N</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}{<span\; class="notranslate">\; \&Sigma;</span>}}<span\; class="notranslate">\; \&lsqb;</span>{\mathrm{<span\; class="notranslate">\; \&epsiv;</span>}}_{\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}}<span\; class="notranslate">\; \&CenterDot;</span>\stackrel{<span\; class="notranslate">\; \&Right\; Arrow;</span>}{\mathrm{<span\; class="notranslate">\; V</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; no</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; D.</span>}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; i</span>}}_{\mathrm{<span\; class="notranslate">\; c</span>}}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; )</span>}^{<span\; class="notranslate">\; *</span>}<span\; class="notranslate">\; \&rsqb;</span>$

Among them, λ ^* is the iteration step size set in the data sending stage, Indicates the number of error signals selected from the N branches in the data transmission stage to participate in the calculation of the tap coefficients, If the number of error signals N _c and are not the same, the iteration step size λ and λ ^* they use also need to be adjusted accordingly.

In the data transmission stage, since the running time of the system has exceeded the error signal delay D, the equalizer tap coefficients can be updated each time.

S207: Determine whether the data processing is completed, if not, return to step S202 to continue processing the next group of parallel signals, and end if the processing is complete.

The simulation verification of the joint parallel method of channel equalization and frequency offset estimation based on LMS in the present invention is carried out below. In the simulation, the number of parallel branches is 256, standard single-mode fiber is used, the fiber transmission distance is about 50km, and the equalizer uses 11 taps.

Firstly, the influence of the number of error signals N _c in the update of the equalizer tap coefficients on the system performance in the initialization stage is simulated. The calculation delay of the equalizer error signal takes 10 clock units, and the delay of the FOE calculation also takes 10 clock units, a total of 20 delay units. In the initialization phase, 12 frames of training data with 1024 symbols per frame and a duration of about 440 ns are used. In the data sending phase, 4 training symbols are inserted into each group of parallel data. Other parameters used in the simulation include a frequency offset of 1G, and the optical signal-to-noise ratio (OSNR) is fixed at 13dB. Table 1 shows the impact of different error signal numbers on system performance in updating the tap coefficients of the equalizer.

N _c BER λ/ _Nc 32 1.7516×10 ^-2 0.2/32 64 4.6241×10 ^-4 0.2/64 128 4.2286×10 ^-4 0.2/128

Table 1

It can be seen from Table 1 that when the optical signal-to-noise ratio is 13dB, only 64 error signals are needed to obtain good system performance.

Then simulate the impact of the number of training symbols R inserted into each group of transmitted symbols in the data transmission phase on the system performance. Two simulations were performed when the OSNR was 12dB and 13dB respectively. Other parameters are the same as those used in Table 1. Table 2 shows the influence of the number of training symbols inserted in each group of sending symbols in the data sending stage on the system performance.

R Bit Error Rate (OSNR 12dB) Bit Error Rate (OSNR 13dB) 2 3.2787×10 ^-2 5.5445×10 ^-4 4 3.1583×10 ^-2 4.0015×10 ^-4 8 3.1293×10 ^-2 4.7803×10 ^-4

Table 2

It can be seen from Table 2 that when the OSNRs are 12dB and 13dB respectively, when the number of inserted training symbols is 4, good enough performance can already be achieved. When specifically designing the burst optical receiver, different numbers of inserted training symbols can be selected according to different system design requirements.

Fig. 4 is a graph comparing the convergence speed of the equalizer between the parallel method and the serial method of the present invention. The OSNR used in this simulation is 13dB, and MSE stands for Mean Squared Error. As shown in Figure 4, although the parallelization of the serial algorithm will reduce the convergence speed of the equalizer to a certain extent, it will not affect the performance after convergence.

Fig. 5 is a graph comparing the convergence speed with and without calculation delay in the joint parallel method of the present invention. The total delay size used in this simulation is 20 clock units. As shown in Figure 5, when there is a calculation delay, the convergence of the equalizer is only delayed correspondingly with the delay of the iterative calculation, and does not affect the performance after the iterative convergence. And if other different delay sizes are used in the simulation, a similar effect can also be obtained, so it can be seen that the present invention has a good tolerance for the size of the calculation delay.

Fig. 6 is a comparison chart of bit error rates between the serial method and the parallel method under different delays of the present invention. As shown in Fig. 6, as long as the optical burst receiver is correctly initialized, the calculation delay in the present invention has no influence on the bit error rate (BER) performance of the optical burst receiver. At the same time, compared with the performance of the ideal serial method, the performance loss due to parallelization in the present invention is only about 0.2 dB.

Although the illustrative specific embodiments of the present invention have been described above, so that those skilled in the art can understand the present invention, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, As long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions and creations using the concept of the present invention are included in the protection list.

Claims (3)

1. A LMS-based channel equalization and frequency offset estimation joint parallel method is characterized by comprising the following steps:

s1: initializing by adopting a training sequence, comprising the following steps:

s1.1: transmitting a training sequence to an optical burst receiver, and performing series-parallel conversion on training sequence signals subjected to coherent demodulation and sampling quantization to obtain N paths of parallel signals; setting equalizer tap coefficient corresponding to the n-th 1-group parallel signal

S1.2: the nth group of parallel signals enter N parallel signal processing branches, each parallel signal processing branch comprises an equalizer and a frequency offset estimation module, and the equalizer of the ith branch obtains an equalized signalWherein k is (N-1) × N + i, i is not less than 1 and not more than N;

s1.3: the frequency offset estimation module estimates the frequency offset based on the known training symbolsFor equalized signalPerforming frequency offset estimation to obtain accumulated phase errorAnd frequency offset estimation

S1.4: estimating the frequency deviation of N branchesAveraging to obtain the frequency deviation estimation average value of the nth group of parallel signals

S1.5: n branches calculate their error signals respectively_n,i：

S1.6: updating equalizer tap coefficients used by the n +1 th set of parallel signals:

${\stackrel{\&RightArrow;}{C}}_{n+1}=\left\{\begin{array}{cc}{\stackrel{\&RightArrow;}{C}}_n& 1\&le;n\&le;D\\ {\stackrel{\&RightArrow;}{C}}_n-\frac{\mathrm{\&lambda;}}{N_c}\&CenterDot;\underset{i_c=1}{\overset{N_c}{\&Sigma;}}\&lsqb;{\mathrm{\&epsiv;}}_{n-D,i_c}\&CenterDot;\stackrel{\&RightArrow;}{V}{(n-D,i_c)}^{*}\&rsqb;& n>D\end{array}\right.$

wherein,respectively representing equalizer tap coefficients used by the n +1 th group and the n < th > group of parallel signals; d represents the delay of the error signal; λ is the set iteration step, which is a positive number; n is a radical of_cRepresenting the number of error signals selected from the N branches involved in the tap coefficient calculation, 1 ≦ N_c≤N，1≤i_c≤N_c；To representThe corresponding observation vector is then calculated,to representConjugation of (1);

s1.7: judging whether the training sequence is processed or not, if not, returning to the step S1.2 to continue processing the next group of parallel signals, and if so, entering the step S2;

s2: entering a data sending stage to process data, and comprising the following steps:

s2.1: the data transmitting end inserts training symbols into the data symbols, and the inserting method comprises the following steps: taking N sending symbols as a group, dividing the N sending symbols into R groups, wherein each group contains one training symbol in N/R sending symbols, and the serial numbers of the R training symbols in the parallel symbols are marked as i_r,1≤r≤R；

S2.2: sending data signals to an optical burst receiver, performing series-parallel conversion on the data signals subjected to coherent demodulation and sampling quantization to obtain N parallel signals, enabling the nth group of parallel signals to enter N parallel signal processing branches, enabling the parallel signal processing branches of the data sending stage to comprise an equalizer, a frequency offset estimation module and a decision module, and enabling the equalizer of the ith branch to process to obtain equalized signals

S2.3: and respectively carrying out frequency offset estimation on the N branches:

when the branch is a training signal, it is directly based on the known training symbolFor equalized signalPerforming frequency offset estimation to obtain accumulated phase errorAnd frequency offset estimation

When the branch is data signal, firstly, the equalization signal is equalizedPerforming phase compensation, the phase compensated signalComprises the following steps:

where d represents the delay of the mean of the frequency offset estimates,represents rounding up; decision module pair signalMaking a decision to obtain a decision signalBased on decision signalsFor equalized signalPerforming frequency offset estimation to obtain a frequency offset estimation valueAnd accumulated phase error

S2.4: estimating the frequency deviation of N branchesAveraging to obtain the frequency deviation estimation average value of the nth group of parallel signals

S2.5: n branches calculate their error signals respectively_n,i：

When the branch is a training signal, i ═ i_rTime, error signal_n,iComprises the following steps:

when the branch is a data signal, the error signal_n,iComprises the following steps:

s2.6: updating equalizer tap coefficients used by the (n + 1) th set of parallel symbols:

${\stackrel{\&RightArrow;}{C}}_{n+1}={\stackrel{\&RightArrow;}{C}}_n-\frac{{\mathrm{\&lambda;}}^{*}}{N_c^{*}}\&CenterDot;\underset{i_c^{*}=1}{\overset{N_c^{*}}{\&Sigma;}}\&lsqb;{\mathrm{\&epsiv;}}_{n-D,i_c^{*}}\&CenterDot;\stackrel{\&RightArrow;}{V}{(n-D,i_c^{*})}^{*}\&rsqb;$

wherein λ is^*Is the iteration step size set by the data transmission phase,indicating the number of error signals selected from the N branches during the data transmission phase to participate in the tap coefficient calculation,

s2.7: and judging whether the data signal is processed or not, if not, returning to the step S2.2 for continuous processing, and if so, ending the processing.

2. The joint parallel method as claimed in claim 1, wherein the specific method of frequency offset estimation comprises the following steps:

s3.1: computing an equalized signalAccumulated phase error of (2):

accumulating phase error when the branch is a training symbolWherein theta is_kRepresenting the phase, phi, of a known training symbol_kRepresenting an equalized signalThe phase of (d);

accumulating phase errors when the branch is a data signalDifference (D)WhereinRepresenting data decision signalsThe phase of (d);

s3.2: calculating frequency deviation estimated value of the branchWhereinRepresenting the accumulated phase error of the (k-1) th signal.

3. Parallel method according to claim 1, characterized in that the insertion position of the training symbols in step S2.1 is N/(R × 2) + x × N/R, x ═ 0, 1.