CN109302362B

CN109302362B - High-speed link equalization method using DFE and CTLE

Info

Publication number: CN109302362B
Application number: CN201811222841.1A
Authority: CN
Inventors: 初秀琴; 张文博; 戴翔宇; 赵国荣; 张超余
Original assignee: Xidian University
Current assignee: Xidian University
Priority date: 2018-10-19
Filing date: 2018-10-19
Publication date: 2020-04-14
Anticipated expiration: 2038-10-19
Also published as: CN109302362A

Abstract

The invention discloses a high-speed link equalization method using an adaptive decision feedback equalizer DFE and a linear continuous time equalizer CTLE, the steps of which are: (1) inputting a waveform; (2) generating a transfer function in the discrete complex frequency domain; ( 3) Obtain the waveform equalized by the linear continuous time equalizer CTLE; (4) Calculate the weight value (5) Calculate the waveform value of each sampling point in the waveform after the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE equalized waveform (6) Obtain the waveform after equalization. The invention effectively improves the generality of the high-speed link equalization, and reduces the computational complexity of the high-speed link equalization.

Description

High-speed link equalization method using DFE and CTLE

Technical Field

The invention belongs to the technical field of computers, and further relates to a high-speed link equalization method using a self-adaptive decision Feedback equalizer DFE (decision Feedback equalization) and a Linear continuous Time equalizer CTLE (continuous Time Linear equalization) in the technical field of computer high-speed link equalization. The invention can realize the equalization of the high-speed link signal and utilize the signals before and after the equalization to carry out signal integrity analysis.

Background

At present, in a high-speed link of a computer, the transmission speed of signals has become faster and faster, which causes significant inter Symbol interference isi (inter Symbol interference) and reflection. Equalization of the high speed link signal is therefore required to recover a usable signal.

The patent document applied by the fifty-fourth institute of electrical and technology corporation of china, "a low-complexity decision feedback equalization algorithm" (application date: 2017.11.28, application number: 201711214380.9, application publication number: CN 107911322A) provides a low-complexity decision feedback equalization algorithm. The method comprises the steps of firstly carrying out frequency domain transformation on a received signal at a receiving end, then carrying out linear equalization conversion to a time domain, inputting the signal to a noise predictor after judgment for noise prediction, thereby reducing noise interference of an undetermined symbol and improving the performance of a communication system. However, the method still has the following defects: the algorithm needs to establish a prediction model according to the characteristics of the channel, and different channels need to establish different prediction models, so that the universality of the high-speed link equalization is reduced.

The patent document of the university in zhongshan provides a channel equalization method of a single carrier system in a combined time domain and frequency domain channel equalization algorithm (application date: 2018.03.20, application number: 201810232190.8, application publication number: CN 108418771 a). The method comprises the following specific steps: (1) at a receiving end, properly segmenting received data in a time domain, and converting time domain calculation into a frequency domain by adopting an overlapping reservation method; (2) performing coarse judgment by adopting an MMSE (minimum mean square error) equalization method; (3) and based on the coarse judgment result, further fine judgment is carried out on the coarse judgment result by adopting a time domain exhaustive search method. The method has the following defects: the algorithm needs to make two decisions, so that the computational complexity of high-speed link equalization is increased.

Disclosure of Invention

The present invention is directed to addressing the deficiencies of the prior art discussed above by providing a high speed link equalization method using an adaptive decision feedback equalizer DFE and a linear continuous time equalizer. The method can realize the equalization of the high-speed link signal, utilizes the signals before and after the equalization to carry out signal integrity analysis, simultaneously ensures the accuracy of the equalizer, reduces the operation complexity of the equalizer and increases the universality of the equalizer.

In order to achieve the purpose, the specific idea of the invention is that firstly, the input waveform is subjected to fast Fourier transform, a transfer function of a discrete complex frequency domain is generated after the transform, and linear continuous time equalization is performed after the generation; then generating an ideal decision waveform, calculating the weight value of each sampling point in the waveform by adopting a self-adaptive algorithm after the ideal decision waveform is generated, and generating tap coefficients of a DFE (decision feedback equalizer); and finally, calculating the waveform value of each sampling point in the equalized waveform of the DFE and the CTLE to generate the equalized waveform.

The method comprises the following specific steps:

(1) inputting a waveform:

(1a) randomly inputting a waveform of a high-speed link as a time domain input waveform;

(1b) performing fast Fourier transform on the time domain input waveform to obtain a frequency domain input waveform;

(2) generating a transfer function of a discrete complex frequency domain:

(2a) converting the zero and pole frequency values of the input linear continuous time equalizer CTLE into a transfer function of a complex frequency domain;

(2b) discretizing the transfer function of the complex frequency domain to obtain a discrete complex frequency domain transfer function;

(3) obtaining the equalized waveform of a linear continuous time equalizer (CTLE):

(3a) multiplying the frequency domain input waveform by a discrete complex frequency domain transfer function to obtain a frequency domain output waveform;

(3b) performing inverse fast Fourier transform on the frequency domain output waveform to obtain a waveform balanced by a linear continuous time equalizer CTLE;

(4) calculating the weight value of each sampling point in the waveform after the interruption:

(4a) taking the middle point of the waveform peak point after the CTLE equalization as the starting point of the waveform, and deleting the waveform data before the starting point to obtain the waveform after the interruption;

(4b) setting all sampling points which are larger than a threshold value in the waveform after the cut-off as high-level values, setting all sampling points which are smaller than the threshold value in the waveform after the cut-off as low-level values, and forming an ideal decision waveform by all the waveforms after the setting is finished;

(4c) calculating the difference value of the waveform value of each sampling point in the waveform after the truncation and the waveform value of each corresponding sampling point in the ideal decision waveform;

(4d) calculating the weight value of each sampling point in the waveform after the truncation by utilizing a self-adaptive algorithm formula and the difference value of the waveform value of each sampling point in the waveform after the truncation and the waveform value of each corresponding sampling point in the ideal decision waveform;

(5) calculating and generating a waveform value of each sampling point in the equalized waveform of the DFE and the CTLE:

(5a) selecting weighted values of the last N sampling points in the waveform after truncation to form tap coefficients of a DFE (decision feedback equalizer), wherein N is more than or equal to 1 and less than or equal to N, and the value of N is determined by the accuracy required by equalization;

(5b) calculating and generating a waveform value of each sampling point in the equalized waveform of the DFE and the CTLE by utilizing a waveform value calculation formula and through a tap coefficient of the DFE;

(6) acquiring an equalized waveform:

and (3) waveform values of all sampling points in the equalized waveforms of the self-adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE are combined into the equalized waveforms of the self-adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE according to a sampling sequence.

Compared with the prior art, the invention has the following advantages:

first, the invention adopts the adaptive algorithm formula to calculate the tap coefficient of the DFE, thus overcoming the problem that different channel prediction models need to be established for different channels due to the fact that the tap coefficient of the DFE needs to be calculated by a channel prediction model in the prior art, and the high-speed link equalization needs to be modified differently for different channels, thereby reducing the universality and expanding the universality of the high-speed link equalization.

Secondly, because no further judgment is needed after the ideal judgment waveform is generated, the defect that the operation complexity of the high-speed link equalization is increased because a time domain exhaustive search method is needed for further judgment after the ideal judgment waveform is generated in the prior art is overcome, and the operation complexity of the high-speed link equalization is reduced while the accuracy of the high-speed link equalization is ensured.

Drawings

FIG. 1 is a flow chart of the present invention;

FIG. 2 is a simulation of the present invention.

Detailed Description

The invention is further described below with reference to the accompanying drawings.

The steps performed by the present invention will be further described with reference to fig. 1.

Step 1, inputting a waveform.

Randomly deriving a waveform file consisting of sampling points at equal intervals from a circuit simulation tool, reading the waveform in the file, and taking the waveform as a time domain input waveform.

And carrying out fast Fourier transform on the time domain input waveform to obtain a frequency domain input waveform.

And 2, generating a transfer function of the discrete complex frequency domain.

And converting the zero point and pole frequency values of the input linear continuous time equalizer CTLE into a transfer function of a complex frequency domain.

And discretizing the transfer function of the complex frequency domain to obtain the transfer function of the discrete complex frequency domain.

The linear continuous time equalizer CTLE is an analog filter, so it needs to convert the pole-zero into a transfer function of the complex frequency domain. And then, converting the transfer function of the complex frequency domain into the transfer function of the discrete complex frequency domain by using a bilinear transformation method, wherein the physical meaning of the transfer function is to convert the analog filter into the digital filter.

And 3, acquiring the waveform of the linear continuous time equalizer CTLE after equalization.

And multiplying the frequency domain input waveform by the transfer function of the discrete complex frequency domain to obtain a frequency domain output waveform.

And performing inverse fast Fourier transform on the frequency domain output waveform to obtain a waveform equalized by a linear continuous time equalizer CTLE.

And 4, calculating the weight value of each sampling point in the waveform after the interruption.

And taking the middle point of the peak point of the waveform after the linear continuous time equalizer CTLE is equalized as the starting point of the waveform, and deleting the waveform data before the starting point to obtain the waveform after the interruption.

Setting all sampling points which are larger than the threshold value in the waveform after the cutting to be high level values, setting all sampling points which are smaller than the threshold value in the waveform after the cutting to be low level values, and forming the waveform after the setting is finished into an ideal decision waveform.

The threshold value is an average value of ideal high and low level values of the DFE of the adaptive decision feedback equalizer.

And calculating the difference value between the waveform value of each sampling point in the waveform after the truncation and the waveform value of each corresponding sampling point in the ideal decision waveform.

And calculating the weight value of each sampling point in the waveform after the truncation according to the difference value of the waveform value of each sampling point in the waveform after the truncation and the waveform value of each corresponding sampling point in the ideal decision waveform by using a self-adaptive algorithm formula.

The adaptive algorithm formula is as follows:

w(k+1)＝w(k)+2μe(k)y_d(h)

w (k +1) represents a weight value of a k +1 th sampling point in the waveform after the truncation, k is 1.. n, n represents the total number of sampling points of the waveform after the truncation, w (k) represents a weight value of a k-th sampling point in the waveform after the truncation, μ represents a step value taking an arbitrary constant value, e (k) represents a difference value between a waveform value of the k-th sampling point of the waveform after the truncation and a waveform value of a corresponding sampling point in an ideal decision waveform, y_d(h) And representing the waveform value of the h sampling point in the ideal decision waveform, wherein the value of h is correspondingly equal to the value of k.

The formula principle of the self-adaptive algorithm is a minimum mean square error algorithm which is provided by B.Widrow, Hoff and the like in 1960 on the basis of a trembling reduction algorithm, and the algorithm is simple and easy to implement, so that the self-adaptive algorithm is widely used. The rule employed by the minimum mean square error algorithm is a criterion that minimizes the mean square error between the desired output value and the actual output value of the equalizer.

The mean square error function is a quadratic equation of the tap coefficients, thereby forming a multi-dimensional hypersolidoid, resembling a bowl-shaped curve, with a unique minimum point. The self-adaptive algorithm is to move the tap coefficient towards the direction of the minimum point at the bottom of the curved surface, and finally the tap coefficient reaches the minimum point at the bottom, so that the optimal tap coefficient value with the minimum error is obtained.

The minimum mean square error algorithm is the optimal algorithm to achieve the above-mentioned motion, and it uses gradient information to analyze adaptive filtering performance and track the optimal filtering state. The tap coefficients are adjusted in the direction of gradient descent.

And 5, calculating and generating a waveform value of each sampling point in the equalized waveform of the self-adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE.

And selecting weighted values of the last N sampling points in the waveform after truncation to form tap coefficients of the DFE of the adaptive decision feedback equalizer, wherein N is more than or equal to 1 and less than or equal to N, and the value of N is determined by the accuracy required by equalization.

And calculating the waveform value of each sampling point in the equalized waveform of the DFE and the CTLE by utilizing a waveform value calculation formula and through the tap coefficient of the DFE.

The waveform value calculation formula is as follows:

wherein, y_e(g) The waveform value of the g sampling point in the equalized waveforms of the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE is represented, g is 1_dAnd (l-i) represents the waveform value of the l-i sampling point in the ideal decision waveform, and the values of f, j and l are correspondingly equal to the value of g.

And 6, acquiring the equalized waveform.

The effects of the present invention can be further illustrated by the following simulations.

1. Simulation conditions are as follows:

MATLAB R2016b simulation software and ADS 2014 circuit simulation software are used in the simulation experiment, the high voltage and the low voltage of the waveform in the ADS circuit simulation software are respectively 1.2V-1.2V, the data rate of the waveform is 8Gbps, the waveform contains 5000 bits, the time length of each bit is 125ps, and the sampling point of each bit is 32. The number of taps of the adaptive decision feedback equalizer DFE is set to 4, the step size is set to 0.001, the zero of the linear continuous time equalizer CTLE is set to 1.26Ghz, and the two poles are set to 19.4Ghz and 9.803Ghz, respectively.

2. Simulation content and simulation result analysis:

the simulation experiment of the invention adopts the equalizing method of the invention and the equalizing method carried by ADS circuit simulation software to respectively equalize the waveform of the receiving end of the high-speed link, and the comparison result is shown in figure 2. Fig. 2 (a) is a waveform diagram before the equalization method is applied to the high-speed link. Fig. 2 (b) is a waveform diagram after the high-speed link is equalized by using the self-contained linear continuous time equalizer CTLE in the ADS circuit simulation software. Fig. 2 (c) is a waveform diagram after equalization of a high-speed link using an adaptive decision feedback equalizer DFE in ADS circuit emulation software. Fig. 2 (d) is a waveform diagram after equalizing the high-speed link using the linear continuous time equalizer CTLE of the present invention. Fig. 2 (e) is a waveform diagram after equalization of a high-speed link using the adaptive decision feedback equalizer DFE of the present invention. The ADS circuit simulation software is a mature commercial software provided by Agilent technologies, Inc. of America, and the self-contained equalization method has been approved by many companies,

the abscissa in fig. 2 represents time in picoseconds. The ordinate in fig. 2 represents the voltage value in volts. The waveforms in fig. 2 (a), 2 (b) and 2 (d) were automatically measured using MATLAB R2016b simulation software, respectively, to obtain data of eye height and eye width of the corresponding waveforms, as detailed in table 1. CTLE in table 1 refers to a linear continuous time equalizer.

TABLE 1 comparison of eye height and eye width after CTLE equalization using the present invention and the prior art

The waveforms in fig. 2 (a), fig. 2 (c) and fig. 2 (e) were measured using MATLAB R2016b simulation software, and the data of the eye height and eye width of the corresponding waveforms were obtained, as detailed in table 2. DFE refers to an adaptive decision feedback equalizer in table 2.

The errors of the eye height and eye width of the waveform after the linear continuous time equalizer CTLE equalization in table 1 and the eye height and eye width of the waveform after the linear continuous time equalizer CTLE equalization in the ADS are calculated according to the following two formulas:

(0.850-0.845)/0.850≈1％

(111.9-111.3)/111.9≈1％

the errors of the eye height and the eye width of the waveform after the linear continuous time equalizer CTLE is equalized and the errors of the eye height and the eye width of the waveform after the linear continuous time equalizer CTLE is equalized in the ADS are about 1 percent. In the field of practical engineering, the error within 10% is an acceptable error, so that the error of 1% is very small, and the method obviously improves the precision of equalization compared with the prior art.

TABLE 2 eye height and eye width comparison table after equalization using DFE of the present invention and the prior art

The errors of the eye height and eye width of the equalized waveform pattern of the DFE of the adaptive decision feedback equalizer in the invention in Table 2 and the eye height and eye width of the equalized waveform pattern of the DFE of the adaptive decision feedback equalizer in ADS are calculated according to the following two formulas:

(0.578-0.5456)/0.546≈5.7％

(112.5-109.4)/112.5≈2.7％

the errors of the eye height and the eye width of the equalized waveform diagram of the DFE of the adaptive decision feedback equalizer in the invention and the eye height and the eye width of the equalized waveform diagram of the DFE of the adaptive decision feedback equalizer in the ADS are respectively about 5.7 percent and 2.7 percent. In the field of practical engineering, the error within 10% is the acceptable error range, so that the errors of 5.7% and 2.7% are smaller, and the method obviously improves the balancing precision compared with the prior art.

Claims

1. a high-speed link equalization method utilizing self-adaptive decision feedback equalizer DFE and linear continuous time equalizer CTLE, is characterized in that, generates the transfer function of discrete complex frequency domain, utilizes the adaptive algorithm formula to calculate each time in the truncated waveform. The weight value of a sampling point is calculated to generate the waveform value of each sampling point in the waveform after equalization by the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE; the steps of the method include the following:

(1) Input waveform:

(1a) Randomly input the waveform of a high-speed link and use it as the time-domain input waveform;

(1b) Fast Fourier transform is performed on the input waveform in the time domain to obtain the input waveform in the frequency domain;

(2) Generate a transfer function in the discrete complex frequency domain:

(2a) Convert the zero and pole frequency values of the input linear continuous-time equalizer CTLE into a transfer function in the complex frequency domain;

(2b) Discretize the transfer function in the complex frequency domain to obtain a discrete complex frequency domain transfer function;

(3) Obtain the waveform after CTLE equalization of the linear continuous time equalizer:

(3a) Multiply the frequency domain input waveform with the discrete complex frequency domain transfer function to obtain the frequency domain output waveform;

(3b) performing an inverse fast Fourier transform on the frequency domain output waveform to obtain a waveform equalized by a linear continuous time equalizer CTLE;

(4) Calculate the weight value of each sampling point in the truncated waveform:

(4a) Take the middle point of the waveform peak point equalized by the linear continuous time equalizer CTLE as the starting point of the waveform, delete the waveform data before the starting point, and obtain the truncated waveform;

(4b) Set all sampling points in the truncated waveform that are greater than the threshold to high-level values, and set all sampling points that are smaller than the threshold in the truncated waveform to low-level values, and combine all the waveforms after setting to form an ideal judgment waveform;

(4c) Calculate the difference between the waveform value of each sampling point in the truncated waveform and the waveform value of each corresponding sampling point in the ideal decision waveform;

(4d) Using the adaptive algorithm formula, calculate the weight value of each sampling point in the truncated waveform by the difference between the waveform value of each sampling point in the truncated waveform and the waveform value of each corresponding sampling point in the ideal decision waveform ;

The adaptive algorithm formula is as follows:

w(k+1)=w(k)+2μe(k)y _d (h)

Among them, w(k+1) represents the weight value of the k+1th sampling point in the truncated waveform, k=1...n, n represents the total number of sampling points of the truncated waveform, w(k) represents the truncated waveform The weight value of the kth sampling point in , μ represents the step value of taking an arbitrary constant value, e(k) represents the difference between the waveform value of the kth sampling point of the truncated waveform and the waveform value of the corresponding sampling point in the ideal decision waveform value, y _d (h) represents the waveform value of the h-th sampling point in the ideal decision waveform, and the value of h corresponds to the value of k;

(5) Calculate and generate the waveform value of each sampling point in the waveform after equalization by the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE:

(5a) Select the weight values of the last N sampling points in the truncated waveform to form the tap coefficients of the adaptive decision feedback equalizer DFE, where 1≤N≤n, and the value of N is determined by the accuracy required for equalization;

(5b) utilize the waveform value calculation formula, through the tap coefficient of the adaptive decision feedback equalizer DFE, calculate and generate the waveform value of each sampling point in the waveform after the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE are equalized;

(6) Obtain the waveform after equalization:

The waveform values of all sampling points in the waveform after the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE are equalized are formed according to the sampling order to form the waveform after the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE.

2. the high-speed link equalization method utilizing adaptive decision feedback equalizer DFE and linear continuous time equalizer CTLE according to claim 1, is characterized in that, the threshold value described in step (4b) refers to, adaptive decision Average of ideal high and low level values for the feedback equalizer DFE.

3. the high-speed link equalization method utilizing adaptive decision feedback equalizer DFE and linear continuous time equalizer CTLE according to claim 1, is characterized in that, described in step (5b), the waveform value calculation formula is as follows:

Among them, y _e (g) represents the waveform value of the g-th sampling point in the waveform after the adaptive decision feedback equalizer DFE and the linear continuous time equalizer CTLE equalize, g=1...n, x(f) represents the truncated waveform The waveform value of the waveform at the f-th sampling point, w(ji) represents the weight value of the ji-th sampling point in the truncated waveform, i=1...N, y _d (li) represents the li-th sampling point in the ideal decision waveform For the waveform value of the sampling point, the values of f, j, and l are correspondingly equal to the value of g.