A clock data recovery method suitable for PAM-N signals

Abstract

The application discloses a clock data recovery method suitable for PAM-N signals, calculates two types of threshold levels after obtaining symbol levels, and performs edge detection on the basis: firstly, a sliding window with a width of w, i.e., a sampling point number of 1 / 2 unit intervals, is set, current levels V current and current indexes I current are found, then sampling points are traversed backward from the current index I current , when a first type of threshold level is crossed, the level and the index of the sampling point are recorded, when a difference between the last recorded index and the current index is greater than w, an effective threshold V validThres is calculated, then a signal jump point is found in the record, if the signal jump point is not found, a sampling point crossing the effective threshold V validThres between the index I last and the current index I current is found again as the signal jump point, finally, a reference clock signal is recovered by interpolation at a precise time. The application can dynamically adapt to multi-level jump characteristics, the detection capability of the PAM-N signal edge is enhanced, time information of all effective edges in the signal can be extracted, the application overcomes edge detection omission, provides a more accurate and complete reference clock signal for PAM-N signal jitter measurement and eye diagram test, and better test performance is achieved.

Description

Technical Field

[0001] This invention belongs to the field of clock data recovery (CDR) technology, and more specifically, relates to a clock data recovery method applicable to Nth-order pulse amplitude modulation (PAM-N) signals. Background Technology

[0002] High-speed serial communication, as the core architecture of modern high-speed interconnect technology, typically consists of a transmitter, a channel, and a receiver. Currently, most mainstream systems use Non-Return-to-Zero (NRZ) codes, where each symbol carries only 1 bit of information. In contrast, PAM-N signals define N discrete level representations for symbols, allowing a single symbol to transmit log₂N bits of information. To achieve the same data transmission capacity, the symbol rate of PAM-N signals theoretically only needs to reach 1 / log₂N of that of NRZ signals, thus significantly reducing signal bandwidth requirements and channel transmission losses (such as skin effect and dielectric loss). This characteristic makes PAM-N technology demonstrate significant application value in high-density data center interconnects, ultra-high-speed optical communication, and 5G millimeter-wave transmission.

[0003] Jitter measurement and eye diagram testing are important tools for evaluating the performance of high-speed serial communication systems. To achieve accurate testing and analysis, CDR technology is needed to generate a reference clock signal for the signal. As the core module of the receiver, CDR ensures system reliability by suppressing transmission jitter, synchronizing the clock, and resampling data. Traditional CDR systems usually rely on a phase-locked loop (PLL) architecture, which consists of a phase detector, a loop filter, and a voltage-controlled oscillator. However, this type of hardware solution has two limitations: (1) rigid architecture: existing solutions are mostly based on fixed circuit design and lack software-based CDR technology. Especially for high-order PAM-N signals, it is difficult to dynamically adapt to multi-level transition characteristics, resulting in edge detection omissions and jitter information loss; (2) nonlinear constraints: the nonlinear characteristics of the phase detector in the PLL can easily cause phase estimation errors, which exacerbates the risk of clock loss in low signal-to-noise ratio or long run scenarios. Although current hardware CDR systems are widely used, their ability to capture the edges of PAM-N signals is insufficient, which directly affects the integrity of jitter measurement. Therefore, there is an urgent need to develop a software-based CDR method to improve the jitter measurement accuracy of PAM-N signals.

[0004] L.Shuang, Y.Peng, T.Feng and D.Yue, "Research and implementation of clock recovery method based on software PLL," 2017 13th IEEE International Conference on Electronic Measurement & Instruments (ICEMI), Yangzhou, 2017, pp.366-370, doi:10.1109 / ICEMI.2017.8265819. and Wu, T., Song, K., Chen, Z., Zhao, H., Yu, H.: A software PAM4 clock data recovery algorithm for high-speed serial communication. IET Commun. 16, 335–347 (2022), doi:10.1049 / cmu2.12351 respectively gave software clock recovery schemes for NRZ (PAM-2) and PAM-4 signals, but they could not be adapted to other PAM-N signals. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a clock data recovery method suitable for PAM-N signals for arbitrary N-order pulse amplitude modulation, providing a more accurate reference clock signal for jitter measurement and eye diagram testing of PAM-N signals, and achieving better test performance.

[0006] To achieve the above-mentioned objective, this invention provides a clock data recovery method for PAM-N signals, characterized by comprising the following steps:

[0007] (1) Symbol level extraction: Extract the levels corresponding to the N symbol bits in the PAM-N signal and record them as symbol levels V from smallest to largest. i , i = 1, 2, ..., N;

[0008] (2) Threshold level calculation: Calculate the first type of threshold level. Type II threshold level And the error limit ε:

[0009]

[0010] (3) Edge detection

[0011] 3.1) Set a sliding window with a width of w, where w is the number of sampling points at 1 / 2 unit intervals, which is half the average number of sampling points for one sign bit;

[0012] Starting from the first sampling point y[1] of the PAM-N signal y[n], find w consecutive samples that are all at a certain symbol level V. i The judgment condition is that the mean of w consecutive sampling points is within a certain symbol level V. i Between the positive and negative error limits ε, the symbol level V i Let this be the current level V. current The index of the center position of the sliding window is denoted as the current index I. current Where n is the index of the sampling point;

[0013] 3.2) From the current index I current Traverse the sampling points backward to find those that pass through the first type of threshold level. The sampling points are recorded to show the levels that cross the first type of threshold level. The index m of the sampling point and the corresponding first-type threshold level Then calculate the difference between index m and the previous record index. If the difference is less than w, continue traversing backwards until the difference is greater than w.

[0014] 3.3) Let the level V last =V current Index I last =I current And calculate the average of the new record index m and the previous record index, and round down the average to get the index. Assign to the current index I current With current index I current Extract the level of the corresponding sampling point Symbol level V i The closest level to i = 1, 2, ..., N The symbol level V i Assign to V current , level V last With the current level V current The mean of V is used as the effective threshold. validThres ;

[0015] 3.4) The first type of threshold level recorded in step 3.2). And find the valid threshold V in the corresponding index m. validThres If the corresponding index has a valid threshold V validThres Then the effective threshold V in the record validThres The corresponding index m is the signal transition point. If no valid threshold V exists... validThres Then in index I lastand index I current Re-find the effective threshold V between validThres The sampling points are used as signal transition points;

[0016] 3.5) Repeat steps 3.2) to 3.4) until all sampling points have been traversed and all valid transition points have been found;

[0017] 3.6) Use cubic spline interpolation between the signal transition point and its subsequent sampling points to find the effective threshold V. validThres The precise transition time t[l], where l is the index of the precise transition time;

[0018] (4) Recover the reference clock signal based on the precise transition time t[l].

[0019] The objective of this invention is achieved as follows:

[0020] This invention applies to the clock data recovery method for PAM-N signals. After obtaining the symbol level, it calculates two types of threshold levels: the first type of threshold level is the average of two adjacent symbol levels, and the second type of threshold level is the symbol level excluding the maximum and minimum symbol levels. Based on this, edge detection is performed: first, a sliding window with a width of w (i.e., 1 / 2 unit interval sampling points) is set to find the current level V. current and the current index I current Then from the current index I current The sampling points are traversed backwards. When a first-class threshold level is crossed, the level and the index of the sampling point are recorded. When the difference between the index and the previously recorded index is greater than w, the effective threshold V is calculated. validThres Then, search for the signal transition point in the record. If it is not found, then search for it in index I. last and current index I current Re-find the effective threshold V between validThres The sampling points are used as signal transition points, and finally, the precise transition time is found through interpolation to recover the reference clock signal.

[0021] This invention provides a clock data recovery scheme for arbitrary N-order pulse amplitude modulation, which can dynamically adapt to multi-level transition characteristics, enhance the detection capability of PAM-N signal edges, and extract the time information of all effective edges in the signal. It overcomes the omission of edge detection and provides a more accurate and complete reference clock signal for jitter measurement and eye diagram testing of PAM-N signals, thus achieving better test performance. Attached Figure Description

[0022] Figure 1 This is a flowchart of a specific embodiment of the clock data recovery method for PAM-N signals according to the present invention;

[0023] Figure 2 This is a schematic diagram illustrating the principle of symbol level extraction;

[0024] Figure 3 This is a schematic diagram illustrating how interpolation finds the precise transition moment;

[0025] Figure 4 This is a specific example of a signal run-length statistics histogram;

[0026] Figure 5 This is a schematic diagram of synchronous clock generation;

[0027] Figure 6 This is a schematic diagram of a software phase-locked loop structure;

[0028] Figure 7 This is the eye diagram of the PAM-2 signal;

[0029] Figure 8 This is the eye diagram of the PAM-3 signal;

[0030] Figure 9 This is the eye diagram of a PAM-4 signal;

[0031] Figure 10 This is the eye diagram of the PAM-5 signal;

[0032] Figure 11 This is an eye diagram of a PAM-16 signal. Detailed Implementation

[0033] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be particularly noted that in the following description, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0034] Figure 1 This is a flowchart of a specific implementation of the clock data recovery method for PAM-N signals according to the present invention.

[0035] In this embodiment, as Figure 1 As shown, the clock data recovery method for PAM-N signals according to the present invention includes the following steps:

[0036] Step S1: Symbol Level Extraction

[0037] Extract the level corresponding to the N symbol bits in the PAM-N signal and denote them as symbol levels V from smallest to largest. i , i = 1, 2, ..., N.

[0038] In this embodiment, a Gaussian mixture model is used to perform cluster analysis on the amplitude values ​​of the discrete data sequence, i.e., the PAM-N signal y[n], to obtain N Gaussian models, such as... Figure 2 As shown, the means of the N Gaussian models, from smallest to largest, are denoted as μ. i Let i = 1, 2, ..., N, and use this as the level corresponding to each symbol, i.e., the symbol level V. i =μ i ,i=1,2,...,N. In this embodiment, as Figure 2 As shown, the order N = 4.

[0039] Step S2: Threshold level calculation

[0040] Calculate the first type threshold level Type II threshold level And the error limit ε:

[0041]

[0042] As can be seen from the above calculation formula, the first type of threshold level is the arithmetic mean of adjacent symbol levels, and the second type of threshold level is equal to the symbol level, but does not include the maximum and minimum symbol levels. The error limit ε can be finely adjusted within a certain range according to the calculation formula.

[0043] Step S3: Edge Detection

[0044] Step S3.1: Find the current level V current and the current index I current

[0045] Set a sliding window with a width of w, where w is the number of sampling points at 1 / 2 unit intervals, which is half the average number of sampling points for one sign bit.

[0046] Starting from the first sampling point y[1] of the PAM-N signal y[n], find w consecutive samples that are all at a certain symbol level V. i The judgment condition is that the mean of w consecutive sampling points is within a certain symbol level V. i Between the positive and negative error limits ε, the symbol level V i Let this be the current level V. current The index of the center position of the sliding window is denoted as the current index I. current , where n is the index of the sampling point.

[0047] The judgment condition is expressed by the formula:

[0048]

[0049] Step S3.2: Locate and record levels that cross the first type of threshold. The sampling points are counted until the difference between two recorded indices is greater than the sliding window width w.

[0050] From the current index Icurrent Traverse the sampling points backward to find those that pass through the first type of threshold level. The sampling points are recorded to show the levels that cross the first type of threshold level. The index m of the sampling point and the corresponding first-type threshold level Then calculate the difference between index m and the previous record index. If the difference is less than w, continue traversing backwards until the difference is greater than w.

[0051] In this embodiment, the method used to find the threshold level that crosses the first type is as follows. The sampling point, i.e., whether index m satisfies:

[0052]

[0053] If this condition is met, then the sampling point corresponding to index m is the one that passes through the first type of threshold level. If the sampling point is not found, continue iterating backwards until the condition is met.

[0054] Step S3.3: Calculate the effective threshold V validThres

[0055] Let level V last =V current Index I last =I current And calculate the average of the new record index m and the previous record index, and round down the average to get the index. Assign to the current index I current With current index I current Extract the level of the corresponding sampling point Symbol level V i The closest level to i = 1, 2, ..., N The symbol level V i Assign to V current , level V last With the current level V current The mean of V is used as the effective threshold. validThres .

[0056] Step S3.4: Based on the effective threshold V validThres Find the signal transition point

[0057] The first type of threshold level recorded in step S3.2 And find the valid threshold V in the corresponding index m. validThres If the corresponding index has a valid threshold V validThres Then the effective threshold V in the record validThres The corresponding index m is the signal transition point. If no valid threshold V exists... validThres Then in index I last and index Icurrent Re-find the effective threshold V between validThres The sampling points are used as signal transition points.

[0058] In this embodiment, the method used to find the effective threshold level V is as follows. validThres The sampling point, i.e., whether index m satisfies:

[0059] (y[m]-V validThres )×(y[m+1]-V validThres )≤0

[0060] If this condition is met, then the sampling point corresponding to index m is the one that passes through the effective threshold level V. validThres If the sampling point is not found, continue iterating backwards until the condition is met.

[0061] If there are multiple signal transition points, it means that there are multiple points nearby that cross the effective threshold V. validThres The sampling points (caused by glitches and noise) are averaged to obtain the average index. The sampling point with the corresponding average index is then found in the PAM-N signal y[n] as the final signal transition point.

[0062] Step S3.5: Repeat steps S3.2 to S3.4 until all sampling points have been traversed and all valid transition points have been found.

[0063] Step S3.6: Interpolate to find the precise transition moment

[0064] The signal transition point is interpolated using cubic spline interpolation with the sampling points before or after it to find the effective threshold V. validThres The precise transition time t[l] is given by the index of the precise transition time.

[0065] In this embodiment, as Figure 3 As shown, y

[14] and y

[30] are signal transition points. Interpolation can be performed using one or more sampling points after them, or by using one or more sampling points before and after them. Interpolation only requires a segment of interpolated data from the subsequent sampling points y

[15] and y

[31] . Then, the effective threshold V is found in the interpolated data. validThres The precise transition times t[1] and t[2]. In this embodiment, the effective threshold V corresponding to the first signal transition point y

[14] is... validThres It is the second type of threshold level, and the effective threshold V corresponding to the second signal transition point y

[30] is the second type of threshold level. validThres This is the first type of threshold level. Interpolation to determine the precise transition time is existing technology and will not be elaborated upon here. In specific implementations, other existing interpolation methods can also be used.

[0066] Step S4: Recover the reference clock signal based on the precise transition time t[l].

[0067] Step S4.1: Determine the average UI duration T symbol

[0068] Differentiate the precise transition time t[l] to obtain the data run RL[l]:

[0069] RL[l]=t[l+1]-t[l],l=1,2,...,L-1

[0070] Where L is the number of signal transition points.

[0071] Create a statistical histogram for the data run RL[l], such as... Figure 4 As shown. In the statistical histogram, the x-axis corresponding to each peak is within the average UI duration T. symbol By finding the x-coordinates corresponding to these peaks near integer multiples of the value, T can be determined. symbol .

[0072] The peaks are identified by their x-coordinates, which are numbered from smallest to largest as T1, T2, T3, ... . The following formula can be used for calculation:

[0073]

[0074] Step S4.2: Synchronization Clock Generation

[0075] When there are multiple symbols between two edges, there is no edge information. In order to determine the edge information of its synchronization clock signal, it is necessary to interpolate the precise transition time t[l] of the PAM-N signal. First, it is necessary to determine the number of synchronization times that need to be inserted between the precise transition time t[l] and the precise transition time t[l+1] based on the time difference between adjacent precise transition times t[l], i.e., the data run RL[l].

[0076]

[0077] Then, synchronization moments are evenly inserted between the precise transition moment t[l] and the precise transition moment t[l+1]. All precise transition moments t[l] are also used as synchronization moments, and a synchronization clock is generated by sorting them by time.

[0078] In this embodiment, as Figure 5 As shown, the synchronization time t sync [p] means:

[0079] t sync [1]=t[1],t sync [2]=t[2],t sync [4]=t[3],t sync [7]=t[4]

[0080]

[0081] Where p is the index of the synchronization time.

[0082] Step S4.3: The software phase-locked loop outputs a reference clock signal.

[0083] Synchronization time t sync [p] serves as the input to the software phase-locked loop, yielding the edge information t of the reference clock signal. ref That is, the recovered clock signal, such as Figure 6 As shown. The coefficients of the loop filter affect the overall loop characteristics of the phase-locked loop and need to be designed according to requirements. Software phase-locked loops are existing technology and will not be discussed further here.

[0084] Figures 7-11 These are eye diagrams for PAM-2, PAM-3, PAM-4, PAM-5, and PAM-16 signals, respectively. Figures 7-11 As can be seen, the method proposed in this invention can dynamically adapt to multi-level transition characteristics, enhance the detection capability of PAM-N signal edges, extract the time information of all effective edges in the signal, overcome the omission of edge detection, provide a more accurate and complete reference clock signal for jitter measurement and eye diagram testing of PAM-N signals, and achieve better test performance.

[0085] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A clock data recovery method suitable for PAM-N signals, characterized in that, Includes the following steps: (1) Symbol level extraction: Extract the levels corresponding to the N symbol bits in the PAM-N signal and record them as symbol levels V from smallest to largest. i , i = 1, 2, ..., N; (2) Threshold level calculation: Calculate the first type of threshold level. Type II threshold level And the error limit ε: (3) Edge detection 3.1) Set a sliding window with a width of w, where w is the number of sampling points at 1 / 2 unit intervals, which is half the average number of sampling points for one sign bit; Starting from the first sampling point y[1] of the PAM-N signal y[n], find w consecutive samples that are all at a certain symbol level V. i The judgment condition is that the mean of w consecutive sampling points is within a certain symbol level V. i Between the positive and negative error limits ε, the symbol level V i Let it be denoted as the current level V current The index of the center position of the sliding window is denoted as the current index I. current Where n is the index of the sampling point; 3.2) From the current index I current Traverse the sampling points backward to find those that pass through the first type of threshold level. The sampling points are recorded to show the levels that cross the first type of threshold level. The index m of the sampling point and the corresponding first-type threshold level Then calculate the difference between index m and the previous record index. If the difference is less than w, continue traversing backwards until the difference is greater than w. 3.3) Let the level V last =V current Index I last =I current And calculate the average of the new record index m and the previous record index, and round down the average to get the index. Assign to the current index I current With current index I current Extract the level of the corresponding sampling point Symbol level V i The closest level to i = 1, 2, ..., N The symbol level V i Assign to V current , level V last With current level V current The mean of V is used as the effective threshold. validThres ; 3.4) The first type of threshold level recorded in step 3.2). And find the valid threshold V in the corresponding index m. validThres If the corresponding index has a valid threshold V validThres Then the effective threshold V in the record validThres The corresponding index m is the signal transition point. If no valid threshold V exists... validThres Then in index I last and index I current Re-find the effective threshold V between validThres The sampling points are used as signal transition points; 3.5) Repeat steps 3.2) to 3.4) until all sampling points have been traversed and all valid transition points have been found; 3.6) Use cubic spline interpolation between the signal transition point and its subsequent sampling points to find the effective threshold V. validThres The precise transition time t[l], where l is the index of the precise transition time; (4) Recover the reference clock signal based on the precise transition time t[l].

2. The clock data recovery method for PAM-N signals according to claim 1, characterized in that, The symbol level extraction in step (1) is as follows: The amplitude values ​​of the discrete data sequence, i.e., the PAM-N signal y[n], are clustered using a Gaussian mixture model to obtain N Gaussian models. The means of the N Gaussian models are denoted as μ in ascending order. i Let i = 1, 2, ..., N, and use this as the level corresponding to each symbol, i.e., the symbol level V. i =μ i ,i=1,2,...,N.

3. The clock data recovery method for PAM-N signals according to claim 1, characterized in that, In step 3.2), the method for finding the threshold level that passes through the first type is as follows. The sampling point, i.e., whether index m satisfies: If this condition is met, then the sampling point corresponding to index m is the one that passes through the first type of threshold level. If the sampling point is not found, continue iterating backwards until the condition is met.

4. The clock data recovery method for PAM-N signals according to claim 1, characterized in that, In step 3.4), the effective threshold level V is found using the following method. validThres The sampling point, i.e., whether index m satisfies: (y[m]-V validThres )×(y[m+1]-V validThres )≤0 If this condition is met, then the sampling point corresponding to index m is the one that passes through the effective threshold level V. validThres If the sampling point is not found, continue iterating backwards until the condition is met.

5. The clock data recovery method for PAM-N signals according to claim 1, characterized in that, In step 3.4), if there are multiple signal transition points, it means that there are multiple points nearby that cross the effective threshold V. validThres The sampling points are then averaged to obtain the average index. The sampling point with the corresponding average index is then found in the PAM-N signal y[n] as the final signal transition point.

Images