A PAM4 adaptive equalization structure and method based on ADC architecture

By employing a fixed-coefficient third forward equalizer and adaptive module optimization coefficients in the PAM4 receiver, the problems of large area and power consumption in the traditional structure are solved, the performance of DFE and MLSD is optimized, and the system achieves efficient operation.

CN123123083APending Publication Date: 2026-11-17CHENGDU RENXIN TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611430224.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-09-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In traditional PAM4 receivers based on ADC architecture, the forward equalizer design is complex, resulting in large area and power consumption overhead. Furthermore, CDR and FFE interfere with each other, affecting the signal-to-noise ratio and DFE and MLSD performance.

Method used

The PAM4 adaptive equalization structure based on ADC architecture is adopted, including a first forward equalizer, a third forward equalizer, and a decision feedback equalizer. The coefficients of the third forward equalizer are fixed, and the coefficients are optimized through an adaptive module and a multiplier module to simplify the system design.

Benefits of technology

The area and power consumption have been reduced, the performance of DFE and MLSD has been optimized, and different operating modes have been adapted to achieve the best system performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123083A_ABST
    Figure CN123123083A_ABST
Patent Text Reader

Abstract

This invention relates to the field of communication technology, specifically to a PAM4 adaptive equalization structure and method based on an ADC architecture; connected to an analog-to-digital converter; including a first forward equalizer, a second forward equalizer, a third forward equalizer, a decision unit, a decision feedback equalizer, a maximum likelihood sequence detector, a clock data recovery module, and a selector; the first forward equalizer performs equalization processing on the output signal of the analog-to-digital converter, equalizing inter-symbol interference other than the first coefficient; the third forward equalizer equalizes the first coefficient with a fixed coefficient; by replacing the forward equalizer set for the clock data recovery module with the third forward equalizer with a fixed coefficient, the coefficients are fixed and do not require adaptation, the structure is simple, and the area and power consumption are greatly reduced; and the sampling point position and system parameters are determined by the coefficient a1, and optimization methods are provided in both DFE and MLSD modes to optimize system performance.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of communication technology, and more specifically, to a PAM4 adaptive equalization structure and method based on an ADC architecture. Background Technology

[0002] With the continuous increase in SerDes (serializer / deserializer) rates, higher-order modulation such as PAM4 and PAM8 are becoming increasingly widely used. In receivers based on ADC (analog-to-digital converter) architectures, after the received signal is sampled and digitized by the ADC, it needs to undergo equalization processing to eliminate inter-symbol interference (ISI) caused by the channel in order to recover the original signal from the transmitter. Commonly used equalization techniques include forward equalizers (FFE), decision feedback equalizers (DFE), and maximum likelihood sequence detectors (MLSD).

[0003] In traditional PAM4 receivers based on ADC architecture, a dual-path structure is typically used. The upper path signal is equalized by FFE0 and then enters the decision slicer for signal decision. The decision result is provided to the clock data recovery module (CDR) and the adaptive module. The lower path signal is filtered by a longer FFE1 filter and then enters the DFE decision slicer. FFE1 does not include coefficient c1 (i.e., h1 is not equalized). The DFE is a decision slicer that includes a 1-tap DFE coefficient. Both the DFE coefficient and the FFE coefficient need to be adaptive. The output yk of FFE1 can also be provided to the MLSD for decision output. The final output is selected between the DFE and MLSD.

[0004] The transmitted signal, after passing through the channel and ADC, can be expressed as: x(n) = Σh(k)·d(nk), where h is the impulse response of the channel model, and h(1), h(2)...h(-1), h(-2)... are called ISI (Inter-symbol Interference). h(0) is the main tap value of the impulse response, and h(1) is the first backward ISI value. CDR usually adopts the Mueller-Muller algorithm, and after CDR locking, h(-1) = h(1) is satisfied.

[0005] However, the traditional structure has the following problems: First, FFE0 is an equalizer specifically designed for CDR, which requires adaptive adjustment of the tap coefficients, resulting in a large area and power consumption overhead. Second, since FFE0 can make h(-1)=0 and h(1)=0, that is, both satisfy the CDR locking condition, the CDR sampling position can be any point. Therefore, CDR and FFE0 have mutual influence, resulting in the position of h(0) being non-unique. Shifting h(0) to the left can reduce h(-1) and increase h(1), thereby reducing the FFE coefficient to reduce the amplification of noise by FFE, which is beneficial to increasing the signal-to-noise ratio after FFE1, but it will reduce h(0), which may worsen the signal-to-noise ratio. At the same time, increasing the ratio of h(1) / h(0) will increase the DFE coefficient, increase the DFE error propagation, and degrade the performance. Third, the error propagation characteristics of MLSD are also related to the ratio of h(1) / h(0). If the ratio is too large, the performance of MLSD will degrade. Therefore, a suitable h(0) sampling point is needed to balance various performance indicators. Summary of the Invention

[0006] To address the aforementioned shortcomings of existing technologies, this invention provides a PAM4 adaptive equalization structure and method based on an ADC architecture, aiming to reduce area and power consumption while ensuring the performance of DFE and MLSD.

[0007] The specific implementation details of this invention are as follows:

[0008] A PAM4 adaptive equalization structure based on ADC architecture is connected to an analog-to-digital converter; it includes a first forward equalizer, a second forward equalizer, a third forward equalizer, a decision unit, a decision feedback equalizer, a maximum likelihood sequence detector, a clock data recovery module, and a selector.

[0009] The input terminal of the first forward equalizer is connected to the output terminal of the analog-to-digital converter, and the output terminal of the first forward equalizer is connected to the input terminals of the second forward equalizer and the third forward equalizer.

[0010] The input terminal of the decision unit is connected to the output terminal of the third forward equalizer, and the output terminal of the decision unit is connected to the input terminal of the clock data recovery module.

[0011] The input of the decision feedback equalizer is connected to the output of the second forward equalizer, and the output of the decision feedback equalizer is connected to the selector.

[0012] The input of the maximum likelihood sequence detector is connected to the output of the second forward equalizer, and the output is connected to the selector.

[0013] The first forward equalizer is used to equalize the output signal of the analog-to-digital converter and equalize the inter-symbol interference other than the first coefficient, where the first coefficient is the first backward inter-symbol interference value of the channel impulse response.

[0014] The third forward equalizer is used to equalize the first coefficient with a fixed coefficient;

[0015] The first forward equalizer does not include the first postscript tap coefficient.

[0016] To better realize the present invention, the third forward equalizer further includes an adaptive module, a multiplier module, and an adder;

[0017] The input of the adaptive module is connected to the output of the first forward equalizer, and the output is connected to the input of the multiplier module; the output of the multiplier module is connected to the input of the adder.

[0018] The adaptive module includes one adaptive module connected in series; the multiplier module includes one multiplier connected in parallel.

[0019] The inputs of the multipliers are connected between adjacent adaptive modules, and their outputs are connected to the inputs of the adders.

[0020] To better implement the present invention, the coefficient Ck of the third forward equalizer is further defined as (-a1)^k, where a1 = h1 / h0, h0 is the main tap value of the channel impulse response, and k is the coefficient index.

[0021] To better realize the present invention, further, a1∈{0, 1 / 8, 1 / 4, 3 / 8, 1 / 2, 5 / 8, 3 / 4,7 / 8}.

[0022] To better implement the present invention, the input of the maximum likelihood sequence detector further includes the output of the second forward equalizer, h0, and h1, where h0 is a fixed value and h1 = h0 * a1.

[0023] Based on the aforementioned PAM4 adaptive equalization structure based on an ADC architecture, and to better implement this invention, a PAM4 adaptive equalization method based on an ADC architecture is further proposed, implemented based on the aforementioned PAM4 adaptive equalization structure based on an ADC architecture; including the following steps:

[0024] The output signal of the analog-to-digital converter is equalized by the first forward equalizer to equalize the inter-symbol interference outside the first coefficient, where the first coefficient is the first backward inter-symbol interference value h(1) of the channel impulse response.

[0025] The output of the first forward equalizer is provided to the third forward equalizer;

[0026] The first coefficient h1 is equalized by the third forward equalizer with a fixed coefficient, where the coefficient of the third forward equalizer is Ck = (-a1). k Where a1 = h1 / h0, and h0 is the main tap value of the channel impulse response;

[0027] Decision feedback equalization is performed using the output of the third forward equalizer, and / or maximum likelihood sequence detection is performed using the output of the first forward equalizer.

[0028] To better realize the present invention, the invention further includes an optimization step a1:

[0029] In DFE mode, a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum threshold count value as the optimization objective. The threshold count value is the number of signals that fall between the upper and lower threshold values ​​near 0 after the decision feedback equalizer is equalized.

[0030] To better realize the present invention, it is further characterized by including an optimization step a1:

[0031] In MLSD mode, the sender continuously sends training sequences, and a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum number of bit errors as the optimization objective.

[0032] The present invention has the following beneficial effects:

[0033] (1) The present invention eliminates the forward equalizer designed for the clock data recovery module in the traditional structure and replaces it with a third forward equalizer. The coefficients of the third forward equalizer are fixed and do not need to be adaptive. Moreover, the coefficients are finite discrete values, which makes the structure simple and greatly reduces the area and power consumption.

[0034] (2) The present invention determines the sampling point position by a1=h1 / h0. Adjusting the coefficient of a1 can make the system achieve the best performance. a1 is also used as the DFE coefficient and the channel parameter of MLSD, which simplifies the system design.

[0035] (3) This invention provides two a1 optimization methods: the DFE mode takes the minimum threshold count value as the optimization target, and the MLSD mode takes the minimum bit error rate as the optimization target. These methods can adapt to different working modes and optimize system performance. Attached Figure Description

[0036] Figure 1 It uses a traditional equalizer structure.

[0037] Figure 2 This is a schematic diagram of the channel impulse response.

[0038] Figure 3 This is a schematic diagram of the channel response after passing through FFE0.

[0039] Figure 4 This is a schematic diagram of the PAM4 adaptive equalization structure based on the ADC architecture proposed in this invention.

[0040] Figure 5 The FFE1 provided for this invention a A schematic diagram of the channel response after the signal.

[0041] Figure 6 The fixed provided by the present invention FFE Structural diagram.

[0042] Figure 7 The flowchart for optimizing a1 in DFE mode.

[0043] Figure 8 This is a schematic diagram of the signal and threshold after DFE equalization.

[0044] Figure 9 A flowchart for optimizing a1 in MLSD mode. Detailed Implementation

[0045] To more clearly illustrate the technical solutions of the embodiments of the present invention, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments, and therefore should not be regarded as a limitation on the scope of protection. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0046] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set up," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0047] like Figure 1 The diagram shows a traditional equalization structure. The upper path passes through the forward equalizer FFE0 and then enters the decision unit for signal decision. The decision result is given to the clock data recovery module (CDR) and the adaptive structure. The lower path passes through the longer forward equalizer FFE1 and then enters the DFE decision unit. The FFE1 filter does not include the first back tap coefficient c1.

[0048] The DFE decision maker is a decision maker containing 1 tap of DFE coefficients. Both the DFE and FFE coefficients need to be adaptive.

[0049] The output yk of the forward equalizer FFE1 can also be used as the decision output for the maximum likelihood sequence detector MLSD. The final output is selected between the DFE decision unit and the maximum likelihood sequence detector MLSD.

[0050] The above structure is a general structure, including a forward equalizer (FFE), a decision unit, a DFE decision unit, an adaptive structure, and a clock data recovery module (CDR).

[0051] The TX signal, after passing through the channel and ADC, is represented as follows:

[0052] =

[0053] h is the impulse response of the channel model, such as Figure 2 As shown in the impulse response, h(1), h(2)...h(-1), h(-2)... are called ISI.

[0054] FFE0: = ;

[0055] D0(n) = sign[ ]; cdrpostn is the number of FFE0 post equilibrium, cdrpren is the number of FFE0 pre equilibrium, cffe0(k) represents the coefficient of the FFE0 structure at position k.

[0056] The impulse response after FFE0 is as follows: Figure 3 As shown, the range of both cdrpren and cdrpostn is 0.

[0057] The clock data recovery module (CDR) uses a Mueller-Müller CDR, which, after locking, satisfies h(-1) = h(1). Since FFE0 can make h(-1) = 0 and h(1) = 0, both satisfying the CDR locking condition, the CDR sampling position can be any point. Therefore, CDR and FFE0 influence each other, resulting in the position of h(0) being non-unique. Shifting h(0) to the left can decrease h(-1) and increase h(1), thereby reducing the FFE coefficient to reduce the amplification of noise by FFE, which is beneficial to increasing the signal-to-noise ratio after FFE1. However, it will decrease h(0), which may worsen the signal-to-noise ratio. At the same time, increasing the ratio of h(1) / h(0) will increase the DFE coefficient, increase DFE error propagation, and degrade performance. A suitable h(0) sampling point is required.

[0058] FFE1: = ;

[0059] dfe: = D(n)=[ ];

[0060] [ [Indicates to] The verdict for PAM4 is -3, -1, 1, 3. is the dfe coefficient.

[0061] The inputs to MLSD are y(n), h(0), and h(1). Its error propagation characteristics are also related to the ratio of h(1) / h(0). If the ratio is too large, the performance of MLSD will decrease.

[0062] Example 1:

[0063] This embodiment proposes a PAM4 adaptive equalization structure based on an ADC architecture, such as... Figure 4 As shown, it is connected to an analog-to-digital converter; it includes a first forward equalizer, a second forward equalizer, a third forward equalizer, a decision unit, a decision feedback equalizer, a maximum likelihood sequence detector, a clock data recovery module, and a selector;

[0064] The input terminal of the first forward equalizer is connected to the output terminal of the analog-to-digital converter, and the output terminal of the first forward equalizer is connected to the input terminals of the second forward equalizer and the third forward equalizer.

[0065] The input terminal of the decision unit is connected to the output terminal of the third forward equalizer, and the output terminal of the decision unit is connected to the input terminal of the clock data recovery module.

[0066] The input of the decision feedback equalizer is connected to the output of the second forward equalizer, and the output of the decision feedback equalizer is connected to the selector.

[0067] The input of the maximum likelihood sequence detector is connected to the output of the second forward equalizer, and the output is connected to the selector.

[0068] The first forward equalizer is used to equalize the output signal of the analog-to-digital converter and equalize the inter-symbol interference other than the first coefficient, where the first coefficient is the first backward inter-symbol interference value of the channel impulse response.

[0069] The third forward equalizer is used to equalize the first coefficient with a fixed coefficient;

[0070] The first forward equalizer does not include the first postscript tap coefficient.

[0071] Working principle: This embodiment removes the traditional forward equalizer FFE0 and adds a third forward equalizer, fixed. FFE (Fixed coefficients FFE) The structure is simple, significantly reducing power consumption and area compared to the forward equalizer FFE0. The original forward equalizer FFE1 is split into a first forward equalizer FFE1. a Second forward equalizer FFE1 b First forward equalizer FFE1 a The result y0(n) is given to the CDR path of the clock data recovery module above.

[0072] The splitting method is: FFE1 a The tap is less than the original FFE0 tap by c1, so the following formula is FFE1. a In the principle formula, the tap positions range from -cdrpren to cdrpostn, but do not include the position where k=1 (where k=1 represents c1). Since the original FFE0 is eliminated, there are no concerns about power consumption or area cost; here, FFE1... a Increasing the coefficient range can help reduce ISI and thus reduce CDR jitter.

[0073] FFE1 b The principle formula is as follows. According to the formula, it can be seen that it includes the original FFE1 excluding FFE1. a The other parts.

[0074] FFE1 a : =

[0075] FFE1 b : =

[0076] First forward equalizer FFE1 a The coefficient c1 is not included, meaning h(1) will not be equalized. The third forward equalizer is Fixed. FFE Responsible for equalizing h(1). After passing through the first forward equalizer FFE1 a Post-channel response such as Figure 5 As shown.

[0077] First forward equalizer FFE1 a The result includes h(1), the third forward equalizer is fixed. FFE The third forward equalizer is responsible for equalizing h(1), where a1 is a fixed coefficient and can take one of [0, 1 / 8, 1 / 4, 3 / 8, 1 / 2, 5 / 8, 3 / 4, 7 / 8]. Due to the characteristics of the Clock Data Recovery (CDR) module, the third forward equalizer is fixed. FFE The equilibrium response is represented as heq , where h eq (-1)=h eq (1) = 0, and the first forward equalizer FFE1 a This makes h(-1)=0. Therefore, regardless of the value of the a1 coefficient, according to the principle of CDR, CDR will adjust the phase, which will make the equalized h... eq (1) = 0. Meanwhile, a1 is responsible for balancing h(1) to 0. According to the zero-forcing algorithm, the ratio of h(1) / h(0) is equal to a1. With a1 fixed, the ratio of h(1) / h(0) is fixed, i.e., h(1) = h(0) * a1. Figure 2 In the impulse response shown, the point h(0) corresponding to the left shift is... l h(1) l Shift the corresponding point h(0) to the right. r h(1) r The ratio of h(1) / h(0) is different when shifting left and right, a1 l >a1 r That is, the location of the sampling point can be determined by the size of a1.

[0078] Example 2:

[0079] This embodiment is based on the above embodiment 1, such as... Figure 6 As shown, a specific embodiment of the third forward equalizer is described in detail.

[0080] The third forward equalizer includes an adaptive module, a multiplier module, and an adder;

[0081] The input of the adaptive module is connected to the output of the first forward equalizer, and the output is connected to the input of the multiplier module; the output of the multiplier module is connected to the input of the adder.

[0082] The adaptive module includes one adaptive module connected in series; the multiplier module includes one multiplier connected in parallel.

[0083] The inputs of the multipliers are connected between adjacent adaptive modules, and their outputs are connected to the inputs of the adders.

[0084] The coefficient Ck of the third forward equalizer is (-a1). k h(0) is the main tap value of the channel impulse response, and k is the coefficient index.

[0085] Where a1∈{0, 1 / 8, 1 / 4, 3 / 8, 1 / 2, 5 / 8, 3 / 4, 7 / 8}.

[0086] Working principle: 1) and 2) below can naturally satisfy the objective of the zero-forcing algorithm. Since cdr makes h eq If (1) = 0, then c1 can satisfy the goal of the zero-forcing algorithm with any value. c2, c3... after c1 can be calculated according to the zero-forcing algorithm.

[0087] 1) The relationship satisfied by the Mueller-Muller CDR is h eq (-1)=h eq (1) = 0,

[0088] 2) ZF (zero forced) algorithm objective: h eq (1) = 0;

[0089] According to the zero-forcing algorithm, Fixed FFE The coefficients Ck are as follows:

[0090] Ck=(-a1) k

[0091] Meanwhile, in the original zero-forcing algorithm, c1=-h(1) / h(0), which is based on the absence of cdr. Due to the existence of cdr and the fixed value of c1, cdr finds the phase that satisfies c1=-h(1) / h(0) from the impulse response, so that after cdr is locked, h(1)=h(0)*a1 is satisfied.

[0092] The maximum length l is selected according to the size of a1. The longer the length, the smaller the residual ISI. In this embodiment, l=7.

[0093] Meanwhile, h(1) = h(0) * a1, where a1 is the dfe coefficient. In MLSD, h(0) is a fixed value, and h(1) can also be calculated by h(0) * a1.

[0094] The other parts of this embodiment are the same as those in Embodiment 1 above, so they will not be described again.

[0095] Example 3:

[0096] This embodiment is based on any one of Embodiments 1-2 above, such as Figure 7 , Figure 8 , Figure 9 As shown, a specific embodiment of the PAM4 adaptive equalization method based on an ADC architecture is described in detail, including the following steps:

[0097] The output signal of the analog-to-digital converter is equalized by the first forward equalizer to equalize the inter-symbol interference outside the first coefficient, where the first coefficient is the first backward inter-symbol interference value h(1) of the channel impulse response.

[0098] The output of the first forward equalizer is provided to the third forward equalizer;

[0099] The first coefficient h(1) is equalized by the third forward equalizer with a fixed coefficient, where the coefficient of the third forward equalizer is Ck=(-a1). k , where a1=h(1) / h(0), and h(0) is the main tap value of the channel impulse response;

[0100] Decision feedback equalization is performed using the output of the third forward equalizer, and / or maximum likelihood sequence detection is performed using the output of the first forward equalizer.

[0101] It also includes the a1 optimization step:

[0102] In DFE mode, a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum threshold count value as the optimization objective. The threshold count value is the number of signals that fall between the upper and lower threshold values ​​near 0 after the decision feedback equalizer is equalized.

[0103] In MLSD mode, the sender continuously sends training sequences, and a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum number of bit errors as the optimization objective.

[0104] Working principle: In this embodiment, the size of a1 affects the performance of dfe and MLSD as described above. At the same time, the structure of a1 is optimized, and the two modes of dfe and MLSD are different.

[0105] In Dfe mode: the signal amplitude is adjusted to a fixed value by c0 of FFE1. Then the optimal a1 value can be determined by minimizing the noise value. From the minimum to the maximum, the a1 corresponding to the minimum noise is the optimal a1, where c0 is the main tap coefficient in FFE.

[0106] Initially, a1 is the minimum value, and the threshold count is calculated.

[0107] After DFE equalization with a threshold count value near 0, the signal falls above the upper threshold up. thd and lower threshold dn thd The number between them, up_thd and dn-thd are illustrated in Figure 8.

[0108] If a1 does not reach its maximum, a1 is increased by 1 / 8, and then the count returns to the threshold.

[0109] The process ends when a1 reaches its maximum. The a1 corresponding to the minimum threshold count value during the scanning process is the optimal a1.

[0110] MLSD mode: The signal after MLSD does not contain noise, so a1 cannot be optimized based on the noise value. The transmitter sends the training sequence, calculates the bit error rate after MLSD, and uses the lowest bit error rate as the optimization target for a1. The process is as follows:

[0111] 1) The sending end continuously sends training sequences until a1 optimization is complete.

[0112] 2) Start with a1 as the minimum value and count the number of bit errors.

[0113] 3) If a1 does not reach its maximum, increase a1 by 1 / 8 and then return to the error statistics.

[0114] 4) The process ends when a1 reaches its maximum. The a1 corresponding to the minimum number of errors during the scanning process is the optimal a1.

[0115] The other parts of this embodiment are the same as any one of the above embodiments 1-2, so they will not be described again.

[0116] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Any simple modifications or equivalent changes made to the above embodiments based on the technical essence of the present invention shall fall within the protection scope of the present invention.

Claims

1. A PAM4 adaptive equalization structure based on an ADC architecture, connected to an analog-to-digital converter; characterized in that, It includes a first forward equalizer, a second forward equalizer, a third forward equalizer, a decision unit, a decision feedback equalizer, a maximum likelihood sequence detector, a clock data recovery module, and a selector; The input terminal of the first forward equalizer is connected to the output terminal of the analog-to-digital converter, and the output terminal of the first forward equalizer is connected to the input terminals of the second forward equalizer and the third forward equalizer. The input terminal of the decision unit is connected to the output terminal of the third forward equalizer, and the output terminal of the decision unit is connected to the input terminal of the clock data recovery module. The input of the decision feedback equalizer is connected to the output of the second forward equalizer, and the output of the decision feedback equalizer is connected to the selector. The input of the maximum likelihood sequence detector is connected to the output of the second forward equalizer, and the output is connected to the selector; the first forward equalizer is used to equalize the output signal of the analog-to-digital converter and equalize the inter-symbol interference other than the first coefficient, where the first coefficient is the first backward inter-symbol interference value of the channel impulse response. The third forward equalizer is used to equalize the first coefficient with a fixed coefficient; The first forward equalizer does not include the first postscript tap coefficient.

2. The PAM4 adaptive equalization structure based on ADC architecture according to claim 1, characterized in that, The third forward equalizer includes an adaptive module, a multiplier module, and an adder; The input of the adaptive module is connected to the output of the first forward equalizer, and the output is connected to the input of the multiplier module; the output of the multiplier module is connected to the input of the adder. The adaptive module includes one adaptive module connected in series; the multiplier module includes one multiplier connected in parallel. The inputs of the multipliers are connected between adjacent adaptive modules, and their outputs are connected to the inputs of the adders.

3. The PAM4 adaptive equalization structure based on ADC architecture according to claim 1, characterized in that, The specific structure of the first forward equalizer is as follows: = ; in, is the output of the first forward equalizer, cdrpostn is the number of back tap equalizers of the first forward equalizer, cdrpren is the number of front tap equalizers of the first forward equalizer, k is the set coefficient index, and cffe1(k) represents the coefficient of the first forward equalizer structure at position k.

4. The PAM4 adaptive equalization structure based on ADC architecture according to claim 3, characterized in that, The specific structure of the second forward equalizer is as follows: = ; in, This indicates the front tap boundary of the second forward equalizer. This indicates the boundary of the back tap of the second forward equalizer.

5. The PAM4 adaptive equalization structure based on ADC architecture according to claim 1, characterized in that, The coefficient Ck of the third forward equalizer is (-a1). k k is the set coefficient index.

6. The PAM4 adaptive equalization structure based on ADC architecture according to claim 1, characterized in that, a1∈{0,1 / 8, 1 / 4, 3 / 8, 1 / 2, 5 / 8, 3 / 4, 7 / 8}。 7. The PAM4 adaptive equalization structure based on ADC architecture according to claim 1, characterized in that, The inputs of the maximum likelihood sequence detector include the output of the second forward equalizer, h(0) and h(1), where h(0) is a fixed value and h(1) = h(0) * a1.

8. A PAM4 adaptive equalization method based on an ADC architecture, implemented based on the PAM4 adaptive equalization structure based on an ADC architecture as described in claim 1; characterized in that, Includes the following steps: The output signal of the analog-to-digital converter is equalized by the first forward equalizer to equalize the inter-symbol interference outside the first coefficient, where the first coefficient is the first backward inter-symbol interference value h(1) of the channel impulse response. The output of the first forward equalizer is provided to the third forward equalizer; The first coefficient h1 is equalized by the third forward equalizer with a fixed coefficient, where the coefficient of the third forward equalizer is Ck = (-a1). k ; Decision feedback equalization is performed using the output of the third forward equalizer, and / or maximum likelihood sequence detection is performed using the output of the first forward equalizer.

9. The PAM4 adaptive equalization method based on an ADC architecture according to claim 8, characterized in that, It also includes the a1 optimization step: In DFE mode, a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum threshold count value as the optimization objective. The threshold count value is the number of times the signal falls between the upper and lower threshold values ​​near 0 after equalization by the decision feedback equalizer.

10. The PAM4 adaptive equalization method based on an ADC architecture according to claim 8, characterized in that, It also includes the a1 optimization step: In MLSD mode, the sender continuously sends training sequences, and a1 is scanned from the minimum value to the maximum value, increasing by 1 / 8 each time. The optimal a1 is determined with the minimum number of bit errors as the optimization objective.