A pipeline analog-to-digital converter calibration method

By embedding a tree-structured decoder and a gain error calibration module in the first two stages of the pipelined analog-to-digital converter, the noise and gain error problems caused by capacitor mismatch are solved, thereby improving the signal-to-noise ratio and spurious-free dynamic range and reducing hardware resource consumption.

CN119906425BActive Publication Date: 2026-11-10SHANGHAI INST OF MICROSYSTEM & INFORMATION TECH CHINESE ACAD OF SCI
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202411846811.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-12-16
Publication Date
2026-11-10
Estimated Expiration
2044-12-16

AI Technical Summary

Technical Problem

Existing pipeline analog-to-digital converters (ADCs) struggle to simultaneously improve signal-to-noise ratio and spurious-free dynamic range in the face of noise and gain errors caused by capacitor mismatch. Existing calibration methods suffer from wasted hardware resources or reduced signal-to-noise ratio.

Method used

A tree-based random matching module is embedded in the first two sub-stages of the pipelined analog-to-digital converter. Combined with gain error calibration and mismatch noise cancellation modules, capacitor mismatch calibration and noise cancellation are performed using pseudo-random signals. Variable step size control coefficients are used to improve calibration accuracy and speed.

Benefits of technology

It effectively eliminates capacitor mismatch noise, improves the signal-to-noise ratio and spurious-free dynamic range, reduces hardware resource consumption, and enhances signal processing accuracy and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119906425B_ABST
    Figure CN119906425B_ABST
Patent Text Reader

Abstract

This invention relates to a pipelined analog-to-digital converter (ADC) calibration method, comprising the following steps: S0 setting a target calibration sub-stage, the target calibration sub-stage including the first two pipelined sub-stages of the pipelined ADC; S1 embedding a tree-based random matching module between the Flash ADC and DAC arrays within each target calibration sub-stage, so that the digital code output of the current pipelined sub-stage is randomly matched with the sampling capacitor; S2 constructing a gain error calibration module to perform gain error calibration at each target calibration sub-stage based on the subsequent quantization output, obtaining the corresponding quantization output; S3 constructing a mismatch noise cancellation module to perform mismatch noise cancellation on the quantization output, and adding the obtained quantization output to the digital code output of the corresponding target calibration sub-stage after gain error calibration. This invention can simultaneously improve the signal-to-noise ratio and spurious-free dynamic range of the pipelined ADC.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of analog-to-digital converter technology, and in particular to a pipeline analog-to-digital converter calibration method. Background Technology

[0002] Pipeline analog-to-digital converters (ADCs) are widely used due to their ability to simultaneously achieve high speed and high accuracy. Unlike noise in subsequent stages of a pipeline ADC, noise introduced in the first stage cannot be suppressed using the characteristics of the pipeline, and therefore often becomes the dominant noise. Furthermore, as integrated circuit dimensions shrink, performance degradation caused by parasitic effects and device mismatch becomes increasingly apparent.

[0003] Pipeline ADCs rely on precise capacitor ratios for conversion. Mismatch between the sampling and feedback capacitors introduces input signal-related errors during conversion, resulting in high-energy harmonic characteristics in the spectrum and deteriorating spurious-free dynamic range. Simultaneously, capacitor mismatch also introduces gain errors (including linear and nonlinear errors), manifesting as over-range or curve distortion in the transmission curve, leading to code loss and spectral spurs. This problem is exacerbated when the operational amplifier's gain and bandwidth are insufficient.

[0004] For capacitor mismatch in Pipeline ADCs, a common method is front-end calibration. However, this method requires interrupting the normal operation of the ADC, and the chip can only function after calibration is complete. Another approach utilizes the statistical characteristics of pseudo-random signals for capacitor mismatch calibration, but this requires inputting the random signal along with the signal source, which reduces the input signal range—not always acceptable. Dynamic component matching techniques can also be used to convert the harmonics caused by mismatch into random noise, but this increases the noise floor, sacrificing the signal-to-noise ratio to improve spurious-free dynamic range, and does not fundamentally solve the problem. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a pipeline analog-to-digital converter (ADC) calibration method that can simultaneously improve the signal-to-noise ratio and spurious-free dynamic range of the pipeline ADC.

[0006] The technical solution adopted by this invention to solve its technical problem is: to provide a pipeline analog-to-digital converter calibration method, comprising the following steps:

[0007] S0 sets a target correction sub-stage, which includes the first two pipeline sub-stages of the pipelined analog-to-digital converter;

[0008] S1 embeds a tree-based random matching module between the FlashADC and DAC arrays within each target correction sub-stage, so that the digital code output by the current pipeline sub-stage is randomly matched with the sampling capacitor.

[0009] S2 constructs a gain error calibration module to quantize the res output of each target correction substage. backend Perform gain error calibration at this level to obtain the corresponding level of quantization output res. bkcal ;

[0010] S3 constructs a mismatch noise cancellation module to separately process the quantized output res bkcal Perform mismatch noise cancellation and then convert the resulting quantized output res bkcal&MNC Add the digital code of the corresponding target correction sub-stage output after gain error calibration.

[0011] Furthermore, the random matching module includes:

[0012] The PRBS module is used to generate a pseudo-random signal. The first two bits of the pseudo-random signal are respectively fed into the corresponding high-precision DAC in the DAC array for processing to obtain a gain error calibration signal.

[0013] The tree-structured decoder module includes multiple hierarchical decision units, which are used to perform multi-level modulation of the digital code using all bits of the pseudo-random signal except for the first two bits. Each bit of the generated modulated random signal is connected to the corresponding low-precision DAC in the DAC array for processing.

[0014] Furthermore, the tree decoder module is a multi-level tree structure with hierarchical decision units as nodes. Each parent node in this multi-level tree structure has two child nodes, and the digital code input to each child node is the digital code output by its parent node. The pseudo-random signal input to each child node consists of uncorrelated different bits.

[0015] Furthermore, the hierarchical decision maker is used for:

[0016] Input numeric code X k,r Perform parity determination;

[0017] The result of the parity determination is modulated using a 1-bit pseudo-random signal to output a modulated random signal s. k,r ;

[0018] Using input numeric code X k,r and modulated random signal s k,r The output of the current level decision maker is generated as follows

[0019]

[0020] Where k is the level of the current level decision maker in the tree structure, and r is the order of the current level decision maker in level k.

[0021] Furthermore, the target correction sublevel is a pipeline sublevel with a 9-level structure, and the tree decoder module is a three-level tree structure.

[0022] Furthermore, the quantization output res bkcal Mismatch noise cancellation includes:

[0023] For the quantized output res bkcal With any of the modulated random signals s k,r The product is smoothed, and then the corresponding capacitance mismatch coefficient Δ is estimated based on the processed result. s,r ;

[0024] Obtain all of the modulated random signals s k,r The corresponding capacitor mismatch coefficient Δ s,r The capacitor mismatch error e is calculated. DAC1 =ΣΔ s,r ×s k,r ;

[0025] The capacitor mismatch error e DAC1 With the quantization output res bkcal The quantized output res is obtained by adding them together. bkcal&MNC .

[0026] Furthermore, the smoothing process is achieved by employing one of the following: exponential moving average, simple moving average, and weighted moving average.

[0027] Furthermore, the hardware logic representation of the gain error calibration module is as follows:

[0028] Post-quantization output res bkcal =Post-quantization output res backend ×W+Gain Error Calibration Signal PN, where W is the gain error calibration coefficient.

[0029] Furthermore, the gain error calibration coefficient W is updated based on the following criteria.

[0030] W[n+1]=W[n]

[0031] +μ vs ×Gain error calibration signal PN [Post-stage quantization output res] backend -Gain error calibration signal PN]

[0032] Where, μ vs Convergence coefficient.

[0033] Furthermore, the convergence coefficient μ vs Configure it using the following method:

[0034] Initially set to a large step size;

[0035] When the gain error calibration coefficient W converges to near the steady-state value, it is set to a small step size.

[0036] Beneficial effects

[0037] Due to the adoption of the above-mentioned technical solutions, this invention has the following advantages and positive effects compared with the prior art: This invention transforms the unknown capacitor mismatch coefficient into a calculable constant by using a tree-shaped decoding structure controlled by a pseudo-random sequence; the mismatch noise cancellation module utilizes exponential moving average to effectively save hardware resources and has low computational cost; this invention overcomes the disadvantage that the harmonics related to the input signal are scattered into the noise floor after using dynamic element matching technology on the sampling capacitor, resulting in a decrease in signal-to-noise ratio, and can simultaneously improve the signal-to-noise ratio and spurious-free dynamic range; before performing mismatch noise cancellation, this invention performs gain error calibration, improving the data accuracy fed into the noise cancellation module, and the gain error calibration module uses variable step size control coefficients, which can simultaneously consider the convergence accuracy and speed of the gain error. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the pipeline analog-to-digital converter structure according to an embodiment of the present invention;

[0039] Figure 2 This is a diagram of the internal structure of a 9-level pipeline sub-level according to an embodiment of the present invention;

[0040] Figure 3 This is a tree-shaped DEM DAC structure diagram according to an embodiment of the present invention;

[0041] Figure 4 This is a functional diagram illustrating the hierarchical decision-maker according to an embodiment of the present invention;

[0042] Figure 5 This is a schematic diagram of the gain error calibration module and mismatch noise cancellation module according to an embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram illustrating the gain error calibration and mismatch noise elimination principle of an embodiment of the present invention. Detailed Implementation

[0044] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.

[0045] The present invention relates to a pipeline analog-to-digital converter calibration method, comprising the following steps:

[0046] S0 sets a target correction sub-stage, which includes the first two pipeline sub-stages of the pipelined analog-to-digital converter;

[0047] S1 embeds a tree-based random matching module between the FlashADC and DAC arrays within the target correction substage, so that the thermometer code output by the current pipeline substage is randomly matched with the sampling capacitor.

[0048] S2 constructs a gain error calibration module to quantize the Res output of the subsequent stage of the target correction substage. backend Perform gain error calibration to obtain the quantized output Res bkcal ;

[0049] S3 constructs a mismatch noise cancellation module to process the quantized output Res. bkcal Perform mismatch noise cancellation and then convert the resulting quantized output Res bkcal&MNC .

[0050] like Figure 1 The diagram shows a block diagram of a 14-bit pipeline ADC that applies Gain Error Correction (GEC) and Mismatch Noise Cancellation (MNC). The first two stages employ calibration and noise cancellation techniques. To accommodate the calibration circuitry, the first two stages utilize a 9-level pipelined sub-stage, while the subsequent two stages use 5-level sub-stages. The 9-level sub-stage is shown below. Figure 2 As shown.

[0051] Compared to conventional redundant substages, this invention randomizes the one-to-one correspondence between the thermometer code output by the comparator group and the sampling capacitor, thereby eliminating the correlation between output error and input. The thermometer code output by the FlashADC is randomly selected for sampling after passing through a dynamically matched random decoder. The randomization method employs a tree-structured decoder, as shown in the diagram. Figure 3As shown in the diagram, signals PN1 and PN2 are processed by two sets of Nb DACs (high-precision DACs), and the resulting signal PN is used for gain error calibration. Here, Nbit represents the DAC precision, determined by the capacitor C connected to the random signal. random and a single sampling capacitor C s0 The relative size is determined, such as but This value can be flexibly adjusted according to the magnitude of the injected random signal.

[0052] Figure 3 S k,r As a hierarchical decision maker, its function is as follows: Figure 4 As shown. After the input signal is parity checked, the output is 0 (representing an even input) or 1 (representing an odd input). This result is then modulated by a 1-bit pseudo-random signal (this pseudo-random signal is a binary sequence that outputs -1 or +1, with a mean of 0) and then outputs s. k,r Enter the numeric code and s k,r The adder generates the input for the next level decision unit.

[0053]

[0054] The calculation process for gain error calibration and mismatch noise cancellation techniques is as follows: Figure 5 As shown. Where Res backend It is the post-quantization output of the pipeline ADC, Res bkcal It is the quantized output after gain error calibration, Res bkcal&MNC It is the quantization output after gain error calibration and mismatch noise cancellation, as detailed below.

[0055] like Figure 6 As shown, taking the first-level calibration as an example, the outputs of the FlashADC and DAC are as follows:

[0056] AD1 = Vin + e ADC1 DA1 = Vin + e ADC1 +e DAC1

[0057] The op-amp's input r1 is

[0058] r1=PN-e ADC1 -e DAC1

[0059] When insufficient gain or bandwidth of the op-amp leads to gain error, the final output r of the MDAC will be affected. g1 It can be represented as: r g1=4(1+a1)r1, where a1 is the total error coefficient caused by the gain error and setup error due to the op-amp gain and bandwidth. The output after further quantization is:

[0060] res backend =(1+a1)r1=(1+a1)(PN-e ADC1 -e DAC1 )+e ADC2 / 4

[0061] Among them, e ADC1 and e DAC1 These represent the errors of the Flash ADC and DAC at this level, e ADC2 This is the error in the subsequent stage.

[0062] This output is sent to, as follows: Figure 5 The gain error calibration module shown:

[0063]

[0064] when hour:

[0065]

[0066] When e DAC1 When the noise is absent or eliminated, this result is close to the ideal quantization output of the subsequent stage. This involves DAC mismatch noise cancellation, i.e., e DAC1 Performing gain error calibration before noise cancellation can accelerate the estimation speed and stability of the noise cancellation module. The updating of the coefficient W in the gain error calibration module depends on the following criteria:

[0067]

[0068] Due to the zero-mean property of pseudo-random numbers PN, and e ADC1 e DAC1 Unrelated, there are:

[0069]

[0070] W will eventually converge to μ in this estimation module vs μ is related to convergence speed and convergence accuracy. vs The larger the step size, the faster the speed but the lower the accuracy. A variable step size algorithm is adopted, using a large step size in the early stage to quickly converge to near the steady state value, and then using a small step size to improve the convergence accuracy.

[0071] The output after gain error calibration is sent to the mismatch noise cancellation module for further estimation of e. DAC1 .

[0072] like Figure 3 As shown, after the DAC undergoes tree-based random decoding, the previously unknown mismatch noise related to the input signal (introducing input-related harmonics, which can be dispersed into the noise floor by the DEM) will have the following form:

[0073]

[0074] Where Δ k,r It is a constant determined by capacitance mismatch, s k,r It is a known set of uncorrelated random signals. The output, after gain error calibration, is sent to the mismatch noise cancellation module, and the result is compared with the s given by the hierarchical decision unit. k,r Perform the relevant calculations and apply an exponential moving average; in this example, s k,r There are 7 groups, so 7 channels are needed to estimate Δ. k,r Taking channel 1 as an example, the estimation principle is as follows:

[0075]

[0076] EMA(res bkcal ×s 3,1 )=Δ 3,1 ×s 3,1 2

[0077] s 3,1 It is a binary sequence (-1 or 1) with a mean of 0, therefore s 3,1 2 =1, and the mismatch coefficient Δ can be calculated at the end. 3,1 .

[0078] 7 by estimating Δ k,r Afterwards, further with s k,r By multiplying and summing the results, the mismatch error e can be estimated. DAC1 .

[0079] Adding this value to the output after gain error calibration yields a quantized output that eliminates mismatch noise and gain errors.

[0080]

[0081] final: The quantification results are close to the ideal results.

Claims

1. A method for correcting a pipelined analog-to-digital converter, characterized in that, Includes the following steps: S0: Set the target correction sub-stage, which includes the first two pipeline sub-stages of the pipelined analog-to-digital converter; S1: Embed a tree-based random matching module between the Flash ADC and DAC arrays within each target correction sub-stage, so that the digital code output by the current pipeline sub-stage is randomly matched with the sampling capacitor; S2: Construct a gain error calibration module to quantize the output of each target correction substage. Perform gain error calibration at this level to obtain the corresponding level of quantization output. ; S3: Construct a mismatch noise cancellation module to separately process the quantized output. Perform mismatch noise cancellation and quantize the resulting output. Add the digital code of the corresponding target correction substage output after gain error calibration; The random matching module includes: The PRBS module is used to generate a pseudo-random signal. The first two bits of the pseudo-random signal are respectively fed into the corresponding high-precision DAC in the DAC array for processing to obtain the gain error calibration signal. ; The tree-structured decoder module includes multiple hierarchical decision units, which are used to perform multi-level modulation of the digital code using all bits of the pseudo-random signal except for the first two bits. Each bit of the generated modulated random signal is connected to the corresponding low-precision DAC in the DAC array for processing. The tree decoder module is a multi-level tree structure with hierarchical decision units as nodes. Each parent node in the multi-level tree structure has two child nodes, and the digital code input to each child node is the digital code output by its parent node. The pseudo-random signal input to each child node is an uncorrelated different bit. The hierarchical decision-maker is used for: Input numeric code Perform parity determination; The parity determination result is modulated using a 1-bit pseudo-random signal to output a modulated random signal. ; Using input numeric codes and modulated random signals The output of the current level decision maker is generated as follows in, This refers to the current level of the decision maker within the tree structure. For the current level decision maker at the level The sorting within.

2. The correction method according to claim 1, characterized in that, The target correction sublevel is a pipelined sublevel with a 9-level structure, and the tree decoder module is a three-level tree structure.

3. The correction method according to claim 1, characterized in that, The quantization output Mismatch noise cancellation includes: For the quantization output With any of the modulated random signals The product is smoothed, and then the corresponding capacitance mismatch coefficient is estimated based on the processed result. ; Obtain all of the modulated random signals Corresponding capacitor mismatch coefficient The capacitor mismatch error was calculated. ; The capacitor mismatch error With the quantization output The quantized output is obtained by adding them together. .

4. The correction method according to claim 3, characterized in that, The smoothing process is achieved by employing one of the following: exponential moving average, simple moving average, and weighted moving average.

5. The correction method according to claim 1, characterized in that, The hardware logic representation of the gain error calibration module is as follows: in, This is the gain error calibration coefficient.

6. The correction method according to claim 5, characterized in that, The gain error calibration coefficient Updated based on the following criteria in, is the convergence coefficient.

7. The correction method according to claim 6, characterized in that, The convergence coefficient Configure it using the following method: Initially set to a large step size; When the gain error calibration coefficient When the convergence reaches the vicinity of the steady-state value, a small step size is set.

Citation Information

Patent Citations

  • Error calibration method and device and analog-to-digital conversion circuit

    CN116781076A