Study On High-performance SAR ADC Background Calibration Technology

The trend of upgrading electronic products has prompted ADCs to develop towards high performance and high energy efficiency.As the first choice for low-power ADCs under advanced technology,SAR ADC faces many problems in the process of moving towards high-speed and high-precision.Due to the extraordinary speed,power efficiency,and integration level of large-scale digital circuits in advanced technology,and the relatively complete theoretical system of digital signal processing,it is possible to solve the difficult problems in high-performance ADC design with the digitally-assisted algorithms.Through the analysis of SAR ADC principle and error,this paper presents a background calibration algorithm based on the "Split" structure for the static capacitance mismatch of high-precision SAR ADC.Compared with the traditional foreground self-calibration technology and background calibration technology based on known signal injection,it reduces the additional analog circuit overhead and the impact of signal injection on the dynamic range of the ADC.At the same time,for the more diverse and complex errors caused by some new improvements or structures in high-speed and high-resolution SAR ADCs,using the powerful nonlinear fitting and generalization capabilities of the neural network,a low-speed,a feed-forward neural network background calibration algorithm with a low-speed reference ADC is proposed.And for the calibration requirements of high-performance ADCs,a suitbable error back propagation algorithm and neural network structure are designed to achieve a low complexity and high efficiency neural network background calibration.Through the simulation of the two algorithms,the results show that under the background calibration algorithm based on the "split" structure,the ENOB of the two 14-bit sub-ADCs with static capacitive mismatch is increased from 8.11 bit and 8.05 bit to 13.84 bit,the SNR is increased from 53.5d B and 53.0d B to 85.01 d B,and the entire calibration can achieve convergence at about 19600 samples,which verifying the effectiveness of the algorithm.Based on the background calibration technology of the neural network,when the input signal frequency is close to the Nyquist sampling frequency,the ENOB of a 14-bit 500 Msps ADC with error is increased from 7.85 bit to 13.08 bit.The SNR and SFDR are increased from 57.2d B and 49.7d B to 80.40 d B and 106.80 d B,the entire calibration only requires about 40,000 samples of data,which can improve the overall non-ideality caused by various errors within the ADC and has high calibration efficiency.Under different frequency input signals,the ENOB of the calibrated ADC kept above 13 digits,which verifying the effectiveness of the algorithm.