A Dynamic Nonlinear Mismatch Correction Method for TIADC Based on GRU Network

In the study of high-speed and high-precision analog-to-digital converter (ADC), time-interleaved ADC (TIADC) is one of the important implementation methods. In high-precision TIADC, calibration of mismatch, especially dynamic nonlinear mismatch, is of great research value. Aiming at the dynamic nonlinear mismatch in the TIADC systems, a dynamic nonlinear mismatch calibration method based on gated recurrent unit (GRU) network is studied here. The calibration process is included two parts: error detection, and error correction. In the error detection part, series of dynamic nonlinear error is inserted into a standard sine wave signal. This sine signal with dynamic nonlinear error is used as test signal for the TIADC system. Its output and the ideal input (standard sine signal) are sent to train the GRU network. In the calibration process, the actual input signal is fed into the TIADC system, the mismatch of the input signal is calibrated by the trained GRU network. By these training and calibration process, an 18-bit 1GS/S 4-channel TIADC system is simulated, and the dynamic nonlinear mismatch is calibrated well. Average SNR has increased from 30.84 to 100.65 dB, ENOB from 4.80 to 16.43 bit, and SFDR from 33.20 to 126.68 dBc. Compared with RNN and LSTM methods, this method has higher detection rate and accuracy. The results show that the calibration method of GRU network has fast detection speed, high detection accuracy and calibration accuracy.