Signal Reconstruction In Time-interleaved ADC System

The rapid development of electronics and information technology dramatically increases the operation frequency and complexity of signal in various electronic systems. As the basis of signal digitalization, high speed and accurate sampling technology is the key technique to obtain and analyze tiny characteristics of complex signal in order to distill useful information. Limited by lagged chip technology, even the speed of fastest flash analog to digital converter (ADC) falls far behind the pratical request. The time-interleaved ADC (TIADC) system achieves the multiplication of sampling rate by multiplexing a bank of slow ADCs operating in parallel. Though the sampling rate of TIADC is greatly higher than single ADC, the difference between ADC chips and clock signal in each channel generates the channel mismatches, which include the gain, offset and time mismatch. The channel mismatches produce distortions in the output spectrum of TIADC, and degrade significantly the resolution of the sampling system, which disbennifit the research of specific characteristics of complex transient signal. So the output signal should be properly reconstructed, in order to decrease channel mismatches and guarantee the high accuracy of high speed TIADC.The signal reconstruction of TIADC usually was divided into two steps which contain obtaination and calibration of channel mismatches. Obtaination of channel mismatches could be carried out by measurenment or estimation. Calibration of channel mismatches includes two ways—off line static calibration and on line dynamic calibration. The measurement, estimation and dynamic calibration of channel mismatches are demonstrated and analyzed in this thesis. Based on theory and technology such as least square, expectation-maximum algorithm, interpolation fit and hybrid filter bank, uniting the pratical request and research work of project"×××wide band and high accurate digitaizer", the main works and contributions of this thesis are as follows:1. The hybrid filter bank (HFB) model of TIADC was built, and an approach using least square method (LS) to measure the channel mismatches of TIADC was proposed. Applying LS and interpolation filter to process output samples of TIADC, the formulas to measure the channel mismatches were deduced. Meanwhile, the Cramer-Rao Lower Bound (CRLB) was deduced when there were quantization error and spurious distortions in sample channel, in addition to harmonic distortions in the test signal. Accuracy of the method was verified by comparing channel mismatch variances to CRLB in MonteCarlo simulation. The effectiveness and accuracy of this method in actual applications were verified in the measurement of channel mismatches in 400MSPS 12-bit high speed digitizer.2. To satisfy the requirement of time mismatch's realtime calibration, estimation algorithm of time mismatch by expectation-maxmum algorithm (EM) was presented. According to the relationship between frequencies of input and distorted signals caused by time mismatch, DFT and EM were used to subgroup the output signal, and the expression of output spectrum was applied to obtain time mismatch. With no limitations on the channel number of TIADC, the algorithm can be implemented in the background. In the simulation and experiment, it was demonstrated that no matter the input signal was single tone or multi-tone, the algorithm can obtain time mismatch effectively.3. To the dynamic calibration of channel mismatches, gain and offset mismatch can be simply calibrated by multiplier and adder, thus emphasis was focused on the dynamic calibration of time mismatch, which equalized to the reconstruction problem of periodic nonuniform sampling. Talor series were used to linearize the complex inversion of time varied filter in traditional reconstruction method to obtain coeffecients of reconstruction filter. In the case that timing mismatch was changed, the timing mismatch was updated by above mentioned calculation algorithm without recalculation of the filter coefficients.4. Considering one channel as the reference channel, the calibration model based on reference channel was built, and reference error calibration method was proposed, as a compensation for reconstruction filter calibration method. Without estimating channel mismatches and using filters, the method can efficiently complete dynamic calibration by look-up-table. Aiming at the limitation that the coverage of address is inadequate, an improved sheme method was studied. Finally , simulations and experiments demonstrated that such approach can effectively compensated channel mismatches'calibration in the whole range of input frequency, furthermore, it can also independently be applied to dynamically calibrate the channel mismatches.