Research On Accurate Signal Reconstruction Of Time-Interleave-based Wideband Acquisition Systems

As the core unit that transforms the analog signal into digital value,the data acquisition system has been widely used in many areas such as communication,nuclear physics,deep space exploration,measurement instruments,etc.With the development of technology,signals in these fields become more complex,which have a wider bandwidth and more remarkable transient characteristics.To test these signals,acquisition systems are required to have a higher performance.However,limited by the integrated circuit technology,it is hard to satisfy the requirements of acquisition systems by using only one analog-to-digital converter(ADC)chip,especially for the needs of the sampling rate.Therefore,the time-interleaved(TI)sampling architecture is proposed to provide multiple times of sampling rate compared to a single ADC.Basically,the TI sampling structure is a kind of parallel sampling structure.The sampling mismatches are inevitable in the TI structure,for the parameters of analog circuits in different sampling paths cannot be consistent in real systems.Although there are lots of studies on TI mismatches,they are more focused on the mismatches in narrowband acquisition systems or TIADC.In wideband acquisition systems,the frequency selectivity of sampling mismatches and the arbitrariness of each channel frequency response greatly increase the difficulty in mismatch estimation and calibration.Besides,the non-ideal frequency response error caused by the unevenness of magnitude response and the nonlinearity of phase response in wideband acquisition systems further increases the distortion in sampling results of wideband signals.The existence of these two kinds of errors makes the sampling results can not accurately reproduce the original input signal.In order to minimize the differences between the sampling result and the input signal,these mismatches and errors must be corrected,so that the calibrated signal is an accurate reconstruction of the original input.This dissertation is devoted to solving the problem of accurate signal reconstruction in broadband TI-based acquisition systems,and is mainly carried out from the following aspects:(1)The TI sampling models based on generalized channel frequency responses are analyzed.The relationship and differences among sampling models based on the linear hybrid filter banks,the period time-varying filter and the fixed mismatches are discussed,which provides reliable theoretical support for the following research on mismatch estimation and calibration.Besides,system errors are divided into sampling mismatches and non-ideal frequency response errors according to the causes of errors.Their influences on system performance are also analyzed,which provides a basis for the evaluation of calibration results.(2)The estimation methods of two kinds of frequency responses required by error correction are studied.As for the sampling mismatches,the estimation method based on the spectrum analysis is modified according to the relationship between the linear hybrid filter bank sampling model and the fixed mismatch sampling model.This improvement solves the fault estimation when the signal period is less than the sampling interval.The relative frequency responses are all estimated by using a series of sinusoids whose frequencies cover the whole bandwidth.As for the non-ideal frequency response error,it is necessary to estimate the difference between the system frequency response and the ideal frequency response,that is,to obtain the differential frequency response.The magnitude response of the differential frequency response is measured by using the frequency sampling method.And the phase response is acquired by comparing the spectrum of the sampled fast slope signal and a reference fast slope signal.To reduce the inaccuracy of phase estimation in high-frequency components caused by noises in the estimation of the phase response,a method that aligns and averages signal in the frequency domain is introduced.By using a two-step positioning method that simplified the nonlinear optimization problem in the alignment process,and combining a onedimensional search method,the optimal alignment of the spectrum in the whole bandwidth is achieved.The proposed method doesn’t need a precise trigger module,which reduces the requirement of hardware resources for error estimation.(3)The theory of sampling mismatches calibration for arbitrary channel frequency responses is studied.The calibration is based on the idea of reconstructing the mismatches first and then eliminating them form the sample result.However,by studying the properties of the period time-varying filter,the proposed calibration structure simplifies the error reconstruction and error elimination steps,and directly obtains the calibration result from the sample data.Due to the use of approximate conditions in the derivation of error reconstruction theory,there is still residual error remains in the calibration result.Therefore,the dissertation further studies the cascading form of the calibration structure to reduce residual error.By deducing the cascade form of periodic time-varying filters,it is concluded that the cascade of periodic time-varying filters can be equivalent to a single periodic time-varying filter,and the numerical relationship of them is also obtained.The significance of this conclusion is that the cascaded calibration structure can be equivalent to the single-stage calibration structure by replacing the filter parameters in the singlestage calibration structure.Thus,the residual error can be further reduced,which ensures more accurate signal reconstruction results and system performance such as higher ENOB or SFDR without consuming additional filter implementation resources.(4)The design and implementation methods of arbitrary frequency response filter in the time domain and frequency domain are respectively researched.As for the filter design in the time domain,the arbitrary frequency response is realized by a linear phase FIR filter that approximates amplitude frequency response,and an all-pass filter that approximates group delay response.A fractional delay filter is also used to align the data after filtering in the calibration of sampling mismatches.As for the filter implementation in the frequency domain,the influences of spectrum leakage and data length are analyzed.A frequency domain filtering method based on overlapping frames is proposed to reduce the deterioration of the filtering effect caused by spectrum leakage.At last,a TI-based digital storage oscilloscope with 20 GSPS sampling rate and 8GHz bandwidth is designed and implemented.The effectiveness of the proposed error estimation and calibration method is verified in the oscilloscope.To minimize the manual operation times in the error estimation process,a software with semi-automatic error estimation and automatic calibration parameter generation functions is designed.Meanwhile,to increase the speed of digital signal processing,the proposed error calibration structure is realized in the hardware.The system performances before and after calibration show that over the system bandwidth the ENOB and the SFDR increased by2.56 bits and 23.44 d B on average.The fluctuation range of magnitude response decreased from 7.59 d B to 0.94 d B,and the system rising time improved from 52.03 ps to 45.62 ps.After calibration,the performance of the designed oscilloscope improves obviously.The calibrated signal is the more accurate reconstruction result of the analog input,which proves the validity of the proposed accurate signal reconstruction technique for the real TI-based wideband acquisition systems.