Channel mismatch compensation in multichannel sampling circuits with weighted integration

摘要

Sampling circuits (SCs) with internal antialiasing filtering radically improve sampling procedure enabling a significant increase in dynamic range, agility, and scale of integration of the analog and mixed signal portions (AMPs) of the equipment. Since these SCs are inherently multichannel, the channel mismatch mitigation is one of the most important problems of their development. There are three possible approaches to this problem. The first approach includes design and implementation measures reducing the mismatch: placing all the channels on the same die, simplifying weight functions, appropriate circuit design, etc. This approach should always be used, but it is not always sufficient. The second approach is based on preventing any overlap of the sampled signal and mismatch error spectra. This enables rejection of the error spectrum in the digital portion (DP) of the equipment. Although the opportunity to totally reject the mismatch error spectrum is very attractive, it requires increase in sampling frequency proportional to the number of channels. This limits the area of application for the second approach. The third approach is adaptive compensation of the SC channel mismatch in the DP. This approach is most universal and flexible. Estimation of the SC channel mismatch can be performed either during the operation mode using the input signal (blind estimation), or using a special calibration signal. This paper analyses adaptive compensation of the SC channel mismatch with blind estimation. Since this compensation is performed in the digital domain, it is as accurate as the mismatch estimation. If the input signal used for the channel mismatch estimation is a stationary stochastic process, the accuracy of the estimation is proportional to the estimation time for a given type of the input signal. The requirements for the accuracy of the mismatch estimation are determined by the resolution of the analog-to-digital converter (A/D) used in the equipment. In the paper--, the time necessary for the channel mismatch estimation is determined as a function of the number n of A/D bits for several types of the input signals. It has been shown that the required estimation time is proportional to (2n - 1)2, for all possible probability distributions of the input signal. Thus, this time grows very fast when the number of A/D bits increases. As a result, blind channel mismatch compensation becomes problematic when the number of A/D bits exceeds 12. The required estimation time also depends on the probability distribution of the receiver input signal. Among signals used in practice, the shortest estimation time corresponds to the signals that can be approximated by the sinewave with random phase, and the longest estimation time corresponds to the signals with Gaussian distribution.