Digital Bandwidth Interleaving (DBI) is a parallel sampling architecture that effectively extends the bandwidth of data acquisition systems. However, phase delay calibration between adjacent subbands has become a critical factor affecting signal reconstruction performance. This paper models the inter-subband phase delay calibration in the DBI architecture as a robust regression model, proposing a robust regression algorithm that utilizes the overlapping-band phase difference to precisely estimate the linear phase delay and phase offset. The algorithm can effectively resist the interference from outliers caused by local oscillator phase noise, sampling clock jitter, and non-linear phase. The proposed scheme was verified in a DBI system with an 80 GS/s sampling rate and 16 GHz bandwidth. The experimental results show that, compared to traditional architectures, the proposed scheme effectively compensates for inter-subband phase misalignment, improving the spurious-free dynamic range (SFDR) from 58.73 dBc to 63.94 dBc, and maintains an effective number of bits (ENOB) of 6.12b at a 16 GHz input. The proposed scheme minimizes hardware complexity while effectively enhancing anti-noise capability, confirming its practicality in ultrahigh-speed acquisition systems.
Dai et al. (Sun,) studied this question.