An Adaptive DFE Using Pattern-Dependent Data-Level Reference in 28 nm CMOS Technology

Abstract

The sign-sign least-mean-squares (SS-LMS) algorithm has been widely used to adapt both decision feedback equalizer (DFE) tap-coefficient and data-level reference. However, the adaptation process suffers from insufficient adaptation in the presence of pre-cursor inter-symbol-interference (ISI), resulting in eye-diagram destruction. A novel pattern-dependent data-level reference (PD-DLR) adaptation scheme is proposed to alleviate the problem. The PD-DLR is acquired by averaging the received signal when the adjacent two digitized data are both "1". With only an extra pattern filter working at low speed, the proposed scheme can be easily extended from existing adaptive DFE. Simulation results show about 20%, 43%, and 80% improvement in recovered eye height with normalized pre-cursor ISI of 0.14, 0.23, and 0.31 respectively and better noise immunity over its conventional counterparts. The effectiveness of the scheme is also verified in a 28 Gbps non-return to zero modulation adaptive DFE designed in 28 nm CMOS technology.