Browse Prior Art Database

Real-Time Sampling Point Adjustment

IP.com Disclosure Number: IPCOM000013760D
Original Publication Date: 2001-Mar-01
Included in the Prior Art Database: 2003-Jun-18

Publishing Venue

IBM

Abstract

High speed serial links generally have a very tight jitter budget. The eye closing at the receiver input is typically in the 75% range. The individual bit however has a considerably larger valid time. This may be verified by measuring a pseudo eye diagram with the trigger signal generated from the incoming data stream (in conventional eye diagrams the clock is used as trigger signal). In high speed serial links, the incoming data stream is oversampled in many cases in order to have several points for data detection and the phase synchronization control loop. The idea of real time sampling point adjustment is based on the above observations: The edges of the incoming data stream are used for a real time and adaptive shift of the sampling point towards its optimum. If no edge is detected, the previous sampling position or an medium setting may be used. Majority voting may further be used to enhance high speed jitter suppression. The concept of real time sampling point adjustment is demonstrated in the figure below. It is assumed, that a ring oscillator generates six sampling phases and a phase rotator is able to shift all six phases to any position. In each decision cycle, the two last samples from the previous cycle are used as additional information about the history of the incoming data stream. This way, a total of eight samples are used. Pipelined logic is used to generate an up/down signal for the phase rotator and to select the output data bit. It is the goal of the algorithm that processes the eight samples to center the edges on the first fife samples (read: at sample 3) and the data on the last six samples (read: on sample 6). In order to have information about the data edges and the data levels at the same time, the time interval used for analysis is approximately 140% of the data bit period. This large analysis interval is divided into two overlapping ranges: The first 2/3 are tested for the occurrence of a data edge and the last 3/4 are used for potential data sampling. The result of the edge test is used for the determination witch of the sampled data is used in the final selection.