Noise Reduction in Jitter and Phase Noise Measurements

Noise Reduction in Jitter and Phase Noise Measurements

Using Keysight Infiniium Real-Time Oscilloscopes

Introduction

Keysight Technologies recently implemented a new noise reduction technique in our Infiniium real-time oscilloscope’s TIE (time-interval error) and single-sideband phase noise measurements that dramatically improves sensitivity. This document describes how this feature works to directly measure clock jitter that was previously obscured by the oscilloscope’s noise floor. It also describes how to make the best use of this new feature, assuming the reader is already familiar with conventional TIE and phase noise measurements.

What it Does

Real-time oscilloscopes are commonly used to measure jitter on clock signals. Sometimes however, the clock’s jitter is too low for the oscilloscope to measure it because the measurement result is dominated by the oscilloscope’s voltage noise. Keysight’s noise reduction method measures a clock signal using two different oscilloscope channels simultaneously, and then computes the TIE or phase noise using a cross correlation technique. The computation removes the oscilloscope’s noise and jitter that is uncorrelated between or unique to each oscilloscope channel.

Figure 1 demonstrates the improved sensitivity of a phase noise measurement using the new noise reduction technique. The measurement compares the phase noise a 100 MHz output from an 81134A pulse generator. Memory 1, (m1) shows pulse generator’s true phase noise. This trace was measured using an E5052B signal source analyzer and then saved to the oscilloscope’s Memory 1. Memory 4, (m4) was measured using a DSOZ334A real-time oscilloscope without noise reduction. Memory 2, (m2) was measured using a DSOZ334A real-time oscilloscope with noise reduction. Without noise reduction, the integrated jitter from 150 kHz to 40 MHz of the oscilloscope measurement was 1.58 ps rms. With noise reduction, the integrated jitter of the scope measurement dropped to 760 fs rms. The signal source analyzer reported an integrated jitter of 750 fs rms.

The new noise reduction technique also works on TIE measurements. Figure 2 shows the noise reduction technique’s improvement on a TIE trend measurement. The bottom trace, (m2) shows a conventional TIE measurement trend waveform of a 10 GHz clock signal. The top trace, (mt) shows a TIE measurement trend of the same clock signal using the new noise reduction technique.

How to Use It

Connections

Noise reduction needs to digitize the SUT (signal under test) using two different input channels simultaneously.

Single-ended signals

The best way to split single-ended signals into two copies is by using passive power splitters or dividers because they don’t add noise or jitter to the measurement. Resistive dividers are a little simpler to use than reactive dividers because they don’t have a low frequency cutoff, and a generally flatter frequency response. Resistive dividers do attenuate the signal more than reactive dividers (-6 dB instead of -3 dB), but that rarely matters for clock signal measurements because clock signals are usually so large that needed at least 3 to 6 dB of oscilloscope attenuation anyway. Much less common is to split a single-ended signal into a differential signal using a passive balun transformer.

You can also use buffers or amplifiers to fan-out your SUT, provided you ensure the buffer or amplifier doesn’t add appreciable jitter of its own to the measurement. Even the two differential outputs of a differential buffer can be used if the buffer doesn’t have significant common-mode noise. Any common mode noise from a fanout buffer will be correlated across the two copies of the SUT and therefore will be added to the measurement result.