measurement of external jitter for true snr estimation of ... · – run jitter analysis at every...

28
2/17/2005 Measurement of External Jitter for True SNR Estimation of A/D Converters Turker Kuyel, PhD Senior Member of the Technical Staff Design Manager Data Acquisition Products High Performance Analog

Upload: others

Post on 28-Sep-2020

21 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Measurement of External Jitter for True SNR Estimation of A/D Converters

Turker Kuyel, PhDSenior Member of the Technical Staff

Design ManagerData Acquisition ProductsHigh Performance Analog

Page 2: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Outline

• What is jitter • Why is it important • How can it be reduced • How can it be measured • How can it be removed• References

Page 3: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

What is time jitter?

• Time Jitter is a measure of temporal noise– A signal must be periodic to have jitter– The period of a signal can be estimated within a time interval

• Uncertainty principle implies ∆F* ∆T=1• A long time period is required to estimate a signal period with

low uncertainty– Once the period is estimated (known), we can talk about

ideal “zero crossings”– Jitter is the temporal deviation of the actual zero crossings

from the ideal zero crossings.• Analyzed statistically• Can be deterministic or random

Page 4: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

In what forms timing jitter can occur?

• Period of the signal might drift with time– Phase locked loop, close-in phase noise

• There may be random variations of the zero crossings• Can be normally distributed• Other distributions can be observed

• There may be deterministic variations of the zero crossings• Can be dependent on the architecture of the periodic waveform

generator: • For example: Every 4Nth clock can be Gaussian around the desired

period, but 4N+1, 4N+2, 4N+3 clocks can have deterministic errors.

• Usually the answer is “all of the above”

Page 5: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Do we care?

• Yes! • Jitter is critical in high-speed digital

– 30ps rms jitter can be considered as “bad” for SERDES

• Jitter is critical in high-performance data converters– 0.3ps rms can be considered as “bad” for ADS55XX– ADC applications require 2-3 orders of magnitude better jitter

than high-speed digital applications– Random and deterministic Jitter makes Signal-to-Noise

Ratio (SNR) of an ADC (DAC) look bad.– Deterministic jitter makes Spurious Free Dynamic Range

(SFDR) of an ADC (DAC) look bad

Page 6: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Jitter noise can exceed quantization noise

• Quantization noise of an N-bit ADC

• Jitter noise of an N-bit ADC

• Find ‘quantization noise equivalent’ jitter)2log(20max επfSNR jitter −=

jitterrmsfrequencyinputf

f N

__

,1021

3.0

==

<

ε

πε

76.1*02.6max += NSNR onquantizati

Page 7: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Jitter Tolerance of Nyquist ADCs (ps rms)

0.000.000.000.0060.0250.100.401.616.4100MHz

0.000.000.010.060.2521.014.0316.16410MHz

0.010.040.160.632.5210.140.31616451MHz

0.10.41.66.325.210140316126451100KHz

14166325210084032161286451210KHz

24bit22bit20bit18bit16bit14bit12bit10bit8bitFin

Page 8: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Critical Example: ADS5500

• 14-bit 125MSPS ADC – Sampling speed is irrelevant

• Fin = 70MHz• SNR = 70dB typical @ Fin = 70MHz• Close to 12-effective-bits performance• SNR TESTING:

– Total jitter must be under 400 fempto-seconds !!!– Can we guarantee that it really is ???– If not, chances are, we are measuring jitter effects. We are

not measuring true SNR of ADS5500.

Page 9: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Critical Example: ADS5500

• What needs to be under 400 fempto-seconds?• In other words, what does “total jitter” mean?• TOTAL JITTER is the RSS SUM of:

– External jitter of the clock source– External jitter of the input source– Internal jitter of the ADC

• S/H aperture jitter• Internal clock jitter

Page 10: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Critical Example: ADS5500

• Key Specifications: SFDR and SNR

• SNR is affected by random and deterministic jitter– Random jitter show-up as a raised noise floor– Deterministic jitter show up a “forest” of small tones.

• SFDR is affected by deterministic jitter– Some of the deterministic jitter tones could be so high, it

could present the “worst spur”.

Page 11: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Where do we look for jitter?

• Order of Importance:1) Clock generator (ATE). This can be terrible.2) Input generator (ATE)3) Output drive load (di/dt) on the DUT board4) Bypass and drive circuitry for Vref+ and Vref-5) Chip design6) PCB Layout. A simple PCB layout with a thick single ground

plane is the ‘last’ place to chase jitter. Do not start with this one.

Page 12: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REDUCE jitter?

• Internal to ADC:– Differential clock architecture– Differential input architecture– Fast clock rise/fall times.– Good chip layout, minimizing supply bounce and

coupling.– Lowest possible digital signal amplitude (LVDS)– Lowest possible digital signal rise/fall times.– Good pinout. Lowest possible inductance.

Page 13: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REDUCE jitter?

• Frequency reference:– Use an oven controlled crystal oscillator. Better than an

“atomic clock” over short time frames. Data capture time is usually around 100us.

– Phase-lock the clock generator, input generator and capture instrument to the oven-controlled-crystal.

– Wenzel Associates, www.wenzel.com makes the lowest jitter OCXOs, according to AeroFlex Comstron. Comstron is the synthesizer in Teradyne’s Catalyst test systems.

– On an ATE, lock to the 10MHz reference of the ATE. If the ATE does not synchronize to its frequency reference, one can find the correct capture window with an edge placement search

Page 14: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REDUCE jitter?

• Input Generator:– Use crystal, pll, mixer based systems for lowest

jitter: HP8662 has the lowest close-in phase noise. HP8644B has better noise floor at higher frequencies. HP8644B is also relatively clean from non-harmonic spurs.

– Wenzel Associates can design custom arrays of crystal oscillators for fixed frequencies, for absolute lowest jitter (-173dB noise floor).

– Use a band-pass filter, crank the input amplitude.

Page 15: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REDUCE jitter?

• Clock Generator:– Use HP8644B or alike.

• PN9000 measured HP8644B jitter to be 0.47ps. This includes PN9000 noise also.

– Use a steep narrowband band-pass filter to reduce jitter. A crystal filter is the most suitable.

– Route the clocks differentially.– Gain the clocks on-chip.– To observe the effects of gaining the clock, HP8133A clock

generator could be used (60ps rise time). HP8133A needs a secondary HP8644B to set the frequency arbitrarily. HP8133A jitter is measured around 1ps rms.

Page 16: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How can we MEASURE jitter?

• Fine-coarse counter based time domain techniques: Wavecrest DTS, 10ps accurate

• Flash ADC based digital scopes: LeCroy, Tektronix, etc. 8-bits 4GSPS, 2-3 ps accurate.

• PLL and mixer based techniques: Comstron PN9000, 0.2ps accurate. Difficult to use. Performance as good as its frequency synthesizer

• CommsADC based TI methods, 0.2ps accurate• Better than 0.1ps ? Pure Research. Alfio Zanchi and

Yiannis Papantonopoulos reported 25 fs.

Page 17: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we measure TOTAL jitter?• Various methods are available at TI.

– Dual-SNRs (Burns, Kao, Kuyel) (1)– Dual-Histograms (Kuyel) (2)– Unwrap-FFT (Pearson. Method uses same dataset) (3)

• To sensitize to aperture jitter and clock jitter, band-pass filter the input, and use an Fin near the band-edge of the input S/H. This means heavy undersampling of high Fin to increase sensitivity to jitter. Assuming input is ideal, gives aperture jitter + clock jitter.

• In 1997, TLC876 (10bit 20MHz ADC with 250MHz BW) measured the Teradyne HSD50 jitter around 12ps rms (1).

Page 18: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Undersampled Sine Wave Capture

Page 19: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Low-slew-rate Histogram

Page 20: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

High-slew rate Histogram

Page 21: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Time Differential Sampling

RF Splitter

Low-Jitter Sine-wave

Transformer

Transformer

Termination

Termination

ADC CLOCK

ADC INPUT

Page 22: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REMOVE jitter?

• Test for random vs. correlated (deterministic) jitter.• If random, it can be removed.• If deterministic, tones can be deleted after careful

investigation.• Frequency synthesis based systems are known to

generate random jitter.

Page 23: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we tell random vs. deterministic?

• The difference between Pearson’s and Kuyel’s techniques gives the deterministic jitter in an rms sense.

• Undersample the capture waveform. – Run jitter analysis at every mKth sample– Run jitter analysis at every mK+1th sample– Run jitter analysis at every mK+2 sample– See if there is any difference in jitter readings.– If there is, then deterministic jitter is present

• If the jitter is mostly random, then it could be removed. OCXO jitter is random.

Page 24: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we REMOVE random external jitter?

• Measure total jitter with one of the available methods (dual-SNRs, dual-hist, FFT-unwrap)

• Remove total jitter noise from the SNR calculations• Measure internal jitter using Time Differential

Sampling technique• Add internal jitter noise back into the SNR

calculations

Page 25: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we measure Internal Jitter?• Internal jitter is usually very low: TLC876 <1ps rms• A high performance ADC should read ~0.1ps rms• TIME DIFFERENTIAL SAMPLING (2):

– Use the same generator as signal and as clock.– The capture waveform will be a noisy DC signal– Delay the clock until minimum-slew-rate DC is captured. This

gives an indication of random noise in the system. This noise is NOT internal jitter.

– Delay the clock until the maximum-slew-rate DC is captured. This gives system noise + jitter noise.

– Internal_jitter_variance = MAXSR_variance –MINSR_variance

– This technique can be extended to multiple points and curve fitting results in more robust estimates (9)

Page 26: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

How do we remove EXTERNAL jitter?

• Run a random/deterministic analysis first• Measure TOTAL JITTER with dual-histograms• Measure INTERNAL JITTER with TDS• Subtract TOTAL JITTER and add INTERNAL JITTER

• Assuming even the best test setup, we may gain 1-2dB SNR, starting from ADS5500 and higher

Page 27: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

Some Jitter Measurement Results (1997-2004)• A580 HSD channel: 11ps @ 20MHz (Measured with dual SNRs,

TLC876)• A580 HSD channel: 9ps @ 40MHz (Measured with dual-hist,

using THS1240, t0 divide-by-four)• A580 super-clock. 4.5ps @ 40MHz (Measured with dual-hist,

using THS1240)• Catalyst HSD channel: 33ps @ 65MHz (Measured with dual-

hist, using AD6640. Highly deterministic nature)• Catalyst pico-clock: 1.1ps @ 65MHz (Measured with dual-hist,

using AD6640)• HP8644B: 0.470ps @ 65MHz (measured with PN9000,

unfiltered)• HP8131A: 0.850ps @ 65MHz (Measured with PN9000)

Page 28: Measurement of External Jitter for True SNR Estimation of ... · – Run jitter analysis at every mKth sample – Run jitter analysis at every mK+1th sample – Run jitter analysis

2/17/2005

REFERENCES

• 1) US 6,240,130 B1 (dual-SNR technique, Burns, Kao, Kuyel)• 2) US 6,640,193 B2 (dual-hist, TDS, and removal techniques, Kuyel)• 3) TI-Patent Pending (FFT-unwrap method, Chris Pearson)• 4) LeCroy Digital Oscilloscopes “Accuracy in Time Jitter Measurements

with LeCroy Oscilloscopes”, LeCroy Corporation• 5) Aeroflex Comstron Inc. “PN9000 Automated Phase Noise

Measurement Systems”• 6) Aeroflex Comstron Inc.”Phase noise theory and Measurement”,

Apps note.• 7) Wavecrest Corporation, “Digital Time Systems”• 8) Rosing et al., “Off-chip Diagnosis of Aperture Jitter in Full-flash A/D

Converters, Feb. 1999, Jetta, pp. 1-8.• 9) Zanchi and Papantonopoulos, “Measuring Sub-picosecond Jitter”,

CommsDesign, March 09, 2004• 10) Wenzel Associates web page www.wenzel.com• 11) Aeroflex Comstron web page www.aeroflex.com