Copyright © 2012 Agilent Technologies Spurious Measurements: Optimizing for Speed and Accuracy Ben Zarlingo Product Manager, Signal Analysis Microwave & Communications Division
Copyright © 2012 Agilent Technologies
Spurious Measurements:
Optimizing for Speed and Accuracy
Ben ZarlingoProduct Manager, Signal Analysis
Microwave & Communications Division
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
This Presentation
Spurious Measurements Context, Background
Measurement Overview, Principles
Specific Case Examples, Suggestions
Performance Optimization Techniques
Taking Advantage of Modern Analyzer Features
References, Additional Information
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
The Challenge: Acquiring Information
Optimize By Using Your Knowledge, Expertise, Priorities
Gather Needed Information; If Extra Then
Trade Extra For
– Faster measurement time
– Lower measurement/equipment cost
If More Information Needed, Consider
– Equipment availability
– Cost
– Time
Cost
Time
Information
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Noise Converts ~ Deterministic Value to Statistical Distribution
GSM (MSK)
SNR 60+ dB
* CCDF varies significantly with span at low SNRs
CCDF
1 MHz Span*
1%
1 dB/div
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Separating Signals from Noise
Signal/Noise Variation with RBW
Amplitude envelope vs. time
Best RBW is one matched to signal
Best ability to separate analyzer noise from signal
RBW (log)
SNR
(dB)
Noise-like
Real signals
can be one
type or
combination
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Information:
What You Have vs. What You Need to Gather
Ask, Answer Questions to Bound the Task
Information Types
– Amplitude accuracy, variance
– Frequency accuracy, variance
– Time domain parameters, cause
– Other: bandwidth, modulation, isolation/crosstalk
What You Know vs. What You Need
– Know the spur frequency? If not, does it matter?
– Measure actual amplitude or only limit test?
– Accuracy, variance needs for amplitude, frequency
Presence or Absence of Other Signals
– Waveguide below cutoff will filter off lower frequencies
– Presence of lower frequencies can affect attenuation, preamp use
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
“Classic” Spectrum Analyzer Spurious
Measurement Setup: 2.51 dB SNR Benefit
Peak Detection
– Avoid missing signal that falls between display “bins” or “buckets” and
reduce effect of time-varying spurs
Video Averaging: Use Narrow Video Bandwidth Filter
– Trace averaging yields same benefit, but slower variance reduction
– Squaring the Average vs. Averaging the Square
– Under-measure noise 1.05 dB by squaring the average
– Log display processing (log amplification)
– Under-measure noise by 1.45 dB
Total “Noise Reduction” = 2.51 dB; Other Processing
Approaches Produce Same -2.51 dB Under-Response to
Noise or Noise-Like Signals
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Detector Types, Averaging Scales Understanding Enough
Processing, Data Reduction Always Done
Signal Characteristics Affect Results
Basic Settings (RBW, Span, Sweep Time) Affect Results,
Even With Same Detector
Know Enough To:
– Use your knowledge of your signal
– Understand results, know what you are getting
– Make good measurements
– Identify signal problems, measurement errors
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Traditional Spectrum Analyzer Block
Diagram (very simplified)
Analog signal processing, but principles apply to DSP
Order of envelope detector and log amp may be reversed
Digital technologies (ADC, DSP) moving to the left (input)
Input
Signal
Processor
and Display
Log AmpEnvelope
Detector
RBW
Local
Oscillator
Atten
VBW
ADCDetector
Mode
Log
Linlog
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Display Detectors:
Peak, Negative Peak, Sample
Input is the IF envelope detector output (log or linear scaled)
Each output represents only one IF envelope value
Peak and neg peak detectors bias the output value, perform
some data reduction
Detector
Mode
Time
Volts
Peak
Neg Peak
Sample
Display points or buckets
Peak
Neg Peak
Sample
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Averaging Detectors
Each output represents 2 or more IF envelope values
Reduces variance faster than single-value detectors;
reduce variance further by increasing sweep time
Good choice for noise, noise-like, or unknown statistics
Time
Volts
Average value
for this bucket
Samples of envelope detector output
Detector
Mode
Average
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Averaging Processes in
Spectrum/Signal Analyzers
Reduce variance for noise, noise-like signals (noisy CW)
Critical to understand which parameter is being averaged
Consider all averaging processes which affect the meas.
Important to ensure the consistency of all averaging processes involved
Processor
and Display
Log AmpEnvelope
Detector
RBW
VBW
ADCDetector
Mode
Log
Lin
Limited BW, but Other
BWs typically narrower
Video Bandwidth
Filtering or Smoothing
Averaging
or RMS
Detectors
Trace-to-Trace
Averaging,
Noise Marker
log
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Averaging Processes & Scales
Four Processes in Most Analyzers– Trace Averaging
– Video Bandwidth Filtering
– Average Detector
– Noise or Band Power Markers (average across buckets, a subset of displayed trace)
In Agilent X-Series, PSA All are Locked to 1 of 3 Scales– Power (W)
– Voltage
– Log power (dB, dBm)
Locking Determined by selected “Average Type”
Locking Not Present on Some Analyzers
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Averaging Scales—Log, Lin, Power
Logarithmic averaging scale—use for CW signals
– Traditional log scale measurement
– Narrow VBW, trace averaging
– Better accuracy, dynamic range for low-level CW signals
Voltage averaging scale—use for voltage envelopes
– Traditional linear scale measurement
– Measure rise/fall times of pulsed RF (TDMA) signals
Power averaging scale—use for time-varying signals
– Measure average power and report result on a dB scale
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Potential Averaging Errors
Example: Averaging Power, dB vs. W
Average of the log is not equal to a log of the average
Narrow VBW or trace averaging performs an average of the
log, an error in measuring time-varying signals
Instead, average the power or report the RMS value of the
voltage of the signal when measuring time-varying signals
0 dBm
6 dBm
1 mW
4 mW Average of the log of the power = 3 dBm
(0 dBm + 6 dBm)/2
Log of the signal’s average power = 3.98 dBm
(1 mW + 4 mW)/2 = 2.5 mW
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Log Scale Compression/Expansion
Assume: Average noise power -150 dBm/Hz
10-15 W/kHz ENBW, 1 fW/kHz
Measurement 1: Add 0.75 fW
Result: -147.6 dBm +2.4 dB
Measurement 2: Subtract 0.75 fW
Result: -156.0 dBm -6.0 dB
Log average: -151.8 dBm
Power average: -150.0 dBm
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Log Scale Bias Benefit for CW Meas.Log scale better for single CW signal near noise
Video BW filtering
or averaging dB
See Agilent AN1303
Noise-free meas SNR 1 dB; pwr avg SNR 1 dB; log avg
1 dB SNR Example
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Non-CW Spur, Close to Noise
Indicators
– Spur noisy
– Spur close to noise
– Spur appears wide
(compared to RBW)
– Multiple concerns
for accuracy
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Narrower RBW Improves S/N
SNR Improved
Lower Measured Power
– Accurate result?
– Signal wider than RBW filter?
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Band Power Marker, Averaging
Use Techniques Independentof Signal Statistics– Band power marker compensates
for signal spread
– Longer sweep, average detector to reduce variance
Spur With Residual FM– Noise-like signal
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
DANL vs. Other Measurement Priorities
Attenuation = 0 dB Optimal for Sensitivity
Attenuation = 10+ dB for Optimal Accuracy
Attenuation = 4-6 dB a Compromise
Problems with Impedance Match
– Use both preamplifier and attenuator?
Problems with Distortion, Compression
Chance of Analyzer Input Damage
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Basic Spectrum Analyzer Cautions
Don’t Overdrive the Mixer/Front End
– A concern whenever the signal bandwidth is much larger than RBW
– Displayed power may understate actual signal power
• Understand total power of modulated or noise-like signals
Don’t Overdrive the Log Amplifier
Overload Exception
– If distortion products are known and not a problem for you
– Overload to improve DANL
– Add your information to improve information gathering
Use the Appropriate Display Detector
Avoid Inconsistent or Incompatible Averaging Methods
Video Averaging: Narrow VBW and/or dB Scale Trace Avg.
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Techniques for Reducing DANL,
Improving Dynamic Range
Reduce Attenuation
Add Preamp
Reduce RBW
Add External Filtering
Better/Shorter Cables, Connectors, Move Analyzer
– 1 dB Gained/Lost = 1 dB Noise Figure
Time Averaging (where possible, not same as meas. avg.)
Measurement Processing (take advantage of Moore’s Law)
Noise Power Subtraction/Noise Correction
Noise Floor Extension (NFE)
– Leverages knowledge of analyzer/circuit behavior (adds information)
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Understanding Signal Analyzer Dyn. Range
Key Factors
In the presence of a carrier:
Want high dynamic range
Formula to Estimate:
(TOI – DANL {in dB}) x 2/3rds
Example: (at 1 GHz, PXA)
+20 – (-155) = 175 x 0.667
= 116.7 dB
PXA Specs for TOI and DANL
(From Agilent PXA Specs Guide)PXA Specifications Reference:
PXA Specifications Guide
Dynamic Range section.
Part Number N9020-90017
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Preamplifiers
NF of Preamp Should be Less than NF of Analyzer
Benefits
– Improved sensitivity
– May be built-in, full frequency range
– Internal preamps are switchable
– If internal, gain can be included in analyzer calibration
– External preamps can be optimized for your
specific needs
Drawbacks
– Best when input is small signals only
– May limit TOI, increase distortion
– May have limited bandwidth
– Calibration less convenient if external
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
IF Gain Setting
Set to High for Small Signals, Lowest Analyzer Noise
Set to Low for Large Signals, Lowest Distortion
You Know More
About Your
Signal than the
Analyzer Does
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Digital Filter ShapeBetter shape factor, biggest selectivity benefit for different signal
levels
Equivalent selectivity at a wider, faster-sweeping RBW
digital filters swept an additional 3-4x faster
30 kHz Digital Filter
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
28
Digital IF Ref Level and Amp Scaling
Logging and scaling processes now DSP
No need to adjust Ref Level
faster, simpler programming
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Sweep Speed Tradeoffs
Analog Filter Default: RBW2/2
Analog Filter, Accuracy mode: RBW2/4 to RBW2/8
Digital Filter Default: RBW2 (with accuracy)
Sweep Speed Problem
– Narrow sweeps with slow sweep times from narrow autocoupled RBWs
– Wide sweeps with narrow RBWs to reduce noise level
Solution: Switch to FFT Mode with “Best Speed” Rules
– Small tradeoff in dynamic range for large improvement in sweep speed
Sweep Rate (Hz) RBW2
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Swept & FFT Measurements
FFT– Faster for narrow RBWs, typically narrower
spans
– Modern analyzers can provide FFT results directly or step/concatenate FFTs and process to provide swept-equivalent results
– Always used for captured signals/recordings
Swept– No span limit
– Flexible bandwidths, detectors, averaging
– Benefits of classic setup apply
– Segmented or zone sweeps
Choosing– Analyzers, applications can choose automatically
– Manual choices for speed, dynamic range, optimize for specific situation
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Take Advantage of Instrument Processing
Thousands or Millions of Measurements/Second,
Processed Locally
Reduce Data Bus Traffic
– Ideally a table of results or simple go/no-go
Offload System CPU, Local Processing Instead
– Measurement processing
– Measurement setup changes (frequency ranges, RBWs, limits)
Available Local Processing
– List sweep
– Zone sweep, zone span
– Dedicated spurious emissions mode
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Harmonics and Spur Search Approaches
Use traditional swept measurement and Marker Peak
Search function
Use “1-button” Harmonic Distortion and Spurious
Measurements (Power Suite)
Use “List Sweep”, with pre-defined list of frequencies
that correspond to known harmonic frequencies
Use traditional swept measurement with Peak Table
readout
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Example: Spurious Emissions Mode
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
List Sweep (SCPI Only)
Eliminate Unnecessary Overhead associated with setting up individual measurements
Multiple Zero Span measurements at multiple frequencies
Different RBW, VBW, Sweep times (etc) at different points
Peak, Mean average power results
Up to 2000 Points
Page 34
PXA Early Engagement Visits
Agilent Confidential
March 13, 2009
X-Series Signal Analyzer Measurement Speed Reference:
App Note 1583: Maximizing Measurement Speed with the Agilent’s X-series Signal
Analyzers Lit Number 5989-4947EN
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Peak Table
Read out up to 2000
peaks above a threshold
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Processing Improvements for Sweep Speed
2 to 3 times speed improvement for FFT sweeps with
large spans and narrow RBWs (common use case for
spur searches).
Changes implemented that sped up FFT sweeps:
Halved number of points in FFTs while maintaining same
amplitude accuracy.
Enabled more parallel processing in FFT sweep
algorithms.
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Time Averaging: An Overlooked Technique
Vector (Time) Average to Remove Noise– Coherent (time-triggered) time domain averaging
– Requires repeated signal, triggering
– FFT, not swept mode operation
Coherent vector average of sets of time samples creates a time-averaged time record
For each sample the average of synchronized measurements converges to the noiseless value
Q
IPhase
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Time Averaging Example
Trace average (power)
reduces variance only Time average
reduces noise
Coherent averaging
calculates spectrum
from averaged time
samples
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
CW Signal Measured Near Analyzer
Noise Floor
Actual S/N
Displayed
S/N
CW Signal
Apparent
Signal
Ampl & Freq
Axes Expanded
Example: No Noise Subtraction or Noise Correction
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Noise Subtraction, “Noise Floor Extension”NFE Calculations in Agilent PXA Improve DANL
Analyzer Noise Power is Calculated/Subtracted in Real Time
3 dB Error
without NFE
“No” Error
Improved Noise Floor or
Displayed Avg Noise Lvl.
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Known vs. Unknown Signals &
Noise Floor Extension or Noise Subtraction
Spur Location
Known
Spur Location
Unknown
No NFE or
Noise Subtraction
Using NFE or
Noise Subtraction
“Classic
Case”
Zero Span
Avg Detector
Pwr Avg Scale
Peak Detector
VBW Filtering
Log Avg Scale
Zero Span
Avg Detector
Log Avg Scale
Peak Detector
VBW Filtering
Pwr Avg Scale
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Gated Sweep for
Time-Selective Spurious Measurements
Gate lengthDelay
Gated Sweep Applicable to Wide, Narrow Spans
Easier Setup with Modern Analyzers
Flexibility Similar to
Non-Gated Meas
– RBW, VBW
– Detectors
– Sweep times
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Time-Varying Spurs:Time Capture, Recording, Post-Processing
Gapless, Real-Time 250-500 Msamples
– Change CF & span
after capture
– External, internal triggers,
pre/post trigger delay
– Capture data is
continuous, real-time
– Maximum span limited
by IF/ADC BW
43
Spectrum
IF Time
Spectrum
RF Envelope
IF Time
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Gap-Free Analysis of Freq, Amplitude, Time
Band-Selective, Level-Selective Triggers, Holdoff
Pre-Trigger Delay Captures Events Before they Happen
Relate Spurs to:
– LO Switching
– Transients (pwr, etc.)
44
Spectrum
IF Time
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Low Noise Path DANL Improvement
3 dB
@3.6 GHz
8-10 dB
@26 GHz
Example:
Spur Search
20-50x faster
at 18 GHz
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Trade Low Noise Path DANL for Speed
Noise FloorImprovement
7-8 dB
SpeedImprovement
~32x
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
References, Additional Information“Spectrum Analyzer Measurements and Noise” Joe Gorin, Agilent Technologies
Application Note 1303, lit. number 5966-4008E
“Spectrum Analyzer Detectors & Averaging for Wireless Measurements” Joe Gorin &
Ben Zarlingo http://www.agilent.com/cm/wireless/pdf/detector_averaging.pdf
“Using Noise Floor Extension in the PXA Signal Analyzer” Agilent application note, lit.
number 5990-5340EN
Webcast “Use capture, playback & triggering to completely analyze a signal” http://www.home.agilent.com/agilent/eventDetail.jspx?cc=US&lc=eng&ckey=1976536&nid=-
11143.0.00&id=1976536&pselect=SR.GENERAL
“Spectrum Analysis Basics” Agilent application note AN-150 lit. number 5952-0292EN
“Optimizing RF and Microwave Spectrum Analyzer Dynamic Range” Agilent
application note AN-1315 lit. number 5968-4545E
“Achieving Amplitude Accuracy in Modern Spectrum Analyzers“ Joe Gorin,
Microwaves & RF Magazine Sept. 2008 http://www.mwrf.com/Article/ArticleID/19724/19724.html
“Swept and FFT Analysis” Agilent PSA product note, lit. number 5980-3081EN
“Bringing New Power and Precision to Gated Spectrum Measurements” Wright, Gorin,
Zarlingo, High Frequency Electronics, August 2007 http://www.highfrequencyelectronics.com/Archives/Aug07/HFE0807_Zarlingo.pdf
Copyright © 2011 Agilent Technologies
© Agilent Technologies 2012
Questions?
49