Beruflich Dokumente
Kultur Dokumente
Introduction
• Noise theory review
• Phase noise basics
Phase noise theory
Analyzing circuit phase noise in GoldenGate
Comparing GoldenGate to SpectreRF
Comparing GoldenGate to measurements
Conclusion
Central-limit theorem
• Large sample of independent random variables with arbitrary probability
density, tends towards a Gaussian
Stationary Noise
• A random process is stationary if it does not change over time
• Can be represented as a Gaussian distribution
• In frequency domain, noise is not correlated
Cyclostationary Noise
• Noise that varies as a function of time, but in a periodic fashion
• Time-varying operating points modulate noise
• Causes multiplication in time domain, convolution in frequency
• Mixes noise up to harmonics of oscillation frequency
• Resulting overlapping noise sidebands are correlated
– Needs to be accounted for in analysis
– Can still be analyzed in linear fashion
Phase Noise in GoldenGate
Agilent Restricted
15 Sep 2006
Phase Noise Basics
v(t ) e Vˆ0 vˆ(t ) e j 0t p (t )
t
where p(t ) 0 ()d
0
Through mathematical manipulation, this can be re-written as
v(t ) e Vˆ0 a(t ) e j 0t p ( t )
where a(t ) is the amplitude modulation
and p(t ) is the phase modulation
Phase Noise in GoldenGate
Agilent Restricted
15 Sep 2006
Phasor Representation
p(t)
Voltage noise
• Combination of phase and voltage jitter Magnitude of V0
v(t ) e Vˆ0 vˆ(t ) e j 0t p (t )
t
where
p(t ) 0 ( )d
0
We can show that the difference between that and a noise-free waveform is
Vˆ ( f ) vˆ(t )e j 2ft dt
0
Vˆ ( f )
A( f ) jP( f )
Vˆ
0
This shows that sideband noise is approximately a linear combination of AM and PM noise.
Vˆ ( f )
A( f ) jP( f )
ˆ
V0
GoldenGate v4.0
• Three NCTs – AM, PM, and DSB
• AM, PM NCT
– Shows contributions of each element in dBc/Hz
• DSB NCT
– calculated from SB noise results
– Noise is computed as
SB Noise3dB20log (peaksignal)
Noise 10 [V 2 /Hz]
10
DSB
Desired
Channel
Interfering IF
2
Desired IF
LO VCO
This is the reverse of how GoldenGate computes noise for autonomous circuits
SpectreRF uses a linearized model
• Based on PSS simulation
• Sources of error include lack of steady state, convergence error, limitations of linearized
model
SpectreRF always uses cyclostationary noise
• GoldenGate currently uses stationary noise for oscillators
• If results don’t match in flicker noise region, change type from “auto” to “cyclostationary”
Common questions
• Q: In SpectreRF I see sideband noise higher than 0dBc/Hz. Will I see that
in GoldenGate?
• A: YES, but we’re correct because we compute PM noise. PM noise can become very large
at low offset because of continued accumulation of phase close to the carrier. Technically,
units of PM noise are [dB rad/Hz]. Because our SB noise is an approximation combining AM
and PM noise, our SB will also go above 0dBc/Hz at low offset.
Spectrum analyzer
• Measures rms noise power in a single sideband, normalized to a 1Hz
bandwidth
• Compares this to the rms signal power measured at the carrier frequency
• Corresponds directly to single sideband noise