Choosing a Phase Noise
Measurement Technique
Concepts and Implementation
Terry Decker • Bob Temple
RF & Microwave Measurement
Symposium and Exhibition
2
Terry Decker, received her BA in Physics from Carleton College in Northfield, Minnesota in 1980 and a BSEE from the
University of Arizona in 1981. She worked as a microwave engineer for Hughes Aircraft Company on the Phoenix and
AMRAAM programs for 4 years. She is currently a product marketing engineer responsible for phase-noise measurement systems at the Spokane Division of Agilent Technologies.
Bob Temple, received his BA in Physics from Harvard University in 1961, and his MSEE in 1965, and PhD EE in 1971
from the University of Colorado in Boulder. Thesis topic was The Operation and Frequency Stability Measurement
of a Hydrogen Cyanide Beam Type Maser at 88.6 GHz.
His career with Hewlett-Packard began in December 1969 at the Loveland Division designing the frequency synthesis
loops for the 3320 and 3330 Frequency Synthesizers. He was co-project manager for the 3585A Spectrum Analyzer and
the inventor and project manager for the 3047A Spectrum Analyzer System for making comprehensive phase noise and
spectral purity measurements. He transferred to the Spokane Division in 1981 and supported phase noise measurements using the 3047A/11740A Phase Noise Measurement Systems working both within the Company and with customers. He is currently the Project Manager for the Agilent 3048A Phase Noise System.
3
Slide 1
There are many techniques for measuring the phase noise from a source or added by a device. How well each of these
methods works depends on both the technique and the characteristics of what is being measured. This presentation
will examine the advantages and disadvantages of using several of the most prevalent methods when measuring the
phase noise of typical devices. One technique using a phase detector to demodulate the phase noise from the carrier
signal will be covered in detail along with a hardware implementation based on this method.
Agenda
* Basic Phase Noise Measurement Concepts
Direct Spectrum Measurement
Demodulation Techniques
Phase Demodulator
Residual or Added Noise Measurements
Single Source Measurements
Phase Detector with Two Sources
Reference Source
Voltage Controlled Source Tuning Requirements
Measurement Optimization
Measurement Examples
4
Slide 2
There would be no need to discuss the measurement of phase noise if all sources produced perfect sinewave signals
and if two-port devices were not capable of adding phase noise to a signal. The deviations from the pure sinewave
signal need to be quantified as a first step to determining their effect on the end results. In this equation, representing
the signal voltage with respect to time, e(t) represents amplitude variations or amplitude modulation of the signal
and ø(t) represents the phase fluctuations modulating the ideal linear phase change of the signal. There are two
fundamental ways to measure these perturbations of the signal: the first is to look at the signal directly on a spectrum
analyzer and the second is to demodulate the f luctuations of the carrier for analysis at baseband.
On a spectrum analyzer, the sum total of all the instabilities of a signal appear as sidebands on either side of the carrier.
The spectral density of these sidebands, S
v(vo
±f), can be read directly for a given offset. Demodulating the amplitude,
phase or frequency fluctuations produces a time-domain voltage analog of these fluctuations for measurement and
analysis. The analysis of this baseband signal can produce the spectral density of the amplitude fluctuations, S
A
(f), of
the phase fluctuations, S
ø
(f), or of the frequency fluctuations, Sv(f). Note that the spectral densities of phase and fre-
quency fluctuations are directly related by the square of the offset frequency.
Basic Phase Noise Measurement Concepts
V (t) = V
O
[ 1+
e(t)
] sin [2π v
O
t + ø (t)]
II
A(t) v (t) = v
O
+
1 dø (t)
Direct Spectrum Demodulate, then analyze
S
v (v
O
± f) SA (f) Sø(f) Sv (f) = f 2 Sø(f)
V
O
2π dt
Direct Spectrum Analysis
If AM << Pm L(f) = SV(v0±f)
Po
Agilent 3561A/62A Dynamic Signal Analyzer
Agilent 3582A Spectrum Analyzer
Agilent 3585A Spectrum Analyzer
Agilent 8566A/B Spectrum Analyzer
Agilent 8568A/B Spectrum Analyzer
5
Slide 3
The quantity that is usually measured in phase noise analysis is L(f), the single sideband phase noise of a signal.
This quantity is the noise power due to the phase fluctuations of the signal in a 1 Hz bandwidth at an offset f Hz
from the carrier normalized to the total signal power. If the AM noise is much less than the PM noise, L (f) is read
directly from the CRT of the spectrum analyzer as the relative level of the noise sidebands compared to the carrier
power. Corrections are necessary to normalize the results for a 1 Hz bandwidth and to account for the logarithmic
scaling of the spectrum analyzer. In addition, for a measurement of only the signal’s noise, the phase noise sidebands
to be measured must be greater than the spectrum analyzer’s own noise sidebands by about 10 dB. The spectrum
analyzers listed here are commonly used for a direct spectrum measurement of phase noise because they have
synthesized local oscillators (except the 3582A and 3561A which perform a Fourier conversion of the signal) to
prevent their own drift from affecting the result.
Demodulate, Then Analyze
If (ø2(t))<< 1 L(f) = Sø (f)
2
Agilent 5390A Frequency Stability Analyzer
Agilent 8901A/B Modulation Analyzer
Agilent 8902A Measuring Receiver
Agilent 3047A/11740A Spectrum Analyzer System
Agilent 3048A Phase Noise Measurement System
Agilent 11729C Carrier Noise Test Set
6
Slide 4
The phase noise on a carrier can be demodulated for analysis with a baseband spectrum analyzer to get the spectral
density of the phase modulation S
ø
(f). The single sided phase noise, L (f), can be calculated from the spectral density of
the phase fluctuations, S
ø
(f), (or frequency fluctuations, Sv(f) = f2x Sø(f) ) if the mean square phase fluctuations <ø2(t)>
are small relative to one radian. Listed here are some of the instruments that are used to do this demodulation and
analysis of phase noise.
7
Slide 5
Caution must be exercised when L(f) is calculated from the spectral density of the phase fluctuations, S
ø
(f), because
of the small angle criterion. This plot of L(f) resulting from the phase noise of a free-running VCO illustrates the
error that can occur if the instantaneous phase modulation exceeds a small angle. Approaching the carrier, L(f) is
obviously an invalid approximation of the actual phase noise as it reaches a relative level of +35 dBc/Hz at a 1 Hz
offset (35 dB more noise power at a 1 Hz offset in a 1 Hz bandwidth than the total power in the signal).
The –10 dB/decade line is drawn on the plot for an instantaneous phase deviation of 0.2 radians integrated over any
one decade of offset frequency. At approximately 0.2 radians the power in the higher order sidebands of the phase modulation is still insignificant compared to the power in the first order sideband, which ensures the calculation of L(f) is
still valid. Below the line the plot of L(f) is correct; above the line L(f) is increasingly invalid and S
ø
(f) must be used to
represent the phase noise of the signal.
8
Slide 6
Another way to represent the instability of a signal besides S
ø
(f) or L(f) is with a plot of the spectral density of
frequency fluctuations, S
v
(f). As illustrated before, Sv(f) is equal to f2x Sø(f) because v(t) is the derivative of ø(t). These
two graphs are from the same data with the left one a plot of S
ø
(f) and the right one a plot of the square root of
S
v
(f). The graph of the square root of Sv(f) indicates the power spectral density of the frequency modulation (FM)
noise the signal has on it. A measure of the spectral density of the FM noise versus the offset from the carrier would
be important in the design of an FM system, for example.
9
Slide 7
Let’s take a look at the direct spectrum method of measuring phase noise with a variety of spectrum analyzers.
Agenda
Basic Phase Noise Measurement Concepts
* Direct Spectrum Measurement
Demodulation Techniques
Phase Demodulator
Residual or Added Noise Measurements
Single Source Measurements
Phase Detector with Two Sources
Reference Source
Voltage Controlled Source Tuning Requirements
Measurement Optimization
Measurement Examples
10
Slide 8
As listed previously, there are a number of spectrum analyzers that will display the single sideband phase noise, L(f),
of a signal. With the exception of the Agilent Technologies 3582A and 3561A which perform a Fourier conversion, the
spectrum analyzers listed here have synthesized local oscillators to prevent the drift of the analyzer from affecting the
measurement of the phase noise sidebands. The Agilent 3048A is a phase noise measurement system that consists of an
interface box for frequency conversion and amplification, the Agilent 3561A Dynamic Signal Analyzer, a controller, and
software to run the measurement and produce the resulting graphs.
The 3048A system software provides direct spectrum measurements with the sub-Hz resolution of the 3561A
for carrier frequencies <100 kHz. It will set up the 3561A, measure and plot the resulting noise voltage.
11
Slide 9
One important criterion for choosing a local oscillator for the downconversion of signals to baseband frequency for
analysis is that the LO should not drift. The local oscillators listed here are synthesized to reduce their frequency
drift to a multiple of a highly-stable crystal reference oscillator. An alternative to the single conversion to baseband
using the mixer in the 3048A interface box is to do a preliminary downconversion using the Agilent 11729C Carrier
Noise Test Set. As explained later, this dual conversion method can produce better sensitivity when measuring the
phase noise of signals in the frequency range of 1.3 to 18 GHz. For signals above 18 GHz there is a millimeter version,
Option H33 to the 11729C. This option allows access to a very clean mm signal to downconvert the test signal to
the nominal range of the 11729C.
12
Slide 10
Throughout this presentation are a series of graphs illustrating the single sideband phase noise, L(f), of various sources
on plots covering an offset frequency range of 0.01 Hz to 40 MHz and down to a relative amplitude level of –180 dBc/Hz.
These graphs will provide a common format for comparing measurement techniques to the typical types of sources that
are measured. On the graph given here, L(f) is plotted for several types of oscillators ranging from a free-running VCO
(Agilent 8684A) to a highly-stable 10 MHz crystal oscillator used as the reference oscillator in many synthesized signal
generators. L(f) for the spectrum analyzers is overlayed on the graph to indicate which analyzer could be used to display the phase noise of typical sources.
Two measurement limitations for each spectrum analyzer are illustrated on this graph. The first is the analyzer’s internally generated noise floor. For the superheterodyne spectrum analyzers (Agilent 8566A/B, 8568A/B, and 3585A), the
phase noise of the analyzer’s synthesized local oscillator determines its sensitivity at offsets of less than approximately
1 MHz. Beyond a 1 MHz offset the noise of the analyzer’s IF circuitry sets its noise floor. The resolution of the Fourier
conversion and internal amplifiers determines the sensitivity of the 3582A. The second measurement limitation illustrated here is the minimum offset frequency specified by the analyzer. The superheterodyne spectrum analyzers are limited by their internal LO feedthrough to the IF Circuitry to a minimum offset of approximately 20 to 100 Hz. The 3582A
has measurement capability to within 0.2 Hz of the carrier due to the high resolution of its Fourier conversion process.
13
Slide 11
This is an example of the benefits of analyzing a microwave signal downconverted by the 11729C to an IF that is
then input for measurement on the 3561A Dynamic Signal Analyzer within the 3048A System. The measurement at the
upper left covers a 500 Hz span at 10.0 GHz and took approximately 1 second to complete on the 3561A. Sweeping the
8566A/B over the same range with a 10 Hz bandwidth would require 15 seconds during which any signal drift could
affect the results, and the resolution of low-level sidebands would be much more limited. Discrete sidebands are clearly
resolved with this technique. The frequency span can be decreased for better resolution until, as in the 10 Hz span of
the lower right plot, the carrier frequency is changing too much for this measure of single sideband phase noise to be
valid. The carrier instability exceeds the small angle criterion that L(f) depends on and a different measurement technique is required; one that determines the spectral density of the phase fluctuations rather than the power in the phase
noise sidebands.
Agilent
14
Slide 12
This list summarizes the limitations of using the direct spectrum measurement technique to measure phase noise.
Spectrum analyzers are valuable tools and widely used for fast, qualitative looks at the stability of a signal.
Direct Spectrum
Measurement Limitations
• Cannot separate AM and PM noise
• AM noise must be << PM noise
• Inadequate dynamic range for many sources
• Cannot measure close in to a drifting carrier
• Valuable for qualitative quick evaluation
15
Slide 13
Next let’s take a look at several measurement techniques that demodulate the phase fluctuations of the signal for
measurement and analysis.
Agenda
Basic Phase Noise Measurement Concepts
Direct Spectrum Measurement
* Demodulation Techniques
Phase Demodulator
Residual or Added Noise Measurements
Single Source Measurements
Phase Detector with Two Sources
Reference Source
Voltage Controlled Source Tuning Requirements
Measurement Optimization
Measurement Examples
16
Slide 14
Listed here are several systems that demodulate the phase noise of the signal in different ways. The 5390A Frequency
Stability Analyzer translates counter readings of the frequency of a signal over a period of time into the equivalent level
of phase noise. The 8901A/B Modulation Analyzer and 8902A Measuring Receiver employ an FM discriminator to demodulate the phase noise of a signal. The 3048A Phase Noise Measurement System can be used in several ways to analyze
phase noise, one of which is with an internal phase detector to mix the signal under test with synthesized oscillator.
Demodulation Techniques Related
to Specific Instruments or Systems
Agilent 5390A Frequency Stability Analyzer
Agilent 8901A/B Modulation Analyzer
Agilent 8902A Measuring Receiver
Agilent 3048A Phase Noise Measurement System
17
Slide 15
Although this system is now obsolete and cannot be ordered, it is instructive to examine the approach that was used
to measure phase noise. The counter in the system was used to measure the difference frequency of the signal under
test and a reference source. If the reference source is sufficiently more stable than the test signal and the test signal
does not drift during the measurement, variations of the difference frequency represent frequency (or phase) instability of the test signal. The system software compiles a series of readings of this difference frequency and calculates
the Allen or the Hadamard variance to determine the phase noise of the signal. This measurement approach can
yield phase noise data very close to the carrier with very good sensitivity if a low frequency beatnote is used.
Several significant limitations are inherent with this measurement technique. One is that the two sources used must be
offset to produce the beatnote to be counted. To overcome this problem, an option to the system was created to add a
second mixer such that the two oscillators of the same frequency to be compared were mixed with a third source at a
different frequency. With this variation the difference in period of the two beatnotes is measured and translated into
the corresponding phase noise. If the sources were of equal stability the result would be the combined phase noise of
both sources (the instability of the third source cancels out with this method).
To produce a valid phase noise measurement this system required a nondrifting signal to measure. Also, as this is
essentially a digital form of phase noise measurement with a series of discrete readings, aliasing is encountered such
that data at high offset frequencies is folded down to lower offsets according to the measurement rate. This aliasing of
the high offset phase noise would increase the phase noise readings at low offsets. This produced a requirement that
the phase noise of the signal under test be decreasing rapidly as the offset frequency increases so that the phase noise
power folded over to the lower offsets would not be significant.
Agilent
Agilent
Agilent
Agilent
Agilent
18
Slide 16
The sensitivity of the 5390A System and the offset range that could be measured were a function of the beatnote
frequency that was used. Excellent sensitivity was available with a beatnote of 10 Hz but the offset range was limited
to less than 1.6 Hz. This limitation is acceptable for measuring precision frequency oscillators used as time standards.
With increasing beatnote frequency, the 5390A System had a range of usefulness for measuring various sources but in
general could not produce a phase noise measurement out to the noise f loor of the oscillator under test.
Agilent
19
Slide 17
The 8901A/B Modulation Analyzer and 8902A Measuring Receiver convert the frequency fluctuations of a signal into
voltage variations with a frequency discriminator. The discriminator output can be connected to a spectrum analyzer
for a display of the spectral density of the phase noise over a range of offset frequencies or the noise can be integrated
over a bandwidth. A correction is made for the calibration constant of the discriminator to achieve calibration. This
calibration constant can be entered into the Agilent 3047A or 3048A System software for an automatically calibrated
output. The phase noise of the 8901A/B or 8902A Internal Local Oscillator is lowest for an input frequency below
300 MHz. For signals below 300 MHz the 8901A/B or 8902A sensitivity is maximized as is indicated on the next slide
of system sensitivity. An advantage of using a frequency discriminator approach as with the 8901A/B or 8902A is that
a certain amount of signal drift can be tolerated in making a valid measurement of the spectral density of phase noise.
Shown here are several methods for downconverting signals into the range of the 8901A/B or 8902A.
20
Slide 18
The curve for the 8901A/B or 8902A at 10 MHz on this graph is the sensitivity of the discriminator used in the analyzer
and actually extends to an offset of approximately 200 kHz for input signals above 10 MHz. At 1.28 GHz the phase noise
of the internal local oscillator of the 8901A/B or 8902A limits the sensitivity. This sensitivity is sufficient to measure
the phase noise of some free-running oscillators as indicated.
Agilent