AN-741

APPLICATION NOTE
One Technology Way • P.O. Box 9106 • Norwood, MA 02062-9106 • Tel
T : 781/329-4700 • Fax: 781/326-8703 • www.analog.com

Little Known Characteristics of Phase Noise

by Paul Smith

INTRODUCTION νOUT
There is a wealth of information available on the topic of ∆ν
phase noise, its characteristics1, how it can be measured2,
and how it affects system performance3. It is well known TIME
that phase noise in oscillators and clocks becomes one ∆τ
of the limiting degradations in modern radio systems.
However, most of the traditional analyses concentrate Figure 1.
on degradations to sine wave signals in single carrier
Traditional sampled data SNR analyses use Figure 1 as
radio systems. The effects of phase noise on multicarrier
an aid to determine how noise on a clock generates an
receivers, wideband systems, or digital radios are very
error in the sampled data. From this it is seen that
rarely discussed. This application note will address
some of the rarely discussed issues related to phase
noise in sampled data systems. It will focus primarily ∆ν( t ) = ∆t × ν'out ( t )
on multicarrier radios, wideband signals, and under-
sampled radio architectures.
{ } {
E ∆ν2 ( t ) = E ∆t 2 × ν'out (t )} 2

E {∆ν ( t )} = E {∆t } × E {ν' (t )},

2 2 2
out
PHASE JITTER IN SAMPLED DATA SYSTEMS
zero mean, independence
The easiest way to calculate the SNR degradations
Therefore,
incurred by phase noise in a sampled data system is to
convert phase noise to phase jitter. This is most easily
accomplished by recognizing that a time delay is the {
σ 2err = σ 2t × E ν' out ( t ) ,
2
}
same as a phase delay at a given frequency. Extending where σ 2t is in (rms seconds)
2

this concept and writing it in terms of noise power yields

Equation 1. From this it is seen that the noise power is a function of
the jitter power and the power in the signal derivative.
σ2θ = ω 2clk σt2 ,
The SNR of a signal sampled with a jittery clock is
where σ θ = phase noise in rms radians (1)
defined as
σt = phase jitter in rms seconds
ω clk = clock frequency in radians /sec
SNR sig =
power in signal
=
σ 2out
=
{
1 E ν out t
2
( )}
(3)
E {ν' (t )}
2 2
That is, for a given jitter error, a higher frequency signal power in noise σ err σ t
2
out
will have more phase error. The term σ θ is the total
integrated phase noise of the clock4 and defines the For example, in a single sine wave,
clock SNR by
ν out (t ) = A sin ω Ot
( )
(2)
SNR clk (dB ) = –10 log σ 2θ ν' out (t ) = A ω O cos ω Ot
Therefore,
Thus, Equation 1 relates the total integrated phase noise,
or clock SNR, to the total jitter in the clock. Phase noise
and clock jitter are two different ways to look at the same
{ 2
}
E ν out (t ) = power in ν out (t ) =
A2
2

E {ν' (t )} = power in ν'

phenomenon. ω 2O A 2
out (t ) =
2
out
2

REV. 0
AN-741
Using Equation 3, Many modern radio systems don’t use narrow-band
carriers. Modulated data often occupies a fairly wide
A2
spectrum. In order to determine how click jitter effects
1 1 1
SNRsig
si = 2 22 2 = 2 2 (4a) the SNR for such systems, it is convenient to assume
σt ω O A σt ω O
the data has zero mean and a flat spectrum uniformly
2 distributed between fL and fH , fL < fH as shown in Figure 2.
When squared and integrated over its bandwidth, the
1 total signal power σ 2out is obtained.
SNR sig = , for a single carrier system. (4b)
4π 2fO2 σ 2t
σout
fH – fH
This is the standard SNR equation for a single sine wave
sampled by a clock with jitter and can be found in many
FREQ
publications5. Intuitively what is happening is that higher fL fH
frequency signals have larger slew rates. This results
in larger voltage changes as the sample time changes. Figure 2.
It should be remembered that quantization noise and
One form of Parseval’s theorem states that the power
thermal noise must also be added to this to obtain the
of a signal in the time domain equals the power of the
total noise out of a data converter.
signal in the frequency domain. That is,
Extending this to a multicarrier signal is a simple matter.
+∞ 1 +∞ +∞
Using the same procedure as before with out defined as ∫ |ν(t )| dt = ∫ |g (ω )| dω = ∫ |g (f )| df
2 2 2

a summation of n equal amplitude sine waves, –∞ 2π – ∞ –∞

where |g (f )|2 is the power spectral density in Watt / Hz .

{
E ν out (t )
2
} =
nA 2
2 In addition, using the differentiation theorem of the Fourier

A (ω )
2 2 transform, which states that the Fourier transform of a
+ ω 22 + ...ω 2n
E {ν' (t )} = derivative is just the Fourier transform of the original
2 1
out
2 function multiplied by iω , as shown below,

nA 2 [ ]
ℑ ν' (t ) = i ωℑ ν(t ) [ ]
1
SNR sig = 2 2 2 22 =
(
σ t A ω 1 + ω 2 + ...ω 2n ) and combining this with Parseval’s theorem, it is seen
that the power in ’(t ) is the same as the power in ig
2 (), as described below,
1 n n
=
2
(
σ t ω 1 + ω 2 + ...ω n
2 2 2
4 π σ t ∑ f i2
2 2
) n
+∞
( ) 1
∫ ν' (t ) dt = 2π ∫ |iωg (ω )| dω =
2
2
+∞

–∞ –∞
1 +∞ 2 +∞
ω |g (ω )|2 dω = ∫ (2πf ) |g (f )|2 df
2
This is relative to the entire signal, ν out . When referenced ∫
2π – ∞ –∞
to only one of the carriers, the SNR becomes
1 σ out
SNR sig = ,
4π 2
σ 2t ∑ f i2 For g (f ) = f H – f L only between fL and fH (and zero
n
(5)
everywhere else), this becomes
for a sin gle carrier in a multicarrier system .

{ }
Compared to the single carrier case, Equation 4b, the fH 2
σ
E ν' out (t ) = ∫ (2πf )
2 2 out
df
denominator has n more frequency terms. The SNR on
a per carrier basis (i.e., dBc) has been degraded by
(f – f )
fL H L

E {ν' (t ) } =
2 fH
σ
approximately 10 log(n). However, in a data converter
∫ (2πf ) df
2 out 2
out
each carrier may need to be reduced by 10 log(n) to f –f H L fL
20 log(n), depending on signal statistics, in order to keep 4 π (f – f )σ 4 π (f )
E {ν' (t ) } =
2 3 3 2 2 2
2 H L out H + fH fL + fL2 σ2out
from clipping the quantizer. This, in effect, raises the =
3(f – f )
out
quantization and thermal noise floor by up to 20 log(n). H L 3
Thus, jitter may contribute less to the overall SNR than
in the single carrier case. Quantization and thermal noise Using Equation 3,
may become more dominant. 1 3
SNR sig = 2
(
σ t f H + f H f L + f L2
2
)
–2– REV. 0
 AN-741
This is the SNR resulting from a flat, wideband signal 2

between fL and fH being sampled by a clock with jit ter 1 f  (7a)

SNR sig = 2  clk 
σ t . As a sanity check, setting fL = fH = fO (i.e., all the σ θ  f sig 
power resides in a single frequency term fO ) results in
the same expression as Equation 4b for the single fre- Using Equation 2, this can be written as
quency case.
2
An alternative expression is obtained by letting fL = f 
SNR sig = SNR clk  clk  (7b)
fO –BW /2 and fH = fO +BW /2. For this case the expression  f sig 
becomes
The SNR of the resulting sampled signal is the same as
1
SNR sig = , the SNR of the clock but scaled by the clock and signal
 BW 2 
4π
4 π 2 σ 2t  fO2 + (6) frequency ratio. As the signal frequency gets higher,
 12  the SNR degrades in a 20 log fashion. This illustrates
for a flat signal centered at fO having bandwidth BW . why undersampled systems (i.e., one in which the
signal frequency occupies one of the higher Nyquist
Again, as a sanity check, when BW =0 the result matches bands) require clocks with much better phase jitter than
the single carrier case in Equation 4b. baseband systems. In fact, performance in IF-sampling
digital radio architectures are often limited by clock
A consequence of all this math is that as long as fO>10BW,
BW
BW,
phase noise, not data converter performance.
the bandwidth of the signal can probably be neglected.
Treating the modulated signal as a single carrier will give Although not apparent from Equation 7b, the spectral
virtually equal results. However, if this is not true, then shape of the clock phase noise is superimposed on the
using the single carrier approximation will give results sampled data, as illustrated in Figure 3. This can intui-
that are too optimistic. tively be seen by modeling the sampling process with
a mixer. As shown in Figure 4, when a clock with phase
This entire discussion has focused on sampled data
noise  is applied to a mixer, the output contains two
systems, but the effects of aliasing have not been men-
mixing products, each of which contains the full phase
tioned. All of the equations derived above assume there
noise  of the clock. Although this simplistic model does
is no aliasing. The bandwidth of the jitter is considered
not show the scaling factor described in Equation 7, it
to fall entirely (and conveniently) into a single Nyquist
is useful to show how the phase spectrum of the clock
zone. If the jitter is bad enough, and the signal close
shows up on the resulting signal.
enough to a Nyquist edge, the noise caused by jitter can
alias back in band, degrading SNR even further. This ef- RV sin((fclk
Vout = 1/2[RV clk + fsig)t + (ff ))
R cos(fsigt)
+ RV sin((fclk
clk – fsig)t + (ff ))]
fect is illustrated in Figure 3. A similar problem exists
with clock feedthrough. If the signal is close to the clock,
phase noise from the clock can directly leak to the output, V sin(fclk
clkt +(f ))
degrading the noise floor.
Figure 4.
SIGNAL
SIGNAL
SNR
ALIAS This can easily be tested by phase modulating a clock
DEGRADATION and feeding it into an ADC. By applying different signal
frequencies, Equation 7 can also be verified. An AD9430
ADC was clocked at 61.44 MHz with a clock phase modu-
lated such that the first sidebands were –60dBc. Figures
5a, 5b, and 5c show the results of the experiment.
fSIG fALIAS
fs/2
s/2
Figure 5a shows the results with a 3.84 MHz input. The
Figure 3. clock modulation components can be seen as the two
small spurs clustered close to the signal fundamental.
PHASE NOISE IN SAMPLED DATA SYSTEMS According to Equation 7, the clock modulation spurs
In addition, nowhere in the preceding discussion did the should be –60 – 20 log(61.44/3.86) = –84 dBc. The results
spectrum of the clock phase noise come into play. All in Figure 5a are quite close to this number.
that was considered was the total jitter (in rms seconds) Figure 5b shows the results with a 65.28 MHz input. This
which was calculated from the total integrated phase is in the 3rd Nyquist zone. The FFT shows the baseband
noise using Equation 1. To see how the phase noise spec- alias in the same location as the 3.84 MHz signal in
trum of the clock affects the sampled data spectrum, Figure 5a (i.e., 65.28 MHz – 61.44 MHz = 3.84 MHz). Here
it is most convenient to use a single sine wave signal. fclk ~ fsig and the –60 dBc clock spurs can readily be seen
Combining Equation 1 with 4a yields Equation 7a. superimposed on the signal, also at –60 dBc. This is to
be expected according to Equation 7b.

REV. 0 –3–
AN-741
Figure 5c shows the results with a 124.72 MHz input Thus, it appears the clock spectrum does indeed appear
in the 5th Nyquist zone. This frequency is about twice around the sampled signal with a scaling factor described
that of Figure 5b and according to Equation 7 the spurs by Equation 7. However, so far, all of the preceding dis-
should increase about 6 dB, which is what is seen. cussions have not differentiated between ADCs and
0
DACs. Do DACs exhibit the same characteristic seen by
–10 ADCs? A similar experiment was run on an AD9744 DAC
–20 using a 61.44 MHz clock phase modulated to give –40 dBc
–30 sidebands, generating an 11 MHz sine wave. The results
–40 over five Nyquist bands are shown in Figure 6.
–50
�
������
–60 �����
��� �����
–70
+ 3
–80 ���
2 4 5 6
–90
���
–100
���
–110
–120 ���

–130 ���
0 2.5 5.0 7.5 10.0 12.5 15.0 17.5 20.0 22.5 25.0 27.5 30.0
FREQUENCY (MHz)
���

Figure 5a. ���

� ���
��� ����
��� ������������ ��������� �������������

���
Figure 6.
���
��� The Sinc function inherent in DAC outputs can clearly be
�
��� seen. But what is happening with the clock spurs? These are
���
�
clearly seen at each of the output images but the amplitudes
�
���
� � � don’t increase as they did with the ADC. Relative to full scale,
��� the spur amplitudes remain constant.
����
����
There are several ways to look at this. When viewed in
���� dBc, as the signal frequency goes up, the modulation
���� spurs get worse in the same manner as described by
� ��� ��� ��� ���� ���� ���� ���� ���� ���� ���� ���� ����
��������������� Equation 7b. The Sinc function applies to both the signal
amplitude and the induced clock phase noise. Calculat-
Figure 5b. ing the spur amplitude relative to each carrier (i.e., in
� dBc), Equation 7b is a good description.
���
Alternatively, the Sinc characteristic is defined as
���
���
 πf sig 
��� sin  
���  f clk 
�
��� πf sig
���
� � � � f clk
���
�
��� The amplitude of the noise is given by the reciprocal of
���� Equation 7a.
2
����
N  f sig 
���� = σ 2θ  
����
S  f clk 
� ��� ��� ��� ���� ���� ���� ���� ���� ���� ���� ���� ����
���������������
That is, the noise is directly proportional to clock phase
Figure 5c.

–4– REV. 0
 AN-741
signal. For example, if a spur is 10MHz away from the
noise and signal frequency. Squaring the Sinc function
carrier, look to see if there is a 10MHz oscillator some-
(because power spectral densities are being examined)
where in the system. If so, this frequency is most likely
and multiplying these two to get a composite noise
leaking in through the clock.
transfer function out of the DAC yields

 πf sig  SUMMARY
sin2   This application note examined the relationship between
N  f clk  phase noise and jitter, deriving the SNR degradations
= σ 2θ
S π2 that occur when a signal is sampled by a clock with
jitter. The results were extended to multicarrier and
The periodic nature of the nulls caused by the sinusoid
wideband modulated data systems. Clock phase noise
still exist. However, the denominator of the Sinc function
spectral issues were then dealt with, examining the
is what causes the roll-off at higher frequencies. This
resulting spectrum at the output of ADCs and DACs.
attenuation has been exactly cancelled by the increas-
Finally, the results are applied to debugging a system
ing phase noise at higher frequencies described by
that may have unusual spurs that would otherwise be
Equation 7b. Thus, the phase noise out of a DAC will not
unaccounted for.
grow at higher frequencies.
NOTES
APPLICATION TO SYSTEM DEBUGGING 1
W. P. Robins, Phase Noise in Signal Sources (London:
Besides the obvious issues revolving around design-
Peter Peregrinus Ltd., 1982).
ing systems to minimize signal degradations, there 2
“Understanding and Measuring Phase Noise in the
are several other consequences to these results worth
Frequency Domain,” Hewlett Packard, Application Note
mentioning. These are related to finding the source of
207, 1976.
mystery spurs and noise. For instance, if the noise floor 3
Stanley J. Goldman, Phase Noise Analysis in Radar
rises at the DAC output, it is most likely not caused by
Systems Using Personal Computers (New York: John
clock phase noise. It may be digital coupling into the
Wiley & Sons, 1989).
output circuitry. 4
“VCO Phase Noise,” Mini-Circuits, Application Note #2.
If a spur exists in a sampled signal, a good test to see if it 5
“Linear Design Seminar,” Analog Devices, Inc., Nor-
comes from the clock is to change the signal amplitude. wood, MA, 1995, pp. 5-20.
Analog distortion terms will change at twice (2nd order
distortion) or three times (3rd order distortion) the rate
of the signal amplitude change. Spurs due to nonlinear-
ity in the quantizer may not change at all, or if they do
change, they will change unpredictably, when the signal
amplitude changes. On the other hand, spurs due to the
clock will change dB for dB with the signal.
When trying to identify the source of a spur in a sampled
data signal, look not only at the explicit spur frequency,
which could be caused by a signal directly coupling into
the output, but also at the frequency offset from the

REV. 0 –5–
–6–
–7–
 AN04985–0–8/04(0)

© 2004 Analog Devices, Inc. All rights reserved. Trademarks and registered trademarks are the property of their respective owners.
–8–