|
Home Up Softrock Lite 6.2 Adventures in Electronics and Radio Elecraft K2 and K3 Transceivers Updates Current Products Prior Products - no longer available Documents Book Software Updates
| |
|
Programmable Canned Oscillator Phase
Noise
I've been intrigued with the small custom frequency
programmed oscillator modules, such as Cardinal Components' CPP series units.
http://www.cardinalxtal.com/docs/series/CPP.pdf. The CPP series is a
one-time programmable module, with 1 Hz steps and is modestly priced at $8.18
each from DigiKey, programmed to your frequency.
If, and it's a big if, these modules have acceptable phase
noise they would be extremely handy for BFO or 2nd oscillators in one-off or low
volume designs. A few months ago, I purchased several of these oscillators for
21.4 MHz and was disappointed at their broadband noise output. This page
provides phase noise comparisons of three oscillators:
- Cardinal CPPT1-H5RT one-time programmable oscillator at
21.412 MHz.
- Raco 20 MHz crystal controlled canned oscillator
module.
- Home brew crystal oscillator at 21.418 MHz.
- Z90 Digital Panadapter in signal generator mode (AD9851
DDS, similar in design to the DDS-60)
Revision History
Original
12 July 2009. Expanded to include Si570, AP3S and CPPC7 oscillators
Table of Contents
Test_Setup
Home_Brew_Colpitts_Oscillator_-_21.418_MHz.
Raco_Oscillator_Module_(Crystal,_non-programmable)_20_MHz.
Cardinal_CPP_Programmable_Oscillalator_Module_21.412_MHz.
Partial_Conclusions
Zero_Offset_Phase_Noise,_Volts/Square_Root_(Hz)
Comments
Updates_and_Additions_of_July_2009
Oscillators_Tested
KE5FX_PN_Software_Run
Notch_Filter_Test_Setup
HP8640B
Si570
AD9851
Conclusions |
Test Setup
The photo below shows the home made crystal oscillator at
left (Colpitts design, with buffer and power amplifier stage), a fixture to hold
canned oscillator modules with the Raco oscillator installed and the Cardinal
programmable oscillator. |
 |
The test equipment setup is illustrated below. The pads between the oscillator
under test and the mixer input were sized to provide -10 dBm input into
the mixer for each oscillator. The ZP1-MH mixer requires +13 dBm local
oscillator drive, provided by an HP 8640B signal generator operating at +19 dBm
output, with a 6 dB pad. The mixer's output connects to an HP 3562A Dynamic
Signal Analyzer, operating in spectrum analyzer mode.
The 8640B's frequency was set to 50 KHz above the
oscillator under test's frequency to provide a 50 KHz beat note output into the
HP 3562A Dynamic Signal Analyzer.
This method of phase noise measurement, of course,
actually measures the combined noise of the HP 8640B and the O.U.T. Although the
HP 8640B is a low phase noise signal source, its phase noise is not zero. Hence,
don't take the crystal controlled oscillator data as representing the oscillator
noise only. The Cardinal programmable oscillator's noise is sufficiently large,
compared with the 8640B, that its noise is the dominant factor in the 3562A's
display, however.
At 50 KHz offset, we may expect the 8640B's phase noise to
be in the -150 dBc/Hz range. John Grebenkemper, KI6WX, has provided phase noise
measurements of his HP 8640B, as well as his Elecraft K2 transceiver, at
http://home.pacbell.net/johngreb/k2phasenoise.pdf.
|
 |
|
The data below is taken at a center frequency of 50 KHz,
corresponding to the 50 KHz beat note between the 8640B and the O.U.T. Since the
oscillator input power into the mixer is approximately the same (±1 dB) the
signal levels may be directly compared. Data is shown for spans of 100 KHz, 10
KHz and 1 KHz, for each oscillator type.
Home
Brew Colpitts Oscillator - 21.418 MHz.
100 KHz Span |
 |
|
10 KHz Span |
 |
|
1 KHz Span |
 |
Raco Oscillator Module (Crystal, non-programmable) 20
MHz.
100 KHz Span |
 |
|
10 KHz Span |
 |
|
1 KHz Span |
 |
Cardinal CPP Programmable
Oscillalator Module 21.412
MHz.
100 KHz Span |
 |
|
10 KHz Span |
 |
|
1 KHz Span |
 |
|
Z90 Digital Panadapter Signal Generator Output
My Z90/91 digital panadapter has an AD9851 DDS local
oscillator. The AD9851 design is similar to the DDS-60 and should have similar
phase noise response. The Z90 includes an auxiliary signal generation mode and
the following plots show its phase noise performance at 21 MHz using the same
test setup as the other tests.
100 KHz Span |
 |
|
10 KHz Span |
 |
|
1 KHz Span |
 |
|
Partial Conclusions My home brew
Colpitts oscillator, designed without paying particular attention to low phase
noise, has slightly lower phase noise than the Raco oscillator module, most
noticeable when viewed with 1 and 10 KHz spans. The Z90's AD9851 DDS-based
oscillator is not too bad for close-in phase noise, but exhibits the typical
spurious responses found in this chip, most noticeable in the 100 KHz view.
Slight changes in frequency result in changes in spurious levels so these
results are only examples of one particular operation mode.
The Cardinal one-time programmable oscillator has much
inferior phase noise, viewed at all spans. At 100 KHz span, we see broadband
noise approximately 25 dB above either crystal controlled oscillator. When
viewed close-in, at 1 KHz, the one-time programmable oscillator does not look as
bad, relative to the crystal oscillators, but still shows increased noise
levels, of perhaps 8 - 10 dB over either crystal oscillator's performance.
The image below shows three oscillators, 100 KHz span,
stacked onto one picture. The stack order is: Cardinal : Raco : Home brew
oscillator. The image shows there is a small but perceptible difference between
the Raco and home brew crystal oscillators and that the Cardinal one-time
programmable oscillator has a huge broadband noise disadvantage.
|
 |
|
The image below stacks the Cardinal programmable oscillator
(red), the Z90's AD9851 DDS (green) and the discrete crystal oscillator (blue).
The data shows that the AD9851's phase noise is not too bad, but its discrete
spurious outputs are much more of concern. (The Z90, along with the DDS-60, uses
an inexpensive 30 MHz crystal time base. A higher quality time base with lower
jitter will improve the AD9851's phase noise, but will not do anything for
discrete spurious responses.) |
 |
|
Zero Offset Phase Noise, Volts/Square Root (Hz)
Traditionally, phase noise plots are presented with a zero
offset, and the vertical axis in volts / square root Hz, scaled in dBc, i.e., dB
below the carrier level. The plots below are in that format, but the vertical
reference point is not scaled to 0 dB = carrier level. However, since the mixer
input level is approximately the same in all plots, the relative phase noise
levels are correct.
One further point—the data is presented log frequency
scale from 1 Hz to 100 KHz. Since there is some drift in both the 8640B signal
generator (even though it is phase locked) and the oscillator under test, data
below about 10 Hz is suspect and should be disregarded.
The data is the average of 16 sweeps.
The plot caption Y axis reads "V/ Hz." The
plotting program does not reproduce the square root symbol and the Y axis should
read V/Sqrt(Hz).
And, of course, as with all the data on this page, the
phase noise presented is the composite of the oscillator under test and my
HP 8640B signal generator.
|
|
Z90 Oscillator—AD9851 DDS |
 |
|
Discrete Crystal Oscillator |
 |
|
Cardinal CPP One-Time Programmable Oscillator |
 |
|
Comparison of Three Oscillators |
 |
|
Comments As with the other
data, the Cardinal one-time programmable oscillator has significantly greater
phase noise, particularly in the range > 1 KHz carrier offset.
The AD9851's phase noise is not bad, but has a number of
discrete spurious outputs.
Caveats:
- The data presents the composite phase noise of the
O.U.T. and my HP8640B signal generator.
- Data below 10 to 20 Hz should be disregarded as it
reflects short-term frequency drift as well as phase noise.
- The Y axis 0 dB reference point does not represent 0
dBc.
|
|
Updates and
Additions of July 2009 Jeff, AC0C sent a couple of newer design inexpensive
synthesized crystal oscillators for me to look at, with the thought that they
might be usable in a Softrock receiver as a substitute for a custom quartz
crystal or a less expensive single frequency alternative to the Si570.
That request lead me to purchase the Si570 oscillator and
controller kit developed by John, K5JHF, and Kees, K5BCQ,
http://www.qsl.net/k5bcq/Kits/Kits.html. I've since built the kit and my
review of the kit and the Si570 is at Si570
Kit from K5BCQ.
The two "synthesizers in a can" oscillators Jeff provide
are updated versions of the Cardinal CPP oscillator analyzed above. For details
on the Si570, see my review of the kit.
Since my first work on this page, I've expanded my test
equipment collection and also have given further thought to phase noise
measurements. Accordingly, I revisited some of earlier oscillators as well as
the new ones.
Oscillators Tested
- Si570. A detailed datasheet for the Si570 can be downloaded at
https://www.silabs.com/products/clocksoscillators/xo/Pages/default.aspx by
selecting the Si570/571 in the "Resources" tab.
- CPPC7, a smaller surface mount version of the
CPPT1-H5RT oscillator, also from Cardinal.
- AP3SLJ, an even smaller surface mount oscillator,
manufactured by Abracon.
- Si570. See the references above for details.
- HP 8640B, the "gold standard" for low phase noise
analog signal generators.
- AD9851 direct digital synthesis chip, as implemented
in my Z90 panadapter.
The two surface mount synthesizers are tiny. The photo
below shows both installed "dead bug" style on a test fixture. The CCPC7 module
at the right is small, but the AP3S at the lower left is absolutely tiny. To
provide a reference scale, the four screw heads are for 4-40 thread and the
surface mount bypass capacitors are 1206 size parts. The AP3S synthesizer is
smaller than a 1206 capacitor!
|
 |
|
KE5FX PN Software Run
My first measurement is based on John Miles's
PN program
that collects phase noise data with one of a variety of spectrum analyzers
operating under
HPIB control. The image below shows the phase noise measurements taken with the PN program and my Advantest R3463 spectrum analyzer.
To see whether the measurements are limited by the R3463,
particularly for close-in measurements, I looked at three signal generators, one
crystal oscillator, the three "synthesizers in a can" and the Si570.
The plot below shows the result. (Click on the plot image for a larger
version). Two conclusions can be drawn from this image:
- The Advantest R3463 is not suitable for phase noise measurements where
the oscillator under study has low phase noise. For example, at 100 KHz from
the carrier, all the low noise oscillators measured, such as the 30 MHz
crystal oscillator or the 8640B, should have phase noise in the -150 to -180
dBc/Hz range. The R3463 measures these at -120 dbC. (dBC/Hz means dB below
the carrier, i.e., the signal being tested, normalized to a 1
Hz wide measuring bandwidth.)
- All three "oscillators in a can" have such high phase noise as to render
the infirmities of the R3463 immaterial. None of the three
"oscillators in a can" seem suitable for even a moderate performance
receiver. Hence, these three devices will not be further examined.
|
 |
|
Notch Filter Test Setup
Of the methods I've used to examine oscillator noise, the one
I currently favor is the crystal notch filter approach, illustrated below.
This consists of an 8 MHz coupled resonator bandpass
filter of 300 KHz bandwidth, with a very narrowband notch in the center. The
notch is formed by shunting 4 carefully matched crystals across the four coupled
resonator inductors. The signal being observed is adjusted in frequency so that
the carrier is removed by the narrow notch, with what remains being phase noise
and broadband noise.
The notch depth exceeds 110 dB, so when the carrier is
notched down that level, it's quite easy to see the residue which consists of
broadband noise and phase noise. This setup is quite a bit more sensitive than
looking at the beat note output, as the HP 3562A dynamic signal analyzer has
only about 80 to 90 dB dynamic range.
|
 |
The image below shows the frequency response of the notch filter, measured with
an HP 8752B vector network analyzer. The upper black trace is 1 MHz wide,
centered at 8 MHz, whilst the lower blue trace is 10 KHz wide, centered on 7996
KHz, the crystal notch center.The 8752B measured
notch depth is -102 dB, but this measurement is limited by the 8752B's dynamic
range. Careful measurements with the 8640B indicate the deepest notch depth is
approximately 115 dB.
|
 |
The 3 dB bandwidth of the notch is 18 KHz. Thus, phase noise
within ±10 KHz or so of 7996 KHz will display lower than it actually is.
|
|
HP8640B
As a reference, the first oscillator tested with the notch is
my HP 8640B. The 8640B's output is set to +14 dBm to match the Si570's output
level. At an offset of 25 KHz from the notched carrier, the spectrum analyzer
reads a noise level of -135 dBm/Hz, or -149 dBC/Hz. There's an additional 2 dB
bandpass filter loss that should be factored into this measurement, so the net
measured noise at 25 KHz offset is -147 dBC/Hz.
This measurement is in good agreement with 8640B
measurements made by John Grebenkemper, KI6WX, at
http://home.pacbell.net/johngreb/k2phasenoise.pdf (Page 4 measurements for
8640B at 7 MHz (40 meter band.)
|
 |
|
Si570
The spectrum analyzer capture below is the same setup, but with the Si570 as the
signal source. Again the carrier is notched below the noise floor of the
spectrum analyzer.
At an offset of 25 KHz, the noise level is -121 dBm/Hz.
For a carrier level of +14 dBm and 2 dB filter loss the corresponding phase
noise is -134 dbC/Hz. As a reference point, KI6WX measured Elecraft's K2 phase
noise on the 40 meter band at approximately -140 dBC/Hz at 25 KHz offset.
At an offset of 10 KHz, the measured noise is -115 dBm/Hz,
corresponding to a phase noise of -128 dBC/Hz. As a point of reference, KI6WX
measured the K2's phase noise (receive) at 10 KHz offset as between -120 dBC/Hz
and -138 dBC/Hz, depending on the frequency band selected.
Closer in data in this plot require adjusting for the
non-negligible notch loss but it's clear that the Si570 is quite a decent
performing device, head and shoulders above the other "synthesizers in a can"
looked at and, at frequencies more than ±100 KHz from center, extremely low in
phase noise. |
 |
|
AD9851
The last oscillator examined is the Z90's AD9851, via its
auxiliary signal generator output mode. The Z90 uses a relatively inexpensive 30
MHz time base and the AD9851 is one generator or more behind AD's current DDS
chips. The AD9851 is also used in the popular DDS-60 board.
The image below shows the 7996 carrier notched into the
noise, but note the large number of discrete spurious signals. These are an
unfortunate problem in DDS devices, due to finite phase accumulator length and
D/A finite resolution and step errors. In this case, the strongest spurious
signals at ±38 Khz or so from the carrier are -72 dBm, or about 75 dB below the
carrier. Although the phase noise at 25 KHz offset is quite low, the large
number of discrete spurious signals (which extend far beyond the 200 KHz span in
the spectrum analyzer image) are troublesome.
|
 |
|
Conclusions
HP's 8640B is still the low phase noise source of choice, but
the Si570 is an excellent performing synthesizer, particularly considering its
modest price and small size. The AD9851 generation of DDS chips show significant
infirmities with respect to spurious discrete signals.
The one-time-programmable "synthesizers in a can" devices
remain unusable for serious receiver purposes. |
|
|
|
|
|
|
|
|
|
|
|