DIY Low offset Phase Noise Analyzer (Erik Kaashoek)
To Bob kb8tq
Figure Of Merit sounds like a useless number. I have a different
approach that yields immediate and useful results. Before I explain
my method, let me introduce myself.
In 1970, I invented, and Memorex patented, the original
zero-deadband phase-frequency detector. You can see it in page 3 of
my '234 patent at https://patents.google.com/patent/US3810234A/
This invention soon led to another invention of tremendous
significance to today's world.
In 2014, researchers published a study in the journal Supercomputing
Frontiers and Innovations estimating the storage capacity of the
Internet at 1e24 bytes, or 1 million exabytes.
When I started working for Memorex, an IBM 2314 disk pack could
store 29.2 million bytes. At that rate, today's internet would
require 1e24/29e6=3.44e16, or 34,400,000,000,000,000 IBM 2314 disk
drives. This is an impossible number. Other estimates give equally
outrageous numbers.
The problem in those days was improvements in disk drive capacity
were basically trial and error. This is a slow and very expensive
business.
My new invention allowed peering into the hard disk and separating
out all the variables that affect performance. With this
information, researchers could see the effect of changes and quickly
optimize the performance. This allowed the tremendous improvement in
tape and disk drive capacity that now allows the internet to store
all the needed data.
You can see how this invention works in the Katz paper at
https://tinyurl.com/2bmuz3n2
Now for my new method.
The schematic for a phase-frequency detector is shown in
DBAND2S.PNG. In operation, a pulse arrives at the DATA pin and pin
U1Q goes high. Then a pulse arrives at the VCO pin and pin U2Q goes
high.
This allows the NAND gate to bring the CLR signal low, which
immediately resets both d-flops.
The result is shown in ZERODB.PNG. It is a very narrow pulse with
both d-flops superimposed.
This is the basis for my new approach. Simply tie both inputs of the
PFD together and measure the noise spectrum of the output. (Of
course, you have to ensure that both outputs match at zero error.)
Once you have the PFD noise, you can enable the loop and measure the
total noise spectrum. Then simply subtract the PFD spectrum to get
the OCXO noise. If you have two identical VCXO's, each one
contributes half the noise.
I don't know if this method would work with a double-balanced mixer.
The problem is a DBM requires quadrature signals, so the noise is a
function of the OCXO noise as well as the mixer diodes. But the OCXO
noise is what you are trying to measure.
This method works with the PFD since only a single pulse is needed
to activate both d-flops, so you are measuring only the PFD noise.
Et Voila.
Now that you can measure the OCXO noise, you might want to try your
hand at designing an oscillator with minimum noise. You immediately
run into a problem. The high Q of the crystal means the oscillator
takes a very long time to start up.
I solved this problem in my OSC.ZIP file at
https://tinyurl.com/2p9yrxmy
Steve Wilson is me. Just start at the README.TXT file and you are on
your way.
Now that you are a fully qualified Time-Nut, you might be interested
in some of the following papers:
Rohde, 1994 How to improve phase noise by multiple varicaps in parallel
http://www.arrl.org/files/file/Technology/ard/rohde94.pdf
Leeson Equation
http://rfic.eecs.berkeley.edu/~niknejad/ee242/pdf/eecs242_lect22_phasenoise.pdf
Oscillator Phase Noise: A Tutorial
Thomas H. Lee, Member, IEEE, and Ali Hajimiri, Member, IEEE
http://smirc.stanford.edu/papers/JSSC00MAR-tom.pdf
Hajamiri
Virtual Damping and Einstein Relation in Oscillators
https://authors.library.caltech.edu/523/1/HAMieeejssc03.pdf
Hi
Regardless of what you call the “ 1 Hz normalized noise “ of a digital
phase detector, it does predict what the noise floor does on it as the
reference frequency is changed over some reasonable range. This has
been demonstrated a lot of times and on a lot of different parts.
Based on a number of RF designs using them ( and using gates for RF
purposes ) the basic gate is what is at fault here. They are noisy and that
noise changes with frequency. Frequency goes up / noise goes up. There
are very good reasons for this.
Getting a gate with a noise figure below 6 db is highly unlikely …. That
is what you would have to do in order to make a gate based circuit measure
a lower noise floor than the DBM based approach. Folks have spent a lot
of time searching for the magic “zero noise gate”.
The sine wave component present at the DBM output at 2X the input
frequency ( in the case of the phase noise test setup) are way higher
than the highest noise you are after. You put in 10 MHz or 100 MHz and
you go up to maybe 100 KHz on the noise. With a sound card, even
getting to 100 KHz is going to be a challenge. 20 KHz may be the max.
Knocking down the 2 x Fin component with a low pass filter is pretty easy.
Indeed the sound card or audio spectrum analyzer likely has some filtering
already. The design and implementation of an adequate LPF is far from the
biggest challenge that the person building the circuit will face.
Indeed 1/F noise and noise corners do matter. All of the above has been
simply talking about noise floor. Gates have significant 1/F issues along
with their other “features”. This carries over to the detectors based on
them. As the gate speed goes up ( and the floor typically comes down),
the 1/F corner normally moves up ….
Bob
On Jul 11, 2022, at 8:05 AM, Mike Monett via time-nuts time-nuts@lists.febo.com wrote:
To Bob kb8tq
Figure Of Merit sounds like a useless number. I have a different
approach that yields immediate and useful results. Before I explain
my method, let me introduce myself.
In 1970, I invented, and Memorex patented, the original
zero-deadband phase-frequency detector. You can see it in page 3 of
my '234 patent at https://patents.google.com/patent/US3810234A/
This invention soon led to another invention of tremendous
significance to today's world.
In 2014, researchers published a study in the journal Supercomputing
Frontiers and Innovations estimating the storage capacity of the
Internet at 1e24 bytes, or 1 million exabytes.
When I started working for Memorex, an IBM 2314 disk pack could
store 29.2 million bytes. At that rate, today's internet would
require 1e24/29e6=3.44e16, or 34,400,000,000,000,000 IBM 2314 disk
drives. This is an impossible number. Other estimates give equally
outrageous numbers.
The problem in those days was improvements in disk drive capacity
were basically trial and error. This is a slow and very expensive
business.
My new invention allowed peering into the hard disk and separating
out all the variables that affect performance. With this
information, researchers could see the effect of changes and quickly
optimize the performance. This allowed the tremendous improvement in
tape and disk drive capacity that now allows the internet to store
all the needed data.
You can see how this invention works in the Katz paper at
https://tinyurl.com/2bmuz3n2
Now for my new method.
The schematic for a phase-frequency detector is shown in
DBAND2S.PNG. In operation, a pulse arrives at the DATA pin and pin
U1Q goes high. Then a pulse arrives at the VCO pin and pin U2Q goes
high.
This allows the NAND gate to bring the CLR signal low, which
immediately resets both d-flops.
The result is shown in ZERODB.PNG. It is a very narrow pulse with
both d-flops superimposed.
This is the basis for my new approach. Simply tie both inputs of the
PFD together and measure the noise spectrum of the output. (Of
course, you have to ensure that both outputs match at zero error.)
Once you have the PFD noise, you can enable the loop and measure the
total noise spectrum. Then simply subtract the PFD spectrum to get
the OCXO noise. If you have two identical VCXO's, each one
contributes half the noise.
I don't know if this method would work with a double-balanced mixer.
The problem is a DBM requires quadrature signals, so the noise is a
function of the OCXO noise as well as the mixer diodes. But the OCXO
noise is what you are trying to measure.
This method works with the PFD since only a single pulse is needed
to activate both d-flops, so you are measuring only the PFD noise.
Et Voila.
Now that you can measure the OCXO noise, you might want to try your
hand at designing an oscillator with minimum noise. You immediately
run into a problem. The high Q of the crystal means the oscillator
takes a very long time to start up.
I solved this problem in my OSC.ZIP file at
https://tinyurl.com/2p9yrxmy
Steve Wilson is me. Just start at the README.TXT file and you are on
your way.
Now that you are a fully qualified Time-Nut, you might be interested
in some of the following papers:
Rohde, 1994 How to improve phase noise by multiple varicaps in parallel
http://www.arrl.org/files/file/Technology/ard/rohde94.pdf
Leeson Equation
http://rfic.eecs.berkeley.edu/~niknejad/ee242/pdf/eecs242_lect22_phasenoise.pdf
Oscillator Phase Noise: A Tutorial
Thomas H. Lee, Member, IEEE, and Ali Hajimiri, Member, IEEE
http://smirc.stanford.edu/papers/JSSC00MAR-tom.pdf
Hajamiri
Virtual Damping and Einstein Relation in Oscillators
https://authors.library.caltech.edu/523/1/HAMieeejssc03.pdf
<DBAND2S.PNG><ZERODB.PNG>_______________________________________________
time-nuts mailing list -- time-nuts@lists.febo.com
To unsubscribe send an email to time-nuts-leave@lists.febo.com
Hi,
On 7/11/22 18:05, Mike Monett via time-nuts wrote:
To Bob kb8tq
Now for my new method.
The schematic for a phase-frequency detector is shown in
DBAND2S.PNG. In operation, a pulse arrives at the DATA pin and pin
U1Q goes high. Then a pulse arrives at the VCO pin and pin U2Q goes
high.
This allows the NAND gate to bring the CLR signal low, which
immediately resets both d-flops.
The result is shown in ZERODB.PNG. It is a very narrow pulse with
both d-flops superimposed.
This is the basis for my new approach. Simply tie both inputs of the
PFD together and measure the noise spectrum of the output. (Of
course, you have to ensure that both outputs match at zero error.)
Once you have the PFD noise, you can enable the loop and measure the
total noise spectrum. Then simply subtract the PFD spectrum to get
the OCXO noise. If you have two identical VCXO's, each one
contributes half the noise.
I don't know if this method would work with a double-balanced mixer.
The problem is a DBM requires quadrature signals, so the noise is a
function of the OCXO noise as well as the mixer diodes. But the OCXO
noise is what you are trying to measure.
This method works with the PFD since only a single pulse is needed
to activate both d-flops, so you are measuring only the PFD noise.
A few comments.
First of all, there is no problem with quadrature signal on DBM, since a
standard PI-loop lock drives it into quadrature, so no extra quadrature
hardware needed and all very simple.
Secondly, noise differs very distinctly from systematic signals in how
convergence work. You need to average longer for good confidence bounds
on noise level compared to a systematic. The uncertainty of the noise
vs. frequency will make suggested delta-processing much less rewarding
unless you average over a long time.
Also, as working close-in, flicker noise is likely to be a major issue,
and I doubt the DFF and PFD is optimized for that, rather the opposite.
Using a delta-approach have proven hard to make useful, because
averaging time is what is used to cram the last dB of noise out only
after as good as possible detection has been used.
Cheers,
Magnus