Re: [time-nuts] Effect of EFC noise on phase noise
On Tuesday, August 02, 2016 01:03:58 PM Gerhard
Hoffmann wrote:
Am 02.08.2016 um 09:14 schrieb Bruce Griffiths:
... and a screen dump for the LTspice challenged
;-) Gerhard
Uually a zero is needed in the bias stabilisation loop to
eliminate low frequency gain peaking.
Bruce
Am 02.08.2016 um 22:24 schrieb Bruce Griffiths:
Uually a zero is needed in the bias stabilisation loop to eliminate
low frequency gain peaking.
2 Meg across the 3.3uF remove the peaking. Nevertheless the circuit
still has
problems with its time constants. Weird things happen if you let the
transient analysis
run for a minute simulated time (must increase n cycles for the source).
Probably I'll accept a higher low frequency corner for now to get reasonable
simulation times.
Remember, that is not a publication but a snapshot of work in progress.
Some part values may be just the result of "what happens if..." experiments.
Or that I have a large bag of 3.3u foil capacitors and would like to use
them up first, if possible. :-)
regards, Gerhard
Gerhard wrote:
... and a screen dump for the LTspice challenged
BTW, it can be a lot easier to create multiple, parallel devices in
LTspice by using the "m" command. (Multiple series devices -- passives
and diodes -- can be created with the "n" command.) The schematics look
cleaner, too. (You want to annotate them so everyone knows that the one
symbol represents "m" parallel devices.)
See http://www.linear.com/solutions/6085, "Combining Multiple Model
Instances Into One Symbol."
Best regards,
Charles
One further point regarding noise from the EFC voltage: The varactor in
the oscillator will necessarily have a rather high resistance in series
with it, which adds a certain amount of unavoidable Johnson noise.
Also, the "other end" of the varactor is not generally grounded --
rather, it is connected (through a high-ish resistance) to an internal
reference voltage, which has its own noise.
Taking the HP 10811 as an example, one end of the varactor is connected
to an internal +6.4v reference through 100k ohms, and the other is
connected to the external EFC voltage, also through 100k ohms. The
resistors alone set a noise density floor of about 90nV/sqrtHz at the
oven temperature, while even a "low noise" 6.4V zener diode operated at
1mA has a noise density in the low thousands of nV/sqrtHz (in this case,
filtered by 11k ohms and 6.8uF). The filter has a LP characteristic
with a corner frequency of ~2Hz, but (1) the noise below that frequency
is still a very real concern for phase noise, and (2) it's only a
one-pole filter, so the 10Hz noise is still in the 1000nV/sqrtHz range.
The point of all this? The external EFC voltage doesn't have to be
heroically quiet to remove it as a practical source of phase noise.
That said, be careful about radiated fields and poor PSRR inducing or
coupling voltages onto the EFC line -- such interference can be
thousands of times larger than the Johnson noise, and can cause spurs on
the oscillator output.
Best regards,
Charles
Around 35 years ago, I worked with the guys
who designed and manufactured the 10811. There
are a couple of things here that don't add up,
subject to remembering stuff from a LONG time
ago:
-
Back in those days at least, there were
vendors who supposedly specialized in providing
low noise zener diodes. The particular breakdown
voltage of zener diodes was important. IIRC,
at low voltages, it is a true "zener" diode and
at higher voltage it is merely an avalanche diode.
The physics are somehow different. There is also
a "magic" voltage where the tempco happens to be
+2mV/degree C, in which case you can cancel it
out with a series junction diode. I believe they
even sold combination diodes with both the zener
and the temperature compensation diode in one
package. 6.4V is not far from the magic voltage,
FWIW. Anyway, what I was led to believe is that
certain JEDEC 1N___ part numbers, with suffixes
indicating noise properties, from particular
vendors had much lower than average noise. Thus
if a run of the mill zener diode has 1,000's of
nV/sqrtHz of noise, these "golden" diode might
have only 100's, or even dozens. At one time I
had some copies of some fairly detailed lab notebook
pages detailed research by engineers that I
have a lot of confidence in. -
In all the work with 10811's, E1938A's, and
the 5071A, there was never any indication that
the zener diode and associated resistors, etc
made any contribution whatsoever to phase noise or Allan
deviation. If there was any such effect, I can
guarantee that we would have heard about "hero
experiments" proving it, and we would see
10811's with "EFC-ectomies" used as references
in phase noise and AD test systems.
An indication of the level of scrutiny during
the 10811 design, I submit 2 examples: there
is a 10 Meg resistor across the crystal that
supposedly prevents DC charge from building
up in the crystal due to cosmic ray hits
There was an extensive witch hunt that found
ultimately that lubricating oil in the piston
trimmer migrated around and caused "aging".
I'm not sure what the fix was other than not
lubricating the cap. Maybe a different type
of oil.
Rick
On 8/3/2016 7:11 AM, Charles Steinmetz wrote:
One further point regarding noise from the EFC voltage: The varactor in
the oscillator will necessarily have a rather high resistance in series
with it, which adds a certain amount of unavoidable Johnson noise. Also,
the "other end" of the varactor is not generally grounded -- rather, it
is connected (through a high-ish resistance) to an internal reference
voltage, which has its own noise.
Taking the HP 10811 as an example, one end of the varactor is connected
to an internal +6.4v reference through 100k ohms, and the other is
connected to the external EFC voltage, also through 100k ohms. The
resistors alone set a noise density floor of about 90nV/sqrtHz at the
oven temperature, while even a "low noise" 6.4V zener diode operated at
1mA has a noise density in the low thousands of nV/sqrtHz (in this case,
filtered by 11k ohms and 6.8uF). The filter has a LP characteristic
with a corner frequency of ~2Hz, but (1) the noise below that frequency
is still a very real concern for phase noise, and (2) it's only a
one-pole filter, so the 10Hz noise is still in the 1000nV/sqrtHz range.
The point of all this? The external EFC voltage doesn't have to be
heroically quiet to remove it as a practical source of phase noise.
That said, be careful about radiated fields and poor PSRR inducing or
coupling voltages onto the EFC line -- such interference can be
thousands of times larger than the Johnson noise, and can cause spurs on
the oscillator output.
Best regards,
Charles
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
Rick wrote:
There are a couple of things here that don't add up,
subject to remembering stuff from a LONG time ago
One source of noise I didn't mention is leakage current in the varactor
itself. They are actually pretty noisy devices because of this. It's
entirely possible that the varactor noise swamps all the other sources
of noise. If that is the case, my conclusion (EFC voltage doesn't need
to be heroically clean) is unaffected. Note that it is a known practice
to bias EFC pins to one extreme, to minimize leakage current and
hopefully reduce phase noise.
- Back in those days at least, there were
vendors who supposedly specialized in providing
low noise zener diodes. The particular breakdown
voltage of zener diodes was important. IIRC,
at low voltages, it is a true "zener" diode and
at higher voltage it is merely an avalanche diode.
Traditionally (1950s through mid-80s), the zener effect predominated in
"zener diodes" of less than about 5.5v, and the (noisier) avalanch
effect in "zener diodes" above that voltage. More recently, there has
been some slight juggling of the exact number for some newer series of
diodes.
There is also
a "magic" voltage where the tempco happens to be
+2mV/degree C, in which case you can cancel it
out with a series junction diode. I believe they
even sold combination diodes with both the zener
and the temperature compensation diode in one
package. 6.4V is not far from the magic voltage
It's about 5.5v to 5.6v, so "compensated" zeners (like the 1N82x series)
were about 6.2v.
Anyway, what I was led to believe is that
certain JEDEC 1N___ part numbers, with suffixes
indicating noise properties, from particular
vendors had much lower than average noise. Thus
if a run of the mill zener diode has 1,000's of
nV/sqrtHz of noise, these "golden" diode might
have only 100's, or even dozens.
I don't think the improvement was ever anywhere near two orders of
magnitude. I'm not sure it was typically even one OOM. But in any
case, the noise figure I quoted is the maximum rated noise for one of
those golden parts, not just some random ~6v zener diode.
Best regards,
Charles
Hi Rick:
My first engineering job was working on Tunnel Diode amplifiers at microwave frequencies (when transistors only worked
at audio frequencies). The bias circuit was a 5.1 Volt Zener and a series diode just as your describe for temperature
stability. There was also a BALCO (+ temp coefficient) and VEECO (-temp coefficient) resistor for tweaking the
temperature response. Note that the operating point on a TD is around 1 volt, i.e. between the peak voltage and the
valley voltage. A pot is included to set the stage gain. http://www.prc68.com/I/Aertech.shtml#TDA
http://www.prc68.com/I/Aertech.shtml#TDA
When I worked at HP in the Kobe Instrument Division (component test instruments) they had the 4352 VCO tester. It was
used in conjunction with one of the low phase noise frequency synthesizers that typically has a "two man lift" sticker
and the production line had small cranes to lift them. The programmable DC (EFC) supply was specially designed to have
a very low noise level. It's the same programmable supply that is an option on the 4395A. I sure would like to find
one of those to add to my 4395A.
http://www.prc68.com/I/4395A.shtml
--
Have Fun,
Brooke Clarke
http://www.PRC68.com
http://www.end2partygovernment.com/2012Issues.html
The lesser of evils is still evil.
-------- Original Message --------
Around 35 years ago, I worked with the guys
who designed and manufactured the 10811. There
are a couple of things here that don't add up,
subject to remembering stuff from a LONG time
ago:
-
Back in those days at least, there were
vendors who supposedly specialized in providing
low noise zener diodes. The particular breakdown
voltage of zener diodes was important. IIRC,
at low voltages, it is a true "zener" diode and
at higher voltage it is merely an avalanche diode.
The physics are somehow different. There is also
a "magic" voltage where the tempco happens to be
+2mV/degree C, in which case you can cancel it
out with a series junction diode. I believe they
even sold combination diodes with both the zener
and the temperature compensation diode in one
package. 6.4V is not far from the magic voltage,
FWIW. Anyway, what I was led to believe is that
certain JEDEC 1N___ part numbers, with suffixes
indicating noise properties, from particular
vendors had much lower than average noise. Thus
if a run of the mill zener diode has 1,000's of
nV/sqrtHz of noise, these "golden" diode might
have only 100's, or even dozens. At one time I
had some copies of some fairly detailed lab notebook
pages detailed research by engineers that I
have a lot of confidence in. -
In all the work with 10811's, E1938A's, and
the 5071A, there was never any indication that
the zener diode and associated resistors, etc
made any contribution whatsoever to phase noise or Allan
deviation. If there was any such effect, I can
guarantee that we would have heard about "hero
experiments" proving it, and we would see
10811's with "EFC-ectomies" used as references
in phase noise and AD test systems.
An indication of the level of scrutiny during
the 10811 design, I submit 2 examples: there
is a 10 Meg resistor across the crystal that
supposedly prevents DC charge from building
up in the crystal due to cosmic ray hits
There was an extensive witch hunt that found
ultimately that lubricating oil in the piston
trimmer migrated around and caused "aging".
I'm not sure what the fix was other than not
lubricating the cap. Maybe a different type
of oil.
Rick
On 8/3/2016 7:11 AM, Charles Steinmetz wrote:
One further point regarding noise from the EFC voltage: The varactor in
the oscillator will necessarily have a rather high resistance in series
with it, which adds a certain amount of unavoidable Johnson noise. Also,
the "other end" of the varactor is not generally grounded -- rather, it
is connected (through a high-ish resistance) to an internal reference
voltage, which has its own noise.
Taking the HP 10811 as an example, one end of the varactor is connected
to an internal +6.4v reference through 100k ohms, and the other is
connected to the external EFC voltage, also through 100k ohms. The
resistors alone set a noise density floor of about 90nV/sqrtHz at the
oven temperature, while even a "low noise" 6.4V zener diode operated at
1mA has a noise density in the low thousands of nV/sqrtHz (in this case,
filtered by 11k ohms and 6.8uF). The filter has a LP characteristic
with a corner frequency of ~2Hz, but (1) the noise below that frequency
is still a very real concern for phase noise, and (2) it's only a
one-pole filter, so the 10Hz noise is still in the 1000nV/sqrtHz range.
The point of all this? The external EFC voltage doesn't have to be
heroically quiet to remove it as a practical source of phase noise.
That said, be careful about radiated fields and poor PSRR inducing or
coupling voltages onto the EFC line -- such interference can be
thousands of times larger than the Johnson noise, and can cause spurs on
the oscillator output.
Best regards,
Charles
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to
https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
Am 03.08.2016 um 18:30 schrieb Richard (Rick) Karlquist:
- Back in those days at least, there were
vendors who supposedly specialized in providing
low noise zener diodes. The particular breakdown
voltage of zener diodes was important. IIRC,
at low voltages, it is a true "zener" diode and
at higher voltage it is merely an avalanche diode.
The physics are somehow different. There is also
It is still important.
In these noise density plots that I made in January
you can see the NXP BZX84 family. 0 dB is 1 nV/sqrt hz
so the 2V7 and 3V3 versions are near 2 nV/sqrt Hz. Methinks
that is really good, given that the best low noise op amps
feature a input-referred noise density of 0.9 nV/sqrt Hz.
With rising voltage, 4V7, 6V8.. the diodes get worse and worse
and the LM399 is the complete catastrophe.
Well, in comparison.
The light blue line is an ancient 3.3V glass diode.
BZX84 are cent devices in SOT-23. Note their low 1/f corner.
The lowest dark blue line is the measurement limit.
Bias was 10 NiMH cells via a 1 or 2 KOhm wire resistor.
The bias added no visible noise.
BTW Prof. Zener sued against the manufacturers that they
should not use his name in vain for avalanche diodes
because it was not "his" effect. They settled on the name
Z-Diode for the form of the characteristic curve but in the
long term it was a Pyrrhus victory.
<
https://www.flickr.com/photos/137684711@N07/24411798996/in/album-72157662535945536/
regards,
Gerhard, dk4xp
On Wed, 3 Aug 2016 14:01:30 -0400, you wrote:
Rick wrote:
...
Anyway, what I was led to believe is that
certain JEDEC 1N___ part numbers, with suffixes
indicating noise properties, from particular
vendors had much lower than average noise. Thus
if a run of the mill zener diode has 1,000's of
nV/sqrtHz of noise, these "golden" diode might
have only 100's, or even dozens.
I don't think the improvement was ever anywhere near two orders of
magnitude. I'm not sure it was typically even one OOM. But in any
case, the noise figure I quoted is the maximum rated noise for one of
those golden parts, not just some random ~6v zener diode.
Best regards,
Charles
I wonder if the difference could be design and processing.
Avalanche type transient voltage suppressors and avalanche rated
rectifiers are optimized to have a very uniform junction so that all
areas of the junction break down at the same voltage. Normal
rectifiers have a softer breakdown knee because different areas or
spots break down before others which leads to hot spots and poor power
handling capability.
How much would this affect the noise performance at low voltages where
a reference diode's temperature coefficient is well controlled?
Variations in processing might explain the wide variations in noise
and the golden parts.
Hi
Picking some random 10811-ish numbers:
10 MHz output
5V EFC range
1.6 ppm total EFC range
10 Hz offset from carrier
If you put in 300 nv of noise, in a 1Hz bandwidth, you get around -146 dbc of phase noise. Your
OCXO would be doing very well at 10 MHz to run -135 dbc phase noise at 10 Hz. The same
300 nv RMS gets you -166 at 100 Hz. That’s still pretty well below the expected noise floor. At
1 Hz, you are at -125 and still pretty well below the expected floor.
I don’t think I’m going to win any “low noise” competitions with a voltage reference that has
a noise density of 0.3 uv / Hz. If your OCXO has less total swing, the noise voltage to generate
the stated phase noise goes up….
Bob
On Aug 3, 2016, at 10:11 AM, Charles Steinmetz csteinmetz@yandex.com wrote:
One further point regarding noise from the EFC voltage: The varactor in the oscillator will necessarily have a rather high resistance in series with it, which adds a certain amount of unavoidable Johnson noise. Also, the "other end" of the varactor is not generally grounded -- rather, it is connected (through a high-ish resistance) to an internal reference voltage, which has its own noise.
Taking the HP 10811 as an example, one end of the varactor is connected to an internal +6.4v reference through 100k ohms, and the other is connected to the external EFC voltage, also through 100k ohms. The resistors alone set a noise density floor of about 90nV/sqrtHz at the oven temperature, while even a "low noise" 6.4V zener diode operated at 1mA has a noise density in the low thousands of nV/sqrtHz (in this case, filtered by 11k ohms and 6.8uF). The filter has a LP characteristic with a corner frequency of ~2Hz, but (1) the noise below that frequency is still a very real concern for phase noise, and (2) it's only a one-pole filter, so the 10Hz noise is still in the 1000nV/sqrtHz range.
The point of all this? The external EFC voltage doesn't have to be heroically quiet to remove it as a practical source of phase noise.
That said, be careful about radiated fields and poor PSRR inducing or coupling voltages onto the EFC line -- such interference can be thousands of times larger than the Johnson noise, and can cause spurs on the oscillator output.
Best regards,
Charles
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.