Empathy List Archives

BG

Bruce Griffiths

Sat, Oct 24, 2015 12:11 AM

On Friday, October 23, 2015 03:27:39 PM Charles Steinmetz wrote:

Gerhard wrote:

What do you consider a run-of-the-mill comparator? LM139, LMV7219,
AD8561, ADCMP580?

For squaring 1-10MHz sine waves, the LT1719 and 1720 are the best
that I've found. This is a matter of how much internal hysteresis
the comparator has, how smoothly the internal hysteresis acts, how
much gain the part has and how it is structured, how fast it is (both
risetime and propagation delay), whether it suffers from thermal
feedback from the output stage to the input stage, how much internal
ground bounce it has, and a host of other die-level issues.

The LT1719 uses bipolar input supplies, so ground can be the
reference voltage and also be in the center of the input common-mode
range. This is always quieter than biasing the inputs to the middle
of a single supply, which is usually done with the 1720 and many
other comparators. (Like other single-supply comparators, the 1720
will work referenced to ground with only a positive supply -- but (i)
the inputs are then at the very edge of the input common-mode range,
and (ii) you can only drive the input 100mV below ground, so you have
to figure out how to clamp the input signal. It is much easier to
just use an LT1719 with +/-5v on the input stage, and it works better,

too.)

Page 22 of the LT1719 datasheet shows the simplest possible circuit
and discusses its performance.

What optimizations? I have seen the "Wenzel" circuit in cheapish
frequency counter inputs in
the late seventies, maybe with a diode bridge added as input

protection..

Use medium-speed transistors, bias both bases from the same low-noise
voltage reference such as an LM329, capacitively couple the emitters,
and use a higher supply voltage, for starters. I use MPSH81/MMBTH81s
and a power supply of around 20v (see attached schematic) for a
reasonably optimized implementation. Other transistors can be used,
but I've found that the H81s work better for squaring 1-10MHz sine
waves than anything else I've tried -- they hit the sweet spot of the
bandwidth/gain tradeoff and have a nice flat gain vs. current
characteristic.

so it shows that one can replace filtering by signal power :-)

It's a matter of the slew rate of the input sine wave at
zero-cross. Once you reach the critical slew rate for any particular
input architecture, the comparator is hard-switched fast enough that
it doesn't spend significant time in the linear region generating noise.

Any objections against the AD9901 phase comparator?

That should work fine.

Best regards,

Charles

A resistor from point A to ground in the Wenzel style shaper you attached
has little effect on the output symmetry due to C4. However it does allow
the output amplitude to be adjusted.

Bruce

On Friday, October 23, 2015 03:27:39 PM Charles Steinmetz wrote: > Gerhard wrote: > >What do you consider a run-of-the-mill comparator? LM139, LMV7219, > >AD8561, ADCMP580? > > For squaring 1-10MHz sine waves, the LT1719 and 1720 are the best > that I've found. This is a matter of how much internal hysteresis > the comparator has, how smoothly the internal hysteresis acts, how > much gain the part has and how it is structured, how fast it is (both > risetime and propagation delay), whether it suffers from thermal > feedback from the output stage to the input stage, how much internal > ground bounce it has, and a host of other die-level issues. > > The LT1719 uses bipolar input supplies, so ground can be the > reference voltage and also be in the center of the input common-mode > range. This is always quieter than biasing the inputs to the middle > of a single supply, which is usually done with the 1720 and many > other comparators. (Like other single-supply comparators, the 1720 > will work referenced to ground with only a positive supply -- but (i) > the inputs are then at the very edge of the input common-mode range, > and (ii) you can only drive the input 100mV below ground, so you have > to figure out how to clamp the input signal. It is much easier to > just use an LT1719 with +/-5v on the input stage, and it works better, too.) > > Page 22 of the LT1719 datasheet shows the simplest possible circuit > and discusses its performance. > > >What optimizations? I have seen the "Wenzel" circuit in cheapish > >frequency counter inputs in > >the late seventies, maybe with a diode bridge added as input protection.. > > Use medium-speed transistors, bias both bases from the same low-noise > voltage reference such as an LM329, capacitively couple the emitters, > and use a higher supply voltage, for starters. I use MPSH81/MMBTH81s > and a power supply of around 20v (see attached schematic) for a > reasonably optimized implementation. Other transistors can be used, > but I've found that the H81s work better for squaring 1-10MHz sine > waves than anything else I've tried -- they hit the sweet spot of the > bandwidth/gain tradeoff and have a nice flat gain vs. current > characteristic. > >so it shows that one can replace filtering by signal power :-) > > It's a matter of the slew rate of the input sine wave at > zero-cross. Once you reach the critical slew rate for any particular > input architecture, the comparator is hard-switched fast enough that > it doesn't spend significant time in the linear region generating noise. > > >Any objections against the AD9901 phase comparator? > > That should work fine. > > Best regards, > > Charles Charles A resistor from point A to ground in the Wenzel style shaper you attached has little effect on the output symmetry due to C4. However it does allow the output amplitude to be adjusted. Bruce

CS

Charles Steinmetz

Sat, Oct 24, 2015 1:38 PM

The schematic I posted yesterday of the optimized Wenzel-style
squarer did not specify capacitor types for C1, C3, and C4. All
three should be C0G/NP0, and C4 must be C0G/NP0. The attached
schematic is revised accordingly.

Best regards,

Charles

The schematic I posted yesterday of the optimized Wenzel-style squarer did not specify capacitor types for C1, C3, and C4. All three should be C0G/NP0, and C4 *must* be C0G/NP0. The attached schematic is revised accordingly. Best regards, Charles

GH

Gerhard Hoffmann

Sun, Oct 25, 2015 11:43 PM

Am 23.10.2015 um 21:27 schrieb Charles Steinmetz:

Use medium-speed transistors, bias both bases from the same low-noise
voltage reference such as an LM329, capacitively couple the emitters,
and use a higher supply voltage, for starters. I use

This evening, I have measured some oh so noisy zeners and was pleasantly
surprised.
Other than one 40+ year old glass diode all of them were pretty low noise.

Since the LM329 was mentioned above, I tried it also. Another surprise.
HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER???

Bias for the DUTs was a 1K wire resistor from 10 NiMH cells. The bias
source does not
add any visible noise. The "shorted" trace shows the noise floor of the
system. That is
about 220 pV/sqrt Hz. FFT analyzer is an Agilent 89441A. It has strong
1/f noise that is
barely hidden by the preamp. 0 dB = 1nV/sqrt Hz

MPSH81/MMBTH81s and a power supply of around 20v (see attached
schematic) for a reasonably
optimized implementation. Other transistors can be used, but I've
found that the H81s work better for squaring 1-10MHz sine waves than
anything else I've tried -- they hit the sweet spot of the
bandwidth/gain tradeoff and have a nice flat gain vs. current
characteristic.

Are there differences between Fairchild and Motorola?

regards, Gerhard

Am 23.10.2015 um 21:27 schrieb Charles Steinmetz: > > Use medium-speed transistors, bias both bases from the same low-noise > voltage reference such as an LM329, capacitively couple the emitters, > and use a higher supply voltage, for starters. I use This evening, I have measured some oh so noisy zeners and was pleasantly surprised. Other than one 40+ year old glass diode all of them were pretty low noise. Since the LM329 was mentioned above, I tried it also. Another surprise. HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER??? Bias for the DUTs was a 1K wire resistor from 10 NiMH cells. The bias source does not add any visible noise. The "shorted" trace shows the noise floor of the system. That is about 220 pV/sqrt Hz. FFT analyzer is an Agilent 89441A. It has strong 1/f noise that is barely hidden by the preamp. 0 dB = 1nV/sqrt Hz > MPSH81/MMBTH81s and a power supply of around 20v (see attached > schematic) for a reasonably > optimized implementation. Other transistors can be used, but I've > found that the H81s work better for squaring 1-10MHz sine waves than > anything else I've tried -- they hit the sweet spot of the > bandwidth/gain tradeoff and have a nice flat gain vs. current > characteristic. Are there differences between Fairchild and Motorola? regards, Gerhard

CS

Charles Steinmetz

Tue, Oct 27, 2015 12:50 PM

Gerhard wrote:

Since the LM329 was mentioned above, I tried it also. Another surprise.
HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER???

0 dB = 1nV/sqrt Hz

The datasheet for both NS and LT LM329s shows broadband noise of
<70nV/sqrtHz, with a corner at ~100Hz, rising to a touch over
100nV/sqrtHz at 10Hz. That is consistent with the LM329s I have used.

So, according to the datasheet, the 10Hz noise ought to be a touch
above 40dB on your graph and the curve should level off above 100Hz
at about 30dB -- if I have done my dBs right. Still somewhat higher
than most of the other diodes you tested, but not outlandishly so.

Your data seems to be much worse than this, with a 1/f noise corner
at ~10kHz (!!), rising to what appears to be about 1mV/sqrtHz (!!) at
10Hz. Is it possible you got fakes? (I know someone who bought a
quantity of LTZ1000s that turned out to be fakes. Not just abused
parts dismounted with a blowtorch and with new leads welded on --
actual, never-were-real-LTZ1000s. And they came from a second-tier
distributor, not an ebay scammer, although further investigation
revealed that the distributor had been dropped by LT some time
prior.) Or is the one you tried just broken? In any case, the
results you obtained are several orders of magnitude worse than the
specs for the part.

Are there differences between Fairchild and Motorola? [MPSH81/MMBTH81]

Probably, although I don't know if any differences there are would
have a noticeable effect on the performance of the circuit. My
current stock of H81s is all Fairchild. I know I have used Motorolas
in the past, and never had any reason to question whether there was a
significant difference -- but I never compared them side-by-side,
either. But I don't think that Motorola/ONsemi has made them for a
long time, anyway. Indeed, I see now that even the Fairchild
through-hole part (MPSH81) seems to be gone. The SMD part (MMBTH81)
seems to be available still.

Best regards,

Charles

Gerhard wrote: >Since the LM329 was mentioned above, I tried it also. Another surprise. >HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER??? > >0 dB = 1nV/sqrt Hz The datasheet for both NS and LT LM329s shows broadband noise of <70nV/sqrtHz, with a corner at ~100Hz, rising to a touch over 100nV/sqrtHz at 10Hz. That is consistent with the LM329s I have used. So, according to the datasheet, the 10Hz noise ought to be a touch above 40dB on your graph and the curve should level off above 100Hz at about 30dB -- if I have done my dBs right. Still somewhat higher than most of the other diodes you tested, but not outlandishly so. Your data seems to be *much* worse than this, with a 1/f noise corner at ~10kHz (!!), rising to what appears to be about 1mV/sqrtHz (!!) at 10Hz. Is it possible you got fakes? (I know someone who bought a quantity of LTZ1000s that turned out to be fakes. Not just abused parts dismounted with a blowtorch and with new leads welded on -- actual, never-were-real-LTZ1000s. And they came from a second-tier distributor, not an ebay scammer, although further investigation revealed that the distributor had been dropped by LT some time prior.) Or is the one you tried just broken? In any case, the results you obtained are several orders of magnitude worse than the specs for the part. >Are there differences between Fairchild and Motorola? [MPSH81/MMBTH81] Probably, although I don't know if any differences there are would have a noticeable effect on the performance of the circuit. My current stock of H81s is all Fairchild. I know I have used Motorolas in the past, and never had any reason to question whether there was a significant difference -- but I never compared them side-by-side, either. But I don't think that Motorola/ONsemi has made them for a long time, anyway. Indeed, I see now that even the Fairchild through-hole part (MPSH81) seems to be gone. The SMD part (MMBTH81) seems to be available still. Best regards, Charles

GH

Gerhard Hoffmann

Tue, Oct 27, 2015 5:03 PM

Am 27.10.2015 um 13:50 schrieb Charles Steinmetz:

Gerhard wrote:

Since the LM329 was mentioned above, I tried it also. Another surprise.
HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER???

0 dB = 1nV/sqrt Hz

The datasheet for both NS and LT LM329s shows broadband noise of
<70nV/sqrtHz, with a corner at ~100Hz, rising to a touch over
100nV/sqrtHz at 10Hz. That is consistent with the LM329s I have used.
So, according to the datasheet, the 10Hz noise ought to be a touch
above 40dB on your graph and the curve should level off above 100Hz at
about 30dB -- if I have done my dBs right. Still somewhat higher than
most of the other diodes you tested, but not outlandishly so.

Your data seems to be much worse than this, with a 1/f noise corner
at ~10kHz (!!), rising to what appears to be about 1mV/sqrtHz (!!) at
10Hz. Is it possible you got fakes? (I know someone who

It's bad, but not that bad. 10 times the voltage is 20 dB more. So

0dB = 1nV/sqrt Hz
20 dB = 10 nV
40 dB = 100 nV
60 dB = 1uV
80 dB = 10 uV/sqrt Hz

The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus the
claimed 75 nV typical
but the corner frequency is really bad. I bought them last year from
Digi-key, still with
National as manufacturer on the bag, not TI.

I'll test a few more next weekend.

regards, Gerhard

Am 27.10.2015 um 13:50 schrieb Charles Steinmetz: > Gerhard wrote: > >> Since the LM329 was mentioned above, I tried it also. Another surprise. >> HOW CAN THEY MAKE SUCH A NOISY SUBSURFACE ZENER??? >> >> 0 dB = 1nV/sqrt Hz > > The datasheet for both NS and LT LM329s shows broadband noise of > <70nV/sqrtHz, with a corner at ~100Hz, rising to a touch over > 100nV/sqrtHz at 10Hz. That is consistent with the LM329s I have used. > So, according to the datasheet, the 10Hz noise ought to be a touch > above 40dB on your graph and the curve should level off above 100Hz at > about 30dB -- if I have done my dBs right. Still somewhat higher than > most of the other diodes you tested, but not outlandishly so. > > Your data seems to be *much* worse than this, with a 1/f noise corner > at ~10kHz (!!), rising to what appears to be about 1mV/sqrtHz (!!) at > 10Hz. Is it possible you got fakes? (I know someone who It's bad, but not that bad. 10 times the voltage is 20 dB more. So 0dB = 1nV/sqrt Hz 20 dB = 10 nV 40 dB = 100 nV 60 dB = 1uV 80 dB = 10 uV/sqrt Hz The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus the claimed 75 nV typical but the corner frequency is really bad. I bought them last year from Digi-key, still with National as manufacturer on the bag, not TI. I'll test a few more next weekend. regards, Gerhard

CS

Charles Steinmetz

Wed, Oct 28, 2015 3:02 AM

Gerhard wrote:

It's bad, but not that bad. 10 times the voltage is 20 dB more. So

0dB = 1nV/sqrt Hz
20 dB = 10 nV
40 dB = 100 nV
60 dB = 1uV
80 dB = 10 uV/sqrt Hz

So I got ~40dB and ~30dB right, but screwed up and turned 10uV into
1mV. How'd I do that?

The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus
the claimed 75 nV typical
but the corner frequency is really bad. I bought them last year from
Digi-key, still with
National as manufacturer on the bag, not TI.

I'll test a few more next weekend.

Even Digi-Key is not immune from counterfeits. I received some
counterfeit voltage references last year, but before I used any I got
a message recalling them.

Thinking on it a bit more, your results look like what one might
expect if someone put up bandgap reference dice in fake LM329
packages. On the other hand, they could be from a lot of real NS
parts that had serious fab issues or catastrophic storage issues.

In any case, if the others you have are like that one, throw them out
and get some new ones that meet spec (and just in case, buy from
another vendor, and perhaps another manufacturer). You can buy them
direct from Linear Technology http://www.linear.com/purchase/LM329,
but only in the plastic TO-92 package (all the 329s I've used have
been in the hermetic package, but those appear to be history now).

Best regards,

Charles

Gerhard wrote: >It's bad, but not that bad. 10 times the voltage is 20 dB more. So > >0dB = 1nV/sqrt Hz >20 dB = 10 nV >40 dB = 100 nV >60 dB = 1uV >80 dB = 10 uV/sqrt Hz So I got ~40dB and ~30dB right, but screwed up and turned 10uV into 1mV. How'd I do that? >The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus >the claimed 75 nV typical >but the corner frequency is really bad. I bought them last year from >Digi-key, still with >National as manufacturer on the bag, not TI. > >I'll test a few more next weekend. Even Digi-Key is not immune from counterfeits. I received some counterfeit voltage references last year, but before I used any I got a message recalling them. Thinking on it a bit more, your results look like what one might expect if someone put up bandgap reference dice in fake LM329 packages. On the other hand, they could be from a lot of real NS parts that had serious fab issues or catastrophic storage issues. In any case, if the others you have are like that one, throw them out and get some new ones that meet spec (and just in case, buy from another vendor, and perhaps another manufacturer). You can buy them direct from Linear Technology <http://www.linear.com/purchase/LM329>, but only in the plastic TO-92 package (all the 329s I've used have been in the hermetic package, but those appear to be history now). Best regards, Charles

BG

Bruce Griffiths

Wed, Oct 28, 2015 7:10 AM

Charles
I have a bunch of LM329's purchased from a large distributor.I guess I should make some noise measurements.Some have even been installed in low noise regulators.
Another item to add to the to do list( repair mill, complete dual fibre point diffraction interferometer, rebuild RF PN interferometer.... ).Probably a good excuse to build that low noise JFET input preamp with the somewhat unusual input coupling architecture.
Bruce

Bruce

 On Wednesday, 28 October 2015 6:00 PM, Charles Steinmetz <csteinmetz@yandex.com> wrote:

Gerhard wrote:

It's bad, but not that bad. 10 times the voltage is 20 dB more. So

0dB = 1nV/sqrt Hz
20 dB = 10 nV
40 dB = 100 nV
60 dB = 1uV
80 dB = 10 uV/sqrt Hz

So I got ~40dB and ~30dB right, but screwed up and turned 10uV into
1mV. How'd I do that?

The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus
the claimed 75 nV typical
but the corner frequency is really bad. I bought them last year from
Digi-key, still with
National as manufacturer on the bag, not TI.

I'll test a few more next weekend.

Even Digi-Key is not immune from counterfeits. I received some
counterfeit voltage references last year, but before I used any I got
a message recalling them.

Thinking on it a bit more, your results look like what one might
expect if someone put up bandgap reference dice in fake LM329
packages. On the other hand, they could be from a lot of real NS
parts that had serious fab issues or catastrophic storage issues.

In any case, if the others you have are like that one, throw them out
and get some new ones that meet spec (and just in case, buy from
another vendor, and perhaps another manufacturer). You can buy them
direct from Linear Technology http://www.linear.com/purchase/LM329,
but only in the plastic TO-92 package (all the 329s I've used have
been in the hermetic package, but those appear to be history now).

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.

Charles I have a bunch of LM329's purchased from a large distributor.I guess I should make some noise measurements.Some have even been installed in low noise regulators. Another item to add to the to do list( repair mill, complete dual fibre point diffraction interferometer, rebuild RF PN interferometer.... ).Probably a good excuse to build that low noise JFET input preamp with the somewhat unusual input coupling architecture. Bruce Bruce On Wednesday, 28 October 2015 6:00 PM, Charles Steinmetz <csteinmetz@yandex.com> wrote: Gerhard wrote: >It's bad, but not that bad. 10 times the voltage is 20 dB more. So > >0dB = 1nV/sqrt Hz >20 dB = 10 nV >40 dB = 100 nV >60 dB = 1uV >80 dB = 10 uV/sqrt Hz So I got ~40dB and ~30dB right, but screwed up and turned 10uV into 1mV. How'd I do that? >The wideband noise at, say, 46 dB would be 200 nV/sqrt Hz versus >the claimed 75 nV typical >but the corner frequency is really bad. I bought them last year from >Digi-key, still with >National as manufacturer on the bag, not TI. > >I'll test a few more next weekend. Even Digi-Key is not immune from counterfeits. I received some counterfeit voltage references last year, but before I used any I got a message recalling them. Thinking on it a bit more, your results look like what one might expect if someone put up bandgap reference dice in fake LM329 packages. On the other hand, they could be from a lot of real NS parts that had serious fab issues or catastrophic storage issues. In any case, if the others you have are like that one, throw them out and get some new ones that meet spec (and just in case, buy from another vendor, and perhaps another manufacturer). You can buy them direct from Linear Technology <http://www.linear.com/purchase/LM329>, but only in the plastic TO-92 package (all the 329s I've used have been in the hermetic package, but those appear to be history now). 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@lists.febo.com

Unified VCXO Carrier Board