Yep, it supports the big C (padded out with increasingly smaller caps) in
general wins. For two low pass filters, one with say 100nF and one with
10nF, same fc, the 100nF filter will have 10 times less noise power, or
sqrt(10) less rms noise. Near DC is another story.

On Mon, Aug 1, 2016 at 5:10 PM, Bob Camp kb8tq@n1k.org wrote:

Are the EFC inputs all directly DC coupled to the varactor diodes
making them high DC impedance?

I always thought they should bring the varactor or EFC ground out as a
separate pin but I assume that since they do not, ground noise at
least within the oscillator does not limit performance.

In the past when I have had low frequency ground noise, I have either
used an instrumentation amplifier at the load or used a high impedance
current source with a load resistor to ground at the load.  But I have
difficulty imagining either being used to drive an EFC input because
close proximity of the driving circuit allows it to use the oscillator
ground as a single point ground.

I wonder if there is anything to be learned by studying how the old
varactor based (parametric) operational amplifiers were used.

On Mon, 1 Aug 2016 16:32:35 -0400, you wrote:

Hi

Ultimately the EFC signal gets to one or more varicap diodes. It likely goes
through a bias or attenuator network to get there. Playing with the resistors
in the network allows the manufacturer to produce parts with consistent EFC
properties.

The pinout of your standard OCXO and it’s single ground lies almost entirely
with the OEM customers who buy a lot of OCXO’s. To them a minimum number
of pins means a minimum amount of hassle. Given the choice between a multi
ground part and a single pin, the choice will alway go to the single pin. “We just hook them
all to the same point anyway … why the extra pins?”….. Presented wth an app note
and the explanation of why, the answer is still “we don’t have time to design that in”.

Yes it’s a scary world out there.

Bob

"I'm working on a new amplifier based on IF3602 or BF862 FETs that can
use 10u foil only."
Similar to that published by Groner in Linear Audio?
Bruce

On Tuesday, 2 August 2016 12:12 PM, Gerhard Hoffmann <dk4xp@arcor.de> wrote:

Am 01.08.2016 um 22:16 schrieb David:

I beg to differ. Voltage references are not that wonderful. Bandgaps
live from amplifying
small voltage differences and stable Zener references at 6 to 7 Volts
are plagued by
avalanche noise. And that includes the LT6655 band gap.

WRT short term stability all of these are eclipsed by 2.7 / 3.3 volt
zeners and by LEDs.
Even the LT6655 gains a lot of noise performance from an active filter
with AD797 /
ADA4898 op amps and even resistors and 6V/1000uF Nipon Chemi SMD
electrolytics.

I have made some absolute noise voltage measurements:
< https://www.flickr.com/photos/137684711@N07/albums/72157662535945536 >

LEDs and Zeners are measured with bias from a 1 or 2k wire resistor and a
14V NiMH battery. I find the HLMP6000 LED really impressive and the
LT3042 regulator.

The preamp is 20 ADA4898 op amps in parallel ( i.e. 220pV/sqrtHz), the 0
dB line is 1 nV/sqrt Hz.
Everything was fed from batteries in a box in box in a box and then
after +80 dB passed to
an 89441A vector signal analyzer.

The input capacitor of the preamp is 20 times 10uF WIMA foil, that is
not enough for the
low frequency corner as we do not see the real 1/f noise below 20 Hz.
What we see looks more like GR noise, spectrum-wise, and it is really
the insufficient shorting
of the 10K bias resistor through the input source and coupling cap.

I have bought some wet slug tantals as proposed by Jim Williams (see
below), 10000 uF bring
the right 1/f behaviour but at very substantial cost :-(    At least for
small input voltages alu
electrolytics do not seem to make a difference. I did not test large
voltages.

I'm working on a new amplifier based on IF3602 or BF862 FETs that can
use 10u foil only.
When it's done I'll repeat these measurements.

JW has the added handicap that he wants to keep the the long term and
absolute stability
of his reference and so cannot afford any voltage drop on a series R. We
do not share that
problem on an EFC line because the C stands for control and if the
voltage there does
never change for some other reason we have probably made a bad decision
with regard
to loop gain.

And large resistors may feature more noise voltage, but that increases
only with the
root of the resistance. The filter corner drops in a linear way, so a
large resistor may
really help. The tiny noise voltage of a reasonable resistor must be
seen anyway in
the context of say, a 10811A that tunes +- 1 Hz for 10 Volts on the EFC.

regards, Gerhard

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 02.08.2016 um 02:42 schrieb Bruce Griffiths:

I know that Groner exists from some web site, but had no personal contact.
Also I don't read Linear Audio other than 2 articles from Scott Wurcer
that I've bought.
I highly regard Scott, he's the father of the AD797 after all and AD fellow.
I also had some conversations with him on that.

The preamp will be classical.  Some JFETs in parallel, no source
resistors except
half an Ohm for feedback, more would add to the input noise voltage.
Cascode with a Zetex bipolar (or whatever they are called now).
OP37 for loop gain. feedback from OP37 output for 50 or 60 dB gain.
Post amplifier to 80 db or so.
Without the cascode, the 1 MHz is not possible. It does not help that
the feedback limits the voltage excursions on the drain.

I'm not yet sure about the effective input capacitance. I get abt. 1 or
1.5 MHz
bandwidth from a low impedance source. A few nF on the input capacitance
would be ok, in the end I want it after a ring mixer for phase noise
measurements
but I get unreasonably more in simulation, depending on if I measure it from
upper frequency corner with a larger input resistor or the resonance
frequency
with an added inductor.

You can get the LTspice file if you like.

regards, Gerhard

Ja Gerhard, bitte schicke mir den LTSpice file,

Danke im Voraus und
73
KJ6UHN
Alex

On 8/1/2016 6:34 PM, Gerhard Hoffmann wrote:

Gerhard, please send me the LTspice files,  so I can compare it with  some of my circuits.
The Groner preamp uses feedback to adjust the dc input at the input (paralleled) FET gate(s) to adjust the dc output to zero.  It uses a small value input coupling capactor with a high value resistor to bias the JFET (parallled BF862's). I had simulated a similar preamp architecture some years before the Groner article.
I suggested a similar preamp architecture on EEVBlog  but the responses indicated that most of the respondents appear incapable of rational analysis and just regurgitate the conventional approach.

Bruce

On Tuesday, 2 August 2016 6:19 PM, Alex Pummer <alex@pcscons.com> wrote:

Ja Gerhard, bitte schicke mir den LTSpice file,

Danke im Voraus und
73
KJ6UHN
Alex

On 8/1/2016 6:34 PM, Gerhard Hoffmann wrote:

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.

Gerhard wrote:

As you recognize, these FETs need significant drain-source voltage to
give their promised performance (8v at the absolute minimum, IMO, and
they really should have 10v).  You might try a folded cascode topology
(sometimes called a "reverse cascode") to increase the D-S voltage.  It
is a little more complicated, but should not have any material impact on
the noise figure if it is done properly.

Best regards,

Charles

Am 02.08.2016 um 09:14 schrieb Bruce Griffiths:

Some remarks:

1. don't get confused from my comments / aide-memoires in the source.

2. The transient simulation will wake up perfectly biased at t=0. If you
   want to
   see the bias integrator working you must use the .ic statement to set the
   integrator capacitor voltage to a start value, typically 0V. We have
   discussed
   that already in the transformer inrush current thread.

3. The IF3602 is now available from Mouser.
   <
   http://www.mouser.de/ProductDetail/InterFET/IF3602/?qs=%2fha2pyFaduiJyeWAJl%2f%252bHc05SUFyPikuEx3OzRizeOU%3d

It does not look like the symbol picture there. The price is somewhat
obscene. Paralleling a lot of them
has some aspects of hybris, but at least the simulator does not care.
Also the input capacitance is huge.
2 dozens of BF862 may be much cheaper and have less capacitance.
The Interfet data sheet is a joke. I did ask for something better but
got no answer.

4. I have fiddled with the BF862 model to make it fit to the XSpice
   included in my Altium Designer.
   You may use the one from NXP directly.

5. The noise of a JFET depends mostly on its capacitance and gm. More Gm
   is better, it is best
   at maximum current. That would be Idss. Idss varies all over the place
   with JFETs and and with the IF3602
   in special. 1:20 has been observed. In an ideal world, gm varies with
   the square root of the current, and
   the input referred noise voltage varies with the square root of gm, so
   the noise gets better with the 4th
   root of the current. This is not a strong dependence and a diminuishing
   ROI. More transistors at the same
   current are better than one, but the input cap is larger. One gets the
   full advantage of more transistors
   only with n times the current.

6. Given the price of the transistors I won't skimp on a heat sink and
   let them sweat a little bit.

7. Using IF3602 instead of IF3601 does not mean that you get matched
   pairs in the sense most people
   are used to. But probably you won't get opposite outliers. The pairs
   cost more than 2 singles, but
   are probably more consistent. Because of the 4th root it does not pay to
   regulate the bias of the
   transistors individually unless you have a truly pathological case.
   Singles are easier to cool.

The measured Id vs. Vgs of my first 4 samples is in fet_preamp_notes.pdf.
Somebody has played with my data and extrapolated Idss. I wonder how he
got the table out of the pdf,
enriched it and put it back in again. Some people are real hackers. :-)

8. Don't use more than 3 or 4 Volts of Vdd. It only adds heat and impact
   ionization in the channel.
   That results in some gate current that hurts you with a 100 Meg bias
   resistor, noise-wise.
   Avoiding gate current is even more important since it rises steeply with
   temperature.

A Cascode is probably a must, alone b/c of the Miller. The feedback to
the sources reduces the
swing at the drain, but with the cascode it works better.

9. The amplifier is AC coupled, but you can easily go down to 100 mHz.
   That won't impair your
   Leeson plots close to the carrier. Take the dc for quadrature control
   from a different cheap
   fet op amp. Your mixers may thank you with a longer life.

10. The version with the 20 paralleled op amps is at
    < http://www.hoffmann-hochfrequenz.de/downloads/lono.pdf  >
    In the same directory there is a .zip file with Gerber files and updated
    schematics.

11. some pictures more at
    <
    https://picasaweb.google.com/103357048842463945642/LowNoisePreamplifiers?authuser=0&feat=directlink

12. I've got first silicon, ehem, FR-4 of my OCXO carrier board and also
    20 MTI-260 5 MHz
    oscillators, still packaged in 20 gps-less Lucent units. While using
    multiple op amps
    works so nicely, 16 sloooowly phaselocked MTI-260 that are wilkinson-ed
    together should
    result in a pretty phase noise plot. Especially now that BVAs are no
    longer easily available.
    Also, 1 of 16 cannot enforce a sudden phase jump; it would be slowly
    corrected back
    by its PLL.

If you need Lucent spare parts other than MTI-260s, ask me.

regards, Gerhard  DK4XP