Gerhard Hoffmann wrote:

Gerhard

RF grounding all three electrodes of the BC860 is not good practice.
The BC860 will tend to oscillate when you do this unless the ESR of the
100uF base bypass cap is large enough.
Some early microprocessor reset circuits buffered the RC reset timing
circuit with an emitter follower which burst into RF oscillation as the
RC network charged up.
The cure is simple omit the 100uF base bypass cap or at least leave an
unbypassed resistor of sufficient value in series with the BC860 base.

You can improve the phase noise somewhat if the time constants
associated with the BC860 base and collector bypass circuits are reduced
so that the BC860 reduces the collector current noise of the BFG196 in
the dc - 100kHz or so spectral region.

The major problem with this circuit is the large dc current flowing in
the transformer primary increases the output distortion significantly.
This may make it difficult to extend the frequency response down to 5MHz
without using a large custom wound transformer.

Minicircuits and similar transformers tend to have a coupling
coefficient much greater than 0.99.

The purpose of the heavy bypassing of the bases of the BFG31 transistors
is to reduce the low frequency noise at the BFG31 bases, this reduces
the amplifier close in phase noise.
It is also necessary to use some active filtering of the power supply if
one is to achieve low close in phase noise.

To maximise reverse isolation the individual amplifiers will need to be
enclosed in RF shields.
Have you measured the reverse isolation?

For maximum phase stability the BNC connectors should replaced by
threaded connectors such as TNC, SMA , N etc.

If you are using thick film resistors replace them with thin film
resistors if you want low close in phase noise.
Have you measured the phase noise?

The output device will saturate or at least have increased distortion if
the output isnt terminated in 50 ohms whilst the input is driven at +13dBm.

Have you measured the input and output VSWR or reflection coefficients?
With real transformers the value of the 200 ohm resistor may need to be
adjusted to minimise the output reflection coefficient.

Bruce

Bruce Griffiths wrote:

Gerhard

Your layout could be a little cleaner.
Use a 47 ohm resistor in series with each of the emitter of Q4 and use an 82 ohm resistor in series with each emitter connection of Q3.

More space between Q2, Q3, Q4 would be conducive to higher reverse isolation.
The bypass capacitor pads need to have lower inductance connections to the ground plane (several vias in parallel) and the base.

A linear RF path would also be useful, the connection from Q2 emitter to Q3 emitter can be a 42 ohm section of microstrip and the connection from Q3 collector to Q4 emitter can be a section of 23 ohm microstrip.

Instead of cramming 6 or 7 amplifiers onto a relatively small board, use a larger board and spread them out more leaving space for RF shielding to improve reverse isolation.

For 5MHz and 10MHz the other NIST design has significantly higher reverse isolation which can be improved somewhat along with reduced distortion by using a 2 transistor input stage. No dc flows in the input and output transformers reducing distortion. It is also possible to design the output stage so that it wont saturate when driving a high impedance load with a + 13dBm input.

Bruce

Gerhard Hoffmann wrote:

Gerhard

The base bias divider network needs to be improved to ensure that the
reverse isolation isn't compromised by the divider network:
Spit R2, R7, R9 into 2 equal (330 ohm) resistors connected in series and
bypass all the resistor junctions with 100nF X7R or NP0 caps.
Bypass the bases of Q3 and Q4 with large value tantalum electrolytics
(or plastic capacitors) as well.

Bruce

On Mon, 22 Sep 2008 15:52:19 +1200, you wrote:

A standard electrolytic? The ESR is less than stellar, see my musings in

http://www.hoffmann-hochfrequenz.de/downloads/experiments_with_decoupling_capacitors.pdf

and BC transistors are AF types.
OK, getting ft is like drinking from a hydrant in Si planar.

The question is more if the 100r * 100u do anything to de-noise
the LED voltage. Close to the carrier probably not. The problem is
that you nowhere get exact numbers to base a decision on.

Yes, but don't tell anybody about that, because Rohde claims he has
a patent for that IIRC (forgetting a capacitor :-) .

So be it custom. Anything less than 300 uH in par will ruin the flat frequency
response, and the parasitics of a large L + coupling cap will do the same :-(
And with the flat response, the delay stability will be gone, too.

This was an area of doubt for me, upto now.
Experimenting with LT-Spice brought some insights.

(btw available for free from http://www.linear.com/designtools/software/switchercad.jsp
Highly_recommended! Even in the sense of free beer, not free/open source.)

Noise gain of a voltage fed to the base of common base stage is proportional
to Zc/Ze. The Impedance at the emitter of Q3 is quite low, so the
noise voltage at the base reference divider is propagated to the output
just like an input voltage to the BFG196.  --> AF decoupling needed for Q3
(and the space is reserved on the board)
www.hoffmann-hochfrequenz.de/downloads/noise_base_Q3.gif
Generator input is 0 dB, solid lines = amplitude, dotted lines = phase.
The AF noise itself may not make it to the output, but it may be
modulated onto the carrier.

The second cascode stage, Q4, is fed from the high source impedance of the
Q3 common base stage. The emitter current is forced into Q4 without
much influence of the Q4 base voltage. The noise voltage at the base
from the resistive divider is pretty much suppressed, so heavy
decoupling of Q4.base should not pay.
www.hoffmann-hochfrequenz.de/downloads/noise_base_Q4.gif

Thinking about it, in this circuit only the second CB stage delivers
"full" isolation.

Yes, but that's a different board. I have built some regulators in the
style of Walt Jung of Analog Devices, and they look quite promising.
(clever idea, feeding the reference from the regulated output)
They feature 2 or 3 nV / sqrt Hz.

http://waltjung.org/PDFs/Regulator_Excels_In_Noise_and_Line_Rejection.pdf

www.hoffmann-hochfrequenz.de/downloads/IMG_0341__jung_pos_ref_1024_q75.jpg
(voltage reference and positive regulator)

The space and the holes are there already, but I still need the access.

not yet.

The next iteration will have to live with BNC because R&S smpd, hp8662A,
SNA-33 & friends all have BNCs and the holes in the 19" front plate of
my GPS disciplined XO are drilled already.
OK, we'll get optional SMAs, too. (the singe channel proto has them already.)

Susumu NiCr 0.1% from Digikey, next delivery..

Not yet. Still bootstrapping. If it's worth measuring, it
cannot be done with the spectrum analyzer.
I've got an offer from a 3rd party to have it measured
and I'll accept that :-)

output return loss is in the .pdf
35 MHz must happen to be the sweet spot in the eyes of the ZRB2 bridge.

Input is RC only, should be trimmable. The next version will be changed to
have equal input delays for all channels, so measuring input RL now
would be a waste of time.

thanks for the proposals!

regards, Gerhard, dk4xp

(The week doesn't have enough end. And it's half past 2, again. Good Night!)

Gerhard Hoffmann wrote:

There usually not too much point in low pass filtering the LED voltage
as the noise of a forward biased LED is very low unless of course the
LED has poor reliability (high LED voltage noise is a very good
predictor of poor reliability).
Of course the finite slope resistance of the LED will allow some of the
power supply noise to appear across it.

Its a bit more than omitting a base bypass cap as it has been
recommended for decades (well before Rhode came up with his not too
original idea - this was obvious once NIST demonstrated that extending
the bandwidth of a current regulation loop helped suppress low offset
frequency phase noise) that a base bypass cap for the current regulator
transistor not be used to avoid oscillation.

You are likely to find that its only practical to cover the 80-120Mhz
region as the NIST and Spectradynamics distribution amplifiers do.

You need to read the phase noise references listed at the bottom of:
http://www.ko4bb.com/~bruce/IsolationAmplifiers.html
http://www.ko4bb.com/%7Ebruce/IsolationAmplifiers.html
Essentially noise voltages between collector and base modulates the
collector output capacitance and hence the modulates the output signal
phase shift.
It can also modulate the emitter current and hence the transistor ft
which also phase modulates the output signal.
Other transistor parameters can also be modulated by noise voltages from
the power supply, resistors etc.

It has some severe limitations for most of the simulations I have done.

Not true it still modulates the collector base capacitance of Q4.

Typically a CB stage has 40dB or more reverse isolation (at low
frequencies with a low impedance connection from base to ground) whilst
an emitter follower may have 10db less isolation (depends on hfe at the
frequency for which the reverse isolation is measured. Using a
darlington or Sziklai pair will improve the reverse isolation over that
of an emitter follower).

These need to be supplemented with on board filtering as they aren't
quite as quiet as you need.
Either the NIST style darlington buffered RC low pass filter (one per
amplifier) and/or a modified (stabilises the shunt transistor re by
making its collector current approximately PTAT) version of Wenzel's
active power supply noise filter can be used.

A TNC connector should fit in the BNC holes.
TNC to BNC adapters are available when you need to connect to a BNC.

1% metal film would suffice.

You can easily measure the phase noise for low offset frequencies using
a low noise mixer with appropriate (not 50 ohm) IF termination followed
by a low noise (audio frequency) preamp driving a sound card. A 24 bit
sound card is ideal, however 16 bit sound cards just need a little more
preamp gain. No need for a PLL just split the output of a low noise OCXO
or similar source drive the mixer LO port with one output and the
isolation amplifier with the other whilst the isolation amplifier output
drives the mixer RF port. You will need to adjust the phasing between
the LO and Rf signals so that they are approximately in quadrature by
using a suitable length of coax or other means. You can even take
advantage of the 2 channel (stereo) sound card inputs to do get well
below the mixer noise and/or sound card noise floor by using cross
correlation techniques.

For higher offset frequencies (>100kHz) a similar setup with a 50 ohm IF
port termination plus a high gain low noise amplifier can be used to
drive the input of a spectrum analyser to measure the phase noise.

Bruce

have done.

You might bring those up with Mike Engelhardt (the author).  He doesn't miss
many tricks.

The Jung article at Gerhard's link claims 3 nv/root-Hz at 1 kHz.  Wenzel's
page claims 20 nv/root-Hz at 1 kHz.  What figures would be expected from the
modified version you're talking about?

What's the current thinking re: FFT window functions for noise measurement?
Does it matter what you use, as long as the window's equivalent noise
bandwidth is factored in?

-- john, KE5FX

John

The amplifiers can couple via the power supply impedance if one isnt
careful.
When one is trying to achieve crosstalk below -100dB this can be an
important path.

The transistor version of Wenzels circuit (without the opamp) is much
better than that.
The real issue is the noise below 1kHz.
Extending the low frequency cutoff sufficiently low is difficult but not
quite impossible.

Yes it does matter particularly at low offset frequency where the noise
bandwidth of the window approaches the offset and the noise process is
sufficiently divergent.
At higher offset frequencies the window function is much less critical see:

http://tf.nist.gov/timefreq/general/pdf/112.pdf

Bruce

In message qlcld41iisd7o6jq43o4tpbi3bgievgr6h@4ax.com, Gerhard Hoffmann write
s:

Be aware that LED's work both ways: current->light and light->current.

Their noise spectrum is influenced by the ambient light, in particular
flourecent lights, other discharge types, and, ironically, modulated
LED lamps.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

In message qlcld41iisd7o6jq43o4tpbi3bgievgr6h@4ax.com, Gerhard Hoffmann write
s:

Be aware that LED's work both ways: current->light and light->current.

Their noise spectrum is influenced by the ambient light, in particular
flourecent lights, other discharge types, and, ironically, modulated
LED lamps.

--
Poul-Henning Kamp      | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG        | TCP/IP since RFC 956
FreeBSD committer      | BSD since 4.3-tahoe
Never attribute to malice what can adequately be explained by incompetence.

Poul-Henning Kamp wrote:

Poul

In practice this isnt usually a significant problem with a forward
biased LED as the photocurrents are relatively small compared to the LED
bias current.
One can always use a light shield or coat the LED in opaque epoxy.

Bruce

Tom Clark, K3IO wrote:

Tom

Another technique is to null the AC magnetic field at the transformer
using a set of coils and a servo loop.
However this isnt always a particularly practical/inexpensive/simple
solution.

The AC field modulates the transformer inductances and hence the phase
shift.
Thus in the absence of shielding or field nulling a transformer with a
ferromagnetic core will suffer from this problem.
Ferrite core chokes will also suffer from inductance modulation by the
AC magnetic field.

One is either left with using RC coupling which makes it difficult to
achieve power gain from a CB stage and the dc load resistors can
contribute significant close in phase noise.
One may need to resort to using an emitter follower to drive the load.
The key factor in keeping the close in phase noise down at low offset
frequencies is to keep the dc gain from input to output low.
Where input means not just the actual RF input but includes any active
bias regulation circuits.
It is relatively easy to ensure that the dc gain and low frequency from
the input transistor base to the output transistor collector load
resistor is relatively low by using a large value resistor (capacitively
bypassed for RF) connected in series between the input emitter follower
and the first common base stage. With such a resistor one can dispense
with the bias regulator transistor and use a divider tap  buffered by a
low frequency emitter follower if necessary to determine the dc base
voltage of the input transistor. The drawback with such an approach is
the increased power supply voltage and the dissipation in the bias and
load resistors. With a 200 ohm collector load resistor and 45mA
collector current there will be a 9V dc voltage drop across it (its will
also dissipate about 400mW).

If the isolation amplifier dc collector current were increased to
80-90mA or so (may require paralleling the outputs of 2 amplifiers) then
it can easily drive 1V rms into a 25 ohm load.
This would allow a capacitively coupled 50 ohm load to be driven whilst
ensuring the output stage reflection coefficient is relatively low.
Since the collector current regulation circuit ensures that the output
current that the dc gain from any of the amplifier transistor bases to
the output transistor collector is low, it is well worth retaining the
bias regulation transistor. However the dc and low frequency gain from
the bias regulation transistor base to the outpout transistor collector
load will be about 1X. This can be reduced by increasing the bias
current regulator circuit effective reference voltage (about 1V if a RED
LED is used). Using a light shield or encapsulating the led in opaque
epoxy is probably a worthwhile precaution to avoid photocurrents and
consequent incidental phase modulation at twice the mains frequency and
its harmonics.

Alternatively narrowband techniques (tuned circuits with air core
inductors) could be used however the resultant phase shift tempcos may
be unacceptable.

Transformers aren't entirely ruled out as one could use an air cored
Guanella balun transformer to couple the output CB transistor collector
to the load.
The bandwidth of such a transformer is significantly less than a
transformer with a ferrite or other ferromagnetic core but may be
acceptable for a frequency distribution amplifier.
Such transformers tend to be bulky and are thus more practical at 100MHz
than 5MHz or 10MHz.

Bruce

Bruce Griffiths wrote:

Tom

Attached partial circuits showing the output coupling circuit using
either air core Ruthroff or Guanella Transformers show that these
devices short out the 200 ohm output resistor in Q4 collector for DC and
low frequencies. This is desirable as it reduces the dc and low
frequency gain from any of the isolation amplifier transistor bases
(including the bias regulation transistor) to the output transistor
collector to near zero, minimising the close in phase noise contribution
from these noise sources. Another advantage of using air core
transformers is that they do not saturate so that dc flowing in the
windings merely heats them a little due the finite dc resistance of the
windings.

The Guanella configuration is bulkier but is likely to have a much wider
bandwidth.

The air core transformers will need to be shielded to minimise crosstalk
between isolation amplifiers, however shielding is required in any case
the shields just need to be a bit larger.

To maximise physical separation between amplifier outputs whilst
minimising the physical separation of the amplifier inputs a star layout
where the individual isolation amplifiers radiate from the central input
may be useful.

Bruce

time-nuts-bounces@febo.com wrote on 09/24/2008 08:44:30 PM:

[snip]

TNCs will fit the same hole as will BNCs.

Joe Gwinn

On Fri, 26 Sep 2008 16:21:49 -0400, you wrote:

I won't change the 8662A; as long as it works I will not open it.
Hopefully for a long time. Its 10811 is probably better than
the MTI 230 in the GPS, noisewise.

I would like to limit this  to BNC, N, SMA and maybe APC / K  where it
cannot be avoided.
The number of converters required goes up with (n+1)**2,
the +1 for male/female. And that's optimistic.

regards, Gerhard

On 9/26/08 2:36 PM, "Gerhard Hoffmann" dk4xp@hoffmann-hochfrequenz.de
wrote:

If you need >18GHz performance, then I think 2.4mm is a better choice than
K. It's awfully easy to screw up the center conductor of a K connector with
a SMA that's not properly assembled.

Tom Clark, K3IO wrote:

Tom

Another problem with ferromagnetic materials such as transformer cores
and magnetic shields is magnetostriction.
The resultant vibration can modulate component parameters.

Bruce

On Thu, 25 Sep 2008 14:08:51 +1200, you wrote:

Yes, but capacitors have their problems, too. Ceramic ones will
misbehave like piezos, converting vibration to voltage. C0G or NP0
will be to small for AF; foil caps will be inductive over a wide range;
electrolytics won't be there if you need them in winter; Sanyo's
contraindication list for OSCONS brings chemotherapy to mind.

I'd like to hear more about this; I could evade to ADS.

In simulation with LTspice, a CB stage brings about 60 dB, which is
optimistic because parasitics in the decoupling caps are not modeled (yet).
The CC stage brings close to nothing, let's say 6 dB, because the
forward biased BE diode does its best to keep the BE voltage drop
at 0.6 V constantly. So most of what is delivered to the emitter will
make it to the base, too.
Total reverse isolation in simulation was abt. 130 dB with ideal
capacitors. Maybe I need another CB stage.

In real live, I measured abt. 100 dB reverse isolation from output
to input upto 20 MHz this evening. Above that, it gradually became worse
and at 60-80 MHz there was some kind of pole / notch. That was completely
without shields and with coax cables that were not really network analyzer quality.

Splitting the bias resistors in two and decoupling in the middle brought nothing,
neither in simulation nor in measurement; but it might make a difference
when the shields are soldered in.

There isn't much choice. NiCr SMD is meant for precision.
1% resistors are thick film, usually.

The PLL is already here. And the preamp, highpasses etc in the
Wenzel appnote style. I changed it to true differential to fight offsets,
with the obvious inflation in 2SK369 FETs. Matching these was no fun.
This must be replaced with something more repeatable. The power consumption
of the relays does no good to the offset, either.

http://www.hoffmann-hochfrequenz.de/downloads/IMG_0009__1000_q50.jpg

I have crammed 100 meters of Aircom-plus cable into a 6 HU 19" box for
407 nsec delay. I can barely lift it :-).
The variable delay is under construction. Coax relays with
cable delays in binary steps under computer control. Still a lot to do.
Two 16 bit 2.5 MSPS digitizers are left over from an earlier project.
Does anybody out there have Matlab or C code for the three-cornered-hat?

To return to the amplifier: I'll probably go the macho way without
output transformer. With twice the bias current I should be able
to develop half the voltage into 25 instead of 200 Ohm. It works
in simulation. I'm approaching the BFG31's limit of 100 mA here.
Maybe I need 2 in parallel. I searched today for fatter alternatives
to the BFG31: nothing.

Having a wideband 25 Ohm load is a drawback at AF; the load could be
paralleled by a choke. As long as the choke is fairly high impedance,
it will do much less harm at the 25 Ohm than at the 200 Ohm level.
I think there are 1uH chokes w/o ferrite in 1206.
A few of them would be enough and transformer coupling does not matter.

Gerhard

Gerhard

Yet another reason for temperature control?

For low voltages one could look at supercaps for filtering as some of
these have relatively low esr and low leakage.
However these are somewhat bulky.

I'll look at this again when I finish some fibreglass work on a 24"
telescope and sort out simulation of software implementations of sigma
delta and MASH DACs.

The simulated reverse isolation depends critically on accurate
transistor models.
This is one aspect that can vary widely between simulators.
60dB is indeed optimistic.
S21 for some of the older HP microwave transistors in CB configuration
is listed at 0.01 to 0.005 at 100MHz on the datasheet.

Surely the CC stage has somewhat higher reverse isolation than that.
When the emitter current is modulated by a signal connected to the
output of a CC stage via a resistor (50 ohm) then the base current is
usually  somewhat smaller than the emitter current.
Ib = Ie/(hfe+1).
Even at 100MHz at 4GHz ft transistor has a current gain of about 40 or
so (provided this doesn't exceed the dc gain) so the reverse isolation
of a CC stage at this frequency should be around 32 dB.
At frequencies somewhat below ft the reverse isolation of a CB stage is
limited by the Early effect which produces a varying Vbe for a fixed Ic
and varying Vcb.
This can be reduced by regulating the emitter voltage with another
transistor, however this reduces the high frequency reverse isolation.
This technique is very useful with a CB input stage as it also
significantly reduces the distortion (provided the collector load
impedance isnt too large)

Adding another CB stage will increase the phase noise floor noise slightly.

Spectradynamics achieve about 110dB reverse isolation over the 80-120Mhz
band using a similar amplifier topology in their HPDA100 distribution
amplifier.

My simulations indicate otherwise when the parasitics of real components
are included.
You could try measuring the attenuation of the bias network itself when
RF is injected at the transistor base taps.
Do this without the transistors present.

Digikey list 1% and 0.5% Susumu thin film surface mount resistors in
their catalog.
Physically larger resistors can be used in the voltage divider string.

If you don't need to go below 0.1Hz or even 0.01Hz then you can either
use an integrator or AC coupling between stages to eliminate the
amplified offset from the 2SK369's.
A differential input stage makes it much easier to reject low frequency
(<1Hz) power supply noise than when using a single ended input stage.
You should disable the PLL for component phase noise measurements and
just adjust the phase with a length of cable or similar.

There is some evidence that the flicker noise of JFETs as well as BJTs
can be much lower than specified on datasheets if one shields the
devices from air currents and stabilises its temperature (temperature
regulation for frequencies below 1 mHz , large thermal time constant
enclosure for frequencies above this). Have you measured the preamp
noise spectrum in the flicker region?
You could use latching relays to overcome the dissipation problem as
these only need to be powered long enough to latch in the new state.

Since the IF port impedance of a mixer (even if terminated in a
capacitor for very low noise) is relatively low, a bipolar input preamp
using an SSM2220 should work well just add an integrator to control the
~1mV residual input offset . NB for low gains you may  want to connect
a  resistor in series with a capacitor between the collectors rather
than reducing the collector load resistors as Enrico suggests.

Power dissipation in the transistors may be a problem at 80mA unless 2
amplifier outputs are connected in parallel.

You would probably achieve better choke performance if you wind you own.
If the frequency were higher a 1/4 wave shorted 50 ohm transmission line
could be used to short out the 25 ohm resistor for dc and low frequencies.

Bruce