Hello Gerhard, everyone,

I made a test using the THS3491 on a slightly modified PCB from my "FDA"
design [1].

At first with 200 Ohm feedback/gain resistors all I got was (pulsed, in
bursts) self-oscillation (around 150MHz or so).
The datasheet [2] does however warn about this and suggests 976 Ohms for a
gain of 2V/V, so I went with 1k.
That got rid of the self-oscillation - but now I am left with severe (16 dB
@ 230 MHz) gain-peaking! (see attached figure)
The datasheet shows a 2V/V gain-plot with only minor peaking at >700MHz.

the noise, so increasing them further for more stability is not great for
AM or PN.
I did not bother with the phase-noise probe yet - better to get rid of that
gain-peaking first (?).
Is there anything else that can be done, besides increasing Rf/Rg for more
stability?
Anything obviously wrong with the layout? It does work fine with LMH6702
and LMH6609.

Thanks for any suggestions!

Anders

[1] https://www.ohwr.org/projects/pda-8ch-fda-8ch/wiki/wiki
[2] http://www.ti.com/lit/ds/symlink/ths3491.pdf
[3]
http://www.anderswallin.net/2015/12/frequency-distribution-amplifier-v2-simulations/
[image: ths.jpg]

On Sat, Nov 17, 2018 at 1:00 AM Gerhard Hoffmann dk4xp@arcor.de wrote:

Has anybody tried the AD8003 (triple video line driver amplifier) for use as
distribution amplifiers?  It looks to me like this part might be a decent
choice.

On caveat, however:  don't succumb to the temptation to reduce the circuit
bandwidth by putting a capacitor across the feedback resistor- here be
dragons!

Each section is a current feedback opamp, intended for use in a
non-inverting
2X voltage gain configuration (leading to unity gain when a back-termination
resistor equal to Zo is used).  It should be equally good for either 10 MHz
or for
PPS pulses and the like.

On concern I have is that there might be some difficulty with driving a
non-
terminated transmission line (by mistake, of course).

One other thing is that this part uses an "exposed paddle", which needs
to be soldered down to a well-grounded pad beneath the device for both
electrical and thermal reasons.  For those in possession of a suitably-
modified toaster oven, this should not be a big issue.

Dana    K8YUM

On Sat, Dec 15, 2018 at 11:03 AM Anders Wallin anders.e.e.wallin@gmail.com
wrote:

as

I don't have any NF or residual PN numbers handy for the triple-amp
versions, but the single-amplifier part from the same family (AD8000) is
somewhat noisier than the LMH6702, with an NF around 22.9 dB for the former
versus 17.5 dB for the latter when both are tested with Av=+2.

The AD8000 has the advantage of an enable pin, and it's also available in a
QFN package that may be a bit more convenient to use on a PCB.  The AD8000
is also good for about 0.5 dB more output power than the LMH6702 can deliver
at 10 MHz.  But that's about it; the LMH6702 is preferable in all other
cases from what I've seen.

Being a triple opamp, the AD8003 would be comparable to the LMH6733, which
also includes enable/disable pins.  I haven't done any benchmarking with the
AD8003 but would expect comparable results, with the LMH6733 a bit quieter.

Re: feedback resistors, the data sheet for a given CFB amp will recommend an
optimum value, often tied to a particular Av, and it's best not to deviate
too far from that.  I've always found that 237 ohms works well with the
LMH6702 regardless of gain, but as noted, the THS3491 wants a much larger
resistor value.

When tested with Av=+5 -- meaning overall gain of +8 dB with 6 more dB lost
in the series termination -- the LMH6702 measures 15 dB NF and the THS3491
17.3 dB.  These figures are similar when observed with either a traditional
NF meter or a residual AM/PM noise measurement.  Neither of these parts
contributes enough 1/f noise to degrade the performance of an oscillator, so
they are both excellent choices for distribution amplifiers in the -165
dBc/Hz neighborhood.  The nod goes to the THS3491 for such applications,
IMHO.  It's not that much noisier than the LMH6702 and the additional output
power is very welcome.

I don't see a good use case for the AD800x family unless you need a
single-amplifier package with an enable pin.  There's nothing really wrong
with them, but they are noisier than the competition and they don't excel in
any other areas AFAIK.

Not an issue, they will happily drive anything (or nothing) through a 50R
series termination.

True, and all of these parts can also be mounted dead-bug style for
experimenting and prototyping.  I keep several LMH6702 amps around in
Hammond boxes, mounted on bare copper with small cermet gain control
trimpots and powered from NiMH AA cells.  They are dirt cheap, rock stable,
and very handy.  If I build any more of these amps I will probably switch to
THS3491s and small lithium batteries.

more

I can 't tell much from your photo but it sounds like you may have some
stray C near the inverting input terminal.  It's a good idea to open up the
ground plane near the - input with all amps in this class.  Watch the output
capacitance upstream of the termination resistor as well.  These parts are
all well-behaved, but the data sheet guidelines have to be followed more or
less religiously.

-- john, KE5FX
Miles Design LLC

A few more notes on the THS3491:

http://www.ke5fx.com/ths3491.htm

I had tested it a few weeks ago, but hadn't gotten around to gathering and sorting through the plots.

-- john, KE5FX
Miles Design LLC

So the THS3491 is an LMH6702 replacement except
for a slightly elevated noise floor.  Why would
anyone not want to use the LMH6702?  What does
the THS3491 have going for it?

Rick

On 12/16/2018 11:02 PM, John Miles wrote:

Stylish dead-bug! ;)
On my board I increased Rf/Rg to 4k7, to get rid of the gain-peaking.
With 4k7 resistors I see +20 dBc/Hz worse phase-noise with the THS3491
compared to LMH6702 :(  (image attached)
If noise power scales with the resistors then it explains about 4k7/200R =
14 dB, for one amp, and perhaps 17 dB for two amps. Still missing 3dB
somewhere in that reasoning...

John: would you say if one can use the suggested 976R Rf/Rg resistors then
below -160dBc/Hz for a 10MHz carrier is possible for a gain of +2?

At least we can say THS3491 is much more tricky to work with than LMH6702
so far... (?)

Anders
[image: ths2.png]

On Mon, Dec 17, 2018 at 9:04 AM John Miles john@miles.io wrote:

Power output is what caught my eye.  After the loss in the series termination, the LMH6702 struggles to put out +15 dBm at 10 MHz without distortion, and you lose a bit more P1dB at 100 MHz and up.  +15 dBm is a pretty reasonable spec for a distribution amp, so you'd like some headroom if possible.

Running the LMH6702 at +/-6 helps a bit, as would using two of them with an output transformer, but a single THS3491 can output +20 dBm on +/- 15V rails by itself.  (There's a note in the data sheet that mumbles something about a +/- 100 mA long-term limit due to electromigration, but that still allows a solid +18 dBm after the series R.)

-- john, KE5FX
Miles Design LLC

The 3 dB figure I mentioned on the page was just a fudge factor to make the numbers come out with the resistors I tried.  I didn't test with 4.7k.  Looks like the NF in your case is 45 dB while my ad-hoc approximation would predict +34, so it clearly doesn't hold up at that end of the scale.  Where does the 200R come from?

The 27 dB noise figure with Rf=Rg=976 is awkward.  If your NF is 44 dB with 4.7k, you'd need +27 dBm to stay 17 dB above the kTB PN floor at -177, so that's out of the question, and with the 27 dB noise figure at 976 ohms you still need +10 dBm.  Disclaimer: no coffee yet, sanity-check all numbers. :)

You do seem to be getting somewhat worse results than I am, and it would be interesting to drill down and  figure out why.  The peaking behavior is excessive -- certainly if you have to use Rg=4.7k to tame the amplifier, you can rule out the part for any low-noise applications -- and what's up with the flicker noise below 10 Hz, I wonder?  I haven't seen any excess 1/f noise at all, and -115 at 1 Hz is a lot.  It might be OK for a low-performance distribution amp, but good OCXOs will be degraded, and I'd assume it's impacting your overall phase stability as well.  Need to find out what's up there.

-- john, KE5FX
Miles Design LLC

Am 17.12.18 um 16:22 schrieb Anders Wallin:

I can't claim any personal experience with the THS3491. They are on

my next DigiKey order, but the data sheet calls for Rf=976 Ohms

(chipscale package)  and for 2K1 in the SO-8 package. And the SO-8

still has at least 2 dB peaking in the data sheet. The huge difference

of 976 Ohms versus 2100  Ohms just for the package (and still close

to fail) makes it clear that the SO-8 is not made for low-gain
applications,

and that with a board that's made for it. (GND removal under the input)

Already at 5V/V it seems to be well-behaved to 600 MHz and then

dropping without a peak, and that without outrageous resistors.

Maybe it's possible to augment Rfdbk with ferrite beads @>200 MHz

but they are pretty good Hi-Q inductors at low frequencies.

BTW already the LMH6702 can be somewhat picky  with regard to

decoupling; a cap directly from VCC to VEE is much more

important than anything to GND, esp. for distortion.

Op amps don't care much about the concept of GND; they are happy

when their inputs are equal.

Has anybody succeeded to massage the Spice model so that it fits

LTspice?

A distribution amplifier that can be used today for 5V 1PPS into 50 Ohms

and tomorrow for 18 dBm at 100 MHz should be quite usable if one gets

the noise down.

regards, Gerhard