LMH6551: single ended swpt ac to diff output

Morning Robin, 

I think your are slightly imbalanced right now, here are some corrected R values using the THS4561 FDA model, 

The FDA device type has an active impedance effect if running single to diff. That design flow is described in several of the FDA datasheets (not the LMH6551 as I had not developed when it was introduced). Page 29, but this applies to all FDA's, the table of R values in the schematic above follows this solution flow. 

There is no TINA model for the LMH6551 on the TI web site, just Pspice, there is one in the TINA V11 library, here is that dropped into the THS4561 circuit, this is quite flat through 100MHz and very close to ideal gain of 1. (0dB). The solution is showing exact R values, use closest available. 

And this file with the LMH6551, 

LMH6551 single to diff.TSC

That orginal Aug. 2011 article is getting hard to find, but here is part 1, 

https://studylib.net/doc/18053452/dc-coupled--single-to-differential-design-solutions-using...

Hi Robin,

if you connect a 50R source to J7, the impedances seen by the +input and -input of LMH6551 are not balanced and the LMH6551 will not work well. See figure 29 of datasheet.

Also, why do you need 1200µF AC coupling caps at the outputs?

Can you tell more about the application?

Kai

Oops, Michael is faster than me

Only because I focus on just a few of these, 

Hello Michael:

Very much appreciate your time and effort in responding in such a useful depth about the FDA and  its confounding operational peculiarities...been there for a long time.

There are some FDAs that apparently won't require so many resistors...for a "price".

I will like to cover a response to Kai here even though his response deserves separate attention!

First off, the application IS CONTROVERSIAL. Yet I will try to explain "why" FDA.

This circuit  gets input from a network analyzer- 50 Ohm source, always swept mode with small signal levels- no more than 10mV.

Reason: this is used to "inject" a signal in the feedback loop of a dc-dc converter designed using TI chip. It does not matter which one...the purpose is to obtain gain margin/phase margin  from the  Bode plot

Now: you would say what is the problem...use any one of the commercially available Network Analyzers for the purpose.

Here is the problem: our modules -most likely will be used by large system houses- where engineers integrating these modules, MAY NOT /NEED NOT be expected to have /buy such +$15000 equipment. Yet they hopefully will have some Vector Network Analyzers. Most of these cannot handle any DC common mode at all, some may do up to 5V ...you will damage $100,000 equipment if such DC  is applied to its channels.

The amplified "signal" across the injecting resistor with DC across  provides the gain/phase information.

Even with xmfr, you need DC rejection - whose value determines the low cut-off of the Bode Plot.

We used to make such "wide band Baluns" back when: those cores are no longer available. Also will be >$10 or so.

Yet, 100% of such applications use unanimously an isolation transformer. I have had some discussions here earlier and that hon. Person stated to stick with what works- isolation xmfr.

Here are the reasons for deviation.

#1 such transformers used to cost upwards $300( no longer available). Would offer a lower cut-off no better than a few kHz, if that.  We need to demonstrate module stability down to 100Hz or close.

    Nothing to beat FDA ability in this mode. HP 8751 or similar will go down to 5Hz if not using the s-parameter set. We will be using that NA.

#2 xmfr as Balun is not guaranteed. It is what will be obtained without adding many extra parts. FDA can be tuned to get 0 output CM.- adding 0.1 cent parts. So FDA output can be directly connected to NA channels.

In order to protect from damage, we do have clamping diodes

#3  xmfr approach is not a low profile repeatable approach. Our approach is low profile, repeatable low cost ever.

#4 Signal source impedance is very stable 50 Ohm, and the terminating floating nodes across have 10 Ohms. Such  large step down is always non-ideal with xmfr.

FDA has no such problem. At least up to 10 MegHz scan.

That said, TINA-TI model is not giving 0  output CM . 

THIS IS the key feature why I am trying this circuit. Modeling is important before we make the board.

In fact, this is for our "EVALKIT"...hoping to sell a gazillion to TI!

LOL

Thes edays, if customer does not have access to NA, we can setup a zoom/Teams session to walk him/her through the whole process while he/she has her's powered up.

Your review and comments will be appreciated. 

I will simulate the circuit using TINA-TI  asap. 

The reason for high decoupling capacitors is clear from the above, Kai, I hope.

r

Thanks Roben for all the detail, if output CM rejection paramount, make sure you match your resistors as closely as you can afford. it is easy in sim to see the output CM signal by putting equal resistors to a midpoint and probe there. Then running a monte carlo on your resistor tolerance will show you this effect. 

One other point I had thought about earlier - most network analyzers have a blocking cap in the source output, not having that on the non-signal side might interoduce DC output offsets. Those will be blocked off at the test signal injection point by the output blocking caps, but I wonder if there might be some other effects of interest. without that matching blocking cap on input side, there will be an imbalanced noise gain on each side (which generates CM output) until the source cap shorts out. 

I think I understood you system description as converting the RF source of some test instrument to a different source with minimal CM content. 

Wow: so fast ,I have not finished simulations...LOL

Is it possible to get almost 0 output CM?..or some but equal value ? ...I will check if most NA can handle 1.5V which impressed me a lot! 3/2 duh!

Does not happen for any other value.

Now the caps from the injection points.

Theoretically, the unequal value will create Vsig_upper_node  and Vsig_lowr_node  signals at ch1 and ch2. Plus some CM from the FDA.

If FDA can be nulled, as I am hoping, then (sig_upper-sig_lowr)/NA-sig Bode plot will be close to ideal, won't you think so?

re

Sounds right, again sims can show all these effects pretty quickly

And incidentally, it doesn't sound like you are after a very high output swing. There are quite a few lower max supply FDA's that might also work. The THS4541 comes to mind. Again, much faster than your 10MHz max target, but it seems you are looking for signal perfection in that 10MHz span and extra bandwidth is useful, that would be a +/-2.5V or even lower depending on your max output swing needs. 

Oh and I found an old FDA parametric table for the TI/NSM parts - that middle section might be of interest to you - the upper section of higher supply parts shows the LMH6550 row, with the LMH6551 a related but slower part, The THS4541 is about twice as fast with 1/3 the input voltage noise level. The lowest block is for DC precision FDA's, where the newer THP210 has not been added. 

Hello Michael:

Thanks again for the list. Very helpful indeed.

I picked up THS4120 from the TINA list. Simulated with your values excepting 1k feedback values, 10mV NA signal, +5/-4.7( neg from TPS604000), lo and behold, output CM is almost 0 (0.1 mV mas o menos), this output is something the NA/loop stability measurement setup can use!

So we are going for it: need 1 channel chip: is there? THS4120 seems a quad...have to check in  ds.

thnx a million for helping here: was of immense value; you got to know.

r

Hi Robin,

this appnote might interest you:

https://www.omicron-lab.com/fileadmin/assets/Bode_100/ApplicationNotes/DC_DC_Stability/App_Note_DC_DC_Stability_V3_3.pdf

Kai