Bypassing

yep, the small cap resonates with the parasitic series L of the big cap and makes a nice tank...

paralling caps worked in the old days when the "big cap" was a low Q pig... parallel caps doesn't work so well these days when a 0.1uf cap is a high Q low L SMT. And for the same reasons, it's no longer needed.

Mark

This guy reckons there is a sort of synergistic advantage in the values being identical.

(see the "50 to 500MHz" section:)

(I posted a link to him before, you were not impressed IIRC)

I just did a simulation with two capacitors (maths too hard at the moment). And he seems to be right. The circuits with different values all have resonant peaks in impedance. When they are all the same there are none, just a single series-resonant "zero". Either way, they are all the same at high and low frequencies (given the same total capacitance).

So your rule of thumb is right, use the maximum capacitance you can get (cheaply, in a given package), and all the same values.

Version 4 SHEET 1 880 680 WIRE 176 96 64 96 WIRE 272 96 176 96 WIRE 176 128 176 96 WIRE 272 128 272 96 WIRE 64 160 64 96 WIRE 176 240 176 208 WIRE 272 240 272 208 FLAG 176 304 0 FLAG 272 304 0 FLAG 64 240 0 SYMBOL ind 160 112 R0 SYMATTR InstName L1 SYMATTR Value 10n SYMBOL cap 160 240 R0 SYMATTR InstName C1 SYMATTR Value {C} SYMBOL ind 256 112 R0 SYMATTR InstName L2 SYMATTR Value 10n SYMBOL cap 256 240 R0 SYMATTR InstName C2 SYMATTR Value 100n SYMBOL current 64 160 R0 WINDOW 123 24 116 Left 0 WINDOW 39 0 0 Left 0 SYMATTR InstName I1 SYMATTR Value 0 SYMATTR Value2 AC 1 TEXT 128 392 Left 0 !.ac dec 100 1e6 100e6 TEXT 0 8 Left 0 !.step param C 20n 100n 20n

--

John Devereux

Over the last years I learned a lesson. The more recent an app note is the more careful I become regarding the claims and suggestions in there.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

47pF in an 0402 won't have much different inductance than the 0603. I always use largish capacitance in a small package as the rule. I do mix values but the bigger one is always something like a 100uF tant at one place on the PCB. At high frequencies, they look like resistors across the supplies.

The designer might have looked at the THS4303 datasheet. It shows, on the first sheet, a 47 pF bypassing capacitor in parallel with a series

0.1 uF capacitor and 30.1 ohm resistor.

This is a very unusual configuration. It suggests to me that the guy who wrote the datasheet experimented and found that this configuration worked for him when other configurations didn't--in his particular set-up. Why else use the series resistor?

This suggests that the part has some stability problems that even the applications engineer had a hard time dealing with.

The datasheet also shows some larger (22 uF) caps bypassing the supply lines that are isolated from the op-amp by ferrite beads. This is probably a very good idea. If the designer left out this feature, it might have led to your problems.

Also, if the circuit doesn't use the recommended output isolation resistor, you could have ringing.

Maybe he did, but that doesn't make a lot of sense in a truly wideband application.

This opamp has an 18 GHz *closed loop* gain-bandwidth product. It has a right to be pickey.

Application engineers do the strangest things.

We did use beads on the supply lines to keep various amps from talking to one another through the power pours. We have an absurd amount of GHz gain jammed into a small box, and not outright oscillating is a victory itself.

The two big caps in parallel, on a smallish local power pour, bead isolated, work fine.

John

And all that for under four bucks. TI came out with some great amps. Now if they only had piped out the IN- directly ...

[...]

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

That node probably can't be brought to a pin.

On our latest gadget, we aren't grounding the IN- pin, but bypassing it hard and driving it from a trimpot + c-load opamp, to use it as a DC offset trim. Wish us luck.

The TI rep has told us that standard analog remains a very important part of their biz. Looks like they got beat up on cell-phone parts (huge volume, zero margin) so are going back to customers like us for revenue. They do have the fastest 30-volt process on the planet.

John

I was hoping pin 12 could do that with pin 13 VS+, but probably you are right, this can become iffy.

That's one of those situations where your new PADS guys can test his RF skills :-)

Yes, and on one project I sure was glad they do.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

Girl. The Brat, in fact. Her first board was a 4-layer, 1.5 GHz, G=100 amplifier. Wish us luck.

The best pcb layout people I've known were women.

Lest I repeat myself... do NOT use THS3062!

John

Historically I tended to ignore the app' note suggestions, until running up against the AD8099. Simply not possible to make it stable unless every single component around the chip is as specified on the datasheet. There's page after page after page of these recommendations. Must have had big stability problems during the chip design as even the output pin had to be routed next to the inputs to avoid parasitics. Phase margin looks next to zilch. On the other hand, a faster, replacement EL5166 worked like a dream, with just 100n decouplers.

So she did come back and join your company? Great! Or did she feel guilty about having bought that Jeep from, ahem, savings?

I also worked with several women who were excellent layouters. But my current layouter is a guy, around 60, tons of experience. When we discuss an RF critical layout after giving him my schematic and netlist we usually talk about RF-tight areas for one minute and then another four about politics. Boards come out like clockwork and luckily we are on the same page in terms of politics :-)

I never use CFB amps. When I need this kind of gusto I always go discrete. Much cheaper, too.

Last higher voltage process amp I used was the THS4021. Worked like a champ even though it was slightly lower in BW than the marketeers said, but good enough.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

She's working for us now. She managed a pretty big programming project (with a terrible programmer) successfully, and is doing a bunch of marketing sort of stuff... ads, press releases, links. Apparently many business schools require a couple of years of work experience, so she may put in a couple of years here, then go for an MBA.

She really likes pcb layout, incidentally.

A lot of her classmates/teammates are still unemployed, or doing scut work.

But how do you get analog precision from discrete amps? I mean millivolt offset,

Maybe they should open up their search a bit. Back at my university you had to get a minimum of six months of industry work under your belt. Else no masters degree. There were times when such jobs were scare and I ended up working a lot in foreign countries. Gives a student a great boost in experience (but they may have to sit down and learn a language).

If needed I use two tricks: Clamping for DC offset restore and controlled FETs for gain auto-calibration. But most of the time DC performance is irrelevant and then I often resort to PIN diodes for gain cal. If it has to be really cheap then stuff like band switches or something with a somewhat useful PIN characteristic.

Yeah, but clients got zinged a few times by availability issues and "not so favorable" pricing. However, for smaller volume production CFB could be the ticket.

--
Regards, Joerg

http://www.analogconsultants.com/

"gmail" domain blocked because of excessive spam.
Use another domain or send PM.

This remains true, so far as I'm aware (checked up on it only a few months back, in fact.) Called a "capstone" here.

Jon

When it comes to bypasses, you can't overcapacitate!© ;-)

Cheers! Rich