Op-Amp noise figures: Generally >=10dB?

That matches what I know. A/D converters have disappointing noise characteristics, too. If you want low noise, you've got to have some discrete transistors (or at least MMICs, which are often little more than transistors with built-in bias) in there someplace.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" gives you just what it says.
See details at http://www.wescottdesign.com/actfes/actfes.html

There are some lower supply voltage versions that would deliver better than that. 2nV/rtHz is way too high for such jobs. OTOH why spend north of five bucks on an opamp that is single-sourced when a couple of dimes would do for a transistors stage?

MMIC are basically glorified RF transistors with the bias resistors already in there. So far I've never really warmed up to one though, on account of many of them being single-sourced and not giving me a compelling advantage over transistors.

Quote "Today, the performance of wide-band op amps is making them viable alternatives to more traditional open-loop amplifiers like monolithic microwave integrated circuits (MMICs) and discrete transistors in communications design."

Uhm, well, besides noise there is one major roadblock here: $$$

--
Regards, Joerg

http://www.analogconsultants.com/

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

There are a good number of folks out there who seriously shy away from circuits with transistors. People who will run away screaming at the sight of a bare transistor doing anything more complicated than switching a ton of current will get warm fuzzies from an empty box on a schematic with an obscure part number that looks like an IC.

I suspect that a good part of the MMIC's appeal is that they are integrated circuits (however loosely that term is used in this case) and so they must be 'better characterized', 'safer' and 'easier to understand' than transistors.

Sigh. No kidding. I've watched people insist on struggling with 1GHz gain-bandwidth product amplifiers that just want to sit there and scream at several hundred MHz trying to do a job that could be easily handled by a 2N3904 or surface-mount equivalent and a few discretes.

But hey - if it weren't for folks like that, _you_ might have to fire up your chain saw and make fake totem poles to sell to tourists, assuming you could find the wood.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Do you need to implement control loops in software?
"Applied Control Theory for Embedded Systems" gives you just what it says.
See details at http://www.wescottdesign.com/actfes/actfes.html

Hi Joerg,

By our standards many of them are quite cheap these days, e.g.,

Thanks, Tim, that's what I wanted to hear.

It's the same with food. I'll never understand why people buy ready-made pizza that is expensive, must often be re-heated and doesn't taste nearly as good as made from scratch. In about five minutes we'll have lunch. Home-made bread, of course. I can already smell it.

I still remember the words of an engineer after looking over my shoulder: "You mean, that's it?"

That client initially was considering dozens of beefy Analog Devices DSP for the circuit that I was tasked with. I don't remember what they cost, just that I almost choked.

And yes, there are a bunch of 2N3904 in there.

Yep :-)

--
Regards, Joerg

http://www.analogconsultants.com/

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

If you must do SMT assembly locally, yes, MMIC can occasionally win. For production in Asia, not a chance.

They have pretty good specs but look at the company name. Defense -> lotsa $$$.

--
Regards, Joerg

http://www.analogconsultants.com/

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

This is all low-volume stuff, 100% designed/manufactured/assembled/tested here in the U.S. In fact, other than some PCBs it's 100% in Oregon even. :-) Some of these sorts of boards will have dozens or even hundreds of little amplifiers or RF switches like this strewn around, and a tech who has to find and fix one bad MMIC or PIN diode will typically spend somewhere between 15 and 60 minutes doing so, so there's a large emphasis on minimizing the raw number of parts as well (since the failures are about 90% due to bad solder joints).

OK, but even for someone experienced such as yourself... wouldn't you probably spend a week or more designing & testing something comparable out of discrete transistors?

---Joel

down

LT1028 is pretty good in a 50 ohm system. But no semiconductor is going to give a really low noise figure in RF apps unless you tune the input; the impedances are usually all wrong... bipolars are too low, fets are too high.

monolithic

MMICS can be great. A $1 mmic has 15 dB of gain dc-8 GHZ, (approximately) 50 ohms in/out, needs one external resistor and two blocking caps, and is unconditionally stable. That's hard to do with discretes.

John

We use a Hittite mmic that's $200. But it swings over 5 volts p-p, dc to 20 GHz.

John

It does look good; thanks for the tip.

As I recall that you have a reflow oven, pick and place machine, etc., so are you doing all of your board stuffing in-house? Or does some of it go to contract manufacturers?

---Joel

Depends on what these amps are supposed to do. If it takes the techs this long you might want to design in some more testability. Examples:

a. Split VDD rail plus a few ten milliohms to each: Not my favorite but you may get away with it. Then the tech can quickly check the voltage drop. If bank #6 reads low then the fault must be in that bank.

b. Weak resistive coupling from a common RF rail: If you can stomach a wee bit of crosstalk (can be kept to -60dB or even less) run resistors to a common rail. Then the tech can send RF juice into that little rail and just rattle off the outputs. Oops, #17 is dead ...

c. H/E field probes: Very handy to sniff out where RF is and isn't. Requires self-training and an acceptance of the fact that not all effects can be easily explained but are still valuable. A bit like a dowsing rod.

d. Automated testing: I am a big fan of this one. MUX, PIN diode, whatever switches in test signal, PC rattles off outputs. That's what I've done since my very first job in industry. Basically the tech presses a button, sips some coffee, then looks at the screen when a beep and a red flag showed up. It would tell him/her to replace R129 and re-test. Not all techs like that. One told me he felt like the Maytag man.

Sure, but that goes for the opamp or MMIC solution just as it does for the transistor. The good thing about transistors is that you can easily spot a bad one because the bias stuff is external. All you really need for that is a $2.99 Harborfreight meter (one client actually does that!).

--
Regards, Joerg

http://www.analogconsultants.com/

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

down

at

monolithic

Yes, for such wideband apps MMIC are great. As long as they aren't much above a buck.

Do you know one that'll swing a few hundred volts into a highly reactive load at 74HC speeds? Just kidding but I could really use that right now.

--
Regards, Joerg

http://www.analogconsultants.com/

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

IIRC they do a lot in-house. If you guys need a good contract place let me know. But it would be in Kahleefohniah. And I am only going to tell if you guys stop giving us so many tickets on I-5 :-)

--
Regards, Joerg

http://www.analogconsultants.com/

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

Hi Joerg,

The 15 minute estimate was based on some 5-10 minutes to find/verify the error and then another 5-10 to physical desolder a part, clean up the pads, and solder down a new part. The hour estimate is based off of having to figure out which part is bad... some of our designs have a bunch of switches, amps, or whatever in fancy series/parallel/cascade arrangements and (at least as currently designed) some faults can not be immediately pinpointed to a particular component. (We once had a big switch matrix where we, well, I -- with the approval of my boss -- decided to drop the DC blocking caps between the FET switches to save on layout time/parts count/etc., and electrically it was OK since everything was at the same DC bias anyway. But unfortunately, if you had ONE bad switch that was, e.g., shorted to ground, it knocked out almost every single switch in the box since it would upset the neighboring switche's DC biases, and those dead switches would then upset their neighbors, and so on throughout the entire matrix until every single switch didn't work. ^$*&^$$&&*$$%!@!@# In retrospect deleting those caps was a seriously poor choice on my part... we spent far more money in time finding shorted switches than we ever would have on a few hundred $0.01 caps per board...)

Thanks for your idea on making troubleshooting easier. We do a fair amount of automated testing now, although it stops at the point of getting a reading back from a network or spectrum analyzer and telling the tech what was being tested at the time (e.g., which amplifier bank or filter section or whatever) so he knows where to start looking for problems.

Mmm... seems to me that if you need, e.g., 20dB gain to 3GHz (or even just

30MHz), buying a $0.79 MMIC and slapping it down on the board is orders of magnitude faster than designing such an amplifier yourself?

---Joel

We do roughly half our pcb stuffing in-house. We do protos, small batches, and sometimes bigger batches if things are slow. If we're busy and have 20+ boards in a kit, we send them out. We do BGAs in house and have 100% yield so far.

We do all final assembly and test in-house.

We have two identical Essemtec semi-auto p+p machines, an Essemtec oven, a couple of pneumatic paste stencil machines, and the cleaning machine that's valuable if only because it drives prongie wild.

John

es down

and

tage

end up

thing

The AD797 is a similar beast with a slightly less cranky output stage

-- Bill Sloman, Nijmegen

...

I'm curious what sort of circuits you're using that you only get as low as 10dB NF. For an off-the-air HF receiver you shouldn't need anything better than that, but you may be doing something different. But in any event, getting below 5dB assuming you're happy with fairly high gain doesn't appear to be difficult. I recently put an HF receiver into production that does better than 10dB, using an op amp as the first amplifier stage, and that includes the loss in some input filtering and switching; it's under 7dB at the amplifier stage input, and lower would have been possible (but wasn't needed).

Cheers, Tom

Cheers, Tom

You can also do that if you use amps that are in essence opamps but not marketed as such. Mostly VGA, and Analog Devices carries a nice selection. So far I went discrete anyhow in >80% of designs. While I do trust AD regarding non 2nd-source discrete was lower in cost and I did eke out a tad more in SNR. Mostly ultrasound where you are in the 3MHz to 50MHz range but must get as low in noise as possible. In ultrasound there is no man-made static and stuff to contend with, it's whisper quiet.

--
Regards, Joerg

http://www.analogconsultants.com/

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