Need ideas for high bandwidth gain stage..

Ok DIY gurus, this time I'm in need of an opamp that will give me over 60mhz of bandwidth at a gain of 10+. I'm not talking datasheet buzzword bandwidth either.

Yep.

The amp needs to be flat from 25mhz to 60mhz.

The only opamps I can find that can give me the true bandwidth at G=10 are current feedback, but there is a problem..

This has to be single supply(5v) AND I need to bias the midpoint to VCC/2 (2.5v).

I would normally use an inverting arrangement and bias the non-inverting input on a voltage feedback opamp but that doesn't work right for a currrent feedback opamp. Since the inputs can(and usually do) have different impedances as well as the voltage offsets due to this, I just can't seem to simulate a happy circuit. I've found a National app note on this but I can't seem to get their circuit working either, it shows a current source biasing the NI input(the circuit is inverting) but it won't simulate for me on Orcad or Circuitmaker and the breadboard version doesn't do anything either..

Any ideas on another route to try?

Instead of shifting your rail, how about biasing your input signal up with a Wheatstone bridge?

I don't know what parts you have looked at, but I don't see where there should be a significant problem if you have a CF amp with adequate GBW and the proper impedances for the feedback R and input divider R (s).

I don't know what the latest CF low voltage parts are these days, but I suspect there is something out there that would be suitable. Video needs have driven such development maybe?

Since the aformentioned R values will probably be small, you will probably be advised to make your half-supply divider impedance and the input R value to "ground" the same---i.e., if you need 200 ohms to "ground" use a 400-400 divider from +5 to common as the input divider R.

PS: I could roll one tailor-made out of those old Harris HUF arrays, although even in SMD the lead inductances make things a bit challenging.

But there must be something that exists already that is fully integrated.

Risking to be stating the ovbious: if you can't find one that does it all in one go then split the gain over two (or more, there's an optimum number of stages to be found). A dual opamp might still give you a small footprint.

BTW, let's say MHz i.s.o. mhz, I assume you're not meaning milli.

BTW #2, got confused by circuit-simulator yesterday myself: had both p & P prefixes for the node-voltages. As we know p (pico) = 10^-12 but I had to look up that 'P': it's peta (the next step after Tera): 10^15.
Might look like insane values and indeed they were, but it wasn't a real-world circuit, just some model of something. The number-dynamics still a bit high indeed, but hey, that simulator does fine.


Bye,

10^15eter

How flat--1 dB or 0.01 dB..?

As Brad said, there should be no problem in using CFAs with single supply.

Also check the video amplifier section of the manufacturers webpage, perhaps there's one that fits as well.

The inputs can (and usually do) have different impedances as well as the voltage offsets due to this.

Click to expand...

The inputs must have different impedances! And there is no inherent need that this would generate an offset.

Samuel

OK, i've been super busy lately but here is a blurb from an AD appnote that sums up the biggest difference for CFB vs. VFB opamps.

The noninverting input is the high-impedance input of a unity gain buffer, and the inverting input is its low-impedance output terminal. The buffer allows an error current to flow in or out of the inverting input, and the unity gain forces the inverting input to track the noninverting input. The error current is mirrored to a high impedance node, where it is converted to a voltage and buffered at the output. The high-impedance node is a frequency-dependent impedance, Z(s), analogous to the open-loop gain of a voltage feedback amplifier;

Click to expand...

In short the INV input is not really an input at all.. It can be forced to be an "input" but only to modify how the NI input works. This is where things get funky for simulation (for pre-built models). As for offsets on the NI input, the opamp attempts to maintain zero voltage on the NI input so applying a voltage without a constant current causes the opamp to act strangely.

I have more but I don't have time to type more!

It's true the inverting input of a CF amp is low impedance (actually the lower the better). But that is why they call it current feedback---the actual feedback signal from the typically pure resistive feedback and input components (at that input only!) is best understood as a current.

Although the equivalent offset voltage between the two inputs, the non-inverting ~high-Z and the inverting low-Z, tends to be a bit larger in CF amps, that is rarely a limitation in the high freq apps to which they are applied.

If you need noninverting gain of times 10 and a half-supply reference, select a feedback R value per the manufacturer's recommendation, and a 1/9 as big feedback divider R. Then make the actual feedback divider out of two physical resistors in series from V+ to ground, with each equal to 2/9 R feedback. You should be home free if the source is already at 1/2V+.

If not, make a conventional 1/2 supply and a.c. couple to a suitable input d.c. reference R at the n.i. input.

Alternatively, you could make a fancier low-Z 1/2 supply, and run the 1/9R from it as well as the bias R for the n.i. input if required. That would be more likely to give you low overall d.c. offset, but it's a lot more complicated.

There should be no need for biasing with current sources, at least for any CF amps I've seen.

BTW there may also be some voltage-feedback opamps with sufficient GBW if CF still spooks.

If you need noninverting gain of times 10 and a half-supply reference, select a feedback R value per the manufacturer's recommendation, and a 1/9 as big feedback divider R. Then make the actual feedback divider out of two physical resistors in series from V+ to ground, with each equal to 2/9 R feedback. You should be home free if the source is already at 1/2V+.

Click to expand...

Trying this on the bench already!

I'll let you know how it works.

Will an Analog AD8013 work for you? It runs on a single +5V supply. How much gain are you looking for?

If you take a look at the gain flatness chart, page 8, you will see that at a gain of 10, the amp's -3db point sits dead in the center of the frequency range that I need the flatness in. To answer a question posed earlier, I would like this within .5db.

Ok here's some results..

I am using the AD8009 currently(yes pun intended...)..

I can simulate an inverting stage with a vcc/2 offset just like any regular VFB opamp. Orcad and circuitmaker agree on this.

The stage has AC coupling both in and out, 20ohm series, 200ohm feedback, 10k termination before the output coupling due to the offset of the next device's input.

Without going into the lengthy testing details, it doesn't work. I either get a severely deformed output or nothing at all(railed to VCC).

I'm perplexed. All the senior guys say this should work too..

I've even tried setting up the circuit with trimmers so that I can trim in-circuit and I cannot get a happy medium.

I've started from scratch 3 times..

I thought I had a handle on this but apparently I don't!

One thing is clear though, Orcad and circuitmaker model their CFB opamps exactly the same as VFB opamps.

Without building stuff I guess I would produce a circuitmaker model using fast discretes, if ADI has been nice enough to provide one of their "simplified" schematics. It really ought to work.

I guess you could try temporarily a +/- 2.5V rails setup, with lots of close-in high frequency bypassing, and see if it is something bizarre about the bias details.

I see they don't say what is inside.

Are the resistors SMD and really close to the amp? That would worry me too, as any lead inductance is going to be significant at 1GHz where this thing still has gain.

Now you are driving n.i., right? Then you would have actually 22.1 ohms (closest 1%) net to common as the feedback divider for Av = +10, or with the arrangement I suggested, 44.2 ohms from +5 to inv in, 44.2 ohms from inv in to ground. That's maybe not a long-term practical solution as you are burning up 57 mA from the 5V supply, but it should allow getting started. You might have to brute force filter the +5 (maybe ferrite slab and low-ESR/ESL ceramic chip cap?). Then have a second higher Z divider for the n.i. input if you are not d.c. coupling. You will presumably need an impedance match there, actually, for your I assume 75 ohm Z source??

Anyway, it sounds like parasitics rather than a d.c. bias problem. That amplfier is insanely fast---note the different values recommended for the feedback R values depending on package!

I just don't think the part I have will support an inverting setup. I can find nothing in the datasheet regarding this whereas other datasheets for other models do. I'm also weary of the input resistance, 110k for NI input, 8R for INV input.. :roll:

I suppose more testing is in order..

I thought you said you needed a gain of +10.

If it weren't for d.c. offset and limited output swing, etc., you could do inverting 20dB with a single fast transistor---simple common-emitter stage with R-C adjusted in the emitter and collector load R. NXP and others have some...

I misspoke earlier---the Harris arrays (and who knows whose now) were the HFA3046 and HFA3128.

But I still think you are on the right track with an integrated part. Would simply need to know more about the details: input Z, output Z, output signal swing, permissible power consumption (I know the 5V part), acceptable second and third, etc.

The input resistance numbers ARE the input Z, not the division network for gain.

I'm also weary of the input resistance, 110k for NI input, 8R for INV input.. Rolling Eyes

Click to expand...

That's from the datasheet, it's spec'd as input resistances to ground...

My portion of the design needs input Z of 50R, output can be for HiZ(feeding CMOS). The signal swing needs to be a max of 3vP-P and the power consumption should be as low as possible but isn't as important as precision. I need a pretty fast settling time, absolute max of 50ns to .1%. I haven't figured out harmonic content requirements yet, I need to jig something up and measure how much phase noise I get. The bad thing is that this part was highly suggested to me since other folks are designing around it as I type. I have built their circuits and those aren't working..

I guess I should explain what I am doing here, this circuit is gain for the output of a DDS(AD9911)(200mvP-P 50R loaded) to 2vP-P 50Rloaded into the REF port of a PLL, the AD4113. I found that during testing the difference between a 200mv(-10.5dbm) signal and a 2v(8dbm) signal is about 10dbm of phase noise at 10khz from the carrier. I also found a fairly linear region of the DDS output around 25-60mhz which also gives me the range to sweep the REF port in order for the PLL to give me the full range of voltage for VCO control.

That's right, I'm locking the PLL and sweeping the REF to get a smooth linear sweep of the CW!

It's working so far but I need to get more phase noise out of the system before I can finalize the design.

Anyway, I built up the NI gain circuit(straight from the datasheet) on a PCB today and I have a strange triangle output regardless of input Z/amplitude and regardless of output Z/gain.

I'm going to try some different amplifiers.

I wish Atlanta were an easy drive :grin:

But I think I understand your needs now at least: d.c. accuracy is important (?), gain needs to be flat, noise low. Input Z is a nice resistive 50 ohms and output Z doesn't matter since it is coupled intimately to a small pure capacitance. Polarity is inverting (?).

yes on most of those!

seems almost impossible eh? Fun so far even though I've been pulling my hair out.

While I'm technically driving CMOS, the AD4113's REF input generates an offset and therefor needs to be AC coupled per the datasheet. I haven't figured out how to use this to my advantage so the AC coupling is staying for now. That's the only hitch to driving it.

I haven't figured out how flat the gain has to be yet but I expect I should stay well within .5-1db so that I don't have to change any PLL settings on the fly. I'm more concerned with settling time so that we don't get any lag during fast transitions. There will be times where I have to tune from one end of the RF spectrum to the other within 2ms. Right now, the PLL/DDS/VCO combination is sweeping full spectrum under 160us! The DDS can actually go much faster but that is the limit I have found for the PLL loop.

Anyway, I feel like a chump not being able to get any incarnation of my AD8009 protos working..

If you had a 1GHz parasitic would you be able to see it directly?

So you are just wanting to reduce phase noise with a presumptively quieter x10 amp, so you can operate the AD9911 in a favorable region. It sounds like you can couple the AD9911 directly because it sits at 2.5V? Then you need the output of the amplifier to sit in the middle for the sake of voltage swing, but as it is a.c. coupled to the PLL you don't care about d.c. precision.

And you don't really care about polarity of the gain.

It sounds like you could actually a.c. couple the output of the AD9911 if it helped reduce circuit complexity.