Opamp Recommendations? 3GHz GBW, low input C

[madshaman]

Hi, I'm looking for a low noise opamp with as little input capacitance as possible (who isn't...) that has a GBW of 3GHz.

Any recommendations?

[David Hess]

If you are looking for a 3 GHz GBW, then you are not going to have a lot of options as far as noise and input capacitance.

TI has one but I would not really consider it 3 GHz.

[T3sl4co1l]

Is this, by any chance, more of a buffer thing than an op-amp thing?  You can use op-amps for an awful lot of things, things that you may not need.

Tim

[madshaman]

I need to amplify a signal with a high output impedance, I'd consider buffering before amplification if the buffer's inputs had less capacitance.  I'd prefer needing only one device instead of two, additional noise not-withstanding.

Is this what's being suggested?

I'm open to alternative ideas; I'm not an expert by any means.

[dannyf]

Jfet buffer plus cfb opamp.

hopefully you have exceptional skills in layout or it will be a nightmare for you.

generally, high speed opamps should be avoided.

[David Hess]

Oscilloscope 1 megohm inputs which use a JFET or MOSFET for a buffer are typically good up to 300 to 500 MHz maximum because of input capacitance at which point their input capacitance yields 25 ohms of reactance matching the impedance of the 50 ohm terminated source used for testing:

300 MHz @ 25 ohms = 21 picofarads
500 MHz @ 25 ohms = 12 picofarads

High impedance active probes do better then this by having much lower input capacitance, 1 picofarad or lower, which points to the requirements of a wide bandwidth high input impedance amplifier; use a low capacitance RF JFET or MOSFET source follower followed by a transconductance amplifier.  The modern version of this is exactly what dannyf suggests, an FET buffer followed by a current feedback amplifier.

If good low frequency performance is desired, then the offset voltage and alpha of the buffer will need to be compensated for.

Note that the input impedance varies significantly with frequency dropping to a low value at high frequencies.  If you want a more constant input impedance that can be lower, then an inverting amplifier might be more appropriate.

[Marco]

What kind of amplification do you need? What input capacitance is low enough?

Would say a THS4303 do? (10x fixed gain, 1.6MOhm ll 1pf.)

[madshaman]

What kind of amplification do you need? What input capacitance is low enough?

Would say a THS4303 do? (10x fixed gain, 1.6MOhm ll 1pf.)

Yes, it fits the bill perfectly.  I only need 4-5x gain and was looking to handle signals to a minimum of 300Mhz, but hoping to push it higher.

A tiny bit pricey, but not for what it can do.

Thanks!

[madshaman]

@David, thank-you for taking the time with a very good and concise explanation of this situation in general, and the common solution.

For someone like myself, it's really helpful.

I have built circuits with discrete components before, but I get a bit chicken as my forays creep up into higher frequencies; at least at first.

Also, thanks for bringing my attention to the issue of non-constant input impedance over frequency.

[madshaman]

@dannyf

Thanks for the recommendation which makes perfect sense.

For this particular problem a fixed-gain amp is perfect, but I can definitely agree that a high impedance buffer followed by a cfb amp will get me both a high impedance input AND excellent high frequency amplification.

P.S. I can't say I have remotely exceptional layout skills; part of me masochistically hopes I have some problems ^^'

[T3sl4co1l]

@David, thank-you for taking the time with a very good and concise explanation of this situation in general, and the common solution.

For someone like myself, it's really helpful.

I have built circuits with discrete components before, but I get a bit chicken as my forays creep up into higher frequencies; at least at first.

Ironically, it's the exact opposite problem which you face!

The double irony, I guess, is if you have a "cost no object" project, it doesn't really matter if you use a $1 transistor or a $10 op-amp.  Even a gold plated part for $20+.  So, if you can afford it, or it's a one-off anyway, "good engineering practice" isn't at all a concern.

JFETs are hardly available nowadays anyway, let alone at that frequency; you'd be looking at a GaAsFET or PHEMT, fT > 10GHz let's say.

Tim

[dannyf]

A tiny bit pricey, but not for what it can do.

For this type of projects, having realistic expectations / spec gets you 90% to the finish line.

[David Hess]

JFETs are hardly available nowadays anyway, let alone at that frequency; you'd be looking at a GaAsFET or PHEMT, fT > 10GHz let's say.

JFETs are alive and well (and cheap) from producers like NXP although you need the secret JFET characteristics decoder ring to evaluate their specifications:

http://www.nxp.com/products/rf/transistors/mosfet/jfets/n_channel_junction_field_effect_transistors_for_general_rf_applications/#products

Ft = Gm / 2 pi C is a reasonable estimate and figure of merit in this case.

[madshaman]

@subolg123

Thanks for taking the time to provide your input.

I agree that one should understand what one is doing and one should be able to mathematically model one's circuits.

Please allow me to be frank as well.  I'm not providing exact details of what I'm working on because it's a proprietary project for my own lab; I'm sure as a microwave engineer, you're cobbled with NDAs much like I was [edit: (when I worked for a salary as a software engineer)]; it's no different just because it's my own lab.

I hope those who do independent work and hope to generate income from that work understand where I'm coming from and I'm not rubbing anyone's fur the wrong way.  I'm not suggesting any of my work is particularly stelar, but it is mine.

Please tell me if I'm out of line.

I realise you are trying to help and dedicated your own free time to do it, so please don't worry about offending me.

Re: Montreal.  I live in Ottawa with my wife so we're close to each other.  Do you have a home lab?

[madshaman]

@David, thank-you for taking the time with a very good and concise explanation of this situation in general, and the common solution.

For someone like myself, it's really helpful.

I have built circuits with discrete components before, but I get a bit chicken as my forays creep up into higher frequencies; at least at first.

Ironically, it's the exact opposite problem which you face! :

Please explain, I'm interested.

The double irony, I guess, is if you have a "cost no object" project, it doesn't really matter if you use a $1 transistor or a $10 op-amp.  Even a gold plated part for $20+.  So, if you can afford it, or it's a one-off anyway, "good engineering practice" isn't at all a concern.

JFETs are hardly available nowadays anyway, let alone at that frequency; you'd be looking at a GaAsFET or PHEMT, fT > 10GHz let's say.

Tim

This is for a prototype which I'm hoping to turn into a product.  I'm not sure if this is common practice in the application engineering world (I come from a software engineering background), but I generally try to take the pragmatic: "make the whole thing work correctly first (with as few potential points of failure and as little complexity as possible), then make it better/faster/cheaper/more-efficient while still meeting the functional spec".

I might not always get the: "with as few potential points of failure and as little complexity as possible" part absolutely perfect, but I find this approach works really well for me; especially when I'm doing something new.

[David Hess]

@David, thank-you for taking the time with a very good and concise explanation of this situation in general, and the common solution.

For someone like myself, it's really helpful.

I have built circuits with discrete components before, but I get a bit chicken as my forays creep up into higher frequencies; at least at first.

Ironically, it's the exact opposite problem which you face! :

Please explain, I'm interested.

I think he means that switching to integrated solutions is usually done to avoid the difficulties of using discrete parts.

[T3sl4co1l]

Ah, a good analogy then: an op-amp is like a... Python's libraries let's say.  Powerful, massive, sometimes buggy, sometimes overengineered, often expensive.  The computational expense being memory and speed, though sometimes financial expense as well.

The main downsides to an op-amp are the extra noise (there are some surprisingly low noise op-amps, only not quite as good as single PHEMTs or whatever), bandwidth (in a given range of parts, I suppose) and financial cost.

One example of buggy hardware I've heard of is DSL line drivers.  They're spec'd like normal op-amps or buffers, but only work in a narrow part of the (specified!) operating range.  Like +/-15V supplies with less than 5V working range, or undocumented latchup behaviors.  Analogous to buggy, poorly written, poorly documented libraries.

Tim

[David Hess]

The main downsides to an op-amp are the extra noise (there are some surprisingly low noise op-amps, only not quite as good as single PHEMTs or whatever), bandwidth (in a given range of parts, I suppose) and financial cost.

Even the expensive ones are cheap considering what they do and what they replace.  On the other hand, I would really like to have integrated high performance matched transistor arrays more available.

One example of buggy hardware I've heard of is DSL line drivers.  They're spec'd like normal op-amps or buffers, but only work in a narrow part of the (specified!) operating range.  Like +/-15V supplies with less than 5V working range, or undocumented latchup behaviors.  Analogous to buggy, poorly written, poorly documented libraries.

The ones I am familiar with operate down to DC and act like operational amplifiers with high power output stages.  I am not quite surprised about latchup problems given the hazardous environment of driving a long POTs line though even with transformer isolation.  Interfacing with telephone lines is not for the faint of heart.

[T3sl4co1l]

The main downsides to an op-amp are the extra noise (there are some surprisingly low noise op-amps, only not quite as good as single PHEMTs or whatever), bandwidth (in a given range of parts, I suppose) and financial cost.

Even the expensive ones are cheap considering what they do and what they replace.  On the other hand, I would really like to have integrated high performance matched transistor arrays more available.

Only one that I know of;
http://www.digikey.com/product-detail/en/HFA3046BZ/HFA3046BZ-ND/936233

Tim

[dannyf]

One example of buggy hardware I've heard of is DSL line drivers.  They're spec'd like normal op-amps or buffers, but only work in a narrow part of the (specified!) operating range.  Like +/-15V supplies with less than 5V working range, or undocumented latchup behaviors.

ADSL line drivers are usually CFB opamps with high current capabilities, like ad815. They are exceptionally capable, extremely fast, and a nightmare for your layout engineers.

I have yet to encounter any of the issues you mentioned above. The only issues I have with those chips are poor dc performance, typically low input impedance, and stability, the latter two are design issues rather than device issues.

[David Hess]

One example of buggy hardware I've heard of is DSL line drivers.  They're spec'd like normal op-amps or buffers, but only work in a narrow part of the (specified!) operating range.  Like +/-15V supplies with less than 5V working range, or undocumented latchup behaviors.

ADSL line drivers are usually CFB opamps with high current capabilities, like ad815. They are exceptionally capable, extremely fast, and a nightmare for your layout engineers.

Anything which is both fast and supports high output current is a nightmare for layout designers.   High output current in the form of low impedance exasperates parasitic inductance just like high impedance circuits exasperate parasitic capacitance.  Decoupling is tricky at high currents and high frequencies and doubly so when both are involved.  Many integrated amplifiers like these have measurably degraded performance in through-hole DIP packages compared to surface mount packages because of the extra lead inductance.

I have yet to encounter any of the issues you mentioned above. The only issues I have with those chips are poor dc performance, typically low input impedance, and stability, the latter two are design issues rather than device issues.

It is not surprising that they suffer from relatively low input impedance on their high impedance input.  Low power fast and low noise voltage feedback amplifiers suffer from the same problem.  Current feedback amplifiers are intended to both drive and be driven by low impedance sources.  If you want higher input impedance, then add a buffer.

[T3sl4co1l]

Only rumor, I don't know personally.  Ref:
https://groups.google.com/d/msg/sci.electronics.design/CjJg265gZro/qE7sIxqxxSIJ

Tim

[dannyf]

I don't know how much wait to put on that post.

Not everyone of the THS family is CFB. Some of them are quite interesting. THS6012 (adsl line driver) / TPS6120 (headphone amp). THS4012 is another one: a CFB that is almost a VFB.

[David Hess]

I don't know how much wait to put on that post.

Not everyone of the THS family is CFB. Some of them are quite interesting. THS6012 (adsl line driver) / TPS6120 (headphone amp). THS4012 is another one: a CFB that is almost a VFB.

The only ways the THS4012 resembles a current feedback amplifier are its speed, output capability, and being built on a complementary process.  Internally it is a standard voltage feedback amplifier topology.  Note that its slew rate is relatively slow for its bandwidth.

Linear Technology makes some single stage fast voltage feedback operational amplifiers that are built like a current feedback hybrid amplifier covering the range of 12 MHz 400 V/us (LT1354) to 400 MHz 2500 V/us (LT1818).  Essentially they just add a high input impedance buffer to the low impedance input of a current feedback amplifier.  The CFB resistor which sets the bandwidth is internal.