Fast Sample-and-Hold Circuit >2GHz

[luky315]

I was looking for a way to sample-and hold an input signal (0 to 3.3V) with a bandwidth of at least 2GHz. Back in the days I did something similar with a much bigger bandwidth with an RF Diode Bridge, but that circuit has limitations. Is there a modern way or even a faster Sample-and-Hold IC than for example the AD783? I am looking for something that is in between the AD783 and a custom diode bridge in terms of complexity and performance.

[berke]

You should give some details about what you want to do, because at 2 GHz it's not happening.

Just think of the S/H input.  That's a digital signal.  But you can't naively toggle a CMOS line at 2 GHz.  Clock period is 500 ps.  You have two edges separated by 250 ps.  Let's be generous and say you can spend 1/5th of that on rise/fall times, that's 50 ps.  Say you're using very low voltage 0.7V CMOS.  That's a dV/dt of 14 GV/s.  Even with a tiny stray capacitance of C=3pF you'll be moving I=C dV/dT=42 mA currents, which over 50 ps is 840 MA/s.  Got a 5 nH stray inductance?  That'll be V=-L dI/dt =-4.2 V but the logic threshold is at 0.35V.

Single-ened CMOS control is not working.  That means we're looking for a discrete S&H IC with an aperture of 500 ps and an LVDS or PECL S/H control input.  Who's gonna buy that?  But what's the market for that?  Non-obsolete ADCs come with their own S&H circuits.  What are the remaining applications?  I don't know, my guess is that they will be very niche high-speed analog but not RF applications for which the analog wizards can probably come up with solutions based on high-speed current-mode op-amps, a couple FETs or whatnot.

That being said the fastest discrete S&H I found is the TI OPA615 https://www.ti.com/product/OPA615

The amplifier BW is ~700 MHz however that doesn't mean you can sample any random signal at that frequency, it will depend on the signal levels and the source impedance for example, plus there is the settling time.

[Marco]

Doesn't meet the voltage requirements and the pricing is just silly, but the fastest you can easily buy is hmc1061lc5 AFAICS.

[berke]

Doesn't meet the voltage requirements and the pricing is just silly, but the fastest you can easily buy is hmc1061lc5 AFAICS.

Holy cow.  I guess that's how they make those super-fast scopes?

There is a slightly more affordable one at 4Gsps, the HMC760LC4B which goes for only 434€ at Digikey (they have 1 in stock).

[tggzzz]

Doesn't meet the voltage requirements and the pricing is just silly, but the fastest you can easily buy is hmc1061lc5 AFAICS.

Holy cow.  I guess that's how they make those super-fast scopes?

Er, no. At least not directly.

Consider the Keysight Infiniium UXR–B Series, which go to 256GS/s for 110GHz real-time bandwidth. https://www.keysight.com/us/en/assets/3123-1313/data-sheets/Infiniium-UXR-B-Series-Oscilloscopes.pdf

Not something you would float so that you could "safely" measure the mains (Yes, Virginia, I am using the word "safely" in a non-standard way!)

[berke]

Er, no. At least not directly.

Consider the Keysight Infiniium UXR–B Series, which go to 256GS/s for 110GHz real-time bandwidth. https://www.keysight.com/us/en/assets/3123-1313/data-sheets/Infiniium-UXR-B-Series-Oscilloscopes.pdf

That spec sheet doesn't tell much.  What do they use then, if not multiple "slow" ADCs with fast S/H amps?  Mixers??

[luky315]

From the specs I would look in the region of the older sampling scopes like the Tektronix TDS820 from 30 years ago - But I need just the sampling part with some level adjustment and impedance conversion ("highest reasonable" input impedance and 0 to 3.3V signal with some over - and undershoot)

[David Hess]

A diode bridge sampler or traveling wave gate, which is the transmission line version of a diode bridge, is still how it is done.  There are other circuits which become feasible on a high performance integrated circuit process, but they are not going to be available unless you want to design and fabricate an IC.

If the sampler does not have enough input range, then attenuation will be required.  Is that a problem?

[luky315]

No, I don't want to design an IC, therefore I'm looking for a) an IC to base the circuit arround or b) some proven and working design. I built a lot of sampling bridges back in the days at the university, but the repeatability was very bad - a lot of implementations didn't work well

[Marco]

Holy cow.  I guess that's how they make those super-fast scopes?

By the time you make a thousand scopes, you're probably better off getting some SiGe/InP/GaAs foundry make an ASIC at those prices.

[max-bit]

Building a fast sampling system is not that easy.
Firstly, the input circuits and the probe itself (here only active).
Then you have an amplifier (plus the appropriate circuit dividers that form the signal to the ADC converter.
You can use several transducers with a lower sampling frequency, working with a phase shift (and this is still done today)
Then, the data acquisition and processing system needs to be implemented, and this is unlikely to be done without the use of fast FPGA systems (oscilloscope manufacturers often use ASIC systems).
If you want to know something more, provide the parameters of the signal you want to meas and what parameters you want to achieve in this sampling system.
Oh, sampling oscilloscopes are a different story, such oscilloscopes are good for repetitive signals

[berke]

You can use several transducers with a lower sampling frequency, working with a phase shift (and this is still done today)

Is this done in practice with overlapping aperture windows and deconvolution?  Or are fast samplers a must?

[David Hess]

Oh, sampling oscilloscopes are a different story, such oscilloscopes are good for repetitive signals

Exactly, the sampler does not need to be fast, but it must have high bandwidth.

Bandwidth of the sampler is normally limited by its RC time constant; it takes time to charge the sampling capacitance and this depends on the source impedance.  When the desired bandwidth is higher, then the sampler's aperture time becomes smaller than the RC time constant and the capacitance is only partially charged (low sampling efficiency), but this situation can be handled as well and is how sampling oscilloscopes achieve high bandwidth way beyond what the RC time constant in their sampling system can support.  2 GHz was not that high for this type of circuit even in the 1970s.  Discrete implementations with through-hole parts can easily achieve twice that, but it requires some design expertise.

The bandwidth of the sampler depends primarily on its aperture time, so the problem now becomes how to make a fast sampling strobe with a narrow width.  If you know the bandwidth, then you can calculate the needed aperture time, and the reverse.

[luky315]

I know how to make pulses with a few 100ps, but this is not a simple circuit and it has to be trimmed by hand etc.
There are a few fast S&H ICs out there, for example https://www.fmaxtech.com/fx331c-30ghz-sh.html and a
But as I said I am really looking for something a bit lower end with lower cost.