-
Does something like this exist? (wideband unity-gain buffer)
Posted by
Moon Winx
on 06 May, 2017 14:42
-
Hi,
I work as an engineer who designs/develops/maintains measurement systems in a primary standards lab with a focus on DC & low frequency parameters. Many times I've came across a need to measure a signal in a circuit without affecting the signal, an impossible task, but I can't help but think there has to be better instruments to do this than what's currently commercially available. There are high input impedance active probes built for oscilloscope applications and unity-gain buffer amplifiers built for general applications, but nothing I've found is what I would call "metrology grade": some device that Vout = Vin to within 1 ppm from DC to 50 MHz or so. So, does something like this exist? Would it be impossible to build? Is the technology just not there or is there just a lack of applications/demand to produce it?
-
#1 Reply
Posted by
rob77
on 06 May, 2017 14:48
-
DC to 50Mhz within 1ppm accuracy ? i would call that mission impossible
-
#2 Reply
Posted by
Andreas
on 06 May, 2017 14:58
-
He shurely mixed up dB with ppm
With best regards
Andreas
-
#3 Reply
Posted by
TiN
on 06 May, 2017 15:07
-
Thermal voltage converters are closest to metrology, but those are tricky to use, and not high impedance. Also 1ppm over 50M band is quite sci-fi call. 10 ppm - doable, just expensive.
-
#4 Reply
Posted by
chris_11
on 06 May, 2017 15:59
-
I can understand that you look for 1ppm at DC but at any frequency above 10Hz 0.1% or even 1% and at higher frequencies 0.1-1dB would do for signal intergrity. If you are interested in nonlineariy at frequency you go with spectral analysis.
You can easily build a combo amplifier with an AZ (LTC2057 style) plus a wideband AC pass amplifier. There are some apps from Jim Williams (i.e. AN21) if you need help on those.
-
#5 Reply
Posted by
ap
on 06 May, 2017 16:10
-
The problem with wideband (50MHz) it that you have to cope with input impedance/reflection/cabling issues, while at DC other properties are important. So you will have to drive low impedance loads. Also depends on our setup. For your purposes, the question is if you even need one item that combines this properties in one, i.e. what is the real application.
One implementation that had to, and that tried to solve the problem (with some parts no longer available), are the amplifiers in the Fluke 792 for the low ranges. Have a look at the schematics there. With some refinement with today's parts, some improvement may certainly be achieved.
-
-
I would have referred to the LTC papers like AN21, AN47 etc. too.
Another look at the difficulties of achieving such specs is offered by horizontal amplifier/divider circuits of good analogue oscilloscopes. They use very complex networks for piecewise compensation and are yet not even close to the specs you gave. Nuehrmanns 'Standardschaltungen der Industrie-Elektronik' from 1979 presents some examples with extensive explanations. Tektronix published some papers on such designs too.
The model 5113 preamplifier from Ametek (formerly EG&G) is certainly a extremely elaborate piece of hardware, but it is limited to DC to 1MHz and there is no mention of 1ppm.
It is certainly not because a lack of applications if such a thing doesn't exist as a instrumen, not to mention a building block or component. A lot of problems might have been solved differently if it were available. Any actual techlology attempting such feats is using signal processing and digital/SW-based compensation, so what you get, is a quite complex system and certainly not something that might be called a buffer.
But if I'm wrong, PLEASE write me up for a couple!
-
#7 Reply
Posted by
babysitter
on 06 May, 2017 19:14
-
In-circuit is hard, but you might take a single part of a DUT and embed it into two "matching" circuits on the ouput and input side. Considering power levels, Use your metrology-grade RF source, let it fire into the input matching circuit, thru your DUT, the output-side circuit and use a really good e.g. rf level meter (e.g. two dummy loads with a temperature sensor bridge, supposing that equal effective voltage gives equal temperature, or look into bolometers, whatever, or a VNA/SNA with suitable specs), then remove the embedded DUT and see what RF behaviour you get. You might be closer to your goal then.
BR
Babysitter
-
#8 Reply
Posted by
Moon Winx
on 06 May, 2017 23:28
-
Thanks for the informative replies. This gives me a good idea on the practicality and complexity of trying to develop such a device and I'll look at the app notes mentioned. 1 ppm is a goal, but 10 ppm through 1 MHz is acceptable, and above that linearly rising to 500 ppm @ 50 MHz.
-
-
Strays (L & C) make those numbers impossible unless you have a carefully defined mechanical system (shielding, routing & materials) and can say, "under such and such conditions the accuracy is..."
-
#10 Reply
Posted by
chris_11
on 07 May, 2017 07:22
-
Who on earth needs 10ppm @ 1MHz ? That's 0.000087 dB. A RG58 coax will reach this damping @ 1MHz at a length of 4.2mm. Good luck with that.
-
#11 Reply
Posted by
CalMachine
on 07 May, 2017 12:02
-
Who on earth needs 10ppm @ 1MHz ? That's 0.000087 dB. A RG58 coax will reach this damping @ 1MHz at a length of 4.2mm. Good luck with that.
A lot of metrologists, actually. What he is wanting would help cut down the time needed to characterize the AC portion of an 8.5 digit calibrator, tremendously.
-
#12 Reply
Posted by
Moon Winx
on 07 May, 2017 16:48
-
Who on earth needs 10ppm @ 1MHz ? That's 0.000087 dB. A RG58 coax will reach this damping @ 1MHz at a length of 4.2mm. Good luck with that.
A lot of metrologists, actually. What he is wanting would help cut down the time needed to characterize the AC portion of an 8.5 digit calibrator, tremendously.
Yes, you're correct. We have a Josephson arbitrary waveform standard that produces perfect waveforms (on chip @ 4.3 K), but cannot drive any sort of load. Having a extremely accurate buffer would allow the measurement of these waveforms with a variety of instruments without requiring a high input impedance. For this application, 50 MHz bandwidth is way overkill as the waveform generator's useful output is limited to around 100 kHz. As you can probably guess, a Fluke 792A is perfect for measuring this, but we are now generating waveforms with greater amplitude than the amplified ranges of the 792A can safely handle.
-
#13 Reply
Posted by
CalMachine
on 07 May, 2017 16:52
-
Who on earth needs 10ppm @ 1MHz ? That's 0.000087 dB. A RG58 coax will reach this damping @ 1MHz at a length of 4.2mm. Good luck with that.
A lot of metrologists, actually. What he is wanting would help cut down the time needed to characterize the AC portion of an 8.5 digit calibrator, tremendously.
Yes, you're correct. We have a Josephson arbitrary waveform standard that produces perfect waveforms (on chip @ 4.3 K), but cannot drive any sort of load. Having a extremely accurate buffer would allow the measurement of these waveforms with a variety of instruments without requiring a high input impedance. For this application, 50 MHz bandwidth is way overkill as the waveform generator's useful output is limited to around 100 kHz. As you can probably guess, a Fluke 792A is perfect for measuring this, but we are now generating waveforms with greater amplitude than the amplified ranges of the 792A can safely handle.
Oh wow! That's awesome to have a JAWG. How large of signals are you generating?
-
#14 Reply
Posted by
daqq
on 07 May, 2017 17:15
-
(note: I am inexperienced in these areas, so sorry if I'm speaking nonsense)
If the goal is to amplify the output of a high impedance known waveform, would it not be possible to use the high impedance output as a sort of a master reference and then compare it to a signal you are generating by some other generator and automatically adjust the generator when adjustments are needed?
You would no longer need an amplifier, 'just' a difference amplifier/comparator or some other clever arrangement.
Sort of like a PLL, only for more parameters?
-
#15 Reply
Posted by
Moon Winx
on 07 May, 2017 17:19
-
Oh wow! That's awesome to have a JAWG. How large of signals are you generating?
I'm not going to lie, it is awesome to be able to play around with it. I think I'm probably the only one in the world whose work is focused on the application side of it rather than the development of the standard itself, meaning I didn't have to earn a Ph.D. in physics to work on the project.
I can produce good waveforms up to 1 V (rms) DC & 1 Hz to about 100 kHz. The group that developed the system have synced two of them together and can produce 2 V rms. The ultimate goal is to reach 10 V so it can replace the DC Josephson standards and one system could define both DC and AC volts.
-
#16 Reply
Posted by
chris_11
on 07 May, 2017 17:53
-
What is a rough figure of your source impedance of the josephson source over frequency?
You might be able to run an AZ amp below and above its own chopper frequency.
-
#17 Reply
Posted by
Moon Winx
on 07 May, 2017 19:51
-
What is a rough figure of your source impedance of the josephson source over frequency?
You might be able to run an AZ amp below and above its own chopper frequency.
I don't want to guess and be wrong, so I'll need to figure it out. I tune an impedance matching pad when producing the higher frequencies and I want to say the resistance is around 73 ohms (the output signal path is a sequence of twisted pair into coax cable), but I need to go back and check, Obviously the source of the signal is superconducting, so the ultimate output impedance will be very low, the input impedance of the measuring device is really high and the matching pad tries to cancel some of the transmission line mismatch. Your idea of using compensation current is a good one, and so good that it is actually used already.
It is a big reason for the recent increase in amplitude capability of the standard.
-
#18 Reply
Posted by
chris_11
on 07 May, 2017 20:17
-
Good that you have a "RF" compatible source impedance. Makes it a lot easier. Those ultra flat gain of 1 (buffer) asks for an enormous amount of bandwidth in the buffer amp and those tend to oscillate if the input impedance goes high at any point. So you have to find an very fast amp which is very small, stable at gain of one and in a combo with an AZ. And takes +/-12 to +/-15V supply which is difficult to find since most RF processes with three digit GHz FT are low voltage. Small because the tau/delay of the output to inverting input connection need to be very short. A mm is long here.
-
#19 Reply
Posted by
ap
on 08 May, 2017 08:59
-
If the upper frequency limit is really only 100kHz there is no need for a FT in the GHz range and also a transmission line is not an issue. At these frequencies it should be possible to do output sensing. Will be tricky too. But will lead to a much better accuracy than characterizing a transmission line.
I would recommend you team up with an FAE team at a chip vendor such as e.g. LT or AD. Am sure they will love that challenge even though they will not sell many chips.
Also, have you talked to the JJA design guys in Germany, they do something similar (AC), I guess you know them anyway so no need for me to say the name here.
-
#20 Reply
Posted by
chris_11
on 08 May, 2017 19:32
-
For a classic dominant pole compensated opamp with 120dB DC gain you would need about 28MHz GBW and with a more realistic 100dB DC Gain (seldom found on high speed opamps) you need a 100MHz GBW to meet a 10ppm gain error @100kHz. Those opamps have internal transistor FT's well in the GHZ range. If we talk about 1MHz multiply those figures by 10. A LT1128 with the help of an AZ should do it. But that is not necessary the most forgiving part in a combo amplifier, but has large Avol and decent GBW and works on +/-15V. Simulation shows (if you can trust the models) that the 10ppm gain error is around 380kHz. LT1028 is not gain +1 stable.
-
-
Also, have you talked to the JJA design guys in Germany, they do something similar (AC), I guess you know them anyway so no need for me to say the name here.
Sorry, could you please be more specific? Web search on JJA +design and JJA +elektronik came out negative!
-
#22 Reply
Posted by
Henrik_V
on 15 Jun, 2017 13:32
-
-
#23 Reply
Posted by
tszaboo
on 15 Jun, 2017 13:50
-
Thanks for the informative replies. This gives me a good idea on the practicality and complexity of trying to develop such a device and I'll look at the app notes mentioned. 1 ppm is a goal, but 10 ppm through 1 MHz is acceptable, and above that linearly rising to 500 ppm @ 50 MHz.
Reality does not work that way. We have instruments with 10M or 50 Ohm input impedance, and we have sources with 0 or 50 Ohm output. A multimeter measuring a reference voltage is 0 ohm output to 10 Meg input. Since 0/10000000 is a small value, therefore the error due to the loading is small. In RF, if your source impedance is not 50 ohm, you get reflections. 50 MHz is RF. So you need a 50 Ohm system. Now, I dont know about you, but I find it hard to to even create a 50 Ohm source impedance, over the frequency range, over temperature, and so on.
-
#24 Reply
Posted by
Moon Winx
on 16 Jun, 2017 04:38
-
Thanks for the informative replies. This gives me a good idea on the practicality and complexity of trying to develop such a device and I'll look at the app notes mentioned. 1 ppm is a goal, but 10 ppm through 1 MHz is acceptable, and above that linearly rising to 500 ppm @ 50 MHz.
Reality does not work that way. We have instruments with 10M or 50 Ohm input impedance, and we have sources with 0 or 50 Ohm output. A multimeter measuring a reference voltage is 0 ohm output to 10 Meg input. Since 0/10000000 is a small value, therefore the error due to the loading is small. In RF, if your source impedance is not 50 ohm, you get reflections. 50 MHz is RF. So you need a 50 Ohm system. Now, I dont know about you, but I find it hard to to even create a 50 Ohm source impedance, over the frequency range, over temperature, and so on.
Say you have a perfect 50 ohm source, perfect 50 ohm transmission line and a perfect 50 ohm tee connected to the end. The tee has two outputs that are the exact same length from the center of the tee. You connect a perfect 50 ohm load to one side, what do you see at the open end of the other side of the tee that is an open circuit? My hope is it would be the same signal being absorbed by the 50 ohm load, but I'm not sure.