JFET+DC servo oscilloscope input buffer

[Martinn]

Hi all,

while designing my own (low frequency, AD8065) front end I often came across the classical JFET/DC servo combination. It has been mentioned in this forum a number of times, e.g.
https://www.eevblog.com/forum/projects/diy-oscilloscope-(yet-again)/msg2783246/#msg2783246
Picoscope on the other hand seems to use integrated buffers (even replacing a JFET solution): https://www.eevblog.com/forum/testgear/oscilloscope-input-noise-comparison/msg2908904/#msg2908904
I was wondering - why still bother with discrete solutions if there is a ADA4817 (1 GHz BW, 4 nV/rtHz)? Lower cost? Lower noise?
Interestingly, in a teardown (I think Dave) I saw that even my shiny new RTB2004 scope seems to employ the classical LMH6518+JFET buffer (?) frontend (and just one attenuator relay!).

So purely for fun I think I'd like to build a JFET/DC servo buffer and see what performance I can get. There's no point in going further (PGA, ADC, FPGA...) as the effort would get excessive and I already have a really nice 300 MHz scope.
I am looking for design hints: There's some in the LMH6518 datasheet and there are a number of reverse engineered Chinese scope schematics around. What would be a modern JFET choice? Any other points to look out for? As I only have a 300 MHz scope for verification there's little point trying wider bandwidth (OK, I have a spectrum analyzer, but no proper RF signal generator, just a TinySA Ultra). Maybe I'd try to beat the AD8065 in terms of noise (7 nV/rtHz).

Thanks! Martin

[Marco]

You probably know this, but just for bystanders. Check out "Wideband Amplifiers", chapter five has a fair bit to say on them. Margan has it on his website.

[moffy]

The Jim Williams edited book: "The Art and Science of Analog Circuit Design" in Section 2: Signal Conditioning in Oscilloscopes ..., has a detailed 500Mhz front end for an oscilloscope and the design choices and reasons. The author uses the BF996S dual gate mosfet and the Internet Archive has a copy at: https://ia804701.us.archive.org/6/items/fe_The_Art_And_Science_Of_Analog_Circuit_Design_PCB/The_Art_And_Science_Of_Analog_Circuit_Design_PCB.pdf

If nothing else it is an interesting read.

[Martinn]

You probably know this, but just for bystanders. Check out "Wideband Amplifiers", chapter five has a fair bit to say on them. Margan has it on his website.

No, didn't know - thanks! Impressive math usage. First 100 pages Laplace transform introduction. I love math, I just wish I had more time to more deeply work with it. Unfortunately on my daily work, * and / cover 95% of the math complexity (with an occasional sine or sqrt).

The Jim Williams edited book: "The Art and Science of Analog Circuit Design"
If nothing else it is an interesting read.

I have this book, just took it out of the shelf and it has a marker in this exact chapter! Unfortunately the author mostly elaborates solid state attenuators.
As you say, interesting read at least. I also have "Analog Circuits Design", but there is a limited volume of anecdotes one can digest.

[moffy]

I have this book, just took it out of the shelf and it has a marker in this exact chapter! Unfortunately the author mostly elaborates solid state attenuators.
As you say, interesting read at least. I also have "Analog Circuits Design", but there is a limited volume of anecdotes one can digest.

As you say there are only so many anecdotes that one can deal with; "As my grandpappy used to say ....." I forgot.

[MasterT]

Decision depends on adc, most JFET analog wizardy was oriented on analog scope or 8-bits. For 12-bits and up, especially if scope would have FFT option distortion is the primary factor, so good quality /low THD in 20-100 MHz range dictate IC choise. Discrete  JFET likely to have 2-5% THD or so, -30 -40 dBc at the best.

[David Hess]

while designing my own (low frequency, AD8065) front end I often came across the classical JFET/DC servo combination. It has been mentioned in this forum a number of times, e.g.

...

I was wondering - why still bother with discrete solutions if there is a ADA4817 (1 GHz BW, 4 nV/rtHz)? Lower cost? Lower noise?

Noise should be lower with the discrete implementation, but the ADA4817 has surprisingly good noise.  A MOSFET should be noisier than a JFET.

The ADA4817 uses series feedback with an input differential pair, so overload recovery is slower and the dynamic response has more aberrations than a FET follower, although low frequency distortion will be better because of feedback.  To get a cleaner response, bandwidth must be limited to what looks like 400 MHz so the ADA4817 is slower, but getting good performance at 400 MHz or higher is tough no matter what technology is used.  Full power bandwidth of the ADA4817 is much slower although this should not matter, but see below.

Note that JFET and MOSFET oscilloscope inputs are typically limited to 500 MHz because of fixture input capacitance, and the same fixture limitations will apply to the ADA4817.  Raw performance of a JFET input is reflected in active probes which achieve 1+ GHz.

Interestingly, in a teardown (I think Dave) I saw that even my shiny new RTB2004 scope seems to employ the classical LMH6518+JFET buffer (?) frontend (and just one attenuator relay!).

If only one input attenuator is used, then the slew rate and full power bandwidth requirements are proportionally increased, by 10 times in a modern (1970s and later) design.  With 2 attenuators the font end only has to handle a single range of 0.5 volts peak-to-peak (50 millivolts per division), but with one input attenuator it has to swing 5 volts which makes a huge difference with great demands on slew rate and full power bandwidth.

This is why oscilloscopes like the Rigol DS1000Z series have bandwidth which varies with signal level.  They have insufficient slew rate and full power bandwidth to support the higher signal levels from using one input attenuator at their specified bandwidth.

[Martinn]

Thanks for the interesting explanations!

If only one input attenuator is used, then the slew rate and full power bandwidth requirements are proportionally increased, by 10 times in a modern (1970s and later) design.

Interesting definition of modern!
Not that I'm an expert, but I have the impression that most current front-ends have one relay max. Together with a PGA like LMH6518 this seems to be sufficient. With an oscilloscope the 10:1 probes always help to extend input range, so you get away with a lower input range. For my 10 MHz frontend I don't support passive probes (not necessary for this frequency range and constant impedance calibration of the attenuators seems messy). Instead I used a 10x and 100x attenuator, giving me 1/10/100/1000 attenuation.

Anyway I wonder how useful >500 MHz high impedance frontends are. From my experience, measurements with passive probes over a few 100 MHz get very difficult and it's much more accurate to use active FET probes (fantastic: Tek P6245).
I also noticed that in the 7000 line there is only a single (7104) mainframe with 1 GHz bandwidth. Maybe only the nuclear weapons developers needed this speed (with single pulse visibility)?

[David Hess]

If only one input attenuator is used, then the slew rate and full power bandwidth requirements are proportionally increased, by 10 times in a modern (1970s and later) design.

Interesting definition of modern!

Not that I'm an expert, but I have the impression that most current front-ends have one relay max. Together with a PGA like LMH6518 this seems to be sufficient. With an oscilloscope the 10:1 probes always help to extend input range, so you get away with a lower input range. For my 10 MHz frontend I don't support passive probes (not necessary for this frequency range and constant impedance calibration of the attenuators seems messy). Instead I used a 10x and 100x attenuator, giving me 1/10/100/1000 attenuation.

That was confusing on my part.  I mean modern instruments in the 1970s and later had two 10x input attenuators, but more recent instruments tend to have only one which considerably raises the slew rate and full power bandwidth requirements of their input buffer and amplifiers.

Anyway I wonder how useful >500 MHz high impedance frontends are. From my experience, measurements with passive probes over a few 100 MHz get very difficult and it's much more accurate to use active FET probes (fantastic: Tek P6245).

High impedance passive probes work to 500 MHz if the signal source is of a suitably low impedance and a coaxial connection is made, but an active probe is better.

I also noticed that in the 7000 line there is only a single (7104) mainframe with 1 GHz bandwidth. Maybe only the nuclear weapons developers needed this speed (with single pulse visibility)?

Nuclear weapons testing was a big application for the 7104, but the 7000 series mainframes only had high impedance inputs up to 200 MHz if you exclude the specialized 350 MHz logic analyzer plug-in.