Sean, I've been working on a PIC-based logic analyzer that will also
double as a DSO so I've been following this thread closely and have been
evaluating a variety of wideband amps. I've seen a lot of excellent advice
here. I'm looking for a usable bandwidth of 20Mhz since my logic analyzer
will sample around 40Mhz. Like you, I'm building this for myself and I'd
like to share it. This is a spare time project but a lot of low-level work
is comming together.

   I've looked at the MAX4107 that you mentioned. You might also want to
look at the 4100 and 4102 which have less bandwidth, 3ua bias, and are
pin-compatible. In any case, I strongly recommend a +/- 5 to 15V supply and
an FET source follower. Also, a 1M/~20pf input to accommodate common scope
probes.

   There's an old but classic, FET probe circuit with a 10M/8pf input and a
50ohm output in the "Circuit Design Idea Handbook" (1972) from EDN magazine.
It was designed by a (then) Sony engineer intended as an impedance converter
for use with 50ohm inputs. The original bandwidth was 20Mhz but you can
increase that by using an updated FET. The probe required +/- 6V (+/- 5V
would be fine) and was built in a metal `cigar' tube with a 6-32 threaded
stud to accept common probe accessories that are still available.

   Linear Technology has an excellent application note in their 1993 "Linear
Applications Handbook Volume II". This is AN47; "High Speed Amplifier
Techniques. A designer's Companion for Wideband Circuitry". This covers a
wealth of information including the "ABC's of Probes" contributed by
Tektronix. One application is a 100Mhz servo controlled FET input amp with
a 10M/3pf input and 100pa bias current. The example has a gain of 10. It's
based on the LT1223 and is intended for probe and ATE pin amplifier
applications. The LT1223 works fine with flash A/D's. Checking Digi-Key, the
components cost around $15. It requires a +/- 15V supply and Burr-Brown has
an nice self-contained, isolated, +5V in +/-15V out, converter in a 14 pin
DIP for $11 from Digi-Key.

   I've been running SPICE models for several amps and configurations. So
far I've been using Linear and Maxim's models. I recently received the v5.1
update to Electronics Workbench and the results have been fairly close to
vendor data. The Bode plotter comes in real handy here. `Gawd', I use to do
that by hand on graph paper with a slide rule... I'm getting ready to test
the FET probe and Linear's servo FET amp. I'll probably go with a modified
version of the Linear example. I still have'nt decided on an A/D. Which one
are you using? If you want, I can run some simulations based on your
front-end and the A/D input resistance/capacitance.

   I'm sorry you have already committed to a pc-board but I strongly suggest
you keep the option open to modify it even if you have to cut a trace or two
and solder short wires point-to-point. One good option is to use a standard,
small (ie: 1 x 1.5"), protoboard with a pad-per-hole and a ground plane on
one side. These are cheap and you could use gold-plated pc pins/receptacles
(See Digi-Key, pg 107) from the pads where your amp would go to the daughter
board without much problem. This would also allow you to try out other
circuits by plugging-in different daughter boards.

   One important engineering lesson to be learned here is never pass a design
to production until it's been prototyped and thoroughly tested or production
will take you out back and string you up from the highest tree ;-)

   - Tom

At 04:46 PM 1/19/98 -0500, you wrote:
>Hello all,
>
>Thanks again to all who made suggestions about my oscilloscope design. To
>those who asked if any further info or a general description, I will post
>one on my website as soon as I am more certain about how the front end will
>turn out. Just briefly, it consists of an input network with variable
>gain/attenuation, feeding a 40MS/s semi-flash ADC. The output of the ADC is
>grabbed by a single 15ns SRAM chip whose address pins are driven by a pair
>of 8-bit 100MHz 74F579 counter chips. The board has an interface which
>allows an external computer (PIC based) to stop/start conversion runs, and
>read out the data from the ram. The board, when completed, should be
>capable of 5 to 40 MS/s conversion rates, with a 50MHz input bandwidth
>(50MHz just to be safe, although 20MHz is the max.
>BW without aliasing) and it can store up to 64Ksamples at a time before the
>data needs to be read out of the ram.
>
>
>I want to clarify an earlier question that I asked:
>
>Does anyone know of an amplifier (op-amp style) with something
>approximating these specs?
>
>* Less than 10nA input bias current
>* OK to operate off +5 and gnd (not necessarily input range incl. GND, just
>doesn't need more than 5v spread from Vcc to Vee)
>* Less than 10mV input offset voltage
>* Same pin-out as 741 (not that I designed this around the 741, but the
>amp I currently am trying to use has same pin out)
>* 1 dB gain flatness up to 50MHz at Av = +10V/V
>* Voltage feedback
>* Can drive up to 40pF cap load with moderate (< 100 ohm) series resistor
>
>If not, does anyone know of a buffer that can at least provide a flat gain
>of 1 and has the rest of these characteristics?
>
>I know that some have suggested that I read various documents on scope
>input design, and I thank you for you suggestions, and I hope to do this,
>however, I already have a board prototyped for this circuit (I guess I was
>a bit premature, I never saw an op amp with an input bias current greater
>than about 100 nA, so I never thought to check the input bias current spec
>on the op amp I was using, which, by the way, is the only thing wrong with
>the op amp that I am using.) I have not finished soldering in the
>components, though, so I could switch op-amps if there was one which was
>pin-compatible. Also, does anyone know of a way to some how cancel out
>input bias current without having to know the source impedance?
>
>If none of these things can be accomplished, I will finish construction
>with the max input impedance I can get (abt 2K) and then make an adaptor
>that will go between the scope and the probe to buffer the input. This
>wouldn't be so bad, after all, I am just doing this as a learning experience.
>
>Thanks again to all, and I appreciate everyones patience that I am tying up
>the list with something that is only remotely PIC related.
>
>Sean