Hello all, this is my first post.
The classic Datron 4000 did employ multiple series connected zeners in a complex 4S 2P (two parallel connected banks of four series zeners), but these were standard 1N829A low temperature coefficient zener diodes.
Unfortunately modern buried-zener references are NOT simple zeners, but rather integrated transistor-zener devices designed to lower the dynamic output impedance and improve temperature stability. In the the AD1399 this has been carried to such an extreme degree, that I doubt that you will be able to get away with connecting even 2 in series without oscillation. Even if such a series chain doesn't oscillate, the loop stability issues could still cause gain peaking such that the noise gets worse instead of better.
So let's go back over some of the other options ...
Other than wiring references in series, there are basically three commonly used methods to scale reference voltages:
1) Integer ratio scaling using a switched-capacitor charge transfer device like the LTC1043.
2) Arbitrary ratio scaling using expensive high-stability ultra-precision resistors.
3) Arbitrary ratio scaling using Pulse Width Modulation.
Each of these has advantages and disadvantages:
1) The LTC1043 is fairly low cost, and has amazingly stable performance, but is limited to integer ratios.
2) Precision resistors are a proven technology, but the needed high-precision wire-wound or metal-foil resistors are expensive and may have long lead times.
3) PWM circuitry can give good performance, but there can still be some slight temperature related drift as switching rise times and RDSon change over temperature. Also, most micro-controller PWM blocks lack the needed resolution to achieve PPM levels of voltage resolution without resorting to high-range, low-range, hacks that can add additional error sources and drift.
In the last few months I have been tinkering with an alternate fourth option that I think has a lot of promise because it has basically ALL of the advantages listed above and none of the disadvantages. Like Andreas, I am rather enamored with the LTC1043, and stocked up a while back when they were only a few dollars each, and I am tickled to find that there is a way to apply them not only for integer multiplication, but ALSO in ultra-high-precision NON-INTEGER-MULTIPLICATION.
So without further adieu...
4) This fourth option is based on running a pair of LTC1043 switched capacitor devices in a variation of the "Frequency-Controlled Gain Amplifier" circuit shown on page 14 of the current LTC1043 datasheet. (Note: Two LTC1043 are required because, although a single LTC1043 has two sections, they can not be independently clocked).
What makes this circuit truly ideal is that it works by charge transfer, with the charges transferred being directly proportional to the frequency ratio which can be VERY precisely controlled. Drift in the capacitors can also effect the output voltage, but this is easily managed. To achieve PPM levels of accuracy all we have to do is make the two LTC1043 charge transfer capacitors identical low temperature-coefficient, low-leakage parts (polystyrene 0.01uf).
I have the circuit wired for the non-inverting configuration (as noted below the schematic), which is really nice because it allows simple self-calibration to manage initial capacitor mismatch error (and future re-calibration to cancel any long term differential drift of capacitor values).
The self-calibration uses the second precision zero drift op amp in the ADA4522-2 as a comparator (I know, op-amps are not recommended as comparators, but it works just fine with the 10k resistors recommended in the datasheet). The calibration circuit is simply this comparator wired across between the input reference voltage input and the multiplied output voltage output. Then to calibrate the Frequency-to-Gain factor, we slowly sweep the frequency-ratio through the range which should give unity gain and note the precise frequency ratio at which the comparator toggles. Then to reset the calibration, we just adjust the applied in-service ratio as needed based on the unity-gain ratio - simple!
This means that ten years from now, even if one of the capacitors were to drift up by 10%, and the other down by 10% (causing a huge uncorrected error of 20% !), simply running the above calibration would still instantly restore the corrected post-calibrated frequency-to-gain transfer function error to less than one PPM!
I don't want to post my prototype schematic because it's an ugly lashed up proof-of-concept kludge based on a AD9854 DDS running off an internally multiplied 30MHz source, with a discrete 12 bit ttl divider chain from the same 30MHz clock giving the other frequency (7324.21875 Hz). The varible DDS frequency and fixed 7.32kHz frequency feed through level shifters into the two LTC1043's.
The DDS works well, and gives me parts-per-billion frequency ratio tweaking capability (corresponding to nanovolt voltage resolution), but it's overkill in this application and I am working to replace it with a simple $4 Si5351 clock generator module.
I would be interested to hear from Andreas and some of the other forum regulars to see if anyone has tried this approach before.
Y.A.R.E
Home
Help
Search
Login
Register
