ad736 input stage calibration

[kokodin]

Hello
the project i was working on for a couple of days is an true rms to dc converter atachment for my meter
it is basically the stock battery operation ad736 aplication with additional opamp on the output for a custom input compensation
in short i put 9 V into voltage devider, i get out 300mV that gows trough rms converter i got 300mV out and then i increase it to 900mV via opamp for a easy readout on any cheap meter

With dc it works fine, with small frequency it matches my ut139c.
i made small 555 based square wave generator with range from ~ 700hz-8kHz and 10-100khz and begin to calibrate my converter
But i can't figure it out
When i calculate rms value on paper it is smaller than what i get out and i can't calibrate devider down anymore, at least not linear
when i get flat response it is about 15% bigger than what i have calculated on paper, and i probed the signal with my cheap analog scope to see if the aplitude changes with frequency (it isn't) it reads dc voltage just fine (well not spot on but close, there is little offset depending on polarity) I still might change trimmer caps for bigger or smaller but i rather not

So i don't know what i am doing wrong

I was going to check with it the voltage of my crt tv heater filement, but when i calibrate it crazy way it can read up to 20 volts or as low as 2
is there any fool proof way to calibrate something like that?
i attach my input devider network schematic along with measured coax cable capacity ad736 is feed from that trough 47k resistor

[kokodin]

ok i think i got it
after thinking about it i powered up my generator from two different constant voltage suplys
one was 6VDC
2nd one was 9VDC
i also calibrated my scope by fine adjustment to display each voltage propertly (3 , 4.5 , 6 , 7.5 , 9 , and 12VDC) and calibrated probe at x10 setting to have no bumps on the square wave
then i count peak to peak value on the screen , which is hard to do on old portable analog scope with 3 inch screen :] i also cheked for duty cycle with my ut139c via a capicitor. (it was 49,5 to 50,5 square wave) i added another capicitor to slow down the generator to 350hz but even then ut139c wasn't much help as a true rms meter. since my waveform was a square wave, and i am only interested in ac value i devided Vpp value in half, and since it is square wave that should be it for rms value, right?
then i tuned my converter along full bandwidth to get consistant values, while monitoring the generated form on the scope
Since the waveform was almost constant in height for both generators (350-8000HZ and 11-90khz) along full bandwidth i tuned both trim caps to match calculated rms value along that bandwidth, Since i previously tuned the dc i was almost sure it would remind accurate if the input voltage change, so i checked it again with lover voltage square wave, and it was reading fine.

I have no idea how accurate my measurments are, but for my tv ideal heater voltage should be around 6,3V rms +-5% and i measured 6,67-6,7V with my meter whitch is about 6-7% too much but i also made the screen geometry different and the crt itself was replaced, so i am surprised how close it is actually. And i know for a fact thet they are suposed to be higher than normal, comparing to different tv with identical crt screen, i did not do much damage to yet.

I also learn, that there is no point in calibrating a meter basing on a battery operated generator, because with battery discharge, the voltage output drops before you finish the adjustments.

[Kleinstein]

The divider is rather high impedance. This could cause trouble, as the input capacitance of semiconductor devices can be nonlinear and thus voltage dependent. It might already be enough to increase the capacitance at the divider and reduce / eliminate the 90 pF at the input.

Normally one would adjust the compensation caps at such a combined divider just like one does it with the scope probe: apply a clean square wave and look for an undistorted waveform after the divider. So no RMS converter involved yet.

[David Hess]

I have calibrated RMS converters this way using a precision square wave but it can be tricky.

Make sure the duty cycle is 50%.  It may be necessary to include a duty cycle feedback loop where the average output is compared to half of the peak-to-peak output voltage.

If you have a high impedance input divider, then it needs to be frequency compensated which probably means measuring its buffered output with an oscilloscope.

[kokodin]

90pf is the coax cable itself, one meter of 50ohm wifi antena cable to be exact. i have used it just because this is a purpouse build thing for crt tv measurments, and those things are a little bit noisy. therefore i used the best shielded concentric cable i had to isolate converter from any unwanted interference (deflection coils, switching power suply from the 80's, or high voltage) Basically converter sits 1 meter away from tv and cable goes at 90 degree to the set, so coax might not be required, but given high inmedance input i would rather avoid any noise on the same frequency from the deflection coils.

Measured ac form is a triangular peak with some flat after it, with no dc offset, around 35Vpp (peak to flat) at 15,6khz frequency if i remember corectly
(efective  ideal 700mA at 6,3V RMS, but this varies with screen geometry, ac line voltage, and set output voltages from the switching power suply, so +-5% is a ballpark 6-6,65V for full deflection pal picture )

As i mention before my scope is around 30years old semi-portable (can be powered from 12V a lead-acid battery) analog thing, with around 10MHz bandwidth and 3 inch soviet union screen. so my voltage readouts on that thing are only as accurate as a reference to known to be accurate value in dc coupling mode. Duty cycle of my generator was also confirmed by counting the divisions on the scope in 10x magnyfied view. Calculations were done in spreadsheet program, and then compared to the actual output of the device. then both adjustable caps were used to get a flat response, consistent with what i got on the scope screen and calculations done from it.

the only problem with this method is , i have no way of verifying it , other than by comparsion to not tampored with tv set with identical crt, and similar design (which i also have)

if that is any indication of properly working converter, the readout stays consistent in both polarisations ac and dc coupling modes
well i will test my other tv's and adjust that voltage acording to that reference. I replaced an old crt in this tv and it looked like the old one was overvolted for the last bit of brightness so i expected higher voltage.
 i can also check what is recomended voltage for my other smaller working set and check this out for reference., and there is another one, more modern crt tv on the attic i can check that one too. so i probably will be fine with accuracy under that 5% error.

I only freeked out when i initilly calibrated it with the scope connected to the output of the divider (and efectively connecting another divider to it in series with the probe) i got flat output on the screen of the scope full bandwidth , but when it comes to measurments it showwed around 20V. Because the methodology was incorect. i belive i got it better this time.

edit*****
the other tv with identical screen tube shows voltage arond 6,12-6,34V depending on a signal type and brightness (for pal signals, ntsc signal shows 6,09V which is not surprising considering less return lines and different screen height- the things that this setup feeds off ), so yeah, it should show this much, and it does. There is also constant 0,04V error, so i should subtract that from the result, and we are spot on.