Electronics > Metrology

Repair Datron/Wavetek 4910 Voltage Standard

(1/4) > >>

zlymex:
Hello fellow voltnuts, I had a very rare and interesting repair example that I'd like to share.
We(Chinese volt-nuts) once became very interested in Datron/Wavetek 4910 and one of us(who's nick name is grn for short) in Beijing bought a partially bad 4910 as shown by this seller's photo.

The thoughts was that the channel 4 may be a simple failure such as power supply. Even the 4th channel cannot be repaired, it can be used as a perfect tool to reverse engineering the PWM part and the rest of three channels are still make it a very good reference. He sent it to another one of us in Guangzhou for repair and the problem confirmed:

It was quickly find out that the 12V power supply of channel 4 was damaged(no output). After replacement of some parts, it seemed problem solved:

However, when he measured the output by 3458A/1271 and also against his own 10V standard, the drift/noise was much larger than expected. He cannot figured it out and sent to me for examination. I hooked it up to my own system(2-pole-by-17-throw DIY scanner https://www.eevblog.com/forum/projects/diy-low-thermal-emf-switchscanner-for-comparisons-of-voltage-and-resistor-stand/msg610769/ plus 3458A and my other 10Vs) and here is the result of 24 hrs test:

Measured by my 3458A in 2 second interval(NPLC=50), one cycle to read all these 17 references takes 17*2=34 seconds. There are more than 2500 data points for every reference.
Left vertical axis represents the temperature in degree C and the right vertical axis is the voltage in V.
The purple line is the room temperature(sometimes I deliberately alter it by air conditioning or open up windows)
For each set of 17 voltage data, every data is first divided by the measured averaged output of my 4910(therefore it not shown on the chart) and multiply by 10. In this way, the drift and error of the 3458A is largely eliminated.
The figure of 10 may be slightly adjusted for different references to show them on the same condensed chart, but is a constant for that reference. I don't care about absolute values at this time. Also, when the measured reference is an 7V standard, I'll multiply by its nominal value(say 7.09876). In this way, it will show the relative performance all at 1ppm-per-horizontal-line level.
It's obvious shown at the bottom 5 lines that grn's 4910 is not stable at all.

To my surprise when I looked it over, the case bottom unexpectedly deformed and shorted to two screws A and B which is internally connected to the common Low of all the output terminals. This was probably caused by ill handling during packaging/transportation. I regarded this as a design flaw because it has no insulation inside that portion of the case for ventilation, the two screws should be connected internally to case/ground rather than Lo. Guys if you have an 4910, check it out.

After the re-formation and insulation, I made another test of more than 24 hrs:

It seemed promising that almost every lines back to its track except channel 4 that apparently temperature related. Further calculation shows that it had a -0.63 ppm/C temp-co which was much larger than specified. I then located the internal 7V point and soldered a wire outside for measurement, here is the test result:

The 7V(marked by grn-7V) is perfect while the corresponding 10V is drifting with temperature, the problem must lies in the PWM section.
Looking at the servicing manual, there are two power supplies for the PWM circled with red and blue respectively, and the red 10V(A) is somewhat buffered from blue 10V(B):

Again to my surprise, 10V(A) measured at 11.24V and changing all the time. I found out that D301 was shorted. Replaced it with three 1N4148 in series and the 10V(A) remained the same at 11.24V!
I soldered another two wires to pin3 and pin6 of U302, the voltage of pin3 was 10.0V but pin6 was 11.9V. By this time I de-soldered the U302 and that gave me an even bigger surprise: pin3 and pin4 were crossed!

No wonder, the problem is inherent and no one even noticed after all these years. As Dave pointed out in his blog #727, this is the correct way to kill an opamp. Replacing this U302/LM101AH(yes, military grade) and C311 with an AD707AH(industrial grade) is straight forward. C311 is the compensation for LM101 but not necessary for AD707.

Although looked stable, now the output of channel 4 becames 10.00071V, 71ppm higher than nominal. After re-adjust ch4 to the average of the other 3, one more test of 29 hours:

The light purple line is back to its track and so much for the repair. Temp-co is very small beyond my capability.

Standard deviation of channel 4: 0.030ppm, that's great! This is calculated by taking 100 reading, each reading takes 2 seconds sample time and 32 seconds interval. This may regarded as the hint for low frequency rms noise(of the reference and 3458A combined). As a comparison, standard deviation of the averaged output is measured 0.025ppm, one of my 732B is also measured 0.025ppm, noise of one channel of 4910 is specified as 0.04ppm, 3458A is specified as 0.014ppm(NPLC=50).
Here is the photo of that 4910 being calibrated at NIM(looks lousy but is the top in China), It lives happily ever after. ^-^

Hypothesis:
1. When this 4910 was assembled in the factory, pin3 and pin4 of U302 was ill positioned and soldered on to the board
2. When powered on and adjusted to 10V at the factory, this channel 4 performed good regardless 10V(A)=11.24V either because Datron/Wavetek did not make extended test or the room temperature remains constant so they did not find out the problem.
3. When this unit shipped to the customer and powered up for a long time, there exist a quick path for current to go through: from 12V(B), through D301 and D302, pin3 to pin4(actually is pin4 to pin3, input protection diode) to the ground. Soon the D302 was overheated and shorted, This make the power supply overload and damaged some components resulting no output at all for ch4.
4. When my friend repaired the power supply, it burn that U302 down(pin4 through pin3 were opened).

btw, D301, D302 and R307 are used for starting up of the buffer and hence starting up the whole PWM section.

5. When I replaced D302, there are four diodes connected in series with -100ppm/C tempco for 10V(A) and resulting its PWM output with -0.63ppm/C temp-co.

Sorry about my poor English, not my mother tongue. Here is the specification of 4910:

TiN:
:-+ Highly appreciate you posting in here, as us, unfortunate people who cannot read chinese miss lot of great discussion for volt-nut topics over bbs.38hot.net, especially your countless threads. Me myself learnt lot of things just from looking at photos in those threads. Wish only to see more.

P.S. you can insert images in post directly by using [ img]URL[/ img] with square brackets tag ;)

P.P.S. perhaps we can see thread about 600A current source build one day?

zlymex:

--- Quote from: TiN on March 03, 2016, 04:35:46 am --- :-+ Highly appreciate you posting in here, as us, unfortunate people who cannot read chinese miss lot of great discussion for volt-nut topics over bbs.38hot.net, especially your countless threads. Me myself learnt lot of things just from looking at photos in those threads. Wish only to see more.

P.S. you can insert images in post directly by using [ img]URL[/ img] with square brackets tag ;)

P.P.S. perhaps we can see thread about 600A current source build one day?

--- End quote ---
Thanks very much for the compliment and encouragement, I have edited the images and now they are okay.

I'll for sure post more here. As for that 600A, progress is very slow, actually stopped for sometime. However, I'll start posting by small current sources such as 1mA, 100mA and gradually build up. :)

Vgkid:
Thanks for the post, very informative.
The smaller curent source will be very interesting.

quarks:
Great post and very surprising failure.
Thanks a lot for sharing.