Yokogawa WT1600 Digital Power Meter

[capt bullshot]

Collecting some posts from the TEA thread here, since this unit starts to get "interesting" - probably a teardown, fault analysis and (hopefully) repair will follow:

The Yokogawa WT1600: Typical Yokogawa knobs and button design. If you know their oscilloscopes, this should look familiar. The user manual found by your favourite internet seach engine reveals first rev. from 2001, this should approximate the date when the unit hit the market. It was discontinued ten years later in 2011 according to the Yokogawa website.



The cal sticker is dated 2010, so it'll be interesting to see how accurate this unit still is.



Alas, only one input element is fitted, and no other options.
For all the timenuts, the BNC labelled "EXT CLK" is not a 10MHz reference input, but rather an input to synchronize the unit to e.g. the fundamental frequency of measurement signals. Useful e.g. if you have quickly jumping amplitudes or heavily distorted signals. Normally the unit synchronizes its internal measurements to one of the input elements, U or I - one can select this through the menus.

[capt bullshot]

Some early notes from the "Fun with (Ice)-TEA" dept.:

Trying to test the accuracy vs. bandwidth of the WT1600, starting with the voltage channel.
This thing has a quite impressive bandwidth / accuracy specification, so I wanted to see if I can set up some measurements to show this.
Starting with the Fluke calibrator, I've applied various DC voltages to check the basic DC accuracy. Looks well within spec.

Next thing is AC voltages. The Fluke can do from 50Hz up to 50kHz, at limited voltages (the higher the frequency, the lower the voltage). For 10V AC, covering the range from 50Hz to 50kHz it looks spot on, too.

Now for the more interesting part: I've set up the high speed amp. It can put out 50Vrms @ 1MHz, but unknown accuracy. So I'd need to measure these signals at a reasonable level of trust into my instruments. Turns out, this isn't that easy. The 34401A does a good job up to some 100kHz, and the scope shouldn't have any issues with the bandwidth at least.

So this is the setup I'm experimenting with now:



Fluke 5100B
33120A
34401A
NF 4005
RTB2004

I've got some other duties now, so stay tuned for more results tomorrow or maybe in the evening (CEST).

[capt bullshot]

First results from todays "Fun with (Ice-) TEA dept.":

So I decided to use the 33120A and the amplifier as the signal source, and the RTB2004, HP34401A as meters for some plausibility checks.
The RTB2004 has a pretty nice feature, one can get its live screen onto a computer just by your favourite browser. The WT1600 has an VGA output that was captured with a grabber device. Finally the 34401A ouput was captured through a simple serial terminal.
This way I have all the interesting readings side by side on the computer screen.



I've started measuring at 100Hz, and stopped at 1.5MHz.
Each step I took a screenshot, like these:




Using human eye powered OCR, I've transferred the interesting measurements into a simple plain text file and made a small graph using gnuplot:




Below maybe 200kHz, I've used the 34401A reading to adjust the 33120A output level, above that the cycle RMS reading from the scope. The scope was connected through a standard 10:1 probe, whose gain error at low frequency (100Hz) was adjusted by fine tuning the scale factor. One can see, the probe isn't perfectly compensated, it drops at 1kHz and peaks at 100kHz a bit.
At the second screen shot (taken at 1.5MHz), one can see the amplifier is at its limits and starts distorting the waveform.
The WT1600 current channel shows 0mA at 100Hz, then 9.24mA at 1.5MHz - this is channel to channel crosstalk due to the high frequency.

[capt bullshot]

Next measurement:
Fluke 5101B as source, 10V from DC to 50kHz:





HP34401A behaves as expected, with some dip in the frequency response at low frequencies. Best results around 1kHz.
WT1600 also behaves as expected, especially no (or very little) difference from DC to 50Hz. This is where the all digital signal processing and RMS calculation can show its capabilities. Basically, if you get your DC offset stable, you can adjust the measurement channel using a stable DC reference and it'll show accurate RMS values up to the bandwidth limits of the analog front end.
Interestingly, one can see the values drop until 10kHz, then showing a jump to higher values at 20kHz.

[capt bullshot]

More WT1600 fun: Cough, cough "Husten, wir haben ein Problem" Edition (German dad joke pun)

Now testing the current measurement input, using the Fluke calibrator and the Tek current clamp:



This is the lowest range (10mA), with 10mArms applied:



100mA range:



This is the largest range (5A), 1.999Arms applied:



Doesn't give too much confidence into the measured values, does it?
Basically, the current measurement using the direct (shunt) input is off all over the place, with different offset and gain errors over the ranges. The unit has an BNC input for current measurement, this input looks OK on a first glance.

The block diagram doesn't give too much info of where the range switching happens in particular, so there'll be some teardown and fault analysis activity in the next days.

[capt bullshot]

The block diagram doesn't give too much info of where the range switching happens in particular, so there'll be some teardown and fault analysis activity in the next days.

Hmmm... but if the external BNC is OK, the fault can't be too far the rabbit hole?

For now, I'd guess there are independent ranging circuits for BNC and shunt that aren't drawn in the above picture. Interestingly, measured values are too small  @10mA, while too large @5A range, and there's always an offset. One can "null" the offset, but O.L. message doesn't go away and gain error stays.
Teardown of the input module will reveal the truth (hopefully).
Maybe calibration data are corrupt, but there's no error message - Cannot find any information regarding how to enter calibration / adjustment modes anyway. Holding the shift key while turning the unit on enters the firmware update mode (a 3.5" floppy disk with firmware is required), I might try all the other keys.

[capt bullshot]

As I discovered yesterday, holding the shift key while powering up brings up the software updater:



There's more keys that do "something" when they are pressed while powering on.

A debug monitor (cursor right)



Wonder where to attach the console / debugger, since there's a GPIB connector only at the rear side of the unit.

Initialize (Reset or Display)



None of these revealed anything useful, nor changed the behaviour of the current measurement channel.

[capt bullshot]

So it's teardown (tuesday) time now 

Removing the top and bottom covers show its build quality, but don't give easy access to the electronics boards:




There's a battery, looks healthy at 3.6V. Of course, one never knows when these will fail, since the voltage remains at a stable plateau and drops rapidly at their end of life.

[capt bullshot]

First view on the input board (Element):

They're using top notch stuff only, and the board looks pretty good.
The ADC (LTC1603) is a 16Bit, 250kSps one, the DSP is an Analog Devices Sharc. This is quite a beast, it has the typical ADI DSP architecture and can do floating point arithmetics. Remember, it's not a todays jellybean MCU, but rather a powerful DSP back in the days (Date code on the chip is 2004, but afair they were introduced in the 90's).

[capt bullshot]

Now, let's have a closer look on the current input channel:

There's the shunt (0.1 Ohm, 4wire), mounted on the bottom heatsink. The top heatsink apparently isn't used here, most probably it's required for the 50A input variant.



Some kind of compensation network at the sense side of the shunt. Note the coaxes wound through ferrite cores, helping with common mode suppression.



Here's top and bottom of the input amplifier with the shields removed:





One can clearly see there's been some repair attempt done.

The interesting stuff is the LT1222 and that seven pin SIL device (2SK 389 dual J-Fet). There's an OPA27, too. My bet is on a composite amplifier architecture, with the JFet and LT1222 providing a low noise high frequency path, while the OPA27 does the low frequency / low drift DC job. Too bad there's no schematic or service manual, I'd really like to see this circuit.

For further troubleshooting, I guess I have to find a way to power up and operate this board. Looking at the board-to-backplane connector, an extender is neither an easy job nor easily available.

[capt bullshot]

Just out of curiosity, the voltage input channel:



See the small copper areas above R72, R73? These are opposite to the resistive input voltage divider string and most probably do some frequency response compensation magic here.
The rest of the circuit looks different from the current channel, too. One thing is in common - there's a shift register (BU4094) driven through some slow speed optocouplers controlling all the analog switches and muxes for the range switching.

[capt bullshot]

So, here we go with some insights:

This is a partial schematic I've drawn on the input structure, in particular the switching and protective circuit for the BNC and the shunt.
The BNC enters the schematic top left - signal goes through a resistive divider with some frequency compensation caps.
The unknown (K52 marking) SOT23 devices look and feel like J-Fets used as switches. The OPA27 buffers the downscaled signal.
Then some DG444 analog switches route either the scaled down BNC input or the voltage across the shunt to the main amplifier (to Q101).
The shunt input circuit features some protection diodes plus a not yet investigated additional circuitry along a tap into the protection diodes.



So clearly both signals (BNC and shunt measurement) are routed through the same amplifier / gain switching circuit. The BNC input signal is scaled down to the same level as the shunt signal before input selection takes place.

As all ranges at the BNC input work just fine, it's an easy conclusion that the main amplifier / gain switching is working fine, too. So the troubleshooting can now continue along the area of the protection and the uninvestigated circuitry.

BTW - the 51R in the shunt signal path had been replaced by the former repair attempt. So I'd guess, the unit has seen some serious overcurrent event.
As far as I did test the shunt, this important component still looks OK.

[capt bullshot]

Continue with live probing - at the moment by soldering short wires to interesting points and putting the board back into the unit (I dislike this method).





So what can I see now:
- no influence of that unknown circuitry around the protection diode on the shunt signal (by probing that 51R resistor).
- The JFet / LT1222 amplifier has switchable gain and appears to work in BNC input mode as well as in direct shunt measurement mode
- as for the largest ranges there's no visible signal at this amplifiers input and output, there might be another signal path

Interestingly, while switching the lower ranges (while the signal is routed through this amp), one can see this amplifier behave exactly the same in both input modes. No gain or offset errors visible.
This points to a conclusion that I don't want to draw right now, so I'm trying to find some more useful test points now.

[capt bullshot]

Moving on, next test point is the primary winding of the isolation transformer.

The unit isolates its signals by the means of transformers, as one can see from the block diagram, too. There's three transformers per channel: One does the supply job, next transfers the carrier frequency from the digital circuitry to the front end amplifiers and the third transfers the modulated analog signal to the ADC circuits.

So I'm tapping the output transformer, using two channels of the oscilloscope as a poor mans differential probe. Works nicely here.
On the scope screen the modulated signal, it's a very simple scheme: the signal gets inverted with each half period of the carrier. So there's no DC across the transformer, and demodulation is easily done by an synchronous demodulator.



This is just one screenshot, I did one at each input range and input mode for comparison: This signal is absolutely as expected, switching gain through the ranges, no gain error, no offset error.

Conclusion: The isolated side of the current measurement path works fine, both in BNC and shunt measurement mode.

Still there's some large errors (gain and offset) displayed at the instruments screen when switching to shunt mode. What might be wrong now?
I suspect, there's no difference in the analog signal path on the non-isolated side between BNC and shunt mode. One thing left: Digital calibration constants - without knowing how to access these or how to adjust (calibrate) the instrument I'm at a road block now.

For the brave, one might check the ADC input once more, I'll see if I can find a suitable tap.

Edit: Confirmed, there's a perfect signal at the ADC input, no offset or gain issues. So it's a matter of adjusting / calibrating the unit. I couldn't find any calibration menu, nor description of how to do this through the GPIB, so I'm at the end of this journey for now.
Anyway, depending on the method to calibrate the unit, it can be quite challenging to provide the necessary signals.

[TERRA Operative]

For service mode, try switching on the instrument while pressing display button. After start up, press shift and then the display button.

https://www.eevblog.com/forum/metrology/yokogawa-wt1600-power-meter-calibration-mode/

[capt bullshot]

For service mode, try switching on the instrument while pressing display button. After start up, press shift and then the display button.

https://www.eevblog.com/forum/metrology/yokogawa-wt1600-power-meter-calibration-mode/

Thanks a lot!
This works and gives access to the calibration menu. I shall setup up a suitable cal rig tomorrow and report here.

[TERRA Operative]

No probs As mentioned in the TEA thread, I'll send an email directly to Yokogawa Japan and ask for service manuals etc. 

[capt bullshot]

Good news on well it works now wednesday:

Thanks to the tip by TERRA Operative, I have now access to the calibration menu at the WT1600.
To calibrate (adjust to be correct) the shunt input, one needs a 60Hz current source at various currents from 50mA to 5A. 50mA to 2A (1.99999A indeed) does the Fluke 5100B. For 5A, I went to the basement and fetched this amplifier from a storage shelf, along with the 1mOhm shunt resistor.
Using these, the Tek clamp on CT and two 34401A, I was able transfer the 2A from Fluke to 5A through this amplifier and the 33120A generator. Now the WT1600 shows pretty good results, and nicely agrees with the second Fluke 5101B. This gives at least some confidence into this non-traceable home lab calibration.
I didn't touch the calibration of the voltage and BNC input ranges, since I found them good enough in spec at former tests.

[Ice-Tea]

Coming together rather nicely! 

[capt bullshot]

So, its test it thursday: Update from the "Fun with (Ice-)TEA" dept.

One of the key features of a precision power meter is phase alignment between voltage and current over a wide frequency range.
If you're measuring power at an phase angle near 0°, a small phase deviation between voltage and current will cause large errors in Q (reactive power).
The other way 'round, if you're measuring near 90°, the real power will be near zero while there might be quite large reactive power, and small phase errors will cause large errors in real power (P).

Every half-decent power meter does the real power calculation by multiplying u(t) by i(t) and integrating the result over a multiple of input period time. Calculation of Q might depend on the actual implementation or might not be available at all.

So for my tests, it was the natural choice to conduct the measurements at 90° phase between voltage and current over a large frequency range.

This is the first setup:



The dual channel AFG provides output signals at 90° phase. One channel is fed directly into the voltage path of the instruments. The other is routed through the high speed amplifier and a 50 Ohm load resistor into the current measurement paths. The dummy load resistor provides non-reactive response to a few MHz, it's not that good but good enough for this test.

DUT are the WT1600 and the Clarke-Hess model 2330 power meter.

This is the plot from 10Hz to 800kHz, read out real power of these instruments. Apparent power seen by the meters is about 1.27VA during the tests. Calculating errors is left to those playing along at home as an exercise to the reader, I've attached a text file with the raw values.

[capt bullshot]

Next one, the WT1600 vs. the Voltech PM3300:

[capt bullshot]

Last, but not least: WT1600 vs. Tek A6302 plus AM503 and RTB2004 Math and Measurement features:





Note:
The NF 4005 amplifier has a specified frequency response to 1MHz, so one has to expect more or less relevant phase errors at frequencies above maybe 200kHz. So I don't know wether the -0.7W deviation at the end of the frequency axis (800kHz) is caused by the power meter or by the amplifier / dummy load.
At least all the power meters agree to some degree, so this points to the amplifier causing the phase shift, otherwise the Tek current probe and the scope should have enough bandwidth to not cause a significant error here. I haven't investigated why the phase shift at this combo is into the other direction (positive real power displayed instead of negative).

I've deskewed the current probe vs, the voltage channel, so I'd put some trust into the scope measurement, otherwise it's the only measurement that largely disagrees with the power meters ...
So no idea ATM which one is more correct.



Edit / P.S.: Now that I'm looking at the finished post, it's quite obvious I did the deskewing wrong, leading to a phase lead at the current probe / scope setup. So it's this last measurement that's lest trustworthy.

[capt bullshot]

Rinse and repeat friday:
Did the above measurement again, with proper deskew. Now it looks plausible:



Anyway, let's face it: It's rather difficult to properly test these high bandwidth power meters vs. their specifications (TBH, I can't be bothered to compare my results to the actual specs right now). It looks plausible, and I'd trust Yokogawa to keep their specs, while the Voltech specs might be rather optimistic (at least as far as I remember).

[_Wim_]

Next one, the WT1600 vs. the Voltech PM3300:

Very interesting tests! This makes we want to buy such a high speed amplifier, but prices on Ebay are 

I am not sure how these power meters manage to measure up to 1 MHz. As far as I can see, the Voltech has a max sample rate of 333KHz, and the Yokogawa of only 200kHz. As both units measure voltage and current directly and do all calculations in the digital domain, this would mean they use some sort of equivalent time sampling making use of the repetitive signal. Or do they use another trick?

I have a Voltech PM6000, but never actually did such extensive testing. Maybe I also give this a try in the future when I find some time for it...

[capt bullshot]

Regarding the sampling frequency / bandwidth: There's no trick at all, it just works by the nature of the matter

For calculating the basic values (Urms, Irms, P), the signal frequency isn't of concern at all. One just integrates over an integer multiple of the period time, no matter what the actual signal or sampling frequency is.
It's some kind of ETS, though it's not necessary to reconstruct the waveform as the sampling oscilloscopes do.

For example, in the frequency domain, a 300kHz signal gets aliased to 100kHz if a sampling rate of 200kHz is used. For Urms / Irms / real P computation it doesn't matter wether the original signal @300kHz or the aliased signal @100kHz is seen by the calculation, as long as both (U and I) signals undergo the same conversion and keep their phase relationship.
There's one corner case: a signal frequency of an exact integer multiple (including 1) of the sampling frequency gets aliased to DC, this can lead to large errors or at least confusion. Didn't check this with the Yokogawa, but I've seen instruments avoiding this by changing the sampling frequency slightly (afair the Clarke Hess does so).

The WT1600 has a harmonics analysis mode, too. Here's the whole thing a bit different: As you want to apply (usually) a FFT with rectangular window to the input signal, the sampling frequency gets synchronized to the signal frequency by means of a PLL circuit (At least this is my guess from reading the manual). This mode is quite limited in terms of fundamental frequency, maybe max. 1kHz afair. Usually you want to switch an aliasing filter into the signal path to avoid the aliasing here.