Valhalla 2701c schematics + firmware V3

[RaymondMack]

I dumped the ROM today and compared with the V3.0 file uploaded earlier. Structurally, they look similar, which is good. And while there are some general similarities, the contents are quite different. Mine has data written up till 0x1485 while the other FW ends at 0x132B. So there is a little bit more written to the ROM.

I'm fairly certain that my unit has a different FW version (not just the IT2 option), so that could contribute to some of the differences.

Does anyone know the date of the instrument using the V3.0 firmware? My unit is somewhere between 1991 and 1993.

I added the UV EPROM's datasheet and a readme file with the firmware. Have fun hacking.

[lowimpedance]

Well I uploaded version 3 as part of the OP from my still to be repaired unit which has the white paint front as well. So probably similar vintage, will have to dig it out of the cupboard to check some date codes.
Thanks for the extra firmware, certainly worth investigating further when I have a working unit.

[Le_Bassiste]

as a side note, upgrading from V1.0 to V3.0 will get you a "no dAtA" error message on power-on, indicating that the cal constants are either missing or corrupted.
in fact, the output voltages will be out of cal after the upgrade. this message goes aways after a full cal/adjust run, though.
fwiw, i'm using an AT28C64B-150 as replacement for the 2764A.

[lowimpedance]

Does anyone know the date of the instrument using the V3.0 firmware? My unit is somewhere between 1991 and 1993.

 Had a look and mine and it has parts ranging from 88 to 94 dates, so approximately the same time period at best guess !.

[RaymondMack]

Thanks for looking. Yours seems to be a year newer, so maybe it has newer firmware? I suppose I could get a AT28C64B EEPROM that Le_Bassiste suggested and see if the V3.0 FW has the IT2 code in it. I'm not sure I would trust leaving a AT28C64B in any equipment where it's used for the FW (seeing how it only has 10 years data retention), but for testing they look pretty neat to have around.

I noticed that you started the thread in 2017! What do you think is wrong with your calibrator?

[lowimpedance]

 The high voltage FET string is damaged , gets rather hot !,  from memory. Isolating the regulator pass transistors TR 6 and 12 allowed me to test the operation of the micro system etc which looks ok.
I will have another go at it soon when I get a few other jobs out of the way, and will post what needed replacing here for general info for others.
 Interesting thought that if version 3 might have all the required code, would be easy to check with those parts already populated. Otherwise now having your firmware it maybe time to investigate the BOM for that option.

[RaymondMack]

Well, after soldering in the new bleeder resistors and radial electrolytic capacitors (of which all had failed in one way or another) the calibrator turned on and started working. DC volts was pretty close to bang on, maybe 10 or 20 counts out from my HP 3456A using some Pomona low thermal banana plug cables connected to the sense terminals. The current source option appeared to be a little noisy and was out of spec by about 400 counts. Visually, the current source circuit looks okay--there isn't any discoloration of the board from overheating or any obvious faults. And again, it appears to work, if a little noisy on the last digit.

While repairing the unit, I noticed some further heat related discoloration near the opto isolators around a TO-92 5V regulator and around the TO-220 +15V regulator a little further away as well as the big TO-3 5V regulator for the digital side. While turned on, all three of these parts get pretty damn hot. Is this normal? Or does this imply I have further issues that need to be resolved? Maybe I should replace the TO-92 with a TO-220 and put heat sinks on the TO-220 +-15V and the soon to be TO-220 5V reg near the optos.

The digital +5V side also had discoloration around the two rectifier diodes, but these now seem to run cool to the touch. I believe the 15000 cap was shorted, so it makes sense they had gotten toasty previously.

After working on this unit, I feel that the engineers at Valhalla had put in as much effort as possible to cut costs when designing this calibrator. At least three voltage regulators are either under specified or lack the appropriate heatsinking required to reliably perform their job. Almost every electrolytic capacitor is too close to high temperature parts while many resistors are under sized and likely not even rated for the applied voltage across them. Then there is the issue of the smaller transformer making faint arcing noises... And the list goes on! I would give these calibrators a C+: They work, but their construction and performance is just at the edge of useable.

[Le_Bassiste]

While repairing the unit, I noticed some further heat related discoloration near the opto isolators around a TO-92 5V regulator and around the TO-220 +5V regulator a little further away as well as the big TO-3 5V regulator for the digital side. While turned on, all three of these parts get pretty damn hot. Is this normal? Or does this imply I have further issues that need to be resolved? Maybe I should replace the TO-92 with a TO-220 and put heat sinks on the TO-220 +-5V and the soon to be TO-220 5V reg near the optos.

The digital +5V side also had discoloration around the two rectifier diodes, but these now seem to run cool to the touch. I believe the 15000 cap was shorted, so it makes sense they had gotten toasty previously.

in my unit,  D212,D213 (5V supply rectifiers), IC201(5V regulator), IC202( +15V regulator) get well beyond 60°C, and there's some discoloration around D212/13. as for IC116 (5V regulator for the PWM isolators), i only can confirm indirectly, as i'm using an SOT223 on my opto-piggyback, which is somewhere in the 45°C range. what worries me are the D212, D213, IC201 cooking away, and i don't know how to tackle that. exchanging the diodes with schottkys would just move the saved losses over to IC201, which is already on the edge. IC201 is by far the hottest running part in my unit. my unit is running on 230 VAC, and i'm thinking about openening the linkage at PL202 and putting a high power chassis-mount resistor in series there to bring the secondary voltages down just a bit. that would hopefully cool off IC201 and then allow for the replacement of the diodes with schottky types.

i agree with your assessment that the entire unit is pretty much "auf kante genäht". funnily enough, the much older FLUKE 343 is running circles around it when it comes to noise and linearity performance. however, i am quite fond of the GPIB programmability and the straight-forward and easy to grasp concept, which makes it a fun playground for understanding precision equipment fundamentals and testing diy improvements. bottom line, though not flawless, its a keeper for sure. is it worth 10 grand? hell, no.

p.s.: as for the output noise, electrole pointed out that this is due to PWM-filtering issues and that the (LF-)noise can be improved by modifying the filter around IC8.
here's what i got after simply increasing R43,44 to 100k each:
(K2002: DC @ 20V, 10 NPLC, no FIL, SDEV after 200 readings.  AC @ 200 mV, no FIL)
setpoint: 10.0V
FLUKE 343A: SDEV = 1.5 uV, VACRMS = 50uV (for comparison, my K2002 has its floor at 35 uV)
VS2701C   : SDEV = 2.85 uV, VACRMS = 0.93 mV(!)

[Electrole]

I suppressed the PWM residuals to some extent in my 2701C, partly by changing the PWM low-pass filter, and partly by inserting a shield between the ends of the resistor pair R38A and R38A.

PWM-filter: I changed both C6 and C7 from 1 µF to 2.2 µF. This should give about 4 times better attenuation at the PWM chopper frequency, but I found to my disappointment that the reduction of the PWM noise was much smaller. I suspect the PWM transients are difficult to tame in the 2701C layout, which does not seem to have received sufficient attention to good layout practices.

R38 shield (check out the pictures): In the 2701C one end of R38A is connected to a node which has full PWM amplitude, and one end of R38B is connected to the feedback node of the chopper amplifier IC5. For some reason Valhalla decided to locate these two ends next to each other. As a result, the PWM transitions couple to IC5. The shield is an attempt to lower this coupling. Again, the improvement by the shielding was not large, about 3 dB, but at least the improvement shows that there is an issue with coupling. The cost of the shield is next to nothing, so I would recommend adding it.

The PWM is only one of the sources of noise in the 2701C, and most noise is caused by the power supply switcher. This is difficult to get around, and real improvements will require another (non-switched) fixed-voltage supply, and getting rid of the 1200 V range, much in line with the LN-option Valhalla offered.

[RaymondMack]

So I just ran a few tests and I think I'll need to do most of these modifications to lower the noise on the output. Using my Racal Dana 5002, I measured 6 mVrms AC ripple on the output at full bandwidth with the Valhalla set to 10V. With the 100kHz filter enabled that dropped to 0.441 mVrms, so there is a lot of HF content in the noise. Flipping the isolate switch to earth ground, the noise jumps to 7.3 mVrms. Using my HP 3400B I measured the same 7.3 mVrms wideband noise (it's input is earth referenced and has no option for ground isolation).

While looking at photos of other people's units, I noticed that some have two orange capacitors connected between Low Sense and Low Out to earth/chassis ground. Why would some units have these installed and other not? I have a couple of these exact same caps sitting in a parts bin...I guess it couldn't hurt to try installing them. Hopefully it helps with attenuating some of the HF noise on the output.

I also need to look at the slew rate of the switcher and check for ringing. I'm also curious if changing out the series connected 1N4007s with proper HV diodes will reduce the HF noise. Though that might just be a drop in the bucket with regards to noise sources in these units.

[Electrole]

Interesting that not all units have the two capacitors installed.
I just checked my unit, and it has two orange 20 nF / 1600 V film capacitors mounted in parallel between chassis and the LO-SENSE node.
There is one unused pad for LO-SENSE, and another for LO-OUT, but only the pad for LO-SENSE is used for the capacitors; There is no capacitor connected between LO-OUT and chassis. Both are connected between chassis and the LO-SENSE node.
I assume these capacitors were mounted to suppress some emission of switching noise from the power supply, but the issue is that the power supply pollutes the calibrator with switching noise in the first place.
If you add similar capacitors, or one sufficiently large, let us hear if this changes your noise measurements.

[RaymondMack]

Thanks. I couldn't see how they were connected and assumed that LO-SENSE also had a capacitor installed.

I'll keep everyone updated on the changes as I incrementally do the suggested mods. Hopefully I can tame the noise somewhat. And eventually find a better calibrator... These 2701Cs are pretty disappointing to be completely honest.

[RaymondMack]

I installed the 2x 20nF, 1.6kV capacitors. Tried to make it look factory done, but ended up touching the edge of a relay with my soldering iron . But other than that goof, I think it looks good. It's pretty surprising I had these exact caps in my junk bin.

Anyways, I hooked everything back up the same as before and actually saw a minor reduction in noise as measured by the Racal Dana 5002 and HP 3400B.

Wide band noise measured with the 3400B dropped to 2.55 mVrms while the Racal Dana saw a much less impressive 4.27 mVrms isolated and 4.92 mVrms when earth referenced. Using the 100kHz filter on the 5002 saw a minor reduction as well 0.217 mVrms isolated and 0.275 mVrms when earth referenced. Interestingly, looping the coax through a ferrite choke (split clamshell type) drops the isolated 100kHz measurement to 0.130 mVrms and the wideband noise to 2.56 mV. For whatever reason, the 3400B shows very little change with the ferrite. My guess is there is some sort of ground interaction between the 2701C and 5002.

So it helped, but certainly wasn't a magic bullet.

With the above in mind, I think a common mode ferrite tube could help to reduce the HF noise content if placed before the small ferrite toroids on the front panel wiring.

Next up will be the 2.2uF filter capacitors and then checking the slew rate and dampening of the switching supply.

[RaymondMack]

I was trying to figure out why the 5002 and 3400B were getting different readings, as they usually agree on most measurements and I found that there is about a 100 mVrms to earth ground from either terminal of the 2701C. I need to lift the 2x 20nF capacitor and see how that affects this common mode AC signal, but there is indeed some ground loop interaction that is throwing off my measurements.

Previously I measured the output noise with a Tektronix DMM4050 to around 1.24 mV using the slow filter. After the capacitor mod it now reads 0.07 mV. So it does look like it actually helps with the noise.

If someone else could measure the AC voltage between each terminal of their 2701C to earth/chassis ground while on the 10V range I would appreciate it.

[Electrole]

I did a few RMS measurements on my 2701C this morning.

NOTE: The power switcher in my 2701C is not the original. I made a new one from scratch as the switcher was blown in my 2701C when I got it.
This is likely going to impact the switching noise generated, and thereby the RMS measurements.
I hope that other owners of a 2701C (with the original switcher) can make RMS measurements, so that we may compare.
Anyway, here are my numbers:

Between 2701C chassis and any of the four output terminals, 20 V range and 10 V setting, activate or standby:
About 215 mV measured with a Racal-Dana 9300B
About 217 mV measured with a Rohde & Schwarz URE (same reading in different BW settings => Must be dominated by LF noise)

Between the output terminals, 20 V range and 10 V setting
Operate:
Rohde & Schwarz URE, full BW: 1.054 mV (drops to 0.968 mV when connecting the chassis to the negative terminal)
Rohde & Schwarz URE, 100 kHz BW: 0.274 mV
HP 3400A: About 1.05 mV
Racal-Dana 9300B: About 1.29 mV
Standby:
Rohde & Schwarz URE, full BW: 1.048 mV (drops to 0.945 mV when connecting the chassis to the negative terminal)
Rohde & Schwarz URE, 100 kHz BW: 0.268 mV
HP 3400A: About 1.02 mV
Racal-Dana 9300B: About 1.28 mV

The larger reading by the Racal-Dana 9300B can be explained by its wider bandwidth (-3 dB @ 57 MHz) compared to the HP 3400A and the R&S URE.

The measurements are sensitive to the wiring of the setup. By merely touching the instruments I could make the values change. This could be an indication of switching noise being radiated from both the output terminals and the power line. During the measurements the 2701C and the RMS meters were connected to the same power outlet strip with ground, and all power cables used had ground.
On my web page https://dabbledoo.weebly.com/valhalla-2701c.html there are a few more noise measurements.

Besides this, your measurements in 100 kHz bandwidth are not far from mine, but you seem to have more noise in a wide bandwidth. Maybe that is related to the some of the components I added to the power switcher. For instance, I have a 680 nF / 300 VAC noise suppressing capacitor connected across the mains power going to the switcher. If you try something similar make sure that the capacitor is a Class X safety capacitor.

[Electrole]

Just a few additional observations:

The measured noise level of my 2701C depends on what's connected to the mains elsewhere in the lab!
By switching on an LED lamp having a switch-mode driver the measured level increases some 10 %, even though the lamp and its driver are located 2 m away.
The effect can be reduced by adding clamp-on ferrites to the output and/or to the mains inlet of the 2701C.
This was observed using the Rohde & Schwarz URE RMS Voltmeter connected to the same mains outlet strip as the Valhalla 2701C, and with 1 m coaxial cable between the 2701C and the URE.

Without making further investigations I cannot tell how much noise is being picked up by the measurement setup and how much really comes from the 2701C. However, the measured noise level is without doubt higher from the 2701C than from my Fluke 5440B using the same setup. The 5440B gives 316 µV at full bandwidth of the URE, whereas the 2701C gives about 1 mV (at least my units do). The noise floor of the URE is about 78 µV at full bandwidth, so the actual figures will be lower.

I also tried to de-activate the switching pre-regulator in the 2701C and replaced the supply with an external supply.
This did to my surprise NOT reduce the noise, at least not in the specific setup with 10 V output in the 20 V range, and no load. Clearly, the switcher is not the sole source of the elevated noise we see.
I'm still thinking the seemingly disorganized layout of the 0V(a) reference plane in the 2701C is partly to blame for the noise issues and puts a limit on how much we may reduce the noise   

As a final remark, I have for a while used a combined common-mode and differential-mode filter on the output of the 2701C in order to reduce the noise.
This turned out to be necessary in order not to overload my EM N1a nanovoltmeter when making measurements between the 2701C and other DC-sources!

[RaymondMack]

Thanks for performing those measurements Electrole, I really appreciate the time you took to make them.

I did a second round of tests to see how the two 20nF capacitors affected the output of the 2701C.

All measurements are wideband RMS using the 3400B with the 2701C set to the 20V range and 10V output. The 3400B is using a 1 meter RG223 BNC cable with three ferrite CM chokes attached to the cable during assembly and a Pomona BNC to 4mm banana plug adapter on the end. The 2701C and 3400B are connected to the same mains power strip with about 400 uV between chassis grounds (measured by connecting the 3400B HI to the 2701C's chassis). So ground loop current shouldn't be a concern.

With the 2x 20nF capacitors disconnected (screw connecting the ring terminal to chassis ground removed):

AC common mode at each of the four terminals to earth/chassis ground was about 2.5V. This seems rather high.

2701C Polarity  |        Connections         | Sense Terminals | Output Terminals
---------------------------------------------------------------------------------
   Normal  (+)  | 3400B HI/LO to 2701C HI/LO |     6.5 mV      |     6.6 mV           
   Normal  (+)  | 3400B HI/LO to 2701C LO/HI |     7.5 mV      |     7.3 mV           
---------------------------------------------------------------------------------
   Reverse (-)  | 3400B HI/LO to 2701C HI/LO |     7.2 mV      |     7.3 mV           
   Reverse (-)  | 3400B HI/LO to 2701C LO/HI |     6.8 mV      |     6.6 mV 

With the 2x 20nF capacitors connected:

AC common mode at each of the four terminals to earth/chassis ground was about 86 mV. Much lower, but still higher than I'd like. I think another bypass capacitor should be located closer to the source of the common mode AC signal.

2701C Polarity  |         Connections        | Sense Terminals | Output Terminals
---------------------------------------------------------------------------------
   Normal  (+)  | 3400B HI/LO to 2701C HI/LO |     1.6 mV      |     2.5 mV           
   Normal  (+)  | 3400B HI/LO to 2701C LO/HI |     3.1 mV      |     3.5 mV           
---------------------------------------------------------------------------------
   Reverse (-)  | 3400B HI/LO to 2701C HI/LO |     3.2 mV      |     3.2 mV           
   Reverse (-)  | 3400B HI/LO to 2701C LO/HI |     1.5 mV      |     1.8 mV

Since the 3400B's LO terminal is earth referenced, we can see that orientation of the connection to the 2701C affects the measured noise. This leads me to believe the location Valhalla chose to place the 2x 20nF bypass capacitors does not seem optimal. Once I get this sorted I'll try adjusting the filter capacitors C6-8 and possibly try swapping the ICL7650S with an LTC1052 or TLC2652A.

[Le_Bassiste]

...Once I get this sorted I'll try adjusting the filter capacitors C6-8 and possibly try swapping the ICL7650S with an LTC1052 or TLC2652A.

the pin-layout of IC5 basically "supports" 8DIP packages, so it's easily doable. trying different chopper amps would require a socket in place of IC5 in order to enable exchange of DUTs. however, the presence of a socket alone could induce enough thermal noise and drift that it would easily mask any positive effect of the DUT on the overall performance of the calibrator.
i tried exactly that with an ADA4522-1 in my 2701C and i couldn't find any improvement over the original ICL7650S (output drift, SDEV). so, either the ADA4522 wasn't the best choice in the first place, or i screwed it because of the socket that i was using. i ended up with removing the socket and putting the ICL7650S (a fresh one, that is) back in.
looking forward to read about your findings!

as for the low frequency noise in the region of  x .. 120 Hz, there's one interesting phenomenon that i could observe while taking DCV readings with less than 10 NPLC and no FIL: the fluctuation of the measurement was periodic. this suggests that the LF output noise is indeed stemming from the PWM stage rather than mains hum, as the clock of the PWM is running at approx. 90 Hz, thus escaping the filtering of the mains-synched NPLC settings.
this also suggests that the mains pre-regulator is a major noise contributor in the HF region, but not so much in the LF portion of the noise, where the PWM dominates.

[Electrole]

I thought I'd better do a fresh FFT of the signal from the 2701C. Again, the 2701C was set to 10 V in the 20 V range.
I amplified the output signal (AC-coupled) by an EG&G model 113 amplifier in balanced mode, set to 100 x gain, and digitized with a National Instruments USB-4431.
The sampling rate was set to 20 kHz and 20 k samples so that I focus on frequencies up to 10 kHz, and I used the RMS averaging function in NI's "Power Spectrum and PSD.vi".
I have included two spectra: One with a logarithmic frequency scale from 0.1 Hz to 10 kHz, and one with a linear frequency scale between DC and 2 kHz.
You can clearly see the mains related components (50 Hz here in Europe) and the PWM-related components.
The graphs are scaled so that the 100 x gain is compensated for. The levels therefore represent what we see on the 2701C output.
Le_Bassiste, you may be right about the mains components being dominant at high frequencies, but it's hard to tell from this FFT included here.
I may try to do another FFT later ...   

A while ago, I considered finding a better substitute for the chopper amplifier, but that was before I realized how noisy the 2701C is.
I have no measurements of the 2701C with the original pre-regulator switcher, but the data provided by RaymondMack suggest that it may even be noisier (though there may be differences caused by our measurement setups). The effort is therefore better spent trying to lower the noise in a 2701C, I believe.

[Le_Bassiste]

I have included two spectra: One with a logarithmic frequency scale from 0.1 Hz to 10 kHz, and one with a linear frequency scale between DC and 2 kHz.
You can clearly see the mains related components (50 Hz here in Europe) and the PWM-related components.
The graphs are scaled so that the 100 x gain is compensated for. The levels therefore represent what we see on the 2701C output.
Le_Bassiste, you may be right about the mains components being dominant at high frequencies, but it's hard to tell from this FFT included here.
I may try to do another FFT later ...   

thanks for the log diagram! yeah, i can see now that my wording is misleading, because in your FFT diagram the AC mains components are indeed the major contribution in LF range. however (and that was the point i was trying to make), on a DMM that has its sampling synched with AC mains and NPLC, AC mains components (fundamentals and harmonics) will be "punched out" of the spectrum, thus making the PWM (@ 90 Hz) the main contributor in the DMM readings, aka SDEV.
as for your filter optimization around IC8: did you have the chance to test different opamps?  i found an OPA134 to give at least some improvement (10%, iirc) over the LF356.

[Electrole]

Inspired by the observations with the OPA134 made by Le_Bassiste I measured the noise of the Valhalla 2701C when using different op-amps for IC8 in the low-pass filter: LF356N (used by Valhalla), and these following substitutes: AD743JN, AD711AQ, OPA277PA, LT1677IN8 and LT1793ACN8. These were selected among single op-amps with low current noise I had readily available. I did not have the OPA134 available.

The noise in 10 kHz bandwidth was measured with NI USB-4431 with EG&G model 113 (set to 100 x gain) as pre-amplifier, with the 2701C set to 10 V in its 20 V range. I found that the measured noise was essentially the same for all op-amp used in this test, around 29.2 to 29.8 mV, with no solid correlation to the op-amp used.

The wide-band noise (about 20 MHz) was measured with Rohde & Schwarz URE, with the 2701C set to 10 V in 20 V range. This was also essentially the same for all op-amps used in this test, about 1.36 mV - 1.41 mV (which is higher than my previous measurements as the mains related noise using this setup depends on what else is connected to my mains power strip!) Again, not really any correlation to the op-amp used in this test.

The interesting part, however, is the low-frequency noise. At low frequencies the 1/f noise performance of the op-amp used for IC8 makes a difference. I measured the low-frequency noise with an EM Electronics model N1a nanovoltmeter (via a filter between the 2701C and the N1a to suppress noise above 10 Hz, and common-mode noise), with the 2701C set to 10 µV in the 2 V range. The voltage from the N1a was digitized by a Fluke 8842A at its medium measurement rate. An FFT was made on filtered data in LabVIEW. The LF356N used by Valhalla gives the highest noise of the op-amps used in this test. The following summarizes the findings, expressed as the standard deviation, and the observed peak-to-peak voltage, for two different frequency spans:

LF356N
Std. dev. = 305 nV / 2.33 µVpp 0.01 Hz - 1 Hz
Std. dev. = 269 nV / 2.31 µVpp 0.1 Hz - 1 Hz

AD743JN
Std. dev. = 199 nV / 1.77 µVpp 0.01 Hz - 1 Hz
Std. dev. = 162 nV / 1.48 µVpp 0.1 Hz - 1 Hz

AD711AQ
Std. dev. = 217 nV / 1.79 µVpp 0.01 Hz - 1 Hz
Std. dev. = 184 nV / 1.41 µVpp 0.1 Hz - 1 Hz

OPA277PA
Std. dev. = 204 nV / 1.51 µVpp 0.01 Hz - 1 Hz
Std. dev. = 163 nV / 1.23 µVpp 0.1 Hz - 1 Hz

LT1677IN8
Std. dev. = 257 nV / 2.11 µVpp 0.01 Hz - 1 Hz
Std. dev. = 222 nV / 1.82 µVpp 0.1 Hz - 1 Hz

LT1793ACN8
Std. dev. = 249 nV / 1.91 µVpp 0.01 Hz - 1 Hz
Std. dev. = 200 nV / 1.58 µVpp 0.1 Hz - 1 Hz

Based on this, I think it would sense to replace the LF356 with AD743, AD711 or OPA277. Other op-amps may be work equally well, the question is where the 1/f noise kicks in and that the op-amp is fine with the actual impedance level of the filter. Some low-noise op-amps with high current noise will not produce good results.

The OPA134 has low current noise, but its 1/f voltage noise starts to creep up below 100 Hz, so it would be interesting to find the noise at low frequencies.
Le_Bassiste: What was the frequency span of your measurements when you observed the improvement compared to LF356N?

[RaymondMack]

Thanks for taking those measurements Electrole! I have some OPA277PA sitting in a drawer, should be an easy swap. Though maybe I should put a gold plated collet socket for comparison tests. I have an EG&G 5113 preamp and an Agilent 89410A VSA that can do FFT up to 10MHz for comparison. Probably not the best setup, but it should work.

Unfortunately further progress from me is halted since I killed one of the HV MOSFETs while probing around with my DMM. I believe my unit has an issue with the smaller transformer. With the 2x 20nF capacitor removed I can hear audible buzzing from the small transformer used to power the mainboard. When the bypass capacitor is installed the buzzing greatly quiets down, but doesn't completely go away. I need to put one of my wideband current probes on the capacitor to see what the ground current looks like.

When probing with my DMM4050 in AC volts (approx. 1 megohm AC coupled to earth ground) the compliance light in the current source would start flashing/dimly glow and random relays would start switching on and off. The 2701C uses opto isolators with the digital side connected to earth ground, so I don't think it has to do to the digital side. It must be one of the protection mechanisms on the analog side getting triggered. This would happen when trying to measure the AC common mode at node "L" on C211 as well as transformer nodes E9 and E10 on the edge of the PCB. In particular, the HV MOSFET died while probing E9, which caused the relays and LEDs to start randomly clicking/light up. It seemed like an oscillation. After disconnecting the DMM, relay RLJ clicked on and off about once per second . If I remember correctly, I had disconnected the 2x 20nF capacitors and everything from the front panel terminals. So the only current path would be leakage / arcing through the transformer itself.

While killing my 2701C I did discover that the primary source of my CM noise is not related to the switching supply. I believe that the bypass capacitor should be located at node "L" but while probing I had the issues discussed above. Looking at the schematic, I don't see how I killed the MOSFET or why the entire analog circuit goes haywire when probing E9, E10 or "L". Node "L" is connected to earth ground by the 2x 20nF when set to normal polarity through relays RLD and RLG. This is why the noise goes up when the polarity is reversed: node "L" is disconnected from the bypass capacitors and the CM signal passes through the HI lead of my 3400B to earth ground. Moving the bypass capacitor closer to "L" should--in theory--help with my noise issue as it shortens the path length and ignores the polarity reversal relays.

I should have some replacement MOSFETs here soon. There may be a few more parts that need to be replaced, as I haven't done an exhaustive search for bad parts yet. Thankfully the analog side is fairly simple, so it shouldn't be too hard to get it working again.

[RaymondMack]

I finished repairing the 2701C today. All that had failed was a single MTP1N95 MOSFET. That said, I decided to replace all five with FQP2N90 from onsemi so the threshold voltages of each part would match. The FQP2N90 are a reasonably close match for the MTP1N95 and seem to work fine so far.

While checking the calibrator over, I noticed that some of the buzzing near the transformers appears to come from C201, a 1kV 10nF Y5U ceramic part. I'm going to try swapping it out tomorrow and see if I can reduce the audible noise. While I'm at it, I'll check for ringing and slew rate.

After that, I'll add in a collet socket for IC8 and run some tests with the stock LF356 vs an OPA277PA. I couldn't find any axial 2uF polypropylene capacitors to replace C6, C7 and C8 but I do have some 1uF parts that I can solder in parallel on the underside of the board for testing. If they help, I'll order some proper replacements.

I just noticed that this is the same filter topology as the active filter module used in the Fluke 8505/6A. If all goes well here, I may try replacing the LM301 with OPA277PA in my 8506A's active filter after I finish fixing the current shunt module.

[Le_Bassiste]

Le_Bassiste: What was the frequency span of your measurements when you observed the improvement compared to LF356N?

unfortunately, i don't have such nifty gear as yours in my possession. i'm directly measuring the 10 V output of the 2701C in 2-wire mode with a keithley 2002 set to 10 NPLC, no FILT, and let it collect 200 samples. when testing the OPA134, that set of samples had SDEV = 2.85 µV.
for comparison, an LTZ1000 measured in the same manner directly at the output (Dr. Frank's design) gives SDEV = 1.0 µV. that's about as good as it gets with my setup. a FLUKE 343A comes out at SDEV = 1.5 µV with above settings.

for wide-band noise measurement, i'm using the 2002 simply in ACRMS mode, 200 mV, no FILT. baseline of the 2002 is about 30 µVRMS. the 2701C @ 10 VDC gives approx. 0.87 mVRMS after warm-up. it doesn't have Y-caps on the LO SENSE (yet), but i inserted 50 µH toroidal inductors on each of the four front output wires, and a 470 pF cap across + and - output terminals. it wasn't equipped with any EMI measures when i got it, and had about 2 mVRMS before the modification. again, a 343A would read just about 50 µV with the same settings.

[RaymondMack]

I got busy with work and a few other projects so I haven't had much time to work on the 2701C until last night. I replaced C201 with a more modern ceramic part and the audible buzzing associated with the old capacitor went away. I then held a short length of flexible plastic tubing up to my ear to safely "probe" around and find the remaining source of the buzzing, and it seems that aside from normal transformer hum on the small transformer, the switching transformer is the primary culprit. I may pour some epoxy in the large gap between the laminations and windings on each side of the transformer, since that is where my probing identified as the primary source of the buzzing. For some reason I thought the smaller transformer was louder than it is .

I was going to probe the switcher but it turns out I didn’t have any alkaline AA batteries and my NiMh batteries were slightly too fat to fit in my Yokogawa 701921 differential probe. So measurements / further adjustment of the switcher circuit will have to wait until I can remember to buy some more batteries.

As for other news, I did some simulation in Tina-TI and think that something like the OPA189 is actually a better replacement than the OPA277. Simulations of the OPA134 were disappointing, as it was worse than the LF356. It is interesting to note that the noise of the LF356 peaks at 5Hz with the default filter capacitors and can be tamed somewhat by lowering C8 to 0.5uF and increasing C6,7 to 2uF and C4 to 1uF. From the simulation results, I think I'll just increase C6,7 to 2uF and use an OPA189. I still want to do comparison testing with the OPA277 and LF356 for validation. It could be that the switching noise of the PSU swamps out the noise of the voltage reference circuit.

Step response testing shows that the settling time increases by about 72% when C6,7 = 2uF. With the OPA189 (or similar AZ opamp) C6,7 could be lowered somewhat and still see a measurable improvement in reference noise without excessively increasing the calibrator's settling time.

(Note: All macro models except for the LF356 were downloaded from TI. So these should be fairly representative of real noise performance in the given filter circuit. Color Key: Maroon = OPA134, Blue = OPA189, Gold = OPA277, Green = LF356.)

Also, I believe the LTC1052 is the better replacement for the ILC7650s. It either matches or beats the ILC7650s in most DC parameters and notably has both lower current noise and voltage noise. The current noise of the LTC1052 at 10Hz is 0.6 fA/sqrt(Hz) vs. 10 fA/sqrt(Hz)  (or 0.01 pA/sqrt(Hz) as per the datasheet) and the voltage noise from DC to 10 Hz is 1.5uV verses 2uV. So it promises to be an all-around improvement over the ILC7650s. The TLC2652A has better long term stability and Vos temp co but has worse voltage noise and the current noise is only specified for 4 fA/sqrt(Hz) at 1kHz making me believe it is higher than even the ICL7650s at low frequencies.