Scanner/Multiplexers for voltage references

[Echo88]

https://www3.panasonic.biz/ac/e_download/control/relay/signal/catalog/mech_eng_txs.pdf TXS2SA-L2-4.5V
https://www.buerklin.com/medias/sys_master/root/h5c/hea/9313902428190/8873768026142.pdf AGQ210A4H
https://content.kemet.com/datasheets/KEM_R7002_EC2_EE2.pdf  EC2-3NU
https://www.te.com/commerce/DocumentDelivery/DDEController?Action=showdoc&DocId=Data+Sheet%7F108-98001%7FZ.1%7Fpdf%7FEnglish%7FENG_DS_108-98001_Z.1.pdf%7F5-1462037-5 5-1462037-5
All spec a minimum load (10µA 10 mV DC / 100μV/ 10μA), probably just to be on the save side legally and otherwise DMMs couldnt measure small voltages reliable due to negligible current over the switching relays.

[3roomlab]

i butchered the omron pdf chart ...
very messy but the lines seem to tell a story
in general, it looks like there are 2 groups of performers, the better ones have a deliberate steeper upward curve

the lines ending in blobs indicate pdf does mention about low thermal emf ability. the 274P pdf mention that it has a special low thermal model. but which model? is it a typo?
the 2 MEDER reeds are 1billion ops @ <5v switching.
has anyone seen a S4EB DC performance curve published by panasonic?

edit SEE BELOW POST for updated butchered pic

[dietert1]

Assuming the gold-plated contacts are good for dry switching, one could invent another problem. What i have seen in the past are relays using composite contacts where the signal needs to pass another contact/oxide barrier. For example the solder pins are made from thicker sheet than the contacts themselves. Oxide is known to cause EMF problems. Maybe one needs to switch some current every now and then to clear the relays. Easy to check whether a certain type of relay has this "feature", if you have one. Anyway, before using a relay outside specs, one could take one apart to understand whether it may be good enough or not.

Regards, Dieter

[kleiner Rainer]

Dieter, there is a good resource about relays and their contacts:

https://www.panasonic-electric-works.com/pew/de/downloads/technical_information_relay_de.pdf (sorry, only in german)

There was an excellent relay handbook published by SDS Relais , one of the predecessor companies, but i could not find a pdf-version yet.

My experience with their relays is twofold: my company used about a million of them, and as a ham radio operator i can testify, that they are able to switch microvolt signals well - the RX-TX-transfer switch has to, otherwise the very weak received signals would not be able to pass the relay contact.

BTW it is a bad idea to try to clean a gold plated contact with high current - in most relays the gold film is very thin for obvious reasons. Relays in telecom apps used a concept named "wetting current" to avoid dry switching, but adding a small voltage drop across the contact while being unimportant for AC applications is not desirable for metrology.

For improved reliability i recommend to look for bifurcated contacts, this gives very high reliability due to the fact that you are switching two contacts in parallel.

Panasonic relays that seem usable for metrology application are TQ (thermal EMF specified max. 2uV for the SMD version), AGN (bifurcated contacts with AgPd, Au clad) and DS (Au clad Ag contacts).

Greetings,

Rainer

[Echo88]

Afair the STLT-suffix in the G6AK-274P-STLT-US-DC5 marks the "Low Thermal"-version.
Seems Omron discarded all datasheet-versions where the LTST-suffix was described.
Funny enough they are still actively sold by e.g. Mouser/Digikey.

Anyway: even the Low Thermal versions of non-latching relays will have more TEMF than latching relays due to their construction.
Latching relays switch with one short pulse and the heat in the relay decays, while non-latching relays will always have a thermal gradient during switching and therefore TEMF in the relay contacts all the time, despite good thermal symmetrical construction.

[kleiner Rainer]

For non-latching relays there is a trick to reduce heating: the pick-up voltage is much higher than the drop-out voltage. Often the drop-out voltage is 10% of the nominal operating voltage. So option one is to use a parallel RC circuit is series to the coil and select the cap so that the charging current pulse is sufficient to close the relay while the resistor just barely holds the relay closed. Option two is described here:

http://www.ko4bb.com/getsimple/index.php?id=how-to-operate-24v-relays-from-12v

Hope that helps.

Greetings,

Rainer

[dietert1]

Of course, it would be extremely stupid to derive from the datasheet requirement of some micro-amperes the destruction of relay contacts by overcurrent.
As stupid would be the assumption that the average user of a DVM would notice a thermal EMF of 1 or 2 uV, as he/she would use banana plugs instead of copper lugs.
Before commenting here, people should try to understand what branadic is trying to achieve.

Unreliable dry switching was new to me until i saw it in active speakers for home use. Output relays need to be strong for several amperes to survive drive currents for the bass channels and they would fail for the tweeter, since during normal home use tweeter currents are to small to clear the relay contacts. So after some years those speakers would sound dumb and one could solve the problem by turning up sound level for some seconds. Nowadays one can use MOSFET switches instead.

Regards, Dieter

[Echo88]

Okay, you were the first to mention minimum load for the relays branadic used, as if that would be in any way relevant to the problem.
Then talking about missing TEMF-terms in the relay datasheet, as if that would change the fact that the TEMF in branadics scanner comes from the external temp gradient due to the surrounding electronic heat sources.

And then comparing some several ampere/inductance load switching relays with DMM-switching and claiming dry switching problems.

Regards to you too. 

[3roomlab]

For improved reliability i recommend to look for bifurcated contacts, this gives very high reliability due to the fact that you are switching two contacts in parallel.

Panasonic relays that seem usable for metrology application are TQ (thermal EMF specified max. 2uV for the SMD version), AGN (bifurcated contacts with AgPd, Au clad) and DS (Au clad Ag contacts).

Greetings,

Rainer

i stumbled on something interesting

there are 2 version of TQ pdf
TQ2 states 1A life is 0.2million with therm emf ability
TQ2S states 1A life is 0.5million no therm emf ability stated
both look like current pdf can be downloaded from the usual octopart etc
but in the TQ2 pdf, it include TQ2S as part of the "total" lineup. it is a confusing story 

this confusion almost infects TXS2 / TX2S. the TXS2 emf spec is 0.3uV. like the S and no S naming problem, the TX2S has no thermal ability stated. try read this line a few times, it helps in the confusion. now which is the model with low emf? 2S ? X2? S2? 2X ?
then there is TE, sometimes it is listed as part number, sometimes it is listed as model number.

(edit maybe all good DMM should have firmware to count relay ops to advise user of reliability issues)
updated butchered pic

bonus article : relay contact adhesion effect
maybe the contacts collided too hard !

[NWerner]

By chance I opened a TXS2-L2-4.5 just yesterday.

in the attached photo one may identify
bifurcated contacts and some contact blocks.

Aside from those blocks all other conductors
have a reddish colour (maybe phosphor bronze or CuBe ?)

[dietert1]

Here i have some results from our HP 3457A with 44492A reed relay scanner (10 channels). Since the relays are in two rows of five relays, the result looks very similar to branadics Prema 2080. Channel 10 is an exception - it sits next to the power supply. I think i could use some tape to close the 44492A on that side to improve that channel. Of course - if one has more channels than needed, one can use a subset. Resolution of the HP 3457A is 10 nV, stdev was 45 nV during a 4 hour run.

When you see a spec of "maximum thermal EMF" in a relay data sheet, you already know that is strange. Thermal EMF is a product of a physical characteristics and temperature gradient. EMF = K * dT. When switching Hi and Lo in a DVM scanner, it gets something like EMF = K * dT(Hi) - K * dT(Lo). Assuming the construction of both relay contacts is similar. There won't be a maximum, unless you specify the maximum gradient. Maybe what they mean is the gradient from self heating, but then again: What is the self heating of a bistable relay?

One of the main problems of those scanners has been discussed and the solution has been shown before by Tin: Don't use connectors at the border to the cold outside, but solder thin shielded cables to the relay board and make a loop of the cables inside the scanner.

Regards, Dieter

[MiDi]

When you see a spec of "maximum thermal EMF" in a relay data sheet, you already know that is strange. Thermal EMF is a product of a physical characteristics and temperature gradient. EMF = K * dT. When switching Hi and Lo in a DVM scanner, it gets something like EMF = K * dT(Hi) - K * dT(Lo). Assuming the construction of both relay contacts is similar. There won't be a maximum, unless you specify the maximum gradient. Maybe what they mean is the gradient from self heating, but then again: What is the self heating of a bistable relay?

I think T-EMF given in uV is meant to be due to self-heating between contact pins.
In contrast if it is given in uV/K I would assume it is due to external temperature difference between contact pins.
For a bistable version self-heating is negligible or at least decays after switching, so not sure what T-EMF in uV implies for those.
With 2 contacts involved the max. T-EMF could be worst case twice as high or best case cancel each other.

[dietert1]

After applying some mods to the HP3457A with built in relay scanner i reduced spread of thermal EMF errors to +/- 400 nV after heat-up of some hours. The mods i tried are:
- Move the floating power supply regulators to the aluminum sheet below the board.
- Cover onto the reference module
- Remove inner signal connector and solder teflon wires to scanner board.
There are more mods on my list.

One can subtract the remaining offsets per channel. Changes in ambient temperature can be corrected by declaring one of the relays a thermocouple and using its temperature dependent residual voltage for compensation of the other channels. Then the residual thermal EMF errors remain below +/-50 nV - a reasonable result. By the way the first two rounds after turning on the DVM with scanner give a channel spread of +/- 56 nV (stdev) with 155 nV p2p. Whatever you do, that scanner won't get better than this.

Regards, Dieter

[dietert1]

Here i have the results of a calibration as outlined above. Some hours at the beginning of a continuous run were used to calculate the calibration, with channel 6 of 1 .. 10 as reference ("thermocouple"). This channel was found to be representative for the others.

Then a 24h test was evaluated using that calibration. HP 3457A noise was reduced by application of 21x boxcar lowpass filters. Later i will try to reduce the protection resistors at the meter input. And run the DVM without Autozero to collect twice as many readings. The device should go into a cabinet with a bit of protection against air drafts.

Residuals with low thermal shorts exhibit standard deviations of 13 to 22 nV for the other nine channels. p2p statistics appears normal. Channel 10 is still worse than the others. Don't know yet what will happen when the lab temperature rises to 30°C in summer, yet for the time being the proposed calibration method works. The HP3457A can be used as a nullmeter for monitoring a set of voltage references to 0.01 ppm.

In the second diagram i stacked the curves by 50 nV per channel to show individual logs better. Y scale is correct only for first channel.

Regards, Dieter

[Kleinstein]

The 3457 is not very suitable as a null-meter. The resistors in the input protection contribute quite a bit too the noise, but I would expect the actual input amplifier to contribute about as much.
The noise shown shows quite a bit if low frequency noise - e.g. thermal fluctuations. The efffective bandwidth for the average over 21 readings at 100 PLC is quite low, at some 1/84 Hz. So some 40 nV/sqrt(Hz) from the resistors (102 K) would only be 4.5 nV of noise over this bandwidth. This is quite a bit less than the obersvend 15 nV RMS.

The fluctuations could be at the scanner or possibly at the voltmeter inputl - there are already relays to switch from front to back/scanner and to isolate the  low voltage path when a higher votlage is measurend.

Using the non AZ mode would only make the noise worse for a slow reading (like 10 PLC).  Swtiching to the reference channel is slow and thus a rather low effective frequency for the 1/f noise. So the higher reading rate would not compensate for more 1/f noise. The non AZ mode is usually only good with very fast readings << 1 PLC.

[z01z]

Dieter, there is a good resource about relays and their contacts:

https://www.panasonic-electric-works.com/pew/de/downloads/technical_information_relay_de.pdf (sorry, only in german)

There's also an english version: https://www.panasonic-electric-works.com/pew/cz/downloads/technical_information_relay_en.pdf

[dietert1]

Many years ago i bypassed the HP 3457A front end for a certain project. Our HP 3457A has U111 in a socket and there is a pigtail from one of the 51K protection resistors to plug there and feed the input terminal to the ADC directly. For a null meter application one could replace the front end by an integrated chopper amplifier with the correct gain. Monitoring voltage references does not require extremely low leakage.

I used the scanner in two wire ohm mode to put a small current into the relay contacts before. Anyway, for me nine channels with 80 .. 100 nV p2p are useful and i did not expect it from that old meter.

Regards, Dieter

[dietert1]

Meanwhile i repeated the low thermal short test for a DIY scanner i made last year. It has a FTDI UM245R module and a XC9572XL CPLD and drives an 8x4 matrix of 32 coils in 16 bistable Axicom relays. There are four pnp transistors as pull-ups. The eight pull-downs are CPLD pins (5V tolerant). Although the relays are general purpose in contrast to the Coto relays of the HP 44492A, the DIY scanner is superior, as it remains cold. The relay coils are 180R and operate with 20 msec pulses. Average total power consumption in all 16 relays is about 1 mW during this test. Two coils get operated every five seconds to advance the scanner. While the span in the 44492A was +/- 400 nV after some mods, this time it is more like +/- 150 nV. Relays 1 and 16 are significantly high.
When calibrated as discussed above, this DIY scanner operates at p2p = +/- 20 nV over 24 hours (15 channels).

Regards, Dieter

[MegaVolt]

When calibrated as discussed above, this DIY scanner operates at p2p = +/- 20 nV over 24 hours (15 channels).

Is there a link to the topic with the project?

[dietert1]

No there hasn't been a thread, but it was mentioned before: https://www.eevblog.com/forum/metrology/hp-mux-e1476ae1442a-cards-for-ref-scans/msg3080729/#msg3080729.
You may notice that i increased pulse timing to 20 msec as recommended in the Axicom datasheet. And channel enumeration was rotated by 180° after some tests with a resistor ladder. Design of a board for that MUX is not yet finished. The board will definitely be bigger than that hand wired prototype as i want to increase distance between drive circuitry and the relays. Also i want to reserve a region of the board for a thermal clamp, with the relays on one side and all external connections on the other side of the clamp. The board will have a connector for the controller, as others may prefer an Arduino or the like instead of the CPLD

Regards, Dieter

[dietert1]

Here i have a test of a spare Axicom bistable dual coil relay. One coil is driven using a 100 uF capacitor - yellow trace. Blue trace shows the inactive coil. One can see the EMF with indication that the contact movement finishes within about 3 or 4 msec. That EMF should be taken into account when designing drive circuitry for the scanner.

Regards, Dieter

[dietert1]

Meanwhile i finished the next version of a USB scanner with Axicom bistable relays as mentioned above. After spending two days on the design of that 4-layer board, i made three of those for our lab. Later i added a small mezzanine with 2x 74AC138 to make the control interface fit into a DSUB-9. The image shows the thermal clamp made of two 8x12 mm aluminum bars, one above the board, the other one below.
In a first test today the channel to channel variations remained within +/- 6 nV p2p without any compensation or calibration. And the 4x4 geometry is still visible as a pattern - so maybe there is still room for improvement.

Regards, Dieter

PS: In fact the decoders are 2x 74ACT138 to implement translation from 3.3 to 5 V at the same time. For the tests i made a STM Nucleo-L432KC as a controller, external to the box. Communication with host is over USB CDC (virtual com port that is part of the ST-LINK debugger).

[dietert1]

Testing a scanner in the low nanovolts with a HP 3457A requires a lot of data. The instrument has a resolution of 10 nV but a noise level of about 43 nV (standard deviation at 100 PLC with AutoZero).
Meanwhile i collected 3000 rounds with the new revision of the scanner. Standard deviation of the 16 channel averages came down to 1.1 nV, with +/- 2 nV p2p. As a cross check i disconnected the scanner from the 3457A input terminals and wired a low thermal short instead. After collecting another 3000 rounds i got 0.6 nV StDev of the channel averages with +/- 1.2 nV p2p. These numbers are as expected from the basic 3457A noise. The diagram shows both sets of data. The visible 20 nV shift between them is very low frequency "noise" of the 3457A (e.g. ambient temperature drift). Average noise level per channel was 42.8 nV rms when measuring through the scanner and 43.2 nV rms with the one low thermal short at the input terminals.
Conclusion: With a 95 % confidence level detection limit of +/- 3.3 nV there are no imperfections of this scanner, neither channel offset nor enhanced noise. Good enough for monitoring voltage references.

Regards, Dieter

PS: You may have guessed it, channels 9 .. 11 (the most suspect ones in the blue plot if at all) are the four relays where the screw in the thermal clamp is missing. Need to fix that asap.

[ch_scr]

Just arrived Fluke 2300A Thermocouple Scanner. Instrument is copyright 1979 and has transformer isolation and SMPS for the 5V rail. Propably to reduce thermal gradient. Scanner card has draft shield for input connections, draft shield on bottom side and what looks like draft shield between relais board and main scanner pcb? Also copper sheet below relais sub-pcb. Input screw connectors are on big block of what looks like brass, whole scanner cards weighs 713g (25oz). Unit has shielded cage around scanner cards, should block interference from transformer as well and even out thermals. Interesting enough, seems like muxed signal is routed through backplane/front panel pcb, main pcb, communication pcb and out the ribbon cable / db connector on the back  where it would connect to thermometer unit.

[ch_scr]

Took apart the scanner board. Screws holding the bottom draft shield hold a coil board on top. Between coil board and reed relais is a 0.6mm thick copper sheet. Below is an isolating plastic and reed relais in little plastic tubs, soldered to ceramic interposer boards.