How to measure -60V to 1000V

Maybe...

http://cds.linear.com/docs/en/design-note/dn219f.pdf

24bits is A LOT of bits...
alot-alot

Man noise will be an issue..

-methods

2^10 = 1024 bits
2^24 = 16,777,216 bits

Uh... 16 MILLION bits ?!?!?!?

so... uh... if we divide 1000V down to 5V using the best damn resistors on the planet... like... super expensive 0.1% or even tighter... we are on a 200:1 ratio

5V / 16M = 300 nano Volt per bit
Bump it by 200 to get uh.. 60uV resolution on 1000V?
My math off?

If I assume two orders of magnitude error (which is realistic...) we are at 6mV on a signal that swings 0V to 1000V

Did I miss something?
Is this an $8 solution ?

Eh... it even has a MUX on the front to do 4 or 8 readings...
Of course they need to be referenced... but for reading across a contactor.

Hrm... makes one scratch the head and think

-methods

uh.... WTF?

Is this a list of 32bit ADC's?>?
http://www.linear.com/parametric/Analog-to-Digital_Converters_(ADC)

Is that even frucking possible?

Last time I was at linear (2001) they were bragging about having the first 14bit... and that was a big deal.

Unbelievable.
I just dont see how they can manage the noise with such tiny values.

Aliens have definitely landed.
If I buy one of these, measure 0-1000V, have like... 1% accuracy, then run it through a $3 digital isolator to my uC.... I am so winning... and we will have proof of alien technology on earth.

4,294,967,296

So.... eh... thats 4.2 BILLION bits
divided into 5V... thats about 1 nano volt of resolution

Man.. 3 years shoveling shit and look how much I missed.

-methods

This may not be relevant, because I don't know the importance of all the various specs to be able to compare them to what you're discussing here, but audio recording has been using 24-bit analog to digital conversions since at least the 1990s, AFAICR.

For instance, the GadgetLabs Wave 8*24, which I have had since it came out, uses a Crystal Semiconductor 24bit ADC
http://www.farnell.com/datasheets/32444.pdf

GL schematics:
https://bitbucket.org/antoinedeschenes/gadgetlabs-wavepro/src/2ae767c1736da2f09e58d1eb3cdde06134b686e1/Hardware/Schematics/?at=master


32-bit recording has also been done, but I don't remember anything about that hardware, and it's much more recent. (though the software has been able to do it a lot longer, it's just padding the recorded data when hardware doesn't provide that depth)


btw, the links you give to Linear pages don't bring up anything but their category list--I think their page must be using cookies and scripts to keep the data on your browser without making a transferrable URL others can look at.

Very nice.
Thanks Amberwolf

The history of ADC can be confusing... as at first they were supercomputers. Here is some hardcore history
http://www.mukhanov.com/uploads/ADC-History-Ch7.8_100yearsSC.pdf

The real trick is getting rid of false readings...
For the right price and enough engineering I am sure that in the 90's it could be done. What we are talking about now is turn-key chips with few peripherals and predictable output. As the bandwidth goes up the need for gnarly anti-aliasing filters goes down... which solves all sorts of complexity issues.

The primary difference between the early days and what you can grab now is settling time on the ADC as well. Thats how long you have to wait for the internal capacitance to charge before you can switch channels or take the next reading.

Note that the Audio devices barely meet Nyquist limit for a 20hz to 20khz spectrum (which would be 40khz...) so they run up to only 50Khz

In the time domain that would be 20uS ... still plenty fast for our purposes. Orders of magnitude faster than is needed. Voltage probably needs to be read once a second in a BMS.

Looks like Linear is doing 10Mhz (0.1uS settling time) for up to 18 bits and 1000sps on 32 bit.
http://www.linear.com/product/LTC2500-32

The actual settling time and how fast you can sample are often masked in data sheets. For the sheet provided... Looks like: 50khz sampling rate. Thats "supa-fast!" for our needs.

Both of those are astounding bits of equipment.

The take-home here is that measuring HV can be very simple with a $10 part and a 2 resistor divider. No need for scaling, multiplexing, multi-pole filters, or anything fancy.
I can say that 10 years ago... eh... 2007ish... this was not an option.
Perhaps because of factors other than technology. Perhaps availability or cost.

I would still be HIGHLY suspicious of noise.
It is quite possible that to meet the requirements of the datasheet you would need:

1) Dead quiet circuit
2) Temperature control
3) Very high or very low impedance

Stuff like that...
Which all sound good till you try to first:

1) Get it running on the bench
2) Get it running in a system
3) Get it running in a system that is running

I know that with most high end Audio... linearity and consistancy is if the utmost importance... so much of that gear "warms up" to a known temperature and holds at that temperature. Thats the only way to get all the fussy chips to play nice together in a consistent way.

Ok... I am supposed to be working...

-methods

methods said:

Ok... I am supposed to be working...

Click to expand...

Reviewing requirements is working, right?

Thanks for the details of what is important in these chips--that's stuff that sounds familiar so I might have known that back when I came out of DeepFry as a tech, but I don't specifically remember it.


As for operating temperatures/etc, well, most of the stuff would be used in controlled climates in recording studios or bedrooms--who wants to be belting out tunes in anything other than pure comfort?

Don't need all that stuff on stage in the heat/rain/etc. for concerts, for instance (and if you do want to record them with it you can still have it all backstage in the control room).


For quiet circuits...the GL8*24 is still the quietest (with no input signal or with one) recording box I've ever tried. Everything else has always had excess hiss or other noises from it. I don't know what the specific difference is, but whatever it is, I wish I could still be using it (have to use this laptop for everything for now, and it's a PCI-based board).

It's such a good unit that there has been an ongoing driver development (by users) even today, despite the company itself going under before they even released the WinXP drivers I specifically picked this one to use it with.

Linear Technology is great for understanding the advancements in digital filtering... Nyquist limit and aliasing are so very important.
I would say filtering is 90% of the equation and sample-n-hold is the other 10%

Speaking of which (work) the cost-benefit analysis of doing both HV measurement and isolation tests came up with the following:

1) USE a COTS PCB from a low cost auto-ranging DMM ($10)
2) Power it from Isolated DC-DC and communicate with it via digital isolator ($10)
3) Multiplex it with SPST 1KV rated reeds @ $3 each, 6 needed minimum $20
4) HV considerations so a larger PCB - piggybacked

Engineering of a reliable HV measurement is not on the table... so...

Combined with CAN communciation, isolation mesurement, precharge, and contactor drive

Available signals required are B-, Chassis, P+ (B+ AFTER contactor), AUX power/ground (12V or any other isolated signal)... B+ required for Precharge
All signals avaialble

Matrix is

B- to Chassis
B- to any AUX
P+ to Chassis
P+ to any AUX

Where dual failure (both B+ and B- to Chassis or AUX) is detected as a voltage divider
Where 100Mohms is way more insulation that we can reasonably expect when counting all the datasheets up in the system
Where the measurement is diode isolated (for the Linear ADC approach) and bi-polar for the DMM approach (eliminating measurements)
Where no induced voltage is required (HiPot by nature) and no current or resistance measurements are to be trusted (but could be used)

where... wait... why am I writing my engineering notes online?
Oh yea... because I am not being paid... and when I am not being paid I work open source.

Speaking of being paid - I have 4 contracts now

One OEM Startup in Fremont
One Robotics in Palo Alto
One Sevcon related in La Selva Beach
One LabVIEW (by far my favorite at the moment)

So... as soon as I can make the BMS contract pay... we are off and running with a full schedule of 40hrs a week.

Currently the BMS project pays only in parts, support, design review, production help, and outlet
Its a HORRIFIC effort to get right in the face of multiple OEM examples and a few COTS examples...
Where looking to make Lego parts which can play nice with others, impede no one, possibly make it into an OEM lineup, and not cause us to go broke.

thanks,
-methods

As for high voltage low current
AC coupled
High Frequency...

Eh... remember when we used to put our hands on static electricity machines?
I think it is important to test only for what will directly harm 80% to 90% of people
If someone is especially sensitive or has electronics in their body... well.. they get to take extra precaution

Just like we cant make every double diamond snowboarding ski-lift handicap accessible... we cant make everything 100% safe in all conditions all the time.
The purpose of safety is to detect and eliminate gross negligence which is likely to harm

What I am looking for is hard faults or shorts in the system

Yes there will be capactive coupling
Yes there will be conductivity in the hundreds of megs
Yes there will be high frequency HV shooting around

The question is if it will kill you or stutter your heart
Not if it will zap you or if you can feel it.

Let me tell you about feeling electricity... felt a lot of it..
Current kills not voltage
Current requires an ass-load of capacitance to kill you - more than parasitics

Often.. ONE MAY WANT TO INTENTIONALLY TIE HV SYSTEMS TO CHASSIS OR OTHER THINGS WITH A KNOWN LARGE RESISTANCE IN EXCESS OF REQUIREMENTS

Yea... nobody agrees. Thats fine

They will be connected.
Just matters whether its consistent or inconsistent... whether it changes over time... what the value is... what the RLC components are
I would just choose to fix an acceptable value
Make that connection in an acceptable place
Accept that the parallel sum of all things will add up to some leakage

Make sure it cant harm anyone or cause electrolysis and stuff like that
Perhaps that will help with the shoot-through which is totally unpredictable and tied in to all sorts of transient things... which... I cant afford to detect.

Pretty soon we are going to have HV DC all over the place
The economy will drive down the price
Just like we have exposed 240V plugs that will frocking KILL YOU super quick... we will have 300V and 500V and 1000V around

We just need to understand it
Never fear it
Be rational about it

Business people, the media, well meaning alarmists, computer dorks, and the mom's walking group should have no say in it.
Guys and Gals who actually work with HV on a daily basis and understand the true risks of shock, harm, and death should be pressed to do the minimum for protection of life.

Regulation can kill things...
And I can tell you that a lot of industries would like to regulate us right now...
And anything over about 48V is treated as nuclear waste...

So
Yea

OEM's are spending $100k each on producing $30k cars and doing everything they can to make sure its a success
This does not need to set set precedence... as some will surely overshoot the mark.

Just remember
Static electricity machine
Shockers and stun guns and tazers
Remember your hair floating up

Thats about where I want to set the BARE MINIMUM limits... where devices are rated by stages of safety... where the minimum is basically... "may cause your hair to float".. and where folks can choose to hit higher levels... like "you can let your baby suck on this".

Second order protection... like mounting and human barriers...

Different regulations for stationary verses mobile power storage

Yea... food for thought.

-methods

Point?

I suspect you can SIGNIFICANTLY drive down cost by accepting a known connection between B- and Chassis.

Proof: It is there anyway... even if it does not show up in a Vuegraph, schematic, power point, etc. If it is not there now it will appear there after years in service, ESPECIALLY in wet environments. One thing we dont understand at all is how all this stuff is going to age - in harsh environments. Most of what we know is from protected environments.

I vote for some accelerated aging tests

Grow green crystals and stuff

-methods

methods said:

Point?

I suspect you can SIGNIFICANTLY drive down cost by accepting a known connection between B- and Chassis.

Proof: It is there anyway... even if it does not show up in a Vuegraph, schematic, power point, etc. If it is not there now it will appear there after years in service, ESPECIALLY in wet environments. One thing we dont understand at all is how all this stuff is going to age - in harsh environments. Most of what we know is from protected environments.

I vote for some accelerated aging tests

Grow green crystals and stuff

-methods

Click to expand...

Whatever isn't potted, and even if it is potted if it's not using anti-capilary wire or solid bare copper potting pass-throughs dies rapidly. As I think you're looking for low power pass throughs for voltage measurement signal, step 1 would be solid core and/or anti-capilary wiring, and the best potting for your application completely encapsulating everything that isn't a needed pass through.

No, don't connect HV- to chassis.

Its not hard to measure voltage, this does it for +/-650v in a safe way
https://eyrie.io/board/88ebffa8546c4bd0b7d9c5444c0bad18?active=schematic&sheet=1&x=6593271&y=7579995&w=14601299&h=7973194&flipped=false

methods said:

Proof: It is there anyway... even if it does not show up in a Vuegraph, schematic, power point, etc. If it is not there now it will appear there after years in service, ESPECIALLY in wet environments. One thing we dont understand at all is how all this stuff is going to age - in harsh environments. Most of what we know is from protected environments.

Click to expand...

I told you, *every* DC charger measures resistance from HV- and HV+ to chassis before delivering current. If it fails it won't charge, I think it fails if measures <500kohm. There are reasons not to play with this.

marcos said:

There are reasons not to play with this.

Click to expand...

Sounds like a reason to redesign....

-methods

I've been working with some 720vdc to 960vdc systems lately for grid energy storage packs. It's been terrifying to hand assembly RnD prototypes and doing testing where every move you make with a tool or meter is only a single mis-touch away from being your least earthly sensation.

The capacitive coupling is startlingly low ESR between the enclosure around a HV pack by virtue of being a low-impedance metal shell isolated by some dielectric material isolating the internal cell stack. In some tests of measuring the current impulse of shorting the galvanically isolated case capacitance to either pack positive or negative easily peaks >1kA (for a very short duration of course as the capacitance drains so rapidly.)

Yes capacitance on these HV packs would get super hairy. Particularly those with aluminium enclosures so close to the full potential of the pack. Working on live stuff (if you must) at these voltages is serious stuff.

As for measuring voltage on a 1000 V pack, I use a voltage divider on Voltron and power the display form an isolating DC/DC converter. Prior to this I used a pair of AA batteries stuffed under the dash

I guess anything grid-related they're used to kV so don't think twice about using it. However, I imagine it's actually a different kettle of fish with a battery system because it's always "live" unless cleverly designed with multiple contactors/disconnects to break up the series string.