Help interfacing Current Transformers to ADC

[Joerg]

I was refering to the prices I found for the DIN mount units with all
the electronics built in and a mV signal ready for DAQ units.

Those will be expensive for two reasons. One, they need to go through
all the UL cert hoops plus TUEV, CE and all that. Two, its a small
market and they still have to amortize those cert costs over a
reasonable period of a few years.

That will not be a problem, as I will have to supply power to the rest
of the electronics on the project (some opamps, line drivers, and otehr
stuff)

There used to be a neat trick, mainly employed in set-top tuners when
the UHF band was introduced. Remember when TV was just a few VHF
stations, Andy Griffith and all that? If there is always a certain
minimum current you can use a CT to supply the power to such electronics.

 

[Tim Wescott]

I am not sure what "baggage" comes with the precision rectifier but am
doing this as much for learning purposes as I am to complete my
automtion project (a learning experience in its own right). That
op-amp circuit looked pretty simple, though 4 diodes is MUCH simpler
and cheaper I guess. I suppose a rectifer package would be much better
suited than 4 discrete diodes?

You have to work to get it right, and to understand it. The circuit
looks simple, may work well in the lab, then give you fits in the field.

I only hope to get a reasonably close picture of the power consumption
of each device. I am looking to log more of a trend than the actual
watts. I mean I would like the results to be resonably close... but I
am nto going to lose sleep over a few percent.

I kepe seeing these warnings and will heed them. I am just curious as
to the damage that can be done without the burden. The measuring
device will be a permantent part of each circuit. I don't suppose the
rectifier is considered safe as the burden and a burden resistor is
required for safety no matter what?

A current transformer is just a regular transformer with a great big
current step-down ratio. This means that it has to have a great big
voltage step _up_ ratio. With the burden resistor connected the thing
presents a very small impedance to the primary (the burden resistance
divided by the turns ratio squared), so the voltage appearing across the
primary is very small. Should the burden resistor fail the transformer
will present an impedance across the primary that's limited by the
transformer's primary inductance, and it'll step up whatever voltage it
gets. The consequence of this is that you'll see a _huge_ voltage
across the output terminals.

So make sure you have something that'll accept the current from your
transformer! A bridge followed by a resistor and cap will form a burden
'resistor', with more opportunities for something to fail and cause
problems. You could back this up with back-to-back zener diodes, if you
wanted to feel safer.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

 

[Joerg]

That is an interstesting solution, however I fear that 20 micros and
the supporting electroinics would get somewhat pricey and take up a lot
of space. A small Atmel or PIC for each CT maybe? Or a larger multi
input unit for a bank fo CTs?

Not really. Some of the Texas MSP430 were targeted at that very market.
A low cost example is the MSP430F2013, about $1.70 in 1k qties. Can run
sans crystal, bare bones, comes with a 16bit ADC on chip.

 

[Tim Wescott]

There's no need to sample fast or to alias the waveform. The object is
to gather statistics on the waveform, not to reproduce it, so Nyquist
doesn't apply; any old scattered bunch of samples will do. I sold many
thousands of electric meters that did true RMS volts and amps, power,
power factor, KWH, KVAH, all that stuff, 16 power channels with a 6803
uP, sampling each channel at about 27 Hz.

And it's easy to offset the sine waves roughly midscale on a unipolar
ADC and software auto-zero the DC part out.

John

I like that solution. What do you do if someone hooks it to a sick
generator that's running at 54Hz?

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

 

[Ancient_Hacker]

You could make your own current transfomers- start with say a cheap
modem transformer ($1.99 at jameco). Wind a very few turns of 10-amp
capable wire over the windings. Voila, a current transformer.

Better start with just one and see how well it works. (It might be too
low a reactance or might saturate)

 

[John Larkin]

I like that solution. What do you do if someone hooks it to a sick
generator that's running at 54Hz?

Well, that can be nasty. The longterm average values will be right,
but you can make some nasty wobbles in the short-term data. I was
working with utility power, which is dead steady.

There's some interesting math here. My sample rate was actually 27.xxx
(I forget the exact number) which actually walked a 60 Hz sinewave in
some prime number of equally-spaced phase angles over 256 samples, in
a sort of random order. That turned out to have very steady
measurements when averaging those 256 samples in clusters, even when a
bunch of harmonics were present. In the frequency domain, my sampling
impulse train was dancing between line harmonics up to a pretty high
frequency. Nowadays, of course, it's easy to just sample a lot faster
to avoid so many opportunities for harmonic embarassments.

I even did an algorithm to reorder the samples and plot the waveform!

John

 

[beananimal]

CSE187L 20 pcs @ $2.34 each @ Digikey - $46.8 + S/H.
Never saw this post (just the reply) I guess i should stop relying on
google groups and go back to a real news reader.

I have used these with success as well. However, they are not very linear
over a wide range with a bridge rectifier. It is actually fairly simple to
shift the output AC from the burden resistor so it stays within the
positive input range of an A/D on a PIC.

Would it then be better just just pony up the extra $1-2 per channel
and use either a better CT (like 6$ window 1000 turn type?) and/or the
ideal recitifer as shown in the first post?

Select the burden resistor so it produces 5 V P-P at maximum current, and
tie one end to ground. Add a resistor about 20x that value to the A/D, and
another of the same value to +5VDC, and the sine wave is centered on a 2.5
VDC level.

I think I understand this and how it will work. I will order some parts
this afternoon and get the breadboard out. Is there a particular
rectifier that I should be looking for?

A simple software routine takes the absolute value, and you have
a precision rectifier in PIC code, with only three resistors per input.
Sample at about 300/sec and read average or true RMS.

Though I am a programmer (VB mostly) I will am new to PICS so imagine
this will be another learning experience alltogether. It sounds simple
enough though. Poll the A/Ds and do some math. Send the result to the
serial port or whatever.

Another way to do this is with an active 2.5 VDC reference for one side of
all the CTs. Use a burden resistor for 2.5 V P-P max, and a high value
resistor (and cap) to protect the A/D inputs from spikes. You can get a PIC
with at least 8 A/Ds for under $5.

This sounds a bit more complicated, is there a benefit?

For a very sophisticated single IC solution, ST Micro's Electric Metering
IC is very interesting. See:

http://www.st.com/stonline/products/literature/ds/10853/stpm01.pdf

$10 each is not so bad, but I fear the the component count for each
chip looks VERY high. Interesting product but I think it pushes the
cost and complexity a bit furhter than I am able to deal with.

 

[Tim Wescott]

beananimal wrote:



The lower measurement range will be non-linear and somewhat temperature
dependent with a plain old rectifier. Doesn't have to be a bridge BTW.
As to rectifier packages I haven't used one in designs for at least ten
years. Discretes are cheaper and with SMT machine stuffing being rather
low cost there is typically not much of an advantage for a reduced parts
count.


Then a passive solution might suffice.


A burden resistor is required for a CT. In the 50/60Hz world you might
be somewhat safe if it fails because the core would saturate but it
might fry the rectifier before that happens and then your electronics
could think there is no current. When the core saturates it almost
becomes a core-less inductor where its inductance drops to a very small
value.

In RF apps I have seen things like burnt up traces, transistors and
diodes that blew off their plastic parts.

I understand that in serious power transmission circles clamping your
current transformer onto a line without a burden resistor can be pretty
dramatic. There will be smoke, sparks, flame, possible injury and
almost certain laughter when your co-workers find out what you did.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

 

[Tim Wescott]

Well, that can be nasty. The longterm average values will be right,
but you can make some nasty wobbles in the short-term data. I was
working with utility power, which is dead steady.

There's some interesting math here. My sample rate was actually 27.xxx
(I forget the exact number) which actually walked a 60 Hz sinewave in
some prime number of equally-spaced phase angles over 256 samples, in
a sort of random order. That turned out to have very steady
measurements when averaging those 256 samples in clusters, even when a
bunch of harmonics were present. In the frequency domain, my sampling
impulse train was dancing between line harmonics up to a pretty high
frequency. Nowadays, of course, it's easy to just sample a lot faster
to avoid so many opportunities for harmonic embarassments.

I even did an algorithm to reorder the samples and plot the waveform!

John

That's exactly what I was thinking of, actually -- I just didn't want to
do the math to figure out the ratio. With a good multiplexing ADC you
should be able to go quite a ways down in frequency to allow a small
microprocessor to keep up with 20 channels, and still get good measurements.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

 

[Joerg]

Hello Tim,

I understand that in serious power transmission circles clamping your
current transformer onto a line without a burden resistor can be pretty
dramatic. There will be smoke, sparks, flame, possible injury and
almost certain laughter when your co-workers find out what you did.

The case where the folks at a client did that there was no laughter. It
had thoroughly fried their "golden unit" :-(

 

[Joerg]

Hello Tim,

Who said anything about 20 microprocessors? How about one, with an
analog multiplexer going into one ADC? Or one ADC with a built-in
multiplexer?

I believe the OP wants each to reside in its own rail-mount enclosure so
he'd need one uC per unit. A master-slave situation could become messy
there and might cause some code issues. But a 430F2013 is under $2.
That's not much more than a 16bit stand-alone ADC. I just priced out
something similar and was stunned how expensive those ADC still are. Not
much to be had under a buck. In some situations it's tempting to use
such a uC just as a muxed AD converter.

Still, in the OP's situation I'd try to get it done analog. It's most
likely not a really big deal.

 

[beananimal]

I believe the OP wants each to reside in its own rail-mount enclosure so

he'd need one uC per unit. A master-slave situation could become messy
there and might cause some code issues.

At this point, I am willing to try whatever is "best". As a previous
poster stated, this may be someplace between easiest and cheapest. In
the end I have to understand what I am building.

(3) 8 channel boards would work just fine, as that is how the outputs
of my controller will be grouped. 3 branch circuits, 8 channels to a
branch circuit. I am capable of laying out the basic boards and
etching and drilling them.

But a 430F2013 is under $2.
That's not much more than a 16bit stand-alone ADC. I just priced out
something similar and was stunned how expensive those ADC still are. Not
much to be had under a buck. In some situations it's tempting to use
such a uC just as a muxed AD converter.

Thats all well and good, but I know NOTHING about programming a TI and
my assembly is more than rusty. The PIC or ATMEL will be challenging
enough as it is.

Still, in the OP's situation I'd try to get it done analog. It's most
likely not a really big deal.

Nope, not a big deal at all. I just want to find an easy enough way to
read the power consumption of all of my devices and log them into my
software package.

This is for a home automation project (a reef aquarium and some other
stuff), there is no R&D money or commercial interests. It is just my
desire to learn electronics and build something useful on the
process... IE Convergence of my two hobbies (reefkeeping and
electronics).

I understand what multiplexing is, as well as sampling rates and a lot
of other uC stuff. However the implementation is in the experience and
details. Neither of which I have a firm grasp on.

My intent is to get guidance on the best path to take. From that point
I will need to research and learn about each technology or aspect of
electronics theory that I will need to implement the chosen path. Some
of this is clearly old hat or second nature to you guys, to me it is a
struggle. I am amazed at the amount of interest this thread has drawn
and am impressed with all of the help, if not more confused than when i
started! So far there seems to be several different opinions on how
this should be done, but I am not in the knowledgable position to be
able to choose or differentiate between the pros and cons of each.

 

[Paul E. Schoen]

Never saw this post (just the reply) I guess i should stop relying on
google groups and go back to a real news reader.

The post contained an image of the transformer. It was 105 kB. I don't
think Google does binary image insertions. My OE worked fine.

Would it then be better just just pony up the extra $1-2 per channel
and use either a better CT (like 6$ window 1000 turn type?) and/or the
ideal recitifer as shown in the first post?

There are pros and cons to everything. The nonlinearity of the cheaper CT
with a rectifier is not too terrible (maybe 5 to 10%), and that can be
compensated for in the PIC with a lookup table for about 2% accuracy over
about a 10:1 range. Larger CTs are usually designed for 5 amp output,
although I have seen some 100 mA. Digikey also has some PCB mounted 1000:1
CTs with a hole for one or more turns of wire, for about $5 in quantity 25.
They are made by Amveco.

I think I understand this and how it will work. I will order some parts
this afternoon and get the breadboard out. Is there a particular
rectifier that I should be looking for?

This idea was for a non-rectified solution. You do a virtual rectification
by taking the absolute value in firmware. For an 8 bit converter, any value
128 and above (bit 7=1) is positive. If bit 7 is zero, complement the value
and add one. Now you have a 7 bit rectified waveform. You probably want to
start with at least 10 or 12 bits, however, so you can get good readings at
5% of maximum.

Though I am a programmer (VB mostly) I will am new to PICS so imagine
this will be another learning experience alltogether. It sounds simple
enough though. Poll the A/Ds and do some math. Send the result to the
serial port or whatever.

You can accumulate 60 readings over a 0.2 second period in a 16 bit
register and shift right to divide by factors of two until you have the
precision you need for your measurement. For true RMS, you need to square
each reading and add to the register (24 or 32 bit), then divide by the
number of readings and take the square root. There are software algorithms
for square roots with PICs.

For best line frequency rejection and smooth readings, total over an
integral number of half-cycles. 0.2 seconds (60 readings at 300/sec) is 12
cycles at 60 Hz and 10 cycles at 50 Hz.

This sounds a bit more complicated, is there a benefit?

For a single PIC with 8 or more channels, you only need to make one active
reference at 1/2 Vcc using an op-amp and two resistors. Then each channel
needs only the burden resistor and an RC filter to the A/Ds. If you are
making multiple independent units, then other approaches may be better.

Paul

 

[Tim Wescott]

beananimal wrote:

-- snip --

My intent is to get guidance on the best path to take. From that point
I will need to research and learn about each technology or aspect of
electronics theory that I will need to implement the chosen path. Some
of this is clearly old hat or second nature to you guys, to me it is a
struggle. I am amazed at the amount of interest this thread has drawn
and am impressed with all of the help, if not more confused than when i
started! So far there seems to be several different opinions on how
this should be done, but I am not in the knowledgable position to be
able to choose or differentiate between the pros and cons of each.

I would suggest at this point that you list all of the alternatives as
you understand them, then list all of the pros and cons of each as you
understand them. To these lists append what you don't know about each
alternative that may be blocking you from making a decision.

If one alternative clearly stands out as better _for you_, even if it's
just because it's the only one that you really understand and can
implement -- do it.

If there's just things that you don't understand that you can resolve by
asking questions here -- do that, on different threads if it's different
enough from this one.

If there's just things that you need to build a little breadboard to
understand -- do that, and ask us why things smoked when they did .

I'm going to take the liberty of listing some of your options -- you
should make your own, though:

Box A:

Choice A1: Use a real current transformer, a rectifier and a filter. Og
the neanderthal EE would like it, and it'll work even better now than it
did in 1950*

Choice A2: Use a hall-effect 'transformer', a precision rectifier and a
filter. Og would be confused, but might accept that it would work after
he's sacrificed a few chickens to the Analog Gods.

Choice A3: Sample the AC and compute the current. Computationally
intensive, but needs little circuitry. Og would run away screaming, or
beat you up -- but he was going to do that when you started bringing the
microprocessors on anyway.

Box B:

Choice B1: Use one ADC and a honking big analog multiplexer. Look in
your data books for CD405x and 406x parts for the multiplexer. You can
use the ADC on your one processor, and control the multiplexers from
there. Precision? It's a good thing you don't need it.

Choice B20: Use 20 ADCs (perhaps one per processor) and do the
multiplexing digitally. With enough comm software and open-drain
outputs you could have one uP per processor, but at a higher price than
a handful of CD405x parts.

Choices B2 through B19: Mix ADCs and analog multiplexers as you see fit.

Now find all the combinations of choices from box A and box B (there's
at least 6), and see which ones you like.

* I'd say 1920, but they didn't have good solid-state rectifiers in 1920.

--

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

Posting from Google? See http://cfaj.freeshell.org/google/

"Applied Control Theory for Embedded Systems" came out in April.
See details at http://www.wescottdesign.com/actfes/actfes.html

 

[Joerg]

beananimal wrote:

At this point, I am willing to try whatever is "best". As a previous
poster stated, this may be someplace between easiest and cheapest. In
the end I have to understand what I am building.

Then I'd suggest building it from scratch. I'd even wind my own CTs,
maybe from a bunch of small surplus transformers. Experiment a little,
see if a passive rectifier will do and then scale the output to what
your uC or PC wants.

(3) 8 channel boards would work just fine, as that is how the outputs
of my controller will be grouped. 3 branch circuits, 8 channels to a
branch circuit. I am capable of laying out the basic boards and
etching and drilling them.



Thats all well and good, but I know NOTHING about programming a TI and
my assembly is more than rusty. The PIC or ATMEL will be challenging
enough as it is.

They'll work, too. But when you know one uC it's easy to hop to another,
almost like renting a car where you get a different one every time.

Nope, not a big deal at all. I just want to find an easy enough way to
read the power consumption of all of my devices and log them into my
software package.

Into a PC? Then there would be easier methods. Way easier.

This is for a home automation project (a reef aquarium and some other
stuff), there is no R&D money or commercial interests. It is just my
desire to learn electronics and build something useful on the
process... IE Convergence of my two hobbies (reefkeeping and
electronics).

I understand what multiplexing is, as well as sampling rates and a lot
of other uC stuff. However the implementation is in the experience and
details. Neither of which I have a firm grasp on.

Only experiments will teach here. But be careful to leave the 120V AC
side alone, especially with water around there. Make sure there is a
GFCI and every other required protection.

My intent is to get guidance on the best path to take. From that point
I will need to research and learn about each technology or aspect of
electronics theory that I will need to implement the chosen path. Some
of this is clearly old hat or second nature to you guys, to me it is a
struggle. I am amazed at the amount of interest this thread has drawn
and am impressed with all of the help, if not more confused than when i
started! So far there seems to be several different opinions on how
this should be done, but I am not in the knowledgable position to be
able to choose or differentiate between the pros and cons of each.

Thing is, many paths lead to Rome. You might select one where in
hindsight someone say "Oh, I'd have done that an easier way". But
hindsight is 20:20 and you'd still have learned a lot.

 

[beananimal]

Into a PC? Then there would be easier methods. Way easier.

Yes, I have written Visual Basic software to log and control many
aspects of my aquarium. That PC is the end destination for this data.
If you could elaborate, that would be great.

Some features (for those who care to see what I am trying to do):
Temperature is logged via a dozen or so Dallas 1-wire senors. PH and
ORP logging is in the works using op-amps and dallas 1-wire A/D. Metal
Halide lighting is switched based up actual solar cycles for the
desired part of the world. T5 Fluorescent actinic lighting simulates
sunrise and sunset and will be actively dimmed using dallas 1-wire pots
(right from their app note!). LED moonlighting will also follow the
specified lunar cycle, this is done via a USB 64 LED matrix
(phidgets.com). There are a LOT of other features. (24) outlets will be
conrtollable by the software in total. I would like to be able to see
the power consumption of each device on the controller. I can then set
alarms for devices that are operating out of their normal range. All
of this will be available over the internet or via the telephone and
menus driven by DTMF, email alerts etc. Most of the software is done,
I am just struggling with some of the features (the reason for this
thread).

The most critical functions (heating and cooling control) are left to a
RANCO. PCs crash and I can not afford a crash to bring the life
support down. The reason this is not all done with a uC is simple. My
lack of an in depth uC skillset is the biggest reason! I woud be
struggling to get the data to the PC for controll and GUI purposes let
alone use an ethernet interface to maek it available outside my public
network.

With regards to mains voltage and the water, yes I understand they do
not mix well and am trying to take all the proper precuations to
eliminate as much risk as possible.

 

[beananimal]

Box A:

Choice A1: Use a real current transformer, a rectifier and a filter. Og
the neanderthal EE would like it, and it'll work even better now than it
did in 1950*

I have no problem with neanderthal or cave man tools or methods. If
the end result gets me what I want, then I will not complain. I
understand that this not the most accurate way to measure my data. The
combination of a cheap CT and basic rectifier will result in a somewhat
non linear relationship between the actual current and returned DC
voltage. If the non linearity is constant, I can simply use some basic
logic on the PC side to adjsut hte returned values. If this
nonlinearity is not constant and even varies with they type of load on
the CT primary (heater, motor, MH ballast, FL ballast etc)... then I
may be in for more work than it appears at first glance. However, the
whole thing appears at first blush to be easy and cheap to implement.
I can feed the signal form each filter into the 1-wire A/D and grab a
sample when I need it?

Choice A2: Use a hall-effect 'transformer', a precision rectifier and a
filter. Og would be confused, but might accept that it would work after
he's sacrificed a few chickens to the Analog Gods.

I am not sure what the benefit here is other than using a bit more
modern club. From what I gather, my lack of understanding my male the
precision rectifier hard to implement. The component count is somewhat
higher, but the results will be somewhat more accurate if all is done
correctly?

Choice A3: Sample the AC and compute the current. Computationally
intensive, but needs little circuitry. Og would run away screaming, or
beat you up -- but he was going to do that when you started bringing the
microprocessors on anyway.

This takes something to sample and compute the current for each CT.
This is where the small PICs come in with their onboard A/D. From here
I need to get the output to the PC somehow. May get complicated for 24
devices. From what I gather from the previous posts, each CT would be
connected directly to the A/D in each PIC (or a pic with multiple
inputs for multiple CTs). via a few resistors and/or a constant Vref.

Box B:

Choice B1: Use one ADC and a honking big analog multiplexer. Look in
your data books for CD405x and 406x parts for the multiplexer. You can
use the ADC on your one processor, and control the multiplexers from
there. Precision? It's a good thing you don't need it.
From what I gather here, the multiplexing is going to reduce the

sampling rate of each individual CT and therefore the error rate will
be higher. This is getting somewhat complicated and I am afraid it
would take a LOT more outside help to get it done. The upside seems to
be a small component count.

Choice B20: Use 20 ADCs (perhaps one per processor) and do the
multiplexing digitally. With enough comm software and open-drain
outputs you could have one uP per processor, but at a higher price than
a handful of CD405x parts.

Choices B2 through B19: Mix ADCs and analog multiplexers as you see fit.

Now find all the combinations of choices from box A and box B (there's
at least 6), and see which ones you like.

You have been a large help so far. I LIKE the all digital route, but
don't think at this point I have the skills to realize a working
project. To many timing issues for a newbie to pull off! That almost
dictates that I use Choice A and one of hte options there. I am more
comfortable getting data into a PC via an interfaced A/D (1-wire, usb
A/Ds etc) or a heap of them, than I am getting the data into a uC and
trying to figure out how to get the uC to talk to my software.

 

[Yuriy K.]

Never saw this post (just the reply) I guess i should stop relying on
google groups and go back to a real news reader.

Of course you should.
I have couple of those. Project was canceled, but the CSL187L worked for
me just fine. Expected current was in 10A range.

Would it then be better just just pony up the extra $1-2 per channel
and use either a better CT (like 6$ window 1000 turn type?) and/or the
ideal recitifer as shown in the first post?

Ideal rectifier is a way to go. The schematic is simple, and it will
work. Software is very simple.
You don't have the volume to justify long development time, do you?

 

[Yuriy K.]

Then I'd suggest building it from scratch. I'd even wind my own CTs,
maybe from a bunch of small surplus transformers.

Instead of buying off the shelf CT from Digikey for $2.5?
One's labor should be very cheap to beat this price.

CT -- load resistor -- active rectifier -- low pass filter -- 3pcs 8x1
mux -- MCU with ADC.

Simple, cheap, easy to program, will work.

 

[John Woodgate]

The combination of a cheap CT and basic rectifier will result in a
somewhat non linear relationship between the actual current and
returned DC voltage.

You are probably worrying unnecessarily. The rectifier non-linearity
won't have much effect below about 3 V r.m.s. input. Simply choose your
CT and burden resistor to keep the voltage away from the non-linear
region. Don't try to get an output directly within the range of your
ADC, because that WILL be non-linear. Instead, generate more volts and
use a resistive potential divider to tap off the voltage you want.

If the non linearity is constant, I can simply use some basic logic on
the PC side to adjsut hte returned values.

It depends only on the instantaneous voltage applied to the rectifier,
and its temperature, for a fixed value of load resistance on the
rectifier.