Need some help CALIBRATING RCTime results for 12V SLA battery monitor circuit

DavidM · 2008-09-21 08:14

Hi,

I have an RCTime circuit, using an 18NF Green cap & a 680K resistor to do RCTime to measure the voltage of a 12V SLA battery.

I am very close but i cant seem to work out from the data that I am getting, how to "CALIBRATE" or work out some king of formula to scale the RCTime data to REAL voltages I am measuring.

its the last 3 lines of code ( the TEMP variable adjustments) that i need to sort out. everything else is OK and working.

I am using an ADJUSTBLE LAB POWER SUPPLY , so i can dial in any voltage I want to. I have the debugger running ( elsewhere) to view the results on my PC monitor.

the return voltage FORMAT is for example "125" for 12.5 volts

Here is the code that I am currently using..

PRI GetVoltage: BatteryVoltage | Temp
       
    Temp := 0                                       'Initialise Temp
    DIRA[noparse][[/noparse]BatteryPin]   := 1                     'Make CAP pin an OUTPUT
    OUTA[noparse][[/noparse]BatteryPin]  := 0                     'Set CAP pin LOW to CHARGE capacitor              
    WAITCNT(CLKFREQ/500+CNT)           'PAUSE for 2ms
    DIRA[noparse][[/noparse]BatteryPin]   := 0                     'Make CAP pin an INPUT  to Discharge the CAP
    Temp := CNT                                   'TRAP the current Time in Temp

    WAITPEQ( |<BatteryPin, |<BatteryPin , 0 )   'WAIT until a change occurs on CAP pin

    Temp := || ( CNT - Temp )                          'Calculate the time LAPSED    

    Temp := 1_000_000 - Temp              'Invert the value
    Temp := Temp / 1000                       'get rid of some precision
    Temp := Temp - 752                         'adjust to suit values ??????

    BatteryVoltage := Temp

    RETURN BatteryVoltage

Sample results ( showing VOLTAGE IN, RCTime result and difference +/-)

16.0V = 155 -5
14.0V = 140
13.5V = 136 +1
13.0V = 131 +1
12.5V = 126 +1
12.0V = 120
11.5V = 114 -1
11.0V = 108 -2
10.5V = 100 -5
10.0V = 93 -7
9.5V = 83 -12
9.0V = 72 -18
8.5V = 62 -23
8.0V = 48 -32

as you can see my results between 14.0v and about 11.0v is almost spot on, But i believe a 12v SLA battery gets "LOW" below 11v and this is where my RCTIME values start to drift off, dont ask me why the result for 16V goes the other way?????

So,

Can anyone help with how I can calibrate this. NOTE: I do not want to use a LOOKUP TABLE that would be cheating!

.

Regards

Dave M

Leon · 2008-09-21 08:54

The relationship is non-linear. You will have to correct for that. Or use an ADC.

Leon

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle

DavidM · 2008-09-21 09:02

Hi Leon,

So How do I "CORRECT" this? the use of an ADC in this case is NOT an option.

Thanks

Dave M

Paul Baker · 2008-09-21 10:27

You need to find the characteristic curve which describes it and find the inverse equation. There's a utility which can help you with this, http://zunzun.com·it takes some getting used to but it is very powerful for finding characteristic curves.

After messing around on the site·I found a relatively simple equation thats a decent fit:

attachment.php?attachmentid=55923

y = 1.0 / (a + bx + cx²)

where

a =· 1.5180026697294174E-01
b = -5.4288518673659548E-04
c = -2.1622767946101787E-07

x is the·pulse width·and y is the voltage

Not much help unless you plan on including the floating point library. But you can use the equation to generate a lookup table to provide a quick conversion from pulse width to voltage. I would use offseted integers to represent the voltage, say x100 so 12.52V is stored as 1252.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

Post Edited (Paul Baker (Parallax)) : 9/21/2008 10:45:35 AM GMT

Leon · 2008-09-21 10:40

It's zunzun.com (no www).

Leon

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Amateur radio callsign: G1HSM
Suzuki SV1000S motorcycle

DavidM · 2008-09-21 11:50

Hi,

This seems to be getting a bit complicated, would using a larger capacitor value make for a more linear values ?

thanks

Dave M

Mike Green · 2008-09-21 12:34

No

The technique you're using to measure voltage is inherently non-linear.
You want a specific accuracy over a voltage range that's not achievable
without compensating for the non-linearity.

Timmoore · 2008-09-21 16:12

A lookup table is the easiest way, if you subtract 48 offset the value before lookup you need a 128 byte table. If you can live with small inaccuracies then make the table smaller and use linear interopolation, i.e. the curse is now a set of short lines, if +-2 is close to linear then

in is input value
in -= 48 ' remove smallest value
lsb2 := in & 3 ' save lsb 2 bits
in >>= 2 'remove 2 bits from lookup
in := table[noparse][[/noparse]in]
in += lsb2 ' add interopolation

this gives a small 32 byte table. You can proabably just hand pick the table values rather than calculating them.

Erik Friesen · 2008-09-21 17:19

Why is an adc not an option? An mcp3204/8 with a 4.096 reference and a voltage divider from 12v makes it a cinch. I suspect that the adc objects written for the prop would also give you the resolution you want.

I have tried something similar to what you did with a rctime but found the results varied too much dependant on temperature and other variables.

StefanL38 · 2008-09-21 19:00

hello

finding a fitting function can be done with Office-software too.
In Mixrosoft-Excel and in openOffice there is a diagramtype called "XY-Diagramm"
see attached screenshot

Inside the options you can add a regression function
You can choose from different function-types as linear, logarithmic, exponential, potentional
The value R^2 is some kind of an indicator how good the regression-function does fit with
the explizit given values

use the regessionfunction to create two columm of X-Y-Data with a constant stepwidth
next thing is to create a lookuptable with values fom the regression-function

best regards

Stefan

Paul Baker · 2008-09-21 19:04

David, why do you say it's getting difficult? Take the equation I gave you (or any equation which does an adequate job of describing the curve), plug the values 1-255 in, store them in an array, use the pulse width as the index into the array and your done.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

Tracy Allen · 2008-09-21 19:04

The way I would have this hooked up would be the capacitor between ground and batteryPin, and the 680k ohm resistor from the pin to the battery (+). Is that how you have it? Then the time measured is the time it takes for the capacitor to charge from 0 to about 1.65 volts.

If that is the case, a better formula for the inverse RCTime would be:

Temp := K1 / Temp  + K2             'Invert the value, K chosen for best fit.

The reason is that when you model the RC circuit, it looks like a pretty good current source when the battery voltage is much greater than the 1.65 volt Prop threshold (as it is here), and if it were a perfect current source the time would bes directly proportional to K/Temp. This is not a perfect current source, more like the inverse polynomial that Paul posted above, but the cx^2 term may (probably) be neglected. A simple procedure can find the two constants K1 and K2 based on two measurements. I have a BS2 article on the calibration procedure here.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com

Paul Baker · 2008-09-21 19:09

You're correct Tracy, the 2E-7 coefficient of the x^2 term is 3 orders of magnitude smaller than the other terms and can be neglected.

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Paul Baker
Propeller Applications Engineer

Parallax, Inc.

DavidM · 2008-09-22 01:06

Hi,

thnaks everyone for the input,

To Erik Friesen, I already have my circuit boards made, in the future I may do this the way you suggested, but I am stuck with RC Time method for now.

To Paul Baker, It was late at night, so anything to me at that time was going to be complicated, no offence

to Tracy Allen, Yes that the way I have it, I got the idea from your website.
Your idea is that I need to work out the 2 constants and then use your code.

Temp := K1 / Temp + K2

so according to your website is this what I use??

V1: HIGH voltage applied to circuit
RCT1: corresponding raw RCtime value

V2: LOW voltage applied to circuit
RCT2: corresponding raw RCtime value

K1 = 10 * (V1 - V2) * ( RCT1 * RCT2 ) / ( RCT2 - RCT1 )
K2 = 10 * V1 - ( K1 / RCT1)

The spin code should be then,,

BatteryVoltage := K1 / Temp + K2

I will try this out

Thnaks to all

regards
Dave M

DavidM · 2008-09-22 02:10

Ok, Forget everything !

For some reason I NOW GET EXACT RESULTS accurate to about 1 tenth of a volt ( +/- 0.1v) with the following code..

PRI GetVoltage: BatteryVoltage | Temp

Temp := 0 'Initialise Temp
DIRA[noparse][[/noparse]BatteryPin] := 1 'Make CAP pin an OUTPUT
OUTA[noparse][[/noparse]BatteryPin] := 0 'Set CAP pin LOW to CHARGE capacitor
WAITCNT(CLKFREQ/500+CNT) 'PAUSE for 2ms
DIRA[noparse][[/noparse]BatteryPin] := 0 'Make CAP pin an INPUT to Discharge the CAP
Temp := CNT 'TRAP the current Time in Temp

WAITPEQ( |<BatteryPin, |<BatteryPin , 0 ) 'WAIT until a change occurs on CAP pin

Temp := || ( CNT-Temp) 'Calculate the time LAPSED

Temp:= 15500 / ( Temp / 1000 )
BatteryVoltage := Temp + (( 100 - Temp ) / 9 )

RETURN BatteryVoltage

I have tested this between 5.5volts up to 16.0 volts well within the range I need to check out my 12V SLA battery level.

I dont know why this works but it does

Thanks

Dave M

Tracy Allen · 2008-09-22 16:22

Hi Dave,

That math reduces by collecting terms to,

Temp := || ( CNT-Temp) 'Calculate the time LAPSED
BatteryVoltage := 13777778 / Temp + 11

When I plug the values from your first post into my calibration formulas, the ones you set out two posts back, I come up with,
BatteryVoltage := 14527241 / temp + 4
or
BatteryVoltage := 14527241 / (temp/10) + 38 ' one more decimal digit, xx.xx at reference points
That using 16.0=V1 and 10.0=V2 and 93000=RCT1 and 151000=RCT2 as the reference points. (back calculated RCT=CNT-CNT values, some loss of precision)
The Prop allows those nice big 32 bit integers!

▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔▔
Tracy Allen
www.emesystems.com

DavidM · 2008-09-22 21:55

Hi Tracy,

When I tried to use the formulas you suggested I must have "Stuffed-up" my caculation because my results came out nothing like yours, but..

I did come up with my own formula that gives me greate results, check my last post before this one..

Temp := || ( CNT-Temp) 
Temp:= 15500 / ( Temp / 1000 )
BatteryVoltage := Temp + (( 100 - Temp ) / 9 )

did you have a look at this new formula?

regards

Dave M

zunzun · 2008-09-23 17:03

Paul Baker (Parallax) said...
There's a utility which can help you with this, http://zunzun.com it takes some getting used to but it is very powerful for finding characteristic curves.

Thank you for referring my site! Don't forget that the actual curvefitting source code is available at

http://code.google.com/p/pythonequations/downloads/list

if you'd like to use it.

James Phillips
http://zunzun.com

Need some help CALIBRATING RCTime results for 12V SLA battery monitor circuit

Comments