Go Down

Topic: string manipulation and float math problems (Read 1 time) previous topic - next topic

jerseyguy1996

In this piece of code:

Code: [Select]
float f, latminutes, longminutes;
    long latdecCoords, longdecCoords;
    int latdegrees, longdegrees;
   
    //char sLongDecCoords[10], sLatDecCoords[10];
   
    //convert latitude to GPS Decimal format
    f = atof(latit1);
    f /= 100.00;
    latdegrees = (int)f; //isolate degrees
    latminutes = ((f - (int)f) * 100.0)+(atof(latit2)/10000.00); //get minutes and fraction of minutes
    latdecCoords = (long)((latminutes / 60.0)*1000000);  //convert to fractions of a degree
    //latdegrees += latdecCoords;  //combine it all into one float
   
    if(NS[0] == 'S') latdegrees *= -1.0;
   
    //cconvert longitude to GPS decimal format
    f = atof(longit1);
    f /= 100.00;
    longdegrees = (int)f; //isolate degrees
    longminutes = ((f - (int)f) * 100.0) + (atof(longit2)/10000.00);
    longdecCoords = long((longminutes / 60.0)*1000000);
    //longdegrees += longdecCoords;
    if(EW[0] == 'W') longdegrees *= -1.0;
   
   
   
    sprintf(latitude, "%i.%ld", latdegrees, latdecCoords);
    sprintf(longitude, "%i.%ld", longdegrees, longdecCoords);
    Serial.print(longit2);
    Serial.print(",");
    Serial.print(atof(longit2)/10000, 4);
    Serial.print(",");
    Serial.print(latdegrees);
    Serial.print(",");
    Serial.print(latminutes,4);
    Serial.print(",");
    Serial.print(longdegrees);
    Serial.print(",");
    Serial.println(longminutes,4);


and

Code: [Select]
latit1 is equal to "4041"
latit2 is equal to "1271"
longit1 is equal to "07406"
longit2 is equal to "4956"


I get an output that looks like this:

Code: [Select]
4956,0.4956,40,41.1271,-74,6.4954


The problem is longit2.  It is equal to "4956" but for some reason after doing this to it:

Code: [Select]
longminutes = ((f - (int)f) * 100.0) + (atof(longit2)/10000.00);

the decimal piece ends up being .4954, and this is repeatable.....it is always .0002 off from the expected value.  I progressively printed out an evaluation of that statement and it maintains the correct value until I add the arguments in front of the plus sign.  What could be going wrong with this?  The calculation for latitude works fine....just the longitude is the problem.
Arduino Uno;
Mega328

PaulS

1000000 doesn't look much like an int. 1000000UL does, on the other hand, look like an unsigned long.

http://snippets-r-us.com might be able to help you. We can't verify that latit1, latit2, longit1, and longit2 are what you say they are (properly NULL terminated arrays of chars).

Intermediate variables, and Serial.print() statements, might help.

jerseyguy1996

Well I figured out where the problem is but I am not sure how to fix it.  To be honest, for this application this level of precision is ridiculous but it still just bugs me that it is coming out like this.  Basically, these two lines:

Code: [Select]
f = atof(longit1);
    Serial.println(f, 4);


result in the output:

Code: [Select]
7406.0000


and

Code: [Select]
f /= 100.00;
    longdegrees = (int)f; //isolate degrees
Serial.println(longdegrees);


results in

Code: [Select]
74

and

Code: [Select]
longminutes = ((f - (int)f) * 100.0);
Serial.println(longminutes,4);


results in

Code: [Select]
5.9998

which is wrong.....it should be 6.0000 based on the original value of f being 7406.0000.

Does anyone know why it may be losing .0002 on that value?  I don't think I am overrunning any of my datatypes.
Arduino Uno;
Mega328

PaulS

Quote
which is wrong.....it should be 6.0000

No. You are performing floating point arithmetic. The 5.9998 value is perfectly reasonable.

Print out f to 6 decimal places, after you divide it by 100.0, to see that.

KeithRB

"Working with floating point is like moving piles of sand. Every time you move one you lose a little sand and pick up a little dirt."

GoForSmoke

Time to look into fixed-point long int or long long int.

Or if you want kilometers to 6 places them work in millimeters and save the decimal point for displays only. That will save you from a lot of work and confusion while keeping the digits exact down to whatever unit you use.

I find it harder to express logic in English than in Code.
Sometimes an example says more than many times as many words.

jerseyguy1996


Quote
which is wrong.....it should be 6.0000

No. You are performing floating point arithmetic. The 5.9998 value is perfectly reasonable.

Print out f to 6 decimal places, after you divide it by 100.0, to see that.


Ahh.....74.059997.  So is there some fix or is this just something we all accept if we insist on working with floating points (and are too lazy to learn fixed point math)?
Arduino Uno;
Mega328

PaulS

Quote
So is there some fix

Yes. Put the Arduino back in the box, and send it back.

jerseyguy1996


Quote
So is there some fix

Yes. Put the Arduino back in the box, and send it back.


Ok point taken....no need to be snide :smiley-slim:
Arduino Uno;
Mega328

GoForSmoke

For some things float you can do small rounding but it's been ages since I did that.

I find it harder to express logic in English than in Code.
Sometimes an example says more than many times as many words.

retrolefty


"Working with floating point is like moving piles of sand. Every time you move one you lose a little sand and pick up a little dirt."


LOL  :D :D :D

OK, this might be a old well known phrase for some, but I've never heard it before and it sums up my experience and limited knowledge of floating point math. My rule is to just say no to FP in a arduino setting unless it's limited to having to receive numbers from some external source that you can not do anything about.

Just because the gcc compiler offers some functions, doesn't mean you have to use them or if they are even useful in a arduino environment.

Lefty

KeithRB

I got it from Kernigham and Plauger: "The Elements of Programming Style", first copyrighted in 1974. Most of the examples are in FORTRAN and PL/1! But it is still a great read about, well, programming style. They really pick on Daniel McCracken*.

They actually present the quote as from a "wise programmer". It is easy to find the quote in the book since the index lists "floating point numbers as sandpiles" 8^)


*For those that don't know, anyone who learned FORTRAN in the '70s and '80s probably used a text by McCracken.

GoForSmoke

Our FORTRAN book at Drexel had WAT4/WAT5 in the title. Must have been derivative.  ;)

Not that I had the $ to stay anyway. Punched the cards, waited for printouts, learned damned little.

When I got a TI-56 in Jan 78 I learned more in the 1st month than that semester at Drexel. I learned for example how badly I wanted indirect addressing! LOL! Got that in the TI-59.

I find it harder to express logic in English than in Code.
Sometimes an example says more than many times as many words.

jerseyguy1996

Well this works.....basically just got rid of that divide by 100.  It looks like this now.  Anything that I should be worried about doing it this way?

Code: [Select]
//cconvert longitude to GPS decimal format
    f = atof(longit1);
    Serial.println(longit1);
    longdegrees = (int)(f/100); //isolate degrees
    Serial.println(longdegrees);
    longminutes = (f - (longdegrees*100)) + atof(longit2)/10000.00;
    Serial.println(longit2);
    Serial.println(longminutes,6);
    longdecCoords = long((longminutes / 60.0)*1000000UL);
    if(EW[0] == 'W') longdegrees *= -1.0;
Arduino Uno;
Mega328

PaulS

Code: [Select]
    longdegrees = (int)(f/100); //isolate degrees
    Serial.println(longdegrees);
    longminutes = (f - (longdegrees*100)) + atof(longit2)/10000.00;
    Serial.println(longit2);
    Serial.println(longminutes,6);
    longdecCoords = long((longminutes / 60.0)*1000000UL);
    if(EW[0] == 'W') longdegrees *= -1.0;

AFTER you extract data from longdegrees, see if long degrees has the right sign. Fix it, if not. Well, OK. Fine by me.

Go Up