[SDK Question] Distance calc overrun ?

[JMW]

Hi People.

Quick question, probably not quick answer.....

On a distance calculation I'm getting a value of "-1.#IND" intermittently.

It's messing up an autopilot.

Is it an overrun of some kind.

If so, what do I need to do....?

Code used:

dis = sin((latitude)*RAD) * sin((tgtlat)*RAD) + cos((latitude)*RAD) * cos((tgtlat)*RAD) * cos((longitude-tgtlng)*RAD);
  dis = acos(dis);
  dis = (dis)*DEG;
  dis = dis * 6371;

[Urwumpe]

The difference in longitude should be absolute:

fabs(longitude-tgtlng)

[martins]

Quote:

Originally Posted by JMW

 Hi People.

Quick question, probably not quick answer.....

On a distance calculation I'm getting a value of "-1.#IND" intermittently.

It's messing up an autopilot.

Is it an overrun of some kind.

If so, what do I need to do....?

Code used:

dis = sin((latitude)*RAD) * sin((tgtlat)*RAD) + cos((latitude)*RAD) * cos((tgtlat)*RAD) * cos((longitude-tgtlng)*RAD);
  dis = acos(dis);
  dis = (dis)*DEG;
  dis = dis * 6371;

The only function that could potentially throw an error I can see here is the acos. Did you check that the input is valid (range [-1,1])? Even if your calculation of the argument is mathematically correct, there is always the possibility of rounding errors.

Quote:

Originally Posted by Urwumpe

 The difference in longitude should be absolute:

fabs(longitude-tgtlng)

Would that make a difference here, given that cos(-x) = cos(x)?

[Urwumpe]

Quote:

Originally Posted by martins

  Would that make a difference here, given that cos(-x) = cos(x)?

Not really, but the other terms should not exceed 2.0 in any case and not 1.0 if the shortest great circle distance is searched. Was just the first difference to the reference I noticed

[martins]

Also, for the cases where it doesn't fail, does it give you the expected result? I am a bit concerned about those lines:

  dis = (dis)*DEG;
  dis = dis * 6371;

Is that making the approximation sin(x) ~ x (i.e. small distances)? Even then, converting the central angle distance to degrees rings alarm bells here ...

[JMW]

Gentlemen, sorry for the delay, have been checking a few things.

Urwumpe: I've tried 'fabs(longitude-tgtlng)' and it sent the calcs haywire.

martins:
By 'the input' do you mean the result of line 1 ?
That does seem to be within -1 && 1 always.

Yes, when it doesn't read "-1.#IND" it gives a perfect distance.

It's occurring whilst hovering directly over a pad, so very small distances are involved....

I got that formula from a website and also from a post in the forum, but can't remember whose.

[Urwumpe]

Quote:

Originally Posted by JMW

 Urwumpe: I've tried 'fabs(longitude-tgtlng)' and it sent the calcs haywire.

That shouldn't happen at all, since it, as martins pointed out, should be just a superfluous term if cos is properly implemented.



"-1.#IND" is the windows version of telling you "NaN" or "not a number", which occurs, when the result of an floating point operation is undefined.

acos(1) and acos(-1) should be no NaN, but acos(1.0000000000000000000000000000000000000000000 000000000000000000000000000000001) would be one.



Your formula is also mentioned in Wikipedia for great circle distance and in the math book I used at university, but always with an absolute definition of difference in latitude.

[Marijn]

Have a look at this site with formulas. Under the head "Spherical Law of Cosines", it says that you might need a different formula for very small distances: http://www.movable-type.co.uk/scripts/latlong.html

[JMW]

Urwumpe: Apologies, I didn't implement fabs correctly. Is fine now.

Marijn: I'll take a look there later, (RL taking over for now)

Sounds like the small distance might be the culprit.

Thanks all.

---------- Post added at 07:48 PM ---------- Previous post was at 07:30 PM ----------

martins:
What did you use for distance calc in VOR/VTOL mfd ? (if it's not classified)

---------- Post added 17-09-19 at 05:48 PM ---------- Previous post was 16-09-19 at 07:48 PM ----------

martins: Can it be "open source" ? ^ ^ ^ ^ ^
I'm curious because there is about 0.22 km difference (mine greater)
at 2 km, reducing to 0.11 at 1 km, 0.05 at 500 mtrs, down to less than 0.09 at zero.

[Urwumpe]

Quote:

Originally Posted by JMW

  I'm curious because there is about 0.22 km difference (mine greater)
at 2 km, reducing to 0.11 at 1 km, 0.05 at 500 mtrs, down to less than 0.09 at zero.

Which radius did you use and what is the true radius?

[JMW]

6371 km.........

---------- Post added at 06:42 PM ---------- Previous post was at 06:37 PM ----------

True... depends where you are ?

Quote:

The radius of Earth at the equator is 3,963 miles (6,378 kilometers), according to NASA's Goddard Space Flight Center. However, Earth is not quite a sphere. The planet's rotation causes it to bulge at the equator. Earth's polar radius is 3,950 miles (6,356 km) — a difference of 13 miles (22 km).

[martins]

Quote:

Originally Posted by JMW

 What did you use for distance calc in VOR/VTOL mfd ? (if it's not classified)

martins: Can it be "open source" ? ^ ^ ^ ^ ^

Let me just quickly declassify this for you. Here is the relevant VOR/VTOL MFD code:

void Orthodome (double lng1, double lat1, double lng2, double lat2,
				double &dist, double &dir)
{
	double A = lng2-lng1;
	double dlng = fabs(A);
	double dlat = fabs(lat2-lat1);
	if (dlat < 1e-14) {
		dist = dlng;
		dir = (lng2 > lng1 ? PI05 : 3*PI05);
		return;
	} else if (dlng < 1e-14) {
		dist = dlat;
		dir = (lat2 > lat1 ? 0.0 : PI);
	} else {
		double sinA  = sin(A),    cosA  = cos(A);
		double slat1 = sin(lat1), clat1 = cos(lat1);
		double slat2 = sin(lat2), clat2 = cos(lat2);
		double cosa  = slat2*slat1 + clat2*clat1*cosA;
		dist = acos (cosa);
		dir = atan2 (sinA*clat2, clat1*slat2 - slat1*clat2*cosA);
		if (dir < 0.0) dir += Pi2;     // 0 <= dir < 2pi
	}
}

Orthodome (sp->lng, sp->lat, tlng, tlat, adist, bdir);
bdir -= sp->dir;
if      (bdir <  0.0) bdir += Pi2;
else if (bdir >= Pi2) bdir -= Pi2;
dist = adist * sp->ref->Size();

Let me know if you spot a mistake.

[JMW]

A "mistake" ?
No radius !
We're now in sync

[martins]

So, was the discrepancy just down to a difference in radius? It sounded quite significant.

[JMW]

Dunno.

I ended up adding this

dist = dist *6371;
deleting
sp->adist * ref->Size();