Writing a motion cueing software from scratch.

Native Simtools 2.4 “third order” EMA Smoothing:

True!!! Very true!!!

I can also tell you where your butt out-performs your eyes:
It's those moments where the "curvature-comb" pops out the graph. Where acceleration sets on. Hence "acceleration ONSET cueing". By choosing the right filters, you make a deliberate decision on exactly what to "show" your butt; In what way you present the onset of acceleration to your butt.


In these pictures the magenta curve doesn't seem to look too different to the naked eye, does it? It looks just,... well,... curved But the curvature comb shows that the curvature onset (=acceleration onset) is very different. Those moments are what your butt notices a lot better than your eyes.

I made a little comparison between two platforms.
- One with 1st and 2nd order LP filters (graph has corners/radii)
- Another with 2nd and 3rd order LP filters (graph has radii/curvature continuity)



Important:
Note how both platforms start and stop at the same time! However, see how one platform "explodes" into action in surge (corners in graph) while the other transitions more smoothly (radii).

YAY-HOOORAY!!! COOL! Looks very nice.

When you try it out and play around with it, could you do me the favour and see if it makes a difference of wether the CoR is near your neck vs. chest vs. belly? I could come up with good arguments for all. Maybe it even depends on personal preference.

...but certainly nice, smooth motion! Thank you, @yobuddy !!!

I know, I'm kinda over-stressing this whole filter thing,.. but just for completeness' sake:


Dirty

Might be paranoia, but in the animation I feel like I still see the bump! that sudden upward movement looks too sudden to my eye. If you have this implementation though, I imagine I would be able to smooth the onset further?

Ultimately what matters is that you guys are on the case! You are definitely focused here on addressing the single biggest complaint I've had. For a long time seemed like I was the only one bothered by this so I didn't have much hope of it getting fixed in the software. Ian at BFF seemed to think I was crazy when I tried to ask him about this problem. =P

Dirty, In Ian's BFF simulation you can define the CoR as a value in mm above the upper joints.

Now what I'm about to tell you may very well be unique to my sim, or unique to BFF, but I found that if I use this setting as intended for some reason the movement of the sim gets much less smooth. I wind up setting it to a very low value or even zero in the end for smoothness.

A more direct answer though, I have been told that the CoR in most cases should be set approximately the height of your belly button or a couple inches above.

By the way, you guys keep talking about 2.4 but I don't find any hints at just how "soon" we are talking about. Is it "soon", or is it "Soon™"? =P

Wish me luck, I'm finally about to start installing these limit switches. Step one is finding exactly where they need to go. Once they are installed and tested next step is to set up SimTools and start testing and tuning motion!

I would agree, and that is pretty much my approach as well. However, I can well understand the arguments for putting it near your neck (vestibular system) as well. I would love to do a blindfolded test, where someone changed the CoR without the occupant knowing where it is set. I'd love to know if someone could tell if he didn't know beforehand

I'm implementing Butterworth filters, just for testing.
Not sure if they are better.
Need to make some speed tests.
But here's some results in the new program.

In gray, the signal received. A square wave.
In red the filtered values.

Here a result for a 2nd order Butterworth low pass:


And 4th order Butterworth low pass:


EMA LP(LP(LP)) vs Butterworth 4th order (arrows for Butterworth):


Edited: For better clarification...

I am surprised to see the overshoots with your “second order” LPF

Is it a “simple EMA” or a “double EMA” (DEMA) ?

Normally, the overshoots comes from a DEMA math. Saying it because I don’t have any overshoots on a “simple EMA” even with second or third order.

Could you maybe provide a pseudo-code of the math which are in use in your examples ?

Overshoot is only for Butterworth, not EMA
Maybe badly explained.
Only the last graph has superimposed an EMA LP(LP(LP)) with a 4th order Butterworth.
And the EMA as no overshoot.

Great stuff, guys!!
It seems a good time to consider building a 6dof flight sim,,,,,

Thanks

If I understand well, with second, third Butterworth order, you can reach the step destination sooner as with second, third order EMA, but you trade it with an overshoot.

Complex math. Would certainly worth a try exploring its possibilities.

You can tell! I've tried cog above belly button and at seat bottom. The higher above seat bottom the smoother rotations feel since your upper body doesn't move as much. But you also now notice your butt rotating. It might be okay for flight depending on the model being flown but it doesn't feel correct at all when racing. It wasn't on a 6dof however.

@Dirty, as you’re very familiar with excel, could you please in your spare time present a graph with two following scenarios:

1.)
first order EMA filtered with third order EMA
2.)
third order EMA filtered with first order EMA

I am expecting a quiet different entry and exit curve.

Thanks

Oh, I am not an expert on filters. Only a "recreational user", so to speak. Maybe @Tim McGuire or @harwoodr can help out here with some background knowledge?

My arguments:
I'd be very surprised if there were a difference . Let's call Ln = Lowpass Filter of n-th order (e.g. L3 = 3rd order Lowpass)

You want to know if an L1 followed by an L3 is the same as an L3 followed by an L1. So, the question is:

L1 + L3 = L3 +L1 Is this statement true?

We already know that... L3 = L1 + L1 + L1 ...because that is how L3 is defined So, in the first statement we can substitute the L3 with L1 + L1 + L1 and get...

L1 + L1 + L1 +L1 + L1 = L1 + L1 + L1 + L1 ...which is obviously true. Therefore...

L1 + L3 = L3 +L1 must be true as well.

Or in plain English:
A third order lowpass filter is the same as three first order lowpass filters in sequence. If you want to add another identical first order lowpass filter, it doesn't make a difference whether you add it before (L1 + L3) or after (L3 + L1).

Or in plain Math:
EMAs are commutative! (I think)

------------------------------------------

But are they still commutative when they have different filter variables?...

Let's call L(x) = Lowpass filter with filter variable x (e.g. L(0.2) = 1st order lowpass filter with 20% adoption rate per frame)



you can switch back and forth between the images to see the difference. The resulting (blue) output line does not only look identical, it actually is the same values. I've done the math, the difference is zero

For all I can say:

L(a) + L(b) = L(b) + L(a)

Dirty


Note: In this case "+" is not the addition operator. The filter outputs are not being added. I am not 100% sure, but I think "concatenation" or "convolution" are better terms. Maybe someone can clarify that.

Oops,....

You asked for a graph

It doesn't make a difference whether you...
...take a 1st order filter (yellow) and then three more to reach the 4th order filter (red), or...
...take a 3rd order filter (green) and then another 1st order to reach the 4th order filter (red).

What I learned about filters:
-You want to turn corners into radii into transitions? --> Increase filter order.
-You want to "shallow out" the graph and avoid steep gradients? --> Use lower filter variable.

That pretty much sums it up

Nice work Dirty. You have no idea how excited I am for the work you guys are doing. I'm finally going to get rid of the potholes (that's what we call big holes in the road in case that's unfamiliar to anyone) in my virtual air!

Thanks @Dirty for your involvement !

I did some more tests and yes, indeed, it seems that I started to think about the filtering superpositions with wrong assumptions.

However…
As It's time soon to decide what has to be behind SimTools sliders, I am drilling more into the topic:

@yobuddy , what's in the vid can also catch your interest.



You notice 1st order EMA (C-curve) followed by another 1st order EMA with 3 different weighting factor making the response more and more smooth going into the S-curve response.