Rube Goldberg Code

[altenbach]

---

craigc wrote:

You have just solved a problem I have been having with reading phase difference from a Network analyser.  I knew there must be a way of getting the phase's back to (-pi, pi). 

---

Sorry, actually my method only gets you to two quadrants, i.e. [0..pi] for cosine and [-Pi/2..Pi/2] for sine, of course. In order to reconstruct the data for all four quadrants you need more information.

One possibility would be to take the sine and cosine and feed the two outpus to the "Inverse tangent (2 inputs)" (see image)

Of course a simpler solution would be to divide by 2Pi and keep the remainder for an output range of [0..2Pi]. No trigonometry needed. 🙂

You can easily tweak for [-Pi..Pi]. (see image)

Message Edited by altenbach on 11-24-2008 08:57 AM

[TCPlomp]

Why a seperate sine and cosine?

 

Ton

Message Edited by TonP on 11-24-2008 08:51 PM

[RobCole]

Oh ho. Did Altenbach get caught in a minor Rube Goldberg?

     Rob

[altenbach]

---

TonP wrote:

Why a seperate sine and cosine?

---

Good catch. In this case I had the sin and cos already on the diagram from some earlier editing, so it was easier to recycle them. 🙂

[Intaris]

Plus that way you know the Sine and Cosine functions can run in Parallel, possible improving performance on multi-core processors.

Shane. 

[Ben]

---

Intaris wrote:

Plus that way you know the Sine and Cosine functions can run in Parallel, possible improving performance on multi-core processors.

Shane. 

---

Now a benchmark test to prove which method is faster and answer if it will really run in two cores at the same time would be an interesting way to show that here are no chinks (sp?) in our Knight's armour.

But....

So far I have learned that the inverse operations when constructed using mth operations does serve a purpose (In not equal to out).

Ben

[altenbach]

---

Intaris wrote:

Plus that way you know the Sine and Cosine functions can run in Parallel, possible improving performance on multi-core processors.

---

I would not make conclusions without benchmarks. I can think of at least two scenarios where the "combo" primitive would be preferable.

• Sin and cos are strongly related (e.g. sin²(x)+ cos²(x) = 1), so calculating one and derive the other could be cheaper than calculating both. I really don't know how many cpu cycles each would take.
• It is conceivable that the combo version could be written so it internally could use multiple cores. I wonder if 8.6 does that?

I'll do some bechmarks to quantify things. 🙂

[altenbach]

Random sightings.... 

(found here)

Message Edited by altenbach on 11-25-2008 01:23 PM

[RobCole]

---

altenbach wrote:

Random sightings.... 

But you do have to admit that the top down processing of property nodes means that the value is written before it is read.

Okay, there are definitely better ways of doing these types of things which I won't enumerate.

     Rob

Message Edited by Robert Cole on 11-25-2008 02:29 PM

Message Edited by Robert Cole on 11-25-2008 02:30 PM

[altenbach]

---

Robert Cole wrote:

But you do have to admit that the top down processing of property nodes means that the value is written before it is read.

---

Yes, you can even expand on the ideas and calculate 2x+1 as follows. 😄

Message Edited by altenbach on 11-25-2008 01:46 PM