How to calculate what size fixed point type I need?

[David99999]

Might be a stupid question but I know I'm limited on the values I can display when it comes to fixed point type. In my FPGA, I'm calculating power factor. But my fixed point value is coming out to a 1, when in reality its more like 0.98 or 0.97. 

Is there a way to figure out the correct word length for my FXP based on the value I need to display?

[Terry_ALE]

This is a range analysis.  Have you seen the fixed point documents such as: https://zone.ni.com/reference/en-XX/help/371599N-01/lvfpgaconcepts/fpgafixedpoint/

You may need more bits on the fractional side of things.  Can you probe to see where the values saturate?

[GerdW]

Hi David,

when configuring the FXP datatype you will see the range (min/max value) and the resolution (delta) in the properties dialog.

Is there any information else needed?

[David99999]

Well, the thing is is I can't change everything inside the high throughput cosine function. And I THINK this might be where my problem is.

[Terry_ALE]

The output has 52-19-1=32 bits for the fractional side there shouldn't be rounding.  That is more than enough to see the decimal values you mention.  I think something else is your issue.

[David99999]

---

@Terry_ALE wrote:

The output has 52-19-1=32 bits for the fractional side there shouldn't be rounding.  That is more than enough to see the decimal values you mention.  I think something else is your issue.

---

Thank you, I'll keep digging!

[David99999 · Accepted solution]

Alrighty. The problem was that I was inputting radians into the cosine function, and not pi radians. I did not know there was a difference! I needed to divide by pi a second time.

But I do understand FXP's a little better now so thanks!

[rolfk]

---

@David99999 wrote:

Alrighty. The problem was that I was inputting radians into the cosine function, and not pi radians. I did not know there was a difference! I needed to divide by pi a second time.

But I do understand FXP's a little better now so thanks!

---

But you multiply with pi in your snippet so how can you divide a second time?

[GerdW]

Hi David,

---

@David99999 wrote:

The problem was that I was inputting radians into the cosine function, and not pi radians. I did not know there was a difference!

---

What is "pi radians" exactly? What is the difference between "radians" and "pi radians"?

Other question:

In your image there is a "x" input defined as "±,52,19": Do you really need an input range of ±524288 for the SinCos calculation? Why is the range that large for a function basically defined from 0…2pi?

Do you really need a FXP resolution of 32 bits? What kind of data do you want to handle? (Which accuracy is needed for your calculations?)

[rolfk]

I think the problem was that the normal floating point functions expect an input of radians (0 - 2pi) while the FPGA functions expect a normalized input (0 - 1). That fact is documented because is allows a significant more efficient implementation in FPGA fabric. 

The thing is, in the earlier image there was a division by 180 and then a multiplication by pi and then he talks about an additional division by 2pi, which would be simply a single division by 360. And yes if you make sure to limit the input to 0 - 360 respective 0 to 1 you can limit the FXP range to have one integer bit only and reduce the size significantly.

Strictly speaking you can limit the range to [0.0 - 360.0) respective [0.0 - 1.0), so excluding the upper limit and then you only need an FXP with enough fractional part bits. Depending on the accuracy your angle needs/can have a 16,0 FXP number (would be better than 4 digits or 0.1 degree accurate) could be more than enough.