2
$\begingroup$

I want to calculate the probability $P(X\le-0.5)$. I have the probability density function $1/(\pi\sqrt{A^2 - x^2})$. $A$ is supposed to be the square root of noise power, $A = \sqrt{N_0}$, and so changes according to the $N_0$ value given by SNR. The signal power is always $1$, so $\mathrm{SNR} = 1/N_0$. $N_0$ is then $N_0 = 10^{-\mathrm{SNR}(\mathrm{dB})/10}$. I do the following:

1) I write the cumulative distribution function as

$$\int_{-A}^{t} \ \frac{1}{ \pi \sqrt{A^2-x^2}}dx$$

2) Rewrite CDF as

$$\int_{-A}^{t} \ \frac{1}{ \pi A \sqrt{1-(x/A)^2}}dx$$

3) Change variable setting $y=x/A$

4)

$$\int_{-1}^{t/A} \ \frac{1}{ \pi \sqrt{1-(y)^2}}dy$$

Then Mathematica simplifies the equation to

$$ \frac{\frac{\pi}{2}+\arcsin( \frac{t}{A} ) }{\pi}$$

with the condition Re[t/A] <= 1 || t/A is not real.

I want to plot the CDF, but $\arcsin$ only gives real values in the range $[-1,1]$ and is complex outside it. So what can I do for the range of $A$ & $N_0$, in order to get the most range of SNR?

$\endgroup$
  • 1
    $\begingroup$ What did you expect? Your probability density is also complex outside $[-A,A]$, so why shouldn't the CDF be? $\endgroup$ – Lukas Lang Sep 30 '17 at 14:46
6
$\begingroup$
f[x_] = 1/(Pi*Sqrt[A^2 - x^2]);

For the PDF to be real

FunctionDomain[f[x], x]

(* A^2 - x^2 > 0 *)

Verifying that f[x] is normalized

Assuming[A > 0, Integrate[f[x], {x, -A, A}]]

(* 1 *)

The distribution is then

dist = ProbabilityDistribution[f[x], {x, -A, A}, Assumptions -> A > 0];

CDF[dist, x] // Simplify

enter image description here

Show[
 Table[
  Plot[Tooltip[CDF[dist, x], A], {x, -5.5, 5.5},
   PlotStyle -> ColorData["Rainbow", A/5]],
  {A, 1, 5}],
 PlotRange -> {{-5, 5}, {0, 1}}]

enter image description here

EDIT: Added assumption to FullSimplify to get simpler form

CDF[dist, -1/2] // FullSimplify[#, A > 0] &

enter image description here

Plot[%, {A, 1/2, 5}, PlotRange -> {0, 1/2}]

enter image description here

$\endgroup$
5
$\begingroup$

Your PDF is actually a special case of the type I Pearson distribution:

pdist = PearsonDistribution[1, -1, 0, -1, 0, A^2];

Refine[PDF[pdist, x], A > 0]

$$\begin{cases}\frac1{\pi\sqrt{\mathtt A-\mathtt x}\sqrt{\mathtt A+\mathtt x}}&-\mathtt A<\mathtt x<\mathtt A\\0&\mathtt{True}\end{cases}$$

Assuming[A > 0, FullSimplify[FunctionExpand[CDF[pdist, x]]]]

$$\begin{cases}1&\mathtt x\geq\mathtt A\\\frac{2\arcsin\sqrt{\frac{\mathtt x}{2\mathtt A}+\frac12}}{\pi }&-1<\frac{\mathtt x}{\mathtt A}<1\\0&\mathtt{True}\end{cases}$$


As it turns out, ArcSinDistribution[] is built-in:

PDF[ArcSinDistribution[{-A, A}], x]

$$\begin{cases}\frac1{\pi\sqrt{\mathtt A-\mathtt x}\sqrt{\mathtt A+\mathtt x}}&-\mathtt A<\mathtt x<\mathtt A\\0&\mathtt{True}\end{cases}$$

CDF[ArcSinDistribution[{-A, A}], x]

$$\begin{cases}\frac{2\arcsin\left(\frac{\sqrt{\frac{a+x}{a}}}{\sqrt{2}}\right)}{\pi}&-a<x<a\\1 & x\geq a\\0&\mathtt{True}\end{cases}$$

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.