I have the following transform:
$$F(y) = \int_{0}^{\infty} y\exp{\left[-\frac{1}{2}(y^2 + x^2)\right]} I_0\left(xy\right)f(x)\;\mathrm{d}x$$
with the following conditions:
- $f(x)$ and $F(y)$ must be real and positive (they are continuous probability distributions)
- $x,y$ must be real and positive (they are magnitudes)
@Carlo suggested the following approach. Re-arrange such that we have: $$\frac{F(y)}{y}\exp{\left[\frac{1}{2}y^2\right]} = \int_{0}^{\infty} \left(\frac{f(x)}{x}\exp{\left[-\frac{1}{2} x^2\right]}\right) I_0\left(xy\right)x\;\mathrm{d}x$$
Now define:
$$ G(y) = \frac{F(y)}{y}\exp{\left[\frac{1}{2}y^2\right]} \quad\quad g(x) = \frac{f(x)}{x}\exp{\left[-\frac{1}{2} x^2\right]}$$
We can now write: $$G(y) = \int_{0}^{\infty} g(x) I_0\left(xy\right)x\;\mathrm{d}x$$
Make the substitution $y = iq$ along with the fact that $I_0(ixq) = J_0(-xq) = J_0(xq)$ and we have:
$$G(iq) = \int_{0}^{\infty} g(x) J_0\left(xq\right)x\;\mathrm{d}x$$
This is a Hankel transform, which has an inverse: $$g(x)=\int_{0}^{\infty}G(iq)J_{0}(xq)q\;\mathrm{d}q$$
Substituting in $q = \frac{1}{i}y$ and $\mathrm{d}q = \frac{1}{i}\mathrm{d}y$ we obtain:
$$g(x)= -\int_{0}^{i\infty}G(y)I_{0}(xy)y\;\mathrm{d}y$$
Substituting out $G(y), \;g(x)$:
$$\frac{f(x)}{x}\exp{\left[-\frac{1}{2} x^2\right]} = -\int_{0}^{i\infty} \frac{F(y)}{y}\exp{\left[\frac{1}{2}y^2\right]} I_{0}(xy)y\;\mathrm{d}y$$
and finally solve for $f(x)$:
$$f(x) = -\int_{0}^{i\infty} x\exp{\left[\frac{1}{2}(y^2 + x^2)\right]} I_0\left(xy\right)F(y)\;\mathrm{d}y$$
Is this inversion valid given the aforementioned conditions?
How can I perform this integral numerically?
I'm directly measuring $F(y)$ (by normalising the histogram of a large set of values) and so I need to calculate $f(x)$ by doing this integral numerically. However I don't understand how to do this when the data is all real but the integral is over the imaginary axis. Can anyone help?
Thank you!