Complex integral in spectrum estimator

51 Views Asked by At

I've been studying Fourier analysis of time series and in order to find an unbiased and consistent estimator for the spectrum, Barlett proposed a solution in 1953, the so called "smooth covariances estimator".

Problem is, for this estimator, there are some complex integrals with which I am not totally familiar yet. I am having trouble understanding a setp from the book. It says that:

$$ W(\lambda) = \int_{\infty}^{\infty} e^{-i\lambda \tau} \left(1 - \frac{|\tau|}{T}\right)d\tau $$

is the Fourier transform of this function $w(\tau) = 1 - |\tau|/T$, for $|\tau| \leq T $. Now the book states that:

$$ W(\lambda) = T\left[ \frac{sin(\pi T \lambda)}{\pi T \lambda}\right]^2 $$

and I cant figure out how I can get there. Can you guys give me any advice? Thanks!

EDIT

Trying to evaluate the integral, I have:

$$ \int_{-T}^{T}e^{-i\lambda \tau}(1-\frac{|\tau|}{T})d\tau = \int_{-T}^{T}e^{-i\lambda \tau}d\tau - \frac{1}{T} \int_{-T}^{T} |\tau|e^{-i\lambda \tau}d\tau $$

Now what I get is that:

$$ \int_{-T}^{T} e^{-i\lambda \tau}d\tau = \frac{e^{i\lambda T} - e^{-i\lambda T}}{i\lambda} = 2\frac{sin(\lambda T)}{\lambda T}$$ which seems quite reasonable.

Breaking piecewise $\frac{1}{T} \int_{-T}^{T} |\tau|e^{-i\lambda \tau}d\tau $:

$$\frac{1}{T} \int_{-T}^{T} |\tau|e^{-i\lambda \tau}d\tau = \frac{1}{T} \int_{0}^{T} \tau e^{-i\lambda \tau}d\tau - \frac{1}{T} \int_{-T}^{0} \tau e^{-i\lambda \tau}d\tau $$

Since:

$$\frac{1}{T} \int \tau e^{-i\lambda \tau}d\tau = \frac{e^{-i\lambda \tau}(1+i\lambda \tau)}{\lambda^2 T} $$

The resulting integral is:

$$ \frac{1}{T} \int_{-T}^{T} |\tau|e^{-i\lambda \tau}d\tau = \frac{2(\lambda T sin(\lambda T) + cos(\lambda T)-1)}{T \lambda^2 } $$

Computing both integrals together, still my result doesn't match what I expected. Where am I mistaken?