01.09.2014 Views

Project Cyclops, A Design... - Department of Earth and Planetary ...

Project Cyclops, A Design... - Department of Earth and Planetary ...

Project Cyclops, A Design... - Department of Earth and Planetary ...

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Let us now introduce the normalized variables<br />

x = e/Po <strong>and</strong> r = Pr/Po. Then equation (D35) becomes<br />

For large n, we may replace (n - 1)! by<br />

1<br />

p(x) = e -(r + x) lo(2 V_) (D36)<br />

n ! 2rr nn e 12n<br />

n<br />

N_<br />

--n÷--<br />

If we add n independent samples <strong>of</strong> the output the<br />

resulting distribution will be the n-fold convolution <strong>of</strong><br />

p(x). The Fourier transform <strong>of</strong> p(x) is given by Campbell<br />

arid Foster (ref. 1, pair 655.1) as<br />

C(co) - 1 e-' +ico exp<br />

giving<br />

pnfy) =<br />

1<br />

-n(r - 1 + -- + y )<br />

n -n- 1 12n 2<br />

--y e ×<br />

2rr<br />

+ _ + _ + etc<br />

1 ° n 2(n + 1)<br />

Thus<br />

(D39a)<br />

[C(c°]n<br />

-nF<br />

(1 +leo) n exp<br />

<strong>and</strong> by pair 650, op. cit., we find the inverse transform<br />

to be<br />

n-I<br />

2<br />

e-X In -<br />

1 (2 vthr-x)<br />

Actually equation (D39a) is in error by only 0.2% for<br />

n = 1 <strong>and</strong> so may be used to compute pn(y) for any n.<br />

The probability that y is less than a certain threshold<br />

YT may be found by numerically integrating equation<br />

(D39a) from y = 0 to y = YT" This integral gives the<br />

probability that signal plus noise fails to exceed the<br />

threshold YT <strong>and</strong> hence that the signal is not detected.<br />

In the absence <strong>of</strong> a signal, r = 0, <strong>and</strong> equation (D39)<br />

becomes<br />

(D37)<br />

If now we replace the sum x by the arithmetic mean<br />

y = x/n, we obtain<br />

n-I<br />

Pn(Y) = n(_) 2 e-n(_ +y) in -1 (2n,,/ff)<br />

(D38)<br />

(ny) n - I e-nY<br />

pn(y) = n (n- I)! (D40)<br />

This may now be integrated from YT to infinity to<br />

give the probability qn(YT ) that noise alone exceeds the<br />

threshold YT" We find<br />

n-I<br />

Now<br />

in _ _(2nvr_7) = (n_)n<br />

SO<br />

nnyn - le-n(r + y)<br />

Pn('V) = (n - 1)!<br />

Q<br />

- l<br />

nZry + 2(n + 1)<br />

oo<br />

_<br />

k=O<br />

+ n2ry 1m ×<br />

(n2ry)<br />

k!(n-l+k)!<br />

( 1<br />

1+ 3(n+2) +"<br />

k<br />

(D39)<br />

qn(YT) = e -nyT _ (nyT)k (I341)<br />

k=O k!<br />

The threshold level required to give a specified p),a is<br />

found from equation (D41). The signal-to-noise ratio, r,<br />

needed to give a permissible probability <strong>of</strong> missing the<br />

signal is then found by integrating equation (D39a) from<br />

y = 0 toy = yT with trial values <strong>of</strong>r.<br />

GAUSSIAN<br />

APPROXIMATIONS<br />

From the central limit theorem we know thal as n -* o_<br />

both (D38) <strong>and</strong> (D40) will approach gaussian distributions.<br />

Reverting to unnormalized variables, the distribution<br />

approached by the signal plus noise will be centered<br />

at P = Po + Pr <strong>and</strong> will have a variance<br />

193

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!