next up previous
Next: 6 Performance Up: An Analytic Technique to Previous: 4 Amplitude tracking

5 Noise

When noise $ N[n]$ is added in the right-hand side of Eq. (1), the solution in Eq.s (3), (7), and (8) is no longer accurate. In particular, since the noise $ N[n]$ is unknown, $ \cos(\phi-\theta) = (\Vert r-N\Vert^2 - A^2 - B^2)/(2AB)$ is also unknown. For small levels of noise, $ D$ may be defined as in Section II as an approximation for $ \cos(\theta-\phi)$. Or, the following correction can be applied to obtain a slightly improved, unbiased estimate:

$\displaystyle D \triangleq
(\Vert r\Vert^2 - \sigma^2 - A^2 - B^2)/(2AB) \approx \cos(\phi-\theta),$

where $ \sigma^2 = E[\Vert N\Vert^2]$. Similarly, Eq.s (7) and (8) may be bias-corrected with
$\displaystyle \hat A$ $\displaystyle \triangleq$ $\displaystyle {1 \over 2} \left(
\sqrt{X+{\pi\over 2}(Y-X) - \sigma^2}
+ \sqrt{X+ {\pi\over 2}(Z-X) - \sigma^2}\right)$  
$\displaystyle \hat B$ $\displaystyle \triangleq$ $\displaystyle {1 \over 2} \left(
\sqrt{X+{\pi\over 2}(Y-X) - \sigma^2}
- \sqrt{X+{\pi\over 2}(Z-X) - \sigma^2}\right).$  



Jon Hamkins 1999-10-29