Source estimation for identification of anomalous events

74 Views Asked by At

I’m stuck on the following problem.

There are two sources $S_A$ and $S_B$ at the ends of a channel. Both are made up of a white noise component $W_i$ plus an impulsive component $I_i$:

$$ S_A = W_A + I_A $$

$$ S_B = W_B + I_B $$

Two receivers are co-located with the sources. The first one records the signal

$$ X_A = S_A + S_B H_{AB} $$

the second one records the signal

$$ X_B = S_A H_{AB} + S_B $$

where the channel $H_{AB}$ is known and introduces a delay and an attenuation

$$ H_{AB} = exp(-j 2\pi f \frac{x}{v})exp(-\alpha x) $$

In this case one can estimate $S_A$ and $S_B$ simply by:

$$ S_A = \frac{X_A - X_B H_{AB}}{1 - H_{AB}^2} $$

$$ S_B = X_B - S_A H_{AB} $$

The problem is when an anomalous event $N$ appears somewhere in the middle of the channel and propagates towards A and B:

enter image description here

The recorded signals become

$$ X_A = S_A + S_B H_{AB} + N H_{NA} $$

$$ X_B = S_A H_{AB} + S_B + N H_{NB} $$

where both $N$, $H_{NA}$ and $H_{NB}$ are unknown.

The knowledge I have about $N$ is:

  • $N$ is a single impulsive event,
  • Its power is below the power of the noise components of $S_A$ and $S_B$.

If I apply the previous method, $N$ is spread on both $S_A$ and $S_B$.

I need to estimate either the sources $S_A$, $S_B$, or $N H_{NA}$ or $N H_{NB}$.

Any suggestion?