The problem of blind identification of source signals is to estimate the source signals without knowing the transmission channel, nor. the source signals themselves. From first glance, it seems that this is an ill-posed problem. However, it is shown that under the assumption of independent sources with distinct kurtosis, we could identify the source signals up to a diagonal matrix λ and a permutation matrix ρ. We term the estimates waveform-preserving estimates in that in most of the applications, we could use them to represent the original source signals. A blind identification algorithm, based on the second and fourth order moments of the observation signals is presented. Simulation reveals that using the algorithm, we could separate the source signals satisfactorily....[ Read more ]

The problem of blind identification of source signals is to estimate the source signals without knowing the transmission channel, nor. the source signals themselves. From first glance, it seems that this is an ill-posed problem. However, it is shown that under the assumption of independent sources with distinct kurtosis, we could identify the source signals up to a diagonal matrix λ and a permutation matrix ρ. We term the estimates waveform-preserving estimates in that in most of the applications, we could use them to represent the original source signals. A blind identification algorithm, based on the second and fourth order moments of the observation signals is presented. Simulation reveals that using the algorithm, we could separate the source signals satisfactorily.

In some applications, the source signals do not arrive at the sensors at the same time. This happens, for example, in array signal processing when wide-band or near fieId source signals are present. Under this circumstance, we have to modify the system model and perform blind identification in frequency domain. This is referred to as wide-band blind identification problem. A novel approach based on narrow-band blind identification and clustering is proposed. This approach is shown to be robust to the presence of uncertainties such as unknown sensor and channel gains, unknown combinations of near-field and far-field sources, unknown source spectral characteristics.

Besides signal separation, blind identification finds its application in diverse areas. We give in this thesis an application of blind identification on adaptive control problem. A new universal self-tuning compensator is proposed. Making use of a new blind identification algorithm, it is shown that the transfer function of an unknown, possibly time varying plant can be identified using only output. Simulation shows that using this new compensator, self-tuning could be obtained satisfactorily even under abrupt changes in the plant.