Non-Coherent Direction of Arrival Estimation and Localization Based on Sensor Arrays

Direction of arrival (DOA) estimation is an important research area in array signal processing.
Existing works normally assume the phase information obtained at sensors of the array
are accurate and therefore their performance degrades when the phase information is not
reliable. For such a scenario, a new class of non-coherent DOA estimation methods has been
proposed, where only magnitude information at an array of sensors is required. However,
the phase retrieval based methods always require one or more reference signals to resolve
ambiguity issues and fail to exploit the multi-snapshot information effectively. In this thesis,
to efficiently and effectively exploit multiple snapshots usually available at an array, a fasT
grOup sparsitY Based phAse Retreival (ToyBar) algorithm is proposed to solve the noncoherent
DOA estimation problem. To avoid the use of reference signals, an effective array
structure based on two uniform linear sub-arrays is proposed first for non-coherent DOA
estimation. Unambiguous DOA can be found either by exploiting the non-linear property
of sinusoidal function with the aid of the extra measurements provided by the second array,
or applying the ToyBar algorithm to the whole array directly. Instead of using ULAs,
uniform circular arrays (UCAs) can also be employed to overcome the ambiguities arising
in non-coherent measurements. In addition, an off-grid model involved with a bias vector
is proposed and a two-step method based on this model is further developed. Moreover,
a two-dimensional localization method with an off-grid signal model is proposed for the
non-coherent source localization problem based on distributed sensor arrays, where each
platform employs a UCA. Finally, the non-coherent method has been extended to wideband
signals, where the signal model is formulated with convolutional sparse coding (CSC) in the
time domain directly.