Search

This paper considers the objective performance evaluation of kernel-based time-frequency distributions (TFDs) using several concentration performance measures and resolution examination through a deep analysis of time slice plots. On the other hand, the numerical complexity of each TFD is evaluated; a parameter that is particularly critical when real-time implementation is intended. The performance of TFDs based on time-lag kernels with compact support (KCS) namely the Cheriet–Belouchrani (CB), the separable (CB) (SCB) and the polynomial CB (PCB) TFDs is compared to the well-known kernel-based TFDs using several tests on real-life and multicomponent signals with linear and nonlinear frequency modulation (FM) components including the noise effects and the influence of the kernel length. In all presented examples, the time-lag KCS TFDs, and particularly the PCB TFD, provide the best compromise between highest autoterm resolution and interference rejection while still requiring moderate computational costs thanks to the compact support nature of their kernels that reduces the number of points needing computation. On the other hand, the derived distributions do not require any smoothing window neither in time nor frequency in order to achieve the best time-frequency resolution. Furthermore, they have an extremely interesting practical advantage since their adjustment is performed by simply changing a single parameter which is integer for the PCB TFD. [ABSTRACT FROM AUTHOR]

This paper proposes a new effective approach to improve the performance of DOA (Direction Of Arrival) algorithms when bursts affect the array data. The proposed approach is based on the combining of data fusion techniques and the results of theoretical performance analysis of conventional DOA algorithms. For this purpose, the received array data is first split in M time-segments. Then, the DOAs are estimated from each data segment using a conventional DOA algorithm. The obtained estimates are fused using the federated fusion algorithm according to their statistical accuracy obtained from the well-documented performance analysis of the considered algorithm. As proof of concept of the proposed approach, numerical experiments have been conducted by considering the MUSIC algorithm. The obtained results show that the new algorithm outperforms the conventional one in terms of accuracy in a non-homogeneous environment. Therefore, it exhibits enhanced robustness capability. Moreover, it reduces the memory cost and computational complexity which makes it suitable for real time applications. To our knowledge, it is the first time that theoretical performance analysis results are exploited for the derivation of new subspace-DOA methods. [ABSTRACT FROM AUTHOR]