In this paper, we propose a sequential type-based detection scheme for wireless sensor networks in the case of spatially and temporally identically and independently distributed observations. First, we investigate the optimal sequential detection rule of the proposed scheme, and then with the motivation of reducing the computational complexity of the optimal detection rule, we consider an approximation scheme and derive a suboptimal detection rule. We also compare the performances of the type-based sequential detection scheme with those of the non-sequential type-based detection scheme in terms of both average number of observations and total energy consumption, and determine the region of individual node power where the proposed scheme outperforms the non-sequential scheme. In addition, we show that the approximated detection rule provides the similar results as the optimal detection rule with a significant reduction of the computational complexity, which makes the approximated detection rule useful for real-time applications.

This paper studies the necessity of local CFAR processing in CFAR detection with multisensors. This necessity is shown by comparison between centralized CFAR detection and the distributed CFAR detection scheme based on local CFAR processing, under three typical backgrounds and in several cases of mismatching ρ, the relative ratio of local clutter power level in sensors in a homogeneous background. Results show that centralized CFAR processing can not be considered as CFAR without exact prior knowledge of ρ. In addition, even if the knowledge of ρ is available, the great difference among local clutter power levels can also result in severe performance degradation of centralized CFAR processing. In contrast, the distributed CFAR detection based on local CFAR processing is not affected by ρ at all, a fact which was proposed in a previous published paper. Therefore, the CFAR processing must be made locally in sensors for CFAR detection with multisensors.

The optimal coding strategy for signal detection in the correlated gaussian noise is established for the distributed sensors system with essentially zero transmission rate constraint. Specifically, we are able to obtain the same performance as in the situation of no restriction on rate from each sensor terminal to the fusion center. This simple result contrasts with the previous ad hoc studies containing many unnatural assumptions such as the independence of noises contaminating received signal at each sensor. For the design of optimal coder, we can use the classical Levinson-Wiggins-Robinson fast algorithm for block Toeplitz matrix to evaluate the necessary weight vector for the maximum-likelihood detection.