Distributed Detection with Local Least Squares Quantization

In this paper, we consider a parallel topology of the decentralized detection problem and use the least squares quantization (LSQ) method to decrease the computational complexity of the local quantization rule design. The LSQ method operates with a probability density function (pdf), but radar target detection often needs to consider two pdfs, corresponding to the null hypothesis (H
₀
) and the alternative hypothesis (H
₁
). As the null case can be considered as the case of zero signal-to-noise ratio (SNR), we focus on how to set the SNR in quantization for a better detection performance. The optimal fusion rule is designed according to the Neyman-Pearson (NP) lemma. Numerical results show that the LSQ quantization under H
₁
can achieve a better detection performance than that under H
₀
. Meanwhile, a comparison to the NP based quantization indicates that the LSQ method suffers a little detection performance loss but the computational cost is decreased dramatically.