Adaptive subspace nulling based on Karhunen-Loeve expansion

Conventional approaches require at least 2N snapshot vectors for an N-element adaptive array for reliable estimation of the noise covariance matrix used in weight vector selection. In all-digital subspace beamforming, a matrix transformation is performed on each of these data snapshot vectors to reduce their dimensions from N to P, where P<N, thereby reducing the computation burden. The authors propose a new hybrid analog/digital subspace beamforming approach in which A/D conversion at the front end of the sensor array is bypassed. Instead they stay in the analog domain and adapt using the coefficients obtained in the truncated Karhunen-Loeve expansion of the received signal. It is demonstrated via computer simulations that for the same subspace dimension, the hybrid approach outperforms the all-digital fully adaptive and the all-digital subspace approach for large time-bandwidth products