TY - GEN
T1 - Bayesian subspace recovery with approximate prior knowledge for radar detection
AU - Jiang, Yuan
AU - Li, Hongbin
AU - Rangaswamy, Muralidhar
N1 - Publisher Copyright:
© 2019 IEEE.
PY - 2019/4
Y1 - 2019/4
N2 - We examine the problem of subspace recovery when approximate prior knowledge of the subspace is available. The problem originates from a recent work [1], which presents a subspace knowledge learning (SKL) algorithm for subspace recovery by exploiting partial and partly erroneous prior knowledge of the subspace. However, SKL is limited by the assumption that the subspace bases coincide with some basis vectors of a known overcomplete dictionary matrix. In the presence of grid mismatch, SKL may degrade considerably. Another issue is that SKL is computationally quite intensive, since it is an iterative algorithm involving matrix operations in each iteration. To address these issues, we present herein a modified SKL (mSKL) algorithm for subspace recovery which can exploit approximate prior knowledge of the subspace and cope with the grid mismatch problem. The mSKL algorithm is further integrated with generalized approximate message passing (GAMP) which replaces matrix operations in SKL with scalar approximations and is hence computationally efficient. The resulting subspace recovery algorithm, referred to as the mSKL-GAMP, is used to solve a radar detection problem that involves detecting a multichannel target signal in subspace interference. Numerical results are presented to demonstrate the effectiveness of the proposed method.
AB - We examine the problem of subspace recovery when approximate prior knowledge of the subspace is available. The problem originates from a recent work [1], which presents a subspace knowledge learning (SKL) algorithm for subspace recovery by exploiting partial and partly erroneous prior knowledge of the subspace. However, SKL is limited by the assumption that the subspace bases coincide with some basis vectors of a known overcomplete dictionary matrix. In the presence of grid mismatch, SKL may degrade considerably. Another issue is that SKL is computationally quite intensive, since it is an iterative algorithm involving matrix operations in each iteration. To address these issues, we present herein a modified SKL (mSKL) algorithm for subspace recovery which can exploit approximate prior knowledge of the subspace and cope with the grid mismatch problem. The mSKL algorithm is further integrated with generalized approximate message passing (GAMP) which replaces matrix operations in SKL with scalar approximations and is hence computationally efficient. The resulting subspace recovery algorithm, referred to as the mSKL-GAMP, is used to solve a radar detection problem that involves detecting a multichannel target signal in subspace interference. Numerical results are presented to demonstrate the effectiveness of the proposed method.
KW - Bayesian inference
KW - Knowledge-aided processing
KW - Radar signal detection
KW - Subspace estimation
UR - http://www.scopus.com/inward/record.url?scp=85073100281&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85073100281&partnerID=8YFLogxK
U2 - 10.1109/RADAR.2019.8835711
DO - 10.1109/RADAR.2019.8835711
M3 - Conference contribution
AN - SCOPUS:85073100281
T3 - 2019 IEEE Radar Conference, RadarConf 2019
BT - 2019 IEEE Radar Conference, RadarConf 2019
T2 - 2019 IEEE Radar Conference, RadarConf 2019
Y2 - 22 April 2019 through 26 April 2019
ER -