Abstract
The polarimetric detection problem for 2-D vector sensor is reformulated based on quaternion technique. The quaternion-formed detectors are deduced and analyzed in different quaternion proper Gaussian noises, mainly for Q-proper and C-proper cases. The entropy method is applied to verify the equivalence of quaternion and complex second order statistics (SOS). The similar behaviors of quaternion and complex detectors are thus obtained when SOS is known a priori. Specially, in the case of unknown Q-proper SOS, quaternion and complex detection statistics follow the same F-distribution except for the distinctive second degrees of freedom. This means the improved behavior of quaternion adaptive detector can be achieved. Further, the estimate accuracy of SOS is evaluated fairly by Kullback–Leibler divergence. Consequently, the improved performance derives from the prior structure involved in the quaternion estimate. Numerical simulations also verify the analysis.
Similar content being viewed by others
Notes
Note that if the Cayley–Dickson form of quaternion noise is \(\tilde{\mathbf{e}} = {{\mathbf{e}}_h} + i{{\mathbf{e}}_v}\), where the imaginary operators of \({{\mathbf{e}}_h},{{\mathbf{e}}_v} \in {{\mathbb {C}}^{LN \times 1}}\) are \(j\), the quaternion random vector \(\tilde{\mathbf{e}}\) is not Q-proper. The relationship between the form \(\tilde{\mathbf{e}} = {{\mathbf{e}}_h} + i{{\mathbf{e}}_v}\) and the form in this paper is similar to the connection between left-handed coordinate system and right-handed coordinate system.
References
Anderson, T. W. (2003). An introduction to multivariate statistical analysis. Hoboken, NJ: Wiley.
Cover, T. M., & Thomas, J. A. (1991). Elements of information theory. New York: Wiley.
Cui, G., Kong, L., Yang, X., & Yang, J. (2010). Adaptive polarimetric MIMO radar detection. In Intelligent signal processing and communication systems (ISPACS) (pp. 1–4).
de Maio, A., Alfano, G., & Conte, E. (2004). Polarization diversity detection in compound-Gaussian clutter. IEEE Transactions on Aerospace and Electronic Systems, 10(1), 114–131.
Ginzberg, P., & Walden, A. T. (2011). Testing for quaternion propriety. IEEE Transactions on Signal Processing, 59(7), 3025–3034.
Giuli, D. (1986). Polarization diversity in radars. Proceedings of the IEEE, 74(2).
Gogineni, S., & Nehorai, A. (2010). Polarimetric MIMO radar with distributed antennas for target detection. IEEE Transactions on Signal Processing, 58(3), 1689–1697.
Hamilton, W. R. (1843). On quaternions. In Proceedings of the Royal Irish Academy.
Hurtado, M., & Nehorai, A. (2008). Polarimetric detection of targets in heavy inhomogeneous clutter. IEEE Transactions on Signal Processing, 56(4), 1349–1361.
Isaeva, O. M., & Sarytchev, V. A. (1995). ‘Quaternion presentations polarization state. In Proceedings of the 2nd IEEE topical symposium of combined optical microwave earth and atmosphere sensing (pp. 195–196), Atlanta, GA, USA.
Le Bihan, N., & Amblard, P. O. (2006). Detection and estimation of Gaussian proper quaternion valued random processes. In 7th IMA conference mathematics in signal processing.
Le Bihan, N., & Mars, J. (2004). Singular value decomposition of matrices of quaternions: A new tool for vector-sensor signal processing. Signal Processing, 84(7), 1177–1199.
Mandic, D. P., Jahanchahi, C., & Took, C. C. (2011). A quaternion gradient operator and its applications. IEEE Signal Processing Letters, 18(1).
Miron, S., LeBihan, N., & Mars, J. (2006). Quaternion-MUSIC for vector-sensor array processing. IEEE Transactions on Signal Processing, 54(4), 1218–1229.
Nehorai, A., & Paldi, E. (1994). Vector sensor processing for electromagnetic source localization. IEEE Transactions on Signal Processing, 42(2), 376–398.
Picinbono, B. (1996). Second-order complex random vectors and normal distributions. IEEE Transactions on Signal Processing, 44(10), 2637–2640.
Robey, F. C., Fuhrman, D. R., Kelly, E. J., & Nitzberg, R. (1992). A CFAR adaptive matched filter detector. IEEE Transactions on Aerospace Electronic Systems, 28(1), 208–216.
Seberry, J., Finlayson, K., Adams, S., Wysocki, T., Xia, T., & Wysocki, B. (2008). The theory of quaternion orthogonal designs. IEEE Transactions on Signal Processing, 56(1), 256–265.
Souyris, J.-C., & Tison, C. (2007). Polarimetric analysis of bistatic SAR images from polar decomposition: A quaternion approach. IEEE Transactions on Geoscience and Remote Sensing, 45(9), 2701–2714.
Sun, X., Wang, F., & Zhang, K. (2008). Time-delay estimation using quaternion. In The fourth international conference on natural computation (ICNC ’08.) (vol. 5, pp. 642–645).
Tao, J. (2013). Performance analysis for interference and noise canceller based on hypercomplex and spatiotemporal-polarisation processes. IET Radar, Sonar & Navigation, 7(3), 277–286.
Tao, J., & Chang, W. (2012). The MVDR beamformer based on hypercomplex processes. In IEEE, 2012 international conference on computer science and electronics engineering (pp. 273–277), Hangzhou.
Took, C., & Mandic, D. (2009). The quaternion LMS algorithm for adaptive filtering of hypercomplex processes. IEEE Transactions on Signal Processing, 57(4).
Took, C., & Mandic, D. (2011). Augmented second-order statistics of quaternion random signals. Signal Processing, 91, 214–224.
Via, J., Palomar, D. P., & Vielva, L. (2011). Generalized likelihood ratios for testing the properness of quaternion Gaussian vectors. IEEE Transactions on Signal Processing, 59, 1356–1370.
Via, J., Palomar, D. P., Vielva, L., et al. (2011). Quaternion ICA from second-order statistics. IEEE Transactions on Signal Processing, 59(4), 1586–1600.
Via, J., Ram, D., & Santamaria, I. (2010). Properness and widely linear processing of quaternion random vectors. IEEE Transactions on Information Theory, 56(7), 3502–3515.
Zhang, F. (1997). Quaternions and matrices of quaternions. Linear Algebra Its Applications, 251, 21–57.
Zhang, X. (2004). Matrix analysis and applications. Beijing: Tsinghua University Press.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (Grant Nos. 61101173 and 61371184).
Author information
Authors and Affiliations
Corresponding author
Appendices
Appendix 1: The proof of the distribution of QAED detection statistics
In this section, we will give the derivation of the distribution of QAED detection statistics and this derivation is an extension of real multivariate statistics (Anderson 2003). Firstly, we make the transformation
where \(\tilde{\mathbf{z}} = {\tilde{\mathbf{R}}^{ - \frac{1}{2}}}\tilde{\mathbf{y}},{\tilde{\mathbf{z}}_i} = {\tilde{\mathbf{R}}^{ - \frac{1}{2}}}{\tilde{\mathbf{y}}_i},i = 1,2, \ldots K\). Under the \({H_0}\) hypothesis \(\tilde{\mathbf{z}},{\tilde{\mathbf{z}}_i}\) follow \({\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_{LN}})\), under the \({H_1}\) hypothesis, \(\tilde{\mathbf{z}}\) follows \({\mathcal {QN}}({\tilde{\mathbf{R}}^{ - \frac{1}{2}}}\tilde{\mathbf{x}},{\mathbf{I}_{LN}})\), \({\tilde{\mathbf{z}}_i}\) follows \({\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_{LN}})\). \(\tilde{\mathbf{z}}\) and \({\tilde{\mathbf{z}}_i}\) are independent, and \({\tilde{\mathbf{z}}_i}\) is independent to \({\tilde{\mathbf{z}}_j}\) for \(i \ne j\). In order to prove (29), the following lemmas and definitions are put forward.
Definition 1
If a quaternion random matrix \(\tilde{\mathbf{X}} = {[{\tilde{\mathbf{x}}_1},{\tilde{\mathbf{x}}_2}, \ldots ,{\tilde{\mathbf{x}}_m}]_{p \times m}} \in {{\mathbb {Q}}^{p \times m}}\) and the column vectors are independent mutually, the matrix \(\tilde{\mathbf{X}}\) is defined as the sample matrix from quaternion normal distribution \({\mathcal {QN}}({\tilde{\mathbf{u}}_{p \times 1}},{\tilde{{\varvec{\Sigma } }}_{p \times p}})\).
Definition 2
If a quaternion random matrix \(\tilde{\mathbf{M}} = \tilde{\mathbf{X}}{\tilde{\mathbf{X}}^H} = \sum _{i = 1}^m {{{\tilde{\mathbf{x}}}_i}} \tilde{\mathbf{x}}_i^H \sim {\mathcal {Q}}{\mathcal {W}}_p(m,{{\varvec{\Sigma } }},{{\Lambda }})\), where \(\tilde{\mathbf{X}}\) is the sample matrix from \({\mathcal {QN}}({\tilde{\mathbf{u}}_{p \times 1}},{\tilde{{\varvec{\Sigma }}}_{p \times p}})\), the matrix \(\tilde{\mathbf{M}}\) is defined as quaternion Wishart matrix, \({{\tilde{\varvec{\Lambda }} }} = 4m{{\tilde{\mathbf {u}}}}{{{\tilde{\mathbf {u}}}}^H}\) is the non-centrality parameter (\({\tilde{\varvec{\Lambda }}}=0\) means \(\tilde{\mathbf{M}}\) is central Wishart matrix and it is ignored in the term \( {\mathcal {Q}}{\mathcal {W}}_p(m,{{\varvec{\Sigma } }})\) ). Specially, if \(p = 1\), \(\tilde{\mathbf{M}} = \sum _{i = 1}^m {{{\left| {{{\tilde{x}}_i}} \right| }^2}} \sim \frac{1}{4}{\sigma ^2}\chi _{4m}^2(\lambda )\), where \({\sigma ^2}\) is the sample variance, \(\lambda = 4m{\left| u \right| ^2}\) is the non-centrality parameter, and \(u\) is the sample mean.
Lemma 1
For a quaternion invertible matrix \(\tilde{\mathbf{A}} \in {{{\mathbb {Q}}}^{p \times p}}\) and a sample matrix \(\tilde{\mathbf{X}}\) from \({\mathcal {QN}}({\tilde{\mathbf{u}}_{p \times 1}},{\tilde{{\varvec{\Sigma } }}_{p \times p}})\), we have \({\tilde{\mathbf{A}}\tilde{\mathbf{X}}}{({\tilde{\mathbf{A}}{\tilde{\mathbf{X}}}})^H} \sim {\mathcal {Q}}{\mathcal {W}}_p(m,\tilde{\mathbf{A}}\tilde{\varvec{\Sigma }} {\tilde{\mathbf{A}}^H}, 4m{\tilde{\mathbf {A}}\tilde{u}}{\tilde{\mathbf {u}}^{H}} {\tilde{\mathbf {A}}^{H}})\).
Proof
It can be easily deduced by \(\tilde{\mathbf{A}}{\tilde{\mathbf{x}}_i} \sim {\mathcal {QN}}({\tilde{\mathbf{A}}\tilde{u}},{\tilde{\mathbf{A}}\tilde{\varvec{\Sigma }}}{\tilde{\mathbf{A}}^{H}})\). \(\square \)
Lemma 2
For a quaternion matrix \(\tilde{\mathbf{U}} \in {{{\mathbb {Q}}}^{m \times a}},{\tilde{\mathbf{U}}^H}\tilde{\mathbf{U}} = {\mathbf{I}_a}\), and a sample matrix \(\tilde{\mathbf{X}}\in {{\mathbb {Q}}^{p \times m}}\) from \({\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_p})\) (Note that the following proof require covariance matrix is real), we have \({\tilde{\mathbf{X}}\tilde{U}}{\tilde{\mathbf{U}}^{H}}{\tilde{\mathbf{X}}^{H}} \sim {\mathcal {Q}}{\mathcal {W}}_p(a,{\mathbf{I}_{p}})\).
Proof
It is equivalent to prove \({\tilde{\mathbf{X}}\tilde{U}}\) is sample matrix from \(QN({\mathbf{0}},{\mathbf{I}_{p}})\). Define \(\tilde{\mathbf{Z}} = {\tilde{\mathbf{X}}^{H}}\), and the entries in \(\tilde{\mathbf{Z}}\) are independent because covariance matrix is \({\mathbf{I}_{p}}\). Then we prove the row vectors of \(\tilde{\mathbf{Z}}\) are i.i.d random vectors.
Consider \(\tilde{\mathbf{Z}}\), we have \({\text {vec}}(\tilde{\mathbf{Z}}) \sim {\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_p} \otimes {\mathbf{I}_m})\), where \({\text {vec}}( \cdot )\) is the vector operator. We have \(vec({\tilde{\mathbf{U}}^H}\tilde{\mathbf{Z}}) = ({\mathbf{I}_p} \otimes {\tilde{\mathbf{U}}^H})vec(\mathbf{Z})\), therefore
Note that the kronecker operator for quaternion requires the second term \({\mathbf{I}_p} \otimes {\mathbf{I}_m}\) in (59) is real. Then \({\text {vec}}({\tilde{\mathbf{U}}^H}\tilde{\mathbf{Z}}) \sim QN({\mathbf{0}},{\mathbf{I}_p} \otimes {\mathbf{I}_a})\), so the entries in \({\tilde{\mathbf{U}}^H}\tilde{\mathbf{Z}}\) are uncorrelated, then \({\tilde{\mathbf{X}}\tilde{\mathbf{U}}}\) is sample matrix from \({\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_{p}})\). \(\square \)
Lemma 3
For random matrix \(\tilde{\mathbf{M}} = \sum _{k = 1}^K {{{\tilde{\mathbf{z}}}_k}\tilde{\mathbf{z}}_k^H} = \tilde{\mathbf{Z}}{\tilde{\mathbf{Z}}^H} \sim {\mathcal {Q}}{\mathcal {W}}_p(m,{\mathbf{I}_p})\), we partition the matrices \(\tilde{\mathbf{M}}\), \({\mathbf{I}_p}\) and \(\tilde{\mathbf{Z}}\) into \(\tilde{\mathbf{M}} = \left( {\begin{matrix}{{{\tilde{\mathbf{M}}}_{11}}} &{} {{{\tilde{\mathbf{M}}}_{12}}} \\ {{{\tilde{\mathbf{M}}}_{21}}} &{} {{{\tilde{\mathbf{M}}}_{22}}} \\ \end{matrix} } \right) \), \({\mathbf{I}_p} = \left( {\begin{matrix}{{\mathbf{I}_a}} &{} {\mathbf{0}} \\ {\mathbf{0}} &{} {{\mathbf{I}_b}} \\ \end{matrix} } \right) \), \(\tilde{\mathbf{Z}} = \left( {\begin{matrix}{{{\tilde{\mathbf{Z}}}_1}} \\ {{{\tilde{\mathbf{Z}}}_2}} \\ \end{matrix} } \right) \), where \({\tilde{\mathbf{Z}}_1} \in {{\mathbb {Q}}^{a \times m}},{\tilde{\mathbf{Z}}_2} \in {{\mathbb {Q}}^{b \times m}}, a + b = p\), and \({\tilde{\mathbf{M}}_{ij}} = {\tilde{\mathbf{Z}}_i}\tilde{\mathbf{Z}}_j^H\). Define the invertible matrix \({\tilde{\mathbf{M}}^{ - 1}} = \left( {\begin{matrix}{{{\tilde{\mathbf{M}}}^{11}}} &{} {{{\tilde{\mathbf{M}}}^{12}}} \\ {{{\tilde{\mathbf{M}}}^{21}}} &{} {{{\tilde{\mathbf{M}}}^{22}}} \\ \end{matrix}} \right) \),\(\mathbf{I}_p^{ - 1} = \left( {\begin{matrix}{{\mathbf{I}_a}} &{} {\mathbf{0}} \\ {\mathbf{0}} &{} {{\mathbf{I}_b}} \\ \end{matrix} } \right) \). Through principle of the inversion of partitioned matrix, we have \({({\tilde{\mathbf{M}}^{22}})^{ - 1}} = {\tilde{\mathbf{M}}_{2.1}} = {\tilde{\mathbf{M}}_{22}} - {\tilde{\mathbf{M}}_{21}}\tilde{\mathbf{M}}_{11}^{ - 1}{\tilde{\mathbf{M}}_{12}}\),
and \({\tilde{\mathbf{M}}_{2.1}}\) is independent to \(\tilde{\mathbf{Z}}_1\).
Proof
Define the matrix \(\tilde{\mathbf{P}} = (\mathbf{I} - \tilde{\mathbf{Z}}_1^H{({\tilde{\mathbf{Z}}_1}\tilde{\mathbf{Z}}_1^H)^{ - 1}}{\tilde{\mathbf{Z}}_1})\) which is an idempotent matrix, so the characteristic values are \(1\) or \(0\). The quaternion matrix \(\tilde{\mathbf{P}}\) can be decomposed by
where \(\tilde{\mathbf{U}}_1^H{\tilde{\mathbf{U}}_1} = {\mathbf{I}_{m - a}}\).
According to lemma 2, we have
Especially, if \(a = p - 1,b = 1\), we have
and due to the independence of \(\tilde{\mathbf{Z}}_1\) and \(\tilde{\mathbf{Z}}_2\), \({\tilde{\mathbf{M}}_{2.1}}\) is independent to \(\tilde{\mathbf{Z}}_1\). \(\square \)
Lemma 4
For a quaternion vector \(\tilde{{\alpha }}\), quaternion random matrix \(\tilde{\mathbf{M}} = \sum _{k = 1}^K {{{\tilde{\mathbf{z}}}_k}\tilde{\mathbf{z}}_k^H} = \tilde{\mathbf{Z}}{\tilde{\mathbf{Z}}^H} \sim {\mathcal {Q}}{\mathcal {W}}_p(m,{\mathbf{I}_p})\) from \({\mathcal {QN}}({\mathbf{0}},{\mathbf{I}_p})\), we have
where \({\left\| {{{\tilde{\alpha }}}} \right\| ^2}={{\tilde{\alpha }}}^H{{\tilde{\alpha }}}\).
Proof
For a quaternion unitary matrix \(\tilde{\mathbf{B}}\) where \(\tilde{\mathbf{B}}^H\tilde{\mathbf{B}}=\mathbf{I}_p\), \(\frac{{\tilde{\alpha }}}{{\left\| {\tilde{{\alpha }}} \right\| }}\) is the last column of \(\tilde{\mathbf{B}}\). Define the matrix \(\tilde{\mathbf{N}} = {\tilde{\mathbf{B}}^{ - 1}}\tilde{\mathbf{M}}{\tilde{\mathbf{B}}^{ - H}}\), and \({\tilde{\mathbf{N}}^{ - 1}} = {\tilde{\mathbf{B}}^H}{\tilde{\mathbf{M}}^{ - 1}}\tilde{\mathbf{B}}\), where \(\tilde{\mathbf{M}} = \sum _{k = 1}^K {{{\tilde{\mathbf{z}}}_k}\tilde{\mathbf{z}}_k^H} = \tilde{\mathbf{Z}}{\tilde{\mathbf{Z}}^H} \sim {\mathcal {Q}}{\mathcal {W}}_p(m,{\mathbf{I}_p})\).
According to Lemma 1, we have \(\tilde{\mathbf{N}} \sim {\mathcal {Q}}{\mathcal {W}}_p(m,{\mathbf{I}_p})\)
According to Lemma 3, we have
For detection statistic \({\tilde{\mathbf{z}}^H}{(\frac{1}{K}\sum _{i = 1}^K {{{\tilde{\mathbf{z}}}_i}\tilde{\mathbf{z}}_i^H} )^{ - 1}}\tilde{\mathbf{z}}\), we have
According to lemma 3 and lemma 4, we have
and
where \(\mu _0 = 0\) and \(\mu _1 = 4{\tilde{\mathbf{x}}^H}{\tilde{\mathbf{R}}^{ - 1}}\tilde{\mathbf{x}}\).
Hence, Eq. (29) is proved. \(\square \)
Appendix 2: Equivalence of non-centrality parameters
In this subsection, we give the proof that \(\mu _1''=\mu _1'\), where \(\mu _1'' = 2{\tilde{\mathbf{x}}^{cH}}{\tilde{\mathbf{R}}^{ - c}}{\tilde{\mathbf{x}}^c}\) and \(\mu _1' = 2{\mathbf{x}^H}{{\mathbf{R}}^{ - 1}}\mathbf{x}\).
Firstly, according to the matrix theory that \({(\mathbf{A} \otimes {\mathbf{B}})^{ - 1}} = {\mathbf{A}^{ - 1}} \otimes {{\mathbf{B}}^{ - 1}}\), where \(\mathbf{A},{\mathbf{B}} \in {{\mathbb {C}}^{n \times n}}\) are invertible matrices (Zhang 2004). we have
Then according to (69), we have
where \(a = \frac{{{\sigma _1} + {\sigma _2}}}{{4{\sigma _1}{\sigma _2}}}\) and \(b = \frac{{{\sigma _2} - {\sigma _1}}}{{4{\sigma _1}{\sigma _2}}}\).
The first term \({\tilde{\mathbf{x}}^H}{\mathbf{R}}_{st}^{ - 1}\mathbf{x}\) in (70) is given by
where we employ that \(jc=c^*j\) for complex number \(c\) whose imaginary operator is \(i\).
The second term \( {\tilde{\mathbf{x}}^{iH}}\tilde{\mathbf{R}}_{st}^{ - 1}\tilde{\mathbf{x}}\) in (70) is given by
According to (71) and (72), The non-centrality parameter \(\mu \) is given by
The non-centrality parameter \(u'\) is given by
Rights and permissions
About this article
Cite this article
Wang, Y., Xia, W., He, Z. et al. Polarimetric detection for vector-sensor processing in quaternion proper Gaussian noises. Multidim Syst Sign Process 27, 597–618 (2016). https://doi.org/10.1007/s11045-015-0325-8
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s11045-015-0325-8