Can the Threshold Performance of Maximum Likelihood DOA Estimation be Improved by Tools from Random Matrix Theory?

Size: px
Start display at page:

Download "Can the Threshold Performance of Maximum Likelihood DOA Estimation be Improved by Tools from Random Matrix Theory?"

Transcription

1 Advances in Signal Processing 2(): 8-28, 204 DOI: 0.389/asp Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? Yuri I. Abraovich, Ben A. Johnson 2, Institute for Telecounications Research University of South Australia, awson Lakes, SA 5045 Australia 2 Colorado School of ines, Golden, CO 8040 Corresponding Author: ben.a.johnson@ieee.org Copyright c 204 Horizon Research Publishing All rights reserved. Abstract For direction of arrival (DOA) estiation in the threshold region, it has been shown that use of Rando atrix Theory (RT) eigensubspace estiates provides significant iproveent in USIC perforance. Here we investigate whether these RT ethods can also iprove the threshold perforance of unconditional (stochastic) axiu likelihood DOA estiation (LE). Keywords Signal Detection and Estiation, axiu-likelihood Estiation, Rando atrix Theory Introduction The pre-asyptotic perforance of traditional stochastic (unconditional) LE of DOAs for closely-spaced Gaussian signals iersed in white Gaussian noise using an -sensor array and a finite nuber T independent identically distributed (i.i.d.) training saples continues to be the subject of investigations [3 5, 5] despite being a relatively old proble [6]. The ain reason, as forulated in [4], is that in the pre-asyptotic doain, no general non-asyptotic results are [currently] available for the perforance evaluation of the L ethod, and each proble requires a special investigation. As an alternative, instead of focusing on the accurate non-asyptotic analysis, one ay consider the asyptotic odel where both quantities and T grow without bound, while their quotient converges to a fixed finite quantity:, T, T c, 0 < c < () This condition is known as the Kologorov asyptotic condition [6], and underpins a field of analysis referred to as Generalized Statistical Analysis (GSA) or Rando atrix Theory (RT). Of course, in any practical situation, one deals with finite and T values, and therefore generalized G-asyptotic (i.e. asyptotic per the conditions in ()) results ay still not be sufficiently accurate. However, in nuerous studies, it has been deonstrated that estiators that are consistent G-asyptotically are ore robust in the presence of finite saples T than other estiators which are only consistent for T, =constant [6, 0,, 8]. Specifically, let us suppose that i.i.d. observations x,..., x T of rando vector ξ with diension are given, and we wish to estiate soe value φ(r ), where φ is a continuous function of the entries of the (true) covariance atrix R of vector ξ. If T is large and is fixed and does not change as T grows (i.e. the standard asyptotic case), then as an estiator of φ(r ), we ay take φ( ˆR ), where pli φ( ˆR ) = φ(r ) (2) T and ˆR is the standard saple covariance atrix for zero-ean data (= T T xxh ). oreover, if x j, j =,..., T is a set of i.i.d. coplex (circular) Gaussian saples (i.e. x j CN (0, R )), then ˆR and φ( ˆR ) are not only consistent (in the standard asyptotic fraework), but also the axiu likelihood estiate of R and φ(r ) respectively [7]. However, this failiar assertion is not in general true under the G-asyptotic condition

2 Advances in Signal Processing 2(): 8-28, (). There, for a wide range of functions φ(r ), one can find a easurable function ψ of the entries of the atrix R for which pli,t [φ( ˆR ) ψ(r )] = 0. (3) But in general, the functions φ(r ) and ψ(r ) do not coincide, in which case φ( ˆR ) is not necessarily a G- consistent estiate of φ(r ). However, with the help of function ψ(r ), one ay try to find a easurable function g( ˆR ) such that pli,t [g( ˆR ) φ(r )] = 0. (4) and where the distribution of noralized difference g( ˆR ) φ(r ) is asyptotically noral. The function g( ˆR ) is then called a G-estiator [6], and is consistent under condition (). In [0,], this RT ethodology has been used to find G-estiates of eigenvalues for a covariance atrix with known ultiplicity of its eigenvalues. Specifically, for an -variate covariance atrix R, let γ < γ 2 <... < γ be the set of distinct eigenvalues ( ) after accounting for individual ultiplicity K of the true eigenvalues (i.e. = K = ). Associated with each eigenvalue γ, there is a coplex subspace of diension K. This subspace is deterined by an K atrix of corresponding eigenvectors, denoted by E, such that EE H = I K. Note that this specification is unique up to right ultiplication by the orthogonal atrix, and therefore the proble of eigendecoposition for the atrix R R = γ j E j Ej H (5) is ore convenient to forulate as a proble of estiation of orthogonal projection atrices, defined as given the saple covariance atrix ˆR that has the eigendecoposition P = E E H, =,...,, (6) ˆR = ˆλ j Û j Ûj H (7) To use this eigendecoposition to estiate the orthogonal projection atrices P, let K be a set of indices K = K j, K j +,..., K j (8) with the cardinality of K equal to the ultiplicity of the eigenvalue γ, naely K. The classical (and indeed axiu likelihood for ultivariate Gaussian observations) estiator of the -th eigenvalue and orthogonal projection atrices P in (6) is given by One can see that ˆγ L and estiates of γ and P. ˆP L ˆγ L = K L ˆλk ; ˆP k K = k K Û k Û H k (9) are specified as φ( ˆR ) in (2) and indeed are strongly T -consistent (and L) Yet in Theore 2 of [0], X. estre proved that under certain conditions (As-As4), the traditional estiators of eigenvalues and eigensubspaces (ˆγ L and ˆη j L = S H k K Û k Ûk HS 2, respectively, where S and S 2 are two deterinistic vectors) can be shown to converge under condition () (i.e., T with /T = c constant and finite) to: γ L 0 and ˆη j L η L 0 (0) where the (non-rando) values γ L ˆγ L and η L γ L are defined as = γ c r K r γ r () γ r γ and w (k) = η L = w (k)s H E k Ek H S 2 (2) { K r K r γ γ r γ µ γ r µ ( ) γ γ r γ µ γ r µ k = k (3)

3 20 Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? where µ is the -th solution to the following equation in µ: K r γ r γ r µ = c (4) under the convention µ < µ 2 <... < µ. Thus, the traditional L estiators for ˆγ L and ˆη L are not G-consistent estiates, and therefore could potentially be iproved within the RT ethodology. Iproved G-consistent estiates for γ and P under the conditions (As-As4) have been derived by X. estre, Theore 3 in [0]: ˆγ G = T (ˆλ k ˆµ k ), K k K ˆη G = ρ (k)s H ÛkÛ k H S 2 ; (5) where { φ (k), k / K ρ j (k) = + ψ (k), k K φ (k) = ( ) ˆλr ˆµ r r K ˆλ k ˆλ r ˆλ k ˆµ r ψ (k) = ( ) ˆλr ˆµ r ˆλ k ˆλ r ˆλ k ˆµ r r / K (6) and ˆµ ˆµ 2 ˆµ are the real-valued solutions to the following equation in ˆµ: ˆλ k ˆλ k ˆµ = c ; c = T (7) One can prove (see [0]) that for any fixed, as T, we have ˆµ k ˆλ k, T (ˆλ k ˆµ k ) ˆλ k (8) and therefore ˆγ G K k K ˆλk = ˆγ L (9) On the other hand, φ (k) 0 and ϕ (k) 0, which iplies that ˆη G ˆη L, as T. Therefore, the RT ethodology provides a unique set of G-consistent estiates for ˆγ G and ˆη G that tend to the classical LE estiates under traditional asyptotic assuptions ( = constant, T ). Specifically, the G-consistent USIC (or G-USIC) pseudo-spectru estiate ˆη G (θ) = ρ (k)s H (θ)ûkû k H S(θ) (20) has been suggested in [9] and deonstrated significant perforance iproveent copared with the traditional USIC function ˆη L (θ). Recently in [8], we showed that for closely spaced sources, despite the substantial iproveents deonstrated by G-USIC (20) with respect to conventional USIC, the threshold perforance of G-USIC (where the estiator s ean squared error departs rapidly fro the Craér-Rao lower bound as SNR and/or saple support is reduced) reains significantly worse than perforance of the rigorously ipleented (via global search) L DOA estiation. Since the RT-ethodology, and specifically the G-consistent estiates in (5), are able to iprove conventional USIC perforance and differ fro the traditional L eigenvalue and subspace estiates, it is quite legitiate to investigate whether these estiates ay be used to iprove the threshold perforance of L DOA estiation itself. In this correspondence, we try to address this question. Note that for S = S 2 = S(θ) where S(θ) is an -eleent antenna steering (anifold) vector specified by the DOA θ, and for =, ˆη (θ) is the traditional USIC pseudo-spectru function.

4 Advances in Signal Processing 2(): 8-28, G-asyptotic L DOA Estiation (G-LE) Forulation For the stochastic (unconditional) L odel under consideration, the set of T i.i.d. data x j CN (0, R 0 ), j =,..., T is described by the true/exact covariance atrix R 0, coprised of independent planewave sources with direction of arrival θ, ebedded in white noise. Under this odel, the traditional L DOA estiates are found as the global axiu of the stochastic (unconditional) likelihood function (LF) coputed for the odel covariance atrix R(Ω ), uniquely specified by 2 + paraeters in Ω ( source DOAs and powers as well as the noise power) and denoted L[R(Ω )]: ˆΩ = arg ax L[R(Ω )], Ω [ exp[ tr R (Ω ) L[R(Ω )] = ˆR] ] T (2) π det R(Ω ) or, equivalently, as the iniu of the (scaled) 2 log-likelihood function (LLF) ll[r(ω )] = tr R (Ω ) ˆR + log det R(Ω ) Note that for any given R(Ω ), the LLF (22) ay be viewed as the L estiate of the function (22) φ[r(ω ), R 0 ] φ[r 2 (Ω )R 0 R 2 (Ω )] = = tr [R 2 (Ω )R 0 R 2 (Ω )] + log det R(Ω ) (23) with an unknown R 0, here represented by its generic (positive definite Heritian, for T ) L estiate saple covariance atrix ˆR fored fro the training data x j. Since the likelihood function (22) is just φ( ˆR (Ω )) (per the notation in (2)), under the RT ethodology, we need to construct the G-consistent estiate g( ˆR (Ω )) (per the notation in (4)) of the function φ 0 (the portion of φ fro (23) dependent on R 0, which can be siplified to tr R (Ω )R 0 ). Yet, according to Lea 3. in [6] (see eqn. 3.0, p. 82), for a very broad class of epirical covariance atrices that ebraces the coplex Wishart case ˆR (Ω ) = T R 2 (Ω )R 2 0 ĈR 2 0 R 2 (Ω ); Ĉ CW(, T, I ), (24) under the Kologorov asyptotic condition (), the following property is proven: [ pli T, T c tr ˆR ] E( tr ˆR ) = 0, (25) where E( ) is the expectation operator and CW(, T, I ) is the coplex Wishart distribution [7]. This property eans that for the function φ 0, we have a special case where its G-consistent estiate g( ˆR (Ω )) coincides with its L estiate φ 0 ( ˆR): g( ˆR (Ω )) = φ( ˆR (Ω )) = tr [R (Ω ) ˆR] (26) In essence, the property (25) eans that for the ultivariate coplex Gaussian data case, the (scaled) unconditional log-likelihood function (22) ay be treated as the L estiate and at the sae tie, the G-consistent estiate of the function φ[r(ω ), R 0 ] given in (23). Lea 3. fro [6], and ore iportantly, its interpretation as G-consistency of the standard log-likelihood function 3, leaves an apparent contradiction which needs to be resolved. Specifically, since according to Theore 2 fro [0], individual L estiates of eigenvalues are not G-consistent and can be iproved, per (5). At the sae tie, the su of all eigenvalues (i.e. the saple covariance atrix trace) is G-consistent, and therefore cannot be iproved within the RT ethodology 4. This contradiction could be resolved by the direct proof that the su 2 This scaling by the factor / is iaterial for a fixed, but is subsequently required for G-asyptotic analysis with (, T ), /T c, for ll[r(ω )] to reain finite as 3 It is iportant to ephasize that G-consistency of the likelihood function (or the G-USIC function, for that atter) does not iply that the resultant DOA estiates are always G-consistent as well. On the contrary, the fact that the globally optial L DOA estiates (as well as the G-USIC DOA estiates) exhibit perforance breakdown in their respective threshold regions deonstrates that G-consistency of the likelihood or G-USIC function is only a necessary, but not at all sufficient condition for G-consistent DOA estiation. 4 While Theore 3 in [0], provides a unique set of G-consistent estiates (5) within the RT ethodology, it does not resolve questions regarding the broader uniqueness of G-consistent estiates. In this paper, per our title, we confine ourselves to the application of these (strongly) consistent RT estiates for individual eigenvalues and eigenvectors. While the possibility of another ethodology which provides better G-consistent estiates cannot be precluded, to our knowledge, such a ethodology has not yet been developed.

5 22 Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? of G-consistent individual eigenvalue estiates ˆγ G in (5) is strictly equal to the su of saple eigenvalues ˆλ k of ˆR (Ω ), ie. K ˆγ G = ˆλ k = tr [ ˆR (Ω )] (27) = According to (5), we therefore have to prove that Theore Under conditions As-As3 in [0], the estiate K ˆγ G = T ( ˆλ k ˆµ k ) = ˆλ k (28) = is the G-consistent estiate of the trace of R, regardless of the eigenvalue splitting condition As4. The proof of this equivalence is shown in Appendix A, and shows that the trace of a saple atrix ˆR is always the G-consistent estiate of the trace of the original (true) covariance atrix R, even when for soe individual eigenvalues of this atrix, their G-consistent estiates ay not exist due to As4 not being satisfied. As an aside for the interested reader, estre s assuptions As to As3 (the details of which are provided in [0]) ipose general conditions on the covariance atrix and eigenvalues which are always eet under our proble with the saple Wishart atrix ˆR (Ω ), but assuption As4 ay or ay not be et. The physical eaning of the As4 condition, also referred to as the eigenvalue splitting condition, is that the asyptotic eigenvalue distribution of ˆR can be considered as a collection of clusters, each one centered around a different eigenvalue of R = E ˆR = R 2 (Ω )R 0 R 2 (Ω ). The size of the support of these clusters depends on the ratio of the nuber of sensors/antennas to the nuber of saples (ie. the ratio c = /T ) fro which ˆR is constructed. If the ratio /T is high, the clusters with close eigenvalues γ j of R ay erge and be represented by a single estiate with a certain ultiplicity K. If, on the contrary, the ratio /T is low, each saple eigenvalue ˆλ k will be associated with a different distinct cluster, well separated fro the rest of the saple eigenvalue distribution. This analysis provides an answer, therefore, to the ain question posed in the title of this correspondence: Within the existing RT fraework, iproved G-consistent estiates of individual eigenvalues ˆγ G and bilinear fors ˆη G given in (5) cannot ake the traditional log-likelihood function ore G-consistent, and therefore provide any iproveent into the threshold perforance of L DOA estiation, in full correspondence with Lea 3. fro [6] shown in (25) above. To confir this further and support nuerical evaluation, we still ay consider an alternative reconstruction of the G-consistent log-likelihood function using the individual estiates (5) in a anner siilar to the G-USIC derivation in [9]. Assuing that the eigenvalue splitting condition As4 fro [0] is satisfied for all the ( + ) different eigenvalues 5, we ay construct the G-LLF (log likelihood function) as g 0 [R(Ω )] = tr R (Ω ) ˆR G, ˆRG = + ˆλ G j ˆP G j. (29) where, based on (5), G-consistent estiates of the eigenspace P j in R 0 (R 0 = λ P + + j=2 λ jp j ) are given by ˆP G j = ρ j (k)ûkû k H ; ˆR = ˆλ j Û j Ûj H (30) Recall that G-USIC was calculated as per (20), and therefore (29)-(30) reseble this construction, but with soe differences. First of all, in contrast to standard eigendecoposition, we note that individual eigensubspace estiates ˆP j G in (30) are no longer utually orthogonal, though they can be shown to su to the identity atrix since the weights ρ j (k) in (6) su to unity. Therefore, if we introduce the diagonal atrix D(j) diag[ρ j (),..., ρ j ()], we iediately discover that according to (30), we get + ˆR G = Û ˆλ G j D j Û H. (3) While individual (G-consistent) estiates ˆP j G in (30) are different fro the L estiates ˆP j = k K j Û k Ûk H, the eigenvectors in ˆR G L are exactly the sae as in the L estiation ˆR (Theore 9.3. in [3]). 5 Recall that R 0 (as opposed to the whitened covariance atrix R ) consists of an diensional signal subspace and a noise subspace of diension

6 Advances in Signal Processing 2(): 8-28, Let us deonstrate that reconstructed eigenvalues λ R k in + ˆλ G j D j = diag[ˆλ R,..., ˆλ R k ] (at least G-asyptotically) tend to the sae liit as the (not G-consistent) LE estiate ˆγ L li,t /T c ˆλ R k li,t /T c ˆγ L j k : = γ k + c + K r γ r (32) γ r γ k Using (5)-(6), for the noise subspace eigenvalues (ˆλ R,..., ˆλ R ), we get ˆλ R k = ˆλ G k + ˆλ r ˆλ k ˆλ r ˆµ r ˆλ k ˆµ r ˆλ G r [ ] ˆλ r ˆµ r ˆλ k ˆλ, (33) r ˆλ k ˆµ r for k =,..., ( ) and with ˆλ G = ˆλ G 2 =... = ˆλ G, which is identical to ˆλ R k = ˆλ G k + ˆλ k (ˆλ G k ˆλ G r )(ˆλ r ˆµ r ) (ˆλ k ˆλ, k =,..., ( ). (34) r )(ˆλ k ˆµ r ) According to (5), for signal subspace eigenvalues (r ), we get ˆλ r ˆµ r = T ˆλ G r, and therefore we have ˆλ R k = ˆλ G k ˆλ k T (ˆλ G k ˆλ G r )ˆλ G r (ˆλ k ˆλ r )(ˆλ k ˆµ r ). (35) Note that for ˆλ r ˆµ r < ˆλ r and ˆµ r T ˆλ T r when As4 is strongly et. Therefore the difference between (35) and the expression (see () and (32)) ˆλ R k = ˆλ G k ˆλ G k T + (ˆλ G k ˆλ G r )ˆλ G r (ˆλ G k ˆλ G r )(ˆλ G k ˆλ G r ) (36) that strictly tends to γ L k in () under asyptotic condition (), is only a difference of the second order (in T ) agnitude. Therefore, the reconstructed noise subspace eigenvalues in ˆR G (3) G-asyptoatically are not strictly identical, but the fluctuations around its asyptotic value γ L is quite negligible. For signal subspace eigenvalues, these fluctuations are even less significant, which ay be deonstrated by derivations siilar to those above. While it would be ore satisfying if ˆλ R k was found to be strictly equal to the L version ˆγL k, this is not the case, as observed by other researchers in the field [2]. But practically speaking, it is not significant that non G-consistent estiates ˆγ k L are replaced in (29) by (slightly) different non G-consistent estiates ˆλ R k. The ain point is that the reconstruction in (29) using the G-consistent estiates of eigenvalues and eigenvectors (5) should be at best as efficient as the conventional accurate LE DOA estiate. Therefore, the conducted analysis has deonstrated how and why the G-consistent individual estiates of eigenvalues and eigenvectors help to iprove USIC threshold perforance, but still lead to the traditional L perforance. This property is proven rigorously for direct G-consistent estiation of tr (R (Ω )R 0 ), and in first order (in T ) approxiation for the reconstruction approach in (29). Let us now support these theoretical findings by epirical results. 3 Nuerical Evaluation Here, we seek by siulation to deonstrate, using direct exhaustive search for the global axiu of the likelihood function and its reconstructed G-odification, that their DOA estiation perforance in the threshold region is identical. For this analysis, we have selected the sae scenario as in [] with = 8 sensor ULA, T = 0 i.i.d. training saples and two equipower sources (p = p 2 ) with DOAs {0 o, 0.4 o } in white noise power p 0 =. The nuber of sources = 2 is liited by the coputational load associated with the exhaustive -diensional search for the LF global axiu, but still deonstrates the relevant LE threshold behavior. For this scenario, the CRB resolution liit (as predicted by the CRB RSE of 0.2 o, derived in ore detail in []) occurs at an SNR of approxiately 40dB (ε CRB = 40dB).

7 24 Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? Prob ITC under est or USIC failure USIC AIC DL AP 8 ULA, θ = [0, 0.4] deg, T = 0, 000 trials SNR (db) Figure. Saple USIC breakdown for closely spaced sources. Note the SNR gap between reliable detection using ITC and reliable estiation using USIC. Fig. plots the probability of underestiating the correct nuber of sources as a function of SNR for the AIC, DL, and AP inforation theoretic criteria (ITC). The sae figure shows the saple probability of conventional USIC breakdown, which illustrates the well-docuented fact that USIC breaks at SNR values significantly larger than the actual LE threshold [8]. In this paper, we exaine the perforance at ε ITC = 30dB, where Fig. shows the reliable detection of the correct nuber of sources = 2, but the L DOA estiate itself is questionable. It can be separately confired, following the ethodology in [0], that for this scenario, the eigenvalue splitting condition As4, required for G-consistency of the eigenvalue estiation, are et for the 30dB case in this scenario. The fact that this condition is et here coes with no surprise. Indeed, the reliable detection of the correct nuber of sources = 2 at this SNR already iplies that the cluster of noise subspace saple eigenvalues (ˆλ to ˆλ 6 ) is reliably separated fro the cluster of the sallest signal eigenvalue ˆλ 7. Since λ 8 λ 7 for our scenario with closely spaced sources, it is really the separation between the λ 7 cluster and the noise subspace one (λ = = λ 6 = ), which is of concern for evaluation of the As4 condition ULA, θ = [0, 0.40] deg, T = 0, 30dB SNR, G, {U#30} G asy Eig Standard Eig saple pdf Eigenvalue Figure 2. Saple p.d.f.s for the eigenvalues of ˆR G We therefore exaine the eigendecoposition of the G-reconstructed covariance atrix ˆR G. As predicted, the eigenvectors of ˆR and ˆR G are exactly the sae, and only the eigenvalues differ slightly (and then only in the noise subspace). The two subspace eigenvalues are also well separated both fro each other and the noise subspace eigenvalues. While in ˆR the noise subspace is described by a single value (= close but not precisely equal to ˆλ (i.e. eig ( ˆR ) G < ˆλ G ˆλ j ), in ˆR G they reain < eig 6 ( ˆR )). G This is deonstrated in Fig. 2, where we introduce saple p.d.f.s for eig 7 ( ˆR ), G eig 8 ( ˆR ), G and the noise eigenvalues eig n ( ˆR ) G = (eig ( ˆR ), G..., eig 6 ( ˆR )), G where here we treat as the noise subspace eigenvalue in ˆR G any one of its six sallest eigenvalues. The p.d.f.s for ˆλ and eig n ( ˆR ) G practically coincide. While Fig. 2 does deonstrate inor variations in the noise subspace eigenvalues in ˆR G around the single noise subspace eigenvalue ˆλ G in ˆR, caused by finite T and values, these variations are very unlikely to result in any noticeable changes in DOA estiation perforance when ˆR G is used instead of ˆR or ˆR in the likelihood function, siply because even larger variations of noise subspace eigenvalues in ˆR around a single L estiate (= ˆλ j ) in ˆR are also not associated with any DOA L estiation perforance degradation. Yet, since all the conditions (As-As4) of the Theore 3 [0] are et, we can expect that these inor variations do result in iproved individual eigenvalue ˆλ G j and eigensubspace ˆP j G estiation accuracy. To confir this, we repeat the analysis conducted in [0], where to assess the quality of the eigensubspace estiates, estre introduced an orthogonality factor: O(j) = tr [P ˆP j j ] tr [I n P j ] ˆP j. (37) Here P j, j =, 2, 3 are the eigenvectors/subspaces of the true covariance atrix R 0 and ˆP j are the estiates

8 Advances in Signal Processing 2(): 8-28, (either standard L ones or the G-consistent ˆP j G ones). This orthogonality factor is the ratio of the total power of the estiated subspace ˆP j that resides in the true subspace P j to the power of its leakage into the true orthogonal subspace [I P j ] ULA, θ = [0, 0.40] deg, T = 0, 30dB SNR, G, 500 {U#30} Rhat Ghat ULA, θ = [0, 0.40] deg, T = 0, 30dB SNR, G, 500 {U#30} Rhat Ghat saple pdf 0. saple pdf orthogonality orthogonality 3 Figure 3. Saple p.d.f.s for LE and GLE orthogonality factors for the two signal eigenvectors. The estiate ˆP G j (as defined in (30)) is better than the LE ˆP j, defined as ˆP = 6 ÛjÛ H j ; ˆP2 = Û7Û H 7 ; ˆP 3 = Û8Û H 8 if, with high probability, the respective orthogonality factor is greater. In Fig. 3, we introduce saple p.d.f.s for the orthogonality factor O(j) and O(j) G, averaged over the sae 500 onte-carlo trials for the SNR=30dB case. Siilarly to [0], we observe that the G-consistent subspace estiate achieves a uch higher orthogonality factor than the traditional L saple estiator, indicating a better capacity to estiate subspaces, and therefore better accuracy for subspace-based DOA estiation techniques such as USIC. To deonstrate, we copare the results of exhaustive global axiu search for the conventional (noralized) likelihood function (ratio): LR[R(Ω )] = det R (Ω ) ˆRe e tr R (Ω ) ˆR and its G-asyptotic counterpart GLR[R(Ω )] constructed using ˆR G instead of ˆR. (38) 8 ULA, θ = [0, 0.40] deg, T = 0, 30dB SNR, R, 500 {U#30} ULA, θ = [0, 0.40] deg, T = 0, 30dB SNR, G, 500 {U#30} 0 0 saple pdf saple pdf θ (deg) θ (deg) Figure 4. Saple p.d.f.s for LE and GLE DOA estiation at 30dB SNR; a- ˆR, b- ˆR G. This global optiization is perfored in two steps. In step, we find the global extreu over a fine grid on a half-plane (θ < θ 2 ), and in step 2, we conduct local optiization over the paraeters θ < θ 2 as well as the two source powers using the ATLAB finco routine. The white noise power is assued known a priori and set to unity. ore details on the optiization routine and LE results can be found in [, 2, 4]. Due to our convention that ˆθ < ˆθ 2, the DOA saple distributions illustrated by Fig. 4 overlap, since for scenarios with very closely spaced sources, over the 500 trials, the iniu ˆθ 2 was soeties less than the axiu ˆθ, even though on each individual trial θ was less than θ 2. One can see that the distribution for the DOAs derived using the standard LR (38) and the G-asyptotic counterpart are practically indistinguishable, as are any oents of these distributions. oreover, in addition to the identical statistical perforance of the DOA estiates, identical individual estiates were registered in a trial-by-trial basis. In addition to the illustrated results, the sae results were observed at several higher SNRs, and were also the sae when the L covariance atrix estiate ˆR = was used. Finally, in a copletely ad-hoc ˆγL ix and atch approach, we also assigned the G-consistent eigenvalue estiates ˆλ G (the G-consistent noise power estiate), ˆλ G 2 and ˆλ G 3 (5) to the axiu likelihood (utually orthogonal) eigensubspaces ˆP, ˆP2, and ˆP 3 and once again did not record any DOA estiation iproveents, showing that the inor variations in noise subspace eigenvalues in ˆR G with respect to ˆR was not significant in LE. P L

9 26 Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? 4 Concluding Rearks The theoretical results fro RT and their exaination by direct onte-carlo siulations has confired that for Gaussian sources in Gaussian noise, L DOA estiation in the so-called threshold region is not iproved by use of recent results fro Rando atrix Theory. ore specifically, we deonstrated that the Girko Lea 3. fro [6] iplies that for the ulti-variate (coplex) Gaussian data case, the stochastic (unconditional) likelihood function is both a T -consistent and G-consistent estiate of the function / tr [R (Ω )R 0 ], where R 0 is represented by a set of T i.i.d. training vectors x j. We deonstrated that in full accordance with this Lea, an alternative reconstruction of the log-likelihood function using individually G-consistent estiates of eigenvalues and eigenvectors does not iprove DOA estiation threshold perforance, despite the fact that these individual G-consistent estiates are ore accurate than their corresponding L estiates. Thus, the deonstrated in [9 ] iproveent in USIC perforance delivered by the G-consistent estiates of the individual noise eigensubspace is not at odds with the lack of iproveent in LE perforance. While applicable to arbitrary paraetric description of the covariance atrix R 0, these results should not be extended beyond the ultivariate (coplex) i.i.d. Gaussian case without detailed exaination. Finally, it is iportant to stress that our results are confined to the currently (in [0]) unique set of G-consistent eigenvalue and eigenvector estiates, and therefore the broader question on whether the threshold perforance of L DOA estiation can be iproved by soe other technique reains open. Acknowledgeents The authors express their gratitude to Dr. Xavier estre of the Telecounications Technological Center of Catalonia (CTTC) for his guidance on RT results. Appendix Proof of Theore The G-consistent eigenvalue estiators defined in Theore 3 in [0] ((5) in the ain body of the paper) can be restructured as K ˆγ G = T ( ˆλ k ˆµ k ), (4.) = k k and therefore we have to prove that T ˆµ k = (T ) ˆλ k. (4.2) Given eqn. (7) in the ain body of the paper, we ay define T ˆµ k = = ˆλ ˆµ k ˆλ ˆµ k. (4.3) Substitution shows that instead of the for in eqn. (7), ˆµ,..., ˆµ can be introduced as the roots of the following polynoial: Q(µ) = (ˆµ k µ) = (ˆλ k µ) T ˆλ r (ˆλ j µ) j r (4.4) where the right-hand side expression is obtained fro eqn. (7). Evaluating Q(µ) at µ = λ, we obtain (ˆλ ˆµ k ) = T ˆλ j (ˆλ ˆλ j ). (4.5)

10 Advances in Signal Processing 2(): 8-28, On the other hand, observing the for of the derivative of Q (µ): Q (µ) = = and evaluating it at µ = ˆλ, we obtain l= (ˆµ k µ) = k l (ˆλ j µ) + T l= j l ˆλ r (ˆλ j µ) l= l r j r j l (4.6) (ˆµ k ˆλ ) = (ˆλ j ˆλ ) T ˆλ l= k l j l= l j j l (ˆλ j ˆλ ) T ˆλ r r j r j (ˆλ j ˆλ ) (4.7) Dividing both sides of (4.7) by (ˆλ j ˆλ ) and using (4.5) to siplify the left-hand side, we obtain j r, T l= ˆλ ˆλ ˆµ l = T r ˆλ + ˆλ r ˆλ r ˆλ (4.8) (Note that (4.4)-(4.8) follow Appendix IV in in [0]). On the other hand, fro (4.3) we get ˆλ k ˆµ j ˆλ k ˆµ j = ˆλ k ˆλ k (4.9) ˆλ k ˆµ j Fro (4.8), it then follows that ˆλ k ˆλ k ˆµ j = T ˆλ k + ˆλ r ˆλ r ˆλ k (4.0) Since ˆλ k + ˆλ r ˆλ r ˆλ k + = ( 2ˆλ r ˆλ r ˆλ k ) (4.) we get ˆλ k ˆµ j ˆλ k ˆµ j = (T + ) ˆλ k 2ˆλ kˆλr ˆλ r ˆλ k. (4.2) But for ˆλ k > 0, the following identity is true (see [0]) ˆλ kˆλr ˆλ r ˆλ k = 0. (4.3) Therefore, we get T ˆµ k = (T ) ˆλ k (4.4) as required by (4.2).

11 28 Can the Threshold Perforance of axiu Likelihood DOA Estiation be Iproved by Tools fro Rando atrix Theory? REFERENCES [] Y.I. Abraovich, B.A. Johnson, and N.K. Spencer. Statistical nonidentifiability of close eitters: axiu-likelihood estiation breakdown. In Proc. EUSIPCO-2009, pages , Glasgow, UK, [2] Y.I. Abraovich, B.A. Johnson, and N.K. Spencer. Statistical nonidentifiability of close eitters: axiu-likelihood estiation breakdown and its GSA analysis. In Proc. ICASSP-2009, pages , Taipei, Taiwan, [3] F. Athley. Threshold region perforance of axiu likelihood direction of arrival estiators. IEEE Trans. Signal Processing, 53(4): , Apr [4] P. Forster, P. Larzabal, and E. Boyer. Threshold perforance analysis of axiu likelihood DOA estiation. IEEE Trans. Sig. Proc., 52():383 39, Nov [5] A.B. Gershan. Pseudo-randoly generated estiator banks: a new tool for iproving the threshold perforance of direction finding. IEEE Trans. Sig. Proc., 46(5):35 364, ay 998. [6] V.L. Girko. An Introduction to Statistical Analysis of Rando Arrays. VSP, Utrecht, Netherlands, 998. [7] R.A. Janik and.a. Nowak. Wishart and anti-wishart rando atrices. J. Phys. A, 36: , [8] B.A. Johnson, Y.I. Abraovich, and X. estre. USIC, G-USIC, and axiu-likelihood perforance breakdown. IEEE Trans. Sig. Proc., 56(8): , August [9] X. estre. An iproved subspace based algorith for sall saple size regie. In Proc. ICASSP-06, Toulouse, [0] X. estre. Iproved estiation of eigenvalues and eigenvectors of covariance atrices using their saple estiates. IEEE Trans. Info. Theory, 54():53 529, Noveber [] X. estre. On the asyptotic behaviour of the saple estiates of eigenvalues and eigenvectors of covariance atrices. IEEE Trans. Sig. Proc., 56(): , Noveber [2] Xavier estre. E-ail counications. arch 05, 200. [3] R.J. uirhead. Aspects of ultivariate Statistical Theory. Wiley, New York, 982. [4] P. Parvazi, A.B. Gershan, and Y. Abraovich. Detecting outliers in the estiator bank-based direction finding techniques using the likelihood ratio quality assessent. In Proc. ICASSP, volue II, pages , Honolulu, Hawaii, April IEEE. [5] Christ D. Richond. ean-squared error perforance prediction of axiu-likelihood signal paraeter estiation. In Proceedings of the Adaptive Sensor Array Processing (ASAP) Workshop, Lexington, A, -3 ar 998. IT-LL. DTIC Ascension ADA [6] D. C. Rife and R. R. Boorstyn. ultiple tone paraeter estiation fro discrete-tie observations. Bell Syste Technical Journal, 55(9):389 40, Nov 976. [7] S. Zacks. The Theory of Statistical Inference. Wiley, New York, 97. [8] K. Zarifi and A.B. Gershan. Asyptotic perforance analysis of blind iniu output energy receivers for large DS-CDA systes. IEEE Trans. Sig. Proc., 56(2): , Feb 2008.

HIGH RESOLUTION NEAR-FIELD MULTIPLE TARGET DETECTION AND LOCALIZATION USING SUPPORT VECTOR MACHINES

HIGH RESOLUTION NEAR-FIELD MULTIPLE TARGET DETECTION AND LOCALIZATION USING SUPPORT VECTOR MACHINES ICONIC 2007 St. Louis, O, USA June 27-29, 2007 HIGH RESOLUTION NEAR-FIELD ULTIPLE TARGET DETECTION AND LOCALIZATION USING SUPPORT VECTOR ACHINES A. Randazzo,. A. Abou-Khousa 2,.Pastorino, and R. Zoughi

More information

Support recovery in compressed sensing: An estimation theoretic approach

Support recovery in compressed sensing: An estimation theoretic approach Support recovery in copressed sensing: An estiation theoretic approach Ain Karbasi, Ali Horati, Soheil Mohajer, Martin Vetterli School of Coputer and Counication Sciences École Polytechnique Fédérale de

More information

Non-Parametric Non-Line-of-Sight Identification 1

Non-Parametric Non-Line-of-Sight Identification 1 Non-Paraetric Non-Line-of-Sight Identification Sinan Gezici, Hisashi Kobayashi and H. Vincent Poor Departent of Electrical Engineering School of Engineering and Applied Science Princeton University, Princeton,

More information

Feature Extraction Techniques

Feature Extraction Techniques Feature Extraction Techniques Unsupervised Learning II Feature Extraction Unsupervised ethods can also be used to find features which can be useful for categorization. There are unsupervised ethods that

More information

TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES

TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES TEST OF HOMOGENEITY OF PARALLEL SAMPLES FROM LOGNORMAL POPULATIONS WITH UNEQUAL VARIANCES S. E. Ahed, R. J. Tokins and A. I. Volodin Departent of Matheatics and Statistics University of Regina Regina,

More information

Chapter 6 1-D Continuous Groups

Chapter 6 1-D Continuous Groups Chapter 6 1-D Continuous Groups Continuous groups consist of group eleents labelled by one or ore continuous variables, say a 1, a 2,, a r, where each variable has a well- defined range. This chapter explores:

More information

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013).

The proofs of Theorem 1-3 are along the lines of Wied and Galeano (2013). A Appendix: Proofs The proofs of Theore 1-3 are along the lines of Wied and Galeano (2013) Proof of Theore 1 Let D[d 1, d 2 ] be the space of càdlàg functions on the interval [d 1, d 2 ] equipped with

More information

Randomized Recovery for Boolean Compressed Sensing

Randomized Recovery for Boolean Compressed Sensing Randoized Recovery for Boolean Copressed Sensing Mitra Fatei and Martin Vetterli Laboratory of Audiovisual Counication École Polytechnique Fédéral de Lausanne (EPFL) Eail: {itra.fatei, artin.vetterli}@epfl.ch

More information

An introduction to G-estimation with sample covariance matrices

An introduction to G-estimation with sample covariance matrices An introduction to G-estimation with sample covariance matrices Xavier estre xavier.mestre@cttc.cat Centre Tecnològic de Telecomunicacions de Catalunya (CTTC) "atrices aleatoires: applications aux communications

More information

SPECTRUM sensing is a core concept of cognitive radio

SPECTRUM sensing is a core concept of cognitive radio World Acadey of Science, Engineering and Technology International Journal of Electronics and Counication Engineering Vol:6, o:2, 202 Efficient Detection Using Sequential Probability Ratio Test in Mobile

More information

Sharp Time Data Tradeoffs for Linear Inverse Problems

Sharp Time Data Tradeoffs for Linear Inverse Problems Sharp Tie Data Tradeoffs for Linear Inverse Probles Saet Oyak Benjain Recht Mahdi Soltanolkotabi January 016 Abstract In this paper we characterize sharp tie-data tradeoffs for optiization probles used

More information

Block designs and statistics

Block designs and statistics Bloc designs and statistics Notes for Math 447 May 3, 2011 The ain paraeters of a bloc design are nuber of varieties v, bloc size, nuber of blocs b. A design is built on a set of v eleents. Each eleent

More information

An Improved Particle Filter with Applications in Ballistic Target Tracking

An Improved Particle Filter with Applications in Ballistic Target Tracking Sensors & ransducers Vol. 72 Issue 6 June 204 pp. 96-20 Sensors & ransducers 204 by IFSA Publishing S. L. http://www.sensorsportal.co An Iproved Particle Filter with Applications in Ballistic arget racing

More information

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search

Quantum algorithms (CO 781, Winter 2008) Prof. Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search Quantu algoriths (CO 781, Winter 2008) Prof Andrew Childs, University of Waterloo LECTURE 15: Unstructured search and spatial search ow we begin to discuss applications of quantu walks to search algoriths

More information

Detection and Estimation Theory

Detection and Estimation Theory ESE 54 Detection and Estiation Theory Joseph A. O Sullivan Sauel C. Sachs Professor Electronic Systes and Signals Research Laboratory Electrical and Systes Engineering Washington University 11 Urbauer

More information

Polygonal Designs: Existence and Construction

Polygonal Designs: Existence and Construction Polygonal Designs: Existence and Construction John Hegean Departent of Matheatics, Stanford University, Stanford, CA 9405 Jeff Langford Departent of Matheatics, Drake University, Des Moines, IA 5011 G

More information

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution

Keywords: Estimator, Bias, Mean-squared error, normality, generalized Pareto distribution Testing approxiate norality of an estiator using the estiated MSE and bias with an application to the shape paraeter of the generalized Pareto distribution J. Martin van Zyl Abstract In this work the norality

More information

Machine Learning Basics: Estimators, Bias and Variance

Machine Learning Basics: Estimators, Bias and Variance Machine Learning Basics: Estiators, Bias and Variance Sargur N. srihari@cedar.buffalo.edu This is part of lecture slides on Deep Learning: http://www.cedar.buffalo.edu/~srihari/cse676 1 Topics in Basics

More information

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon

Model Fitting. CURM Background Material, Fall 2014 Dr. Doreen De Leon Model Fitting CURM Background Material, Fall 014 Dr. Doreen De Leon 1 Introduction Given a set of data points, we often want to fit a selected odel or type to the data (e.g., we suspect an exponential

More information

The Transactional Nature of Quantum Information

The Transactional Nature of Quantum Information The Transactional Nature of Quantu Inforation Subhash Kak Departent of Coputer Science Oklahoa State University Stillwater, OK 7478 ABSTRACT Inforation, in its counications sense, is a transactional property.

More information

PULSE-TRAIN BASED TIME-DELAY ESTIMATION IMPROVES RESILIENCY TO NOISE

PULSE-TRAIN BASED TIME-DELAY ESTIMATION IMPROVES RESILIENCY TO NOISE PULSE-TRAIN BASED TIME-DELAY ESTIMATION IMPROVES RESILIENCY TO NOISE 1 Nicola Neretti, 1 Nathan Intrator and 1,2 Leon N Cooper 1 Institute for Brain and Neural Systes, Brown University, Providence RI 02912.

More information

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion

Supplementary Material for Fast and Provable Algorithms for Spectrally Sparse Signal Reconstruction via Low-Rank Hankel Matrix Completion Suppleentary Material for Fast and Provable Algoriths for Spectrally Sparse Signal Reconstruction via Low-Ran Hanel Matrix Copletion Jian-Feng Cai Tianing Wang Ke Wei March 1, 017 Abstract We establish

More information

3.3 Variational Characterization of Singular Values

3.3 Variational Characterization of Singular Values 3.3. Variational Characterization of Singular Values 61 3.3 Variational Characterization of Singular Values Since the singular values are square roots of the eigenvalues of the Heritian atrices A A and

More information

Using a De-Convolution Window for Operating Modal Analysis

Using a De-Convolution Window for Operating Modal Analysis Using a De-Convolution Window for Operating Modal Analysis Brian Schwarz Vibrant Technology, Inc. Scotts Valley, CA Mark Richardson Vibrant Technology, Inc. Scotts Valley, CA Abstract Operating Modal Analysis

More information

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization

Support Vector Machine Classification of Uncertain and Imbalanced data using Robust Optimization Recent Researches in Coputer Science Support Vector Machine Classification of Uncertain and Ibalanced data using Robust Optiization RAGHAV PAT, THEODORE B. TRAFALIS, KASH BARKER School of Industrial Engineering

More information

Testing equality of variances for multiple univariate normal populations

Testing equality of variances for multiple univariate normal populations University of Wollongong Research Online Centre for Statistical & Survey Methodology Working Paper Series Faculty of Engineering and Inforation Sciences 0 esting equality of variances for ultiple univariate

More information

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines

Intelligent Systems: Reasoning and Recognition. Perceptrons and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley osig 1 Winter Seester 2018 Lesson 6 27 February 2018 Outline Perceptrons and Support Vector achines Notation...2 Linear odels...3 Lines, Planes

More information

Estimating Parameters for a Gaussian pdf

Estimating Parameters for a Gaussian pdf Pattern Recognition and achine Learning Jaes L. Crowley ENSIAG 3 IS First Seester 00/0 Lesson 5 7 Noveber 00 Contents Estiating Paraeters for a Gaussian pdf Notation... The Pattern Recognition Proble...3

More information

Extension of CSRSM for the Parametric Study of the Face Stability of Pressurized Tunnels

Extension of CSRSM for the Parametric Study of the Face Stability of Pressurized Tunnels Extension of CSRSM for the Paraetric Study of the Face Stability of Pressurized Tunnels Guilhe Mollon 1, Daniel Dias 2, and Abdul-Haid Soubra 3, M.ASCE 1 LGCIE, INSA Lyon, Université de Lyon, Doaine scientifique

More information

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics

ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS. A Thesis. Presented to. The Faculty of the Department of Mathematics ESTIMATING AND FORMING CONFIDENCE INTERVALS FOR EXTREMA OF RANDOM POLYNOMIALS A Thesis Presented to The Faculty of the Departent of Matheatics San Jose State University In Partial Fulfillent of the Requireents

More information

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices

13.2 Fully Polynomial Randomized Approximation Scheme for Permanent of Random 0-1 Matrices CS71 Randoness & Coputation Spring 018 Instructor: Alistair Sinclair Lecture 13: February 7 Disclaier: These notes have not been subjected to the usual scrutiny accorded to foral publications. They ay

More information

Lower Bounds for Quantized Matrix Completion

Lower Bounds for Quantized Matrix Completion Lower Bounds for Quantized Matrix Copletion Mary Wootters and Yaniv Plan Departent of Matheatics University of Michigan Ann Arbor, MI Eail: wootters, yplan}@uich.edu Mark A. Davenport School of Elec. &

More information

A remark on a success rate model for DPA and CPA

A remark on a success rate model for DPA and CPA A reark on a success rate odel for DPA and CPA A. Wieers, BSI Version 0.5 andreas.wieers@bsi.bund.de Septeber 5, 2018 Abstract The success rate is the ost coon evaluation etric for easuring the perforance

More information

CS Lecture 13. More Maximum Likelihood

CS Lecture 13. More Maximum Likelihood CS 6347 Lecture 13 More Maxiu Likelihood Recap Last tie: Introduction to axiu likelihood estiation MLE for Bayesian networks Optial CPTs correspond to epirical counts Today: MLE for CRFs 2 Maxiu Likelihood

More information

Fairness via priority scheduling

Fairness via priority scheduling Fairness via priority scheduling Veeraruna Kavitha, N Heachandra and Debayan Das IEOR, IIT Bobay, Mubai, 400076, India vavitha,nh,debayan}@iitbacin Abstract In the context of ulti-agent resource allocation

More information

Compressive Distilled Sensing: Sparse Recovery Using Adaptivity in Compressive Measurements

Compressive Distilled Sensing: Sparse Recovery Using Adaptivity in Compressive Measurements 1 Copressive Distilled Sensing: Sparse Recovery Using Adaptivity in Copressive Measureents Jarvis D. Haupt 1 Richard G. Baraniuk 1 Rui M. Castro 2 and Robert D. Nowak 3 1 Dept. of Electrical and Coputer

More information

On Constant Power Water-filling

On Constant Power Water-filling On Constant Power Water-filling Wei Yu and John M. Cioffi Electrical Engineering Departent Stanford University, Stanford, CA94305, U.S.A. eails: {weiyu,cioffi}@stanford.edu Abstract This paper derives

More information

W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS

W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS W-BASED VS LATENT VARIABLES SPATIAL AUTOREGRESSIVE MODELS: EVIDENCE FROM MONTE CARLO SIMULATIONS. Introduction When it coes to applying econoetric odels to analyze georeferenced data, researchers are well

More information

Least Squares Fitting of Data

Least Squares Fitting of Data Least Squares Fitting of Data David Eberly, Geoetric Tools, Redond WA 98052 https://www.geoetrictools.co/ This work is licensed under the Creative Coons Attribution 4.0 International License. To view a

More information

A Simple Regression Problem

A Simple Regression Problem A Siple Regression Proble R. M. Castro March 23, 2 In this brief note a siple regression proble will be introduced, illustrating clearly the bias-variance tradeoff. Let Y i f(x i ) + W i, i,..., n, where

More information

Testing the lag length of vector autoregressive models: A power comparison between portmanteau and Lagrange multiplier tests

Testing the lag length of vector autoregressive models: A power comparison between portmanteau and Lagrange multiplier tests Working Papers 2017-03 Testing the lag length of vector autoregressive odels: A power coparison between portanteau and Lagrange ultiplier tests Raja Ben Hajria National Engineering School, University of

More information

Real-time Super-resolution Sound Source Localization for Robots

Real-time Super-resolution Sound Source Localization for Robots 22 IEEE/RSJ International Conference on Intelligent Robots and Systes October 7-2, 22. Vilaoura, Algarve, Portugal Real-tie Super-resolution Sound Source Localization for Robots Keisuke Nakaura, Kazuhiro

More information

A Note on the Applied Use of MDL Approximations

A Note on the Applied Use of MDL Approximations A Note on the Applied Use of MDL Approxiations Daniel J. Navarro Departent of Psychology Ohio State University Abstract An applied proble is discussed in which two nested psychological odels of retention

More information

ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD

ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD PROCEEDINGS OF THE YEREVAN STATE UNIVERSITY Physical and Matheatical Sciences 04,, p. 7 5 ON THE TWO-LEVEL PRECONDITIONING IN LEAST SQUARES METHOD M a t h e a t i c s Yu. A. HAKOPIAN, R. Z. HOVHANNISYAN

More information

e-companion ONLY AVAILABLE IN ELECTRONIC FORM

e-companion ONLY AVAILABLE IN ELECTRONIC FORM OPERATIONS RESEARCH doi 10.1287/opre.1070.0427ec pp. ec1 ec5 e-copanion ONLY AVAILABLE IN ELECTRONIC FORM infors 07 INFORMS Electronic Copanion A Learning Approach for Interactive Marketing to a Custoer

More information

ANALYTICAL INVESTIGATION AND PARAMETRIC STUDY OF LATERAL IMPACT BEHAVIOR OF PRESSURIZED PIPELINES AND INFLUENCE OF INTERNAL PRESSURE

ANALYTICAL INVESTIGATION AND PARAMETRIC STUDY OF LATERAL IMPACT BEHAVIOR OF PRESSURIZED PIPELINES AND INFLUENCE OF INTERNAL PRESSURE DRAFT Proceedings of the ASME 014 International Mechanical Engineering Congress & Exposition IMECE014 Noveber 14-0, 014, Montreal, Quebec, Canada IMECE014-36371 ANALYTICAL INVESTIGATION AND PARAMETRIC

More information

Complex Quadratic Optimization and Semidefinite Programming

Complex Quadratic Optimization and Semidefinite Programming Coplex Quadratic Optiization and Seidefinite Prograing Shuzhong Zhang Yongwei Huang August 4 Abstract In this paper we study the approxiation algoriths for a class of discrete quadratic optiization probles

More information

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis

E0 370 Statistical Learning Theory Lecture 6 (Aug 30, 2011) Margin Analysis E0 370 tatistical Learning Theory Lecture 6 (Aug 30, 20) Margin Analysis Lecturer: hivani Agarwal cribe: Narasihan R Introduction In the last few lectures we have seen how to obtain high confidence bounds

More information

COS 424: Interacting with Data. Written Exercises

COS 424: Interacting with Data. Written Exercises COS 424: Interacting with Data Hoework #4 Spring 2007 Regression Due: Wednesday, April 18 Written Exercises See the course website for iportant inforation about collaboration and late policies, as well

More information

Physics 215 Winter The Density Matrix

Physics 215 Winter The Density Matrix Physics 215 Winter 2018 The Density Matrix The quantu space of states is a Hilbert space H. Any state vector ψ H is a pure state. Since any linear cobination of eleents of H are also an eleent of H, it

More information

Symbolic Analysis as Universal Tool for Deriving Properties of Non-linear Algorithms Case study of EM Algorithm

Symbolic Analysis as Universal Tool for Deriving Properties of Non-linear Algorithms Case study of EM Algorithm Acta Polytechnica Hungarica Vol., No., 04 Sybolic Analysis as Universal Tool for Deriving Properties of Non-linear Algoriths Case study of EM Algorith Vladiir Mladenović, Miroslav Lutovac, Dana Porrat

More information

On Conditions for Linearity of Optimal Estimation

On Conditions for Linearity of Optimal Estimation On Conditions for Linearity of Optial Estiation Erah Akyol, Kuar Viswanatha and Kenneth Rose {eakyol, kuar, rose}@ece.ucsb.edu Departent of Electrical and Coputer Engineering University of California at

More information

EE5900 Spring Lecture 4 IC interconnect modeling methods Zhuo Feng

EE5900 Spring Lecture 4 IC interconnect modeling methods Zhuo Feng EE59 Spring Parallel LSI AD Algoriths Lecture I interconnect odeling ethods Zhuo Feng. Z. Feng MTU EE59 So far we ve considered only tie doain analyses We ll soon see that it is soeties preferable to odel

More information

Identical Maximum Likelihood State Estimation Based on Incremental Finite Mixture Model in PHD Filter

Identical Maximum Likelihood State Estimation Based on Incremental Finite Mixture Model in PHD Filter Identical Maxiu Lielihood State Estiation Based on Increental Finite Mixture Model in PHD Filter Gang Wu Eail: xjtuwugang@gail.co Jing Liu Eail: elelj20080730@ail.xjtu.edu.cn Chongzhao Han Eail: czhan@ail.xjtu.edu.cn

More information

SEISMIC FRAGILITY ANALYSIS

SEISMIC FRAGILITY ANALYSIS 9 th ASCE Specialty Conference on Probabilistic Mechanics and Structural Reliability PMC24 SEISMIC FRAGILITY ANALYSIS C. Kafali, Student M. ASCE Cornell University, Ithaca, NY 483 ck22@cornell.edu M. Grigoriu,

More information

A Low-Complexity Congestion Control and Scheduling Algorithm for Multihop Wireless Networks with Order-Optimal Per-Flow Delay

A Low-Complexity Congestion Control and Scheduling Algorithm for Multihop Wireless Networks with Order-Optimal Per-Flow Delay A Low-Coplexity Congestion Control and Scheduling Algorith for Multihop Wireless Networks with Order-Optial Per-Flow Delay Po-Kai Huang, Xiaojun Lin, and Chih-Chun Wang School of Electrical and Coputer

More information

On the Communication Complexity of Lipschitzian Optimization for the Coordinated Model of Computation

On the Communication Complexity of Lipschitzian Optimization for the Coordinated Model of Computation journal of coplexity 6, 459473 (2000) doi:0.006jco.2000.0544, available online at http:www.idealibrary.co on On the Counication Coplexity of Lipschitzian Optiization for the Coordinated Model of Coputation

More information

arxiv: v1 [cs.ds] 3 Feb 2014

arxiv: v1 [cs.ds] 3 Feb 2014 arxiv:40.043v [cs.ds] 3 Feb 04 A Bound on the Expected Optiality of Rando Feasible Solutions to Cobinatorial Optiization Probles Evan A. Sultani The Johns Hopins University APL evan@sultani.co http://www.sultani.co/

More information

Optimal nonlinear Bayesian experimental design: an application to amplitude versus offset experiments

Optimal nonlinear Bayesian experimental design: an application to amplitude versus offset experiments Geophys. J. Int. (23) 155, 411 421 Optial nonlinear Bayesian experiental design: an application to aplitude versus offset experients Jojanneke van den Berg, 1, Andrew Curtis 2,3 and Jeannot Trapert 1 1

More information

Inspection; structural health monitoring; reliability; Bayesian analysis; updating; decision analysis; value of information

Inspection; structural health monitoring; reliability; Bayesian analysis; updating; decision analysis; value of information Cite as: Straub D. (2014). Value of inforation analysis with structural reliability ethods. Structural Safety, 49: 75-86. Value of Inforation Analysis with Structural Reliability Methods Daniel Straub

More information

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval

Uniform Approximation and Bernstein Polynomials with Coefficients in the Unit Interval Unifor Approxiation and Bernstein Polynoials with Coefficients in the Unit Interval Weiang Qian and Marc D. Riedel Electrical and Coputer Engineering, University of Minnesota 200 Union St. S.E. Minneapolis,

More information

A Simplified Analytical Approach for Efficiency Evaluation of the Weaving Machines with Automatic Filling Repair

A Simplified Analytical Approach for Efficiency Evaluation of the Weaving Machines with Automatic Filling Repair Proceedings of the 6th SEAS International Conference on Siulation, Modelling and Optiization, Lisbon, Portugal, Septeber -4, 006 0 A Siplified Analytical Approach for Efficiency Evaluation of the eaving

More information

Recovering Data from Underdetermined Quadratic Measurements (CS 229a Project: Final Writeup)

Recovering Data from Underdetermined Quadratic Measurements (CS 229a Project: Final Writeup) Recovering Data fro Underdeterined Quadratic Measureents (CS 229a Project: Final Writeup) Mahdi Soltanolkotabi Deceber 16, 2011 1 Introduction Data that arises fro engineering applications often contains

More information

Optimal Jamming Over Additive Noise: Vector Source-Channel Case

Optimal Jamming Over Additive Noise: Vector Source-Channel Case Fifty-first Annual Allerton Conference Allerton House, UIUC, Illinois, USA October 2-3, 2013 Optial Jaing Over Additive Noise: Vector Source-Channel Case Erah Akyol and Kenneth Rose Abstract This paper

More information

Kernel Methods and Support Vector Machines

Kernel Methods and Support Vector Machines Intelligent Systes: Reasoning and Recognition Jaes L. Crowley ENSIAG 2 / osig 1 Second Seester 2012/2013 Lesson 20 2 ay 2013 Kernel ethods and Support Vector achines Contents Kernel Functions...2 Quadratic

More information

Using EM To Estimate A Probablity Density With A Mixture Of Gaussians

Using EM To Estimate A Probablity Density With A Mixture Of Gaussians Using EM To Estiate A Probablity Density With A Mixture Of Gaussians Aaron A. D Souza adsouza@usc.edu Introduction The proble we are trying to address in this note is siple. Given a set of data points

More information

Topic 5a Introduction to Curve Fitting & Linear Regression

Topic 5a Introduction to Curve Fitting & Linear Regression /7/08 Course Instructor Dr. Rayond C. Rup Oice: A 337 Phone: (95) 747 6958 E ail: rcrup@utep.edu opic 5a Introduction to Curve Fitting & Linear Regression EE 4386/530 Coputational ethods in EE Outline

More information

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS

AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Statistica Sinica 6 016, 1709-178 doi:http://dx.doi.org/10.5705/ss.0014.0034 AN OPTIMAL SHRINKAGE FACTOR IN PREDICTION OF ORDERED RANDOM EFFECTS Nilabja Guha 1, Anindya Roy, Yaakov Malinovsky and Gauri

More information

Proc. of the IEEE/OES Seventh Working Conference on Current Measurement Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES

Proc. of the IEEE/OES Seventh Working Conference on Current Measurement Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES Proc. of the IEEE/OES Seventh Working Conference on Current Measureent Technology UNCERTAINTIES IN SEASONDE CURRENT VELOCITIES Belinda Lipa Codar Ocean Sensors 15 La Sandra Way, Portola Valley, CA 98 blipa@pogo.co

More information

Bootstrapping Dependent Data

Bootstrapping Dependent Data Bootstrapping Dependent Data One of the key issues confronting bootstrap resapling approxiations is how to deal with dependent data. Consider a sequence fx t g n t= of dependent rando variables. Clearly

More information

On the theoretical analysis of cross validation in compressive sensing

On the theoretical analysis of cross validation in compressive sensing MITSUBISHI ELECTRIC RESEARCH LABORATORIES http://www.erl.co On the theoretical analysis of cross validation in copressive sensing Zhang, J.; Chen, L.; Boufounos, P.T.; Gu, Y. TR2014-025 May 2014 Abstract

More information

A note on the multiplication of sparse matrices

A note on the multiplication of sparse matrices Cent. Eur. J. Cop. Sci. 41) 2014 1-11 DOI: 10.2478/s13537-014-0201-x Central European Journal of Coputer Science A note on the ultiplication of sparse atrices Research Article Keivan Borna 12, Sohrab Aboozarkhani

More information

Probability Distributions

Probability Distributions Probability Distributions In Chapter, we ephasized the central role played by probability theory in the solution of pattern recognition probles. We turn now to an exploration of soe particular exaples

More information

Supporting information for Self-assembly of multicomponent structures in and out of equilibrium

Supporting information for Self-assembly of multicomponent structures in and out of equilibrium Supporting inforation for Self-assebly of ulticoponent structures in and out of equilibriu Stephen Whitela 1, Rebecca Schulan 2, Lester Hedges 1 1 Molecular Foundry, Lawrence Berkeley National Laboratory,

More information

Nonmonotonic Networks. a. IRST, I Povo (Trento) Italy, b. Univ. of Trento, Physics Dept., I Povo (Trento) Italy

Nonmonotonic Networks. a. IRST, I Povo (Trento) Italy, b. Univ. of Trento, Physics Dept., I Povo (Trento) Italy Storage Capacity and Dynaics of Nononotonic Networks Bruno Crespi a and Ignazio Lazzizzera b a. IRST, I-38050 Povo (Trento) Italy, b. Univ. of Trento, Physics Dept., I-38050 Povo (Trento) Italy INFN Gruppo

More information

Weighted- 1 minimization with multiple weighting sets

Weighted- 1 minimization with multiple weighting sets Weighted- 1 iniization with ultiple weighting sets Hassan Mansour a,b and Özgür Yılaza a Matheatics Departent, University of British Colubia, Vancouver - BC, Canada; b Coputer Science Departent, University

More information

Chaotic Coupled Map Lattices

Chaotic Coupled Map Lattices Chaotic Coupled Map Lattices Author: Dustin Keys Advisors: Dr. Robert Indik, Dr. Kevin Lin 1 Introduction When a syste of chaotic aps is coupled in a way that allows the to share inforation about each

More information

Antenna Saturation Effects on MIMO Capacity

Antenna Saturation Effects on MIMO Capacity Antenna Saturation Effects on MIMO Capacity T S Pollock, T D Abhayapala, and R A Kennedy National ICT Australia Research School of Inforation Sciences and Engineering The Australian National University,

More information

A method to determine relative stroke detection efficiencies from multiplicity distributions

A method to determine relative stroke detection efficiencies from multiplicity distributions A ethod to deterine relative stroke detection eiciencies ro ultiplicity distributions Schulz W. and Cuins K. 2. Austrian Lightning Detection and Inoration Syste (ALDIS), Kahlenberger Str.2A, 90 Vienna,

More information

ASSUME a source over an alphabet size m, from which a sequence of n independent samples are drawn. The classical

ASSUME a source over an alphabet size m, from which a sequence of n independent samples are drawn. The classical IEEE TRANSACTIONS ON INFORMATION THEORY Large Alphabet Source Coding using Independent Coponent Analysis Aichai Painsky, Meber, IEEE, Saharon Rosset and Meir Feder, Fellow, IEEE arxiv:67.7v [cs.it] Jul

More information

RECOVERY OF A DENSITY FROM THE EIGENVALUES OF A NONHOMOGENEOUS MEMBRANE

RECOVERY OF A DENSITY FROM THE EIGENVALUES OF A NONHOMOGENEOUS MEMBRANE Proceedings of ICIPE rd International Conference on Inverse Probles in Engineering: Theory and Practice June -8, 999, Port Ludlow, Washington, USA : RECOVERY OF A DENSITY FROM THE EIGENVALUES OF A NONHOMOGENEOUS

More information

Ştefan ŞTEFĂNESCU * is the minimum global value for the function h (x)

Ştefan ŞTEFĂNESCU * is the minimum global value for the function h (x) 7Applying Nelder Mead s Optiization Algorith APPLYING NELDER MEAD S OPTIMIZATION ALGORITHM FOR MULTIPLE GLOBAL MINIMA Abstract Ştefan ŞTEFĂNESCU * The iterative deterinistic optiization ethod could not

More information

About the definition of parameters and regimes of active two-port networks with variable loads on the basis of projective geometry

About the definition of parameters and regimes of active two-port networks with variable loads on the basis of projective geometry About the definition of paraeters and regies of active two-port networks with variable loads on the basis of projective geoetry PENN ALEXANDR nstitute of Electronic Engineering and Nanotechnologies "D

More information

A NEW ROBUST AND EFFICIENT ESTIMATOR FOR ILL-CONDITIONED LINEAR INVERSE PROBLEMS WITH OUTLIERS

A NEW ROBUST AND EFFICIENT ESTIMATOR FOR ILL-CONDITIONED LINEAR INVERSE PROBLEMS WITH OUTLIERS A NEW ROBUST AND EFFICIENT ESTIMATOR FOR ILL-CONDITIONED LINEAR INVERSE PROBLEMS WITH OUTLIERS Marta Martinez-Caara 1, Michael Mua 2, Abdelhak M. Zoubir 2, Martin Vetterli 1 1 School of Coputer and Counication

More information

Computational and Statistical Learning Theory

Computational and Statistical Learning Theory Coputational and Statistical Learning Theory TTIC 31120 Prof. Nati Srebro Lecture 2: PAC Learning and VC Theory I Fro Adversarial Online to Statistical Three reasons to ove fro worst-case deterinistic

More information

ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER

ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER IEPC 003-0034 ANALYSIS OF HALL-EFFECT THRUSTERS AND ION ENGINES FOR EARTH-TO-MOON TRANSFER A. Bober, M. Guelan Asher Space Research Institute, Technion-Israel Institute of Technology, 3000 Haifa, Israel

More information

A new type of lower bound for the largest eigenvalue of a symmetric matrix

A new type of lower bound for the largest eigenvalue of a symmetric matrix Linear Algebra and its Applications 47 7 9 9 www.elsevier.co/locate/laa A new type of lower bound for the largest eigenvalue of a syetric atrix Piet Van Mieghe Delft University of Technology, P.O. Box

More information

OPTIMIZATION in multi-agent networks has attracted

OPTIMIZATION in multi-agent networks has attracted Distributed constrained optiization and consensus in uncertain networks via proxial iniization Kostas Margellos, Alessandro Falsone, Sione Garatti and Maria Prandini arxiv:603.039v3 [ath.oc] 3 May 07 Abstract

More information

MSEC MODELING OF DEGRADATION PROCESSES TO OBTAIN AN OPTIMAL SOLUTION FOR MAINTENANCE AND PERFORMANCE

MSEC MODELING OF DEGRADATION PROCESSES TO OBTAIN AN OPTIMAL SOLUTION FOR MAINTENANCE AND PERFORMANCE Proceeding of the ASME 9 International Manufacturing Science and Engineering Conference MSEC9 October 4-7, 9, West Lafayette, Indiana, USA MSEC9-8466 MODELING OF DEGRADATION PROCESSES TO OBTAIN AN OPTIMAL

More information

Interactive Markov Models of Evolutionary Algorithms

Interactive Markov Models of Evolutionary Algorithms Cleveland State University EngagedScholarship@CSU Electrical Engineering & Coputer Science Faculty Publications Electrical Engineering & Coputer Science Departent 2015 Interactive Markov Models of Evolutionary

More information

arxiv: v1 [cs.ds] 17 Mar 2016

arxiv: v1 [cs.ds] 17 Mar 2016 Tight Bounds for Single-Pass Streaing Coplexity of the Set Cover Proble Sepehr Assadi Sanjeev Khanna Yang Li Abstract arxiv:1603.05715v1 [cs.ds] 17 Mar 2016 We resolve the space coplexity of single-pass

More information

Hybrid System Identification: An SDP Approach

Hybrid System Identification: An SDP Approach 49th IEEE Conference on Decision and Control Deceber 15-17, 2010 Hilton Atlanta Hotel, Atlanta, GA, USA Hybrid Syste Identification: An SDP Approach C Feng, C M Lagoa, N Ozay and M Sznaier Abstract The

More information

The linear sampling method and the MUSIC algorithm

The linear sampling method and the MUSIC algorithm INSTITUTE OF PHYSICS PUBLISHING INVERSE PROBLEMS Inverse Probles 17 (2001) 591 595 www.iop.org/journals/ip PII: S0266-5611(01)16989-3 The linear sapling ethod and the MUSIC algorith Margaret Cheney Departent

More information

arxiv: v1 [cs.ds] 29 Jan 2012

arxiv: v1 [cs.ds] 29 Jan 2012 A parallel approxiation algorith for ixed packing covering seidefinite progras arxiv:1201.6090v1 [cs.ds] 29 Jan 2012 Rahul Jain National U. Singapore January 28, 2012 Abstract Penghui Yao National U. Singapore

More information

Iterative Decoding of LDPC Codes over the q-ary Partial Erasure Channel

Iterative Decoding of LDPC Codes over the q-ary Partial Erasure Channel 1 Iterative Decoding of LDPC Codes over the q-ary Partial Erasure Channel Rai Cohen, Graduate Student eber, IEEE, and Yuval Cassuto, Senior eber, IEEE arxiv:1510.05311v2 [cs.it] 24 ay 2016 Abstract In

More information

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and

This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and This article appeared in a ournal published by Elsevier. The attached copy is furnished to the author for internal non-coercial research and education use, including for instruction at the authors institution

More information

In this chapter, we consider several graph-theoretic and probabilistic models

In this chapter, we consider several graph-theoretic and probabilistic models THREE ONE GRAPH-THEORETIC AND STATISTICAL MODELS 3.1 INTRODUCTION In this chapter, we consider several graph-theoretic and probabilistic odels for a social network, which we do under different assuptions

More information

Kernel-Based Nonparametric Anomaly Detection

Kernel-Based Nonparametric Anomaly Detection Kernel-Based Nonparaetric Anoaly Detection Shaofeng Zou Dept of EECS Syracuse University Eail: szou@syr.edu Yingbin Liang Dept of EECS Syracuse University Eail: yliang6@syr.edu H. Vincent Poor Dept of

More information

IN modern society that various systems have become more

IN modern society that various systems have become more Developent of Reliability Function in -Coponent Standby Redundant Syste with Priority Based on Maxiu Entropy Principle Ryosuke Hirata, Ikuo Arizono, Ryosuke Toohiro, Satoshi Oigawa, and Yasuhiko Takeoto

More information

Bipartite subgraphs and the smallest eigenvalue

Bipartite subgraphs and the smallest eigenvalue Bipartite subgraphs and the sallest eigenvalue Noga Alon Benny Sudaov Abstract Two results dealing with the relation between the sallest eigenvalue of a graph and its bipartite subgraphs are obtained.

More information