2.8.1 Derivations for Section 2.3.3Derivation of Equation (2.61)Recall that the RLS algorithm for updating the blind linear MMSE algorithm is as follows : Equation 2.247
Equation 2.248
Equation 2.249
Equation 2.250
Equation 2.251
We first derive an explicit recursive relationship between m 1 [ i ] and m 1 [ i -1]. Define Equation 2.252
Premultiplying both sides of (2.249) by , we get Equation 2.253
From (2.253), we obtain Equation 2.254
where Equation 2.255
Substituting (2.249) and (2.254) into (2.250), we get Equation 2.256
where Equation 2.257
is the a priori least-squares estimate at time i . It is shown below that Equation 2.258
Equation 2.259
Substituting (2.248) and (2.258) into (2.256), we get Equation 2.260
Therefore, by (2.260) we have Equation 2.261
where v [ i ] is defined in (2.56). Therefore, from (2.261) we get Equation 2.262
Finally, we derive (2.258) and (2.259). Postmultipling both sides of (2.251) by r [ i ], we get Equation 2.263
On the other hand, (2.247) can be rewritten as Equation 2.264
Equation (2.258) is obtained by comparing (2.263) and (2.264). Multiplying both sides of (2.254) by k [ i ], we get Equation 2.265
Equation (2.255) can be rewritten as Equation 2.266
Equation (2.259) is obtained comparing (2.265) and (2.266). Derivation of Equations (2.62) “(2.69)Suppose that an application of the rotation matrix Q [ i ] yields the following form: Equation 2.267
Then because of the orthogonality property of Q [ i ] (i.e., ), taking the outer products of each side of (2.267) with their respective Hermitians, we get the following identities: Equation 2.268
Equation 2.269
Equation 2.270
Associating A 1 with the first N columns of the partitioned matrix on the left-hand side of (2.62), and B 1 with the first N columns of the partitioned matrix on the right-hand side of (2.62), then (2.268), (2.269), and (2.270) yield Equation 2.271
Equation 2.272
Equation 2.273
Equation 2.274
Equation 2.275
Equation 2.276
A comparison of (2.271) “(2.273) with (2.54) “(2.56) shows that C [ i ], u [ i ], and v [ i ] in (2.62) are the correct updated quantities at time n . Moreover, (2.67) follows from (2.274) and (2.57), (2.68) follows from (2.275) and (2.59), and (2.69) follows from (2.276) and (2.262). 2.8.2 Proofs for Section 2.4.4Proof of Lemma 2.3Denote
Note that the eigendecomposition of H is given by Equation 2.277
Then the Moore “Penrose generalized inverse [189] of matrix H is given by Equation 2.278
On the other hand, the Moore “Penrose generalized inverse H of a matrix H is the unique matrix that satisfies [189] (a) HH and H H are symmetric; (b) HH H = H ; and (c) H HH = H . Next we show that G = H by verifying these three conditions. We first verify condition (a). Using (2.106), we have Equation 2.279
where the second equality follows from the facts that W T W = I N and S T S T = V T V = VV T = I K . Since the N x N diagonal matrix S S = diag ( I K , ), it follows from (2.279) that HG is symmetric. Similarly, GH is also symmetric. Next we verify condition (b). Equation 2.280
where in the second equality, the following facts are used: W T W = I N , S T S T = I K , and V T V = VV T = I K ; the third equality follows from the fact that S S S = S . Condition (c) can be similarly verified (i.e., GHC = G ). Therefore, we have Equation 2.281
Now (2.107) follows immediately from (2.281) and the fact that U T U = UU T = I N . 2.8.3 Proofs for Section 2.5.2Some Useful LemmasWe first list some lemmas that will be used in proving the results in Section 2.5.2. A random matrix is said to be Gaussian distributed if the joint distribution of all its elements is Gaussian. First we have the following vector form of the central limit theorem. Lemma 2.4: (Theorem 1.9.1B in [443]) Let { x i } be i.i.d. random vectors with mean m and covariance matrix S . Then
Next we establish that the sample autocorrelation matrix given by (2.122) is asymptotically Gaussian distributed as the sample size M . Lemma 2.5 Denote Equation 2.282
Equation 2.283
Equation 2.284
Then converges in probability toward a Gaussian matrix with mean and an N 2 x N 2 covariance matrix whose elements are specified by Equation 2.285
Proof: Since given by (2.285) has , and it is a sum of i.i.d. terms ( r [ i ] r [ i ] T ), by Lemma 2.4, it is asymptotically Gaussian, with an N 2 x N 2 covariance matrix whose elements are given by the covariance of the zero-mean random matrix ( r [ i ] r [ i ] T ). To calculate this covariance, note that (for notational convenience, in what follows we drop the time index i ) Equation 2.286
We have Equation 2.287
where the last equality follows from the fact that Equation 2.288
Note that the last term of (2.285) is due to the nonnormality of the received signal r [ i ]. If the signal had been Gaussian, the result would have been the first two terms of (2.285) only (compare this result with Theorem 3.4.4 in [18]). Using a different modulation scheme (other than BPSK) will result in a different form for the last term in (2.285). In what follows we make frequent use of the differential of a matrix function (cf. [421], Chap. 14). Consider a function . Recall that the differential of f at a point x is a linear function such that Equation 2.289
If the differential exists, it is given by L f ( x ; x ) = T ( x ) x , where . Let y = f ( x ) and consider its differential at x . Denote . Hence for fixed x , D y is a function of D x ; and for fixed x , if x is random, so is D y . We have the following lemma regarding the asymptotic distribution of a function of a sequence of asymptotically Gaussian vectors. Lemma 2.6: (Theorem 3.3A in [443]) Suppose that is asymptotically Gaussian; that is ,
Let be a function . Denote y ( M )= f [ x ( M )]. Suppose that f has a nonzero differential L f ( x ; x ) = T ( x ) x at x . Denote and D y ( M ) = T ( x ) D x ( M ). Then Equation 2.290
where Equation 2.291
Equation 2.292
To calculate C y we can use either (2.291) or (2.292). When dealing with functions of matrices, however, it is usually easier to use (2.292). In what follows we make use of the following identities of matrix differentials: Equation 2.293
Equation 2.294
Equation 2.295
Finally, we have the following lemma regarding the differentials of the eigencomponents of a symmetric matrix. It is a generalization of Theorem 13.5.1 in [18]. Its proof can be found in [197]. Lemma 2.7: Let the N x N symmetric matrix C have an eigendecomposition , where the eigenvalues satisfy . Let D C be a symmetric variation of C and denote . Let T be a unitary transformation of C as Equation 2.296
Denote the eigendecomposition of T as Equation 2.297
(Note that if C = C , then W = I N and L = L .) The differential of L at L , and the differential of W at I N , as a function of , are given, respectively, by Equation 2.298
Equation 2.299
Proof of Theorem 2.1DMI Blind Detector Consider the function . The differential of at C r is given by Equation 2.300
where . Then according to Lemma 2.6, is asymptotically Gaussian as M , with zero mean and covariance matrix given by (2.292) [4]
Equation 2.301
Now, by Lemma 2.5, we have Equation 2.302
Writing (2.302) in a matrix form, we have Equation 2.303
with
The eigendecomposition of C r is Equation 2.304
Substituting (2.303) and (2.304) into (2.301), we get
where the last equality follows from the fact that . Subspace Blind Detector We will prove the following more general proposition, which will be used in later proofs. The part of Theorem 2.1 for the subspace blind detector follows with v = s 1 . Proposition 2.6: Let be the weight vector of a detector , and let be the weight vector of the corresponding estimated detector. Then
with Equation 2.305
where Equation 2.306
Equation 2.307
Proof: Consider the function . By Lemma 2.6, is asymptotically Gaussian as M , with zero mean and covariance matrix given by where D w 1 is the differential of at ( U s , L s ). Denote U = [ U s U n ]. Define Equation 2.308
Since T is a unitary transformation of , its eigenvalues are the same as those of . Hence its eigendecomposition can be written as Equation 2.309
where are eigenvectors of T . From (2.308) and (2.309), we have Equation 2.310
Thus we have Equation 2.311
The differential in (2.311) at ( I N , L ) is given by Equation 2.312
where E s is composed of the first K columns of I N . Using Lemma 2.7, after some manipulations, we have Equation 2.313
with Equation 2.314
where we have used the fact that D T is symmetric (i.e., [ D T ] i,j = [ D T ] j,i ). Denote Equation 2.315
Then C y = U T C r U = L . Moreover, we have D T = D C y . Since E { D T } = , by Lemma 2.5 for 1 i,j N , Equation 2.316
Using (2.313) and (2.316), we have Equation 2.317
where (2.317) follows from the fact that Equation 2.318
since it is assumed that a similar relationship holds for U T s a . Writing (2.317) in matrix form, we obtain Equation 2.319
where Equation 2.320
Equation 2.321
Equation 2.322
Finally, by (2.311), . Substituting (2.319) into this expansion, we obtain (2.305). Proof of Corollary 2.1First we compute the term given by (2.120). Using (2.304) and (2.128) and the fact that , we have Equation 2.323
with Equation 2.324
Equation 2.325
Equation 2.326
Equation 2.327
Hence we have Equation 2.328
Next note that the linear MMSE detector can also be written in terms of R as [520] Equation 2.329
Therefore, we have Equation 2.330
Equation 2.331
By (2.130), for the DMI blind detector, we have , and for the subspace blind detector, Equation 2.332
where we have used the fact that the decorrelating detector can be written as [549] Equation 2.333
Finally, substituting (2.328) “(2.332) into (2.119), we obtain (2.132). SINR for Equicorrelated SignalsIn this case, R is given by Equation 2.334
where 1 is an all-1 K -vector. It is straightforward to verify the following eigenstructure of R : Equation 2.335
with Equation 2.336
Equation 2.337
Since A 2 = A 2 I K , we have Equation 2.338
Similarly, we obtain Equation 2.339
Equation 2.340
Substituting (2.338) “(2.340) into (2.132) “(2.135), and defining Equation 2.341
Equation 2.342
Equation 2.343
Equation 2.344
we obtain expression (2.143) for the average output SINRs of the DMI blind detector and the subspace blind detector. |