CHANNEL CAPACITY WITH PERFECT CSIR AND PERFECT CSIT

QR scanner with .netUsing Barcode Control SDK for Visual Studio .NET Control to generate, create, read, scan barcode image in Visual Studio .NET applications.

Power

QR Code 2d Barcode barcode library for .netgenerate, create qr code 2d barcode none with .net projects

p4 = 0

.net Vs 2010 qr code recognizerfor .netUsing Barcode recognizer for visual .net Control to read, scan read, scan image in visual .net applications.

(determines average power constraint)

Barcode barcode library in .netusing .net todisplay barcode with asp.net web,windows application

2 s z2 / l1

.NET Crystal bar code draweron .netusing barcode integration for .net crystal control to generate, create bar code image in .net crystal applications.

Spatial channels

Control qr bidimensional barcode size for .net c#to embed qr code jis x 0510 and qr codes data, size, image with c#.net barcode sdk

2 s z / l5

Web Form qr-code integrationfor .netusing barcode implementation for asp.net web forms control to generate, create qr code iso/iec18004 image in asp.net web forms applications.

(a) w1 Y1

Encoder

VS .NET Crystal ucc.ean - 128 generationon .netusing visual studio .net crystal toget ean 128 in asp.net web,windows application

Encoder . . .

.net Vs 2010 barcode writerfor .netgenerate, create barcode none in .net projects

Channel matrix S

VS .NET gs1 datamatrix barcode encoderin .netgenerate, create data matrix 2d barcode none with .net projects

Receive Matrix A* . . . .

Decoder . . . . Decoder

Control qr code iso/iec18004 data on visual c#.netto get qr-code and qr barcode data, size, image with visual c#.net barcode sdk

w1

Generate barcode for objective-cusing ipad tomake barcode with asp.net web,windows application

Elgenbeamforming m = min {nT, nR}

2D Barcode development with visual basic.netgenerate, create 2d barcode none in visual basic projects

. . . .

Control data matrix 2d barcode image with vb.netusing barcode creation for .net framework control to generate, create data matrix barcode image in .net framework applications.

. . . .

Produce 1d barcode on .netuse asp.net web linear printer toinclude 1d barcode for .net

wm

DataMatrix implement for exceluse office excel gs1 datamatrix barcode creation torender gs1 datamatrix barcode with office excel

Encoder

X nT

.net For Windows Forms qrcode generatorwith .netusing barcode implement for .net windows forms control to generate, create qr-codes image in .net windows forms applications.

Full feedback of channel matrix H

(b) Figure 2.18. Ergodic capacity of MIMO channels with ideal CSIT and CSIR: (a) illustration of spatial water- lling of transmit power (the allocated power on each spatial channel is proportional to the shaded area); (b) optimal transmitter receiver structure.

MIMO LINK WITH PERFECT CHANNEL STATE INFORMATION

the channel encoders Y = [Y1, . . . , Ym*] are passed to the power adaptation matrix Q to adjust the power allocated to individual spatial channels and then to the eigenbeamforming matrix B to decouple the MIMO channel into m* independent vector channels. At the receiver, the received symbols (dropping the symbol index n) can be expressed as Y = HX + Z

= A DB * B Q Y + Z = A DQ Y + Z where B, A are unitary matrices (eigenvector matrix of hh* and h*h, respectively) and D is a diagonal matrix (eigenvalues of hh*) Since A is unitary, there is no loss of information by premultiplying both sides by A*. Hence, the processed received signal R is given by R = A * Y = DQ Y + Z (2.65)

where Z = A*Z is the equivalent-channel noise, which has the same distribution and covariance matrix as Z. From Equation (2.65), the equivalent channel with input Y and output R can be decoupled into m* = rank(h) spatial channels (or eigenchannels). Hence, the optimal receiver structure (as illustrated in Figure 2.18b) consists of a beamforming matrix A* followed by m* isolated channel decoders. 2.7.2 Slow Flat Fading MIMO Channels

Consider a slow at fading MIMO channel with channel input Xn. The channel output Yn is given by Yn = HX n + Zn where the channel fading H remains quasistatic for the entire encoding frame. For a given channel fading realization H = h, the instantaneous channel capacity is given by Ccsit,csir (h) = max I (T; Y, H = h) = max I (X; Y H = h)

p ( T ( h )) p(X h)

(2.66)

where the second equality is due to the fact that given h, the distribution on T induces a distribution on X. In the Rayleigh fading example we considered, the capacity achieving input distribution is circular symmetric complex Gaussian with input covariance matrix e[XX*|h] = Q(h). Note that since we have slow fading, the entire frame spans across a quasistatic fading realization. Together with the average power

SUMMARY

constraint (averaged over an encoding frame), there is no temporal power adaptation across an encoding frame. Hence, the instantaneous channel capacity is given by Ccsit,csir (h) = max log 2 I n R +

Q :tr (Q ) P0

hQh * 2 sz

(2.67)

Using an approach similar to that in the fast fading example, the capacityachieving input covariance matrix Q(h) is given by the spatial power water lling Q = BQB * (2.68)

where B is the nT m* eigenvector matrix of h*h and Q is a m* m* diagonal matrix with m* = min[nT, nR] diagonal elements given by qn ,n

2 sz 1 = m n2 l n +

"n [1, m *]

(2.69)

where m is a constant chosen to satisfy Snqn,n P0. Since the slow fading channel can be regarded as a compound channel [80], we have to make use of CSIT to perform both rate adaptation as well power adaptation.8 The power adaptation is done according to Equation (2.69), which consists of spatial and temporal power water lling. The rate adaptation is done in such a way that the transmission rate R(h) is always smaller than the instantaneous channel capacity Ccsit,csir(h) by e for arbitrarily small e. In this way, all the frame transmission is guaranteed to have zero outage probability. In other words, the presence of perfect CSIT in slow fading channels can guarantee nonzero ergodic capacity. The average capacity over multiple frame transmissions is given by hQh * Ccsit,csir = e max log 2 I nR + 2 Q :tr (Q ) P0 sz where the expectation is taken over h.