Digital Signal Processing Reference
In-Depth Information
s ( n )
m, 0
s ( n )
m
F ( z )
m ,0
C ( z )
m
K
P
z
s ( n )
m, 1
channel
F ( z )
m ,1
K
P
z
s ( n )
m,K
z
1
F ( z )
m , K −1
K
P
new transmitter filters
blocking into K -vector
s ( n )
m
s ( n )
m, 0
H ( z )
m ,0
K
+
P
s ( n )
m, 1
z −1
H ( z )
m ,1
K
P
from other
users
z −1
s ( n )
z −1
m,K −1
H ( z )
m , K −1
K
P
new receiver filters
unblocking
Figure P7.22 .
7.23. Null resistance. The cyclic-prefix system described in Sec. 7.3.2 uses the
equalizer multipliers 1 /C [ k ] , 0
1 , at the receiver. Here C [ k ]
are the M DFT coe cients of the FIR channel (assumed to be of order
L ). In general, some of these can be very small, or even equal to zero.
This creates practical di culties in the implementation, as the noise can
be severely amplified by 1 /C [ k ] . There are two practical ways to overcome
this problem. One is to replace the zero-forcing equalizers 1 /C [ k ]with
MMSE equalizers. The other is to use extra redundancy at the transmitter
to incorporate what is called null resistance. In this problem we describe
the second approach [Liang and Tran, 2002]. Consider the modified cyclic-
prefix system shown in Fig. P7.23. Here the blocked version s ( n )ofsize
M is transformed into a vector r ( n )ofsize N = M + L using a matrix Q .
The vector r ( n ) is treated as if this is to be transmitted. Thus, we perform
N -point IDFT, and insert cyclic prefix, and so forth, as shown.
k
M
1. Express the N
N transfer matrix from s 1 ( n )to y 1 ( n )intermsof
the channel coe cients c ( n ) .
2. Show that the transfer function from r ( n )to v ( n ) is a diagonal matrix
with elements C ( e j 2 πk/N ) , 0
×
k
N
1 . These are the N -point DFT
coe cients of the channel.
Search WWH ::




Custom Search