Haykin_app_F_pp3.

fm Page 31 Tuesday, January 8, 2013 1:18 PM

F.1 Block Interleaving A31

Data
read out
rows

Data
read in
columns

...

...
(a) (b)

Figure F.1 Block interleaver structure. (a) Data “read in.” (b) Data “read out.”

EXAMPLE 1 Interleaving
Figure F.2a depicts an original sequence of encoded words, with each word consisting of
five symbols. Figure F.2b depicts the interleaved version of the encoded sequence, with
the symbols shown in reordered positions. An error burst occupying five symbols, caused
by channel impairment, is also shown alongside Figure F.2b. Note that the manner in
which the encoded symbols are reordered by the interleaver is the same from one word to
the next.

Original A B C D E
sequence
of
encoded A1 A2 A3 A4 A5 B1 B2 B3 B4 B5 C1 C2 C3 C4 C5 D1 D2 D3 D4 D5 E1 E2 E3 E4 E5
words
(a)

1 2 3 4 5
Interleaved
sequence
of A1 C1 D1 B1 E1 A2 C2 D2 B2 E2 A3 C3 D3 B3 E3 A4 C4 D4 B4 E4 A5 C5 D5 B5 E5
words
X X X X
(b)

A B C D E
De-interleaved
sequence
of A1 A2 A3 A4 A5 B1 B2 B3 B4 B5 C1 C2 C3 C4 C5 D1 D2 D3 D4 D5 E1 E2 E3 E4 E5
words
X X X X
(c)

Figure F.2 Interleaving example. (a) Original sequence. (b) Interleaved sequence. (c) De-interleaved
sequence.
Haykin_app_F_pp2.fm Page 32 Friday, December 7, 2012 10:21 AM

A32 Appendix F Interleaving

On de-interleaving in the receiver, the scrambling of symbols is undone, yielding a

sequence that resembles the original sequence of encoded symbols, as shown in Figure
F.2c. This figure also includes the new positions of the transmission errors. The important
point to note here is that the error burst is dispersed as a result of de-interleaving.
This example teaches us the following:
1. The burst of transmission errors is only acted upon by the de-interleaver.
2. Insofar as the encoded symbols that are received are error free, the de-interleaver
cancels the scrambling action of the interleaver.

F.2 Convolutional Interleaving

The block diagram of a convolutional interleaver/de-interleaver is shown in Figure F.3.

Defining the period
T = LN
the interleaver is referred to as an (L  N) convolutional interleaver, which has properties
similar to those of the (L  N) block interleaver.
The sequence of encoded bits to be interleaved in the transmitter is arranged in blocks
of L bits. For each block, the encoded bits are sequentially shifted into and out of a bank of
N registers by means of two synchronized input and output commutators. The interleaver,
depicted in Figure F.3a, is structured as follows:
1. The zeroth shift register provides no storage; that is, the incoming encoded symbol
is transmitted immediately.
2. Each successive shift register provides a storage capacity of L symbols more than
the preceding shift register.
3. Each shift register is visited regularly on a periodic basis.
With each new encoded symbol, the commutators switch to a new shift register. The new
symbol is shifted into the register and the oldest symbol stored in that register is shifted
out. After finishing with the (N – 1)th shift register (i.e., the last register), the commutators
return to the zeroth shift register. Thus, the switching/shifting procedure is repeated
periodically on a regular basis.
The de-interleaver in the receiver also uses N shift registers and a pair of input/output
commutators that are synchronized with those in the interleaver. Note, however, the shift
registers are stacked in the reverse order to those in the interleaver, as shown in Figure
F.3b. The net result is that the de-interleaver in the receiver performs the inverse operation
to interleaving in the transmitter, and so it should.
An advantage of convolutional over block interleaving is that in convolutional
interleaving the total end-to-end delay is L(N – 1) symbols and the memory requirement is
L(N – 1)2 in both the interleaver and de-interleaver, which are one-half of the
corresponding values in a block interleaver/de-interleaver for a similar level of
interleaving.
The description of the convolutional interleaver/de-interleaver in Figure F.3b is
presented in terms of shift registers. The actual implementation of the system can also be
Haykin_app_F_pp2.fm Page 33 Friday, December 7, 2012 10:21 AM

F.3 Random Interleaving A33

Sequence of bits
from
channel encoder
L

Multiplexer 2L To modulator

Synchronizing
sequence

(N – 1)L
Bank of
shift registers

(a)

(N – 1)L

To channel
decoder
(N – 2)L
From baseband
processor involving
(N – 3)L Demultiplexer
channel estimation
and equalization

Synchronizing
sequence

(b)

Figure F.3 (a) Convolutional interleaver. (b) Convolutional de-interleaver.

accomplished with a random access memory unit in place of shift registers. This
alternative implementation simply requires that access to the memory units be
appropriately controlled.

F.3 Random Interleaving

In a random interleaver, a block of N input bits is written into the interleaver in the order
in which they are received, but they are read out in a random manner. Typically, the
permutation of the input bits is defined by a uniform distribution. Let  (i) denote the
permuter location of the ith input bit, where i = 1, 2, , N. The set of integers denoted by
   i  N
i = 1, defining the order in which the stored input bits are read out of the
interleaver, is generated according to the following two-step algorithm:
1. Choose an integer i1 from the uniformly distributed set 𝒜 ={1, 2, , N}, with the
probability of choosing i1 being p(i1) = 1N. The chosen integer i1 is set to be  (i).
Haykin_app_F_pp2.fm Page 34 Friday, December 7, 2012 10:21 AM

A34 Appendix F Interleaving

2. For k  1, choose an integer ik from the uniformly distributed set

𝒜k =  i  𝒜 i  i 1 i 2  i k – 1 
with the probability of choosing ik being p(ik) = 1(N – k + 1). The chosen integer ik
is set to be  (k). Note that the size of the set 𝒜k is progressively reduced for k  1.
When k = N, we are left with a single integer iN, in which case iN is set to be  (N).
To be of practical use in communications, random interleavers are configured to be
pseudo-random, meaning that within a block of N input bits the permutation is random as
described above, but the permutation order is exactly the same from one block to the next.
Accordingly, pseudo-random interleavers are designed off-line; they are of particular
interest in the construction of turbo codes, discussed in Chapter 10.

Notes
1. Interleaving of both the block and convolutional types is discussed in some detail in Clark and
Cain (1981) and in lesser detail in Sklar (2001). For a treatment of interleaving viewed from the
perspective of turbo codes, see the book (Vucetic and Yuan, 2000).
Haykin_app_G_pp2.fm Page 35 Friday, December 7, 2012 10:22 AM

APPENDIX
G The Peak-Power Reduction Problem
in OFDM
In Section 9.11 we discussed the multicarrier transmission technique, namely orthogonal
frequency-division multiplexing (OFDM), which is of particular importance to wireless
communications due to the computational benefits offered by the fast Fourier transform
(FFT) algorithm. However, envelope variations are a frequently cited drawback of OFDM
because of the peak-power limited problem. This problem arises due to the statistical
possibility of a large number of independent subchannels in the OFDM becoming
constructively superimposed, thereby resulting in high peaks. In the literature, the
practical issue of envelope variations is described in terms of the peak-to-average power
ratio, commonly abbreviated as PAPR.1
In this section, we discuss the PAPR problem in wireless communications and how it
can be reduced.

G.1 PAPR Properties of OFDM Signals

Consider a single modulation interval, that is, a single symbol of OFDM, the duration of
which is denoted by Ts. In its most basic form, the transmitted OFDM signal is described by
N–1
st =  sn exp  j2nft  0  t  Ts (G.1)
n=0

where the term f denotes the frequency separation between any two adjacent
subchannels in the OFDM. By definition, the frequency separation f and symbol
duration Ts are related by the time–bandwidth product:
T s f = 1 (G.2)

This condition is required to satisfy the orthogonality requirement among the N

subchannels of the OFDM.
Typically, the coefficients in OFDM, denoted by sn in (G.1) are taken from a fixed
modulation constellation, exemplified by M-ary phase-shift-keying (PSK) or M-ary
quadrature amplitude modulation (QAM) techniques, which were discussed in Chapter 7.
With s(t), in its baseband form, being a complex-valued signal with an amplitude and
phase that characterize it, we may express the time-averaged power of an individual
symbol of the OFDM signal in (G.1) as follows:
T
1 s

2
P = ----- s  t  dt (G.3)
Ts 0

N–1

2
= sn
n=0

A35
Haykin_app_G_pp3.fm Page 36 Tuesday, January 8, 2013 1:23 PM

A36 Appendix G The Peak-Power Reduction Problem in OFDM

where the summation in the second line of the equation follows for Parseval’s theorem,
discussed in Chapter 2. With the OFDM coefficient sn being a random variable, which it is
in a wireless environment, it follows that the time-averaged power P is itself a random
variable. It follows therefore that the ensemble-averaged power of the OFDM signal is
given by the expectation
P av = ⺕  P 
(G.4)
2
= ⺕  s  t   for 0  t  T s
In an OFDM signal based on M-ary PSK, for example, we have s n = 1 for all n. In this
special case, (G.4) yields
P av = N (G.5)

As pointed out previously, the metric of interest commonly used in the literature for
assessing the issue of statistical peak-power variations in the use of OFDM for wireless
communications is the peak-to-average power ratio (PAPR), for which we offer the
following definition:
2
max s  t 
0tT
 = ------------------------------
s
-
P av
(G.6)
2
max s  t 
0tT
s
= ------------------------------
-
2
⺕ st
where, in words, the term in the numerator denotes the maximum value of the
instantaneous power (i.e., peak power) of the OFDM signal measured across the symbol
interval, 0  t  T s, and the denominator denotes average power, hence PAPR. The
formula used in (G.6) refers to the baseband formulation of the PAPR problem.2
Recognizing that PAPR is, in reality, a random variable distributed across each OFDM
symbol, a statistical interpretation of it is useful. To this end, we may express the
probability of the event that an OFDM symbol, denoted by s(t) as defined in (G.1),
exceeds the peak value  p with probability P c as follows:
⺠    p  = Pc (G.7)

To expand on this definition, we say that the PAPR is less than some prescribed value  p
for 100  1 – P c  of the OFDM symbols, in which case we may refer to 100  1 – P c  as a
percentile PAPR.

G.2 Maximum PAPR in OFDM Using M-ary PSK

Consider an OFDM system based on M-ary PSK for its modulation scheme. For this
special application of OFDM, the PAPR is always less than or equal to N, where N is the
number of subchannels. To justify this statement, we first note that for M-ary PSK,
s n = 1 for 1  n  N