Beruflich Dokumente
Kultur Dokumente
ABSTRACT
I I
punctured convolutional codes, long constraint length convolu-
tional codes and punctured Turbo codes. The ARQ schemes Data
+
considered are ARQ with majority voting, ARQ with metric
combining and ARQ with code combining (or Type I11 ARQ). Figure 1: Slot structure of IS-136
The performance of these techniques when used with one-slot
and two-slot interleaving is studied. Results show that highest Encixler Puncturer Interleaver
throughput is achieved when a rate-516 convolutional code is
used with one-slot interleaving and Type I11 ARQ scheme.
Channel
0-7803-3659-3/97
$ 1 0.000 1 997 IEEE 1029
length and decoding becomes impractical beyond a certain con-
straint length. However sequential decoding algorithms can be
used to decode long constraint length convolutional codes. Al-
though the performance of sequential decoding algorithms is
suboptimal, the decoding complexity is linear with the con-
symbols
straint length and, hence, is suitable for long constraint length
codes. The performance of a v = 31 rate-516 convolutional
code is studied when used with a soft-decision Fano sequential
decoder.
1030
4.2 ARQ w i t h P a c k e t Combining rate-1/2 code. With this technique after 2 transmissions, the
FER performance of the system is guaranteed to be at least as
When a decoded packet is deemed erroneous at the receiver, good as the original rate-'1/2 code.
all the branch metrics from the Viterbi decoder are stored and The multiple transmissions can be effectively combined by
a retransmission is requested. The transmitter then transmits using a technique similar 160 that used for the ARQ with packet
an identical copy of its previous transmission. The receiver now combining. Consider combining the first and second transmis-
combines these transmissions using a packet combining rule to sions, or equivalently, determining the branch metrics after the
decode the input bit sequence dk given multiple observations second transmission of the packet. Recall, that when decod-
r p ) ,.., r f ' of the same encoded output sequence X k , where r f ) ing punctured convolutional codes, the branch metrics corre-
is the received signal during the jth transmission and k is the sponding to the punctured symbols are not calculated. When
time index. To understand this better, consider the branch computing the branch metrics during the second transmissions,
metric corresponding to a transition from state SL.-l to state y(')(S', S,k ) can then be computed using
S k . During the first transmission, the Viterbi decoder operates
with the branch metric y(l)(S', S, k) computed as
A = [
1 1 1 1 1 1
, where a 1 represents a trans-
highly correlated. The iriterleaver permutes the code symbol
sequence, making the fades appear uncorrelated. The perfor-
mance of a system largely depends on the interleaving scheme.
mitted bit and 0 represents a punctured bit. The punctured
In general, the larger the interleaver size or the time separa-
output is then modulated and transmitted over the channel.
tion between the fades affecting the bits in the data stream, the
When a packet is received and deemed erroneous, the receiver
better the performance of the system will be.
stores all the branch metrics and requests a retransmission. The
transmitter however, does not transmit an identical copy of
the packet. The transmitter now transmits the bits that were 5.1 One-slot interleaving
punctured during the first transmission. In otherwords, the en-
coded outputs are now punctured using a puncturing pattern Recall that data is transmitted in the form of packets in
. The two transmissions together time slots specified by the IS-136 frame structure. Specifically
1 0 0 0 0 each user transmits over ab 6.67 ms time slot and then waits for
now correspond to a rate-5/12 code, which embeds in itself the 20 ms before the next slot. With one-slot interleaving, data
original rate-1/2 mother code. Recall that in the previous cases, within one slot (260 bits) is interleaved using a 26 x 10 block
the effective rate of the code after two transmissions was 5/12 interleaver. The data is written row-wise and read column wise.
also. The important difference is that, in the previous cases If X O ,X I ,....,2 2 5 9 is the iriput to the interleaver, the output is
after 2 transmissions the effective code is the concatenation of 2 0 ~ x 2 6 ...,
, ~ 2 5 9 .This ensures that the adjacent code symbols are
a rate-5/6 code with a 2 symbol repetition code. The distance affected by fades which are separated by at least 26 symbols.
properties of the concatenated rate-5/12 repetition code are in- However, interleaving is restricted to within one time slot and,
ferior to that of the rate-5/12 code which contains the original hence, the performance cciuld be adversely affected.
1031
5.2 Two-slot interleaving comparison of the achievable throughput in kbps when rate-
1/2, 5/6, and 10/11 codes of constraint length 5 are used with
The major problem with one-slot interleaving is that inter- an AWGN-limited channel at MS velocities of 8 km/hr and 100
leaving is restricted to one time frame and, hence, the fades km/hr and SNR of 14dB. It can be seen that the performance
cannot be effectively uncorrelated at low speeds. With two- of the rate 5/6 code is good both at low and high MS velocities.
slot interleaving, two adjacent packets are combined and inter- We now discuss the performance of ARQ schemes and in-
terleaving schemes, when used with a rate-5/6 code. A plot of
leaved. The receiver then deinterleaves two packets to ether
(8
and decodes them separately. Specifically, if x r ) , ....,x Z s 9 and FER vs SNR for mobile velocities of 8 km/hr (slow speed) and
x('+l) , ...., x (259
k + ' ) represent the encoded outputs corresponding
100 km/hr (high speed) is shown for the CCI limited channel in
to packets at time IC and k+l respectively, then two new packets Figs. 4 and 5 respectively. Note that the performance of Type
are formed using the rule, I11 ARQ is uniformly better than the other two schemes and
is significantly better than the ARQ scheme with majority vot-
ing, especially at low C/I. However, the storage requirement is
greater for Type I11 ARQ. Table 2, provides a sample compari-
son of FER performance, complexity and storage requirements
*.z(le+l) -
- { xi'),
x!'"),
if i mod 2 = 1
if i mod 2 = 0.
for the three ARQ schemes when used with one-slot interleaving
for C/I of 14 dB.
1032
moderate C/I. This is due to the fact that the error-correcting
capability (free distance) of the code is better and, hence, after
distributing the errors, the errors in both the packets can be
corrected. In conclusion, the performance curves for one-slot
and two-slot interleaving exhibit a crossover C/I that depends
not only on the speed but also on the error-correction capabil-
ity of the code used. In particular, the crossover C/I decreases
when the error-correcting capability (free distance) of the code
increases or MS velocity increases. In this case, one-slot inter-
leaving is better than two slot interleaving up to 20 dB C/I at
40.0
speed of 8 km/hr and up to 14 dB at speed of 100 km/hr.
Uncoded
7. Conclusions -fl 1 slot interleaving with majority voting
+e2 slot interleaving with majority voting
30.0 A 1 slot interleaving with packet combining
We have studied the performance of several FECI interleav- 4 2 slot interleaving with packet combining
ing and ARQ schemes for transmitting packet data using IS-136 t 1 slot interleaving with Type 111 ARQ
& + 2 slot interleaving with Type Ill ARQ
TDMA. Results show that short constraint length, punctured I
convolutional codes used with soft-decision Viterbi decoding 2
20,0
provide good performance at moderate complexity. The per- 6
formance of these codes can be improved by using a Type I11 E
-
ARQ scheme. Further, it was shown that one-slot interleaving
is better than two-slot interleaving for low-to-moderate C/I, es-
10.0
pecially at low MS velocities.
References
1
Data Services Task Group, “RLP Protocol Performance 0.010.0 12.0 ’
14.0 16.0 18.0 20.0
Working Paper,” E.I.A/T.I.A TR45.3.2.5/93.07.06. C/I (dB)
2
Comm., vol. 36, pp. 389-400, Apr 1988. 6
S. Kallel, “Complementary Punctured Convolutional 2 40.0
E
(CPC) codes and Their Applications,” IEEE Trans. U
1033