Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Abstract—Cognitive radio is an emerging technology that are also very strong candidates for anti-jamming codes [4],
allows secondary communication in a frequency band originally [5]. The comparison of various erasure correcting schemes
allocated to one or more primary users (PUs). Secondary user in the context of cognitive radio has been discussed in [3].
(SU) usually operates in a low power profile to avoid interference
with the PU and sudden appearance of a PU causes loss of data It shows that, among different erasure correcting schemes,
of the SU. Appropriate erasure correction coding (also called as rateless coding and piecewise linear coding are the two high
anti-jamming coding) is, thus, required to maintain the reliability performing anti-jamming coding schemes in cognitive radio
of the link. In this paper, we propose the random linear coding systems. Piecewise linear coding, proposed in [5], involves
(RLC), which is commonly used in network coding, to be used arranging the packets as columns of a matrix and applying
as an anti-jamming scheme. We also propose a variant of RLC
anti-jamming scheme that requires reduced complexity decoding. the linear block code on each row to generate parity packets
The throughput of both the proposed schemes is analyzed and (encoded columns).
it is shown that high throughput can be obtained at the expense In the emerging area of network coding, the random linear
of very small redundancy by carefully selecting the Galois field coding (RLC) is becoming increasingly popular tool to achieve
size.
multicast capacity [6], [7]. We propose to use RLC as an anti-
jamming technique motivated by the following observations:
I. I NTRODUCTION
• Cognitive radio system with spectrum pooling model
The concept of cognitive radio is based on the notion of [8] used for secondary usage behaves similar to the
secondary spectrum access to improve spectrum utilization. distributed storage system; and jamming behaves similar
This allows a non-primary or unlicensed user, called as a to the storage device failure [3].
secondary user (SU), to dynamically access the unused parts • RLC based schemes perform well in case of uncoordi-
of the spectrum owned by the primary license holder, called a nated distributed storage systems [9], [10].
primary user (PU), temporarily [1]. Unused licensed spectrum
Since decoding in case of RLC requires inversion of a dense
can be detected and allocated to SUs in an opportunistic way
matrix over a finite field, which is computationally complex;
[2]. Though SU can sense the presence of a PU, the PU is
hence we propose reduced complexity RLC (RC-RLC) scheme
not equipped with the sensing capability and is not aware of
for which computationally efficient decoding is possible. The
the existence of the SUs. Thus, if a PU suddenly becomes
throughput performance of the RLC scheme is analyzed for
active in a particular band currently used by SU, it will cause
the cognitive radio set up. Also, performance lower bounds are
collision with the SU and the packets transmitted by the SU
developed for both RLC and RC-RLC schemes by utilizing the
through this primary band may get corrupted due to strong
mathematical tools presented in [9] and [10].
interference from the PU. Such a collision is referred to as
The remainder of the paper is organized as follows. Sec-
jamming. Since SU usually operates in a lower power profile
tion II gives system model along with RLC based anti-
in order to avoid interfering with the PU, the entire packet
jamming coding schemes. Performance analysis is carried out
could get corrupted. The channel coding may not be able to
in Sections III and IV, and numerical results are presented in
recover the data associated with this packet and the packet
Section V. Finally, Section VI concludes the paper.
is considered to be lost. Therefore, anti-jamming coding is
required to recover such a corrupted packet. II. S YSTEM M ODEL
The packet affected by jamming is either received correctly
or completely lost. Thus, jamming can be considered to First, the concept of spectrum pooling is introduced, which
be equivalent to the erasure event. This makes conventional forms the basis for secondary usage model. Then, RLC and
erasure codes like Reed-Solomon (RS) codes and Low-Density RC-RLC based anti-jamming schemes are explained.
Parity-Check (LDPC) codes to be natural candidates as anti-
jamming codes. However, these codes have highly structured A. Spectrum Pooling
construction and hence are less flexible on the number of As discussed in [8], the spectrum pool is formed by dividing
redundant packets and the number of encoded packets. This the whole spectrum where secondary use is permitted, into
makes their direct application unsuitable in cognitive radio [3]. number of subchannels. The SU is assigned a set of subchan-
Fountain codes, which are from the class of rateless codes, nels on which PUs are not active. The subchannel selection
Proof: See Appendix A. This subsection further characterizes the RLC scheme by
Theorem 1 helps us to see the effect of field size on finding the lower bounds on the important parameters: over-
the throughput. It is easy to observe that the throughput head factor and decoding success probability.
is an increasing function of q. Also, for asymptotic field The overhead factor is a commonly used parameter in
size (q → ∞), term II tends to unity and the throughput erasure coding to indicate the quality of the code. It is ratio
is code rate times the term I, which is nothing but the of overhead by number of data packets where the overhead is
probability of loosing ≤ (h = n − k) packets. In other the average number packets that must be received to recover
words, we can say that for very large field sizes, RLC has all the data packets [12].
a perfect decoding threshold on number of lost packets, i.e., Let rmin denote the minimum number of packets that
the decoding failure probability behaves as a step function: should be received for successful decoding. Then, the overhead
P r {E|(l > h)} = 0 and P r {E|(l ≤ h)} = 1. Note that this can be obtained as E {rmin }; the expectation enters into
is the ideal performance that can be achieved using any coding the scenario as the coding framework is probabilistic. For
scheme; for example, RS codes, which have the optimum decoding to be successful, the rank of the r×k matrix obtained
erasure correction capability. by stacking the weight vectors associated with the r received
1 1
RLC
0.9 RLC (LB) 0.9
RC−RLC (LB)
0.8 No Coding
0.8
0.7 0.7
Throughput (η)
Throughput (η)
0.6 h = {1, 3, 5}
0.6
0.5 0.5
0.4 0.4
RLC, p = 0.01
0.3 0.3 RC−RLC, p = 0.01
RLC, p = 0.05
0.2 0.2 RC−RLC, p = 0.05
RLC, p = 0.1
0.1 0.1
RC−RLC, p = 0.1
0 −3 −2 −1 0
0
10 10 10 10 0 2 4 6 8 10 12 14 16 18 20
Jaming Rate (p) Redundancy (h)
Fig. 1. Analytical results for throughput η versus jamming rate p for k = 40 Fig. 2. Analytical results for throughput η versus redundancy h = n − k
and q = 27 . for k = 40 and q = 27 .
0.7
packets should be k. Then, using proposition 5.1 in [9], it is
straightforward to obtain 0.6
h=5
E {rmin } q 0.5
Throughput (η)
. (9)
k q−1 0.4
be obtained as 0.2
RLC
1 RLC (LB)
(1 − Pr {E|r}) ≥ 1 − ,
0.1
(10) RC−RLC (LB)
Ideal
k 0
1 2 3
10 10 10
provided Field Size (q)
r 2 logq (k)
≥1+ , (11) Fig. 3. Analytical results for η versus field size q for k = 40, p = 0.1, and
k k various h.
by using proposition 5.2 in [9].
From the above results it is possible to make following
remarks. clearly outperform the no coding case except when jamming
• By choosing a field size large enough, overhead ratio can rate is very small. Also, one can observe that larger amount
be made arbitrarily close to one, which is the optimum of redundancy must be introduced if the jamming rate is high
case. to obtain better performance.
• It is sufficient to receive just over k coded packets to To better understand the effect of redundancy h on the
recover all the data packets almost certainly (assuming k throughput, Figure 2 is presented for different jamming rates.
to be sufficiently large). It shows that, for a given p, an optimal value of h exists
At this juncture, it is important to highlight the trade- giving the highest throughput and the optimal h increases with
off between field size and complexity: a large field size the jamming rate. Also, it can be seen that both the schemes
achieves minimally additional overhead at the expense of large require just the marginal amount of redundancy.
complexity associated with the inversion of a matrix over a In Figure 3, we analyze the effect of the field size q on the
larger field size. throughput. The binary extension field GF (2l ) is considered
for l = {1, 2, . . . , 10}. We also present the performance of the
V. S IMULATION R ESULTS ideal scheme which can recover the original data packets from
In this section, we evaluate the performance of both the any k received packets with certainty. It can be seen that the
proposed schemes numerically. It is assumed that the message performance of both the schemes approaches the ideal case
to be transmitted consists of 40 packets which are encoded as field size increases. The gap between the lower bound and
into n = k + h packets using RLC or RC-RLC anti-jamming the actual throughput for RLC reduces with the field size.
scheme. We consider the binary extension field GF (2l ), where It is interesting to note that the effect of q follows the law
l is an integer. of diminishing returns; saying that beyond certain value of q
Figure 1 shows the throughput of the RLC scheme and lower the gain in throughput will not be significant. Also, it can be
bound (LB) on the throughput for both RLC and RC-RLC observed that the throughput is close to the ideal for reasonably
schemes as a function of jamming rate p for a field size of small field sizes, making the proposed schemes elegant from
q = 27 and various redundancy (h) values. The throughput the implementation perspective.
of the simple transmission scheme without any anti-jamming Figures 4, 5, and 6 illustrate the simulation results. We have
is also plotted for comparison. We can see that both schemes chosen c = 5n/k + 1 to implement RC-RLC scheme. All the
1 0.8
0.9 RLC
RC−RLC 0.7
0.8
h=5
0.7 0.6
Throughput (η)
Througput (η)
0.6
0.5
0.5
h=1,3,5
0.4
0.4
0.2
0.2 RLC
0.1
RC−RLC
0 0.1
−3 −2 −1 0 0 1 2 3
10 10 10 10 10 10 10 10
Jamming Rate (p) Field Size (q)
Fig. 4. Simulation results for throughput η versus jamming rate p for k = 40 Fig. 6. Simulation results for throughput η versus field size q for k = 40,
and q = 27 . p = 0.1, and various h.
0.8
piecewise linear coding, respectively.
0.7
Throughput (η)
0.6
VI. C ONCLUSION
0.5
0.9
p = 0.01
0.8 0.8
h=5
0.7
Throughput (η)
p = 0.1
Throughput (η)
h=3 0.6
0.6
0.5
0.4
0.4