Documente Academic
Documente Profesional
Documente Cultură
DOI 10.1007/s11276-015-1081-2
123
2072 Wireless Netw (2016) 22:2071–2079
when the requirements of the battery are permissive [11]. processed by the controller before transmitting it by the
The recent works have shown that for long block-lengths, transceiver unit.
the best performing LDPC codes are irregular codes and
these codes can outperform turbo codes of the same block- 3.2 LDPC encoder
length and code-rate [12]. In [13] the authors present a
cross layer analysis of error control schemes considering The LDPC encoder is implemented in the controller. The
routing, medium access and physical layer. This paper Mac Kay’s method is used for encoding. The LDPC code is
compares ARQ and FEC schemes in WSN and proves that encoded using the parity-check matrix directly by trans-
for more hops FEC schemes are better than ARQ scheme. forming it into upper triangular form and using back sub-
Low density parity check codes has its rate less advan- stitution [15]. The encoding procedure for LDPC codes
tage and while comparing it with fixed rate codes it attains consists of three parts: generation of the parity check
Shannon’s limit and it is very useful where Channel State matrix, the pre-processing part (done only once per matrix)
Information (CSI) is not available [14]. and the encoding of given information bits.
Finally, at the end of the survey, it is found that most of
• Generation of parity check matrix H.
the work is in identifying energy efficient error control
• Pre-processing H matrix using row and column
codes for WSN and very few works referred to cross lay-
permutations.
ered error control for WSN. According to the knowledge of
• Dividing the H^ matrix.
the authors it is reported that this work is one of the first of
A B T
its kind to use cross layered parameter to adaptively change • H^ =
C D E
the rate of the rate less LDPC code and hence achieve the
desired performance. where T—lower triangular matrix of size
(m - g) 9 (m - g), B—matrix of size (m - g 9 g), A—
matrix of size (m - g 9 k), C—matrix of size (g 9 k),
3 Cross layered adaptive rate optimized LDPC D—matrix of size (g 9 g), E—matrix of size
(CLARL) (g 9 m - g), The g rows of H left in C, D and E are called
gap of the approximate representation.
The proposed CLARL block diagram is shown in Fig. 1.
• Finding parity check bit vectors xp1 ; xp2
LOW-DENSITY parity-check (LDPC) codes were origi-
• Generation of encoded code word ð s xp1 xp2 Þ.
nally invented and investigated by Gallager [15]. LDPC
codes are block codes with parity-check matrices that
contain only a very small number of non-zero entries, 3.3 Adaptive rate optimizer
hence the name ‘‘low density parity check’’ code.
This module finds the rate of the LDPC code using cross
3.1 Sensor data/sensor sink layered parameters such as channel sate information and
user defined QoS requirements. The ARO is a block at the
These are the nodes in a sensor network, which are MAC layer as shown in Fig. 2. It takes the input from the
responsible for converting the environmental variables to routing layer and physical layer to compute the LDPC code
measure electrical variables and serve the purpose of the rate.
WSN. The sensed data is digitized using the ADC and
123
Wireless Netw (2016) 22:2071–2079 2073
The ARO has two phases the channel study phase and messages along the edges of a Tanner graph. Each Tanner
rate transfer phase. During the channel study phase the graph node works in isolation, only having access to the
ARO transmits test signals to its next node and collects the information contained in the messages on the edges con-
SNR and BER information. From the received parameters nected to it. The message-passing algorithms are also
the ARO computes the coherence time Tc of the channel. known as iterative decoding algorithms as the messages
From the physical layer parameters and the demanded QoS pass back and forth between the bits and check nodes
such as BER and data rate the ARO computes the coding iteratively until a result is achieved (or the process halted).
rate for the LDPC encoder as given in Table 1. A different message-passing algorithms are named for
the type of messages passed or for the type of operation
3.4 OQPSK modulator performed at the nodes. In some algorithms, such as bit-
flipping decoding, the messages are binary and in others,
The Offset Quadrature Phase Shift Keying (OQPSK) such as belief propagation decoding [17], the messages are
modulator maps the input binary signals to an analog signal probabilities which represent a level of belief about the
for transmission. value of the codeword bits. It is often convenient to rep-
resent probability values as log likelihood ratios and when
3.5 Channel this is done, belief propagation decoding is often called as
sum-product decoding since the use of log likelihood ratios
The channel is the medium by which information is allows the calculations at the bit and check nodes to be
transmitted from the transmitter to the receiver. In WSN, computed using sum and product operations.
this is a wireless channel. The addition of noise normally
occurs in the channel. In the simulations, the channel is
modeled as Additive White Gaussian Noise (AWGN) 4.1 SPA decoding algorithm for AWGN channel
channel. The resulting noise added to the system follows
the zero-mean normal distribution, with variance NO/2 and step1 Bit nodes send a message to their connected check
NO is the single-sided noise power spectral density. nodes.
step2 Check nodes calculate a response to their con-
3.6 LDPC decoder nected bit nodes using the messages received from step1.
step3 Bit nodes use the messages they get from the check
The decoder is implemented at the end-user receiving the nodes to decide a bit at that position.
information. The Sum Product Algorithm (SPA) is used for step4 Repeat steps 2 to exit or a certain number of
decoding. The decoding is a process that loops through iterations are reached.
passing messages back and forth along the tanner graph The codeword c = [0 0 1 0 1 1] is sent through an AWGN
[16] till certain conditions are satisfied or a maximum with symbol energy over noise = 1. 25 (0.9691 dB) and
number of passes have occurred. y = [1 0 1 0 1 1] is received. Since the channel is AWGN
the ri for the received code word y and the LLR is calcu-
3.7 Retrieve message from codeword lated by using
ri ¼ 4 Es =N0 yi
This process retrieves the estimated message from the
estimated codeword. and the received signal is
y ¼ ½ 0:1 0:5 0:8 1:0 0:7 0:5
4 LDPC decoding The LLR calculated value is
r ¼ ½ 0:5 2:5 4:0 5:0 3:5 2:5
The class of decoding algorithms used to decode LDPC
codes is collectively termed as message-passing algorithms For the purpose of illustration, we set the maximum
since their operation can be explained by the passing of number of iterations to three and pass in H and r.
123
2074 Wireless Netw (2016) 22:2071–2079
123
Wireless Netw (2016) 22:2071–2079 2075
Table 2 Complexity of
Operation Comment Complexity
determining xp1
AsT Multiplication by sparse matrix O (n)
T 1 ½AsT Multiplication by lower triangular matrix O (n)
E ½T 1 AsT Multiplication by sparse matrix O (n)
T Multiplication by sparse matrix O (n)
Cs
E ½T 1 AsT þ CsT Addition O (n)
F 1 ½E ½T 1 AsT þ C sT Multiplication by dense g 9 g matrix O (g2)
Table 3 Complexity of
Operation Comment Complexity
determining xp2
AsT Multiplication by sparse matrix O (n)
BxTp1 Multiplication by sparse matrix O (n)
BxTp1 þ As T Addition O (n)
T 1 ½BxTp1 þ AsT Multiplication by lower triangular matrix O (n)
check and generator matrices. The complexity of multi- nodes, (E) operations are required. Similarly, to sum all
plying a codeword with a matrix depends on the amount of check nodes, (E) operations are needed.
1’s in the matrix. The overall complexity of decoding one codeword is
In the normal way of computing, the sparse matrix H is therefore proportional to N 9 E. Where N is the number of
written in the form [pT I] via Gaussian elimination. The decoding iterations. For each code word encodes Rn infor-
generator matrix G is calculated as G = [I P]. The sub- mation bits, where R is the code rate and n is the block length,
matrix P is generally not sparse, so that the encoding the decoding complexity per information bit is O (N 9 E/Rn).
complexity will be quite high. Since the complexity grows
at O (n2) even sparse matrices do not result in a good
performance if the block length gets very high. Whereas, in 6 Results and discussion
the Mac Kay’s method of encoding the computational
complexity is O (n), this result in the linear time encoding The time taken for performing encoding and decoding of
system. So, the encoding complexity is less than the normal the LDPC codes for various values of data size n is given in
method of generating G matrix and hence the Elapsed time Table 4.
is also less which results in less energy consumption. From the table it is found that the time taken for
encoding is very less, it is in the order of milliseconds only,
5.2 Decoding complexity whereas for that of decoding the time taken is of 18.7 s
when n is about 2 k bits.
For a parallel message-passing decoder, the decoding
complexity of LDPC codes is proportional to the product of 6.1 Energy consumption
the number of decoding iterations and the number of
arithmetic operations performed per iteration. The com- The energy required for transmitting n bits is given by
putational effort per iteration is proportional to the number Eq. (3). The energy spent in receiving n bits is given by
of edges E in the graph of the code, while the number of Eq. (4). The energy spent for transmitting and receiving n
messages passed per iteration is 2E. Thus, the overall bits using lAMPS-1 mote is shown in Table 5.
complexity is proportional to the total number of messages Etx ¼ TstartPstart þ n=ðRÞðPtxElec þ ðaamp
passed in the iterative decoding process. þ bampPtxÞÞ ð3Þ
The variable-node update for the SPA can be performed
as follows. At each variable node of degree dv, we first where, n—length of the codeword, R—nominal data rate
compute the sum of the dv extrinsic messages and the
channel message; this requires dv additions. To compute 6.2 Energy consumed for receiving n bits
the output message for some particular edge, we then
Erx ¼ TstartPstart þ n=R Pr xElec þ n EdecBit ð4Þ
subtract the extrinsic message received over the same edge
from the computed sum. Thus, computing the m output From the obtained results, it is found that as the number
messages requires 2dv operations. To sum all variable of bits increases the transmitting energy and reception
123
2076 Wireless Netw (2016) 22:2071–2079
-1
10
BER
-2
10
-3
10
0 1 2 3 4 5 6 7
Eb/No (dB)
123
Wireless Netw (2016) 22:2071–2079 2077
energy increases form micro joules to mill joules. Also that Figure 5 shows that the BER performance of the pro-
in the case of 1000 bits the receiving energy is more or less posed algorithm. From the results it is found that as the
equivalent to that of transmitting energy this is because the energy per bit increases, the BER decreases in all the rates
receiver uses SPA decoder which does soft decoding and of the code. It is also found that bit error performance is
the algorithm iterates till the required performance is better for rate 1/3 code compared to rate by about 85 %
achieved. at the Eb/No value of 5 dB. This enhancement is at the cost
of increased redundancy of the LDPC code.
6.3 Bit error rate performance evaluation Figure 6 shows the output code rate of the ARO.
Depending upon the channel characteristics and the
Figure 4 shows the BER performance of the LDPC rate demanded QoS of the user, the ARO adaptively selects a
code with respect to other error correcting algorithms. code rate and maintains the selected code till the average
From the results, it is found that as the energy per bit coherent time of the channel.
increases, the BER decreases for all the codes. It is also Figure 7 shows that the BER performance of the
found that bit error performance is better for rate LDPC CLARL with respect to time. The proposed algorithm
code compared to other code. At the Eb/No value of 6 dB maintains the average BER at 10-3, even when the channel
the LDPC is 14 % better than RS code and 98 % better condition changes. This is due to the adaptive nature of the
than the uncoded.
0.7
ARO output
0.6
0.5
0.4
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50
Time (ms)
-1
10
BER
-2
10
-3
10
-4
10
0 5 10 15 20 25 30 35 40 45 50
Time (ms)
123
2078 Wireless Netw (2016) 22:2071–2079
CLARL to change the code rate dynamically with respect 7. Liu, B., Ren, F., Lin, C., & Ouyang, Y. (2008). Performance
to cross layer parameters. analysis of retransmission and redundancy schemes in sensor
networks. In IEEE international conference on communications
(pp. 4407–4413).
8. Tian, Z., Yuan, D., & Liang, Q. (2008). Energy efficiency anal-
7 Conclusion ysis of error control schemes in wireless sensor networks. In
IEEE international conference on wireless communications and
mobile computing (pp. 401–405).
To attain energy efficiency in wireless sensor networks 9. Bonello, N., Chen, S., & Hanzo, L. (2011). Low-density parity-
(WSN), we proposed the cross layered adaptive rate opti- check codes and their rateless relatives. IEEE Communication
mized LDPC as error control code for WSN. On imple- Survey and Tutorials, 13, 1.
mentation it is found that Mac Kay’s method for encoding 10. Howard, S. L., Schlegel, C., & Iniewski, K. (2006). Error control
coding at low-power wireless sensor networks: When is ECC
has a linear encoding complexity of the order of O (n). The energy-efficient? EURASIP Journal on Wireless Communications
Sum Product Algorithm (SPA) for decoding has O (N 9 E/ and Networking, 2006, 074812. doi:10.1155/WCN/2006/74812.
Rn) complexity. Also, the energy consumption for trans- 11. Hassan, N., Lentmaier, M., & Fettweis, G. P. (2012). Comparison
mitting and receiving n-bits for lAMPS-1 mote are cal- of LDPC block and LDPC convolutional codes based on their
decoding latency. In 7th international symposium on turbo codes
culated for various code lengths and it is found to be 0.687 and related topics (IEEE digital library). August 27–31, 2012.
and 0.585 mJ for n = 1000 respectively. The proposed 12. Awale, R. N., & Rathod, D. P. (2012). An efficient algorithm for
CLARL adaptive changes the coding rate of the LDPC irregular Low Density Parity Check Code with reduced compu-
coder to maintain the required QoS performance. This tational complexity and error floor. In International conference
on communication, information and computing technology
work can be enhanced further by fine tuning the optimizer (ICCICT), Mumbai, India. October 19–20, 2012.
performance using optimization algorithms. 13. Vuran, M. C., & Akyildiz, I. F. (2009). Error control in wireless
sensor networks: A cross layer analysis. IEEE/ACM Transactions
Compliance with ethical standards on Networking, 17(4), 1186–1199.
14. Richardson, T. J., & Urbanke, R. L. (2001). Efficient encoding of
Conflict of interest The authors of this paper declares that there is low density parity-check codes. IEEE Transactions on Informa-
no conflict of interests regarding the publication of this paper. tion Theory, 47, 2.
15. Gallager, R. G. (1963). Low-density parity-check codes. Cambridge,
MA: MIT Press.
16. Tanner, R. (1981). A recursive approach to low complexity
References codes. IEEE Transactions on Information Theory, IT-27,
533–547.
1. Holger, K., & Willig, A. (2005). Protocols, and architectures for 17. Varshney, L. R. (2011). Performance of LDPC codes under faulty
wireless sensor networks. London: Wiley. iterative decoding. IEEE Transactions on Information Theory, 57,
2. Vuran, M., & Akyildiz, I. (2006). Cross-layer analysis of error 7.
control in wireless sensor networks. In 3rd annual IEEE com-
munications society on sensor and ad hoc communications and
networks (SECON) (Vol. 2, pp. 585–594). T. Sasikala is a research scholar
3. Jong-Suk, A., Jong-Hyuk, Y., & Kang-Woo, L. (2007). Perfor- at the Department of Electronics
mance and energy consumption analysis of with FEC codes over and Communication Engineer-
wireless sensor networks. Journal of Communication and Net- ing-Guindy, Anna University
works, 9(3), 265–273. Chennai. She received the
4. Ahn, J. S., Lee, Y. S., Yoon, J. H., & Lee, K. W. (2006). Ana- Bachelor of Engineering Degree
lyzing the effect of a block FEC algorithms symbol size on in Electronics and communica-
energy consumption in wireless sensor networks. In H. Youn, M. tion Engineering from Govern-
Kim, & H. Morikawa (Eds.), Ubiquitous computing systems, ment College of Engineering,
lecture notes in computer science (Vol. 4239, pp. 440–453). Salem, Madras University. She
Berlin: Springer. did her post-graduation in
5. Wen, H., Lin, C., Ren, F., Yue, Y., & Huang, X. (2007). Applied Electronics from Col-
Retransmission or redundancy: Transmission reliability in wire- lege of Engineering-Guindy,
less sensor networks. In IEEE international conference on mobile Anna University; Chennai. She
adhoc and sensor systems (pp. 1–7). is doing her research in faculty
6. Chiasserini, C. F., & Magli, E. (2004). Energy-efficient coding of Information and Communication Engineering, Anna University
and error control for wireless video-surveillance networks. Chennai.
Telecommunication Systems, 26, 369–387.
123
Wireless Netw (2016) 22:2071–2079 2079
123