IET Communications

Research Article

ISSN 1751-8628
Efficient coding schemes for low-rate wireless Received on 13th May 2015
Revised on 7th December 2015
personal area networks Accepted on 15th February 2016
doi: 10.1049/iet-com.2015.0699
www.ietdl.org

Hua Qian 1,2, 3,4 ✉, Shengchen Dai 1,2, 4, Kai Kang 3, Xudong Wang 5
1
Shanghai Institute of Microsystem and Information Technology (SIMIT), Chinese Academy of Sciences (CAS), Shanghai,
People’s Republic of China
2
Key Laboratory of Wireless Sensor Network and Communication, CAS, Shanghai, People’s Republic of China
3
Shanghai Advanced Research Institute (SARI), CAS, Shanghai, People’s Republic of China
4
Shanghai Research Center for Wireless Communications, Shanghai, People’s Republic of China
5
University of Michigan-Shanghai Jiao Tong University Joint Institute, Shanghai Jiao Tong University, Shanghai,
People’s Republic of China
✉ E-mail: qianh@sari.ac.cn

Abstract: The emerging market of Internet of things has created great demand for low-cost, low-power wireless
technologies. Existing IEEE 802.15.4 standard is designed for low-rate wireless personal area networks (LR-WPANs).
However, current standard does not fully utilise the benefit of the code redundancy. In this study, the authors propose
new coding schemes for LR-WPANs with improved coding gain. They first propose a block code based on extended
Bose–Chaudhuri–Hocquenghem (BCH) code that increases the minimum Hamming distance compared with the
existing code used in LR-WPANs. The computational complexity of the encoder and decoder remains about the same.
In addition, by applying the extended BCH code directly to LR-WPANs, the data rate of the system can be increased
without sacrificing coding performance. They further propose a tail-biting convolutional (TBC) code with optimum
generator polynomials for LR-WPANs. The proposed TBC code enjoys significant performance improvement while
preserving the effective code rate as well as a low decoding complexity. Simulation results validate the effectiveness of
the proposed coding schemes.

1 Introduction mapping can be viewed as a (32, 4) block code. The minimum

Hamming distance of this code is 12. The coding gain is ∼2 dB at
As the Internet of things (IoT) attracts more and more attention packet error rate (PER) of 1% [9].
recently, low-cost and low-power wireless communications From channel coding perspective, the coding scheme provided in
become essential to provide ubiquitous connectivity [1]. current standard is not optimal. For general block codes, Griesmer
Constant-envelope modulation technique and low-rate (LR) [10] suggested an upper bound of minimum Hamming distance of
coding schemes are adopted to improve power efficiency and the codes. The minimum Hamming distance must be ≤16 for a
system performance of LR wireless personal area networks (32,4) code, suggesting that extra coding gain is possible. For
(LR-WPANs) at different frequency bands. Among all wireless convolutional codes, the bound of free distance listed in [11]
standards, IEEE 802.15.4 standard plays an important role in IoT limits the system performance. The convolutional codes with
for LR-WPANs. The 2.4 GHz physical layer (PHY) in IEEE maximum free distance can be utilised to improve performance. In
802.15.4 standard supports an over-the-air data rate of 250 kb/s. a word, there is plenty of room to improve the code efficiency for
Direct sequence spread spectrum (DSSS) is applied to improve LR-WPANs.
system performance and the chip rate is 2 Mb/s. The signals are In this paper, we propose new coding schemes with improved
then modulated by offset quadrature phase shift keying (OQPSK) coding gain for LR-WPANs. We propose a block code based on
with half-sine pulse shaping filter [2]. Since this modulation is extended Bose–Chaudhuri–Hocquenghem (BCH) code that
identical to minimum the shift keying (MSK) modulation, the improves the coding gain. Such benefit is obtained without extra
transmitted signals have constant envelope. Such a transmitter computational complexity. By applying the extended BCH code
has advantage in adopting non-linear power amplifiers and directly to LR-WPANs, the data rate can be increased without
achieving high-power efficiency. sacrificing the coding performance. To obtain additional coding
Conventional approaches apply differential detection to detect gain, we further propose a tail-biting convolutional (TBC) code
MSK signals. The performance of such detection schemes is not with optimum generator polynomials for LR-WPANs. System
satisfactory [3–7]. The coherent detection achieves good performance can be significantly improved.
performance. However, it is sensitive to initial phase error or The rest of this paper is organised as follows. Section 2 provides
frequency offset. Dai et al. [8] proposed a robust detector, which system setup for LR-WPANs. We take the coding scheme of the 2.4
provides satisfactory performance with only 1 dB loss compared GHz PHY in IEEE 802.15.4 standard as an example for further
with the coherent detector in additive white Gaussian noise discussion. Section 3 proposes new coding schemes based on
(AWGN) channel. block codes. The proposed coding scheme based on convolutional
In addition to improve the performance of demodulation schemes, codes is presented in Section 4. Simulation results are provided in
the overall system performance can be further improved by fully Section 5 that demonstrates the effectiveness of the proposed
utilising the benefit of the code redundancy. For example, DSSS coding schemes. Section 6 extends the proposed coding schemes
in IEEE 802.15.4 standard is not simple spreading. It provides to other LR-WPANs standards. Finally, Section 7 concludes this
coding gain in addition to spreading gain. The symbol-to-chip paper.

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

& The Institution of Engineering and Technology 2016 915
2 System setup Table 1 Symbol-to-chip mapping table in IEEE 802.15.4 standard
Data symbol Chip values
In this section, we provide a system setup for LR-WPANs. Fig. 1 (u0, u1, u2, u3) (c0, c1, …, c31)
shows the transceiver structure of general LR-WPANs. An LR
encoder is applied to improve the receive sensitivity. The coded 0000 1101 10011100 00110101 001000101 110
bits are then modulated to transmitted signals. To meet the spectral 1000 1110 11011001 11000011 010100100 010
0100 0010 11101101 10011100 001101010 010
mask, pulse shaping filter is applied. The digital baseband signals 1100 0010 00101110 11011001 110000110 101
are then transformed to analogue domain by digital-to-analogue 0010 0101 00100010 11101101 100111000 011
converter. For a single-carrier, narrow-band system in the 1010 0011 01010010 00101110 110110011 100
applications of LR-WPANs, AWGN channel is reasonable 0110 1100 00110101 00100010 111011011 001
1110 1001 11000011 01010010 001011101 101
assumption. At the receiver side, the baseband analogue signals 0001 1000 11001001 01100000 011101111 011
are sampled to digital signals by analogue-to-digital converter. 1001 1011 10001100 10010110 000001110 111
Matched filter is utilised to maximise the signal-to-noise ratio 0101 0111 10111000 11001001 011000000 111
1101 0111 01111011 10001100 100101100 000
(SNR) of the received signals. After synchronisation, the receiver 0011 0000 01110111 10111000 110010010 110
demodulates and decodes the signals. 1011 0110 00000111 01111011 100011001 001
Channel coding is essential to improve the system performance in 0111 1001 01100000 01110111 101110001 100
LR-WPANs. In this paper, we take the coding scheme of the 2.4 1111 1100 10010110 00000111 011110111 000
GHz PHY in IEEE 802.15.4 standard as an example for further
discussion. The encoder for IEEE 802.15.4 standard is shown in
Fig. 2. The information bits are segmented into symbols every 4
bits. The symbol u = (u0, u1, u2, u3) is spread to a 32 bit chip finding new coding schemes with improved coding gain for
sequence, which is denoted by c = (c0, c1, …, c31). Table 1 is the LR-WPANs in this section. This paper is equivalent to finding a
symbol-to-chip mapping table in IEEE 802.15.4 standard. OQPSK set of codewords with large minimum Hamming distance. It is
with half-sine pulse shaping filter is applied. Such modulation is known that the Griesmer bound [10], which is effective for codes
identical to MSK modulation. Since the analogue transmitted with low code rate, provides an upper bound of linear block codes.
signals have constant envelope, power-efficient non-linear power With the Griesmer bound, the minimum Hamming distance dmin
amplifiers can be applied in the transmitter. of linear block codes is bounded by
The encoder used in the standard is not simple spreading. It
provides coding gain. The coding gain for a coded binary OQPSK k −1 
 
d
system can be approximated by [12]
min
≤ n, (2)
i=0
2i


k ln 2
g ≃ 10 log10 Rc dmin − , (1) where ‘⌈·⌉’ denotes the smallest integer that is larger than or equal to
Eb /N0 the input. For (32,4) linear block codes, Griesmer bound suggests
that dmin ≤ 16. On the other hand, the bound table in [13] shows
where k is the length of information bit sequence, n is the length of that linear block codes with dmin = 16 are possible. Our task is to
codewords, dmin is the minimum Hamming distance of the set of find the codes that achieve the Griesmer bound.
codewords, Rc = k/n is the code rate and Eb/N0 is the SNR per bit. We start the search with some well-studied linear block codes.
In IEEE 802.15.4 standard, the encoder maps every four One of the best-known codes, which are attractive for
information bits to a 32 bit chip sequence. The minimum low-to-moderate block length, is BCH code. From the extensive
Hamming distance of the set of codewords is 12. The coding gain list of BCH codes, we observe that BCH codes do not support the
is about 2 dB [9]. On the other hand, this set of codewords is not codeword length n = 32. Instead, the (31,6) primitive BCH code
optimised for coding gain. provides the largest minimum Hamming distance dmin = 15 with
codeword length n = 31 [12]. From [14], we know that by adding
a parity-check bit, the (n, k) binary linear block code with odd
3 Coding schemes based on linear block codes minimum Hamming distance dmin can be extended to a new (n +
1, k) code. Meanwhile, the minimum Hamming distance of the
Since linear block codes are easy to implement and analyse, and their extended code is also increased by 1. By padding a parity-check
performances are similar to that of general block codes, we focus on bit to the (31,6) BCH code, the codeword length becomes 32 and
the minimum Hamming distance of this code increases to 16.
To design a bound-achieving (32,4) block code that is compatible
with the 2.4 GHz PHY in IEEE 802.15.4 standard, we choose a
subset of the (32,6) extended BCH code corresponding to four
independent information bits. A possible way to generate a (32,4)
code based on the (32,6) extended BCH code is shown in Fig. 3.
The input information bits, which are segmented into blocks of 4
bits, can be expressed as u = (u0, u1, u2, u3). In the ‘Bit Pad’
block, u is extended to u′ by padding two bits. These two bits can
be padded in any position of u. Valid combinations can be u′ =
(u0, u1, u2, u3, 0, 1) or u′ = (u0, u1, u2, 0, u3, 1). To be
Fig. 1 Transceiver structure of general LR-WPANs
complied with the IEEE 802.15.4 standard, we emphasise that the
padded bits are not allowed to be all zeros. Since in this case,

Fig. 2 Encoder of the 2.4 GHz PHY in IEEE 802.15.4 standard Fig. 3 (32,4) Code based on the (32,6) extended BCH code

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

916 & The Institution of Engineering and Technology 2016
 Fig. 5 (32,6) Extended BCH code for LR-WPANs
Fig. 4 Proposed (32,4) code based on extended BCH code for LR-WPANs

Similarly, we obtain a 32 bit codeword c = (c0, c1, …, c31) by

all-zero or all-one codewords are possible. The mean transmit power adding a parity-check bit after the (31,6) BCH encoder. To avoid
may be different from other codewords. In addition, the pattern in all-zero or all-one codewords, a whitening block can be applied.
all-zero or all-one codewords may interfere with other systems Next, we will discuss the decoder of the proposed coding schemes.
transmitting in the same frequency band. The ‘BCH Encode’ In short-range communications such as IEEE 802.15.4, AWGN
block encodes the 6 bit sequence u′ with the corresponding (31,6) channel is considered to be the general case [15]. At the receiver
BCH code. At last, by adding a parity-check bit to the output of side, with perfect synchronisation, the received baseband signal
the BCH encoder, we obtain a 32 bit codeword c = (c0, c1, …, can be represented as
c31). The proposed coding scheme can be summarised in Fig. 3.
By studying the codewords of the (32,6) extended BCH code, we
observe that half of the 64 codewords are mutual complements of the ri = si + vi , (3)
other half. All chips between two complementary codewords are
opposite. Each pair of such codewords has the largest Hamming where ri is the received baseband signal at the ideal sampling time, si
distance 32 for the (32,6) extended BCH code. The design of is the ith chip modulated with OQPSK in a 32-chip sequence and vi
encoder and decoder can be further simplified by this property. is the equivalent AWGN with variance σ 2. Since the proposed (32,4)
Fig. 4 shows the updated (32,4) block code. The four information code is based on the (32,6) extended BCH code, existing decoding
bits u = (u0, u1, u2, u3) are separated into two parts: (u0, u1, u2) algorithms of BCH codes are available. However, the performance
and u3. In the ‘Bit Pad’ block, three bits are padded to (u0, u1, of these hard-decision decoding algorithms is not satisfactory.
u2). One valid combination is u′ = (u0, u1, u2, u3 = 0, 0, 1). The With a coherent demodulator for the OQPSK system [8], the
corresponding BCH code and extended BCH code can be information is only contained in one branch and the noise in the
obtained. In this case, only eight codewords are chosen. The bit u3 other branch can be completely ignored. The ith chip in a symbol
is utilised to generate the complementary codewords. When u3 = 1, for coherent demodulator can be described as
we first generate the corresponding codeword with u3 = 0. Then,

we flip all bits in the codeword to get the complementary Re{ri } = s′i + Re{vi }, (i mod 2) = 0,
r′i = (4)
codeword with a logic inverter. It is worth noting that these eight Im{ri } = s′i + Im{vi }, otherwise,
codewords are not generated from the BCH encoder and the
parity-check bit extension. However, it is easy to show that these
eight codewords are valid codewords satisfying the extended BCH where Re{x} is the real part of x, Im{x} is the imaginary part of x and
code. Table 2 gives one possible symbol-to-chip mapping table of s′i [ {−1, +1} is the ith chip modulated with binary phase shift
the proposed (32,4) code. The proposed code has a minimum keying. For a certain codeword c, the corresponding mapped chip
Hamming distance of 16 and performs better than the codewords sequence would be s′ = (s′0 , s′1 , s′2 , . . . , s′31 ). Since Re{vi} and
used in the IEEE 802.15.4 standard. Im{vi} are Gaussian noises with variance σ 2/2, the conditional
The performance of the coding schemes for LR-WPANs can be probability density function of received bit r′i given a prior
further improved. With a given codeword length n, (1) indicates information of transmitted bit s′i can be estimated by
that the coding gain of block codes increases linearly with respect
to the information bit length k when the minimum Hamming 1 ′ ′ 2 2
f (r′i |s′i ) = √ e−(|ri −si | /s ) . (5)
distance dmin is the same. Since the (32,6) extended BCH code ps 2
achieves the same minimum Hamming distance dmin = 16 as the
proposed (32,4) code, the (32,6) extended BCH code as shown in The conditional probability density function of the received 32-chip
Fig. 5 can be directly applied in LR-WPANs to provide better sequence given a prior information of the transmitted chip sequence
coding gain. The information bits are segmented into blocks of 6 s′ can be estimated by
bits, which can be defined by u = (u0, u1, u2, u3, u4, u5).

31
′ ′ 2
f (r′ |s′ ) = e−|ri −si |
Table 2 Symbol-to-chip mapping table of the proposed (32,4) code i=0
based on extended BCH code


31 31
−(|r′i |2 +1) 2r′i ·s′i
Data symbol Chip values = e ×e i=0 , (6)
(u0, u1, u2, u3) (c0, c1, …, c31) i=0

0000 0000 01100101 10111101 010001001 111

1000 1000 01010111 01100011 111001101 000 where r′ = (r′0 , r′1 , r′2 , . . . , r′31 ) is the received soft chip sequence.
0100 0100 01001110 00001100 101101111 011
31 −(|r′i |2 +1)
In (6), the term i=0 e is a constant for all possible
1100 1100 01111100 11010010 000101011 100  ′ ′
0010 0010 01110000 01100101 101111010 101 codewords c. The term 31 i=0 2r i · s x
i is a real number. Since e is a
1010 1010 01000010 10111011 000111110 010 monotonically increasing function when x is real, the
0110 0110 01011011 11010100 010011100 001
1110 1110 01101001 00001010 111011000 110 maximum-likelihood (ML) sequence decoder of the above block
0001 1111 10011010 01000010 101110110 000 code can be written as
1001 0111 10101000 10011100 000110010 111
0101 1011 10110001 11110011 010010000 100
1101 0011 10000011 00101101 111010100 011 ĉ = arg max f (r′ |s′ )
0011 1101 10001111 10011010 010000101 010 c
1011 0101 10111101 01000100 111000001 101 (7)
0111 1001 10100100 00101011 101100011 110 
31
1111 0001 10010110 11110101 000100111 001 = arg max r′i · s′i .
c i=0

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

& The Institution of Engineering and Technology 2016 917
Equation (7) suggests that the optimal cross-correlation between the n − 1 + 21−h}. We have
received soft chips and the modulated codeword gives the ML
sequence decoder of the codeword.  
By adopting the proposed (32,4) code defined in Fig. 4, we can 2l−1
df ≤ min l (K + l − 1)n − 1. (10)
further simplify the ML sequence decoder by computing only half l≥1 2 − 1
of the correlation values since half of the codewords are just the
opposite values of the other codewords.
At the end of this section, we would like to discuss the Exhaust search of generator polynomials, which can be terminated
computational complexity of the proposed block coding schemes. when the free distance of encoder achieves the improved Heller
Comparing the proposed (32,4) code with the code used in the bound, has been executed to find maximum free distance
standard (as shown in Table 1), a simple look-up-table (LUT) can convolutional codes [12]. The maximum free distance
be applied as the encoder. The codewords used in the standard convolutional codes are often deemed as optimum convolutional
have cyclic shift structure that can be applied to further simplify codes in AWGN channel. Frenger et al. [16] presented a family of
the encoder design. Nevertheless, the encoder complexities of both rate-compatible LR convolutional codes with maximum free
codes are low. For the proposed (32, 6) code, the LUT-based distance. Maximum free distance codes with rates ranging from 1/
encoder is slightly more complex than the code used in the 4 to 1/512 and constraint lengths between 3 and 11 are tabulated
standard. On the other hand, if we generate the codewords from in [16]. With these codes, a convolutional code of any code rate
the BCH generator polynomial, the encoder of the (32, 6) code Rc = 1/n can be obtained for different constraint lengths, resulting
can also be simple. in flexible and powerful coding schemes. In this paper, we utilise
In the decoder, the above proposed ML sequence decoder the existing LR convolutional codes for further discussion.
achieves optimal symbol error rate for the proposed block codes as In LR-WPANs, the packet length is usually small. The application
well as the code used in the standard. As we discussed earlier, for of convolutional codes may result in rate loss when transmitting
the proposed (32,4) code, we only need to compute eight additional known terminating bits to flush out the memory. TBC
cross-correlation values between received soft chips and the codes can help to eliminate the rate loss by initialising the memory
candidate codewords. Given the properties of the code used in the state with actual tail bits in the transmitted sequence. It is widely
standard, we conclude that the optimal decoder also needs to used in numerous communication systems such as the worldwide
compute eight cross-correlation values. Thus, the decoding interoperability for microwave access and the long-term evolution.
complexities of these two codes are about the same. For the In IEEE 802.15.4 standard, the maximum length of payload is as
proposed (32,6) extended BCH code, 32 cross-correlation values small as 20 bytes. The TBC codes with optimum generator
need to be computed. The decoding complexity is slightly higher polynomials are suitable coding schemes for short-packet
than the (32,4) code. In summary, the proposed block code applications.
provides additional coding gain to the LR-WPANs systems with Let C be an (n, 1) TBC code, where n is the number of output bits
no or little additional computational complexity. per input information bit. The encoder of TBC code is initialised
with the tail K − 1 information bits in the transmitted sequence,
where K is the constraint length of the code. The output bits per
information bit are generated corresponding to memory state and
4 Coding schemes based on convolutional codes generator polynomials, which can be defined by c = (c0, c1, …,
cn−1).
In Section 3, we propose the improved coding schemes based on At the receiver side, many studies have been made to reduce the
linear block codes. The proposed coding schemes are similar to decoding complexity of the ML decoders for tail-biting trellises
that used in current standard and require minimum changes to the [17–19]. Sub-optimal decoders such as wrap-around Viterbi
standard. The proposed coding schemes achieve the upper bound algorithm [20] and other alternatives [21] are also developed.
of dmin as linear block codes. In practice, the coding gain can be Since decoding algorithms of the TBC code are not the focus of
further improved if we do not restrict the type of codes. One of this paper, we select a low-complexity sub-optimal method in [21]
the alternative codes is convolutional code. In this section, we to decode the TBC code for LR-WPANs. Denote
discuss the possibility of applying convolutional codes to r′ = (r′0 , r′1 , . . . , r′nL−1 ) as the received soft chip sequence, where
LR-WPANs. Optimum generator polynomials of convolutional L is the length of the information sequence. Copying the ending
codes are utilised for LR-WPANs. nT values of r′ to the front of the received sequence, and copying
In general, we assume that the encoder has a code rate Rc = 1/n the leading nH values of r′ to the end of the received sequence,
with constraint length K in LR-WPANs. The asymptotic coding the extended received sequence for TBC code decoder is given by
gain for convolutional codes with soft decision in AWGN channel
can be defined as re = (r′nL−1−(nT−1) , . . . , r′nL−1 , r′0 , r′1 , . . . , r′nL−1 , r′0 , . . . , r′nH−1 ),
(11)
g ≃ 10 log10 (Rc df ), (8)
where T is the length of copied information bits in the tail and H is
where df is the free distance of convolutional codes. For a given code the length of copied information bits in the head. A super trellis can
rate Rc, the asymptotic coding gain is restricted by the free distance be constructed on re. It is safe to assume that the starting and
of convolutional codes. The free distance of convolutional codes is terminating states are known as all-zero states for this trellis.
directly related to the generator polynomials. The Heller bound or Viterbi algorithm can be applied to the super trellis re. The
the upper bound of the free distance of convolutional codes with corresponding decoder output of the middle L bits is the decoder
code rate Rc = 1/n is given by [11] output. The performance loss of this sub-optimal decoder is
negligible when T ≥ 5K and H ≥ 5K [21].
  In IEEE 802.15.4 standard, the chip rate is eight times to that of
2l−1 information bits, which can be viewed as an encoder with code
df ≤ dH W min l (K + l − 1)n , (9)
l≥1 2 − 1 rate Rc = 1/8. For a given code rate Rc = 1/8, the maximum free
distance of improved Heller bound is shown in Fig. 6. The
maximum free distance increases almost linearly with constraint
where ‘⌊·⌋’ denotes the largest integer that is less than or equal to the length K. However, the number of states in trellis, which indicates
input. Assume that dH is minimised  at l = h, that is, the complexity of decoding algorithm, increases exponentially
dH = minh≥1 (2h−1 /2h − 1)(K + h − 1)n . Then, a tighter bound with the constraint length of the code. The constraint length K = 7
can be obtained when dH is odd and dH > (2h−1/2h − 1){(K + h − 1) is chosen as a compromise between performance and complexity

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

918 & The Institution of Engineering and Technology 2016
 extended BCH code, the (32,6) extended BCH code and the TBC
code with optimum generator polynomials in AWGN channel.
Fig. 7 shows the PER performance of different coding schemes in
AWGN channel. The PER performance with respect to SNR per bit
is shown in Fig. 7a. For LR-WPANs, the receive sensitivity is
usually measured by SNR per chip. We redraw Fig. 7a in Fig. 7b
to show the PER performance with respect to SNR per chip. The
dashed lines show PER performance of the (32,4) code used in
current standard; the solid lines show PER performance of the
(32,4) code based on extended BCH code; the dash-dotted lines
show PER performance of the (32,6) extended BCH code; and the
dotted lines show PER performance of the TBC code with
optimum generator polynomials. From Fig. 7a, we observe that at
the PER of 1%, the (32,4) code based on extended BCH code
achieves about 0.5 dB coding gain compared with the (32,4) code
used in current standard. The (32,6) extended BCH code achieves
about 1.8 dB coding gain compared with the (32,4) code used in
current standard. While the TBC code with optimum generator
polynomials achieves about 4.3 dB coding gain compared with the
(32,4) code used in current standard.
Fig. 6 Maximum free distance of the improved Heller bound with code rate
In Fig. 7b, the x-axis changes to SNR per chip, which denotes the
Rc = 1/8
receive sensitivity in practice. Since the (32,4) code based on
extended BCH code and the (32,4) code used in current standard
have the same rate, the performance gain is 0.5 dB in terms of the
[22, 23]. The corresponding generator polynomials of TBC code in sensitivity as well. We would like to emphasise that the encoding
octal format are 117, 127, 135, 135, 145, 155, 171, 173 [16].
Next, we evaluate the encoding and decoding complexities of the
proposed TBC code. The encoder of the TBC involves the operation
of shift registers with length 6. The encoder complexity is low. In the
decoder part, the sub-optimal reduced complexity circular Viterbi
decoding algorithm [21] gives satisfactory performance with
comparable complexity as the regular Viterbi algorithm. Such
decoder has higher computational complexity than the ML
sequence decoder for the block code. However, the computational
complexity of the Viterbi-like algorithm is still acceptable for
LR-WPAN systems with reasonable increase of the logic size and
memory size. We emphasise that the coding gain of the proposed
TBC code is significant comparing with the block code in the
standard.

5 Simulation results

In this section, we explore the PER performance of the proposed

coding schemes. The transceiver structure is shown in Fig. 1. The
encoder block takes random binary input and generates coded bits.
The coded bits are mapped to constellations with OQPSK
modulation. A half-sine pulse shaping filter is applied in the
transmitter. AWGN channel and Rayleigh flat-fading channel are
utilised in the following simulations. At the receiver side, matched
filter is utilised to maximise the SNR of the received signals.
Perfect synchronisation is assumed. For the block code, the ML
decoder in (7) is applied. For the TBC code, a low-complexity
sub-optimal Viterbi algorithm [21] is utilised for decoding. The
detailed description of the decoding algorithm is listed in Section
4. The system performance is evaluated at PER of 1% with
effective payload length of 20 bytes. The number of simulated
packets is 107 in every SNR condition. The simulation setup is
summarised in Table 3.
In the first simulation, we compare the PER performance of the
(32,4) code used in current standard, the (32,4) code based on

Table 3 Simulation parameters

Parameters Values

modulation OQPSK Fig. 7 PER performance of the proposed codes in AWGN channel
pulse shaping filter half-sine
chip rate 2 Mb/s a PER performance of the proposed codes with respect to SNR per bit in AWGN
payload length 20 bytes channel
number of packets 107 b PER performance of the proposed codes with respect to SNR per chip in AWGN
channel

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

& The Institution of Engineering and Technology 2016 919
Fig. 8 PER performance of the proposed codes in Rayleigh flat-fading Fig. 9 PER performance of the proposed codes for IEEE 802.15.4c
channel standard in AWGN channel

and decoding complexities of the two codes are about the same. The
performance of the (32,6) extended BCH code is about the same as
that of the (32,4) code used in current standard. However, with the
(32,6) extended BCH code, the data rate improves by 50%. The
performance improvement of TBC code is significant. The above
three coding schemes have their specific advantages. The (32,4)
code based on extended BCH code improves performance without
extra complexity. The (32,6) extended BCH code improves the
data rate without receive sensitivity loss. The TBC code with
optimum generator polynomials significantly improves the
performance at the expense of extra decoding complexity.
In the second simulation, we repeat the simulation in Rayleigh
flat-fading channel. The PER performance with respect to SNR per
bit is shown in Fig. 8. The definition of each line in Fig. 8 follows
the definition in the first simulation. We observe that the (32,4)
code based on extended BCH code, the (32,6) extended BCH code
and the TBC code with optimum generator polynomials achieves
about 1.0, 4.0 and 8.0 dB coding gain compared with the (32,4)
code used in current standard, respectively. The performance
improvement in Rayleigh flat-fading channel is even more
significant than that in AWGN channel.
Fig. 10 PER performance of the proposed codes for IEEE 802.15.4c
standard in Rayleigh flat-fading channel

6 Discussion

The purpose of this paper is to provide solutions in improving the code, the (16,5) extended BCH code and the TBC code with
coding gain of LR-WPANs. Our proposed coding schemes can be optimal generator polynomials are 0.4, 0.9 and 4.1 dB, respectively.
applied to other LR-WPANs standards as well. For example, IEEE Similarly, Fig. 10 shows the performance with respect to SNR per
802.15.4c standard has similar coding scheme to IEEE 802.15.4 bit of the (16,4) code used in IEEE 802.15.4c standard, the (16,4)
standard. Every four information bits (u0, u1, u2, u3) are mapped code based on extended BCH code, the (16,5) extended BCH code
to a 16-chip sequence [24]. The minimum Hamming distance of and the TBC code with optimum generator polynomials in
this (16,4) code is dmin = 6. This coding scheme does not fully Rayleigh flat-fading channel. The definition of each line follows
utilise the benefit of the code redundancy either. By adopting the definition in Fig. 7. Compared with the (16,4) code used in
similar coding scheme as shown in Fig. 4, we design a (16,4) code IEEE 802.15.4c standard, the performance gains of the (16,4) code
based on extended BCH code, which achieves the upper bound of based on extended BCH code, the (16,5) extended BCH code and
linear block (16,4) code with dmin = 8. The (16,5) extended BCH the TBC code with optimal generator polynomials are 1.0, 3.0 and
code is found to improve the data rate without performance loss. 8.0 dB, respectively.
In addition, the optimum generator polynomials of the TBC code
in octal format with code rate Rc = 1/4 and constraint length K = 7
are 117, 127, 155, 171 [16]. 7 Conclusion
Fig. 9 shows the performance with respect to SNR per bit of the
(16,4) code used in IEEE 802.15.4c standard, the (16,4) code In LR-WPANs, existing coding techniques do not fully utilise the
based on extended BCH code, the (16,5) extended BCH code and benefit of the code redundancy. In this paper, we propose new
the TBC code with optimum generator polynomials in AWGN coding schemes that achieve the largest minimum Hamming
channel. The simulation setup is similar to that in Section 5. The distance dmin or free distance df at the required code rate. In IEEE
definition of each line follows the definition in Fig. 7. Compared 802.154 system, the proposed (32,4) code achieves about 0.5 dB
with the (16,4) code used in IEEE 802.15.4c standard, the improvement in terms of PER performance in AWGN channel and
performance gains of the (16,4) code based on extended BCH 1.0 dB in Rayleigh flat-fading channel. In addition, the design

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

920 & The Institution of Engineering and Technology 2016
complexities of encoder and decoder remain about the same. The 7 Teng, C., Fonseka, J., Dowling, E.: ‘Non-coherent detectors for
coding gain of the (32,6) extended BCH code is 1.8 dB in AWGN quadrature-multiplexed continuous phase modulation signals’, IET Commun.,
2009, 3, (4), pp. 610–619
channel and 4.0 dB in Rayleigh flat-fading channel. With the 8 Dai, S., Qian, H., Kang, K., et al.: ‘A robust demodulator for OQPSK-DSSS
(32,6) extended BCH code, the data rate can be increased by 50% system’, Circuits Syst. Signal Process., 2015, 34, (1), pp. 231–247
with no performance loss. The TBC code with optimum generator 9 Lanzisera, S., Pister, K.: ‘Theoretical and practical limits to sensitivity in IEEE
polynomials significantly improves the system performance by 4.3 802.15.4 receivers’. Proc. IEEE Int. Conf. on Electronics, Circuits and Systems
dB in AWGN channel and 8.0 dB in Rayleigh flat-fading channel. (ICECS’2007), 2007, pp. 1344–1347
10 Griesmer, J.: ‘A bound for error-correcting codes’, IBM J. Res. Dev., 1960, 4, (5),
The proposed coding schemes can be extended to other pp. 532–542
LR-WPANs standards. 11 Daut, D., Modestino, J., Wismer, L.: ‘New short constraint length convolutional
code constructions for selected rational rates (Corresp.)’, IEEE Trans. Inf.
Theory, 1982, 28, (5), pp. 794–800
8 Acknowledgments 12 Proakis, J.: ‘Digital communications’ (McGraw-Hill, New York, 2007, 5th edn.)
13 ‘Bounds on the minimum distance of linear codes and quantum codes’. Available at
http://www.codetables.de, 2007
This work was supported in part by the National Natural Science 14 Simonis, J.: ‘Adding a parity-check bit’, IEEE Trans. Inf. Theory, 2000, 46, (4),
Foundation of China (no. 61231009), the National Key Science pp. 1544–1545
and Technology 863 Project (no. SS2015AA011303), the Science 15 Lee, S., Kwon, H., Jung, Y., et al.: ‘Efficient non-coherent demodulation scheme
and Technology Commission Foundation of Shanghai (no. for IEEE 802.15.4 LR-WPAN systems’, Electron. Lett., 2007, 43, (16),
pp. 879–880
14511100200), and the State Key Laboratory of Rail Traffic 16 Frenger, P., Orten, P., Ottosson, T.: ‘Code-spread CDMA using maximum free
Control and Safety (no. RCS2014K004), Beijing Jiaotong distance low-rate convolutional codes’, IEEE Trans. Commun., 2000, 48, (1),
University. pp. 135–144
17 Qian, H., Wang, X., Kang, K.: ‘A depth-first ML decoding algorithm for tail-biting
trellises’, IEEE Trans. Veh. Technol., 2014, 64, (8), pp. 3339–3346
18 Pai, H., Han, Y., Wu, T.: ‘Low-complexity ML decoding for convolutional
9 References tail-biting codes’, IEEE Commun. Lett., 2008, 12, (12), pp. 883–885
19 Wang, X., Qian, H., Xiang, W.: ‘An efficient ML decoder for tail-biting codes
1 Kranenburg, R., Anzelmo, E., Bassi, A., et al.: ‘The Internet of things’. Proc. First based on circular trap detection’, IEEE Trans. Commun., 2013, 61, (4),
Berlin Symp. on Internet and Society, October 2011, pp. 25–27 pp. 1212–1221
2 IEEE Std. 802.15.4-2006: ‘IEEE standard for information technology – part 15.4:
20 Shao, R., Lin, S., Fossorier, M.: ‘Two decoding algorithms for tailbiting codes’,
wireless medium access control (MAC) and physical layer (PHY) specifications for
IEEE Trans. Commun., 2003, 51, (10), pp. 1658–1665
low-rate wireless personal area networks (WPANs)’, 2006
21 Sung, W.: ‘Minimum decoding trellis lengths for tail-biting convolutional codes’,
3 Wang, C., Huang, J.: ‘A low-power 2.45 GHz ZigBee transceiver for wearable
personal medical devices in WPAN’. Proc. IEEE Int. Conf. on Consumer Electron. Lett., 2000, 36, (7), pp. 643–645
Electronics (ICCE’2007), 2007, pp. 1–2 22 IEEE Std. 802.16-2009: ‘IEEE standard for local and metropolitan area networks –
4 Wang, C., Sung, G., Huang, J., et al.: ‘A low-power 2.45 GHz WPAN modulator/ part 16: air interface for broadband wireless access systems’, 2009
demodulator’, Microelectron. J., 2010, 41, (2), pp. 150–154 23 3GPP TS 36.212: ‘3rd generation partnership project; technical specification group
5 Yu, Y., Meng, X., Xiao, S., et al.: ‘A new low-cost demodulator for 2.4 GHz radio access network; evolved universal terrestrial radio access (E-UTRA);
ZigBee receivers’, J. Electron. (China), 2009, 26, (2), pp. 252–257 multiplexing and channel coding (Release 8)’, 2007
6 Yin, S., Cui, J., Luo, A., et al.: ‘A high efficient baseband transceiver for IEEE 24 IEEE Std. 802.15.4c-2009: ‘IEEE standard for information technology – part 15.4:
802.15.4 LR-WPAN systems’. Proc. IEEE Ninth Int. Conf. on ASIC, 2011, wireless medium access control (MAC) and physical layer (PHY) specifications for
pp. 224–227 low-rate wireless personal area networks (WPANs)’, 2009

IET Commun., 2016, Vol. 10, Iss. 8, pp. 915–921

& The Institution of Engineering and Technology 2016 921