Anda di halaman 1dari 5

This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts

for publication in the WCNC 2008 proceedings.

Hardware Impairments on LDPC coded SC-FDE


and OFDM in Multi-Gbps WPAN (IEEE 802.15.3c)
Ming Lei, Ismail Lakkis∗ , Chin-Sean Sum, Tuncer Baykas, Jun-Yi Wang,
M. A. Rahman, R. Kimura, R. Funada, Y. Shoji, Hiroshi Harada, and Shuzo Kato

Ubiquitous Mobile Communications Group (UMCG)


National Institute of Information and Communications Technology (NICT)
YRP-1 Bldg., 3-4 Hikari-no-oka, Yokosuka 239-0847 Japan
E-mail: minglei@nict.go.jp / minglei8998@hotmail.com

Tensorcom Inc., 10875 Rancho Bernardo Rd. 108, San Diego, CA, USA

Abstract— The Multi-Gbps wireless personal area network advantages such as lower peak-to-average power ratio (PAPR)
(WPAN) using 60-GHz is being standardized in the task group which we will elaborate on later in this paper.
of IEEE 802.15.3c. In the physical (PHY) layer design, there
For choosing the channel coding scheme, the complexity
are two competitive techniques, one is orthogonal frequency
division multiplexing (OFDM) and the other is single-carrier is a priority. It is difficult for Viterbi or Turbo decoding to
frequency-domain equalization (SC-FDE), a.k.a., single-carrier support the target data rates of Multi-Gbps due to its high
block transmission (SCBT). This paper compared the perfor- complexity. Therefore, we mainly considered the systematic
mance of low-density parity-check (LDPC) coded OFDM and block codes, such as Reed-Solomon (RS) and low-density
SC-FDE under hardware impairments including analog-to-digital
converter (ADC) resolution, non-linear distortion induced by 60-
parity-check (LDPC) codes. LDPC and the iterative decoding
GHz power amplifier (PA), and phase noise of 60-GHz PLL algorithm were first proposed by Gallager [4]. However they
circuits. The simulation results show that SC-FDE has much had been silent for more than three decades until MacKay [5]
better performance than OFDM under hardware impairments rediscovered them. Recently, intensive research work has been
of 60-GHz devices and circuits. dedicated to LDPC due to its excellent performance. It has
Index Terms— Multi-Gbps, WPAN, PHY, OFDM, SC-FDE,
been used in IEEE 802.11n [6] and 802.16 [7] standards.
SCBT, LDPC, ADC resolution, non-linear distortion, phase noise,
60-GHz, IEEE 802.15.3c. Compared with the convolutional or Turbo coding, LDPC
shows
I. I NTRODUCTION • stronger error correction capability;
• lower decoding complexity;
Responding to the ever increasing market demands for
• more flexible scalability.
ultra high-data-rate indoor wireless applications such as un-
compressed high-definition video streaming and flash file- Therefore, we considered LDPC as the high-performance error
downloading, the millimeter-wave 60-GHz wireless personal correction coding in PHY design for IEEE 802.15.3c.
area network (WPAN) is being standardized by the IEEE The latest PHY layer design for Multi-Gbps WPAN (IEEE
802.15 Alternative PHY Task Group 3c (IEEE 802.15.3c) [1]. 802.15.3c) can be found in [8][9]. We compared the perfor-
This emerging IEEE 802.15.3c standard targets the data rates mance of PHY layer designs based on SC-FDE and OFDM
of multi-giga bits per second (Multi-Gbps) for short-range in [10]. In [11], we evaluated the MMSE-FDE based on
indoor applications. estimated SNR for Multi-Gbps WPAN.
In the physical (PHY) layer design, there are two compet- However, the performance of SC-FDE and OFDM were not
itive techniques: one is orthogonal frequency division mul- systematically compared under hardware impairments, such
tiplexing (OFDM) and the other is single-carrier frequency- as analog-to-digital converter (ADC) resolution, non-linear
domain equalization (SC-FDE), a.k.a., single-carrier block distortion induced by 60-GHz power amplifier (PA), and phase
transmission (SCBT) [2][3]. OFDM has been a popular tech- noise of 60-GHz PLL circuits. We accomplished this work in
nique for many years and adopted as the core technique in this paper. We followed the Selection Criteria [12] of IEEE
a number of wireless standards. However, using OFDM over 802.15.3c to conduct the computer simulations. The results
the wide bandwidth (several GHz) at 60-GHz could encounter show that SC-FDE has much better performance than OFDM
serious difficulties. In contrast to OFDM, SC-FDE has many under hardware impairments.
The remainder of this paper is organized as follows. In
This work was supported by the Millimeter-Wave 60-GHz WPAN (IEEE Section II, we briefly introduce SC-FDE and OFDM in PHY
802.15.3c) Project at the National Institute of Information and Communi-
cations Technology (NICT), which was funded by the Ministry of Internal layer design for Multi-Gbps WPAN; the hardware impairments
Affairs and Communications (MIC), Japan. and their models are introduced in Section III; in Section IV,

442
1525-3511/08/$25.00 ©2008 IEEE
Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 07:04 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2008 proceedings.

we show the simulation results; finally, we conclude this paper to the ADC resolution. B should be determined based on the
in Section V. compromise among decoding performance, power consump-
tion and complexity.
II. SC-FDE VS . OFDM IN PHY L AYER D ESIGN FOR
M ULTI -G BPS WPAN B. Non-linear distortion
In Fig.1, we show the block diagrams of OFDM and SC- The non-linear distortion is mainly caused by the power
FDE. SC-FDE is very similar to OFDM since both of them amplifier. Each kind of power amplifier has a linear zone. If
perform channel equalization in the frequency domain. Their the signal amplitude goes beyond the linear zone, it will be
difference is that the data detection is performed in frequency distorted. To avoid this distortion, the power amplifier output
domain in OFDM, while it is performed in time domain in power should be backed off to keep the signal amplitude within
SC-FDE. the linear zone as much as possible. We denote the PA output
It was in [2][3] that Sari et al. first compared these two power back-off (OBO) by β in this work.
techniques, OFDM and SC-FDE, in multi-path fading chan- The non-linear distortion is composed of AM-AM and AM-
nels. In comparison with OFDM, SC-FDE has the following PM effects. By using the modified Rapp model [14], the AM-
advantages. AM effect of a 60-GHz IBM SiGe BiCMOS power amplifier
• Lower peak-to-average power ratio (PAPR); (PA) can be modeled by
• Less sensitive to carrier frequency offset (CFO); Vin
FAM −AM (Vout ) = 1 ,
• More robust against frequency-selective fading;
(1 + (Vin /Vsat )2p ) 2p
• Lower dependence on channel coding;
• Lower dependence on interleaving;
where, Vin and Vout are the PA input and output signal voltage,
• Longer transmission range;
respectively. Vsat =2.09V is the saturation voltage level and the
• Allow the use of less expensive power amplifier and less-
coefficient p is 1.6. The AM-PM transform function of the
resolution ADC, etc.. modified Rapp model [14] is
q
SC-FDE is superior over OFDM in many aspects. Moreover, AVin
FAM −P M (θ) = ,
SC (single carrier) is a hardware-friendly technology which is 1 + (Vin /B)q
already matured in algorithms, circuits and systems. Therefore, where, A is -10250, B is 0.0554, and q is 3.5.
the PHY layer design based on SC-FDE is very attractive for
Multi-Gbps WPAN (IEEE 802.15.3c). C. Phase noise
In PHY layer design, we mainly used the popular modu- The phase noise is mainly caused by carrier frequency
lations such as BPSK, QPSK and 8PSK. The channel coding offset (CFO). For OFDM system, it results in inter-carrier
we considered is the low complexity but good-performance interference (ICI) and common phase error (CPE). For SC-
Reed-Solomon codes, RS(255,239,8), and LDPC codes of FDE system, it leads to phase rotation and inter-symbol
different block lengths [8][9]. A rate-7/8 irregular LDPC interference (ISI).
code, LDPC(1152,1008), was used in this work. The parity The phase noise is modeled by
check matrix of this LDPC code is constructed by structured
1 + (f /fz )2
design [6], i.e., the parity check matrix can be partitioned P SD(f ) = P SD(0) ,
into square subblocks (submatrices) and these submatrices are 1 + (f /fp )2
either cyclic-permutations of the identity matrix or null (all- where, the pole frequency fp is 1MHz, and the zero frequency
zero) submatrices. fz is 100MHz [14]. In our work, we denote P SD(0) as θ.
As this paper was written, the standardization of IEEE
IV. S IMULATION R ESULTS
802.15.3c was ongoing, and the PHY parameters were being
subjected to change. The details of the latest progress in PHY The system and simulation parameters are shown in Table I.
layer design, such as channelization, PHY frame structure, By using QPSK, structured LDPC(1152,1008) of rate-7/8, FFT
PHY-SAP data rates, can be found in [8][9]. of 256 and GI of 64 at the symbol rate of 1.632GHz, we
can achieve the data rate of 2.285Gbps at PHY-SAP [9]. The
III. H ARDWARE I MPAIRMENTS synchronization and channel estimation were assumed to be
In actual systems, the link performance will be degraded ideal. FDE was based on MMSE algorithm. The output back-
mainly by following three kinds of hardware impairments. off (β) of power amplifier was 3.0∼5.0dB and the phase
noise (θ) was -96∼-87dBc/Hz@1MHz. Each data packet was
A. ADC resolution loaded with 2K (2048) raw bytes. The other parameters not
The ADC resolution should be kept as small as possible mentioned in Table I followed the Section Criteria [12] of
because increasing of one bit resolution approximately double IEEE 802.15.3c.
the power consumed at ADC. On the other side, ADC resolu- LDPC can be easily applied to SC-FDE system since
tion is critical to the performance of LDPC or Viterbi decoding the calculation of LLR is direct. However, in order to use
since it decides the accuracy of quantized log-likelihood ratio LDPC for OFDM system, we need to consider an additional
(LLR). The quantization bit number (B) is usually equivalent procedure, LLR weighting. This is due to that in OFDM

443

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 07:04 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2008 proceedings.

Fig. 1. Block diagrams of OFDM and SC-FDE.

TABLE I
0
PARAMETERS OF S IMULATIONS 10

Channel model NLOS residential [13] −1


10
(RMS delay=6.24ns)
Symbol rate 1.632GHz −2
10
(Nyquist bandwith)
Roll-off factor 0.3235 −3
10

BER
Channel bandwidth 2.16GHz
Modulation QPSK −4
10
Channel coding Structured LDPC(1152,1008), R=7/8
FFT length (time) 256 (156.86ns) −5
10
GI length (time) 64 (39.22ns)
F−on, W−on
PHY-SAP data rate 2.285Gbps −6
10 F−on, W−off
Synchronization Ideal F−off, W−off
F−off, W−on
Channel estimation Ideal −7
10
FDE MMSE-FDE 0 2 4 6 8 10 12 14 16
E /N [dB]
b 0
Power amplifier model IBM SiGe BiCMOS PA [14]
Output back-off 3.0∼5.0dB
Fig. 2. The effects of LLR weighting (W) and frequency-domain bit
(β = Psat − Pout )
interleaving (F) on BER performance of LDPC coded OFDM (Z = 3σ, B =
Phase noise (θ) -96∼-87dBc/Hz@1MHz 8).
Packet size 2048 bytes (16384 bits)

simulations.
system LLR is generated from the frequency-domain symbols • A. Search and determine the appropriate quantization
and thus is affected by the frequency-selective fading. To zone width (Z) for LLR;
compensate for this fading, LLR must be weighted by the • B. Search and determine the appropriate quantization bit
square module of the channel frequency response (CFR). number (B)1 based on the determined Z;
In Fig.2, we show the effects of using LLR weighting (W) • C. Evaluate the effects of PA output back-off (β) on BER
and frequency-domain bit interleaving (F) functions on BER performance with the determined Z and B;
performance of LDPC coded OFDM. The quantization zone • D. Evaluate the effects of phase noise (θ) on BER
width (Z) is set to an experiential value of 3σ; the quantization performance with the determined Z and B.
bit number (B) is set to 8 which is adequate to make the
quantization noise undetectable. As we can observe from the A. Search and determine the appropriate quantization zone
results, the frequency-domain bit interleaving is not necessary width (Z)
since the bit interleaving / de-interleaving is already embedded The quantization zone width (Z) is measured by σ where
in LDPC coding / decoding procedures. From Fig.2, we can σ is square root of the variance of the LLR values. The LLR
verify that LLR weighting is indispensable for LDPC coded value outside the quantization zone will be clipped. To avoid
OFDM system. Based on these results, we turned on LLR the influence of quantization bit number (B), we set B to 8
weighting and turned off frequency-domain bit interleaving in which is adequate to make the quantization noise undetectable.
the follow-up simulations. For SC-FDE system, neither LLR In Figs.3 and 4, we show the effects of quantization zone
weighting nor frequency-domain bit interleaving is necessary. width (Z) on BER performance of LDPC coded SC-FDE and
OFDM, respectively. The optimal Z for SC-FDE is 3.0σ while
To evaluate the hardware impairments on the link per-
formance, we follow the procedures below to conduct the 1 The quantization bit number is equivalent to the ADC resolution.

444

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 07:04 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2008 proceedings.

0 0
10 10

−1
10
−2
10 −2
10

−3
10
BER

BER
−4
10
−4
10
Qnt. zone width (Z): 0.5σ
Qnt. zone width (Z): 1.0σ Qnt. bit number (B): 2
−5
10 Qnt. bit number (B): 3
−6 Qnt. zone width (Z): 1.5σ
10 Qnt. bit number (B): 4
Qnt. zone width (Z): 2.0σ
Qnt. bit number (B): 5
Qnt. zone width (Z): 2.5σ −6
Qnt. bit number (B): 6
10
Qnt. zone width (Z): 3.0σ Qnt. bit number (B): 7
Qnt. zone width (Z): 3.5σ Qnt. bit number (B): 8
−8 −7
10 10
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
E /N [dB] E /N [dB]
b 0 b 0

Fig. 3. The effects of quantization zone width (Z) on BER performance of Fig. 5. The effects of quantization bit number (B) on BER performance of
LDPC coded SC-FDE (B = 8). LDPC coded SC-FDE (Z = 3σ).
0 0
10 10

−1 −1
10 10

−2 −2
10 10

−3 −3
10 10
BER

BER

−4 Qnt. zone width (Z): 1.0σ −4


10 10
Qnt. zone width (Z): 2.0σ
Qnt. zone width (Z): 3.0σ Qnt. bit number (B): 2
−5
10
−5
10 Qnt. bit number (B): 3
Qnt. zone width (Z): 4.0σ
Qnt. bit number (B): 4
Qnt. zone width (Z): 4.5σ
Qnt. bit number (B): 5
−6 Qnt. zone width (Z): 5.5σ −6
Qnt. bit number (B): 6
10 10
Qnt. zone width (Z): 6.0σ Qnt. bit number (B): 7
Qnt. zone width (Z): 7.0σ Qnt. bit number (B): 8
−7 −7
10 10
0 2 4 6 8 10 12 14 16 0 2 4 6 8 10 12 14 16
E /N [dB] E /N [dB]
b 0 b 0

Fig. 4. The effects of quantization zone width (Z) on BER performance of Fig. 6. The effects of quantization bit number (B) on BER performance of
LDPC coded OFDM (B = 8). LDPC coded OFDM (Z = 4σ).

the optimal one for OFDM is 4.0σ. This also proves OFDM We set B to 5 and 7 for SC-FDE and OFDM, respectively,
signal has larger PAPR than that of SC-FDE. in follow-up simulations.
We set Z to 3.0σ and 4.0σ for SC-FDE and OFDM, C. Evaluate the effects of PA output back-off (β) on BER
respectively, in follow-up simulations. performance
B. Search and determine the appropriate quantization bit In Fig.7, we show the effects of PA output back-off (β) on
number (B) BER performance of SC-FDE and OFDM. For comparison, we
also show the performance without non-linear distortion (linear
In Figs.5 and 6, we show the effects of quantization bit PA). The performance of OFDM is vulnerable to non-linear
number (B) on BER performance of LDPC coded SC-FDE distortion due to its high signal PAPR. In contrast, SC-FDE
and OFDM, respectively. The appropriate value of B should is very robust against non-linear distortion. This unbeatable
effectively remove the error floor and enable the system to advantage allows SC-FDE system to use less expensive power
approach the optimal performance. We observed the appropri- amplifier. Thus, the overall system cost can be reduced.
ate value of B is 5 for SC-FDE. The B value higher than
5 does not lead to significant performance improvement for D. Evaluate the effects of phase noise (θ) on BER performance
SC-FDE. For OFDM, the appropriate value of B is 7. The In Fig.8, we show the effects of phase noise (θ) on BER
OFDM signal has larger dynamic range than that of SC-FDE, performance of SC-FDE and OFDM. The phase noise does
therefore it needs more quantization bit at ADC. not significantly degrade the performance of either SC-FDE

445

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 07:04 from IEEE Xplore. Restrictions apply.
This full text paper was peer reviewed at the direction of IEEE Communications Society subject matter experts for publication in the WCNC 2008 proceedings.

0
10
V. C ONCLUSIONS
This paper compared the performance of LDPC coded
OFDM and SC-FDE under hardware impairments including
−2
10
ADC resolution, non-linear distortion induced by 60-GHz
power amplifier, and phase noise of 60-GHz PLL circuits.
The simulation results show that SC-FDE has much better
performance than OFDM under these hardware impairments.
BER

−4
10
Therefore, SC-FDE is a very promising technique for the
SC−FDE (Linear PA)
SC−FDE (β=3.0dB) emerging Multi-Gbps WPAN (IEEE 802.15.3c) standard.
SC−FDE (β=4.0dB)
−6
10
SC−FDE (β=5.0dB) ACKNOWLEDGMENT
OFDM (Linear PA)
OFDM (β=3.0dB) The authors would like to thank Dr. Makoto Noda and Dr.
OFDM (β=4.0dB) Hiroyuki Yamagishi, SONY Corporation, Japan, for helpful
OFDM (β=5.0dB)
−8
10
discussions. The authors would also like to thank the col-
0 2 4 6 8 10 12 14 16 leagues at the National Institute of Information and Com-
E /N [dB]
b 0
munications Technology (NICT) and the Consortium for the
Practical Application of Millimeter-Wave (COMPA)2 for their
Fig. 7. The effects of PA output back-off (β) on BER performance of LDPC
coded SC-FDE (Z = 3σ, B = 5) and OFDM (Z = 4σ, B = 7). consistent support and encouragement.
0
10 R EFERENCES
[1] http://www.ieee802.org/15/pub/TG3c.html
−1
10 [2] H. Sari, G. Karam, and I. Jeanclaud, “Frequency-domain equalization
of mobile radio and terrestrial broadcast channels,” in Proc. IEEE
−2 GLOBECOM 1994, vol. 1, pp. 1–5, November 1994.
10
[3] H. Sari, G. Karam, and I. Jeanclaude, “Transmission techniques for
digital terrestrial TV broadcasting,” IEEE Commun. Magazine, vol.
−3
10 33, pp. 100–109, Feb. 1995.
BER

[4] R. G. Gallagar, Low-Density Parity-Check Codes, Cambridge, MA:


−4
SC−FDE (No phase noise)
MIT Press, 1963.
10
SC−FDE (θ=−96dBc/Hz) [5] D. J. C. MacKay, “Good error-correcting codes based on very sparse
SC−FDE (θ=−90dBc/Hz) matrices,” IEEE Trans. Inform. Theory, vol. 45, no. 2, pp. 399–431,
−5
10 SC−FDE (θ=−87dBc/Hz) March 1999.
OFDM (No phase noise) [6] “Draft Standard for local and metropolitan area networks - specific
−6 OFDM (θ=−96dBc/Hz) requirements- Part 11: Wireless LAN Medium Access Control (MAC)
10
OFDM (θ=−90dBc/Hz) and Physical Layer (PHY) specifications,” IEEE P802.11n/D2.00,
OFDM (θ=−87dBc/Hz) February 2007.
−7
10 [7] “IEEE Standard for LAN/MAN Part 16: Air Interface for Fixed and
0 2 4 6 8 10 12 14 16 Mobile Broadband Wireless Access Systems,” IEEE Std 802.16e-2005
E /N [dB]
b 0 and IEEE Std 802.16-2004/Cor 1-2005 (Amendment and Corrigen-
dum to IEEE Std 802.16-2004), February 2006.
Fig. 8. The effects of phase noise (θ) on BER performance of LDPC coded [8] H. Harada et al., “CoMPA PHY proposal,” IEEE 802.15-07-0693-03-
SC-FDE (Z = 3σ, B = 5) and OFDM (Z = 4σ, B = 7). 003c, May 2007.
[9] H. Harada et al., “Unified and flexible millimeter wave WPAN
systems supported by common mode,” IEEE 802.15-07-0761-00-
003c, July 2007.
or OFDM when it is not higher than -87dBc/Hz@1MHz. [10] M. Lei, C.-S. Choi, R. Funada, H. Harada, and S. Kato, “Through-
Moreover, it seems OFDM is less sensitive to phase noise put comparison of multi-Gbps WPAN (IEEE 802.15.3c) PHY layer
designs under non-linear 60-GHz power amplifier,” in Proc. IEEE
than SC-FDE. This is due to that we use higher number of PIMRC 2007, Sept. 2007.
quantization bit in OFDM (B = 7) than in SC-FDE (B = 5). [11] M. Lei, R. Kimura, C.-S. Sum, R. Funada, Y. Shoji, H. Harada, and
In other words, the ADC power consumption in OFDM is S. Kato, “MMSE-FDE based on estimated SNR for single-carrier
block transmission (SCBT) in multi-Gbps WPAN,” submitted to IEEE
approximately four times of that in SC-FDE. WCNC 2007, Sept. 2007.
The simulation results show that SC-FDE overwhelmingly [12] A. Seyedi, “TG3c Selection Criteria,” IEEE 802.15-05-0493-26-003c,
outperforms OFDM under hardware impairments, even we July, 2007.
[13] S.-K. Yong, “TG3c channel modeling sub-committee final report,”
loose the conditions for OFDM: IEEE 802.15-06-0195-04-003c, October 2006.
• OFDM needs two more quantization bits than SC-FDE, [14] C.-S. Choi et al., “RF impairment models for 60GHz-band SYS/PHY
simulation,” IEEE 802.15-06-0477-01-003c, November, 2006.
which results in approximately 4 times of power con-
sumption at ADC;
• OFDM needs LLR weighting which leads to additional
complexity.
Therefore, SC-FDE is the more promising technique in sup- 2 In July 2006, the academic research institutes and industrial companies
porting Multi-Gbps transmission in 60-GHz WPAN (IEEE in Japan established a unified consortium body, COMPA, to coordinate the
802.15.3c). contributions to IEEE 802.15.3c.

446

Authorized licensed use limited to: VELLORE INSTITUTE OF TECHNOLOGY. Downloaded on August 2, 2009 at 07:04 from IEEE Xplore. Restrictions apply.

Anda mungkin juga menyukai