You are on page 1of 6

2011 6th International ICST Conference on Communications and Networking in China (CHINACOM)

Fully Reconfigurable Cognitive OFDMA Testbed


Lianxiang Wang1,2 , Zhaoyang Zhang1,2 , Lu Ye1,2 , Benkang Gong1,2 ,Wentao Lou1,2
1. Department of information science and electronic engineering, Zhejiang University, Hangzhou 310027,China.
2. Zhejiang provincial key laboratory of information network technology, Hangzhou 310027, China.
Email: ning ming@zju.edu.cn

Abstract— In this paper, we report our design and imple- as FTP and so on. Or we can directly read file from the
mentation of a testbed for demonstration of cognitive OFDMA hard disk and wrap it into a frame in Sora program. Sora
transmission. Our implementation is based on a powerful fully program is organized as an Ethernet driver under the NDIS
programmable software radio platform, i.e., the Microsoft Re-
search Software Radio(MRSR) platform. Two kinds of nodes framework. The driver implements the OSI lower three layers,
are designed in the testbed, one acts as the primary user(PU) i.e. link layer, MAC layer and PHY layer(including Base
which communicates in the authorized frequency band with Band). In these layers, we can do all the work needed in
dynamic spectrum activity, and the other acts as the secondary our communication protocol without any dependence of the
user(SU) which opportunistically accesses the vacant frequency hardware.
without disrupting the transmission of primary user. To this end,
the SU nodes are equipped with reliable multi-carrier sensing
functionality, as well as some special multi-channel signalling and
processing capabilities. Test results show that the SU nodes can
alter their power, frequency and other parameters dynamically
and communicate efficiently in response to the PU’s spectrum
activity.
Index Terms— reconfigurable; cognitive; OFDMA; testbed;
sensing; aic

I. I NTRODUCTION
Cognitive radio (CR)[1][2] is a wireless technology based Fig. 1. Sora system architecture
on software defined radio to achieve the spectrum sharing
of licensed system and cognitive radio system. Orthogonal
In our system, the cognitive users can find out the vacant
frequency division multiple access (OFDMA) is a multi-
frequency band by frequency detection function, then make
carrier technology that can offer high data rates. Because
a channel assessment of the available band, through the
of its scalable and adaptive feature for air interface and
reasonable usage of the idle band segment to improve the
carrier aggregation, the combination of OFDMA and cognitive
frequency spectrum utilization as shown in Fig.2.
radio has been a promising technology in the future wireless
communication[3][4]. In this paper, we will introduce a testbed Used by SUs Occupied by PUs Used by SUs
of cognitive OFDMA system based on Sora platform.
Sora [5], a software radio platform designed by Microsoft
Research Asia(MSRA), includs hardware and software de- ĂĂ ĂĂ
velop kit as shown in Fig.1. The hardware consists of a high f

performance general computer, a radio control board(RCB)


designed by MSRA, an adapter, a radio board(we use USRP Fig. 2. Carrier occupation form for SUs
daughterboard) and an antenna. RCB is connected to the
computer by high-speed PCIe interface which could reach The remainder of the paper is organized as follows. Section
several Gbps throughput. Through DMA, RCB reads digital II presents the introduction of the system as well as some
samples , which have been modulated into frames but are important modules. The test results are shown in section III.
not synchronized, from the computer memory. The embedded Finally, section IV concludes this paper.
FPGA in the RCB synchronizes the frames and sends them
II. I NTRODUCTION OF THE SYSTEM
to the radio board. Then the digital frames are converted into
analog waves and sent to the air. Usually,the Sora platform In our platform, there are three kinds of nodes. The first one
is implemented as a virtual network card. Source data could imitates the PU transmitter to transmit OFDM signals, while
be retrieved from all existing networking application such the second one and the third one imitate the SU transmitter and
receiver respectively. The signal interactive process of these
This work was supported by the National Hi-Tech Research and Develop- three nodes is shown in Fig.3.
ment Program (863 program) of China (No. 2007AA01Z257), National Basic
Research Program (973 Program) of China (No. 2009CB320405) and National Fig.4 and Fig.5 describe detailedly the worker threads of
Natural Science Foundation Program of China (No. 60802012, 60972057). the SU transmitter and SU receiver. The SU transmitter and
978-1-4577-0101-6/11/$26.00⃝2011
c IEEE

933 978-1-4577-0101-6/11/$26.00 © 2011 IEEE











 

 

  



 



  


 
  


  



Fig. 5. Worker threads for receiver
 

Sensing Data transmiting Sensing Data transmiting




 
Sensing Data receiving Sensing Data receiving

Fig. 6. Sensing and data processing cycle


Fig. 3. Signal interactive process

A. sensing
the SU receiver periodically do sensing and data processing as Sensing is a key function of SUs to identify the vacant fre-
shown in Fig. 6. As the synchronization for the transmitter and quency band. There are three classical detection methods that
the receiver is not easy to realize in our testbed, we choose can be used in frequency detection, i.e., coherent detection,
the transmitter to be an active controller and the receiver to cyclostationary detection and energy detection [6][7][8].The
be a passive follower in the synchronization processing. The first two methods need some prior information of PUs. How-
transmitter uses a timer to control the sensing cycle. If the ever, in our system, the SUs have no pre-knowledge of the
receiver can not capture the frame, it will come to the state PUs, hence, we choose energy detection as the main method
of sensing. In the following context, we will give a detailed to sense the activities of PUs. The traditional energy detector
description for the important parts in the worker threads. can not differentiate the interference and noise from the PUs’
signal energy, so we choose the improved detector algorithm
proposed in [9] to realize the carrier sensing. In [9] the energy
emission of the detected adjacent carriers got by conservative
pre-decision is gradually removed from the carrier to be
detected. Because of the space limit, we can not give a detail in
this paper. More information about this algorithm is presented
in [9]. The calculation of terminate threshold 𝜆1 in [9] as
shown in (1) includes the interference and noise energy density
𝜎0 . ( −1 )
𝑄 (𝑃𝑓 𝑎 )
𝜆1 = √ + 1 𝜎02 (1)
𝑁
The obtaining of 𝜎0 is according to the statistics of envi-
ronment interference and noise, and we update it using each
sensing result. As long as the PUs do not use all the authorized
frequency band, there always exists some vacant carriers that
can be identified in each sensing period. Then we can use
Fig. 4. Worker threads for transmitter the energy of these vacant carriers to update the value of 𝜆1 ,
which can better adapt to the non-stationary of the interference
and noise. Fig.7 is the a kind of frequency occupation form

934
of PUs got from a wireless spectrum analyzer produced by during the subcarrier allocation process. The function of AIC
ROHDE& SCHWARZ . It can scan the spectrum and do real- is to cancel the interference caused by the sidelobe energy
time analysis. As the picture shown, the vertical marker in the of SUs to the PUs, so as to shield the existence of SUs for
middle represents the central frequency 2.422GHz that is seen the PUs as much as possible. Detailed introduction about AIC
from the top left corner. The whole frequency band, ranging algorithm will be given in the last part of this section.Due to
from 2.407GHz to 2.437GHz seen from the bottom has been the dynamical changing of the vacant carrier location, in our
divided into 10 grids and every grid spans 3MHz. So, from the system, we use time-frequency two-dimensional pilot to make
picture, we can see that the PUs have occupied about 10MHz real-time estimation of the channel fading. Then after padding
bandwidth distributed on the two sides of the center frequency. 0 on the carriers that are not used by the SU and making
Fig.8 is the sensing result of SUs, in which ‘1’ indicates the IFFT(Inverse Fast Fourier Transform), we obtain a symbol
carrier is occupied by the PUs, while ‘0’ means the carrier is in time-domain. In order to eliminate the ISI(Inter-Symbol
vacant. The sensing result agrees quite well with the frequency Interference) caused by the delay of multi-channel, we cut
occupation of PUs in Fig.7. the last 1/4 length of the time-domain symbol as cyclic prefix
and add it into the start of the symbol. By now a complete
symbol has been obtained. In order to assemble a frame, we
insert some extra parts to assistant the transmission of the
data package, which include the short preamble, long preamble
and frame header. The short preamble is designed to be used
for frame capture and symbol synchronization, while the long
preamble is mainly used for frequency offset estimation and
first channel fading estimation. The function of frame header
is to carry some modulation information to help the receiver
demodulate the data package.
The design of short preamble and long preamble can refer
to the protocol of 802.11a, which is used for a fixed carrier
allocation scene and the number of IFFT is 64. In our system,
the carrier allocation is dynamically changing, and the IFFT
number is 256 instead of 64. But we find that for any length
of IFFT, as long as we allocate the preamble symbol every 4
Fig. 7. Frequency occupation of PUs getting from instrument carriers, we always can get the cycle time sequence with period
of 𝑁/4, as shown in Fig.10. Furthermore, in order√to adapt
1.4
different available carrier number, we use parameter 𝑁𝑁𝑆𝑈 to
normalize the power on each carrier, which 𝑁 denotes the total
1.2 number of carriers, and 𝑁𝑆𝑈 denotes the number of carriers
used by the SU.
1
Sensing Result

Frame
Header
0.8
Short
Preamble
up- Add
0.6 RF Tx D/A Filtering
conversion Prefix
Long
Preamble
0.4 MACPDU Rateless Padding 0
Interleave mapping
packing Coding and IFFT

0.2
Pilot AIC

0
−60 −40 −20 0 20 40 60
Carrier Index n Fig. 9. The process flow for the transmitter

Fig. 8. Sensing result of frequency occupation of PUs Fig.11 is the spectrum image of the PUs and SUs sharing
the authorized band. The left band and the right band are used
by the PUs, which is the same as Fig.7. Then the two bands
B. Transmitter process flow in the middle of the screen are the spectrum of SUs. From the
The flow of data processing for the transmitter has been spectrum we can see that the SUs properly use the sensed idle
shown in Fig.(9). The data packages that are packaged in carriers to transmit data.
MAC layer are transferred into physical layer to make some AIC algorithm is a feature part in the system design, which
process before forming an OFDM symbol, including rateless is very important for the protection of PUs. Although the CUs
coding[10], interleave, and mapping process. Then the pilot do not use the carriers that occupied by Pus, the interference
and active interference cancellation (AIC) will be inserted caused by the sidelobe emission is still inevitable as shown in

935
0.16

0.14

0.12
'f : carrier interval
'f
CorrValue

0.1 PSD

0.08

0.06

0.04

3.5 2.5 1.5 0.5 0.5 1.5 2.5 3.5


0.02
f
'f
0
50 100 150 200 250
Carrier Index Fig. 12. Interference

Fig. 10. The correlation property of the short preamble


each side of [𝑣1 : 𝑣2]. So the active interference cancelation
carriers locate in [𝑣1−𝑐 : 𝑣1−1] and [𝑣2+1 : 𝑣2+𝑐]. Suppose
the interference generated by data transmitting carriers of SUs
in [𝑣1 : 𝑣2] is 𝑄𝐼 𝑋𝐼 , and the interference generated by the
active interference cancelation carriers is 𝑄𝐶 𝑋𝐶 . Then the
sum interference is
2
𝐼 = ∥𝑄𝐶 𝑋𝐶 + 𝑄𝐼 𝑋𝐼 ∥ (5)

The objective of active interference cancelation is to mini-


mize the sum interference in [𝑣1 : 𝑣2] as
2
𝑋𝐶 = arg min ∥𝑄𝐶 𝑋𝐶 + 𝑄𝐼 𝑋𝐼 ∥ (6)
𝑋𝐶

We can use the Least-Square method [12][13] to solve (6),


and get the solution as:
Fig. 11. Spectrum shared by PU and SU 𝑋𝐶 = −(𝑄𝐻
𝐶 𝑄𝐶 )
−1 𝐻
𝑄𝐶 𝑄𝐼 𝑋𝐼 (7)

Fig.13 is the performance comparison of with/without AIC


Fig.12. algorithm. The data modulated on the carriers for AIC is 𝑋𝐶
Suppose 𝑋 = [𝑋0 , 𝑥1 , ...𝑋𝑁 −1 ] denotes the vector of as shown in (7) in the case of with AIC algorithm, while is
OFDM system in frequency domain, then the system after zero in the case of without AIC algorithm. From Fig.13 we
IFFT transformation can be expressed as can see that the AIC algorithm can reduce the interference
1 ∗ generated by the SUS to PUs at least 10dB.
𝐹 𝑋
𝑥= (2)
𝑁 𝑁 ×𝑁
where 𝑁 is the length of IFFT, (⋅)∗ represents complex 10
Normalized Power Spectral Density£¨dB£©

conjugate operations, and 𝑁1 𝐹𝑁∗ ×𝑁 denotes the matrix of 0


inverse Fourier transform. After 𝜇 times up-sampling to the
signal 𝑥, we can obtain the spectrum model as: −10

1 −20
𝑆 = 𝐹𝑢𝑁 ×𝑁 𝑥 = 𝐹𝑢𝑁 ×𝑁 × 𝐹𝑁∗ ×𝑁 𝑋 = 𝑄𝑋 (3)
𝑁 −30

where 𝑄 = 𝑁1 𝐹𝑢𝑁 ×𝑁 × 𝐹𝑁∗ ×𝑁 is a 𝜇𝑁 × 𝑁 matrix, which −40


is called the spectrum parameter matrix. The element of 𝑄 is
−50
𝑁
∑ −1
𝑛 𝑙 With AIC
𝑄(𝑙, 𝑘) = exp(𝑗2𝜋 (𝑘 − )) (4) −60
Without AIC
𝑛=0
𝑁 𝑢
−70
80 85 90 95 100 105 110 115 120 125
where 𝑙 = 0, 1, 2 ⋅ ⋅ ⋅ 𝑢𝑁 − 1, 𝑘 = 0, 1, 2 ⋅ ⋅ ⋅ 𝑁 − 1, and 𝜇 is Carrier Index n
the sampling factor.
Suppose that the carries to be protected locates in [𝑣1 : 𝑣2]. Fig. 13. The performance of AIC algorithm
The SUs will put 𝑐 active interference cancelation carriers on

936
C. Receiver process flow spacing as it only ranges from −𝜋 to 𝜋. Hence,equation (10)
The flow of data processing for the receiver has been shown can not serve as estimation for quite large frequency offset.
in Fig.(14). However, in our testbed, the frequency offset is not so large,
so the estimation range in (10) is enough. After the frequency
compensation for the long preamble and data sequences in
RF Rx A/D
down-
Filtering
Frame time domain, the cyclic prefix will be removed before the
conversion detection
Short preamble FFT Transformation. Channel estimation and compensation
Symbol
are realized in frequency domain. The channel estimation
synchronization consist of two parts which are based on the long preamble and
long preamble
data
pilot respectively. The principle of these two parts is the same,
Frequency offset
Estimation based on
Frequency which is using the known training sequence to get the channel
compensation
Preamble fading parameters, and then give corresponding compensation
Remove to the data sequence. The pilot is mainly used for the real-
Prefix
time estimating of the channel. Then the last step is to give
FFT some inverse processing to the data according to the processing
in the transmitter, including demapping ,deinterleaving and
Demapping The second
Deinterleave channel
data The channel pre-
compensation
decoding.
Decoding compensation data

Channel
Estimation
Channel Estimation III. T EST R ESULTS
pilot based on Preamble preamble
based on Pilot

In this section, we give the test results of our system. The


Fig. 14. The process flow for the receiver following table shows the system parameters of our platform
including the performance parameters. Fig.15∼Fig.18 present
The function of the software work of the receiver mainly the constellation of the received data and the demodulated data
starts from the frame detection. Frame detection and symbol with QPSK and 16QAM modulation respectively.
synchronization are realized according to the cycle charac-
teristics of the short preamble. The receiver correlates the System parameters Value
received signal with a local preamble sequence. Using this Center frequency 2.422GHz
correlation method the beginning of the frame can be estimated total bandwidth 20MHz
as the correlation will be significantly higher when the received carrier number 128
preamble samples are pushed through the correlator. The IFFT points 256
frequency offset estimation is realized using the long preamble Prefix points 64
which is composed of two same symbols. The samples of symbol cycle 8𝜇𝑠
the received signal 𝑦(𝑘) are modulated by a frequency offset carrier number used by SU 48
with respect to the transmitted signal 𝑥(𝑘). The phase of the Data rate for SU with 16QAM 12𝑀 𝑏𝑖𝑡/𝑠
samples grows continuously in consequence of this frequency Data rate for SU with QPSK 6𝑀 𝑏𝑖𝑡/𝑠
offset Δ𝑓𝑐 : FER for SU with 16QAM 10−3
𝑦(𝑘) = 𝑥(𝑘)𝑒𝑗2𝜋Δ𝑓𝑐 𝑇𝑠 𝑘 (8) FER for SU with QPSK 10−2
where 𝑇𝑠 is the sampling period. The correlation with the From the constellation diagrams we can see that our system
delayed version of the same signal 𝑦(𝑘 + 𝑁 ) yields: can efficiently utilize the vacant frequency band to commu-
𝑁∑
−1 nicate. At the same time, the energy emitted from the SU’s
𝑅= 𝑦(𝑘)𝑦 ∗ (𝑘 + 𝑁 ) carriers to the other carriers is very little, so it will not disturb
𝑘=0
𝑁∑−1 the existing system.
= 𝑥(𝑘)𝑒𝑗2𝜋Δ𝑓𝑐 𝑇𝑠 𝑘 𝑥∗ (𝑘 + 𝑁 )𝑒−𝑗2𝜋Δ𝑓𝑐 𝑇𝑠 (𝑘+𝑁 )
𝑘=0 (9)
𝑁∑−1 IV. C ONCLUSION
2 𝑗2𝜋Δ𝑓𝑐 𝑇𝑠 𝑘 −𝑗2𝜋Δ𝑓𝑐 𝑇𝑠 (𝑘+𝑁 )
= ∣𝑥(𝑘)∣ 𝑒 𝑒
𝑘=0
𝑁∑
−1 In this paper, we introduce a testbed for cognitive OFDMA
2
= 𝑒−𝑗2𝜋Δ𝑓𝑐 𝑇 ∣𝑥(𝑘)∣ system which is realized on a fully programmable softradio
𝑘=0
platform, Microsoft Research Software Radio(MRSR). In our
where 𝑇 = 𝑇𝑠 𝑁 denotes the duration of the correlation. Then testbed, the cognitive nodes can efficiently sense the vacant
the frequency offset can be calculated as follows: frequency band, make an assessment of the idle band and then
1 make a reasonable usage of the available band segment to
Δ𝑓𝑐 = arg 𝑅∗ (10)
2𝜋𝑇 improve the frequency utilization. The test results show that
It is obvious that the estimation in (10) does not correctly our system has a reliable sensing function as well as a robust
deliver the frequency offsets which are larger than half a carrier communication in the vacant frequency band.

937
1000 1000

800 800

600 600

400 400

200 200

0 0

−200 −200

−400 −400

−600 −600

−800 −800

−1000 −1000
−1000 −500 0 500 1000 −1000 −500 0 500 1000

Fig. 15. The constellation for the received data with QPSK modulation Fig. 17. The constellation for the received data with 16QAM modulation

200 150

150
100

100
50
50

0 0

−50
−50

−100
−100
−150

−200 −150
−200 −150 −100 −50 0 50 100 150 200 −100 −50 0 50 100

Fig. 16. The constellation for the demodulated data with QPSK modulation Fig. 18. The constellation for the demodulated data with 16QAM modulation

R EFERENCES [10] MacKay, D.J.C, “Fountain codes,” Communication, IEE Proceedings-,


vol.152, Dec.2005.
[1] J.Mitola III and G.Q.Maguire, Jr., “Cognitive radio: Making software [11] Ze Lin, Zhaoyang Zhang “The Research on Active Interference Can-
radios more personal,” IEEE Pers. Commun., vol.6, no. 4, pp.13-18, Aug. cellation in cognitive OFDMA system,” M . S. Dissertation in Zhejiang
1999. University, 2010.
[2] J.Mitola, Cognitive Radio An Integrated Agent Architecture for Software [12] W. Gander, “Least squares with a quadratic constraint. Number,” Math.,
Defined Radio. PhD thesis, Royal Institute of Technology, 2000. vol. 36, pp. 291-307, 1981.
[3] T.S.Rappaport, A.Annamalai, R.M.Buehrer, and W.H.Tranter, “Wireless [13] G. H. Golub, C. F. Van Loan, “Matrix Computations,” John Hopkins,
communications:Past events and a future perspective,” IEEE Commun. University Press, Third Edition, 1996.
Mag., vol.40, no.5, pp.148-161, May, 2002.
[4] IEEE Standard for Local and metropolitan area networks Part 16: Air
Interface for Broadband Wirelwss Access Systems, IEEE Standard 802.16,
May 2009.
[5] Kun Tan, Jiansong Zhang, Ji Fang, He Liu, Yusheng Ye, Shen Wang,
Yongguang Zhang, Haitao Wu, Wei Wang, and Geoffrey Voelker, “Sora:
High Performance Software Radio using General Purpose Multi-core
Processors,” USENIX NSDI 2009, Apr 2009, Boston, MA. (Awarded Best
Paper).
[6] T. Yucek and H. Arslan, “A Survey of Spectrum Sensing Algorithm for
Cognitive Radio Application,” in IEEE Commu. Surveys Tuts., vol.11,
no.1, pp.116-130, First Quar., 2009.
[7] D. Cabric, S. M. Mishra, and R. W. Brodersen, “Implementation issues in
spectrum sensing for cognitive radios,” in Proc. Asilomar Conf. Signals,
Systems and Computers, pp.772-776, Nov. 2004.
[8] R. Tandra and A. Sahai, “SNR walls for feature de
[9] Lu Ye, Zhaoyang Zhang, Jianmin Zhang, Huazi Zhang, “Carrier Sending
with Self-cancelation of Inter-Carrier Emission in Cognitive OFDMA
System,” Proc. IEEE ICC, 2011.

938

You might also like