Professional Documents
Culture Documents
Regular Paper
Original Contribution
1
INTRODUCTION
and the dimensions, including the doping levels for the two
devices A and B is given in Table I.
TABLE I
DEVICE GEOMETY AND DOPING LEVELS FOR THE TWO DEVICES STUDIED
Parameter
Device A [2]
Device B [5]
5 x 50m2
1.2 mm2
Emitter Width
5 m
7m
5 m
3m
0.2 m
0.8m
2 x 10 19
2 x 10 19
Base thickness
0.1 m
1 m
Base contact
density ( cm -3)
1 x 10 19
8 x 10 19
Base doping
density( cm -3)
2 x 10 18
4.1 x 10 17
Sub-Collector length
3 m
12 m
Collector doping
density ( cm -3)
2 x 10 16
8.5 x 10 15
Bi (N ) =
min,i + max,i (
N g ,i
N
max,i min,i
) i
(1)
T =T0 =300K
where i is for electrons (n) and holes (p). The values of the
constants max,i , min,i N g ,i , and i are presented in Table
II, taken from the literature approximated for the 4H-SiC
material [13-16]. In relationship with Equation (1), the
temperature dependence mobility is modeled by:
T
Bi (N).( )i
T0
i (N,T) = max,i (T0 )
T
1+ Bi (N).( )i +i
T0
Where T0 is the room temperature and
(2)
i ( N , T )
is the
i
i
i
a Data
2.6
0.5
aT 2
,
T +b
(3)
IV.
T
n, p = n, p (T0 )
T0
T
N c, v = N c, v (T0 )
T0
(4)
1.2
Simulated
1.0
(5)
Saturation velocity
Nc (300 K)
Nv (300 K)
SRH lifetime
Surface-recom. velocity
G a in ( N o r m a liz e d )
E g (T ) = E g (0)
3 x 107 cm/s
1.64 x1019cm
3.22 x1019cm3
50 ns
1x105 cm/s
1x104 cm/s
[6]
[6]
[6]
[23]
[21]
[22]
Measured: Device A
0.8
0.6
0.4
0.2
0.0
1.0
1.5
2.0
2.5
3.0
3.5
-1
Temperature 1000/T (k )
Fig. 2. The simulated maximum current gain versus 1000/T
compared with the measured experimental device A [1].
The differential base resistance is an important parameter
for bipolar transistors at high frequency application. In bipolar
junction transistors, the emitter prevents the diffusion of
majority carriers from base (holes) into the emitter, which
results in high electron injection efficiency. The differential
base resistance is obtained from the inverse slope of the base
current (Ib) at a given emitter-base bias (VBE) and temperature
in low injection region, Figure 3. The high doping density of
the base allows the base resistance to decrease without
significantly sacrificing the emitter efficiency. As shown in
Fig. 3, a similar trend is observed experimentally [1,2,3]. The
measured base resistance in device A remains relatively
constant at high temperature. The comparison shows the
overall strength of the device modeling in predicting the
1
Simulation
Measured-Device A
0.8
[5]. Both the room temperature and the data for T=423 K are
shown. The maximum current gain simulated for the device B
is about 25 at room temperature and about 15 at T=423 K
compared with the measured data. The maximum dc gain
simulated is in close agreement with the experimental result.
The measured gain is somewhat lower than the simulated
result at low collector current (low injection) indicating
leakage current in measured data.
0.6
Simulated T=300 K
0.2
30
Simulated T=423 K
0.4
25
4
20
-1
G a in
Temperature 1000/T (k )
Fig. 3. The simulated base resistance versus 1000/T compared with the
measured experimental device A [1].
15
10
5
0
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
20
G a in
16
12
8
Measured: Device A
Simulated
4
0
0
10
15
20
Ib Measured: Device A
Ic Measured: Device A
Ib Simulated
0.10
Ic Simulated
1.E-01
0.09
1.E-02
Simulated T=300 K
0.07
0.06
Simulated: T=423 K
1.E-03
C u rre n t (A )
Current Ib (A)
0.08
Measured:Device B T=423 K
0.05
0.04
1.E-04
1.E-05
1.E-06
1.E-07
1.E-08
0.03
1.E-09
0.02
1.E-10
0
0.01
10
0.00
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
Fig. 7. The simulated room temperature base current and collector current for
the device A as compared with the measured data [1,2,3]. The measured base
current shows leakage current. A maximum dc gain of about 20 is obtained for
this device at room temperature. High level injection is dominated by series
effect.
V.
CONCLUSION
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
ICSRM manuscript no. 470, session ThP3-14, Lyon, France, Oct. 5-10,
2003.
Danialsson, E., Processing and electrical characterization of GaN/SiC
heterojunctions and SiC bipolar transistors, ISRN KTH/EKT/FR01/1-SE, KTH, Royal Institute of Technology, Stockholm, 2001.
Danielsson, E., et al., Solid State electronics, vol. 47 , p. 639, 2003.
Li, X, et al., on the temp. coefficient of 4H-SiC BJT current gain,
Solid State Electronics, vol. 47, pp. 233-239, 2003.
Mnatsakanov,T.T., et al., carrier mobility model for simulation of
SiC-based electronic devices, Electronic Journal: Semiconductor
Science and Technology, 17, No. 9, pp. 974-977, 2002.
Pinto, M. R., Rafferty, C. S., Dutton, R. W., PISCES II: Poisson and
Continuity Equation Solver, Integrated Circuits laboratory, EE
Department, Stanford University, CA., 1984.
Raghunathan, R., and Baliga, B. J., p-type 4H and 6H-SiC HighVoltage Schottky Barrier Diodes, IEEE EDL, vol. 19, no. 3, pp. 7173, 1988.
Caughey, D. M., and Thomas, R.E., Proceedings of IEEE, vol. 55, p.
2192, 1967.
Mnatsakanov T T, Pomortseva L I and Yurkov S. N., Semiconductors
35 394, 2001.
Roschke M and Schwierz F., IEEE Trans. Electron Devices 48 1442,
2001.
Morkoc, H., et al., Large-band-gap SiC III-V nitride, and II-VI ZnSebased semiconductor, J. Appl. Phys. 76 (3), 1, pp. 1363-1398, 1994.
Joshi, R. P., Monte Carlo calculation of the temperature- and fielddependent electron transport parameters for 4H-SiC, J. Appl. Phys.
78(9), pp. 5518-5521, 1995.
Thurmond, C. D.,The standard thermodynamic for the formation of
electrons and holes in Ge, GaAs, GaPs, J. Electrochem., Soc., vol.
122, pp. 1133-1141, 1975.
Sze, S. M., Physics of Semiconductor Devices, John Wiley & Sons,
New York, 1981.
Arora, N. D., Hauser, J. R., and Roulston, D.J., IEEE Trans. ED, vol.
29, p. 292, 1982.
Ayalew, Tesfaye ,SiC semiconductor Devices technology, Modeling,
and Simulation, PhD Thesis, Technischen Universitat Wien, 2004.
Galeckas, A., et al., Appl. Phys. Lett., 79, 365, 2001.
Ivanov, P. A., et al., Factors, limiting the current gain in high-voltage
4H-SiC npn BJTs, Solid State Electronics, vol. 46, pp. 567-572, 2002.
Fardi, H. Z., Modeling the dc gain of 4H-SiC Bipolar Transistors as a
function of surface recombination velocity, Solid State Electronics,
2005.
Van Zeghbroeck Bart, Perez, I., Zhao, F., and Torvik, J., Technology
development of 4H-SiC BJTs with 5GHx fMAX, CS MANTECH
Conference, pp. 223-226 ,Vancouver, Canada, April 24-27, 2006.
Regular Paper
Original Contribution
7
I. INTRODUCTION
Multi-view video capturing, analysis, coding and display
have attracted a lot of attention in recent years since three
Manuscript received on August 1, 2007. This work was supported by Natural
Science Foundation of China (grant 60472100, 60672073), the Program for
New Century Excellent Talents in University (NCET-06-0537), the Key Project
of Chinese Ministry of Education (grant 206059), and the Natural Science
Foundation of Ningbo China (grant 2007A610037).
Yun Zhang was with the Faculty of Information Science and Engineering,
Ningbo University, Ningbo, China, he is now with the Institute of Computing
Technology, Chinese Academic of Science and Graduate School of Chinese
Academic of Science, Bejing, 100080, China (email: zhangyun_8851@163.
com)
Mei Yu is with the Faculty of Information Science and Engineering, Ningbo
University, Ningbo, 315211, China (e-mail: yumei2@yahoo.com.cn). She is
corresponding author of the manuscript.
Gangyi Jiang is with the Faculty of Information Science and Engineering,
Ningbo University, Ningbo, 315211, China (e-mail: jianggangyi@126.com).
View
P
P
P
Time
Time
N time M view
x p
i
(1)
B. Computational Complexity
In H.264/AVC based MVC coding platform, a good
portion of the gain comes from high precise disparity
compensation prediction (DCP) and motion compensation
prediction (MCP), however, at the cost of high coding
complexity and memory consumption. MCP or DCP occupies
most of coding time, so in this paper, we estimate the
computational complexity of a MVC prediction structure by
using the minimum number of reference frames of a coding
cell, PNmin. (Note that the coding structures could expand the
number of referenced frames for high compression
efficiency.) The lager PNmin is, the more complicated the
encoder is. For example, simulcast scheme, as illustrated in
Fig.1(a), adopts P frames that reference at least one frame in
the encoding procedure. Therefore, its PNmin is 30.
C. Memory Requirement
General memory requirements stem from AVC, since all
frames from all views are reordered and treated as one single
video stream by the coder. And the decoded picture buffer
(DPB) in H.264/AVC is used to store the reference frames.
Assume that each scheme adopts the optimal coding order to
minimize the DPB size, represented by DPBmin here. For the
simulcast prediction structure, since P frames are singlereferenced and are coded view-by-view, only one frame has to
be stored in DPB for next frame, i.e. DPBmin equals to 1.
D. View Scalability
View scalability enables the video to be displayed on a
multitude of different displayers, such as multi-view
displayer, stereo displayer or HDTV, and terminals over
networks with varying conditions[3]. Therefore, the decoder
should be able to selectively decode the appropriate number of
views according to the type of display modes with view
scalability. In this paper, we define two cost variables, FSV,
FDV, to represent the average number of compulsorily decoded
frames in a coding cell when single view or double views are
displayed, respectively.
Let On be a set of frames in a coding cell and Xi,j be a set of
compulsory decoded frames when the frame at (i,j) position in
a coding cell is displayed. Thus X i , j On . Suppose j is the
probability for jth view to be chosen to display by the user, and
j,k is the probability for both jth view and kth view to be
accessed. FSV and FDV are defined as
i =1
(2)
FSV =
FDV =
Ntime
Card U X i, j j
j =1
i =1
M view
M view M view
Ntime
Card U ( X
j =1 k = j +1
i =1
i, j
U X i ,k ) j ,k
(3)
(4)
View
View
A. Simulcast
As a reference of evaluating MVC schemes compression
efficiency, simulcast coding structure was analyzed based on
H.264/AVC coding platform, in which each view in multiview video is coded independently [4]. The first frame of each
view is coded as an intra frame (I-frame), the remaining
frames are coded as P-frames, as shown in Fig.1(a). Simulcast
coding can be achieved by independently coding the coding
cell column by column. The simulcast coding structure is a
direct expansion of mono-video coding scheme.
On account of the deficiency of simulcasts RD
performance, Sun et al have proposed a MVC scheme based
on DCP, denoted as IPPP, as shown in Fig.1(b). It is clear
that the key frame of the center view in each group of GOP is
coded as I-frame, while other key frames are coded using
inter-view DCP from center to both sides. The structure
enhances coding efficiency by eliminating inter-view
redundancy of the first time of a coding cell, while maintains
relatively fast RA.
P
P
Time
Time
Bts
Bts
Bts
Bts
Bt
Bts
Bts
Bts
Bts
Bt
Bts
Bts
Ps
Ps
Ps
Ps
Bts
Bts
Bt
Bts
Bts
Bts
Bts
Bts
Bt
Bts
Bts
Bts
Bts
Bts
Pt
Bts
Bts
10
B2
b4
B3
b4
B3
B2
B1
B2
B1
B3
b4
B3
b4
B3
B2
B3
B2
B3
B2
B3
b4
B3
b4
B1
P0
B1
P0
B3
I0
B3
B2
B3
B3
P0
b4
B1
B1
B3
B3
P0
b4
B2
B1
B3
46
38
36
20cm
640480
640480
320240
19.5cm
30mm
3cm
Camera
distance
320240
Simulcast
IPPP
Jeong
PSVP
BSVP
Mpicture
HBP
GoGOP SR
GoGOP MR
40
TABLE II
CODING PARAMETER
Platform
JM8.5 main profile
RDO
Yes
Entropy Coding
Loop filter, CABAC
Search range
32 for CIF/VGA
Motion/Disparity Compensation 1/4Pixel Accuracy, Full Search
QP
18, 24, 30, 36
0.1
(b)
0.3
0.4
0.5
0.6
0.7
Aquarium
Aquarium
42
40
38
Simulcast
IPPP
Jeong
PSVP
BSVP
Mpicture
HBP
GoGOP SR
GoGOP MR
36
34
32
30
28
0.5
(c)
1.5
Entropy (Bit/pixel)
Entropy(Bit/pixel)
2.5
Xmas
Xmas
46
golf2
golf2
46
42
44
Average PSNR(dB)
Average PSNR(dB)
44
Average PSNR(dB)
Average PSNR(dB)
0.2
Entropy(Bit/pixel)
Entropy (Bit/pixel)
44
Average PSNR(dB)
Average PSNR(dB)
42
32
Resolution
Video characteristics
44
34
TABLE I
T EST S EQUENCES FOR M ULTI - VIEW V IDEO
Data Set
flamenco1
flamenco1
48
Average PSNR(dB)
Average PSNR(dB)
11
40
Simulcast
IPPP
Jeong
PSVP
BSVP
Mpicture
HBP
GoGOP SR
38
36
34
GoGOP MR
32
0.1
0.2
0.3
0.4
0.5
0.6
Entropy (Bit/pixel)
Entropy(Bit/pixel)
(a)
0.7
42
40
Simulcast
IPPP
Jeong
PSVP
BSVP
Mpicture
HBP
GoGOP SR
GoGOP MR
38
36
34
32
0.1
(d)
0.2
0.3
0.4
0.5
Entropy (Bit/pixel)
Entropy(Bit/pixel)
0.6
0.7
12
Average PSNR(dB)
Average PSNR(dB)
race2
race2
44
42
TABLE II
PERFORMANCE COMPARISON TABLE
40
Coding
Structure
38
Simulcast
IPPP
Jeong
PSVP
BSVP
Mpicture
HBP
GoGOP SR
GoGOP MR
36
34
32
30
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Entropy (Bit/pixel)
Entropy(Bit/pixel)
(e)
Simulcast
GoGOP SR
GoGOP MR
PSVP
BSVP
MVC with
Mpicture
MVC with
HBP*
Lims
scheme
IPPP
Random
Access Cost
Fav
Fmax
3.0
6
3.6
9
4.6
14
11.0
34
7.5
19
PNmin
DPB
min
30
111
114
58
83
1
16
16
7
7
View
Scalablity
FSV
FDV
7
14
12.6
21.7
15.4
25.2
21
28
21
28
6.0
20
97
16
16
22.4
7.6
16
96#
21
15.2
26.1
3.1
62
12.6
18.2
4.2
34
8.2
15.2
13
14
REFERENCES
[1]
[2]
[3]
[4]
[5]
V. CONCLUSION
Nine typical MVC prediction structures with six categories
were analyzed in detail. And comparisons on compression
efficiency, RA, view scalability and complexity have been
made. From the experimental results, we can obtain that
MVC with HBP is the most efficient in compression, and
simulcast is most suitable for the applications without strong
bandwidth limitations but with low-delay requirements. Other
schemes have a trade-off between compression efficiency and
functionalities by adopting multiple intra frames, sequential
prediction, multiple reference prediction and I-frame centered
structure etc.
For most multi-view video, the majority of correlation is
mainly in temporally preceding frames. However, there is
also a significant amount of dependencies that are better
predicted from the spatial direction. Multiple reference
prediction may reduce prediction residue and improve coding
efficiency, but inordinate and fulsome multiple reference
predictions not only increase complexity and memory
requirements, but also make against view scalability and
random accessibility. Multiple reference prediction should be
used more effectively according to the relationship between
temporal and inter-view dependencies. Additionally, it is
obvious that view-by-view sequential prediction relieves
exposure and occlusion problems but goes against with RA
performance, view scalability. Fortunately, using B picture in
sequential prediction structure may improve RA a bit. And
improvements on RA and scalability can be achieved by
placing I-frame in the center of the 2D GOP. RA can be
improved by adopting multiple I-frames in a 2D GOP while
at the cost of compression efficiency.
In future work, we will do further researches to propose
some smart prediction structures that predict from spatial
domain, temporal domain or spatial-temporal joint domain
according to the spatial-temporal correlation. And the
structure will provide high coding efficiency, low complexity,
reasonable memory requirements and fast RA.
ACKNOWLEDGMENT
Mitsubishi
Electric
Research
Laboratories,
KDDI
R&D
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
15
Engineering, Ningbo University, China. Her research interests include image and
video coding, and visual perception.
Gangyi Jiang received his M.S degree from Hangzhou
University, China, in 1992, and received his Ph.D degree
from Ajou University, Korea, in 2000. He is now a
professor at Faculty of Information Science and
Engineering, Ningbo University, China. His research
interests mainly include digital video compression and
communications, multi-view video coding, image based
rendering, and image processing.
Regular Paper
Original Contribution
16
I. INTRODUCTION
17
/4
/4
RFout
OMN
RFin
Cout
IMN
Cin
Fig. 1. Simplified schematic of an amplifier including bias network (dashed
line) with RF choke.
(b) Port 3
Port 3
lA
/4
Port 1
(c)
rO,RS
/4
B
B
Port 3
RF choke
A
Port 2
Port 1
rO,BS
B
/4
Port 2
A
Port 1
/4
Port 2
Fig. 2. Bias network with RF choke using (a) microstrip stub, (b) radial stub,
and (c) butterfly stub.
Material
RO4350B
Dielectric thickness
Dielectric constant
Dissipation factor
Metal thickness
Metal conductivity
Surface roughness
VCC
TABLE I
PCB PROCESS PARAMETERS
0.254 mm
3.48 0.05
0.0037
0.035 mm
5.8 x 107 S/m
0.001 mm
18
Measured
Simulated
-20
-30
-40
1
V. DISCUSSION
These discontinuities shown in Fig. 4 can be explained by
resonances in the DC-path of the bias network. To verify the
resonance phenomena along the DC path, the input impedance
amplitude at point A (see Fig. 1) when looking into the RFchoke, |ZinA|, is analyzed. Based on the simulated input
reflection coefficient S11, |ZinA| was calculated for the
microstrip stub bias network.
In Fig. 5, comparing |ZinA| to the transmission coefficient, it
can be seen that the discontinuities of the |S21| appear at the
same frequencies where |ZinA| 0. Thus, it can be concluded
that discontinuities of the transmission coefficient are caused
by equivalent series resonance in the DC path. The DC path
from junction point A to Port 3 (see Fig. 2) results in a shortcircuited /2 transmission line resonator.
0
-10
Measured
-20
Simulated
-30
10000
|S21|
|ZinA|
-10
1000
-20
100
-30
10
-40
1
1
Frequency (GHz)
-40
1
Frequency (GHz)
(b) Radial stub, layout length lA = 28 mm and rO,RS = 6.8 mm, see Fig 2b.
Forward transmission (dB)
|ZinA| ()
(a) Microstrip stub, layout length lA = 28 mm and /4=11.5 mm, see Fig 2a.
Frequency (GHz)
0
-10
Measured
-20
Simulated
-30
-40
1
Frequency (GHz)
(c) Butterfly stub, layout length lA = 28 mm and rO,BS = 8.3 mm, see Fig 2c.
Fig. 4. Simulated and measured transmission coefficients of the bias networks
with (a) microstrip stub, (b) radial stub, and (c) butterfly stub.
Fig. 5. Microstrip stub bias network. |S21| and the input impedance at point A
(see Fig. 2 ), |ZinA|.
19
0
-10
-20
-30
Microstrip stub
0
-10
-20
-30
0
-10
-20
-30
VI. CONCLUSION
Radial stub
4
5
Butterfly stub
Frequency (GHz)
Microstrip stub
0
-5
-10
-15
-20
4
5
Radial stub
REFERENCES
2
0
-5
-10
-15
-20
1
4
5
Butterfly stub
10
50
90
6
[1]
[2]
[3]
Frequency (GHz)
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
20
[15] I. Sakagami, Y. Hao, M. Mohemaiti, A. Tokunou, On a transmissionline Butterworth lowpass filter using radial stubs, IEEE International
Symposium on Circuits and Systems, 2002, ISCAS 2002, vol. 3, pp. 867
870.
[16] F. Giannini, C. Paoloni, M. Ruggieri, CAD-Oriented Lossy Models for
Radial Stubs, IEEE Tran. Microwave Theory and Tech., vol. 36, pp.
305 313, Feb. 1988.
[17] F. Giannini, M Salerno, R. Sorrentino, Two-Octave Stopband
Microstrip Low-Pass Filter Design using Butterfly Stubs, 16th
European Microwave Conference, 1986, pp. 292 297.
[18] J. Zhu, Z. Feng, Enhancement of Stopband Rejection of Microstrip
Bandpass Filters by Radial Stubs, International Conference on
Microwave and Millimeter Wave Technology, 2007, ICMMT '07, April
2007, pp. 1 3.
[19] B. T. Tan, J. J. Yu, S. J. Koh, S. T. Chew, Investigation into Broadband
PBG Using a Butterfly-Radial Slot (BRS), Microwave Symposium
Digest, 2003 IEEE MTT-S International, vol.2, 8-13 Jun. 2003, pp. 1107
1110.
[20] B. T. Tan, J. J. Yu, S. T. Chew M.-S. Leong, B.-L. Ooi, A Miniaturized
Dual-Mode Ring Bandpass Filter with a New perturbation, IEEE Tran.
Microwave Theory and Tech., vol. 53, pp. 343 348, Jan. 2005.
[21] R.-J. Mao, X.-H. Tang, F. Xiao, Miniaturized Dual-Mode Ring
Bandpass Filters With Patterned Ground Plane, IEEE Tran. Microwave
Theory and Tech., vol. 55, pp.1539 1547, Jul. 2007.
[22] R. K. Joshi, A. R. Harish, Characteristics of a Rotated Butterfly Radial
Stub, Microwave Symposium Digest, 2006. IEEE MTT-S International,
Jun. 2006, pp.1165 1168.
[23] R. Dehbashi, K. Forooraghi, Z. Atlasbaf, N. Amiri, A Novel BroadBand Band-Stop Resonator with Compact Size, Asia-Pacific
Conference on Applied Electromagnetics, 2005, APACE 2005, Dec.
2005.
[24] Agilent Technologies Inc., Advanced Design System (ADS),
http://eesof.tm.agilent.com/.
Regular Paper
Original Contribution
21
I. I NTRODUCTION
a broad range of mathematical concepts, computational algorithms and technologies. In other words, Cryptology is
a multi-disciplinary subject that covers a wide spectrum of
different disciplines and increasingly involves using a range of
engineering concepts and technologies through the innovation
associated with term technology transfer. These include areas
such as Synergetics, which is an interdisciplinary science
explaining the formation and self-organization of patterns and
structures in non-equilibrium open systems and Semiotics,
which is the study of both individual and grouped signs and
symbols, including the study of how meaning is constructed
and understood [11].
Cryptology is often concerned with the application of formal
mathematical techniques to design a cryptosystem and to
estimate its theoretical security. This can include the use of
formal methods for the design of security software which
should ideally be a safety critical [12]. However, although
the mathematically dened and provable strength of a cryptographic algorithm or cryptosystem is necessary, it is not a
sufcient requirement for a system to be acceptably secure.
This is because it is difcult to estimate the security of a
cryptosystem in any formal sense when it is implemented
under operational conditions that can not always be predicted and thus, simulated. The security associated with a
cryptosystem can be checked only by means of proving its
resistance to various kinds of known attack that are likely
to be implemented. However, in practice, this does not mean
that the system is secure since other attacks may exist that
are not included in simulated or test conditions. The reason
for this is that humans possess a broad range of abilities from
unbelievable ineptitude to astonishing brilliance which can not
be formalised in a mathematical sense or on a case by case
basis.
The practical realities associated with Cryptology are indicative of the fact that security a process, not a product
[13]. Whatever the sophistication of the security product
(e.g. the encryption and/or key exchange algorithm(s), for
example), unless the user adheres strictly to the procedures and
protocols designed for its use, the product can be severely
compromised. A good example of this is the use of the Enigma
[14] cipher by Germany during the Second World War. It was
not just the intelligence of the code breakers at Bletchley
Park in England that allowed the allies to break many of the
Enigma codes but the irresponsibility and, in many cases,
the sheer stupidity of the way in which the system was used
by the German armed and intelligence services at the time.
The basic mechanism for the Enigma cipher, which had
been developed as early as 1923 by Artur Schubius for se-
22
23
24
25
Fig. 1. Alice and Bob can place a message in a box which can be secured
using a combination lock and sent via a public network - the postal service,
for example.
26
Bob invent it and agree upon it. Thus, the principal problem
in symmetric encryption is how Alice and Bob exchange the
key. Irrespective of how strong the cipher and key are, unless
the key exchange problem can be solved in an appropriate and
a practicable way, symmetric encryption always suffers from
the same fundamental problem - key exchange!
If E denotes the encryption algorithm that is used which
depends upon a key K to encrypt plaintext P , then we can
consider the ciphertext C to be given by
C = EK (P ).
Decryption can then be denoted by the equation
P = EK (C).
Note that it is possible to encrypt a number of times using
different keys K1 , K2 , ... with the same encryption algorithm
to give a double encrypted cipher text
C = EK2 (EK1 (P ))
or a triple encrypted ciphertext
C = EK3 (EK2 (EK1 (P ))).
Decryption, is then undertaken using the same keys in the
reverse order to which they have been applied, i.e.
P = EK1 (EK2 (EK3 (C))).
Symmetric encryption systems, which are also referred to
as shared secret systems or private key systems, are usually
signicantly easier to use than systems that employ different
protocols (such as asymmetric encryption). However, the requirements and methods associated with key exchange sometimes make symmetric system difcult to use. Examples of
symmetric encryption systems include the Digital Encryption
Standard DES and DES3 (essentially, but not literally, the
Digital Encryption Standard with triple encryption) and the
Advanced Encryption Standard (AES). Symmetric systems are
commonly used in many banking and other nancial institutes
and in some military applications. A well known historical
example of a symmetric encryption engine, originally designed
for securing nancial transactions, and later used for military
communications, was the Enigma.
B. Asymmetric Ciphers
Instead of Alice and Bob agreeing on a combination number
a priori, suppose that Alice sets her lock to be open with a
combination number known only to her. If Bob then wishes to
send Alice a message, he can make a request for her to send
him an open lock. Bob can then write his message, place it
in the box which is then locked and sent on to Alice. Alice
can then unlock the box and recover the message using the
combination number known only to her. The point here is
that Bob does not need to know the combination number, he
only needs to receive an open lock from Alice. Of course
Bob can undertake exactly the same procedure in order to
receive a message from Alice. Clearly, the processes that are
undertaken by Alice and Bob in order to send and receive a
single message are not the same. The protocol is asymmetric
27
and
c1 c2 c3 = p.
28
using Fermats Little Theorem, i.e. for any integer a and prime
number p
ap = amodp.
Note that this result is strictly dependent on the fact that ed1
is divisible by (p 1)(q 1) making e a relative prime of
(p 1)(q 1) so that e and (p 1)(q 1) have no common
factors except for 1. This algorithm, is the basis for many
public/private or asymmetric encryption methods. Here, the
7 There are some claims that the method was rst developed at GCHQ in
England and then re-invented (or otherwise) by Rivest, Shamir and Adleman
in the USA; the method was not published openly by GCHQ - such are the
realities of keeping it quiet.
29
this does not in any way mean that they are the most secure.
Most government institutions and the military do not reveal
the type of algorithm used in the design of a cryptosystem.
The rationale for this is that, if we nd it difcult to break a
code with knowledge of the algorithm, then how much more
difcult is it to break a code if the algorithm is unknown?
On the other hand, within the academic community, security
in terms of algorithm secrecy is not considered to be of
high merit and publication of the algorithm(s) is always
recommended. It remains to be understood whether this is
a misconception within the academic world (due in part to
the innocence associated with academic culture) or a covertly
induced government policy (by those who are less innocent!).
For example, in 2003, it was reported that the Americans had
broken ciphers used by the Iranian intelligence services. What
was not mentioned, was the fact that the Iranian ciphers were
based on systems purchased indirectly from the USA and thus,
based on USA designed algorithms [45].
The known algorithm approach originally comes from the
work of Auguste Kerchhoff. Kerchhoffs Principle states that:
A cryptosystem should be secure even if everything about the
system, except the key, is public knowledge. This principle
was reformulated by Claude Shannon as the enemy knows
the system and is widely embraced by cryptographers world
wide. In accordance with the Kerchhoff-Shannon principle,
the majority of civilian cryptosystems make use of publicly
known algorithms. The principle is that of security through
transparency in which open-source software is considered to
be inherently more secure than closed source software. On
this basis there are several methods by which a system can be
attacked where, in each case, it is assumed that the cryptanalyst
has full knowledge of the algorithm(s) used.
A. Basic Attacks
We provide a brief overview of the basic attack strategies
associated with cryptanalysis.
Ciphertext-only attack is where the cryptanalyst has a
ciphertext of several messages at their disposal. All messages
are assumed to have been encrypted using the same algorithm.
The challenge for the cryptanalyst is to try and recover the
plaintext from these messages. Clearly a cryptanalyst will be
in a valuable position if they can recover the actual keys used
for encryption.
Known-plaintext attack makes the task of the cryptanalysis
simpler because, in this case, access is available to both the
plaintext and the corresponding ciphertext. It is then necessary
to deduce the key used for encrypting the messages, or design
an algorithm to decrypt any new messages encrypted with
the same key.
Chosen-plaintext attack involves the cryptanalyst possessing
both the plaintext and the ciphertext. In addition, the analyst
has the ability to encrypt plaintext and see the ciphertext
produced. This provides a powerful tool from which the keys
can be deduced.
30
31
breaking the Enigma naval transmissions under their timevariant code-book protocol would have been very difcult. A
British Naval message dated May 10, 1941 reads: 1. Capture
of U Boat 110 is to be referred to as operation Primrose;
2. Operation Primrose is to be treated with greatest secrecy
and as few people allowed to know as possible... Clearly,
and for obvious reasons, the British were anxious to make
sure that the Germans did not nd out that U-110 and its
codebooks had been captured and all the sailors who took part
in the operation were sworn to secrecy. On HMS Bulldogs
arrival back in Britain a representative from Bletchley Park,
photographed every page of every book. The interesting piece
of equipment turned out to be an Enigma machine, and the
books contained the Enigma codes being used by the German
navy.
The U-boat losses that increased signicantly through the
decryption of U-boat Enigma ciphers led Admiral Carl Doenitz
to suspect that his communications protocol had been compromised. He had no rm evidence, just a gut feeling that
something was wrong. His mistake was not to do anything
about it9 , an attitude that was typical of the German High
Command who were certiable with regard to their condence
in the Enigma system. However, they were not uniquely certiable. For example, on April 18, 1943, Admiral Yamamoto
(the victor of Pearl Harbour) was killed when his plane was
shot down while he was attempting to visit his forces in the
Solomon Islands. Notication of his visit from Rabaul to the
Solomons was broadcast as Morse coded ciphertext over the
radio, information that was being routinely decrypted by the
Americans. At this point in the Pacic War, the Japanese
were using a code book protocol similar to that used by the
German Navy, in which the keys were changed on a daily
basis, keys that the Americans had generated copies of. Some
weeks before his visit, Yamamoto had been given the option
of ordering a new set of code books to be issued. He had
refused to give the order on the grounds that the logistics
associated with transferring new code books over Japanese
held territory was incompatible with the time scale of his visit
and the possible breach of security that could arise through a
new code book being delivered into the hands of the enemy.
This decision cost him his life. However, it is a decision
that reects the problems associated with the distribution of
keys for symmetric cryptosystems especially when a multiuser protocol needs to be established for execution over a wide
communications area. In light of this problem, Yamamotos
decision was entirely rational but, nevertheless, a decision
based on the assumption that the cryptosystem had not already
been compromised. Perhaps it was his faith in the system and
thereby his refusal to think the unthinkable that cost him his
life!
The principles associated with cryptanalysis that have been
briey introduced here illustrate the importance of using a
dynamic approach to cryptology. Any feature of a security
infrastructure that has any degree of consistency is vulnerable
to attack. This can include plaintexts that have routine phrases
such as those used in letters, the key(s) used to encrypt the
9 An
32
u(r, t)
2
1
,
D
where D is the Diffusivity, S is a source term and u
is a function which describes physical properties such as
temperature, light, particle concentration and so on with initial
value u0 .
The diffusion equation describes elds u that are the result
of an ensemble of incoherent random walk processes, i.e.
walks whose direction changes arbitrarily from one step to the
next and where the most likely position after a time t is pro
portional to t. Further, the diffusion equation differentiates
between past and future, i.e. t . This is because the diffusing
eld u represents the behaviour of some average property of
an ensemble of many agents which cannot in general go back
to their original state. This fundamental property of diffusive
processes has a synergy with the use of one-way functions
in cryptology, i.e. functions that, given an input, produce an
output that is not reversible - an output from which it is not
possible to compute the input.
Consider the process of diffusion in which a source of
material diffuses into a surrounding homogeneous medium,
the material being described by some initial condition u(r, 0).
Physically, it is to be expected that the material will increasingly spread out as time evolves and that the concentration
of the material decreases further away from the source. The
general solution to the diffusion equation yields a result in
which the spatial concentration of material is given by the
convolution of the initial condition with a Gaussian function.
This solution is determined by considering how the process of
diffusion responds to a single point source which yields the
Greens function (in this case, a Gaussian function) given by
[47]-[49],
Fig. 2. Image of an optical source (left), the same source imaged through
steam (centre) and a simulation of this effect obtained by convolving the
source image with a Gaussian Point Spread Function (right).
u(r, 0) = u0 (r), =
G(r, t) =
4t
n
2
exp
r2
4t
, t > 0, r =| r | .
u(r, t) = n (r)(t)
2
(x + y 2 ) u0 (x, y)
4t
33
(1)n n n
u(r, t)
T
n!
tn
n=1
.
t=T
=D
2n
u(x, y, T ).
t=T
and after substituting this result into the series for u0 given
above, we obtain
u0 (r) = u(r, T ) +
(1)n
(DT )n
n!
n=1
u DT
2n
u(r, T )
u, DT << 1.
2
u(r, t) = S(r, t), u(r, 0) = 0
t
the solution is
u(r, t) = G(r, t) r t S(r, t), t > 0.
If a source is introduced in terms of an impulse at t = 0, then
the system will react accordingly and diffuse for t > 0. This
is equivalent to introducing a source function of the form
S(r, t) = s(r)(t).
Output=Diffusion+Confusion.
(1)n
[(DT )n
n!
n=1
2n
u(r, T ) + D1
u(x, y) =
exp (x2 + y 2 ) u0 (x, y) + n(x, y).
4t
4t
We thus arrive at a basic model for the process of diffusion
and confusion, namely
2n2
S(r)]
34
u(x, y) n(x, y) =
E(f ) =
xPf (x)dx,
E[exp(kf )] =
exp(kx)Pf (x)dx
E[exp(ikf )] =
f1 (t + )f2 ( )d = f1 (t)
f2 (t) = 0
exp(ikx)Pf (x)dx
F1 ()F2 () = 0
and
where
F1 () =
f1 (t) exp(it)dt
and
F2 () =
f2 (t) exp(it)dt.
If each function has a PDF Pr[f1 (t)] and Pr[f2 (t)] respectively, the PDF of the function f (t) that is the sum of f1 (t) and
f2 (t) is given by the convolution of Pr[f1 (t)] and Pr[f2 (t)],
i.e. the PDF of the function
f (t) = f1 (t) + f2 (t)
is given by [55], [56]
Pr[f (t)] = Pr[f1 (t)] Pr[f2 (t)].
Further, for a number of statistically independent
stochastic functions f1 (t), f2 (t), ..., each with a PDF
Pr[f1 (t)], Pr[f2 (t)], ..., the PDF of the sum of these
functions, i.e.
f (t) = f1 (t) + f2 (t) + f3 (t) + ...
Pf (x) =
12 The
i=1
is given by
Pr[f (t)] = Pr[f1 (t)] Pr[f2 (t)] Pr[f1 (t)] ...
dx exp(ikx).
and
Pf (x) = Pf1 (x) Pf1 (x) ...
35
1
X,
0,
| x | X/2;
otherwise
i=1
Fig. 4. Illustration of the Central Limit Theorem. The top-left image shows
plots of a 500 element uniformly distributed time series and its histogram
using 32 bins. The top-right image shows the result of adding two uniformly
distributed and independent time series together and the 32 bin histogram.
The bottom-left image is the result after adding three uniformly distributed
times series and the bottom-right image is the result of adding four uniformly
distributed times series.
6
exp(6x2 /XN )
N
where Pi (x) = P (x), i and N is large. Figure 4 illustrates
the effect of successively adding uniformly distributed but
independent random times series (each consisting of 500
elements) and plotting the resulting histograms (using 32 bins),
i.e. given the discrete times series f1 [i], f2 [i], f3 [i], f4 [i] for
i=1 to 500, Figure 4 shows the time series
1
1
exp
2
2
Gauss(x)dx = 1
s1 [i] = f1 [i],
and
then, since
Gauss(x) exp(ik) exp
2 k2
2
2 k2
2
j=1
N 2 k2
2
so that
N
j=1
Gauss(x) =
1
2N 2
exp
1
2N
36
1 U (k)N (k)
u0 (r) = Fn
| N (k) |2
where N is the Fourier transform of n and U is the Fourier
transform of u. However, this approach requires regularisation
in order to eliminate any singularities when | N |2 = 0 through
application of a constrained deconvolution lter such as the
Wiener lter [52]. Alternatively, if n is the result of some
N (k)
U0 (k) = U0 (k)
| N (k) |2
so that
u0 (r) = n(r)
u(r).
= 1 and
I2 (x, y)
= 1.
37
Fig. 6. Binary image (top-left), uniformly distributed 2D noise eld (topcentre), convolution (top-right) and associated 64-bin histograms (bottom-left,
-centre and -right respectively).
Fig. 5. Example of watermarking an image with another image using noise
based diffusion. The host image I2 (top-left ) is watermarked with the
watermark image I1 (top-centre) using the diffuser (top-right) given by a
uniform noise eld n whose pixel-by-pixel values depend upon the seed used
(the private key). The result of computing m I1 (bottom-left) is added to
the host image for R = 0.1 to generate the watermarked image u (bottomcentre). Recovery of the watermark image I1 (bottom-right) is accomplished
by subtracting the host image from the watermarked image and correlating
the result with the noise eld n.
u0 (r) =
i
then
n(r ri ).
i=1
n(r ri ) =
Pr[u(r)] = Pr
i=1
Pr[n(r)]
i=1
38
1
F
kx ky
,
39
I1
I2
| I1 |2
n I1
I1
I2 i2 .
| I1 |2
40
Fig. 10.
Example of a covert watermarking scheme. i1 (top-left) is
processed with i2 (top-middle) to produce the noise eld (top-right). i2 is
printed at 600 dpi, scanned at 300 dpi and then re-sampled back to its original
size (bottom-left). Correlating this image with the noise eld generates the
reconstruction (bottom-centre). The reconstruction depends on just the host
image and noise eld. If the noise eld and/or the host image are different
or corrupted, then a reconstruction is not achieved, as in the example given
(bottom-right).
Ii = log2 Pi
where Pi = ni /N which is the proportion of the subsets.
Finally, if we consider the most general case, where the subsets
are non-uniform in size, then the information will no longer
be the same for all subsets. In this case, we can consider the
mean information given by
N
I=
Pi log2 Pi
i=1
Pi ln Pi
i
41
Fig. 12. 128-bin histograms for an 7-bit ASCII plaintext u0 [i] (left), a
stream of uniformly distributed integers between 0 and 127 n[i] (centre) and
the substitution cipher u[i] (right).
Fig. 11. A 3000 element uniformly distributed random number stream (top
left) and its 64-bin discrete PDF (top right) with I = 4.1825 and a 3000
element Gaussian distributed random number stream (bottom left) and its 64bin discrete PDF (bottom right) with I = 3.2678.
1
F[Pu0 (x)]
42
Fig. 13. 127-bin histograms for an 7-bit ASCII plaintext u0 [i] (left) after
space-character padding, a stream of uniformly distributed integers between
0 and 255 n[i] (centre) and the substitution cipher u[i] (right).
general Greens function solution. Confusion has been considered through the application of the inhomogeneous diffusion
equation with a stochastic source function and it has been
shown that
u(r) = p(r) r u0 (r) + n(r)
where p is a Gaussian Point Spread Function and n is a
stochastic function.
Diffusion of noise involves the case when u0 is a stochastic
function. Diffusion by noise involves the use of a PSF p that
is a stochastic function. If u0 is taken to be deterministic
information, then we can consider the processes of noise
diffusion and confusion to be compounded in terms of the
following:
Diffusion
u(r) = n(r) r u0 (r).
Confusion
u(r) = u0 (r) + n(r).
43
44
(1)
(2)
(N )
),
(2)
(1)
(2)
(N )
),
(1)
(2)
(N )
45
or
xi+1 = rxi (1 xi )
where r = 1 + kt. This is a simple quadratic iterator known
as the logistic map and has a range of characteristics depending
on the value of r. This is illustrated in Figure 14 which
shows the output (for just 30 elements) from this iterator
for r = 1, r = 2, r = 3 and r = 4 and for an initial
value of 0.118 . For r = 1 and r = 2, convergent behaviour
Fig. 15. Feigenbaum diagram of the logistic map for 0 < r < 4 and
0 < x < 1.
Fig. 14. Output (30 elements) of the logistic map for values of r = 1 (top
left), r = 2 (top right), r = 3 (bottom left) and r = 4 (bottom right) and an
initial value of 0.1.
Fig. 16. Feigenbaum diagram of the logistic map for 3.840 < r < 3.855
and 0.44 < x < 0.52.
46
Fig. 17. Two dimensional phase space analysis of the Lorenz equations
illustrating the strange attractor.
dx2
dx3
dx1
= a(x2 x1 ),
= (bx3 )x1 x2 ,
= x1 x2 cx3
dt
dt
dt
where a, b and c are constants. These equations were originally
derived by Lorenz from the uid equations of motion (the
Navier Stokes equation, the equation for thermal conductivity
and the continuity equation) used to model heat convection in
the atmosphere and were studied in an attempt to explore the
transition to turbulence where a uid layer in a gravitational
eld is heated from below. By nite differencing, these equations, we convert the functions xi , i = 1, 2, 3 into discrete
form xn , i = 1, 2, 3 giving (using forward differencing)
i
dx1
dpx
dx2
= px ,
= x2xy;
= py ,
dt
dt
dt
(n+1)
x1
(n+1)
x2
(n+1)
x3
(n)
(n)
(n)
= x1 + ta(x2 x1 ),
(n)
(n) (n)
(n)
= x2 + t[(b x3 )x1 x2 ],
(n)
(n) (n)
(n)
= x3 + t[x1 x2 cx3 ].
47
associated with chaos for the purpose of encryption: Good mixing transformations are often formed by repeated products of
two simple non-commuting operations [11]. Hopf19 considered
the mixing of dough by such a sequence of non-commuting
operations. The dough is rst rolled out into a thin slab, then
folded over, then rolled, and then folded again and so on. The
same principle is used in the making of a Japanese sword,
the aim being to produce a material that is a highly diffused
version of the original material structure.
The use of chaos in cryptography was not fully appreciated until the late 1980s when the simulation of chaotic
dynamical systems became common place and when the role
of cryptography in IT became increasingly important. Since
the start of the 1990s, an increasing number of publications
have considered the use of chaos in cryptography, e.g. [74][78]. These have included schemes based on synchronized
chaotic (analogue) circuits, for example, which belong to
the eld of steganography and secure radio communication
[79]. Over the 1990s cryptography started to attract a variety
of scientists and engineers from diverse elds who started
exploiting dynamical systems theory for the purpose of encryption. This included the use of discrete chaotic systems
such as the cellular automata, Kolmogorov ows and discrete
afne transformations in general to provide more efcient
encryption schemes [80]-[83]. Since 2000, the potential of
chaos-based communications, especially with regard to spread
spectrum modulation, has been recognized. Many authors have
described chaotic modulations and suggested a variety of
electronics based implementations, e.g. [76]-[79]. However,
the emphasis has been on information coding and information
hiding and embedding. Much of this published work has been
of theoretical and some technological interest with work being
undertaken in both an academic and industrial research context
(e.g. [84]-[90]). However, it is only relatively recently that
the application of chaos-based ciphers have been implemented
in software and introduced to the market. One example of
this is the basis of the authors own company - CrypsticTM
Limited - in which the principle of multi-algorithmicity using
chaos-based ciphers [11], [91] has been use to produce metaencryption engines that are mounted on a single, a pair or a
group of ash (USB - Universal Serial Bus) memory sticks.
Some of these memory sticks have been designed to include
a hidden memory accessible through a covert procedure (such
as the renaming - by delation - of an existing le or folder)
from which the encryption engine(s) can be executed.
Consider an algorithm that outputs a number stream which
can be ordered, chaotic or random. In the case of an ordered
number stream (those generated from a discretized piecewise
continuous functions for example), the complexity of the eld
is clearly low. Moreover, the information and specically the
information entropy (the lack of information we have about the
exact state of the number stream) is low as is the information
content that can be conveyed by such a number stream.
A random number stream (taken to have a uniform PDF,
19 Hopf, Eberhard F. F, (1902-1983), an Austrian mathematician who made
signicant contributions in topology and ergodic theory and studied the mixing
in compact spaces, e.g. On Causality, Statistics and Probability, Journal of
Mathematics and Physics, 13, 51-102, 1934.
x X, k K
48
corresponds to Shannons original requirements for an encryption system in the late 1940s. In both chaos and cryptography
we are dealing with systems in which a small variation of any
variable changes the outputs considerably.
A. Stream Cipher Encryption
The use of discrete chaotic elds for encrypting data can
follow the same basic approach as used with regard to the
application of pseudo random number generating algorithms
for stream ciphers. Pseudo chaotic numbers are in principle,
ideal for cryptography because they produce number streams
that are ultra-sensitive to the initial value (the key). However,
instead of using iterative based maps using modular arithmetic
with integer operations, here, we require the application of
nonlinear maps using oating point arithmetic. Thus, the rst
drawback concerning the application of deterministic chaos
for encryption concerns the processing speed, i.e. pseudo random number generators typically output integer streams using
integer arithmetic whereas pseudo chaotic number generators
produce oating point streams using oating point arithmetic.
Another drawback of chaos based cryptography is that the
cycle length (i.e. the period over which the number stream
repeats itself) is relatively short and not easily quantiable
when compared to the cycle length available using conventional PRNGs, e.g. additive generators, which commence by
initialising an array xi with random numbers (not all of which
are even) so that we can consider the initial state of the
generator to be x1 , x2 , x3 , ... to which we then apply
xi = (xia + xib + ... + xim )mod2n
where a, b, ..., m and n are assigned integers20 , have very long
cycle lengths of the order of 2f (255 1) where 0 f n
and linear feedback shift registers with the form
xn = (c1 xn1 + c2 xn2 + cm xnm )mod2k
which, for specic values of c1 , c2 , ...cm have cycle lengths
of 2k .
The application of deterministic chaos to encryption has two
distinct disadvantages relative to the application of PRNGs.
Another feature of IFS is that the regions over which chaotic
behaviour can be generated may be limited. However, this
limitation can be overcome by designing IFS with the specic
aim of increasing the range of chaos. One method is to use
well known maps and modify them to extend the region of
chaos. For example, the Matthews cipher is a modication of
the logistic map to [92]
xi+1 = (1 + r) 1 +
1
r
xi (1 xi )r , r (0, 4].
Fig. 18.
49
block and the maps that are used to encrypt the plaintext
over each block. Thus, suppose we have designed a data
base consisting of 100 chaotic maps, say, consisting of IFS
f1 , f2 , f3 , ..., f100 , each of which generates a oating point
number steam through the operation
xi+1 = fm (xi , p1 , p2 , ...)
where the parameters p1 , p2 , ... are pre-set or hard-wired
to produce chaos for any initial value x0 (0, 1) say. An
algorithm selection key is then introduced in which two
algorithms (or possibly the same algorithm) are chosen to
drive the block cipher - f50 and f29 say, the key in this
case being (50, 29). Here, we shall consider the case where
map f50 determines the algorithm selection and map f29
determines the block size. Map f50 is then initiated with the
key 0.26735625 say and map f29 with the key 0.65376301 say.
The output from these maps (oating point number streams)
are then normalized, multiplied by 100 and 1000, respectively,
for example, and then rounded to produce integer streams with
values ranging from 1 to 100 and 1 to 1000, respectively. Let
us suppose that the rst few values of these integer streams
are 28, 58, 3, 61 and 202, 38, 785, 426, respectively. The block
encryption starts by using map 28 to encrypt 202 elements of
the plaintext using the key 0.78654876 say. The second block
of 38 elements is then encrypted using map 58 (the initial
value being the last oating point value produced by algorithm
28) and the third block of 785 elements is encrypted using
algorithm 3 (the initial value being the last oating point value
produced by algorithm 58) and so on. The process continues
until the plaintext has been fully encrypted with the session
key (50,29,0.26735625,0.65376301,0.78654876).
Encryption is typically undertaken using a binary representation of the plaintext and applying an XOR operation
using a binary representation of the cipher stream. This can be
constructed using a variety of ways. For example, one could
extract the last signicant bits from the oating point format
of xi . Another approach, is to divide the oating point range
of the cipher into two compact regions and apply a suitable
threshold. For example, suppose that the output xi from a map
operating over a given block consists of oating point value
between 0 and 1, then, with the application of a threshold of
0.5, we can consider generating the bit stream
b(xi ) =
1, xi (0.5, 1];
0, xi [0, 0.5).
Fig. 19. Probability density function (with 100 bins) of the output from the
logistic map for 10000 iterations.
Fig. 20. Illustration of the effect of using multiple algorithms for generating
a stream cipher on the computational Energy required to attempt a brute force
attack.
the form
1,
xi [T, T + + );
b(xi ) = 0,
xi [T , T );
1, otherwise.
where T is the threshold and + and are those values
which yield an output stream that characterizes (to a good
approximation) a uniform distribution. For example, in the
case of the logistic map T = 0.5 and + = =
0.2. This aspect of the application of deterministic chaos to
cryptography, together with the search for a parameter or set
of parameters that provides full chaos for an invented map,
determines the overall suitability of the function that has been
invented for this application.
The ltering of a chaotic eld to generate a uniformly distributed output is equivalent to maximizing the entropy of the
cipher stream (i.e. generating a cipher stream with a uniform
PDF) which is an essential condition in cryptography. In terms
of cryptanalysis and attack, the multi-algorithmic approach to
designing a block cipher introduces a new dimension to the
problem. The conventional problem associated with an attack
on a symmetric cipher is to search for the private key(s) given
knowledge of the algorithm. Here, the problem is to search
not only for the session key(s), but the algorithms they drive
as illustrated in Figure 20.
One over-riding issue concerning cryptology in general, is
that algorithm secrecy is weak. In other words, a cryptographic
system should not rely of the secrecy of its algorithms and
50
respect to the required (maximum entropy) performance conditions through implementation of the appropriate threshold
parameters T and . The design is based on applying the
following basic steps:
XII. C RYPSTICTM
r
3.3725
3.17
2.816
0.9999
0.9990
T
0.5
0.5
0.6
0.5
0.6
+
0.3
0.25
0.3
0.3
0.25
0.3
0.35
0.2
0.3
0.25
51
Fig. 22.
52
any party, is lost, then a new pair of Crypstics are issued with
new encryption engines unique to both parties. In addition to
a two-party user system, Crypstics can be issued to groups of
users in a way that provides an appropriate access hierarchy
as required.
XIII. D ISCUSSION
The material discussed in this paper has covered some of
the basic principles associated with cryptography in general,
including the role of diffusion and confusion for designing
ciphers that have no statistical bias. This has been used as a
guide in the design of ciphers that are based on the application
of IFS exhibiting chaotic behaviour. The use of IFS allow for
the design of encryption engines that are multi-algorithmic,
each algorithm being based on an IFS that is invented, subject
to the condition, that the output stream has a uniform PDF.
The principle of multi-algorithmicity has been used to develop
a new product - CrypsticTM - that is based on the following:
(i) a multi-algorithmic block encryption engine consisting of
a unique set of IFS; (ii) maximum entropy conversion to a bit
stream cipher; (iii) a key that is determined by the le name to
be encrypted/decrypted. The approach has passed all statistical
tests [11] recommended by National Institute of Standards and
Technology (NIST) [95].
Access and use of the encryption engine is based on
utilizing an commercial-off-the-shelf USB ash memory via
a combination of camouage, obfuscation and disinformation
in order to elude any potential attacker. The approach has
been based on respecting the following issues: (i) security is a
process not a product; (ii) never underestimate the enemy; (iii)
the longer that any cryptosystem, or part thereof, remains of
the same type with the same function, the more vulnerable
the system becomes to a successful attack. Point (iii) is a
singularly important feature which CrypsticTM overcomes by
utilizing a dynamic approach to the design and distribution of
encryption engines.
A. Chaos Theory and Cryptography
We have discussed cryptography in the context of chaos
theory and there is clearly a fundamental relationship between
cryptography and chaos. In both cases, the object of study is
a dynamic system that performs an iterative nonlinear transformation of information in an apparently unpredictable but
deterministic manner. In terms of chaos theory, the sensitivity
to the initial conditions together with the mixing property
ensures cryptographic confusion and diffusion, as originally
suggested by Shannon. However, there are also a number
of conceptual differences: (i) chaos theory studies dynamic
systems dened on an innite state space (e.g. vectors of
real numbers or innite binary strings), whereas cryptography
relies on a nite-state machine and all chaos models implemented on a computer are approximations, i. e. pseudo-chaos;
(ii) chaos theory studies the asymptotic behaviour of a nonlinear system (the behaviour as the number of iteration approach
innity when the Lyapunov dimension can be quantied - see
Appendix II in which the denition of the Lyapunov dimension
is based on N ), whereas cryptography focuses on the
Cryptography
Pseudo-chaotic system
Nonlinear transform
Finite states
Finite iterations
Plaintext
Ciphertext
key
Confusion
Diffusion
53
54
Ciphertext
Covertext
Stegotext
Transmission
Note that this approach does not necessarily require the use
of plaintext to ciphertext conversion as illustrated above and
that plaintext can be converted into stegotext directly. For
example, a simple approach to this is to use a mask to delete
all characters in a message except those that are to be read by
the recipient of the message. Apart from establishing a method
of exchanging the mask, which is equivalent to the key in
cryptography, the principal problem with this approach is that
different messages have to be continuously invented in order
to accommodate hidden messages and that these inventions
must appear to be legitimate. However, the wealth of data
that is generated and transmitted in todays environment and
the wide variety of formats that are used, means that there is
much greater potential for exploiting steganographic methods
than were previously available. In other words, the wealth of
information now available has generated a camouage rich
environment in which to operate and one can attempt to
hide plaintext or ciphertext (or both) in a host of data types,
including audio and video les and digital images. Moreover, by understanding the characteristics of a transmission
environment, it is possible to conceive techniques in which
information can be embedded in the transmission noise, i.e.
where natural transmission noise is the covertext. There are
of course a range of counter measures - steganalysis - that
can be implemented in order to detect stegotext. However, the
techniques usually requires access to the covertext which is
then compared with the stegotext to see if any modications
have been introduced. The problem is to nd ways of obtaining
the original stegotext which is equivalent to a plaintext attack.
C. Hiding Data in Images
The relatively large amount of data contained in digital
images makes them a good medium for undertaking steganography. Consequently digital images can be used to hide
messages in other images. A colour image typically has 8
bits to represent the red, green and blue components. Each
colour component is composed of 256 colour values and the
modication of some of these values in order to hide other
data is undetectable by the human eye. This modication is
often undertaken by changing the least signicant bit in the
binary representation of a colour or grey level value (for grey
level digital images). For example, the grey level value 128
55
P (x) exp(ikx)dx
P (k) =
X/2
1
exp(ikx)dx = sinc(kX/2)
X
=
X/2
Q(x) =
i=1
1
k
=1
1
1
1
(k)3 + (k)5 (k)7 + . . .
3!
5!
7!
1
1
1
(k)2 + (k)4 (k)6 + . . .
3!
5!
7!
1
1
1
(k)2 (k)4 + (k)6 . . .
3!
5!
7!
=1N
+
1
1
1
(k)2 + (k)4 (k)6 + . . .
3!
5!
7!
N (N 1)
2!
1
1
1
(k)2 (k)4 + (k)6 . . .
3!
5!
7!
(k)2
(k)4
(k)6
+
...
3!
5!
7!
N (N 1)(N 2)
3!
+...
A PPENDIX I
C ENTRAL L IMIT T HEOREM FOR A U NIFORM
D ISTRIBUTION
We study the effect of applying multiple convolutions of the
uniform distribution
P (x) =
1
X,
0,
2 2
=1N
... +
| x | X/2;
otherwise
i=1
6
exp(6x2 /XN )
N
where Pi (x) = P (x), n and N is large. by considering
the effect of multiple convolutions in Fourier space (through
N (N 1)
2!
4 k 4
6 k 6
+ ...
2
(3!)2
3!5!
N (N 1)(N 2)
3!
=1
k
4 k 4
6 k 6
+N
k
3!
5!
7!
N 2 2
k +
3!
6 k 6
+ . . . + ...
(3!)3
N 4 N (N 1) 4
+
k4
5!
2!(3!)2
N 6 N (N 1) 6 N (N 1)(N 2) 6
+
+
k6 + . . .
7!
3!5!
3!(3!)3
1 2 4
1
c k c3 k 6 + . . .
2!
3!
56
exp(y 2 )dy =
1
N 2 ,
6
1
1
N + N (N 1) 4
120
72
c=
1 2
c =
2
or
c2 =
and
1 3
c =
6
or
4 ,
2n + O(N n1 2n ).
X
2,
X
N k2
24
1
exp XN k 2 exp(ikx)dk
24
exp
XN
k
24
24 ix
XN 2
+ix
6 6x2
e XN
XN
6
XN
2
6
XN
XN k
ix
6 2
6
.
XN
By Cauchys theorem
1
I=
Noting that
6 6x2
e XN
XN
ez dz =
n+1
ln
n=1
n+1
= N .
ln
= lim
n=1
n+1
ey dy
+ix
= c exp(n)
we can write
1
=
n+1
1
I=
2
ln
fn+1 = F (fn ) = f +
then
Q(x) exp
1
=
2
A PPENDIX II
T HE LYAPUNOV D IMENSION
6
exp[6x2 /(XN )]
XN
Pi (x)
i=1
1
1
1
N3
N2 +
N
216
1080
2835
cn =
for large N .
N
N (N 1) N (N 1)(N 2)
+
+
5040
720
1296
c3 =
Q(x) =
1 2
1
N N
36
90
6x2
6
e XN
XN
57
1
1
N
M
M
N
where
1
N N
, M > N ;
, M < N .
N
N = lim
ln
n=1
n+1
ACKNOWLEDGMENT
Some aspects of this paper are based on the research and
PhD Theses of the following former research students of the
author: Dr S Mikhailov, Dr N Ptitsyn, Dr D Dubovitski, Dr
K Mahmoud, Dr N Al-Ismaili and Dr R Marie. The author is
grateful to the following for their contributions to CrypsticTM
Limited: Mr Bruce Murray, Mr Dean Attew, Major General
John Holmes, Mr William Kidd, Mr David Bentata and Mr
Alan Evans.
R EFERENCES
[1] S. Singh, The Code Book: The Evolution of Secrecy from Mary, Queen
of Scots to Quantum Cryptography, Doubleday, 1999.
[2] B. Schneier, Beyond Fear: Digital Security in a Networked World, Wiley,
2000.
[3] B. Schneier, Thinking Sensibly about Security in an Uncertain World,
Copernicus Books, 2003.
[4] N. Ferguson and B. Schneier B, Practical Cryptography, Wiley, 2003.
[5] A. J. Menezes, P. C. van Oorschot and S. A. Vanstone, Handbook of
Applied Cryptography, CRC Press, 2001.
[6] B. Schneier, Applied Cryptography, Second Edition Wiley, 1996.
[7] J. Buchmann, Introduction to Cryptography, Springer 2001.
[8] O. Goldreich, Foundations of Cryptography, Cambridge University
Press, 2001.
[9] J. Hershey, Cryptography Demystied, McGraw-Hill, 2003.
[10] H. F. Gaines, Cryptanalysis, Dover, 1939.
[11] N. V. Ptitsyn, Deterministic Chaos if Digital Cryptography, PhD Thesis,
De Montfort University, 2003.
[12] http://vl.fmnet.info/safety/
[13] http://www.amazon.com/Network-Security-process-not-product
[14] http://en.wikipedia.org/wiki/Enigma Machine
[15] S. Katzenbeisser and F. Petitcolas, Information Hiding Techniques for
Steganography and Digital Watermarking, Artech House, 2000.
[16] N. F. Johnson, Z. Duric and S. Jajodia, Information Hiding: Steganography and Watermarking -Attacks and Countermeasures, Kluwer Academicf Publishers, 2001.
[17] G. Kipper, Investigators Guide to Steganography, CRC Press, 2004.
[18] Hackers Black Book, http://www.hackersbook.com
[19] A. N. Shulsky and G. J. Schmitt, Silent Warefare: Understanding the
World of Intelligence, Brassey, 2002.
[20] R. Hough, The Great War at Sea, Oxford University Press, 1983
[21] P. G. Halpern, A Naval History of World War One, Routledge, 1994.
[22] R. A. Ratcliff, Delusions of Intelligence, Cambridge University Press,
2006.
[23] R. A. Woytak, On the Boarder of War and Peace: Polish Intelligence
and Diplomacy and the Origins of the Ultra-Secret, Columbia University
Press, 1979.
[24] W. Kozaczuk, Enigma: How the German Machine Cipher was Broken,
and how it was Read by the Allies in World War Two, University
Publications of America, 1984.
[25] B. Booss-Bavnbek and J. Hoyrup, Mathematics at War, Birkh user,
a
2003.
[26] B. J. Copelend, Colossus: The Secrets of Bletchley Parks Code Breaking
Computers, Oxford University Press, 2006.
[27] A. Stripp and F. H. Hinsley, Codebreakers: The Inside Story of Bletchley
Park, Oxford University Press, 2001
[28] http://www.gchq.gov.uk/
58
Regular Paper
Original Contribution
59
I. INTRODUCTION
Metal 2: antenna
Metal 3: ground
60
Rigid
Flex
Rigid
(b) Bendable property.
Microstrip lines
Metal 1
Metal 2
Ground
Fig. 1. Printed circuit board structure: (a) detailed cross-section, and (b)
bendable property.
A. Monopole antenna
Fig. 2 shows a circular monopole antenna integrated in the
flex-rigid substrate. As shown in Fig. 2, the ground plane is
integrated in the rigid part. The radiating antenna element is
placed entirely on the flex part of the substrate. The circular
antenna geometry provides good omni-directionality [11].
Polyimide foil
Differential feed-line
Port 2, Metal 2
Port 3, Metal 2
/2
Circular
monopole
antenna
Metal 2 connected
to ground
30 mm
Ground plane
(rigid part)
Single-ended
feed-line
B. Dipole antenna
Fig. 3 shows a circular dipole antenna integrated in the flexrigid substrate. The radiating antenna element is placed
entirely on the flex part of the substrate. Furthermore, the
balun is integrated in the rigid part of the substrate. The balun
utilizes broadside-coupled microstrips [13].
Circular dipole
antenna
Polyimide
foil
Rigid part
3
2
Single ended
feed-line
Fig. 3. Circular dipole antenna.
VSWR
38 mm
Frequency (GHz)
10
11
61
B. Dipole antenna
Fig. 6 shows VSWR simulation of a circular dipole antenna
on the flex-rigid substrate. Fig. 6a shows the VSWR
simulation result without the balun, and Fig. 6b shows the
simulation result with the balun. It is seen that the circular
dipole antenna has a wide impedance bandwidth using the
suggested flex-rigid structure. Furthermore, it is seen that the
balun is the component limiting the bandwidth.
=0
0
0
330
30
-5
-10
-15
300
60
-20
-25
-30 270
90
-25
-20
-15
240
120
-10
4
VSWR
-5
150
180
2
1
210
=0
10
Frequency (GHz)
330
30
-5
-10
-15
300
60
-20
-25
-30 270
90
-25
VSWR
-20
-15
240
120
-10
-5
2
1
210
10
150
180
Frequency (GHz)
330
30
-5
-10
-15
=0
0
0
300
60
-20
-25
-30 270
90
-25
-20
-15
240
120
-10
-5
0
210
150
180
62
S21
S31
-10
-20
-30
-40
-50
Frequency (GHz)
Frequency (GHz)
(a) Simulation of forward transmission.
Phase ()
S21
200
150
100
50
0
-50
-100
-150
-200
S31
Frequency (GHz)
(c) Simulation of phase balance.
Fig. 8. Balun: (a) simulated forward transmission, single-differential port, (b)
simulated forward transmission, single-single port, and (c) simulated phase.
IV. DISCUSSION
Simulation results show that the monopole antenna has a
wide operational frequency band. The monopole antenna is
smaller than the dipole antenna but needs a quite large ground
plane. On the contrary, the dipole antenna only requires a
small ground plane beneath the balun, but the total size is
larger. The simulations of the circular dipole antenna indicate
that the antenna has a typical radiation pattern as expected
from a common dipole antenna. It is also observed (Fig. 6)
that the balun limits the bandwidth if the dipole antenna is
used with a single-ended port. However, this antenna with the
balun (Fig. 6b) has a good bandpass characteristic, covering
the frequency band 3.1-4.8 GHz required by the Mode 1
UWB specification.
V. CONCLUSION
Simulations show that a circular monopole antenna can be
implemented on a flex-rigid substrate with VSWR<2.0 over
the entire UWB frequency bandwidth (3.1-10.6 GHz).
Moreover, a circular dipole antenna implemented using the
flex-rigid substrate can cover the Mode 1 UWB frequencybandwidth (3.1-4.8) at VSWR<1.54 without a balun and
VSWR<1.68 with a balun.
63
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Regular Paper
Original Contribution
64
I. INTRODUCTION
Wireless communication systems require more and
more spectrums, while single module solutions grow in
popularity. An integratable low-profile antenna structure
limits design options, regarding various performance
aspects. For instance, the fractional bandwidth is limited.
A monofilar spiral antenna is known to have wide
bandwidth, i.e., good fractional bandwidth compared to
other planar antennas [1-4]. One known drawback is the
physical size of a plane monofilar spiral antenna, i.e., the
single arm spiral is physically large in terms of its
diameter [3, 4]. Furthermore it is known that even though
the spiral may have very wide impedance bandwidth,
dispersion may cause problems for using spiral antennas
in wideband impulse-systems [1, 2, 5]. However, in
multi-band systems the antennas need only to transmit
and receive a signal with limited bandwidth, i.e., the
width of a single multi-band pulse determines the
maximum frequency spectrum used simultaneously [6].
Flat antenna performance is desired in UWB systems [7].
A terminating resistor at the arm end may be used to
reduce reflections in order to flatten the antenna input
impedance, but it causes signal loss [5]. Our approach is
instead to optimize the physical properties of the antenna
so that the input impedance becomes acceptable close to
a 50 real load. The antennas are intended to be
integrated on a multilayer printed circuit board (PCB). A
concept to reduce the total size of the complete design is
to place components on a layer below the antenna as
proposed in our previous publications [3, 6]. This
requires that the antenna structure is compatible with an
overall module technology. Ultra wideband radio (UWB)
Radius (r)
75 mm
75 mm
75 mm
50 mm
50 mm
50 mm
30 mm
30 mm
30 mm
Layer
RCC
RO4350B
RCC
All layers
All layers
Top and bottom
RO4350B (core)
All layers
All layers
Dimension
Height, h=0.05 mm
(1x) h=1.524 mm
h=0.05 mm
3.480.05
0.004
0.025 mm
0.018 mm
5.8x107 S/m
0.001 mm
Antenna
layer
RO4350B
Air core
hair
Component PCB
Bottom
Component layer
(a) Cross-section.
Component layer
Radiating zone
Antenna
Turn distance, s
Open end
theta ()
Radius, r
Y
phi ()
n/2
Feed point
Fig. 3.
Layout of a monofilar spiral antenna, oriented in the XY
plane, =0.
Via
Antenna PCB
Top
Gnd
(b) Cross-section.
Fig. 2.
Cross-sections of substrate: (a) four layer PCB, and (b) air
core substrate structure.
D. Methods
Design and simulation were done using the ADS
Momentum, which is a 2.5D planar electromagnetic
(EM) simulator. The antennas simulated are of planar
structure therefore the Momentum-engine is a reliable
choice, despite the lack of surface roughness
consideration [12]. The simulations were done with an
infinite ground plane, imposed by ADS.
Voltage standing wave ratio (VSWR) measurements
were done with a Rhode&Schwartz ZVM vector network
analyzer (NWA). Radiation measurements were done in
an anechoic antenna chamber. An HP 8510C vectorNWA together with an HP 8517B S-parameter test set
and the MiDAS 2.01g software was used to measure the
antennas. Linearly polarized horn antennas with known
antenna gain were used as reference antennas.
III. SIMULATION RESULTS
A. Monofilar Spiral Antennas of 50
Fig. 4 shows the layout and VSWR simulation of a
monofilar spiral antenna. To optimize performance, the
real part of the characteristic impedance was calculated to
be 50 , i.e., the line width was calculated to be 3.43 mm
with a substrate thickness of 1.5 mm. Three radii with
three different turn distances for each size of the antenna
were designed and simulated. It is shown that in all three
simulations in Figs. 4a-4c, a more dense turn distance
(s) provides a high fractional bandwidth. As mentioned
in the introduction, in theory the infinite number of turns
gives the infinite fractional bandwidth. However, owing
to the chosen line width and the need of spacing between
turns and limited number of turns, the bandwidth is
limited in our case as seen in Figs. 4a-4c. Fig. 4d shows a
rather good match between simulated and measured data.
5
s=6.5 mm
s=4.5 mm
4
VSWR
65
s=5.5 mm
2
3
2
1
2
Frequency (GHz)
10
11
66
s=4.5 mm
s=5.5 mm
s=6.5 mm
3
2
10
11
Frequency (GHz)
s=6.5 mm
-2
-4
s=4.5 mm
Gain (dBi)
VSWR
s=5.5 mm
5
6
7
8
Frequency (GHz)
(a) Antenna gain, r=30 mm.
10
11
Gain (dBi)
VSWR
4
s=4.5 mm
s=5.5 mm
s=6.5 mm
-4
10
11
s=6.5 mm
-2
2
1
s=5.5 mm
s=4.5 mm
5
6
7
8
Frequency (GHz)
(b) Antenna gain, r=50 mm.
10
11
Frequency (GHz)
VSWR
Simulated
Gain (dBi)
s=4.5 mm
Measured
s=5.5 mm s=6.5 mm
-2
-4
5
6
7
8
Frequency (GHz)
(c) Antenna gain, r=75 mm.
2
1
2
10
10
11
5
6
7
8
9
Frequency (GHz)
(d) Antenna gain, air core, r=75 mm, s=4.5 mm.
10
11
11
Gain (dBi)
2.2 mm air
10
8
6
4
1.5 mm air
2
10
Gain (dBi)
Fig. 4.
VSWR simulation of Monofilar spiral antennas of 50
input impedance: (a) r=30 mm, (b) r=50 mm, (c) r=75 mm, and (d)
simulation and measurement comparison, r=75 mm, s=4.5 mm.
12
Frequency (GHz)
2x-air
1x loss-less
2x
-5
1x
-10
2
5
6
7
8
Frequency (GHz)
10
11
67
Antenna 2.3
VSWR
3
2
3
Antenna 3.1
6
7
8
Frequency (GHz)
s=4.5 mm
5
Frequency (GHz)
(b) Maximum antenna gain.
4
3
10
-4
s=6.5 mm
-2
-6
Fig. 7.
VSWR and antenna gain measurements of two spiral
antennas with a radius of 50 mm: (a) turn distance 6.5 mm, and (b)
maximum antenna gain measured from three sweeps with =0, 45,
and 90, respectively.
3
6
7
8
Frequency (GHz)
10
2
s=5.5 mm
0
-2
-4
-6
s=4.5 mm
5
Frequency (GHz)
(b) Maximum antenna gain.
Fig. 6.
VSWR and antenna gain measurement of two spiral antennas
with a radius of 30 mm: (a) turn distance 4.5 mm, and (b) maximum
antenna gain measured from three sweeps with =0, 45, and 90,
respectively.
3
2
3
6
7
8
Frequency (GHz)
10
VSWR
Gain (dBi)
Gain (dBi)
3
2
3
6
7
8
Frequency (GHz)
10
68
Antenna 1.3
Antenna 1.1
5
VSWR
VSWR
2
3
6
7
8
Frequency (GHz)
3
2
10
4.5 GHz
6
7
8
Frequency (GHz)
3.5 GHz
(dB)
-5
Antenna 1.1
5
-10
6.5 GHz
-20
-25
-60
-40
VSWR
-15
5.5 GHz
0
20
Theta ()
(d) Measured E-field characteristics.
-20
40
60
4
3
2
s=5.5 mm
-2
6
7
8
Frequency (GHz)
10
s=4.5 mm
-4
3.5 GHz
-5
-6
(dB)
Gain (dBi)
10
s=6.5 mm
5
Frequency (GHz)
(e) Maximum antenna gain.
Fig. 8.
VSWR and antenna gain measurements of three spiral
antennas with a radius of 75 mm: (a) turn distance 4.5 mm, (b) turn
distance 5.5 mm, (c) turn distance 6.5 mm, (d) measured radiation
characteristics of the antenna with 4.5-mm turn distance, =90, and (e)
maximum antenna gain measured from three sweeps with =0, 45,
and 90, respectively.
-10
-15
4.5 GHz
-20
-25
-60
-40
0
20
Theta ()
(c) Measured E-field characteristics.
-20
40
60
Gain (dBi)
=0
4
2
0
=90
-2
=45
5
Frequency (GHz)
(d) Measured antenna gain, 2.2 mm air.
69
Gain (dBi)
8
2.2 mm air
6
4
2
1.5 mm air
3
5
Frequency (GHz)
(e) Maximum antenna gain.
Fig. 9.
Air core antenna measurements, r=75 mm: (a) VSWR, 1.5
mm RO4350B core + 1.5 mm air, (b) VSWR, 0.8 mm RO4350B core +
2.2 mm air, (c) measured E-field of the 1.5 mm air core antenna, =90,
(d) measured antenna gain, 0.8 mm RO4350B + 2.2 mm air, and (e)
maximum antenna gain measured from three sweeps with =0, 45,
and 90, respectively.
V. DISCUSSIONS
A study of monofilar spiral antennas for UWB is
presented in this paper. Antennas covering the entire
UWB frequency band of 3.1-10.6 GHz can be realized
with various SWR numbers, but the physical size limits
the fractional bandwidth and determines the centre
frequency. The turn distance was found to be a crucial
factor for the antenna gain, and the optimal turn distance
varies with the radius of the antenna. However,
increasing the turn distance decreases the fractional
bandwidth so a trade-off must be made. Some
missmatching between simulated and measured results is
seen, especially regarding to fractional bandwidth and the
center frequency, see Fig. 4. The difference may come
from the fact that the simulations are done with an
infinite ground-plane while the prototype ground-plane is
only 30 mm larger than the spiral. Moreover the
simulations are done with a more limited number of
points. Momentum is as mentioned not a full 3D
simulator. More advanced EM simulators like Ansoft
HFSS or CST Microwave studio may give more accurate
results.
A single arm spiral antenna can have good
performance over a wide frequency range. However, the
radiation pattern is dependent of the frequency so that the
performance in a certain direction varies throughout the
frequency-range (see Figs. 8d and 9c). The performance
varies owing to the fact that the beam moves around the
z-axis (see Fig 3.) such that several modes interact.
Furthermore, the high frequency ripple in VSWR affects
performance and causes the antenna gain to vary
throughout the frequency-range. These variations are due
to the physical properties of the single arm spiral that
introduces non-real input impedance. Moreover, some
inductance might be introduced by the SMA connector,
i.e., the feeding via through the PCB is not impedance
controlled. A more uniform and high gain can be
achieved when an air core is used (see Fig. 9). This is due
to the fact that with a high gain less signal power reaches
the arm end, i.e., the backward travelling wave gets
suppressed relative to the forward travelling wave.
Furthermore, the air core effectively reduces the high
frequency variation in VSWR. In general an additional
VI. CONCLUSIONS
A study of monofilar spiral antenna for UWB has been
conducted. How the spiral radius, turn distance, and
substrate dissipation factor and thickness affect the
antenna gain were shown. Moreover, antenna gain
improvement with an air gap between the ground plane
and the antenna plane has been shown.
Monofilar spiral antenna solutions for UWB optimized
for an input source impedance of 50 was designed,
simulated and measured. It is shown that a planar
monofilar spiral antenna implemented on a PCB is
suitable for RF module designs.
Monofilar spiral antennas with a radius of 30, 50 and
75 mm were designed, simulated and measured. One
monofilar spiral antenna of these sizes covers the entire
UWB frequency band (3.1-10.6 GHz) at a VSWR<4, <3,
<2, respectively.
Monofilar spiral antennas with a radius of 75 mm
consisting of an air core were also designed, simulated
and measured. One monofilar spiral antenna with the air
core covers the entire UWB frequency band at
VSWR<1.8. The VSWR response is also more stable.
The antenna gain is higher than that from any of the
monofilar antennas without an air core.
REFERENCES
[1]
70
[11] H. Nakano, J. Eto, Y. Okabe, J. Yamauchi, Tilted- and AxialBeam Formation by a Single-Arm Rectangular Spiral Antenna
With Compact Dielectric Substrate and Conducting Plane, IEEE
Trans. on Antennas and Propagation, vol. 50, no. 1, Jan. 2002.
[12] A. P. Jenkins, A. M. Street, D. Abbott, Filter design using CAD.
II. 2.5-D simulation, Effective Microwave CAD, IEE
Colloquium, no. 1997/377, pp. B1/1-B1/5, Dec. 1997.
[13] P. de Maagt, R. Gonzalo, Y.C. Vardaxoglou, J.-M. Baracco,
"Electromagnetic bandgap antennas and components for
microwave and (Sub)millimeter wave applications," IEEE Trans.
on Antennas and Propagation, vol.51, no.10, pp. 2667- 2677, Oct.
2003.
[14] J.-M. Baracco, M. Paquay, P. de Maagt, "An electromagnetic
bandgap curl antenna for phased array applications," IEEE Trans.
on Antennas and Propagation, vol.53, no.1, pp. 173- 180, Jan.
2005.
Magnus Karlsson was born in Vstervik, Sweden in
1977. He received the M.Sc. and Licentiate of
Engineering from Linkping University in Sweden, in
2002 and 2005, respectively.
In 2003 he started his Ph.D. study in the
Communication Electronics research group at
Linkping University.
His main work involves
wideband antenna-techniques and wireless communications.
Shaofang Gong was born in Shanghai, China, in 1960.
He received the B.Sc. degree from Fudan University in
Shanghai in 1982, and the Licentiate of Engineering
and Ph.D. degree from Linkping University in
Sweden, in 1988 and 1990, respectively.
Between 1991 and 1999 he was a senior researcher at
the microelectronic institute Acreo in Sweden. From
2000 to 2001 he was the CTO at a spin-off company from the institute. Since
2002 he has been full professor in communication electronics at Linkping
University, Sweden. His main research interest has been communication
electronics including RF design, wireless communications and high-speed
data transmissions.
Regular Paper
Original Contribution
71
J. C. Chedjou , K. Kyamakya
Institute for Smart-Systems
Technologies,
University of Klagenfurt
Klagenfurt, Austria
e-mails: Jean.Chedjou@uni-klu.ac.at
, Kyandoghere.Kyamakya@uniklu.ac.at
Abstract
INTRODUCTION
72
73
that analog techniques might bring some fresh air to the theory
of computation in the field of non linear dynamics.
Practical problems and advices
Concerning the effects of nonlinearity in an unspecified
sinusoidal oscillator, we mention that they come mainly from
the discrete components (bipolar junction transistor (BJT),
operational amplifier (Opamp), or analog circuits multipliers)
constituting the gain elements. Thus, the effects of nonlinearity
will be differently perceived depending upon the type of gain
element. Some recent scientific contributions [1-4] have
presented analytical approaches to explain the nonlinear
behavior exhibited by classical oscillators using bipolar
junction transistors. The exponential dependence of the current
flowing through the collector of the BJT with respect to the
voltage drop between its base-emitter regions was presented as
the origin of nonlinearity in the oscillators. Bifurcation
structures were presented, showing the coexistence between the
regular states (limit cycles) and the irregular states (chaotic) of
the oscillators.
B.
Offset phenomenon:
74
Saturation phenomenon:
Power transfer:
Time scaling:
IMPORTANT CONTRIBUTION
75
A.
Circuit description
Fig. 1 is the design of the shunt structure of the Colpitts
oscillator under investigation. The bipolar junction transistor
Q1 used in the common-base-configuration, plays the role of
nonlinear gain element. The feedback network consists of the
inductor L and the capacitors C1 and C 2 . These capacitors
act as voltage divider.
f0=
1
2
C1 + C 2
LC 1C 2
(1)
Z input =
h11
h21 + 1
(2)
76
B.
Circuit modeling
V BE
(3a)
I E = I S exp
1
VT
(3b)
IC = FI E
where I S is the saturation current of the base-emitter junction
and VT = 26mV is the value of the thermal voltage at room
temperature. F denotes the common-base short-circuit
forward current gain of the BJT.
We emphasize that the idea is to select the simplest
possible model which maintains the essential features [16]
exhibited by the real circuit. If we denote by I L the current
flowing through the inductor L , and
voltages across the capacitors
Vi (i = 1, 2, 3) the
dI L
= V1 + V2
dt
(4a)
x=
where
dV2
= VCC V1 V2 V3 +
dt
R[(1 F )I E I L I 0 ]
dV
RC3 3 = VCC V1 V2 V3 R F I E
dt
F = 1 (this is
100 F 200 )
with
dimensionless quantities:
(5b)
(5c)
(5d)
= LC 2 , =
C2
L
, 1 =
,
C1
C2
VCC
IS
C2
=
, =
,
, =
, and
C3
VT
R
VT
I0
=
, Eqs. (4) can be transformed into the following
VT
2 =
dx
= y+z
(6a)
d
dy
= 1 .[ x + ( y z ) f ( z )]
(6b)
d
dz
= .( y z ) x
(6c)
d
d
= 2 .[ .( y z ) f ( z )]
(6d)
d
where f ( z ) is the exponential function derived from Eqs. (3)
f ( z ) = ( exp( z ) 1) )
(4b)
(4c)
(6e)
The model described by Eqs. (6) is the one we propose for the
investigation of the dynamical behaviour of the structure
under consideration of the shunt type Colpitts oscillator.
Assuming
(5a)
VT
V
y= 1
VT
V
z= 2
VT
V
= 3
VT
dV
RC1 1 = VCC V1 V2 V3 +
dt
R F I E RI L
RC 2
iL
Chaotic behavior
(4d)
ln[d (t )]
t
t
max = lim
where
(7a)
77
d (t ) =
(x )2 + (y )2 + (z )2 + ( )2
(7b)
and computed from the variational equations obtained by
perturbing the solutions of Eqs. (6) as follows: x x + x ,
y y + y , z z + z , and + . d (t ) is
the distance between neighbouring trajectories [38].
Asymptotically, d (t ) = exp max .t . Thus, if max > 0 ,
max = 0
iL
inductor
for
iL
inductor
for
78
through
the
inductor:
iL
(C1 = 4.7nF )
(C1 = 22.0nF ) , c)- period-4
(C1 = 25nF ) , and d)- chaos (C1 = 47nF )
, b)- Period-2
79
C 2 = 100nF , C3 = 22F ,
L = 470F , and
R1
R3
=
R 2 R 4 + R5
The current I 0 pulled from the load is given by :
I0 =
where
R2
Vi
R1R 5
(8a)
(8b)
I0 =
Vi
1000
Thus,
(8c)
U2 is connected to -12V.
C1
C1
C1
C1
R10 . A 1
80
V.
CONCLUSION
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
81
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
82
Regular Paper
Original Contribution
83
I. INTRODUCTION
from 3.1 to 4.8 GHz and are centered at 3.432, 3.960, and
4.488 GHz, respectively.
Due to the characteristics of the UWB signals, i.e., very low
radiated power (-41.3 dBm/MHz) and large bandwidth (the
minimum bandwidth is 500 MHz), a multi-band OFDM UWB
receiver requires better receiver sensitivity and a lower noise
figure than, for example, an IEEE 802.11a receiver [7]. The
expected receiver sensitivity is around -70 dBm [8], and it can
be achieved by an optimal design of the low-noise amplifier
(LNA) in terms of wideband, near-to-minimum noise figure
and reasonable power gain. Furthermore, an optimal
integration of the entire RF front-end including the antenna
can also contribute to better receiver sensitivity by minimizing
losses. Another challenge of the UWB front-end design is
caused by the problem of narrowband interferers, e.g., out-ofband, in-band, or other unintentional radiation of electronic
devices [9]. In particular, services around 2.4 and 5 GHz
(IEEE 802.15.1, IEEE 802.11a) can hinder the UWB
communication and must be taken into consideration in UWB
front-end implementation. One solution is to filter
interferences at radio frequencies (RF) before or within the
first amplification stage, e.g., LNA [9]. Two different
techniques can be employed to achieve selective operation in
different band groups or, more restrictive, within each
frequency band group, i.e., in different sub-bands. In [10],
LNAs with selective gain-frequency characteristics employ
multi-resonance load networks which shape the LNA transfer
function. These techniques require area consuming and
complex LC (inductor and capacitor) load networks.
Sometimes, they also need load center frequency control
mechanisms by means of noisy switching pulses.
An alternative approach with a multi-band LNA covering
the multi-band UWB is presented in this paper. It is a
frequency-triplexed RF front-end using one 3.1-4.8 GHz LNA
for Band Group 1 UWB systems. The proposed solution
combines a multi-band pre-selecting filter function with the
frequency multiplexing function to connect the three different
RF inputs to only one LNA. The LNA is optimized for a nearto-minimum noise figure and flat gain response. The RF frontend is completely integrated into a four-metal layer printed
circuit board and is dedicated to a complete integration of
UWB antenna-LNA system on the same RF module.
84
F
M
N
UWB LNA
Matching
network
Metal 1-2
Broadside coupling
Edge coupling
Matching
network
LNA
Metal 1
Metal 2
Ground
Metal 4
BPF
sub-band #1
Port 2
/4 @
3.432 GHz
BPF
sub-band #2
Port 3
sub-band #3
/4 @
3.960 GHz
BPF
/4 @
3.488 GHz
RF In
RF Out
S2P
Rstab
Microstrip stubs
Port 4
bias network
Port 3
Port 2
Port 1
Metal 1: Triplexer
Metal 2: Triplexer
Metal 3: Ground
Metal 4: LNA
RO4350B
RO4450B
RO4350B
Fig. 6. Printed circuit board structure.
Dimension
0.254 mm
3.480.05
0.004
Dimension
0.200 mm
3.540.05
0.004
Dimension
0.035 mm
0.025 mm
7
5.8x10 S/m (Copper)
0.001 mm
85
Sub-band #1
Sub-band #2
Sub-band #3
0
-10
-20
-30
-40
-50
-60
-70
-80
2
4
Frequency (GHz)
S21 (dB)
S32 (dB)
S31 (dB)
-20
-30
-40
-50
-60
-70
-80
2
Frequency (GHz)
15
Front-end NF
LNA NF
LNA NF-min
10
20
10
0
-10
-20
-30
-40
-50
-60
-70
Sub-band #1
Sub-band #2
Sub-band #3
Sub-band #1
Sub-band #2
Sub-band #3
Fig. 9. Triplex LNA system, forward transmission: (a) simulation, and (b)
measurement.
15
10
0
3
Frequency (GHz)
Frequency (GHz)
Frequency (GHz)
0
-10
86
Frequency (GHz)
Fig. 10a and 10b show the front-end noise figure simulation
and measurement results, respectively. The LNA noise figure
and the simulated minimum noise figure are also shown. It is
seen that the measured noise figure for the entire system
within each 500 MHz sub-band is kept below 6 dB at the subband center frequency. However, the measured noise figure of
the front-end is larger than the simulated noise figure. The
larger values of the noise figure compared to the simulated
values can be explained by (a) larger insertion loss in the
triplexer, (b) lower amplifier gain.
87
20
Front-end NF
LNA NF
LNA NF-min
15
10
5
0
2
IV. DISCUSSION
Frequency (GHz)
20
Front-end NF
LNA NF
15
10
5
0
2
Frequency (GHz)
Fig. 10. Triplex LNA system, noise figure: (a) simulation, and (b) measured.
V. CONCLUSION
0
S21 (dB)
S32 (dB)
S31 (dB)
-10
-20
-30
-40
-50
-60
-70
-80
2
Frequency (GHz)
S21 (dB)
S32 (dB)
S31 (dB)
-10
-20
In this paper, a new multi-band UWB 3.1-4.8 GHz frontend was presented. It consists of a fully integrated preselective filter and triplexer network and a wideband and flat
gain low-noise amplifier. The UWB front-end is dedicated to
a complete integration of the UWB antenna-LNA system on
the same RF module. Using a microstrip network and three
combined broadside- and edge-coupled bandpass filters, a
multi-band transfer function and RF frequency multiplexing
are achieved simultaneously. The low-noise amplifier has
been designed for a near-to-minimum noise figure and a flat
power gain over the 3.1-4.8 GHz frequency band. The
measured LNA noise figure is below 4 dB while the measured
overall front-end noise figure is below 6 dB. The attenuation
of potential narrowband interference and good isolation
between the three sub-channels are achieved.
REFERENCES
-30
[1]
-40
-50
[2]
-60
[3]
-70
-80
2
Frequency (GHz)
[4]
88
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
December
2005,
www.ecmainternational.org/publications/files/ECMA-ST/ECMA-368.pdf.
T. W. Fischer, B. Kelleci, K. Shi, A. I. Karilayan, E. Serpedin, "An
analog approach to suppressing in-band narrow-band interference in
UWB receivers," IEEE Transactions on Circuits and Systems, Vol. 54,
No. 5, pp. 941-950, May. 2007.
G. Cusmai, M. Brandolini, P. Rossi, F. Svelto, A 0.18-m CMOS
Selective Receiver Front-End for UWB Applications, IEEE Journal of
Solid-State Circuits, vol. 41, Issue 8, Aug. 2006, pp. 1764-1771.
M. Karlsson, and S. Gong, "A frequency-triplexed inverted-F antenna
system for ultra-wide multi-band Systems 3.1-4.8 GHz," to be published
in ISAST Transactions on Electronics and Signal Processing, 2007.
M. Karlsson, P. Hakansson, S. Gong, A frequency triplexer for ultrawideband systems utilizing combined broadside- and edge-coupled
filters, submitted for publication in IEEE Transactions on Advanced
Packaging, 2007.
A. Serban, M. Karlsson and S. Gong, Microstrip bias networks for
ultra-wideband systems, submitted for publication in ISAST
Transactions on Electronics and Signal Processing, 2007.
G. Gonzalez, Microwave Transistor Amplifier Design. Analysis and
Design, Prentice Hall 1997, pp. 344-345.
Regular Paper
Original Contribution
89
I. I NTRODUCTION
90
Fig. 1. Financial time series for the FTSE value (close-of-day) from 02-041984 to 12-12-2007 (top), the log derivative of the same time series (centre)
and a Gaussian distributed random signal (bottom).
91
Fig. 2. Evolution of the 1987, 1997 and 2007 nancial crashes. Normalised
plots (i.e. where the data has been rescaled to values between 0 and 1
inclusively) of the daily FTSE value (close-of-day) for 02-04-1984 to 24-121987 (top), 05-04-1994 to 24-12-1997 (centre) and 02-04-2004 to 24-09-2007
(bottom)
92
Complex systems can be split into two categories: equilibrium and non-equilibrium. Equilibrium complex systems,
undergoing a phase transition, can lead to critical states that
often exhibit random fractal structures in which the statistics of
the eld are scale invariant. For example, when ferromagnets
are heated, as the temperature rises, the spins of the electrons
which contribute to the magnetic eld gain energy and begin
to change in direction. At some critical temperature, the spins
form a random vector eld with a zero mean and a phase
transition occurs in which the magnetic eld averages to zero.
But the eld is not just random, it is a self-afne random eld
whose statistical distribution is the same at different scales,
irrespective of the characteristics of the distribution. Nonequilibrium complex systems or driven systems give rise to
self organised critical states, an example is the growing of
sand piles. If sand is continuously supplied from above, the
sand starts to pile up. Eventually, little avalanches will occur
as the sand pile inevitably spreads outwards under the force of
gravity. The temporal and spatial statistics of these avalanches
are scale invariant.
Financial markets can be considered to be non-equilibrium
systems because they are constantly driven by transactions that
occur as the result of new fundamental information about rms
and businesses. They are complex systems because the market
also responds to itself, often in a highly non-linear fashion, and
would carry on doing so (at least for some time) in the absence
of new information. The price change eld is highly nonlinear and very sensitive to exogenous shocks and it is probable
that all shocks have a long term effect. Market transactions
generally occur globally at the rate of hundreds of thousands
per second. It is the frequency and nature of these transactions
that dictate stock market indices, just as it is the frequency and
nature of the sand particles that dictates the statistics of the
avalanches in a sand pile. These are all examples of random
scaling fractals [20]-[28].
III. D OES A M ACROECONOMY HAVE MEMORY ?
When faced with a complex process of unknown origin, it
is usual to select an independent process such as Brownian
motion as a working hypothesis where the statistics and probabilities can be estimated with great accuracy. However, using
traditional statistics to model the markets assumes that they are
games of chance. For this reason, investment in securities is
often equated with gambling. In most games of chance, many
degrees of freedom are employed to ensure that outcomes are
random. In the case of a simple dice, a coin or roulette wheel,
for example, no matter how hard you may try, it is physically
impossible to master your roll or throw such that you can
control outcomes. There are too many non-repeatable elements
(speeds, angles and so on) and non-linearly compounding
errors involved. Although these systems have a limited number
of degrees of freedom, each outcome is independent of the
previous one. However, there are some games of chance that
involve memory. In Blackjack, for example, two cards are dealt
to each player and the object is to get as close as possible to
21 by twisting (taking another card) or sticking. In a bust
(over 21), the player loses; the winner is the player that stays
A(t) = u(t)
u( )u( t)d.
u(t) =
u(t) | U () |2
U () = F[u(t)] =
u(t) exp(it)dt
93
Fig. 3. Log-power spectra and ACFs of log price changes and absolute
log price changes for FTSE 100 index (daily close) from 02-04-1984 to 2409-2007. Top-left: log price changes; top-right: absolute value of log price
changes; middle: log power spectra; bottom: ACFs.
94
n steps the total length of the path the walker has taken will
be just an. We dene this value as the resultant amplitude A
- the total length of the walk - which will change only by
account of the number of steps taken. Thus,
A = an.
If each step takes a set period of time t to complete, then it
is clear that
A(t) = at.
This scenario is limited by the fact that we are assuming that
each step is of precisely the same length and takes precisely the
same period of time to accomplish. In general, we consider
a to be the mean value of all the step lengths and t to be
the cumulative time associated with the average time taken to
perform all steps. A walk of this type has a coherence from
one step or cluster of steps to the next, is entirely predictable
and correlated in time.
If the same walk takes place in the complex plane then the
phase from one step to the next is the same. Thus, the result
is given by
A exp(i) =
a exp(i) = na exp(i).
n
a exp(in)
n
n/2
sin(/2)
2
n
and we obtain the result
sin
.
95
1
2 ,
0 2;
otherwise.
0,
A(t) = a t.
With a coherent walk we can state that the resulting amplitude
after a time t will be at. This is a deterministic result.
However, with an incoherent random walk, the interpretation
xj =
cos(i )
i=1
j
yj =
exp(im ).
=a
m=1
exp(im )
m=1
exp(im )
m=1
exp(im )
m=1
= a2 n +
exp(ij )
j=1,j=k
(2)
with a = 77 and P = 231 1 and an arbitrary value of x0 the seed. For the coherent phase walk
i =
= a2
sin(i )
i=1
2 xi
16 x
exp(ik ) .
k=1
A = a n.
96
x2
t
u(x, t) = 0
for a diffusivity D = 1 . The eld u(x, t) represents a measurable quantity whose space-time dependence is determined
by the random walk of a large ensemble of particles or a
multiple scattered waveeld or information owing through a
complex network. We consider an initial value for this eld
denoted by u0 u(x, 0) = u(x, t) at t = 0. For example, u
could be the temperature of a material that starts radiating
heat at time t = 0 from a point in space x due to a mass
of thermally energised particles, each of which undertakes
a random walk from the source of heat in which the most
likely position of any particle after a time t is proportional to
97
u(x, t) +
2 2
u(x, t)
u(x, t) +
x
2! x2
where the higher order terms are neglected under the assumption that if is small, then the distance travelled, , must also
be small. We can then write
u(x, t) +
u(x, t) = u(x, t)
t
P ()d
+ u(x, t)
x
1 2
P ()d +
u(x, t)
2 x2
2 P ()d.
P ()d = 1.
1 2
u(x, t) + u(x, t) = u(x, t) +
u(x, t)
t
2 x2
so that
2
u(x, t) =
u(x, t)
t
x2
2
P ()d.
2
D=
2
P ()d
2
2
u(x, t) = D 2 u(x, t).
t
x
u(x, t) = 0
u(x, t + )
u(x, t + ) =
u(x, 0) = w(x)
t
the (dAlembert) general solution is given by [36]
u(x, 0) = v(x),
x+ct
1
1
u(x, t) = [v(x ct) + v(x + ct)] +
2
2c
w()d.
xct
2 P ()d
98
xj =
cos(i ) cos(i )
i=1
j
yj =
sin(i ) cos(i )
i=1
j
zj =
sin(i )
i=1
Fig. 5. Random phase walks in the plane for a uniform distribution of angles
i [0, 2] (top left), i [0, 1.9] (top right), i [0, 1.8] (bottom left)
and i [0, 1.2] (bottom right).
VIII. L E VY P ROCESSES
99
p(x) =
1
2
1
p(x) =
2
1
a
exp(a | k |) exp(ikx)dk =
2 + x2
a
1
f (x) =
1 ikx kq
1
=
e e
ix
ix
k=0
dkH(k)k q1 ek eikx , x
where
1,
0,
H(k) =
k>0
k<0
1
(ik)q
1
x1q
f (x)
and thus
p(x)
q i1q
ix xq
1
x1+q
, x
f (x) =
1
2
1
2
exp[x2 /(4a)],
a
whose rst and second moments are nite. The Cauchy distribution has a relatively long tail compared with the Gaussian
distribution and a stochastic eld described by a Cauchy
distribution is likely to have more extreme variations when
compared to a Gaussian distributed eld. For values of q
5 P L vy was the research supervisor of B Mandelbrot, the inventor of
e
fractal geometry.
q
ix
dkk q1 ek eikx
0
exp(ak 2 ) exp(ikx)dk
q
=
2ix
p(x) =
eikx ek dk
q
1 q1 kq ikx
k e
e
ix ix
k=0
q
x2
q
x2
100
q(q 1)
2x2
H(k)eikx (k q2 ek )dk.
q(q 1) i2q
x2 xq1
so that
1
, x
x1+q
which maps onto the previous asymptotic as q 1 from the
above.
For q 2, the second moment of the L vy distribution
e
exists and the sums of large numbers of independent trials are
Gaussian distributed. For example, if the result were a random
walk with a step length distribution governed by p(x), q > 2,
then the result would be normal (Gaussian) diffusion, i.e. a
Brownian process. For q < 2 the second moment of this PDF
(the mean square), diverges and the characteristic scale of the
walk is lost. This type of random walk is called a Le y ight.
v
p(x)
2
x
t
to the fractional form
2
q
q q
2
x
t
where q (0, 2] and D = 1/ is the fractional diffusivity.
Fractional diffusive processes can therefore be interpreted
as intermediate between classical diffusive (random phase
walks with H = 0.5; diffusive processes with q = 1) and
propagative process (coherent phase walks for H = 1;
propagative processes with q = 2), e.g. [39], [40] and [38]
- references therein. Fractional diffusion equations can also
be used to model L vy distributions [41] and fractal time
e
random walks [42], [43]. However, it should be noted that the
fractional diffusion operator given above is the result of a phenomenology. It is no more (and no less) than a generalisation
of a well known differential operator to fractional form which
follows from a physical analysis of a fully incoherent random
process and it generalisation to fractional form in terms of the
Hurst exponent. Note that the diffusion and wave equations can
be derived rigorously from a range of fundamental physical
laws (conservation of mass, the continuity equation, Fouriers
law of thermal conduction, Newtons laws of motion and so
Fig. 7.
101
q(t)
1
q( )
u(x, t) =
q( )
2
t
q(t)
q
1
u(x, t) =
tq
2
Let
u(x, t) = f (x)n(t).
1
u(x, t) =
2
U (x, ) exp(it)d
and
1
n(t) =
2
N () exp(it)d.
102
When q = 1 and 1 = i i,
q = i(i) 2
1
u(x0 , t) =
2
dxf (x)...
2
+ 2 g(x | x0 , ) = (x x0 )
q
x2
1
2
g(x | x0 , )f (x)dx
U (x0 , ) = N ()
(3)
where [36]
g(x | x0 , k) =
i
exp(iq | x x0 |)
2q
exp( i | x x0 |)
N () exp(it)d.
i
exp(a p)
1
exp(pt)dp = exp[a2 /(4t)],
p
t
ci
we obtain
u(x0 , t) =
dxf (x)
N ()
U (x0 , ) =
2i
1
u(x0 , t) =
2
N ()
exp(i | x x0 |) exp(it)d
i
1
2
dxf (x)
2 t
dxf (x)...
1
2
n(t t0 )dt0 .
t0
n(t | x x0 |)dt
which describes the propagation of a wave travelling at velocity 1/ subject to variations in space and time as dened by
f (x) and n(t) respectively. For example, when f and n are
both delta functions,
1
u(x0 , t) =
H(t | x x0 |).
2
This is a dAlembertian type solution to the wave equation
where the wavefront occurs at t = | x x0 | in the causal
case.
U (x0 , ) =
iM0 N ()
(iq )m Mm (x0 )
1+
2q
m!
M0
m=1
(4)
where
f (x) | x x0 |m dx.
Mm (x0 ) =
103
M (x0 )
M (x0 )
N () or u(x0 , t) =
n(t)
2
2
where
1
2
M (x0 ) =
M0 q 1
2
2
2
iN ()M0
exp(it)d =
2q
M0 1
q
2 2 2
iM2 1
2.2! 2
N ()
q exp(it)d
(i) 2
1
M0
=
q
q
2 (2i)
2
1q
2
q
2
n()
d.
(t )1(q/2)
u(x0 , t) =
M1
n(t).
N () exp(it)d =
2
1
+
2
n()
d.
(t )1(q/2)
k=1
1kq
2
kq
2
...
M2 2 d 2
N ()i(i) exp(it)d =
q n(t)
2.2! dt 2
n()
d
(t )1(kq/2)
M3 q d q
n(t)
2.3! dtq
and
M4 1
i
2.4! 2
1q
2
q
2
q
2
3q
N ()i (i)
3q
2
3q
M4 2 d 2
exp(it)d =
n(t)
2.4! dt 3q
2
M0 (x0 )
1
q/2 (2i)q
2
M2 1
1
=
q/2 (2i)q
2.2!
M1 1
2 2
M3 1
2.3! 2
N ()i
q exp(it)d
(i) 2
M0 q d 2
2 q n(t).
2
dt 2
The second term is the same is in the previous case (for q > 0)
and the third term is
iM2 1
2.2! 2
N ()(i) 2 exp(it)d =
1
2
iN ()M0
exp(it)d
2q
M1 (x0 )
1
n(t) +
2
2
k=1
1q
2
q
2
n()
d
(t )1(q/2)
k+1
(1)
dkq/2
Mk+1 (x0 ) kq/2 kq/2 n(t).
(k + 1)!
dt
I p f (t) =
1
(p)
f ()
d, p > 0
(t )1p
(5)
104
q-value
t-space
-space (PSDF)
Name
q=0
1
t
White noise
q=1
1
||
Pink noise
1
2
Brown noise
1
||q
Black noise
t
R
q=2
TABLE I
N OISE CHARACTERISTICS FOR DIFFERENT VALUES OF q. N OTE THAT THE
RESULTS GIVEN ABOVE IGNORE SCALING FACTORS .
(6)
x0 0
( 1q )
1
2
1
q/2
2 (2i)q ( q )
n()
d,
(t)1(q/2)
where
sign() =
> 0;
< 0.
q > 0;
q < 0.
N ()
.
(i)q/2
Pr
1
t1q/2
n(t) = Pr[n(t)], q 0.
M0 (x0 )
1
=
2 q/2 (2i)q
7 Note
1,
1,
q = 0;
n(t)
2 ,
q/2 q/2
d
2 dtq/2 n(t),
U () =
n(t)dt
1
n(t) isign()N ()
t
n(t)
t(q/2)1 n(t)
q>2
(x) | x x0 |m dx =| x0 |m .
Mm (x0 ) =
n(t)
1q
2
q
2
n()
d
(t )1(q/2)
M1 (x0 )
n(t), q > 0.
(7)
2
Here, the solution is the sum of fractal noise and white noise.
Further, by relaxing the condition 0 we can consider the
approximation
u(x0 , t)
M0 (x0 )
1
q/2 (2i)q
2
1q
2
q
2
n()
d
(t )1(q/2)
105
M1 (x0 )
M2 (x0 ) q/2 dq/2
n(t), q > 0, << 1
n(t) +
2
2.2!
dtq/2
(8)
in which the solution is expressed in terms of the sum of
fractal noise, white noise and the fractional differentiation8 of
white noise.
f ( )wL (, t)d
where
t
1
, L > 0.
wL (, t) = w
L
L
The independent variables L and t are continuous dilation and
translation parameters respectively. The wavelet transformation is essentially a convolution transform where wL (t) is the
convolution kernel with dilation variable L. The introduction
of this factor provides dilation and translation properties into
the convolution integral that gives it the ability to analyse
signals in a multi-resolution role (the convolution integral is
now a function of L), i.e.
FL (t) = wL (t) f (t), L > 0.
In this sense, the asymptotic solution (ignoring scaling)
1
u(t) = 1q/2 n(t), q > 0 x 0
t
is compatible with the case of a wavelet transform where
1
w1 (t) = 1q/2
t
for the stationary case and where, for the non-stationary case,
1
w1 (t, ) = 1q( )/2 .
t
XII. FTSE A NALYSIS USING OLR
N ()
(i)q/2
ln P () = C + q ln
where C = ln c. The problem is therefore reduced to implementing an appropriate method to compute q (and C) by
e(q, C) = ln P () ln P (, q, C)
2
2
u(x, t) = (x)n(t), x 0
P () = q
8 As
or
106
XIII. D ISCUSSION
This paper is concerned with the introduction and theoretical
analysis (in terms of general a solution) associated with the
non-stationary fractional diffusion operator
q(t)
2
q(t) q(t)
x2
t
in the context of a macroeconomic model. By considering a
source function of the type (x)n(t) where n(t) is white noise,
we have shown that, for x 0, the fractional diffusive eld
u(t) at time is given by (ignoring scaling)
u(t, ) =
1
t1q( )/2
n(t)
107
Fractal type
Fractal Dust
Fractal Curve
Fractal Surface
Fractal Volume
Fractal Time
Hyper-fractals
.
.
.
Fractal Dimension
0 < DF < 1
1 < DF < 2
2 < DF < 3
3 < DF < 4
4 < DF < 5
5 < DF < 6
.
.
.
TABLE II
F RACTAL TYPES AND CORRESPONDING
FRACTAL DIMENSIONS
log(N )
.
log(r)
108
L( )
= L( ) 1 ,
ln L( ) ln N [L( ), ]
=
1=
ln
ln N [L( ), ] ln L( )
ln
ln N ()
ln
ln N ()
ln
where N () is taken to be N [F (f ), ] for notational convenience. Thus a piecewise continuous eld has precise fractal
1
, 0,
DF
then, by inspection,
DF = 2 H.
Thus, a Brownian process, where H = 1/2, has a fractal
dimension of 1.5. For higher topological dimensions DT
DF = DT + 1 H.
This algebraic equation provides the relationship between the
fractal dimension DF , the topological dimension DT and the
Hurst dimension H. We can now determine the relationship
between the Fourier dimension q and the fractal dimension
DF .
Consider a fractal signal f (x) over an innite support with
a nite sample fX (x), given by
fX (x) =
109
Thus, since
1
fX (x) =
2
DF = DT + 1 H,
FX (k) exp(ikx)dk,
1
2
PX (k) =
|FX (k)|
X
and
and
gX (x) =
1
f (ax),
aH
0 < x < X;
otherwise.
GX (k) =
gX (x) exp(ikx)dx =
DT
3DT + 2
=
,
2
2
the Fourier dimension being given by q = /2.
DF = DT + 1 H = DT + 1
ACKNOWLEDGMENT
Some of the material presented in this paper is based on
the PhD Theses of two former research students of the author,
Dr Mark London and Dr Irena Lvova. The author is grateful
to Mr Bruce Murray (Lexicon Data Limited) for his help
in the assessment of the nancial data analysis carried out
by the author and to Ms Mariam Fardoost (Merrill Lynch,
London) for her advice with regard to global economics, macro
overviews and market forecasts.
0
X
1
aH+1
f (s) exp
0
R EFERENCES
iks
ds
a
[1]
[2]
[3]
[4]
[5]
1
FX
aH+1
k
a
[6]
1
1
FX
a2H+1 aX
k
a
[7]
[8]
[9]
and, as X ,
[10]
P (k) =
P
a2H+1
k
a
[11]
[12]
1
P
a2H+1
k
a
[13]
[14]
[15]
.
[16]
k 2H+1
1
5
=
.
2
2
The fractal dimension of a fractal signal can be calculated
directly from using the above relationship. This method also
generalizes to higher topological dimensions giving
DF = 2
= 2H + DT .
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
http://www.tickdata.com/
http://www.vhayu.com/
http://en.wikipedia.org/wiki/Louis Bachelier
http://en.wikipedia.org/wiki/Robert Brown (botanist)
T. R. Copeland, J. F. Weston and K. Shastri, Financial Theory and
Corporate Policy, 4th Edition, Pearson Addison Wesley, 2003.
J. D. Martin, S. H. Cox, R. F. McMinn and R. D. Maminn, The Theory of
Finance: Evidence and Applications, International Thomson Publishing,
1997.
R. C. Menton, Continuous-Time Finance, Blackwell Publishers, 1992.
T. J. Watsham and K. Parramore, Quantitative Methods in Finance,
Thomson Business Press, 1996.
E. Fama, The Behavior of Stock Market Prices, Journal of Business Vol.
38, 34-105, 1965.
P. Samuelson, Proof That Properly Anticipated Prices Fluctuate Randomly, Industrial Management Review Vol. 6, 41-49, 1965.
E. Fama, Efcient Capital Markets: A Review of Theory and Empirical
Work, Journal of Finance Vol. 25, 383-417, 1970.
G. M. Burton, Efcient Market Hypothesis, The New Palgrave: A
Dictionary of Economics, Vol. 2, 120-23, 1987.
F. Black and M. Scholes, The Pricing of Options and Corporate
Liabilities, Journal of Political Economy, Vol. 81(3), 637-659, 1973.
http://uk.nance.yahoo.com/q/hp?s=%5EFTSE
B. B. Mandelbrot and J. R. Wallis, Robustness of the Rescaled Range
R/S in the Measurement of Noncyclic Long Run Statistical Dependence,
Water Resources Research, Vol. 5(5), 967-988, 1969.
B. B. Mandelbrot, Statistical Methodology for Non-periodic Cycles:
From the Covariance to R/S Analysis, Annals of Economic and Social
Measurement, Vol. 1(3), 259-290, 1972.
E. H. Hurst, A Short Account of the Nile Basin, Cairo, Government
Press, 1944.
http://en.wikipedia.org/wiki/Elliott wave principle
http://www.olivercromwell.org/jews.htm
B. B. Mandelbrot, The Fractal Geometry of Nature, Freeman, 1983.
J. Feder, Fractals, Plenum Press, 1988.
K. J. Falconer, Fractal Geometry, Wiley, 1990.
H. O. Peitgen, H. J rgens and D. Saupe D, Chaos and Fractals: New
u
Frontiers of Science, Springer, 1992.
P. Bak, How Nature Works, Oxford University Press, 1997.
M. J. Turner, J. M. Blackledge and P. Andrews, Fractal Geometry in
Digital Imaging, Academic Press, 1997.
N. Lam and L. De Cola L, Fractal in Geography, Prentice-Hall, 1993.
J. M. Blackledge, Digital Image Processing, Horwood, 2006.
110
Regular Paper
Original Contribution
111
I. INTRODUCTION
I/Q demodulator
w3
Judgement circuit
RF
LP
Diode Filter
w4
Correlator
RFin
SixPort
LNA
AG
w5
w6
I-data
Q-data
LO
I-data
w4
w5
w6
+
Q-data
+
2a
w3
w4
w5
w6
ADC
I-data
ADC
ADC
DSP
Q-data
ADC
2b
Fig. 2a and 2b. Judgment circuits (a) analog and (b) digital for six-port
recivers.
0.7 slo
slo
Port 3
Port 4
srf
50
Port 5
Port 6
Vin-
+
-
Vin+
330
330
+
560 270
Power divider
90 coupler
0.5 slo + j0.5 srf
0.7srf
0.7 slo
j0.7 srf
75
112
+
-
270
560
Vout
113
P3
LO
Wide band 90
Branch couplers
LP filter
Diodes
P2
P4
RF
input
21 mm
0.004
Metal thickness
25 m
Metal conductivity
58 MS/m
Surface roughness
P2
3.48
Loss factor
12 mm
1 m
3.3 mm
P3
P6
90 branch coupler
P1
Matching networks
P7
Ref
clock
Vector
Signal
Generator
LO
Instrumentation
amplifier
I
I/Q
demodulator
Oscilloscope
Vector
Signal
Analyzer
Power
divider
P4
P5
C. Test set-up
The test set-up for the proposed six-port I/Q demodulator is
shown in Fig. 6. The vector signal generator used in the test
set-up is an SMIQ 06 B from Rode Schwartz. To generate the
LO signal a continuous sinusoidal wave from a ZVM from
Agilent Technologies Inc. is utilized. To analyze the
demodulated signals an wideband oscilloscope and a spectrum
analyzer from Agilent Technologies are used. The vector
signal analyzer software package VSA 89600 from Agilent
Technologies Inc. is used to further analyze the I/Q vector
signals.
-5
S21
S31
S41
S51
-10
-15
-20
-25
3.5
4.5
Frequency (GHz)
114
-5
S26
S36
S46
S56
-10
-15
-20
-25
-1
-2
-3
3.5
4.5
3.5
4.5
Frequency (GHz)
Frequency (GHz)
Fig. 9. Measured S-parameters from the RF-input port (P6) to the output
ports (P2-P5).
0
S61
-5
-10
S21-S31
S41-S51
S26-S36
S46-S56
-15
-20
-25
-30
-35
-40
3.5
4.5
Frequency (GHz)
Fig. 10. Measured leakage from the RF input port to the LO input port.
100
98
96
phase(S21)-phase(S31)
phase(S41)-phase(S51)
phase(S26)-phase(S36)
phase(S46)-phase(S56)
94
92
90
88
86
84
82
80
3.5
4.5
Frequency (GHz)
115
0,2
(a)
0,1
Q-output (V)
0.5
Eye
opening
-0.5
-1
-0.8
-0.6
-0.4
-0.2
0.2
0.4
0.6
0.8
Symbol
0,0
(b)
-0,2
-0,2
-0,1
0,0
0,1
0,2
I-output (V)
Q
Fig. 13. Measured I and Q constellation diagram of a 64-QAM signal at
3.96 GHz.
0.1
0
-0.05
10
12
14
16
18
20
12
14
16
18
20
Q-output (V)
Time (us)
0.1
0.05
0
-0.05
-0.1
0
10
Time (us)
I-output (V)
0.005
Eye
opening
0
-0.005
-0.01
-1
-0.8
-0.6
-0.4
-0.2
0.2
0.4
0.6
0.8
Symbol
Fig. 16. Measured eye-diagram of the Q-output with QPSK modulated
signal at 3.96 GHz, with a 0 dBm LO signal and a RF of (a) 15 dBm and
(b) -30 dBm.
IV. DISCUSSION
0.05
-0.1
0.1
0.05
0
-0.05
-0.1
0
10
12
14
16
18
20
12
14
16
18
20
Time (us)
Q
-output (V)
Q-output (V)
0.01
-0,1
0.1
0.05
0
-0.05
-0.1
0
10
Time (us)
3.1
1.7
5.6
-0.9
-0.3
-0.5
116
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14] E. Marsan, J.-C. Schiel, K. Wu Gailon Brehm and R. G. Bosisio, HighSpeed Carrier Recovery Circuit Suitable for Direct Digital QPSK
Transceivers Proc. of RAWCON 2002, pp. 103 106.
[15] F.R de Sausa and B. Huyart, Carrier Recovery in five-port receivers,
Proc. Euro Conf. Wireless Technology., 2003, pp. 419-421.
Pr Hkansson was born in Karlshamn, Sweden
in 1979. He received his M.Sc. degree from
Linkping University in Sweden in 2003.
From 2004 to 2005 he worked as a research
engineer in the research group of Communication
Electronics at Linkping University, Sweden. In
2005 he started his Ph.D. study in the research
group. His main work involves both wireless and
wired high-speed data communications.