Professional Documents
Culture Documents
Abstract - This report examines optical CDMA communi- asynchronously, without centralized control, and it does
cation techniques with optical orthogonal codes. Simulations not suffer from packet collisions. As a result, optical
that show the desired properties of theses codes and their use CDMA systems have lower latencies than TDMA or
in optical CDMA are reported. Based on the simulations, we WDMA. Furthermore, since time and frequency (or
investigate the properties of optical CDMA. Proba bility of
wavelength) slots do not need to be allocated to each
error is also evaluated.
individual user, significant performance gains can be
I. INTRODUCTION achieved through multiplexing. Also, TDMA and
WDMA systems are limited by hardware because of the
There have been many efforts to take the full advantage slot allocation requirements. In contrast, CDMA systems
of fiber-optic signal processing techniques to establish an are only limited the tolerated bit error rate relationship to
all optical CDMA communication systems since CDMA the number of users, affording the designer a much more
was first applied to the optical domain in the mid-1980s flexible network design [4].
by Prucnal, Salehi, and others [1-3]. Traditional fiber
optic communication systems use either TDMA or To establish the optical CDMA, we have to overcome
WDMA schemes to allocate bandwidth among multiple the code orthogonality problem. Many researchers have
users. Unfortunately, both present significant drawbacks proposed several codes such as prime code, optical
in local area systems requiring large numbers of users. orthogonal code, and so on. In this project, we focus on
optical orthogonal codes (OOC) among those codes. In
In a TDMA system, the total system throughput is section II, we introduce the optical orthogonal codes.
limited by the product of the number of users and their Section III discusses three simulations demonstrating the
respective transmission rates since only one user can principles of optical CDMA. The first one is for two-user
transmit at a time. For instance, if 100 users wish to synchronous channels to understand basics of optical
transmit at 1 gigabit per second, at a minimum the CDMA and the second one is for two-user asynchronous
communication hardware would need to be capable of channels. The third one is for K-user synchronous
sustaining a throughput of 100 gigabits per second, a data channels. Section IV evaluates the probability of error.
rate that would strain even the highest performance
optical networking equipment. In addition, TDMA II. O PTICAL O RTHOGONAL CODES
systems show significant latency penalties because of the
coordination required to coordinate and grant requests for An optical orthogonal code is a family of (0, 1)
time slots from users by the central node [4]. sequences with good auto- and cross-correlation
properties. Thumbtack-shaped auto-correlation enables
Unlike TDMA, a WDMA system allows each user to the effective detection of the desired signal (Fig. 2 c), and
transmit at the peak speed of the network hardware since low-profiled cross-correlation makes it easy to reduce
each channel is transmitted on a single wavelength of interference due to other users and channel noise (Fig. 2
light. A WDMA system could easily support a bandwidth d). The use of optical orthogonal codes enables a large
of one terabit per second, ideal for the needs of a local number of asynchronous users to transmit information
area network. Unfortunately, it is difficult to construct a efficiently and reliably. The lack of a network
WDMA system for a dynamic set of multiple users synchronization requirement enhances the flexibility of
because of the significant amount of coordination among the system. The codes considered here consist of truly (0,
the nodes required for successful operation. To build a 1) sequences (Fig. 1 a) and are intended for “unipolar”
WDMA network with a dynamic user base, control environments that have no negative components since you
channels and collision detection schemes would need to either have light, or you don't, while most document ed
be implemented that would waste significant bandwidth. correlation sequences are actually (+1, -1) sequences (Fig.
1 b) intended for systems having both positive and
Fortunately, an alternative to TDMA and WDMA negative components.
networking schemes, optical CDMA communication
systems , require neither the time nor the frequency An (n, w, ?a, ?c ) optical orthogonal code C is a family
management systems. Optical CDMA can operate of (0, 1) sequences of length n and weight w which satisfy
the following two properties [5]. other time following (1). Fig. 2 (d) shows the cross-
1) The Auto-Correlation Property: correlation between OOC 1 and OOC 2. It always takes
n−1 one and zero at any time following (2). Here 1s are taken
∑x xt=0
t t +τ
≤ λa (1) as auto-correlation constraint, ?a, and cross-correlation
constraints, ?c, since 1s are the lowest value they can be,
for any x ∈ C and any integer t, 0<t<n. and correlation is calculated by convolution sum of two
2) The Cross-Correlation Property: sequences.
n −1
∑x y ≤ λc
5 5
(2)
t t +τ 4.5 4.5
t=0 4 4
3 3
The numbers ?a and ?c are called the auto- and cross- 2.5 2.5
2
correlation constraints. The (0, 1) sequences of an optical 2
1.5 1.5
0.5 0.5
0 0
5 10 15 20 25 30 5 10 15 20 25 30
1.5 1.5
1 1
(a) (b)
5
5
0.5 0.5
4.5
4.5
4
0 0 4
3.5
3.5
-0.5 -0.5 3
3
2.5
2.5
-1 -1
2
2
1.5
-1.5 -1.5 1.5
0 5 10 15 20 25 30 0 2 4 6 8 10 12 14 16
1 1
0
0.5
10 20 30 40 50 60 0
An (n, w, ?a, ?c ) OOC C can be alternatively considered Fig. 2. (a) OOC 1 (b) OOC 2 (c) auto-correlation (d)
as a family of w sets of integers modulo n in which each cross-correlation
w set corresponds to a codeword and the integers within
each w set specify the nonzero bits. For instance, let’s II. O PTICAL CDMA
think of a simple OOC, 1101000, characterized by (7, 3, 1,
II.1.A. Two-User Synchronous Channel
1). Here we can see three 1s as the nonzero bits. Their
positions are 0th, 1st, and 3rd, respectively. Thus 1101000 The two-user synchronous channel simulated here can
can be simply represented by {0, 1, 3} (mod 7). This be characterized by the follows.
notation can simply represent codes instead of
exhaustively describing long (0, 1) sequences. Table I y (t ) = A1b1 s1 (t ) + A2 b2 s2 (t ) + σn( t) A1 = A2 = 1, t ∈ [0, T ] .
(3)
shows some optimal (n, 3, 1, 1) codes in this notation. T w, s1 = s2
ρ =< s1 , s 2 >= ∫ s1 (t ) s2 (t )dt =
n Optimal (n, 3, 1, 1)-codes
0
0 or 1, s1 ≠ s2
7 {0,1,3}
Let an (32, 4, 1, 1) optical code C with 2 codewords be
13 {0,1,4}, {0,2,7}
used. C can be represented by {{0, 10, 13, 28}, {0, 5, 12,
19 {0,1,5}, {0,2,8}, {0,3,10}
31}} mod(32) in Fig. 3 (d) and (e). Thus, the system can
25 {0,1,6}, {0,2,9}, {0,3,11}, {0,4,13}
31 {0,1,7}, {0,2,11}, {0,3,15}, {0,4,14}, {0,5,13} accommodate 2 transmitters simultaneously. Each
37 {0,1,11}, {0,2,9}, {0,3,17}, {0,4,12}, {0,5,18}, transmitter is assigned a w(=4) set from C, i.e.
{0,6,12} transmitter1 is assigned a {0, 10, 13, 28} set and
transmitter 2 is assigned a {0, 5, 12, 31} set. At a
43 {0,1,19}, {0,2,22}, {0,3,15}, {0,4,13}, {0,5,16},
transmitter, every information bit is encoded into a frame
{0,6,14},{0,7,17}
of n(=32) optical chips in the following way. (A chip is an
optical time slot which can assume one of two values: ON
Table I. Some Optimal (n, 3, 1, 1) Codes [5]
of OFF) Let the assigned w set for a particular transmitter
be S={s1, s 2, · · · , s w}. In this case, s 1 is a {0, 10, 13, 28}
Let’s think of C represented by {{0, 10, 13, 28}, {0, 5,
12, 31}} (mod 32) with two code words. Two code set and s 2 is a {0, 5, 12, 31} set. Assume the information
bit is 1. In the corresponding frame, which consists of n
words are shown in Fig. 2 (a) and (b), respectively. C also
optical chips, photon pulses (i.e., ON signals) are sent at
can be represented by (32, 4, 1, 1) in (n, w, ?a, ?c ) notation.
exactly the s 1th, s 2th, · · · and s wth chips (Fig. 3. f and g).
Fig. 2 (c) shows auto-correlation of OOC 1. Its maximum
In the other n-w chips, no photon pulse (i.e., OFF signals).
value is w at the correlation time and one or zero at any
A property of MVG_OMALLOOR
OOC 1 OOC 1
C
A Optical E Optical H J Recovered
+
User 1 • Encoder
• G G Decoder • + • Signal
Optical
Sum • • Optical + Recovered
User 2 •B Encoder • Decoder
• +
•K Signal
F I
I
D
OOC 2 OOC 2
OPTICAL OPTICAL
(a) (a)
5 5 5 5
2.5 2.5
2.5 2.5
2 2
2 2
1.5 1.5
1.5 1.5
1 1
1 1
0.5 0.5
0.5 0.5
0 0
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 0 0
50 100 150 200 250 300 350 50 100 150 200 250 300 350
4.5 4.5
4 4
3.5 3.5
1.5 1.5
0.5
1
0.5
Basically, optimal CDMA scheme is same as radio
0
50 100 150 200 250 300
0
50 100 150 200 250 300
frequency CDMA scheme except using the special codes.
Here, we use the matched filter to convert the received
(f) (g)
signal in Fig. 3 (h) assuming n(t) in (3) is zero. At the
5
3.5
separate the transmitted signals. The decoder consists of a
3 bank of 2 tapped delay -lines, one for each codeword. The
2.5
2
delay taps on a particular line exactly mat ch the signature
1.5
sequence, i.e., the delays between successive taps are
1
0.5
equal to s2-s 1, s 3-s 2, · · · , optical chips, respectively. Each
0
50 100 150 200 250 300 tapped delay-line effectively calculates the correlation of
(h) the received waveform with its signature sequences. In
Fig. 4 (b) and (c), there are five different correlation
Fig. 3. (a) schematic diagram for a transmitting part (b) values, 0, 1, 2, 4, and 5. Because of the properties of
signal at A (c) signal at B (d) signal at C (e) signal at D optical orthogonal codes, the correlation between
(f) signal at E (g) signal at F (h) signal at G different signature sequences is low, 0 and 1. Thus the
delay-line output is high, 4 and 5, only when the intended
On the other hand, if the information bit is 0, no photon transmitter’s information bit is 1. However, a potential
pulses are sent in the corresponding frame, i.e., all OFF problem due to interference can be happened. When
signals are sent. In other words, the codeword set is used correlation value has 2, it is definitely due to the
as the signature sequence of the transmitter. interference. In this case, the value is always below 4 so it
A property of MVG_OMALLOOR
can be discarded by choosing relevant threshold value. scheme for synchronization. Therefore, now we can say
But when total number of users goes up, the cross- that optical CDMA does need no network synchronization.
correlation due to interfering users adds up quickly to
severely degrade the system performance. For instance, II.2. K-User Synchronous Channel
when w is 4 like this case, accommodating 4 users make it
possible to have 4 as the correlation value even if the In the previous section, we investigated a simp le 2-user
intended transmitter’s information bit is 0. To avoid this channel to understand the optical CDMA. In this section,
phenomenon, both high w which can be considered as the we explore problems faced by increasing K (Fig. 6 a).
sum of 1s in the sequence, and long n are required. If we Here we choose 7 as K, and C is (43, 3, 1, 1) having 7 sets,
increase only w fixing n, cross-correlation value due to {{0, 1, 19}, {0, 2, 22}, {0, 3, 15}, {0, 4, 13}, {0, 5, 16},
the interference can be lowered. However, OOC has very {0, 6, 14}, {0, 7, 17}}.
sparse marks to keep the cross-correlation low, i.e. a
number of zeros is much higher than that of ones in the The first issue is a threshold value. As we saw in the
sequence. It means that cross-correlation increases by previous section, interfering signal can be effectively
itself by increasing only w. Therefore, both w and n discarded by setting a relevant threshold value. The
should be increased simultaneously. But this solution also threshold value can be chosen under the following
reveals a drawback, long signal processing time due to condition.
long n. 0 ≤ threshold ≤ w (4)
Finally the transmitted information is extracted by Fig. 6 (d) through (g) show results from several threshold
thresholding the correlator output in Fig. 4 (d) and (e). In values, 1, 2, 3, and 5. As the threshold value goes up, the
this case, they are successfully recovered as intended. recovered signal gets similar to the ideal signal in Fig. 6
Here, 3 is chosen as the threshold values. Threshold issue (c). However, the threshold value can not be over w. If the
will be covered in detail in the section II.2. intended information bit is 1, the correlation value is w.
Therefore, the threshold over w incorrectly converts it to
II.1.B. Two-User Asynchronous Channel 0 when the intended information bit is 1. Fig. 6 (g) shows
the results by choosing threshold over w. We can clearly
In optical CDMA, all users are allowed to transmit at see that a dotted line is missed comparing to Fig. 6 (c).
any time. There is no network synchronization required.
In this section, we simulate a two-user asynchronous Even if the highest value under (4) is chosen (Fig. 6 f),
channel to investigate above statement. To verify no the recovered signal can not be exactly same as the ideal
network synchronization, the same scheme in II.I.A. is pattern. In the Fig. 6 (a), if the intended information bit is
used here except the time delay at F in Fig. 3 (a). Fig. 5 1, the correlation value is w(=3). But as we already
(b) and (c) show the recovered signals from asynchronous discussed in the previous section, the cross-correlation
and synchronous channel, respectively. They agree well due to interfering users adds up very quickly to severely
even if synchronous channel does not use any special degrade the system performance. Here, the correlation
value due to interference is even higher than w(=3). This
1.5
is the reason why the received signal can not be recovered
perfectly.
1
0, 0 ≤ x < 1
(a)
1.5
Therefore, if an optical light intensity (x) is bigger than or
1.5
OOC 1 OOC
G . (a)
• .
OOC K .
OOC
Optical + Recovered
Decoder • + • Signal Hard G’ Optical H’ + J’ Recovered
Limiter • Decoder • + ? Signal
(a)
8 1.5 (b)
7
8 8
6
7 7
1
5
6 6
4
5 5
3
4 4
0.5
2
3 3
1
2 2
0 0
0 50 100 150 200 250 300 350 50 100 150 200 250 300 1 1
0 0
(b) (c) 50 100 150 200 250 300 50 100 150 200 250 300
1.5 1.5
(c) (d)
8 8
7 7
1 1
6 6
5 5
0.5 0.5
4 4
3 3
2 2
0 0
50 100 150 200 250 300 50 100 150 200 250 300
1 1
0 0
(d) (e) 0 50 100 150 200 250 300 350 50 100 150 200 250 300
1.5 1.5
(e) (f)
1.5 1.5
1 1
1 1
0.5 0.5
0.5 0.5
0 0
50 100 150 200 250 300 50 100 150 200 250 300
(f) (g) 0
50 100 150 200 250 300
0
50 100 150 200 250 300
(g) (h)
Fig. 6 (a) schematic diagram of a receiving part for K
users (b) signal at H (c) ideal signal at J (d) signal at J 1.5
threshold=5
0.5
does not exist any more. From these, the thresolding can (i)
much effectively recover the information. Fig. 7 (g) and Fig. 7. (a) receiving part without optical hard-limiter (b)
(h) show the recovered signals without and with the receiving part with optical hard-limiter (c) signal at G
optical hard-limiter, respectively. We can clearly see that (d) signal at G’ (e) signal at H (f) signal at H’ (g) signal
Fig. 7 (h) is much similar to the intended signal (Fig. 7 i). at J (h) signal at J’(i) ideal signal expected at J
Consequently, we can say that the optical hard-limiter can
effectively lower the interference effect.
A property of MVG_OMALLOOR
w=1
0 w=3
10
The probability of error per bit is defined as w=5
w=7
-2 w=9
10
10 -4
Probability of Error
PE=p(LI≥th | b=0)?p(b=0)
-6
10 -8
-10
where PE, LI, th, and b are the probability of error, light 10
-12
Probability of Error
-2
10
K−1− i
1 K − 1 w2 w2
i
K −1
(7)
PE = ∑ 1 −
2 i = th i 2 n 2n
-4
10
10-6
number of users=10
Fig. 8 shows several dependency of PE. In Fig. 8 (a), a number of users=30
number of users=50
10-8
length of code, n, is tested in different values, 200, 500, 1 1.5 2 2.5 3
threshold
3.5 4 4.5 5
10-3
[6] C. F. lam, “To Spread or Not to Spread-the Myths
10-4
of Optical CDMA,” Lasers and Electro-Optics
10
-5
Society 2000 Annual Meeting. LEOS 2000. 13th
-6
n=200
n=500
Annual Meeting. IEEE, vol.2, 2000, pp. 810 -811.
10
-7
n=1000
n=2000 [7] Sergio Verdu. Multiuser Detection. Cambridge
10
1 1.5 2 2.5 3
threshold
3.5 4 4.5 5 University Press, NY, 1998
(a)
A property of MVG_OMALLOOR
III. S TATE ESTIMATION Throughput vs. offered load for different Transmission Scheduling algorithms
1
No transmission scheduling
Assume that several nodes on an optical CDMA LAN Pseudo-cooperative
Selfish
are transmitting simultaneously on different codewords. Each 0.8 Cooperative
Normalized throughput
codeword is shifted by a different phase shift that depends on
the exact instant it was transmitted. The state of the line at 0.6
Abstract— Transmission scheduling is a media access control The throughput of an optical CDMA LAN is limited by
mechanism that prevents degradation of throughput in optical multi-user interference. When several users transmit simul-
CDMA Local Area Networks (LANs) at high offered load. Optical taneously, their packets and hence their codewords overlap.
CDMA is a multiple access technique for broadcast optical
Local Area Networks. The throughput of an optical CDMA When the optical pulses in the codeword overlap, their optical
LAN at high offered load is limited by multi-user interference. power is added. Optical pulses from one codeword can be
Interference Avoidance, a distributed, contention based media detected by receivers tuned to other codewords. As a result
access control mechanism, can prevent throughput degradation receivers may falsely detect their codewords resulting in packet
at high loads. Interference avoidance consists of state estimation errors. These false positive errors increase with offered load,
and transmission scheduling. This work analyzes algorithms for
transmission scheduling under perfect state estimation. The anal- resulting in throughput collapse.
ysis shows that transmission scheduling under specific conditions
can provide upto 30% network throughput at high offered load. Interference Avoidance is a contention media access control
This compares well to non scheduled systems which have close mechanism that prevents throughput collapse in optical LANs
to zero throughput under the same conditions. Simulations show networks at high offered load. It consists of state estimation
that the performance of transmission scheduling is independent of and transmission scheduling. State estimation is a mechanism
codeset length and degrades with increase in codeset weight. The by which nodes on the network estimate the state of the line.
results also show that the performance of transmission scheduling
does not degrade when used with realistic network traffic based Transmission scheduling is a mechanism by which nodes use
on traffic obtained from a real network. the estimated state to schedule their transmissions to avoid
KEYWORDS: Networks, Optical communication, Code divi- packet losses due to interference.
sion multiaccess, Protocols, Access control.
The contribution of this paper is the analysis of transmission
scheduling algorithms for optical CDMA under perfect state
I. I NTRODUCTION estimation. The analysis quantifies the difference between
throughput of systems with and without transmission schedul-
This work considers a shared medium, packet switched ing. The analysis shows that transmission scheduling under
optical CDMA LAN in which several nodes are connected specific conditions can provide upto 30% throughput at high
to a passive star coupler to form an all optical broadcast offered load. In contrast, non scheduled systems have close
network. Each node on the network is allocated an optical to zero throughput under the same conditions. A sensitivity
CDMA codeword to receive on. Optical CDMA codewords study of the transmission scheduling algorithms shows that the
are sequences of zeroes and ones (unipolar codewords) that performance gain depends on certain codeword parameters.
are transmitted asynchronously. The codewords are transmitted The performance is independent of codeword length and
by binary intensity modulation i.e. a one in the codeword is degrades with increase in codeword weight. The performance
represented by pulse of light. Nodes use ON-OFF keying of does not degrade with a traffic model based on traffic obtained
the codeword to transmit binary data. To transmit a 1 bit the from a real network.
codeword is sent and to transmit a 0 bit, an all zeros codeword
is sent. When a node wants to transmit, it tunes its transmitter The paper is organized as follows. Section II provides
to the receiver’s codeword and transmits. The code division background on optical CDMA. Section III-A discusses the
multiplexing allows several pairs of users to communicate motivation for Interference Avoidance. Section III-C discusses
simultaneously. the channel characteristics of optical CDMA and a represen-
tation for the state of the network. Section III-D defines the
1 This material is based upon work supported by the Defense Advanced transmission scheduling algorithms. Section IV analyzes the
Research Projects Agency under contract no. N66001-02-1-8939 issued by the performance of the transmission scheduling algorithms and
Space and Naval Warfare Systems Center (SPAWAR). Any opinions, findings, Section V discusses a sensitivity study of the algorithms.
and conclusions or recommendations expressed in this material are those of
the authors and do not necessarily reflect the views of the Defense Advanced Section VI discusses the related work in this field. Section VII
Research Projects Agency, SPAWAR, or the U.S. Government. discusses the conclusions and future work.
A property of MVG_OMALLOOR
ISI-TR-2006-617 2
0 1 1 0 0 1 0 0 C3
0 1 0 0 1 0 1 0 C2
Control
0 1 0 0 0 0 1 1 C1
0 0 0 0 0 0 0 0 C0 (0 bit)
1 1 1 1 1 3 0 1 Line
Splitter TDL Coupler
Receive
Hard- t=0 1 2 3 4 5 6 7 8 9 10
fiber Photo- Threshold
limiter TDL (a)
detector detector
0 1 1 0 0 1 0 0 C3
TDL
0 1 0 0 1 0 1 0 C2
1 0 1 0 0 0 0 1 1 C1
3 3 1
2
1 1 1 1 1 1 1 1 0 1 1 1 1 1 1 0 0 0 1 1 0 0 C0 (1 bit)
2 1 1 1 2 4 0 1 Line
t=0 1 2 3 4 5 6 7 8 9 10
(b)
0 1 1 0 0 1 0 0 C3
0 1 0 0 1 0 1 0 C2
0 1 0 0 0 0 1 1 C1 Optical CDMA
1 0 0 0 1 1 0 0 Transmitter
C0 (1 bit) Bus
(from node Transmit
2 1 1 1 2 4 0 1 Line processor) fiber
(to coupler)
Transmission
t=0 1 2 3 4 5 6 7 8 9 10 Transmit
scheduling
buffer
module
(a) Receive
fiber
0 1 1 0 0 1 0 0 C3 (from
coupler)
0 1 0 0 1 0 1 0 C2 State State
Bus estimation observation
0 1 0 0 0 0 1 1 C1 module module
(to node
1 0 0 0 1 1 0 0 C0 (1 bit) processor)
2 1 1 1 4 1 1 1 Line
Receive Optical CDMA
buffer Receiver
t=0 1 2 3 4 5 6 7 8 9 10
(b)
Fig. 4. An interference error is caused in C0 in (a). No error is caused when Fig. 5. Block diagram of an Interference Avoidance Network Interface Card.
the same codewords are sent with a different set of chip offsets. (b) shows
codeword C0 delayed by 1 chip time
0.1
TABLE III
T HE OVERLAP SECTION TRANSMISSION SCHEDULING ALGORITHM 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized offered load
1 10
Threshold scheduling Threshold scheduling
Overlap section scheduling Overlap section scheduling
0.9 Pure selfish scheduling 9 Pure selfish scheduling
Aloha-CDMA Aloha-CDMA
0.8 8
0.6 6
0.5 5
0.4 4
0.3 3
0.2 2
0.1 1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Normalized offered load Normalized offered load
Fig. 7. Comparison of the performance of the transmission scheduling Fig. 8. Comparison of the codeword multiplexing of the different transmis-
algorithms based on simulation. The traffic model is Poisson arrivals with sion scheduling algorithms based on simulation. The traffic model is Poisson
exponentially distributed packet lengths. The codeset is (10, 3, 3) and code- arrivals with exponentially distributed packet lengths. The codeset is (10, 3, 3)
words are chosen uniform randomly. For the threshold scheduling algorithm, and codewords are chosen uniform randomly. For the threshold scheduling
the threshold parameter was set to 0.5 algorithm, the threshold parameter was set to 0.5
1
Threshold scheduling
Figure 7 shows the results of simulation for the same 0.9
Overlap section scheduling
Pure selfish scheduling
codeset as described in the analytical results. The results Aloha-CDMA
0.8
are quite similar to the analytical results. All transmission
0.7
scheduling algorithms prevent throughput degradation. Also Packet error rate
codewords with a higher probability of packet error. When B. Effect of varying the weight of the codeset
packets are lost due to interference errors, the higher layers Changing the codeset weight w has two effects: As w
of the protocol stack must recover through some form of is increased, the threshold on the correlation receiver can
Automatic Repeat Request (ARQ) or Forward Error Correction be increased. Increasing the threshold makes the codewords
(FEC). Therefore the threshold and overlap section scheduling more resistant to interference. This is because it takes a
algorithms are better choices because a larger fraction of the larger number of chip overlaps to cause a false positive error.
packets transmitted are transmitted without error. However, with the increase in weight, each codeword causes
more interference with other codewords. An increase in w also
V. S ENSITIVITY ANALYSIS makes it difficult for the transmission scheduling algorithm to
A simulation based study was conducted to allow a deeper schedule codewords on the line without causing interference
sensitivity analysis of the transmission scheduling algorithms. errors. This reduces the number of codewords that can be
The study considered parameters at the physical layer (codeset simultaneously transmitted on the line. The results show that
parameters), the media access control layers (the scheduling as the weight increases the throughput decreases rapidly. The
algorithm parameters) and the traffic model (packet arrival and reduction in the codewords on the line and the increased
size distributions). The objective of this study is to quantify interference offset any gains in the resistance to interference.
the impact of these factors on the transmission scheduling It has been shown [8] that codesets with higher weight have
algorithms. The sensitivity analysis consisted of quantifying better bit error rate characteristics at low loads. At low loads
the: the increase in resistance to interference dominates resulting
in higher throughput for high weight codesets. However as the
• Effect of varying the codeset length.
load is increased, the effect of interference tends to dominate,
• Effect of varying the codeset weight.
resulting in lower throughput for high weight codesets. Fig-
• Effect of different packet size distributions.
ure 11 shows a graph of the network throughput at offered
• Performance under realistic network traffic.
load of 1 vs. the weight.
A. Effect of varying the length of the codeset C. Effect of packet size distribution
Increasing the length of the codeset has several effects. As Figure 12 shows the packet throughput as the average
N increases, the scheduling algorithm can schedule a larger packet size is varied. The traffic model is Poisson arrivals
number of codewords simultaneously on the line. Therefore and exponentially distributed packet sizes. The figure shows
more nodes can transmit in parallel without error. However that the packet size has no effect on packet throughput.
the nodes transmit at a lower data rate. The results show However, studies indicate that real network traffic packet
that the two effects balance each other and the network size distributions may not be exponential [9]. Recent packet
throughput is constant when N is varied. Figure 10 shows statistics obtained from a backbone network [10] exhibit a
a graph of network throughput at an offered load of 1 vs. the trimodal distribution. In one trace, about 70% of the packet
A property of MVG_OMALLOOR
ISI-TR-2006-617 9
1 1
Threshold scheduling Threshold scheduling
Overlap section scheduling Overlap section scheduling
0.9 Pure selfish scheduling 0.9 Pure selfish scheduling
Aloha-CDMA Aloha-CDMA
0.8 0.8
Normalized network throughput at
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 1 2 3 4 5 6 7 8 9 10 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Codeset weight w Normalized offered load
Fig. 11. Comparison of the performance of the transmission scheduling Fig. 13. Comparison of the performance of the transmission scheduling
algorithms as the codeset weight is varied (based on simulation). The traffic algorithms for a trimodal packet size distribution (based on simulation). The
model is Poisson arrivals with exponentially distributed packet lengths. The traffic model is Poisson arrivals with packet size distribution consisting of
codeset length is 100 and κ = 3. Codewords are chosen uniform randomly for 70% 40 byte packets, 20% 1500 byte packets and 10% 500 byte packets. The
the codeset. For the threshold scheduling algorithm, the threshold parameter codeset is (100, 3, 3) and codewords are chosen uniform randomly. For the
was set to 0.5 threshold scheduling algorithm, the threshold parameter was set to 0.5
1
Threshold scheduling
Overlap section scheduling
where short packets tend to experience lower error rates than
0.9 Pure selfish scheduling
Aloha-CDMA long packets [11].
0.8
The squeeze through effect can be demonstrated analytically
Normalized network throughput at
normalized offered load of 1
1 1
Simulation Threshold scheduling
Analysis Overlap section scheduling
0.9 0.9 Pure selfish scheduling
Aloha-CDMA
0.8 0.8
0.7 0.7
at an offered load of 1
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Fraction of short packets Normalized offered load
Fig. 14. The squeeze through effect. The throughput is maximized when Fig. 15. Comparison of the performance of the transmission scheduling
the fraction of short packets is 0.9. The graph shows both analytical and algorithms for a realistic traffic model (based on simulation). The traffic
simulation results. The transmission scheduling algorithm is pure selfish. The model was based on real network traffic traces (see description). The codeset
packet sizes are 50 bytes and 1000 bytes. The codeset is (100,3,3). is (100, 3, 3) and codewords are allocated to addresses. For the threshold
scheduling algorithm, the threshold parameter was set to 0.5
VI. R ELATED WORK systems state is a scalar variable and media access control
is through admission control i.e. the number of simultaneous
Work related to Interference Avoidance can be divided into users is controlled.
four areas: Bit error rate analysis of optical CDMA networks,
optical CDMA codeset design, FEC for optical CDMA and
media access control in optical/wireless networks. VII. C ONCLUSIONS AND FUTURE WORK
Salehi [5], [8] analyzed an optical CDMA based network This work has presented an analysis of transmission
and developed expressions for the bit error rate of a network scheduling algorithms for optical CDMA media access control.
that uses codesets with κ = 1. The analysis also determined The analysis quantified the difference between throughput of
the bit error rate for codesets with different lengths and systems with and without transmission scheduling and showed
weights and with hard-limiting at low loads using Aloha- that transmission scheduling achieved 30% throughput while
CDMA. This work examines these results in the context of non scheduled systems had close to zero throughput. Simula-
transmission scheduling at high offered loads. tions showed that the throughput of transmission scheduling is
The area of optical CDMA code design has focused on con- independent of codeset length. It also showed that an increase
struction of codesets with large size. Chung et al. [3] described in weight can lead to a degradation in the performance of these
several algorithms to construct OOCs. These constructions are algorithms, although the degradation is not as bad as systems
for codes with maximum crosscorrelation parameter κ = 1. without transmission scheduling. Simulations also showed that
Chung and Kumar [13] described a method for construction transmission scheduling prevents degradation when used with
of codes with κ = 2. Several construction methods for OOCs a realistic traffic model based on traffic obtained from a real
are described in [4] and [14] among others. network.
Efforts at reducing packet errors in optical CDMA have Limitations of this work include the fact that it assumes per-
mostly focused on using error correcting codes on top of fect state estimation and neglects errors due to synchronization
optical CDMA. Hsu et al. [15] analyzed the performance of and receiver contention. Future work will explore the impact
slotted and unslotted optical CDMA packet networks. They of realistic state estimation.
developed expressions for the throughput of the network and Work in progress includes a testbed implementation of the
showed performance can be improved using Forward Error transmission scheduling hardware. The testbed demonstrates
Correction (FEC) codes and hard limiters. Muckenheim et a simplified form of threshold transmission scheduling by
al. [16] studied the effect of bit error probability on the packet transmitting bits such that the number of chip overlaps is
error probability and suggested the use of block codes to constrained. Measurements indicate that the bit error rate is
reduce packet errors. They also described a random delay substantially lower for this system than for a system without
protocol to reduce the errors incurred during periods of high transmission scheduling.
activity and showed throughput improvement. The mechanism
detects periods of high activity and defers transmissions. In
contrast Interference Avoidance does not use any FEC and R EFERENCES
schedules packet transmissions to avoid interference.
[1] B. Mukherjee, “WDM based local lightwave networks. Part I: Single
Contention media access control mechanisms such as Car- hop systems,” IEEE Network Magazine, vol. 6, no. 3, pp. 12–27, May
rier Sense and Collision Detection for optical networks have 1992.
[2] A. Keshavarizian and J. A. Salehi, “Optical orthogonal code acquisi-
been studied in [17]. Interference Avoidance in optical CDMA tion in fiber optic CDMA systems via simple search method,” IEEE
is analogous to carrier sensing in single channel shared Transactions on Communications, vol. 50, no. 3, pp. 473–483, March
medium networks. Interference Avoidance can be viewed as 2002.
the result of applying CSMA design principles to the optical [3] H. Chung, J. Salehi, and V. K. Wei, “Optical orthogonal codes: Design,
analysis, and applications,” IEEE Transactions on Information theory,
CDMA physical layer. vol. 35, no. 3, pp. 595–605, May 1989.
Synchronous-CDMA [18] is a CDMA based physical layer [4] O. Moreno, Z. Zhang, P. Kumar, and V. Zinoviev, “New constructions
of optimal cyclically permutable constant weight codes,” IEEE Trans-
for cable networks. Nodes reserve codewords to use in a time actions on Information Theory, vol. 41, no. 2, March 1995.
slot through a centralized controller. This mechanism is similar [5] J. Salehi, “Code division multiple-access techniques in optical fiber
to the Interference Avoidance techniques in that it performs networks - Part 1: Fundamental principles,” IEEE Transactions on
admission control and restricts the number of simultaneous Communications, vol. 37, no. 8, pp. 824–833, Aug. 1989.
[6] P. Kamath, J. D. Touch, and J. A. Bannister, “The need for media access
users on the network. However it is centralized and admission control in optical CDMA networks,” in IEEE Infocom, vol. 4, March
control is static (by codeword assignment). 2004, pp. 2208–2219.
In wireless networks, several mechanisms have been pro- [7] F. A. Tobagi and V. B. Hunt, “Performance analysis of carrier sense
multiple access with collision detection,” Computer Networks, vol. 4,
posed to take advantage of knowledge about channel condi- no. 5, pp. 245–259, Oct. 1980.
tions. State in wireless networks has to account for the chan- [8] J. Salehi and C. Brackett, “Code division multiple-access techniques in
nel noise and multi-path effects. Channel load sensing [19], optical fiber networks - Part 2: Systems performance analysis,” IEEE
Transactions on Communications, vol. 37, no. 8, pp. 834–842, Aug.
[20], [21], MIMO/CSIT based media access control [22] and 1989.
scheduling algorithms [23] are systems where wireless state [9] V. Paxson and S. Floyd, “Wide area traffic: the failure of Poisson
has been used to schedule transmissions. The differences modeling,” IEEE/ACM Transactions on Networking, vol. 3, no. 3, pp.
226–244, 1995.
between these mechanisms and Interference Avoidance are [10] Sprint ATL Labs, “Packet size distributions from an internet backbone,”
the state description and the scheduling algorithms. In these www.sprintlabs.com, March 2005.
A property of MVG_OMALLOOR
ISI-TR-2006-617 12
[11] B. S. Bakshi, P. Krishna, N. H. Vaidya, and D. K. Pradhan, “Improving For any other transmission scheduling algorithm, the same
performance of TCP over wireless networks,” in International Confer- average number of packets are offered for transmission. How-
ence on Distributed Computing Systems (IDCS), May 1997, pp. 365–
373. ever the transmission scheduling algorithm does not allow all
[12] NLANR, “Traffic traces from an OC48 link,” www.nlanr.net, March of these packets to be transmitted. Let the average number
2005. of packets on the line at any point of a receive fiber be
[13] H. Chung and P. V. Kumar, “Optical orthogonal codes - new bounds
and an optimal construction,” IEEE Transactions on Information Theory, Nonline , of which a fraction Pe are lost due to error. The
vol. 36, no. 4, pp. 866–873, July 1990. ratio of the average number of error free packets transmitted
[14] R. Fuji-Hara and Y. Miao, “Optical orthogonal codes: Their bounds and to the average number of packets offered for transmission is
new optimal constructions,” IEEE Transactions on Information Theory,
vol. 46, no. 7, pp. 2396–2406, Nov. 2000. Nonline (1 − Pe )/(λ/µ)
[15] C. S. Hsu and V. O. K. Li, “Performance analysis of slotted fiber- Therefore,
optic code division multiple access (CDMA) packet networks,” IEEE
Transactions on Communications, vol. 45, no. 7, pp. 819 – 828, July T hnorm = (Nonline (1 − Pe )/(λ/µ))ρ
1997. = (Nonline (1 − Pe )/(λ/µ))(λ/µN )
[16] J. Muckenheim, K. Iversen, and D. Hampicke, “Construction of high-
efficient optical CDMA computer networks: Statistical design,” in IEEE = Nonline (1 − Pe )/N
International Conference on Communications, vol. 3, June 1998, pp.
1289–1293.
[17] C. Chae, E. Wong, and R. Tuckker, “Ethernet over passive optical A PPENDIX II
network based on optical CSMA/CD media access technique,” in In- N ORMALIZED NETWORK THROUGHPUT
ternational Symposium on Services and Local access, April 2002.
[18] Cable Television Labs Inc., “Data Over Cable Service Interface Spec- This appendix derives an expression for the normalized
ifications (DOCSIS) 2.0, Radio Frequency Interface Specification,” no. network throughput of an Interference Avoidance based optical
SP-RFIv2.0, April 2004.
[19] A. H. Abdelmonem and T. N. Saadawi, “Performance analysis of spread
CDMA LAN. First, a concise representation of line state which
spectrum packet radio network with channel load sensing,” IEEE Journal allows easy mathematical manipulation is defined. Using this
on Special Areas in Communications, vol. 7, no. 1, pp. 161–166, Jan. state representation, expressions are derived for the number of
1989.
[20] G. Judge and F. Takawira, “Spread Spectrum CDMA Packet Radio MAC
codewords on the line and the probability of error when the
Protocol using Channel Overload Detection and Blocking,” Wireless system is in any state.
Networks, vol. 6, pp. 467–479, December 2000.
[21] K. Toshimitsu, T. Yamazoto, M. Katayama, and A. Ogawa, “A novel
spread slotted Aloha system with channel load sensing protocol,” IEEE A. State representation
Journal on Special Areas in Communications, vol. 12, no. 4, pp. 665–
672, May 1994. The state of the line can be represented by a pair (n0 , n1 )
[22] S. A. Reddy and L. Tong, “Exploiting decentralized channel state infor- where n0 < N and n1 < N . n0 is the number of zeros
mation for random access,” IEEE Transations on Information Theory,
vol. 51, no. 2, pp. 537–561, Feb. 2005. in the true state and n1 is the number of ones. The number
[23] T. Shepard, “Decentralized channel management in scalable multihop of overlaps is nov = N - (n1 + n0 ). The term state will be
spread-spectrum packet radio networks,” PhD. Thesis, no. MIT/LCS/TR- used to refer to this reduced representation of the state of the
670, 1995.
line. The state of the line could be any value (n0 , n1 ) where
0 ≤ n0 ≤ N and 0 ≤ n1 ≤ N . A valid state is defined
A PPENDIX I as a state where n0 + n1 ≤ N . All other states are invalid
O PTICAL CDMA LAN PERFORMANCE METRICS (when n0 + n1 > N ) The initial state is defined as the state
Let the aggregate arrival rate of packets to the network be of the line when no codewords are on the line i.e. (N, 0). A
λ packets/s. If the average packet size is L bytes, the chipping reachable state is defined as a state which can be reached from
rate is B chips/s, then the maximum possible packet arrival the initial state by a series of state transitions due to packet
rate on a single channel network is B/8L packets/s. Therefore, arrivals. The initial state (N, 0) is, by definition, a reachable
the normalized offered load ρ is, state. The set of reachable states depends on the transmission
scheduling algorithm. When there are no codewords on the
ρ = 8Lλ/B line i.e. no node is transmitting packets, the state at a point
on the line is the initial state i.e. (N , 0), i.e. N zeroes and
Consider an optical CDMA LAN where the arrival rate of
0 ones and overlaps. The arrival of a single codeword adds
packets is λ packets/s. The service rate µ (in packets/s) of the
w ‘1 chips’ to the state of the line and the state changes.
network is the rate at which packets are transmitted on the
This is called a state transition. Let the start state of a state
network. For an optical CDMA LAN using a codeset (N ,
transition be (f romn0 , f romn1 ) and the destination state be
w, κ), an average packet size of L bytes, and a chipping
(ton0 , ton1 ). A state transition may be caused only by a packet
rate B chips/s the service rate is µ = B/(8LN ) packets/s.
arrival or departure4. When a codeword is added to the line,
Substituting for B in the expression above,
w 1 chips are added. A ‘1 chip’ could overlap with a 0, 1 or
ρ = 8Lλ/(8LN µ)) = λ/µN an overlap. Let the number of ‘1 chips’ overlapping with 0s,
1s and overlaps be c0 , c1 and coverlap respectively.
Consider an Aloha-CDMA system where packets are trans-
mitted on arrival. Consider any point on a receive fiber. If each 4 The effect of ON-OFF keyed modulation is neglected by assuming that a
packet takes an average time of 1/µ seconds at that point, then packet consists of only 1 data bits i.e. all data bits are ON. Packet arrivals
and departures are assumed to be the sole cause of any state change. This
Little’s law, the average number of packets at that point (on assumption means that the probability of error calculated is higher that the
the line) is λ/µ. true value and the throughput is the worst case throughput.
A property of MVG_OMALLOOR
ISI-TR-2006-617 13
taken to reach that state. Under these circumstances, the state These equations can be solved for the equilibrium state
transition diagram for arrivals and departures is a Markov probabilities Pstate (s0 , s1 ).
chain. Equilibrium probabilities may be found by solving the
balance equations for the system.
B. Admissible transmissions
This subsection describes how to identify admissible state
transitions given a codeset and a transmission scheduling
algorithm.6
1) Aloha-CDMA: For Aloha-CDMA, all transitions are
admissible.
2) Pure selfish scheduling: A transmission is admissible if,
c0 ≥ 1
3) Threshold scheduling: A transmission is admissible if
c0 ≥ 1
tonoverlap ≤ αN
where α is the threshold parameter.
4) Overlap section scheduling: A transmission is admissi-
ble if
c0 ≥ 1
tonoverlap ≤ ton1