Sunteți pe pagina 1din 11

WiMAX Services over Transport Networks

Dayou Qian † , Dirceu Cavendish, Ting Wang

C & C Research Laboratories


NEC USA Inc.
4 Independence Way
Princeton, NJ 08544, USA
† Corresponding author E-mail:dqian@nec-labs.com

Abstract - The merging of long-haul carriers with wireless operators signals significant changes in
telecommunications services in the US. Wireless access is likely to become ubiquitous, while efficient fiber
based transport provides the necessary bandwidth for aggregated traffic transport. Recently, WiMAX wire-
less technology had been tried out in places across the world. WiMAX technology promises a quick and
cheap deployment of broadband and voice services over a non-line-of-sight wireless channel. In addition,
WiMax radios probe the air channel so as to dynamically adapt to atmospheric conditions. This paper
addresses the provisioning of voice and data services over a combined WiMAX/optical network. We ad-
dress issues such as wireless channel call blocking of voice services and transport bandwidth provisioning
schemes vis-a-vis packet loss performance of a wireless/transport network.
Keywords: Orthogonal Frequency Division Multiplexing, Synchronous Optical Network, Wavelength Di-
vision Multiplexing, Optical Transport Networks

I. I NTRODUCTION
This paper addresses the bandwidth management of WiMax air interface and transport network in the
provisioning of voice and data services. WiMax radios support adaptive codings according to atmospheric
conditions, adopting various operation modes. Hence air channel bandwidth varies dynamically. In this
paper, we study how to provide voice and data services over a WiMax over transport network platform.
Specifically, we address bandwidth partitioning between voice and data services, as well as bandwidth
adjustments of provisioned transport pipes in reaction to air channel bandwidth variations. Our contributions
can be summarized as:
• Voice service model to compute call blocking probabilities as a function of bandwidth partitioned for
voice services.
• Bandwidth management scheme for adjusting transport pipes to air channel variations.
For large applicability of our work, we do not specify which transport technology to be used, but as-
sume that transport provisioning and bandwidth adjustments at the transport layer takes 50 msecs to take
effect. This is in-line with next generation SONET/SDH nodes implementing Virtual Concatenation and
Link Capacity Adjustment Schemes.
In what follows, we first model the WiMax air channel adaptive behavior. Then, we propose a discrete
time markov chain to model voice circuits over the wireless channel. For data services, we propose a
bandwidth adjustment algorithm to reflect the air channel bandwidth variations into the transport provisioned
pipe.

II. M ODELING THE AIR CHANNEL

At the physical layer, many wireless channels such as IEEE 802.11/15/16 [9] implement an Adaptive
Modulation and Coding (AMC) scheme. AMC aims at maximizing the data rate by probing the air trans-
mission and adjusting transmission parameters to channel variations, while maintaining a prescribed packet
error rate [1]. Therefore, there are N transmission modes available, each of which used at specific SNR
air channel conditions. Assuming constant power transmission, and partitioning the SNR range into N + 1
non-overlapping consecutive intervals, with boundary points denoted by γn , 0 ≤ n ≤ N + 1 SNR values.
Assuming we know these SNR thresholds for each modulation and coding mode in order to satisfy a given
bit error rate (BER):

mode n is chosen, whenγ ∈ [γn , γn+1 ) (1)


where n is the mode index and γ is the received SNR. The computation of γn values for a given Packet
Error Rate (PER) can be found in [1], hence it is assumed known hereafter.
Similar to [2], at the hardware-radio layer, the average received SNR γ can be expressed as

γ = Pt − Li − PN − Pp (2)
where Pt (dBm) is the transmitter output power; Li (dB) is the implementation loss due to hardware (e.g.,
connecting cable and antenna patterns); PN (dBm) is the receiver noise power, which is related to hardware
noise figure and bandwidth; and Lp (dB) is the path loss due to radio propagation, which is based on well-
established models [6], [7]. Following [2], we adopt the following expression for Lp :

Lp = G1 + G2 log10 fc + G3 log10 d (3)

where G1 , G2 , and G3 are constants that depend on application-specific operating conditions; fc (Hz) is the
carrier frequency; and d(m) is the transmitter-receiver distance.
We next follow [5] in order to model the adaptive air channel. Let Pl,n be the transition probability
between modes l and n. Assuming slow fading conditions so that transitions happen only between adjacent
modes, the probability of transition exceeding two consecutive modes is zero, or:

Pl,n = 0, |l − n| ≥ 2 (4)
The adjacent-state transition probability can be determined by:
Nn+1 Tf
Pn,n+1 = , if n = 0, . . . , N − 1 (5)
P r(n)
Nn Tf
Pn,n−1 = , if n = 1, . . . , N (6)
P r(n)
where Nn is the cross-rate of mode n (either upward or downward), which can be estimated [4] as:

mγn fd mγn m−1 mγn


r
Nn = 2π ( ) exp(− ) (7)
γ Γ(m) γ γ
where fd denotes the mobility-induced Doppler spread. P r(n) is the probability that mode n is chosen, and
is given by [1]:

Z γn+1
P r(n) = pγ (γ)dγ
γn
Γ(m, mγn /γ) − Γ(m, mγn+1 /γ)
= (8)
Γ(m)

where Γ(m, x) := x∞ tm−1 e−t dt is the complementary incomplete Gamma function, and m is the Nak-
R

agami parameter.
From IEEE802.16.2004 Table 338(page 627) [9], we find the SNR thresholds for different modulation
modes based on a target bit error rate of BER = 10−6 .

2
Modulation BPSK QPSK 16QAM 64QAM
Coding rate 1/2 3/4 1/2 3/4 1/2 3/4 2/3 3/4
Data rate(Mbits/s) 1.89 2.84 3.95 6.0 8.06 12.18 16.30 18.36
SNR threshold(dB) 6.2 8.2 9.4 11.2 16.4 18.2 22.7 24.4
Table I: SNR threshold for modulation and coding modes in WiMax, Bandwidth 5MHz

Here after, we assume d = 800(m), fc = 900(M Hz), Li = 5(dB) and PN = −130(dBm). We also
adopt Lee’s path-loss model, so we use Lp = 130(dB) from Fig.2.46 in [7] for the Philadelphia urban area.
We also assume m = 1.0 and fd Tf = 0.01. With the SNR thresholds in Table I, from Eqs. 5, we obtain the
transition matrix for the WiMax channel. Fig. 1 illustrates the WiMax channel behavior in the low power
case, Pt = 20(dB). Notice that for low power case, the channel tends to dwells mostly at the lowest and
highest modes. Long incursions into low rate modes is particularly undesirable for voice services.

WiMax Channel
40

35

30
Capacity (Byte/Frame)

25

20

15

10

0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Frame (2 ms/frame)
Fig. 1. WiMax channel modelling - 25dB power case

III. VOICE S ERVICES OVER W I MAX/ TRANSPORT


WiMax (802.16) fixed wireless technology [9] supports T1/E1 services over wireless OFDM link. A T1
line carries 24 64Kbps voice circuits (1.544Mbps), wereas an E1 line carries 32 voice circuits (2.048Mbps).
As shown in Table I, the total capacity of the wireless channel can vary from 30Mbps to 90Mbps, depend-
ing on the channel mode. Moreover, for voice services, we assume no multiplexing of transport network
resources, that is, each voice circuit in service at the wireless link gets a 64Kbps bandwidth allocation at the
transport pipe. In what follows, we study the call blocking performance of the WiMax channel via a Markov
Chain model, which provides upper bounds, as well as via simulations.

A. Voice Service Finite State Markov Chain Model


We assume A(t) = 1 − eλt to be the PDF of the sojour time between voice circuit requests in the wireless
channel. The assumption is justifiable, provided that the voice channel is used for human conversation, and
not other applications running on top of the wireless voice circuit (e.g., dial-up internet access). For similar
considerations, we assume L(t) = 1 − eµt to be the PDF of the voice circuit service duration.
We here follow an imbedded Markov Chain technique [8]. Let qn be the number of calls in the system at
channel frame boundaries. Moreover, let:

def
pi,j = P [qn+1 = j, qn = i] (9)

3
be the transition probability between states qn+1 and qn . Then, within a fixed time frame F , pi,j can be
written as:

(λF )(j−i+a) e−λF (µF )a e−µF


pi,j = Σia=0 (10)
(j − i + a)! a!

The imbedded Markov Chain for the voice service is depicted in Fig. 2

0,u i-1 c-j,u i-1 c-2,u i-1 c-1,u i-1 c,u i-1
p c,c-1 Pu
i-1,u i-1
p c,c-2 P
u
p c,0 P
u
i-1,u i-1
i-1,u i-1
p c,c Pu
i ,ui-1
p c,c Pu
p c,0 P p c,c-1 Pu i-1,u i
u ,u
p c,c-2 P
u ,u
i ,ui-1
i i-1 i i-1
p c,Nu Pu ,u
i i i
p c,cPu
i ,ui p c,c+2 Pu ,u
i i

p c,c+1 Pu ,u
i i
0,u i c-j,u i c-2,u i c-1,u i c,u i c+1,u i c+2,u i c+k,u i Nui ,u i
p c,c-1 Pu ,u
i i

p c,c-2 P
p c,0 Pu u ,u
i i
i ,ui

p c,c+1 Pu
p c,c Pu p c,c Pu i ,ui+1 p c,Nu P
i+1,u i i ,ui+1 p c,c+2 Pu i+1 ui ,ui+1
i ,ui+1

p c,Nu P
p c,c+2 Pu ,u i+1 ui+1,u i+1
i+1 i+1

p c,c+1 Pu ,u
i+1 i+1 Nui+1,
c,u i+1 c+1,u i+1 c+2,u i+1 c+k,ui+1
u i+1
c : # of voice circuits in service
u i : wireless channel mode
Nu i : Max capacity of wireless channel at mode u , in circuits

Fig. 2. WiMax Markov Chain for Voice Service

where pij is given by Eq. 10, and Pn,n+1 , Pn,n−1 are given by Eqs. (5,6), respectively. Notice that Fig. 2
depicts the states and transition probabilities regarding three channel modes only: ui−1 , ui , ui+1 . For sake
of clarity, the picture focuses on state ui transitions only, hence it does not contain all possible transitions.
Nui , 0 ≤ i ≤ N is the capacity of the air channel when operating at mode i, in number of circuits.
We further define X n as the random variable representing the Markov Chain of Fig. 2 at time n, and πc,u t

as the probability that the system is found at state c, u at time t, then:

t def
πc,u = P [X t = c, u] = P (Ct = c, Ut = u) (11)

If the discrete markov chain is irreducible and aperiodic, then a limit distribution πc,u exists, or

t def
πc,u = lim πc,u (12)
t→∞

Moreover, this stationary distribution is independent of the initial probability distribution of the Markov
Chain states. Therefore, if the transition probability matrix is P , we may write:

t t−1
πc,u = πc,u P (13)

If the initial MC state probability is πc, u0 , then we may iterate Eq. 13 until we converge to a stationary
distribution πc,u . The processing of Eq. 13 can be done by any numerical analysis package, such as MatLab.

4
Let Bt and Dt denote the number of calls blocked and dropped at time t respectively. We can express
them as

Dt = max{0, max{0, Ut − Lt } − Ct } (14)


Bt + Dt = max{0, At + max{0, Ut − Lt } − Ct } (15)

which depends on At ,Lt ,Ut and Ct . We wish to find the stationary behavior of Bt and Dt , as t → ∞.
From equation (13), we have the stationary distribution for (C, U ). Since At and Lt are stationary, the
limiting distribution of At and Lt exist as t → ∞. Letting A := limt→∞ At and L := limt→∞ Lt , we have

P (A = a) = P (At = a) (16)
P (L = l) = P (Lt = l) (17)

and E{A} = E{At } = λTf . From Eq. (14), the stationary distribution of Dt exist and is given by

D: = lim Dt
t→∞
= lim max{0, max{0, Ut − Lt } − Ct }
t→∞
= max{0, max{0, U − L} − C} (18)

Similarly, we can have

B+D : = lim (Bt + Dt )


t→∞
= lim max{0, At + max{0, Ut − Lt } − Ct }
t→∞
= max{0, A + max{0, U − L} − C} (19)

Using Eq. (18), the ensemble-average number of calls dropped per time-unit can be calculated as Eq.
(20).
X
E{D} = DP (C = c, U = u, L = l)
c,u,l
X
= [max{0, max{0, u − l} − c}P (C = c, U = u)P (L = l)] (20)
c,u,l

Using Eq. (19), the ensemble-average number of calls blocked per time-unit can be calculated from Eq.
(20) and (21).
X
E{B + D} = (B + D)P (A = a, C = c, U = u, L = l)
a,c,u,l
X
= [max{0, a + max{0, u − l} − c}P (A = a)P (C = c, U = u)P (L = l)] (21)
a,c,u,l

Based on above equations, we can calculate dropped probability Pd and blocking probability Pb as
PT
Dt E{D} E{D}
Pd : = lim Pt=1
T
= = (22)
T →∞
t=1 At E{A} λTf
PT
Bt E{B + D} − E{D} E{B + D} − E{D}
Pb : = lim PTt=1 = = (23)
T →∞
t=1 At E{A} λTf

5
In summary, given a set of circuit capacities for the air channel modes Nui , 0 ≤ i ≤ N , and a voice
service demand ( λ, µ ), Eq. 22 and 23 provide estimates of the call dropped and blocking probabilities.
In Fig. 3, we study how well WiMax channel supports voice circuits. For the discrete-time call arrival
model, we assume 1/λ ∈ [0.1, 180](sec) to be the mean interarrival time and 1/µ = 180(sec) the mean call
duration. The figure includes both analytical blocking probabilities from the Markov Chain just described,
as well as simulation results, as follows. The wireless channel is simulated based on the model of the
previous section. Calls arrive at the WiMax channel at various rates, and their blocking and dropping rates
are measured. Blocking happens when newly arrived calls find the channel full, whereas dropped calls are
those ongoing calls which need to be dropped due to a reduction on the channel capacity generated by a mode
change. We assume that a circuit can be retained for at most time τ even without the needed bandwidth,
before being dropped. This retaining parameter is customary at cellular networks, with typical values of the
order of hundreds of milliseconds [3]. The simulation is carried out for one week traffic activity.

Voice Traffic Analysis


0.8

Blocking rate (τ=0.3s)


0.7 Dropped rate (τ=0.3s)
Blocking rate (τ=0.5s)
Dropped rate (τ=0.5s)
Blocking rate (τ=0.7s)
0.6 Dropped rate (τ=0.7s)
Call Unsuccessful Rate

0.5

0.4

0.3

0.2

0.1

0
0 20 40 60 80 100 120 140 160 180
Call interarrival time(s)
Fig. 3. WiMax Voice Service Performance

The figure shows that the call blocking is high at heavy loads only, having a fairly flat and long curve for
light loads. The dropping rate, which is the fraction of calls dropped due to channel mode changes is highly
sensitive to the retaining parameter τ , as expected.

IV. DATA S ERVICES OVER W I M AX /T RANSPORT


WiMax (802.16) radios operate over an unlicensed and licensed spectra over longer distances, say 10-20
Km, on Line Of Sight (LOS) and non-line-of-sight (NLOS) modes. Typical total channel speeds are on the
order of 300Mbps. For voice circuits, there is no statistical multiplexing of the various 64Kbps channels
from the wireless link into the transport network. Therefore, the capacity of the transport pipe supporting
voice services should match the maximum number of circuits served by the wireless link. For the data
services, however, statistical multiplexing can be exercised at the wireless/transport interface, since buffers
must be present for the mapping/demapping of datagrams anyway. Moreover, the bandwidth variability of
the wireless access link should be taken into account when provisioning transport pipes, since air quality
does not change drastically and frequently.
In what follows, we explore bandwidth management issues concerning data services in this newly in-
tegrated wireless over transport architecture. We seek to match the air channel bandwidth used for data
services with a transport pipe specifically for data services.

6
A. Data mapping and demapping
By data mapping we refer to the encapsulation of data packets coming from the wireless link into the
transport network. Fig. 4 a) illustrates the process.
By data demapping we refer to the decapsulation of data packets coming from the transport network and
pushed into the wireless link. Fig. 4 b) illustrates the process.
a(t) s(t)
Mapper

wireless link wired transport


a) Wireless over transport mapping

s(t) a(t)
Demapper

wireless link wired transport

b) Wireless over transport demapping


Fig. 4. Packet over Transport - mapping and demapping

A.1 Data mapping


Data packets coming from the wireless link are mapped into a fixed bandwidth provided by the transport
network. The typical scenario is the mapping of data packets into Synchronous Payload Envelope (SPE),
via Generic Frame Procedure (GFP). Moreover, data packets may come from different applications running
over the broadband access service. Therefore, we anticipate data packets arriving with various priorities.
For instance, video streams would have high priority, whereas web browsing would have low priority.
Transport channel provisioning involves the sizing of the channel according with the expected traffic
coming from the wireless link. Moreover, channel mode changes affect the packet arrival process at the
transport edge node.
If we plot the transport bandwidth provisioning and the WiMax channel bandwidth curves, the provi-
sioning curve should be an envelope of the wireless bit rate curve, so as to guarantee no packet drop at the
interface due to congestion, while minimizing buffering. Notice that the wireless bandwidth curve varies
much faster than the transport provisioning curve, because typically, transport bandwidth adjustments can
be done within 50 msec, whereas mode change can happen on a frame by frame, 1 to 2 msec basis. Clearly,
the shape of the provisioning curve depends on the provisioning strategy adopted for the transport network.
Since buffers are expected to be scarce at the wireless/transport interface, the provisioning curve should
never go below the wireless channel bit rate. On the other hand, a large gap between these two curves re-
sults in transport bandwidth wastage. Therefore, an efficient provisioning strategy should seek to minimize
the gap between these two curves while preventing them from intersecting.
Let Rti be the transmission rate of the wireless channel in mode i at frame t. Let T be a time window,
j
defining a sequence R of rates R = Rti , Rt+1 k . We define the uplink average wireless channel
, · · · , Rt+T
capacity as:

Pt+T
up−wl t Rti
Ct+T = (24)
T
Moreover, let Cttr be the capacity of the transport pipe allocated for the uplink wireless channel. Since it
is reasonable to seek to match the wireless and transport channel capacities, we have:

up−wl
Cttr = αCt+T (25)

7
where α is a multiplexing parameter. Notice that fluctuations on wireless channel capacity within a window
T may momentarily have the wireless link deliver more traffic than the transport pipe may accomodate,
hence wireless/transport interface buffers, common at any transport mapping interfaces, should be dimen-
sioned to cope with these transients.
It is not feasible to estimate the wireless capacity via Eq. 24 at every wireless frame boundary, which
happens every 1 msec. Therefore, we propose a “jumping window” mechanism for transport bandwidth
adjustment, as illustrated in Fig. 5.
Air channel BW
estimation epochs

802.16frame

t
T
transport
jumping
window
Fig. 5. Jumping Window Wireless Channel Bandwidth Estimation

The time window T within which Equation 24 is computed, is slid by T /2 frames, before a new estimation
of the air channel bandwidth is made. Let Cttr be the current transport bandwidth provisioned for data
up−wl
services. Once a new value Ct+T /2 is computed, we make provisioning adjustments onoly if there have
been significant changes in the air channel bandwidth. That is, the transport bandwidth adjustment strategy
is:


 αC up−wl wl
if αCt+T tr
tr t+T /2 /2 − Ct > Mup
Ct+T = (26)
/2  αC up−wl if Cttr − αCt+T
wl
t+T /2 /2 > Mdw

where Mup , Mdw are parameters that define a minimum amount of “significant up/down change” in the
bandwidth to warrant an adjustment of the transport provisioned bandwidth. Notice that the proposed trans-
port bandwidth adjustment scheme is based on the air channel capacity, not on the air channel utilization.
That is, the bandwidth adjustment scheme is not based on the traffic load carried within the air channel, or
channel utilization. This means that even if there is no uplink traffic, the transport bandwidth is adjusted
according to the air channel capacity. This is because data traffic is bursty in nature, so the transport pipe,
with its inherent provisioning latency, should not be adjusted based on traffic fluctuations. Otherwise, large
buffers would be required for holding bursty traffic. For instance, if an innactive air channel interface trig-
gered a zero transport bandwidth adjustment, a burst of 50Mbps would require 2.5 Mbits to be stored before
the transport bandwidth be adjusted, assumming a 50msec reprovisioning time.
Uplink dynamic behavior has been studied by simulating the system of Fig. 4 a), with wireless channel
model of section II, and various buffer sizes to accomodate frames during transient mismatches of air and
transport channel bandwidth. The following parameters were used: T = 500msec, Mup = 0.8M bits/sec, Mdw =
2M bits/sec. The simulation runs for 100 hours.
From the figure it is clear that improvements on packet loss performance due to large buffers is limited.
Although too short buffers cause larger packet losses, during momentary wireless channel fading, large
buffers do not seem to help packet loss performance (10M, 100M, and 1G buffer curves coincide). This
shows that buffers help with temporary fading of the WiMax channel only.
Fig. 7 shows the Sonet channel capacity with bandwidth management for uplink, with multiplexing
parameter α = 1.2. From Fig. 6, α = 1.2 is good enough to eliminate packet loss for large buffers. The
average Sonet utilized bandwidth in this case is 8.0229M bits/s, which is just 43.7% of the WiMax channel

8
Uplink Packet Loss Rate v.s. α
0.7
buffer size 100k
buffer size 10M
buffer size 100M
0.6 buffer size 1G

0.5

Packet Loss Rate


0.4

0.3

0.2

0.1

0
0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
α
Fig. 6. Packet Loss Performance of an Uplink WiMax over Transport interface

WiMax Channel (uplink)


25
WiMax channel
Sonet Uplink Channel

20
Capacity (Mbits/sec)

15

10

0
0 2000 4000 6000 8000 10000 12000
Time (ms)
Fig. 7. Uplink Bandwidth Transport Adjustment

bandwidth peak value of 18.36M bits/s. The figure shows that the bandwidth adjustment scheme proposed
follows quite well the wireless channel dynamics, except for a latency due to the assumed 50msec time
taken to re-provision the transport bandwidth.

A.2 Data demapping


At the downlink interface, as far as bandwidth management is concerned, the ideal would be to adjust
the air channel to accomodate the traffic coming out of the transport network. Unfortunately, this is not
feasible. Instead, we adjust the transport bandwidth to match the downlink channel capacity, using a similar
air channel bandwidth estimation as:

Pt+T
dw−wl t Rti
Ct+T = (27)
T
and the transport capacity then becomes:

Cttr = βCt+T
dw−wl
(28)

Hence, a connection that originates and ends at wireless channels will affect the transport bandwidth
according to:

up−wl dw−wl
Cttr = min{αCt+T , βCt+T } (29)

9
The transport bandwidth provisioning dictated by Eq. 29 is based on the rational that it is better to drop
traffic at the entrance of the transport network, where transport network resources have not been used, than
after the traffic has crossed the entire transport network, at the downlink interface.
The downlink system of Fig. 4 b) has been simulated for 100 hours, for various buffer sizes and multiplex-
ing factors (β), with the following parameters: T = 500msec, Mup = 2M bits/sec, Mdw = 0.8M bits/sec.

Downlink Packet Loss Rate v.s.β


0.5
buffer size 1M
buffer size 100M
0.45 buffer size 100K

0.4

0.35
Packet Loss Rate

0.3

0.25

0.2

0.15

0.1

0.05

0
0.9 1 1.1 1.2 1.3 1.4 1.5 1.6
β
Fig. 8. Packet Loss Performance of a Downlink WiMax over Transport interface

Similar to the uplink result, large buffers help only with momentary wireless channel fading, being unable
to reduce packet loss for the case of significant multiplexing (β >> 1).
WiMax Channel (downlink)
20
WiMax channel
Sonet Downlink Channel
18

16

14
Capacity (Mbits/sec)

12

10

0
0 2000 4000 6000 8000 10000 12000
Time (ms)
Fig. 9. Downlink Bandwidth Transport Adjustment

Fig. 9 shows the Sonet channel capacity with bandwidth management for downlink, with β = 0.8.
The average Sonet bandwidth provisioned is 5.0396M bits/s, which is just 27.45% of the WiMax channel
bandwidth peak value 18.36M bits/s. Apart from the latency involved, the transport bandwidth adjustment
scheme follows the dynamics of the WiMax channel closely.
We finally study the relation between the bandwidth adjustment thresholds Mup , and Mdw and packet loss
performance, in Fig. 10. The downlink simulation was run for 1 hour, and for 100Mbytes buffer size, with
β = 1. The figure shows that packet loss performance is more sensitive to the down threshold Mdw than
the up threshold Mup . This is because the WiMax channel fading occurs much faster than the bandwidth
adjustment, which we assume it takes 50 msecs, so the adjustment mechanism needs to be more responsive
to the wireless channel fading than channel capacity increase.

10
Packet Loss Rate with Different Thresholds Parameters

0.25

0.2

Packet loss rate


0.15

0.1

0.05

0
8

6 8
7
4 6
5
4
2 3
2
0 1
Mup (Mbits/sec) M (Mbits/sec)
down
Fig. 10. Packet loss - Threshold sensitivity

V. C ONCLUSION
In this document, we have studied bandwidth resource allocation on an adaptive wireless channel and
transport interface, for serving voice and data traffic. We have devised a discrete time markov chain to esti-
mate call blocking probability of a slow fading wireless channel and compared it against simulation results.
In addition, we have proposed a bandwidth adjustmet scheme for transport provisioned sessions to support
data services originated/terminated at wireless adaptive channels. We have investigated the performance of
these schemes with regard to buffer requirements, as well as the algorithms’ threshold parameters.

R EFERENCES
[1] M. S. Alouini and A. J. Goldsmith, “Adaptive Modulation over Nakagami Fading Channels”, Kluwer Journal on Wireless
Communications, Vol. 13, No. 1-2, pp. 119-143, May 2000.
[2] Q. Liu, S. Zhou, G. B. Giannakis, “Queuing With Adaptive Modulation and Coding Over Wireless Links: Cross-Layer
Analysis and Design”, IEEE Transactions on Wireless Communications, Vol. 4, No. 3, pp. 1142-1153, May 2005.
[3] M. Lundan and I. D. D. Curcio, “Mobile Streaming Services in WCDMA Networks”, Proc. of the 10th IEEE Symposium on
Computers and Communications (ISCC 2005), pp. 231-236, June 2005.
[4] M. D. Yacoub, J. E. V. Bautista, and L. G. de R. Guedes, “On Higher Order Statistics of the Nakagami-m Distribution”, IEEE
Transactions on Vehicular Technology, Vol. 48, No. 3, pp. 790-794, May 1999.
[5] J. Razavilar, K. J. R. Liu, S. I. Marcus, “Jointly Optimized Bit-rate/Delay Control Policy for Wireless Packet Networks with
Fading Channels”, IEEE Transactions on Communications, Vol. 50, No. 3, pp. 484-494, March 2002.
[6] T. Rappaport, “Wireless Communications: Principles and Practice.”, Upper Saddle River, NJ: Prince-Hall Inc., 1996.
[7] G. L. Stuber, “Principles of Mobile Communication.”, Norwel, MA: Kluwer Academic, 2nd Edition, 2001.
[8] L. Kleinrock, “Queueing Systems. Volume I: Theory,” Wiley, 1975.
[9] IEEE Standard 802.16 Working Group, “IEEE Standard for Local and Metropolitan Area Networks part 16: Air Interface for
Fixed Broadband Wireless Access Systems,” IEEE, 2004.

11

S-ar putea să vă placă și