A Survey of Cross-Layer Designs in Wireless Networks

IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION 1
A Survey of Cross-Layer Designs in
Wireless Networks
Bo Fu, Yang Xiao, Senior Member, IEEE, Hongmei (Julia) Deng, and Hui Zeng, Member, IEEE
Abstract—The strict boundary of the five layers in the TCP/IP
network model provides the information encapsulation that
enables the standardizing of network communications and makes
the implementation of networks convenient in terms of abstract
layers. However, the encapsulation results in some side effects,
including compromise of QoS, latency, extra overload, etc.
Therefore, to mitigate the side effect of the encapsulation between
the abstract layers in the TCP/IP model, a number of crosslayer
designs have been proposed. Cross-layer designs allow
information sharing among all of the five layers in order to
improve the wireless network functionality, including security,
QoS, and mobility. In this article, we classify cross-layer designs
by two ways. On the one hand, by how to share information
among the five layers, cross-layer designs can be classified into
two categories: non-manager method and manager method. On
the other hand, by the organization of the network, crosslayer
designs can be classified into two categories: centralized
method and distributed method. Furthermore, we summarize
the challenges of the cross-layer designs, including coexistence,
signaling, the lack of a universal cross-layer design, and the
destruction of the layered architecture.
Index Terms—Cross-layer design, wireless networks, QoS,
security.
I. INTRODUCTION
THE END-TO-END connection in TCP/IP networks is
established through the collaboration of all of the five
layers (application layer, transport layer, network layer, data
link layer, and physical layer), which are designed to maintain
a limited interface between two neighbor layers [1]-[5], [10]-
[20]. The layers can be organized as a top-down or bottom-up
architecture. In either way, the exchange of data and service
calling takes place only between two adjacent layers and forms
a significant black box feature of the TCP/IP model [22].
The black box characteristic in TCP/IP networks leads to
the abstraction of the internal details of each layer. This is
also called information hiding. However, the abstraction of
the internal details may cause side effects in the networks
[22]. Although the strict boundary between the layers makes
the network easy to be deployed, the encapsulation of the
layers prevents some necessary information sharing between
layers. For example, in TCP/IP wireless networks, the layer
abstraction hides the root cause of the connection termination,
and the status about the connection termination is not utilized
for the repair or re-establishment of the connection [14]. If
the wireless channels are noise, the encapsulation in TCP/IP
Manuscript received October 27, 2012; revised April 3, 2013.
B. Fu and Y. Xiao are with Department of Computer Science, The
University of Alabama, Tuscaloosa, AL 35487-0290 USA (e-mail: yangxiao@
ieee.org).
H. (Julia) Deng and H. Zeng are with Intelligent Automation, Inc.,
Rockville, MD.
Digital Object Identifier 10.1109/SURV.2013.081313.00231
networks causes too many connection terminations even when
a connection is temporarily disconnected for a short time,
and the reestablishment of the connection costs a lot of time
because of the reestablishment of the links in all of the
five layers through all the nodes in the link route [44]. To
solve such a problem, cross-layer designs which consider the
interrelationship and the reconfiguration capability of the five
layers are needed to identify and improve the weakest link
in the link route before the connection fails. Furthermore,
an individual TCP/IP protocol normally aims at solving one
specific set of problems without considerations of the endto-
end performance, and therefore the deployment of these
protocols does not always satisfy the increased performance
requirements [22]. A lot of works have confirmed the poor performance
of current TCP/IP implementations without crosslayer
designs in wireless networks [33], [36], [37], [39], [40],
[42], [47], [48].
Therefore, to mitigate the side effect of the encapsulation
between the abstract layers in the TCP/IP model, a number
of cross-layer designs have been proposed. Normally,
any attempt to violate the black box characteristic in the
TCP/IP model is considered as a cross-layer design [22].
Attempting to break the virtually strict boundaries among
the five layers in the current TCP/IP model, the cross-layer
design is an escape from the waterfall-like concept of current
TCP/IP wireless networks. Numerous research efforts have
been presented to provide the solutions to achieve cross-layer
designs in wireless networks. Without cross-layer designs,
only two adjacent layers can achieve the service calling and
the data exchange. Cross-layer designs do not destroy the five
layer structure of TCP/IP networks, but provide the inter-layer
communication between two non-adjacent layers. Moreover,
cross-layer designs may also allow the disclosure of internal
status and parameters that are kept by each layer but now
reveled to the other layers. Cross-layer designs may allow
information sharing among all of the five layers. Furthermore,
cross-layer designs may allow a layer to determine its behavior
based upon the data that it retrieves or receives from the other
layers. Therefore, cross-layer designs imply that each layer is
able to share parameters, status, and other information with
other four layers, without breaking the five layer structure of
computer networks.
Moreover, cross-layer designs allow information sharing
through the layer boundaries to enable the compensation
for the network performance and reliability, e.g., increasing
throughput, reducing latency, and minimizing bit error rate,
by control the input to another layer [22]. Cross-layer designs
are able to make the hidden information (e.g., channel state
information) in each layer visible to other layers [22], [44].
1553-877X/13/$31.00 c 2013 IEEE
2 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION
Many of the cross-layer designs propose alternated methods
in the network layer, the data link layer, or the physical layer
[6], [21], [22]. For example, the cross-layer design in [6], [54]
supports an optimization for delay-sensitive applications such
as the real-time video streaming. The frameworks in [6] and
[21] incorporate the alternative protocol stack to leverage the
flexibility offered by optimization of design parameters. The
cross layer design in [54] incorporates both the data link layer
and the physical layer. There are also many other cross layer
design schemes [24], [55]-[58].
In this paper, a survey on cross-layer designs in wireless
networks is presented. Security, quality of service (QoS), and
mobility are three issues that cross-layer designs consider, and
they can be viewed as three goals of cross-layer designs. To
achieve these goals, a cross-layer design may allow one layer
to exchange and share data with other layers, may allow one
node to exchange and share data with other nodes as well. The
sharing scheme inside one node may be: a) the non-manager
method, which allows one layer directly communicate with
other layers; b) the manager method, which introduces a
vertical plane as a public library of the cross-layer information.
The sharing scheme among the nodes in a network may be:
a) the centralized method, which uses a central node or tier
structure to control the cross-layer information sharing; b)
the distributed method, which organizes information sharing
without a central node. Therefore, we have two kinds of
classifications for cross-layer designs. Cross-layer designs can
be classified into two categories (i.e., the non-manager method
and manager method) by how to share the information among
the five layers [7], [22], [25], [44]. Meanwhile, cross-layer
designs can be classified into two other categories (i.e., the
centralized method and distributed method) by the organization
of the network [7], [22], [25], [44].
However, there are many disadvantages/challenges of crosslayer
designs that are inevitable due to the characteristics
of these designs, including aspects of coexistence, signaling,
overhead, and the lack of a universal cross-layer design.
Firstly, each cross-layer design has its specific cross-layer
communication manner, and thus the coexistence and signaling
are the two challenges that cross-layer designs have to deal
with. Secondly, it is inevitable to result in an extra overhead
when exchanging the cross-layer information in the crosslayer
designs. Thirdly, a universal cross-layer design that
is optimized for all the applications is unlikely existent,
since different applications have distinct requirements for
the cross-layer design. Finally, cross-layer designs destroy
the encapsulation of the layers so that they may turn the
well organized layered architecture to a flat and disordered
design. It becomes difficult to make a modification for one
layer without considering other layers in cross-layer designs.
Therefore, the destruction of the layered architecture might be
the fundamental disadvantage of cross-layer designs.
The aforementioned challenges are not caused by a specific
design, but are common for cross-layer designs. Therefore, we
summarize the challenges of cross-layer designs in this article
and present possible solutions for these challenges.
The organization of this article is as follows. The goals
of cross-layer designs are introduced in Section II. Two
classifications of cross-layer designs are presented in Section
Mobility
QoS
Security
Application Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
Fig. 1. The goals of cross-layer designs: security, QoS, and mobility [7],
[22], [25], [44]. A cross-layer design scheme normally aims at least one of
these three goals.
III. The details of these two classifications are introduced in
Section IV and V, respectively. The challenges of cross-layer
designs are presented in Section VI. Finally, we conclude this
paper in Section VII.
II. THE GOALS OF CROSS-LAYER DESIGNS IN WIRELESS
NETWORKS
The goals of cross-layer designs are modeled as a coordination
model that briefly describes the functionality that
cross-layer designs might support [22], [25]. As shown in
Fig. 1, the coordination model introduces three coordination
planes (including security plane, QoS plane, and mobility
plane) extending across the five TCP/IP protocol layers [22],
[25]. Each coordination plane encapsulates a series of original
designed protocols, revised protocols, or algorithms to support
the cross-layer design functionality and to solve a specific
problem in wireless networks [22]. The coordination model
shows three goals of cross-layer designs: security, QoS, and
mobility. A cross-layer design normally aims to achieve at
least one of the goals that the three coordination planes
represent.
A. Security
The security coordination plane encapsulates the protocols
about security issues across the five TCP/IP layers [22], [80],
[81]. Encryption methods, such as SSH and Wi-Fi protected
access, might be deployed in this plane in a cross-layer
design aiming at security communication [22]. The cross-layer
designs in the following papers contain security plane: [45]-
[48], [59].
A cross-layer design may deploy encryption methods for
security: at the application layer (e.g., SSH and SSL for
the end-to-end encryption), at the network layer (e.g., the
IPSec protocol for the end-to-end encryption), and at the data
FU et al.: A SURVEY OF CROSS-LAYER DESIGNS IN WIRELESS NETWORKS 3
link layer and the physical layer (e.g., IEEE 802.11 wireless
networks) [22]. The authors in [45] point out that the abstract
feature of TCP/IP networks is inadequate and inefficient for
the security assurance in wireless sensor networks (WSN) and
give an overview of the existing cross-layer designs in WSN
for the security purpose.
The importance of cross-layer designs for the security
mechanisms in multi-hop wireless networks is discussed in
[46]. The cross-layer design in [46] shares the parameters
in each layer to avoid multi-layer attacks. The performance
comparison shows that the cross-layer design in [46] results
in less routing overhead and much fewer acknowledgement
packets.
The disadvantages of current security schemes in Wireless
Metropolitan Area Network (WirelessMAN) are reviewed in
[47] and a sub-linear rekeying algorithm with perfect secrecy
is proposed to achieve security goals by using cross-layer
designs inWirelessMAN. The evaluation results show a 4∼7%
performance improvement using the algorithm in [47] than that
not using the cross-layer design, for both message counts and
total communications.
A Cross-Layer Design Network Security Management
(CLDNSM) is proposed in [48] to protect the system security
by aggregating system information from layers and using it
to obtain the optimal security settings. The numerical results
show that the CLDNSM server is self-adapted in the dynamic
environment to maintain and modify system resources. The
CLDNSM security system in the simulation also suffers from
severely system overload and is more appropriate for dynamic
environments than the traditional non-cross-layer model.
In [59], a cross-layer design incorporates a blind video
watermarking method and a middleman detection algorithm
to enhance the medium access control (MAC) layer and the
network layer in wireless networks.
B. QoS
The QoS coordination plane aims at improving the quality
of service in the wireless communication across the five layers
[22]. Due to the characteristic of the physical layer and the
data link layer in wireless networks, the upper layers need
to be aware of the information in the two lowest layers
in order to improve the QoS in certain circumstances [22].
This requirement of the information sharing between the two
lower layers and the three upper layers, however, is not
supported in the current waterfall-like concept of the wireless
network model [22]. The QoS coordination plane aims at
achieve the cross-layer communication in order to improve
QoS. The cross-layer designs in the following papers contain
QoS coordination plane: [2], [3], [25], [26], [28], [30], [33],
[35], [36], [37], [42], [43], [54].
There are plenty of factors affect the QoS in wireless
networks [66], [67], [68], [69], [70], [71], [72], [73], [74],
[75], [76], [77], [78], [79]. For example, the transmission error
in wireless networks is one of the problems that we may solve
by cross-layer designs in order to improve QoS. Transmission
errors, e.g., package loss, in wireless network when using TCP
as the transportation layer protocol are mainly caused by the
bad performance of the MAC layer and the physical layer [22],
[37]. The problems related with transmission errors existing
in cross-layer designs in wireless networks are summarized
in [34]. To reduce transmission errors, the authors in [36]
present a protocol based on Explicit Loss Notification (ELN)
in a wireless network which uses TCP as the transportation
layer protocol. In [36], ELN mechanism is designed to notify
the packet sender the reason for the transmission error of
a packet. This mechanism specially informs the sender that
the transmission error caused by the reasons unrelated to the
network congestion when the network is not congested but
transmission error occurs, so that the sender can schedule its
retransmissions without considering the congestion control. If
the receiver is aware that the transmission error is not due
to the congestion, it sets the ELN bit in the TCP header
and propagates it to the source [36]. In this way, the ELN
mechanism shares the information in between the TCP and
the MAC layers in the sender with the receiver. The evaluation
results show that under certain error rates the performance
improvement by ELN is a factor of 2 and the relative performance
by the ELN is about 100% better than not using the
cross-layer design.
Many applications in wireless networks use TCP as the
transport layer protocol, and TCP is sensitive to the problems
in the lower layers listed in the last paragraph. Channel fading
may cause a route rescheduling. A transmission delay and
high bit error rate may trigger a retransmission in a TCP
connection. In addition, too many retransmissions may cause
the congestion in a busy wireless network. Therefore, a lot of
cross-layer designs improve the QoS by solving the problems
specifically in the wireless link, e.g., channel fading, channel
interference, bit error rate, transmission delay, etc [22].
The authors in [33] propose a forward error correction
mechanism to share with other layers the transmission errors
occurred in the MAC layer and the physical layer. This
mechanism uses forward error correction along with ELN,
in order to improve TCP performance for wireless cellular
networks [33]. The forward error correction scheme is a
solution to reduce the transmission error acknowledged by
the upper layers, and the forward error correction cooperating
with the explicit loss notification mechanism improves TCP
performance by 10% in [33].
Similar as the scheme in [33], another revision of existing
mechanisms is proposed in [35], called hybrid automatic repeat
request, which is a cooperation of forward error correction
mechanism and automatic repeat request. The authors in [35]
optimize the mapping between the signal-to-interference-andnoise
radio and the modulation-and-coding scheme in order to
improve QoS.
An adaptive coded modulation to solve the channel interference
problem is proposed in [42]. The authors in [37]
point that the forward link is the bottleneck of a wireless
network, and one of the reasons why this bottleneck exits
is the interference. The adaptive coded modulation in [42] is
used to solve the flat-fading channel problem. The authors in
[42] present an adaptive modulation technique and the general
principles of combining coset codes, and then apply their
method to a spectrally efficient adaptive Mary Quadrature Amplitude
Modulation (MQAM) technique to obtain the trelliscoded
adaptive MQAM. The simulation results show that the
4 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION
Higher
Layers
Data Link
Layers
Physical
Layers
Higher
Layers
Data Link
Layers
Physical
Layers
Adaptive Modulation and Coding (AMC)
(frame as unit)
ARQ
(packet as unit)
Fig. 2. A cross-layer design that aims at maximizing spectral efficiency
consists of AMC in the physical layer and ARQ in the data link layer [39].
adaptive coded modulation solves the channel interference
problem and comes close to the Shannon capacity limit of
fading channels [42].
Moreover, increasing the capacity of served users is another
way to improve QoS. The capacity of served users is decided
by the spectral efficiency [22]. In order to maximize the
spectral efficiency, the authors in [39] propose a cross-layer
design making use of the adaptive modulation and coding at
the physical layer with a truncated Automatic Repeat reQuest
(ARQ) protocol at the data link layer.
To increase the throughput in wireless networks which
are in time-varying channel conditions, Adaptive Modulation
and Coding (AMC) at the physical layer have been studied
in [39]. The ARQ protocol at the data link layer is used
to mitigate channel fading [35], [39]. As shown in Fig. 2,
the cross-layer design in [39] consists of the AMC in the
physical layer and the ARQ in the data link layer. At the
physical layer, the AMC selector determines the modulationcoding
pair; the AMC controller then updates the transmission
mode at the transmitter; and the coherent demodulation and
maximum-likelihood decoding are used at the receiver [39].
At the physical layer, a selective repeat ARQ protocol is
implemented, and the ARQ controller arranges retransmission
of the packets. The numerical results show a 3 Mb/s increase in
the transmission rate when using AMC with truncated ARQ
[39]. Moreover, the numerical results also imply that AMC
reduces the average packet error rate at the physical layer and
offers a much higher spectral efficiency than not using the
cross-layer design [39].
Similar as [39], the cross-layer design in [40] makes use
of both Hybrid Automatic Repeat Request (HARQ) at the
data link layer and AMC at the physical layer. HARQ is a
combination of ARQ and high-rate forward error-correcting
coding.
Fig. 3 shows the cross-layer structure using AMC in the
physical layer and HARQ in the data link layer in [40]. The
work in [40] is based on IEEE 802.11/16 standards. In the
Higher Layer
Mode Selection
Data Link Layer
Higher Layer
HARQ
Controller
HARQ
Controller
Transmitter
Mod-Coding&
Mode Controller
Receiver
Demod-Decoding&
Mode Selector
Physical Layer
Fading Channel
Channel
Estimator
Feedback Channel
Retransmission Request
Fig. 3. A cross-layer design using AMC and HARQ [40].
data link layer of the sender, the error-correcting codes of
HARQ code the packets from the higher layer. After AMC
controller in the physical layer of the sender receives these
packets, they are processed with the feedback from the AMC
mode selector of the receiver. The AMC selector at the receiver
determines the updated mode to send feedback to the sender.
Then the feedback packets are decoded by the error-correcting
codes of HARQ in the receiver and sent back to the sender.
HARQ controller at the sender arranges the retransmission
of the requested packet that is stored in the buffer [40]. The
numerical results show that a small retransmission number can
obtain the achievable spectral efficiency with less delay and
buffer-size penalties [40].
C. Mobility
The mobility coordination plane aims at guaranteeing the
uninterrupted communication in wireless networks [22], [82]-
[95]. Node movements are common in ad-hoc networks, so
that the events caused by the node movement, e.g., channel
switch and route change, are necessary to be discovered and
solved to assure the communication not to be uninterrupted.
Two handover categories are presented in [22] to describe the
types of node movements: horizontal handover and vertical
handover. The former describes the movement of a node
between access points (APs) of the same wireless access
technology; the latter describes the movement of a node
between APs of different wireless access technologies [23]. In
both categories, the upper layers in a cross-layer design need to
be aware of the events, e.g., channel switch and route change,
taking place in the lower layers, so that the communication
maintained by the upper layers will not be uninterrupted [23].
Channel fading, transmission delay, high bit error rate, and
other failures that decrease QoS may affect the mobility as
well. Therefore, some of the cross-layer designs in the last
FU et al.: A SURVEY OF CROSS-LAYER DESIGNS IN WIRELESS NETWORKS 5
TABLE I
THE GOALS OF CROSS-LAYER DESIGNS
Goal Explanation Examples
Security Security issues across the five TCP/IP layers are
considered in some cross-layer designs. Encryption
methods, such as SSH, Wi-Fi protected access, might
be deployed in a cross-layer design aiming at security
communication.
An efficient sub-linear rekeying algorithm with perfect secrecy achieves security
goals by using cross-layer design in WirelessMAN [47].
A cross-layer design network security management protects system security by
gathering system information from layers and then using it to obtain optimal
security settings [48].
QoS To improve the QoS in the wireless communication
across the five layers, some cross-layer designs
enable the cross-layer communication between the
upper layers (the application layer and the transport
layer) and the lower layers (the physical layer and
the data link layer) [22].
Some cross-layer designs aim at reducing transmission errors that are mainly caused
by the bad performance of the MAC layer and the physical layer in wireless network
when using TCP as the transportation layer protocol [22], [32].
A cross-layer design based on explicit loss notification in [36] increases TCP
performance by information sharing of the lower layers.
Forward error correction mechanism in [33] and hybrid automatic repeat request in
[35] also share the transmission errors occurred in the MAC layer and the physical
layer with other layers.
Mobility Some cross-layer designs aim at guaranteeing the
uninterrupted communication in a wireless network,
since node movement, which would cause channel
switch, route change, and other problems, is common
in wireless networks.
TDMA and FDMA are used for increased the number of served users. Some crosslayer
designs use CDMA/HDR to solve the time slot waste in wireless networks.
sub-section have the goal of mobility as well. The cross-layer
designs in the following papers contain mobility coordination
plane: [25], [34], [37], [42]-[44].
For example, ratio of served users is considered as one
of mobility problems in cross-layer designs. In centralized
networks, e.g., 802.11 Wi-Fi and cellular networks, the number
of served users is limited due to the limited number of
channels, channel interference, etc. Time-Division Multiple
Access (TDMA) and Frequency-Division Multiple Access
(FDMA) methods are developed to solve this problem [22].
It is obvious that TDMA causes a waste of time slots and
the bandwidth even there is no data transmission between the
base station and the mobile station due to the characteristic
of TDMA [22], [34]. Cross-layer designs help to deal with
the time waste problem due to the benefits of the crosslayer
information sharing. The authors in [37] use Code
Division Multiple Access/ High Data Rate (CDMA/HDR)
to build a bandwidth-efficient wireless data service to solve
the aforementioned problem. The main solutions in [37] are
the channel measurement, channel control, and interference
suppression and mitigation. The authors of both [34] and [37]
use the information sharing in cross-layer designs to avoid
wasting of time slots. The estimated maximum achievable
throughput is evaluated in [37] in which a graph of a cumulative
distribution function is present to show an additional
2 dB of margin to account for various losses and to show an
increased throughput.
These three coordination planes (security, QoS, and mobility)
describe the three goals of cross-layer designs. One
cross-layer design scheme normally aims at least one of these
three goals. Table I summarizes the aforementioned crosslayer
designs and their goals.
III. TWO CROSS-LAYER DESIGN CLASSIFICATION
Cross-layer designs allow the information sharing between
any two of the five layers in the TCP/IP model, and may allow
a layer determines its activities based on the information that
it retrieves or receives from the other four layers. Therefore,
cross-layer designs allow each layer to be able to share
its information, including parameters and status with other
four layers, without breaking the five layer structure of the
TCP/IP model. Some cross-layer designs even allow the crosslayer
information sharing between different nodes in wireless
networks.
But how these cross-layer designs achieve the three goals
in the last section? For example, to improve the QoS in
wireless networks, the QoS coordination plane is introduced
as shown in Fig. 1, but a series of questions need to be
answered: Does the cross-layer design need to be deployed
in all the nodes or some of the nodes? Does the crosslayer
design need to be deployed in the protocols in all
of the five layers or some of the layers? Does the crosslayer
design need to just revise the current protocols or build
a totally new architecture? Is a centralized node necessary
for QoS in a cross-layer design? To answer these questions,
we summarize the existing cross-layer designs and present
two kinds of classification for cross-layer designs. The first
classification is that the cross-layer designs can be classified
into two categories by how to share information among the five
layers in one node: the non-manager method and the manager
method. Meanwhile, cross-layer designs can also be classified
into other two categories by the organization of the network for
cross-layer information sharing: the centralized method and
the distributed method. The non-manager method and manager
method are used for sharing cross-layer information in one
node, and the centralized method and the distributed method
are used for sharing cross-layer information among nodes in
a network.
Figs. 4(a) and (b) show the classification based on how to
share the information among the five layers in one node: the
non-manager method and the manager method.
Fig. 4(a) shows the non-manager method, which allows
direct communication between any pair of layers in the TCP/IP
protocol stack [22], [25]. This method does not change the
five layers structure of the TCP/IP model, but changes the
function of the protocols in certain layers by allowing the
direct communication between two layers [22].
Fig. 4(b) shows the manager method, which introduces a
vertical plane as a manager that share data with some (or all)
of the layers in the TCP/IP protocol stack [7], [22], [25]. This
6 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION
Application Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
Vertical Plane
Application Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
(a) Non-manager method
(b) Manager method
Fig. 4. The first classification of cross-layer designs is (a) the non-manager method and (b) the manager method, which are classified by how to share
information among the five layers in one node [7], [22], [25], [44]. The data exchange takes place directly between any two layers in the non-manager method;
there is a vertical plane in the manager method to manage data exchange between the layers.
method does not change the five layers structure of the TCP/IP
model, but changes the function of the protocols in the layers
by allowing the data sharing with the vertical planes [7], [22],
[25].
On the other hand, the cross-layer designs can also be
classified the centralized method and the distributed method
for sharing the cross-layer information among the nodes in a
wireless network.
Fig. 5(a) shows the centralized method, which introduces
a centralized node (sometimes using a base station) or tiers
which are in a hierarchical manner [12], [22]. The centralized
node or tiers are introduced to manage the information sharing
of the five TCP/IP layers between two nodes [12], [22]. The
centralized method is typically used in cellular networks.
Fig. 5(b) shows the distributed method, which does not
include any centralized node or any base station for the crosslayer
information sharing. Since there is no centralized in this
case, the multi-hop path from a node to another is possible
during the cross-layer information sharing [22], [31]. The
distributed method is typically used in ad-hoc networks.
Since the above two kinds of classification in Fig. 4 and
Fig. 5 differentiate cross-layer designs by two standards, they
are independent from each other. Therefore, it is possible
that a cross-layer design belongs to one of the non-manager
method and the manager method, and meanwhile it belongs
to one of the centralized method and the distributed method
as well. We introduce these two classifications in Section
IV and Section V, respectively. Table II summarizes the
aforementioned categories of the cross-layer designs.
IV. CLASSIFICATION OF CROSS-LAYER DESIGNS BY HOW
TO SHARE INFORMATION AMONG FIVE LAYERS IN ONE
NODE
The first kind of the classification of cross-layer designs
composes of the non-manager method and the manager
method, which are classified by how to share the information
among the five layers in one node.
A. Non-manager Method
Fig. 4(a) shows the non-manager method, which allows the
direct communication between any pair of layers in the TCP/IP
protocol stack. This method does not change the five layers
structure of the TCP/IP model, but changes the function of the
protocols in some layers by allowing the direct communication
between two layers [22].
For example, the authors in [30] propose a cross-layer
design framework for 802.16e orthogonal frequency-division
multiple access systems for the performance improvement by
a cross-layer adaptation framework, and they present a design
example of primitives for cross-layer operation between the
MAC and PHY layers as well. This framework is composed
of a user grouper, a MAC scheduler, and a resource controller
[30]. The user grouper classifies the users, the MAC scheduler
determines the scheduling of users and how to schedule packets
in the current frame, and the MAC scheduler handles the
mechanisms for appropriate data transport according to QoS of
each data transport class [30]. The resource controller assigns
frequency bands to each selected user (by applying subcarrier
the allocation algorithm) after the scheduler determines the
FU et al.: A SURVEY OF CROSS-LAYER DESIGNS IN WIRELESS NETWORKS 7
TABLE II
THE CATEGORIES OF CROSS-LAYER DESIGNS
Classification Method Explanation Examples
The first
classification:
by how
to share
information
among layers
in one node
Non-manager
method
The data exchange takes place
directly between any two
layers in the non-manager
method.
A cross-layer architecture in [2] improves the TCP performance by crosslayer
communication directly between the TCP layer and the lower layers,
and therefore it is a non-manager method cross-layer design.
A cross-layer design framework for 802.16e orthogonal frequency-division
multiple access systems in [30] provides performance improvement by crosslayer
communication between layers. There is not a vertical plane that
manages the cross-layer communication.
Manager
method
There is a vertical to manage
data exchange between the layers.
A cross-layer design architecture called ECLAIR in [27] is composed of
tuning layers and optimizing subsystems to achieve cross-layer communication.
ECLAIR functions as the vertical plane that manages the cross-layer
communication.
A cross-layer optimization strategy in [26] uses a cross-layer optimizer as
the vertical plane to improve the performance of video streaming in wireless
networks.
A proactive and adaptive cross-layer reconfiguration scheme in [44] uses
an adaptation interaction interface as the vertical plane in order to achieve
reliable communication in tactical networks.
The second
classification:
by network
organization
for cross-layer
information
sharing
Centralized
method
The centralized method uses a
centralized node or tier which
is in a hierarchical manner
to achieve communication between
nodes. The centralized
method is typically used in cellular
networks.
A scheduling mechanism in [28] achieves cross-layer information sharing in
the Universal Mobile Telecommunications System (UMTS) in which a base
station can be considered as the centralized node in cellular networks.
A cross-layer design in [29] is designed for real-time video applications in
3G the cellular network which has a centralized structure.
M@ANGEL, an autonomic management mobile platform, is composed of
two tiers in a hierarchical manner in [12], and therefore it is considered as a
centralized cross-layer design.
Distributed
method
The distributed method does
not use any centralized node
or tier. The distributed method
is typically used in ad-hoc networks.
A scheme called farcoopt in [1] is a cross-layer design that makes use of
the farthest neighbor to increase QoS, to reduce energy consumption, and to
increase throughput in multi-hop networks. This scheme does not use any
centralized node.
A cross-layer approach in [21] seeks to improve end-to-end performance in
ad-hoc networks without using any centralized node.
A proactive and adaptive cross-layer reconfiguration scheme in [44] is
designed to achieve reliable communication in ad-hoc networks in which
there is no centralized node.
scheduled users [30]. The above procedure is executed in order
to maximize the throughput [30].
The framework in [30] is designed solely for 802.16e orthogonal
frequency-division multiple access systems. It makes
use of the MAC layer and the physical layer by implementing
the cross-layer design. The disadvantage of this framework is
that it does not fully make use of the other three layers. This
framework is not easy to be implemented on other types of
wireless networks.
In a number of cross-layer design methods, Cognitive
Network (CN) is developed as a new type of data networks
which are expected to solve some problems that networks are
facing with [4], [9]. According to [2], [9], [60]-[65], the CN
network is a network with a cognitive process that perceives
the current network conditions, and then plans and acts on
those conditions. To take into account end-to-end goals, the
CN network learns from current adaptations and uses them
to make future decisions [9]. The authors in [2] present a
cross-layer design in CN networks with primary users and
secondary users to maximize the TCP throughput. The authors
in [2] consider spectrum sensing, access decision, physicallayer
modulation, coding scheme, and data-link layer frame
size in CN networks. Secondary users, also called unlicensed
users, can operate in the licensed spectrum bands, but they
are considered lower priority to avoid interference to primary
users in their vicinity [2]. In CN networks, primary users and
secondary users share a block of a spectrum consisting of
several radio channels [2].
Fig. 6 shows the cross-layer architecture in [2] that aims to
increase TCP performance of secondary users. The secondary
users (cognitive sensors) sense the channel to observe the
channel and obtain sense outcomes that are sent from the
physical layer to the TCP layer, as shown in Fig. 6 [2]. Based
on these sensing outcomes, the TCP layer determines the
frame size that is sent to the data link layer, access decision
that is sent to the MAC layer, and modulation and coding
scheme that is sent to the physical layer [2]. The authors
in [2] also present a TCP throughput model, which gives
the calculation of some variables, such as TCP throughput,
bit error rate, roundtrip time, etc. Simulation results in [2]
show that average TCP throughput and spectrum utilization
in this architecture are higher than the traditional schemes. In
the scheme in [2], the TCP layer directly communicates with
other layers, and therefore it is considered as the non-manager
method of cross-layer designs.
The advantage of this scheme is that it uses the TCP layer
as the pivot of the cross-layer information sharing without an
independent vertical plane, and, therefore, this architecture is
concise but enough to handle the secondary users. But it might
be difficult if we try to extend the functionality of this scheme,
because the TCP layer is not able to handle all the behaviors
of other layers.
B. Manager Method
Fig. 4(b) shows the manager method, which introduces one
or more vertical planes that share the data with some (or all) of
8 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION
(a) Centralized method
(b) Distributed method
Fig. 5. The second classification of cross-layer designs is by the organization
of the network for cross-layer information sharing [22], [44]: (a) the
centralized method and (b) the distributed method. The centralized method
uses a centralized node or tier which is in a hierarchical manner to achieve
communication between nodes; the distributed method does not use any
centralized node or tier.
the layers in the TCP/IP model. This method does not change
the five layers structure of the TCP/IP model, but changes
the function of the protocols in the layers by allowing the
data sharing with the vertical planes. The main difference
between the non-manager method and the manager method
is: the former allows direct communications between any two
layers; the latter requires a vertical plane for communications.
For example, the proactive and adaptive cross-layer reconfiguration
scheme in [44] introduces an adaptation interaction
interface which functions as the vertical plane to manage the
cross-layer communication.
A manager cross-layer method is proposed in [25] for the
purpose of solving the problems in the performance of wireless
links and mobile terminals, such as the high error rate of
wireless networks, power saving requirements, unpredictable
QoS in an increasingly dynamic network environment. The
authors in [25] also propose a cross-layer design framework
for 4G networks and summarize the known problems associated
with current strictly layered protocol architecture. A
cross-layer manager is introduced in [25], as shown in Fig.
7. Each layer shares its events with the cross-layer manager
TCP Layer
Data Link Layer
Frame
size
MAC Layer
Physical Layer
Sensing
Outcomes
(including
channel
state)
Access
Secision
Modulation
And Coding
Scheme
Fig. 6. An intra-layer cross-layer design in cognitive networks with primary
users and secondary users to maximize the TCP throughput [2]. By using
cross-layer design, the TCP layer determines the frame size that is sent to the
data link layer, access decision that is sent to the MAC layer, and modulation
and coding scheme that is sent to the physical layer.
which can be considered as the vertical plane in Fig. 4(b),
and the cross-layer manager shares all the state variables in
the layers in TCP/IP and cellular networks.
The authors in [26] proposes a cross-layer optimization
strategy that jointly optimizes the application layer, the data
link layer, and the physical layer of the TCP/IP protocol
stack using an application oriented objective function in order
to maximize the user satisfaction. Aiming at a cross-layer
design for video streaming in wireless networks, the authors
in [26] focus on application-driven optimization. Moreover,
they observe the trade-off between performance and the additional
computation and overhead introduced by its cross-layer
optimization [26]. As shows in Fig. 8, a cross-layer optimizer
is introduced to optimize the parameters shared by the layers.
The cross-layer optimizer jointly optimizes multiple network
layers, makes predictions on their states, and selects optimal
values for their parameters [26]. There are three steps in this
cross-layer working process: 1) Layer abstraction computes an
abstraction of the parameters in each layer [26]. The purpose
of this step is to avoid transmitting too many parameters to
the cross-layer optimizer [26]. The parameters being abstract
include: Source rate, encoding format, compression, FEC,
TDMA time slots, OFDM carriers, directional beams, bit error
rate, frame rate, picture size, net transmission rate, modulation
scheme, channel coding, etc [26]. 2) Optimization reconfigures
the parameters to optimize a specific objective function [26].
3) Layer reconfiguration distributes the reconfigured parameters
to the corresponding layers and requires the layers to
execute their actual operation [26]. Since the authors in [26]
aim at cross-layer designs for video streaming, they present
the detail on how to revise the value of the above parameters
in the video streaming. After layer abstraction, the cross-layer
FU et al.: A SURVEY OF CROSS-LAYER DESIGNS IN WIRELESS NETWORKS 9
TCP/RTP
events
MIPv4 MIPv6
CDMA 802.11 Blue tooth
Cross-Layer
manager
State variables
events
State variables
events
State variables
Fig. 7. An manager cross-layer design scheme for improving the performance of wireless links and mobile terminals [25].
Optimized parameter setting
Cross-Layer
optimizer
Abstraction A
Layer N
D, DT, IT
Optimized parameter setting
Abstraction A
Layer 2
D, DT, IT
Optimized parameter setting
Abstraction A
Layer 1
D, DT, IT
Fig. 8. A cross-layer design that uses a cross-layer optimizer for improving the performance of video streaming [26].
optimizer revises the value of the parameters, and then sends
them back to the five layers. In this way, the five layers behave
under the decisions of the cross-layer optimizer.
One of the advantages of the scheme in [26] is that it
uses the layer abstraction to reduce the work load of the
vertical plane. This scheme improves the video quality by
allocating network resources in its cross-layer design as well.
The disadvantage of this scheme is that it is designed solely
for the video streaming, and it is not easy to expend to other
applications. This is because the parameter revising for one
application may not be suitable for another application. For
example, the video streaming requires low TCP retry value,
but the text transmission can endure long TCP retry limit in the
bad network connection resulted by the channel interference.
The authors in [27] present a cross-layer design architecture
called ECLAIR which is composed of interfaces (called
tuning layers) and optimization algorithms (called optimizing
subsystems) in each layer, as shown in Fig. 9. The tuning layer
provides an interface to protocol data-structures that determine
the protocols behavior [27]. In the TCP/IP model, a protocols
behavior is determined by its control data-structures and the
protocol implementation typically has data-structures for the
control and data [27]. As the tuning layer is the interface to
protocol data-structures, it reads and updates the protocol datastructure
[27]. The authors in [27] give an example: in Linux,
TCP control information is stored in a data structure called
tcp opt, and ECLAIR is able to read and update this data in
order to control TCP behavior. Therefore, the communication
and data exchange between two adjacent layers is conducted
by the tuning layer [27].
Moreover, the optimizing subsystem in each layer contains
the algorithms and data structures for cross-layer optimizations
in that layer [27]. After the optimizing subsystem receives the
control and data information through its tuning layer from
another layer, the optimizing algorithms in the optimizing
subsystem starts to optimize and adapt with the protocols
behavior [22], [27].
The advantage of ECLAIR is that it does not destroy the
five layer structure of the TCP/IP model and each layer has
its tuning layer and optimizing subsystem to conduct crosslayer
functions. This feature makes the structure of ECALIR
clear and easy to be extended. The disadvantage of ECLAIR
is that each layer communicates and exchanges data only with
its adjacent layers, but not all the other layers. The behavior
of a layer is determined by its optimizing subsystem which is
only aware of its (at most) two adjacent layers. In this case,
the application layer is not able to determine its behavior by
the network layer, which might be crucial for the behavior of
the application layer.
The non-manager method allows direct communications between
any two layers; the manager method requires a vertical
plane for the information sharing. Both of them achieve the
cross-layer information sharing. The non-manager method has
to affect the waterfall-like structure of the five layers, since
the non-manager method is able to make any two layers
10 IEEE COMMUNICATIONS SURVEYS & TUTORIALS, ACCEPTED FOR PUBLICATION
Protocol
Optimizer−1
Protocol
Optimizer−2
Protocol
Optimizer−3
Protocol
Optimizer−n

Comments are closed.