Sensors 2017, 17, 1637 2 of 14
Therefore, several cross-layer schemes are proposed to alleviate congestion in the wireless
channel [
5
–
8
]. For example, a cross-layer ECN (Explicit Congestion Notification) scheme was proposed
to perceive the link congestion according to the retransmission counter at the MAC layer. In this
scheme, the retransmission counter at the MAC layer is used as the congestion metric to trigger the
ECN mechanism. ECN is the explicit congestion notification mechanism of the IP layer. When network
congestion occurs, the sender can adjust the congestion window and reduce the sending rate by the ECN
mark from the receiver. Differently from [
9
], Yue Peng et al. utilized the frame service delay at the MAC
layer as the congestion metric to mark or drop packets [
10
–
12
]. WCCP (Wireless Congestion Control
Protocol) was proposed to dynamically adjust the congestion window of TCP according to the channel
business ratio, which presents the extent of channel congestion in a more accurate way [
13
]. These
cross-layer methods improve the efficiency and fairness of wireless ad hoc networks [
14
,
15
]. However,
the retransmission counter and the frame service delay are not precise enough to detect congestion
from the MAC layer, while WCCP is too complex to implement in existing transport protocols.
Therefore, this paper proposes an MAC congestion metric called frame transmission efficiency,
describing channel congestion in a more precise manner. Furthermore, combining this congestion
metric with ECN, we propose a cross-layer ECN mechanism, RECN (ECN and the ratio of successful
transmission delay to the frame service delay in the MAC layer, namely, the frame transmission
efficiency), which is based on frame transmission efficiency and adjusts the transport sending rate
through a standard ECN signaling method, without any modification to transport protocols.
The remainder of this paper is structured as follows. Section 2 describes several congestion
metrics at the MAC layer and proposes a congestion notification scheme RECN. Section 3 gives
the definition and analysis of the frame transmission efficiency to demonstrate the relationship
between the frame transmission efficiency and the link layer congestion. Section 4 gives the protocol
design of RECN. In Section 5, the simulation results show that our proposed congestion notification
mechanism significantly outperforms traditional TCP in wireless multi-hop ad hoc networks. Section 6
is our conclusion.
2. Wireless Channel Congestion Detection
How to detect wireless channel congestion is important for congestion control design in wireless
networks. Recent works have proposed several congestion metrics at the MAC layer.
(1)
Retransmission counter: There exists a correlation between the number of RTS (Request to Send)
retransmissions and channel congestion extent at the MAC layer [
16
,
17
]. In [
4
], when the number
of RTS retransmissions is greater than two, the wireless channel will be deemed as congested, and
the TCP sender will decrease its sending rate. Though this method is very easy to deploy, the
number of RTS retransmissions is not able to describe the congestion state accurately.
(2)
Frame Service Delay (FSD): The frame service delay is the interval from the time that the MAC
layer begins to sense the channel for data transmission to the time that the acknowledgment is
received successfully, which includes collision time and transmission time [
18
]. The greater the
service delay is, the higher the probability of network congestion is. However, the hops that
the flows are traveling through and the frame size greatly affect the service delay. Subsequently,
the threshold value of the service delay is difficult to determine, resulting in the inaccuracy of
congestion estimation.
(3) Channel business ratio: The channel business ratio is defined as the ratio of link layer busy time
(including collision time and data successful transmission time) to the total time [
19
–
22
]. It can be
shown in Equation (1).
R
b
= 1 −
p
i
σ
p
i
σ + p
s
T
suc
+ p
c
T
col
(1)
where
σ
is the length of the backoff time slot,
p
i
is the probability that the observed backoff time slot
is idle,
p
s
is the probability that there is one successful transmission,
p
c
is the collision probability