25-08-2017, 09:32 PM
Crowded Spectrum in Wireless Sensor Networks
Crowded Spectrum in Wireless Sensor Networks.pdf (Size: 330.95 KB / Downloads: 16)
Abstract
With the exciting progress of wireless sensor network
(WSN) research, we envision that in 5-10 years, the world
will be full of low power wireless sensor devices. Due to the independent
design and development, together with the unexpected
dynamics during deployment of co-existing networks and devices,
the limited frequency spectrum will be extremely crowded. Plus,
existing electric appliances like microwaves make the congestion
even worse. This paper proposes to develop new suites of WSN
protocols along three complementary dimensions: (1) to achieve
high communication throughput within a single WSN, (2) to
achieve multi-frequency functionality among overlapping but
cooperative WSNs and (3) to resolve the crowded spectrum issue
caused by any reason, such as random transmitting devices, other
nearby sensor networks, or even co-existing electric appliances.
I. INTRODUCTION
Wireless sensor network (WSN) is an exciting new technology
with application to environmental monitoring, agriculture,
medical care, smart buildings, factory monitoring and
automation, and numerous military applications. A WSN can
also be considered as the underlying infrastructure that will be
an integral part of future ubiquitous and embedded computing
applications. We project that in 5-10 years, (i) many individual
WSNs will be very sophisticated and operating at high levels
of utilization, and (ii) there will exist many thousands, if
not millions, of sensor networks. When this latter situation
materializes, WSNs will overlap and co-exist. One significant
problem is that the majority of WSN research work today
is focused on single frequency systems. To deal with high
performance WSNs and with the projected situation of large
numbers of deployed sensor networks will require multifrequency
systems. In this paper, we present new suites of
protocols for multi-frequency WSNs along three complementary
dimensions. They are: (1) to achieve high performance
for both broadcast and unicast communications within a single
WSN, (2) to support overlapping, but cooperative WSNs, and
(3) to handle noise and the crowded spectrum caused by any
reason, such as random transmitting devices, other nearby
sensor networks, or even co-existing electric appliances.
By incorporating the collection of new solutions, we envision
excellent throughput performance for sophisticated, high
workload WSNs. The WSNs will also be robust to noise, the
crowded spectrum and even to certain degrees of jamming
attacks. We also anticipate a new ability to deploy multiple
overlapping and cooperative WSNs in different time frames.
These cooperative WSN systems will be able to seamlessly
interact to improve overall performance of many applications.
For example, in assisted living facilities the first deployed
WSN system may be a specialized WSN to monitor patients’s
indoors activities and improve their lifestyle and health. Later,
another specialized WSN (perhaps built and sold by a different
company) may be deployed to better monitor both indoors
and outdoors environmental conditions such as temperature,
air quality and hazards such as fire. It may be impractical to
shut down the previous system and create a single new system,
or to reload the older system with new software that results in
a single new integrated system. Having these multiple WSNs
co-exist and interact seamlessly is likely to be a necessary
feature in the future and can result in major additional benefits
to patients. Similar application examples can be described for
embedded systems in environmental and military domains.
The rest of the paper is organized as follows: Section II
presents a solution to achieve high communication performance
within a single WSN. Section III explains multifrequency
support for overlapping but cooperative WSNs.
Section IV analyzes how to handle the crowded spectrum issue
caused by any reason, including co-existing sensor network
devices as well as electric appliances. Finally, conclusions are
given in Section V.
II. ACHIEVING HIGH THROUGHPUT
Media access control (MAC) is an essential part of the
communication stack, and a number of MAC protocols [1] [2]
[3] [4] [5] [6] have been proposed in WSN context, to achieve
high throughput. While these designs demonstrate good performance
in single-channel scenarios, parallel transmission within
a vicinity through multiple channels is not considered, to further
improve the throughput. Since the current sensor devices
provide very limited single-channel bandwidth, 19.2Kbps in
MICA2 [7] and 250Kbps in MICAz [8] and Telos [9], it is
imperative to design multi-channel MACs that can achieve a
higher throughput through parallel communications. Plus, the
CC2420 radio [10] used in MICAz and Telos motes already
provides multiple physical channels, paving the way for multichannel
sensor network MAC designs.
When switching from WSN to general wireless ad hoc
networks, multi-channel MAC designs are not new and have
been well studied. However, due to the reasons discussed below,
these protocols are not appropriate for resource-restrained
sensor network applications. The first reason comes from
different hardware assumptions. A typical sensor device is
usually equipped with a single radio transceiver, which can
not conduct simultaneous transmission and reception, but can
work on different channels at different times. On the contrary,
many MAC protocols in general wireless ad hoc networks
assume more powerful radio hardware. For instance, protocols
[11] [12] are designed for frequency hopping spread spectrum
wireless cards, and protocol [13] assumes the busy-tone ability
for the hardware. Also, some protocols [14] [15] [16] [17]
require the hardware to be capable of carrier sensing on
multiple channels simultaneously. Second, WSNs have very
limited communication bandwidth and the MAC layer packet
size is very small, 3050 bytes, compared to the 512+ bytes
used in general wireless ad hoc networks. Due to such small
data packet sizes, the RTS/CTS control packets in IEEE 802.11
[18] no longer constitute a small overhead that can be ignored.
So protocols [19] [20] [21] that use RTS/CTS for frequency
negotiation, and protocols [22] [23] that are based on IEEE
802.11 are not suitable for WSN applications, even though
they perform well in general wireless ad hoc networks.
To further understand the cost that RTS/CTS control packets
incur in general wireless ad hoc networks versus WSNs, we
choose MMAC [19] as a case study. MMAC is a typical multichannel
MAC protocol proposed for general wireless ad hoc
networks. In MMAC, periodically transmitted beacons divide
time into beacon intervals, each of which starts with a small
ATIM window. During the ATIM window, nodes that have
packets for transmission negotiate frequencies with destination
nodes, using a default frequency. After the ATIM window,
nodes switch to the negotiated frequencies and use IEEE
802.11 for data communication, i.e., exchanging RTS/CTS
before sending out DATA packets. We implement MMAC in
GlomoSim [24] with the same experiment set up as in [19].
0
200
400
600
800
1000
1200
1400
512 256 128 64 32
Aggregate MAC Throughput (Kbps)
Packet Size (bytes)
Packet arrival rate per flow is 10 packets/sec
802.11
MMAC, Beacon Interval = 50ms
MMAC, Beacon Interval = 100ms
MMAC, Beacon Interval = 150ms
Fig. 1. Effect of Packet Size on MMAC
As demonstrated in the result (Figure 1), when the packet
size is large, the MMAC protocol with 3 frequencies and a
beacon interval of 100ms (the default configuration suggested
in [19]) impressively achieves almost twice the throughput of
IEEE 802.11. This result is consistent with that presented in
[19]. However, when the packet size decreases, both MMAC
and IEEE 802.11 obtain diminished performance. The reason
is that the overhead of RTS/CTS control becomes more
prominent when the data packet size is smaller. When the
packet size is as small as 32 bytes, IEEE 802.11 has even
a slightly higher throughput than MMAC. Also, Figure 1
demonstrates that while using a shorter beacon interval (50ms)
helps to some extent, MMAC with 3 frequencies still can not
even outperform IEEE 802.11 with a single frequency, when
the packet size is as small as 64 or 32 bytes. While more
detailed analysis can be found in [25], the main observation
we make here is that while MMAC is a good multi-frequency
MAC protocol for general wireless ad hoc networks where
data packets are usually large, it is not suitable for WSNs
where data packets are much smaller.
Since multi-channel MAC designs for general wireless ad
hoc networks are not adequate for WSNs, the key question
is: what are the essential design considerations for multichannel
MACs in WSNs to achieve higher throughput? In what
follows, we analyze two core aspects: frequency assignment
and media access design.
Frequency Assignment: Since RTS/CTS frequency negotiation
constitutes too high an overhead for bandwidth limited
and small packet size sensor networks, frequency assignment
stands out to be a more promising design choice. During frequency
assignment, neighboring nodes are allocated different
frequencies for unicast packet reception, for supporting of
parallel communication to achieve high throughput. A naive
frequency assignment design is to let each node overhear its
neighbors’ frequency choices, and then choose one of the least
used frequencies for its own data reception. A more sophisticated
design needs to consider the hidden terminal problems
[18], as well as the in-situ reality that the radio interference
range may be greater than the communication range [26]. In
[25], we present a collection of frequency assignment schemes
that demonstrate different merits in different application scenarios,
together with corresponding performance comparison.
Media Access Design: When nodes within a vicinity are
assigned different frequencies for unciast packet reception, the
question of broadcast support is raised. A simple choice is to
interpret a broadcast transmission as multiple unicast transmissions.
Since WSNs usually maintain high node densities
to trade for enhanced system lifetimes, parsing a broadcast as
multiple unicasts actually involves a very high communication
overhead and makes it a poor choice. A better design we
suggest is to assign a default broadcast channel for all nodes
to receive broadcast packets, while at the same time maintain
different channels for unicast packet reception.
There are two general schemes to integrate broadcast and
unicast communications, in such a multi-channel, but singletransceiver
context. First, periodic beacons can divide time
into fixed-length beacon intervals. During each interval, each
node can choose to send/receive a broadcast/unicast packet.
By assigning different priorities for broadcast and unicast
communications, together with carefully designed carrier sense
and backoff schemes, communication correctness can be guaranteed,
and also the throughput can be maximized. Interesting
readers can refer to [25] for details.
Second, without the presence of time synchronization, a
toggle snooping technique can be used instead to provide
efficient broadcast support from the root. The basic idea is
to let a receiver carrier sense on the broadcast and unicast
channels, in an alternating fashion. Whenever it overhears a
signal, it stops toggling between the broadcast and unicast
channels, and stays on the current channel to receive the
data packet. This can either be a broadcast or unicast packet.
Also, the transmitter needs to prepare a longer preamble than
normal, which can cover the time period when the receiver
stays on the other channel for carrier sensing. Due to the space
limits, the specific design details are not presented here.
III. CROSS-NETWORK COOPERATION
In the near future, we envision that multiple WSNs will be
deployed together within the same physical location, serving
different purposes. In this case, detecting co-existing networks
and conducting possible MAC layer cooperation among them
become very critical for reducing cross-network interference
and improving aggregated throughput. In this multi-network
context, the ultimate design goal is to assign different frequencies
to different nodes at different times, achieving the
maximum parallel transmission, at any time and for any
location. Three major properties should be provided by this
design:
Space-Dimension Flexibility: In a single-network case,
we use a “node density” concept, and its value varies from
location to location. In the multi-channel scenario, we introduce
another concept “network density”, which is defined as
the number of networks within a communication range. Its
value also varies from location to location. The frequency
assignment should be differentiated according to different
network densities and node densities.
Time-Dimension Flexibility: The application traffic pattern
varies from time to time, in both single-network and multinetwork
contexts. The environmental noise also varies from
time to time. Plus, new networks are introduced and old
networks fade out dynamically. All these dynamics raise the
need for dynamic frequency adjustment.
QoS Control: QoS control is desired when the available
physical bandwidth is not able to fully support all traffic from
all co-existing networks, for all locations and for all times.
So, a parameter is offered for users to set different priority
values for different networks. At any location at any time, each
network is assigned the bandwidth, according to the ratio its
value over the sum of all values whose networks co-exist
within that specific location. The ratio depends on the number
of competing networks, varying from location to location and
from time to time.
To achieve the forgoing three properties, there is a need
for both static frequency assignment and dynamic frequency
adjustment.
A. Static Frequency Assignment
Let neti, 1 i M, represent the M co-existing networks
in the environment, and let frei, 1 i N, denote the
N non-overlapping physical frequencies. Each network neti
is assigned a comparative priority i. Within any location
and at any time, each network netj competes with locally
co-existing networks, and is supposed to use
j percent of
available frequency spectrum:
j =
j
PK
1 i
(1)
where K is the number of available networks within that
location at that time, and 1 K M.
When a new network is deployed in a space, where existing
networks are running, each node in the existing networks
is called to temporarily switch through the following two
cooperation steps to get its frequency reassigned, and then
switch back to its normal operation.
Differentiated QoS Computation: Each node first conducts
neighbor discovery. Different networks are identified
with different group IDs, shortened as “gID”, which is a
standard field for all TinyOS [7] messages. During neighbor
discovery, each node beacons the following information: ID,
gID and . With this information, each node computes its local
QoS control parameter
, according to Formula 1.
Chained Frequency Decision: With neighbor information
collected, each node makes two decisions: 1) What portion of
frequency range to choose from and 2) What frequency to use.
The chained decisions proceed in the increasing order of gID.
When two nodes tie with gID, the node with the smaller node
ID wins. During the whole process, each node (node ) keeps
the following rules in mind:
1) Node checks all neighbors from which it has not heard
frequency decisions. If gID is the smallest one among
the neighborhood, and also no neighbor has the same
gID and a smaller node ID, node starts its frequency
decision. Otherwise, it waits for its neighbors’ decisions.
2) During node ’s frequency decision, it first decides the
portion of frequency range to choose from. The range
starts where the most recently overheard neighbor stops,
and the length of the range is
× N.
3) When the frequency range is decided such as [Sfre,
Efre], node checks the overheard frequencies that
have been chosen by neighbors from the same network
(they carry the same gID). Node randomly chooses
one of the least loaded frequencies among the range
[Sfre, Efre].
B. Dynamic Frequency Adjustment
Since traffic patterns vary with time, the spectrum usage
must be monitored during runtime. When the spectrum usage
is found heavily imbalanced, dynamic frequency adjustment
is triggered, reassigning nodes from crowded frequencies
to lightly loaded frequencies. If fre() denotes node ’s
frequency and Tra() represents ’s traffic load, the traffic
load for a frequency (FTrai) and the traffic load for a network
(NTraj ) can be calculated as follows:
FTrai = X
fre()=frei
Tra(), NTrai = X
2neti
Tra() (2)
Also, the spectrum imbalance level can be computed as:
ImbLevl =
max{FTrai}
min{FTrai}
(3)
When a node (node ) detects that the imbalance level
ImbLevl is greater than a threshold ImbLevlThr, it triggers
the dynamic frequency adjustment, which repeats the following
process until the imbalance level is below the threshold:
notifying a node from the busiest channel to switch to the least
busy channel. The busiest channel is identified as the channel
that has the maximum FTrai, and is denoted as frebusy.
Among node ’s neighbors that use frequency frebusy, there
may be multiple candidates from multiple networks. Among
these candidates, nodes from a comparatively more aggressive
network should be considered first, which is identified as
having a comparatively larger NTrai
i
value. If again, multiple
candidates exist within this network, the node with the highest
Tra() value stands out. Having located this neighbor, node
informs it to switch to the currently least loaded frequency,
the one that carries the smallest FTrai value.
Hot Potatoesuring dynamic frequency adjustment, each
node individually decides whether the local spectrum usage is
balanced. In some extreme cases, there may be a node that
has extraordinarily heavy bandwidth requirements compared
to others. No matter what frequency this node uses, that
frequency becomes overloaded. Instead of pushing these “hot
potatoes” around, individual nodes can detect them locally, and
keep the imbalance brought by them within the local region.
IV. THE CROWDED SPECTRUM
With the explosive application of 802.11b, 802.15.1 and
802.15.4, we vision that the human world will be full of
electronic devices and most of them work on the same or
overlapping frequency spectrum. The original 802.11 standard
released in 1997 operates within the 2.4 GHz ISM band and
divides it into 78 channels (1 MHz distance). The 802.11b also
uses the 2.4 GHz ISM band and divides it into 14 channels
(5 MHz distance). IEEE 802.15.1 divides the 2.4 GHz ISM
band into 79 1-MHz channels and IEEE 802.15.4 divides it
into 16 5-MHz channels. When these electronic devices, such
as wireless keyboards, wireless PDAs, wireless cell phone
headsets and wireless sensor networks, are bought home and
used in the same building, it is obvious that the 2.4 GHz ISM
band will be congested and overloaded.
What is worse, the widely used electric appliances like
microwaves, can also generate very strong interference. To
obtain a better understanding of the crowded spectrum, in the
presence of electronic devices as well as electric appliances,
we measured the 2.4GHz ISM band spectrum usage with a
HP 8593E Spectrum Analyzer. A Sharp Carousel microwave is
used as a representative electric appliance, which is typical in a
home care sensor network application. Also, a Logitech cordless
2.4GHz PowerPoint presenter is used as a representative
electronic device, which is typical for an office environment.
Figure 2 plots the result.
As shown in Figure 2, the small sinusoidal curve within
2.4GHz and 2.41GHz (adjacent to bottom left), indicates the
power level of the sensor network signals we deployed within
the measured environment. The large mountain like curve,
which lies between 2.43GHz and 2.47GHz (in the middle),
reflects the microwave’s interference. According to the IEEE
802.15.4 specification [27], the 2450 MHz PHY range starts