Design and evaluation of a versatile and efficient receiver-initiated link layer for low-power wireless.

Full text

Page 1

Design and Evaluation of a Versatile and Efficient
Receiver-Initiated Link Layer for Low-Power Wireless
Prabal Dutta†, Stephen Dawson-Haggerty‡, Yin Chen⋆, Chieh-Jan Mike Liang⋆, and Andreas Terzis⋆
†Computer Science & Eng. Division
‡Computer Science Division
University of MichiganUniversity of California, Berkeley
Ann Arbor, MI 48109Berkeley, CA 94720
prabal@eecs.umich.edustevedh@cs.berkeley.edu
⋆Computer Science Department
Johns Hopkins University
Baltimore, MD 21218
{yinchen,cliang4,terzis}@cs.jhu.edu
Abstract
We present A-MAC, a receiver-initiated link layer for
low-power wireless networks that supports several services
under a unified architecture, and does so more efficiently and
scalably than prior approaches. A-MAC’s versatility stems
from layering unicast, broadcast, wakeup, pollcast, and dis-
covery above a single, flexible synchronization primitive. A-
MAC’s efficiency stems from optimizing this primitive and
with it the most consequential decision that a low-power link
makes: whether to stay awake or go to sleep after probing the
channel. Today’s receiver-initiated protocols require more
timeandenergytomakethisdecision, andtheyexhibitworse
judgment as well, leading to many false positives and nega-
tives, and lower packet delivery ratios. A-MAC begins to
make this decision quickly, and decides more conclusively
and correctly in both the negative and affirmative. A-MAC’s
scalability comes from reserving one channel for the initial
handshake and different channels for data transfer. Our re-
sults show that: (i) a unified implementation is possible; (ii)
A-MAC’s idle listening power increases by just 1.12× un-
der interference, compared to 17.3× for LPL and 54.7× for
RI-MAC; (iii) A-MAC offers high single-hop delivery ratios,
even with multiple contending senders; (iv) network wakeup
is faster and far more channel efficient than LPL; and (v)
collection routing performance exceeds the state-of-the-art.
Categories and Subject Descriptors
C.2.2 [Computer-Communication Networks]:
work Protocols
General Terms
Design, Experimentation, Performance, Standardization
Keywords
Link protocols, MAC protocols, wireless sensor networks
Net-
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for components of this work owned by others than ACM
mustbehonored. Abstractingwithcreditispremitted. Tocopyotherwise, torepublish,
to post on servers or to redistribute to lists, requires prior specific permission and/or a
fee.
SenSys’10, November 3–5, 2010, Zurich, Switzerland.
Copyright 2010 ACM 978-1-4503-0344-6/10/11 ...$10.00
1 Introduction
A receiver-initiated link layer is one in which the re-
ceiver triggers communications by first transmitting a probe.
Receiver-initiated protocols have experienced a renewed in-
terest because they offer many benefits over sender-initiated
protocols for low-power wireless: they [17, 33] handle hid-
den terminals better than sender-initiated ones [30, 38, 39];
their low-power probing (LPP) mechanism ([28]) supports
asynchronous communications but avoids the long pream-
bles of sender-initiated low-power listening (LPL, [20, 30])
which run afoul of regulatory standards [2]; they support ex-
tremely low duty cycles [28] or high data rates [33]; and they
support many low-power services including wakeup [12],
discovery [11], broadcast [32], anycast [13], and pollcast [9].
Despite these many benefits, receiver-initiated protocols
face a number of drawbacks as well. Their fundamental syn-
chronization primitive – the probe – costs more than channel
sampling, which means that baseline power draw is higher
than sender-initiated protocols. Their frequent probe trans-
missions can congest the channel and delay data communi-
cations, which affects their scalability under even light traffic
loads. Their use of incompatible probe semantics for differ-
ent services makes concurrent use of those services difficult:
some probes use hardware acknowledgments [13, 28] while
others do not [9, 33]; some probes include only receiver-
specific data [28, 33] while others may also include sender-
specific data [9, 13]; and some probes include contention
windows [13, 33] while others do not [9, 28]. These dif-
ferences raise the question of whether it is possible to design
a general-purpose, yet efficient, receiver-initiated link layer.
In this paper, we present A-MAC, a new receiver-initiated
link layer that shows it is possible to support multiple ser-
vices under a unified architecture, and to do so more effi-
ciently and scalably than prior designs. Thus, we narrow
the gap between sender- and receiver-initiated approaches to
low-power wireless. A-MAC uses the backcast synchroniza-
tion primitive – a probe/ack frame exchange – to determine
quickly, robustly, and in constant time whether inbound traf-
fic is pending [13]. All other services are multiplexed above
the primitive or piggybacked on the probe. To minimize con-
tention between probe and data traffic, A-MAC (optionally)
uses one or more secondary channels to complete data trans-
feraftertheinitialprobe, allowingA-MACtoscalewithden-
sity and load [22, 25]. Section 3 presents the A-MAC design.
1

Page 2

Lacking proper hardware support, A-MAC achieves its
high performance by dynamically reassigning hardware ad-
dresses, making use of hardware address recognition, and
leveraging hardware acknowledgment collisions.
these mechanisms misappropriate addresses, violate stan-
dards, and abuse acknowledgments, the underlying ideas are
more principled, and we believe they highlight new direc-
tions for radio hardware and link protocols. The techniques
allow us to implement unicast, broadcast, wakeup, and poll-
cast using today’s off-the-shelf radios within a unified frame-
work that exposes a standard TinyOS ActiveMessage in-
terface, allowing drop-in use with many existing codebases.
Our description omits asynchronous neighbor discovery due
to space constraints, but supporting discovery is a matter of
systematically scheduling the probe and listen times [11].
Section 4 details our prototype implementation, and the var-
ious mechanisms we employ, to demonstrate the value of
hardware support for a receiver-initiated link layer.
Sections 5 and 6 explore A-MAC’s microbenchmarks and
macrobenchmarks, respectively. Key microbenchmarks in-
clude evaluating the robustness of the fundamental synchro-
nization mechanism (including effects of path delays, path
loss, and neighborhood density). We also provide energy mi-
crobenchmarks for A-MAC’s probe, receive, transmit, and
idle listening energy costs, and we present how these fig-
ures translate to average current across a range of probe and
data periods. We show that A-MAC’s idle listening power
increases by just 1.12× in the presence of interference, com-
pared to 17.3× for LPL and 54.7× for a recent receiver-
initiated MAC, RI-MAC [33]. Our macrobenchmarks show
that A-MAC offers higher single-hop delivery ratios with
multiple contending senders than RI-MAC as well. We also
show that network wakeup is nearly twice as fast as LPL and
uses vastly fewer transmissions, making A-MAC far more
channel efficient. Finally, we show that collection routing
with CTP [18] over A-MAC outperforms the state-of-the-art.
The A-MAC design faces a number of obvious limita-
tions, however. Timing critical operations require low-level
hardware support, which is only partly provided today, ham-
pering broader use. Some of the design choices violate cur-
rent standards (like acknowledging broadcast frames), but
our work shows there are significant gains to be won by do-
ing so. Since communications is receiver-initiated, the basic
primitive is a probe, which means baseline channel usage
scales with node density rather than data rate. For low or
medium density networks, this is not an issue, but for higher
density networks, it could affect latency. Although using one
probe channel and (optional) secondary channels for data
transfer helps significantly, very high neighborhood densi-
ties might also require coordinating probe transmissions [8],
which we do not explore in this paper.
While
2 Related Work
Since radio communications dominate node-level energy
consumption, it is not surprising that a wide range of MAC
protocols have been proposed for low-power wireless net-
works. Low-power links provide a range of service abstrac-
tions, allowing nodes to synchronize with peers, contend for
the channel, discover neighbors, and transfer data.
Depending on which end of a communication link initi-
ates a transfer, a MAC can be classified as either sender-
initiated or receiver-initiated. Among the sender-initiated
protocols, LPL/B-MAC [20, 30], Hui’s MAC [22], SCP [39],
S-MAC [38], T-MAC [37], and X-MAC [5], represent
canonical design points, and Flash [26] represents a link
layer flooding protocol. Among the receiver-initiated proto-
cols, PTIP [15], RI-MAC [33], LPP/Koala[28], Pollcast [9],
Backcast [13], and ADB [32] offer a range of both conven-
tional and more exotic communication abstractions. The rest
of this section compares the abstractions they provide and
the low-power synchronization mechanisms they employ.
B-MAC, S-MAC, T-MAC, X-MAC, and SCP all offer
unicast and broadcast. RI-MAC offers just unicast but ADB
essentially extends RI-MAC to offer a broadcast service.
The Koala system uses low-power probing (LPP) to offer a
receiver-initiated, asynchronous network wakeup. The Flash
flooding protocol uses low-power listening (LPL) to offer
sender-initiated wakeup. Pollcast offers single-hop collabo-
rative feedback, which allows a node to pose true/false pred-
icates to neighbors. Backcast offers an optimized acknowl-
edged anycast service that can implement Pollcast and LPP.
A-MAC offers all of these service abstractions – unicast,
broadcast, flood, wakeup, and pollcast, layered above back-
cast and within a unified link layer architecture.
Low-power wireless protocols must synchronize their
communications either explicitly by scheduling communica-
tion windows or implicitly by sampling or probing for pend-
ing traffic. S-MAC, T-MAC, and SCP all schedule commu-
nication windows: S-MAC uses fixed windows, T-MAC ad-
justs the window size to match the traffic load, and SCP ad-
justs the window size to account for clock drift.
B-MAC, X-MAC, and Hui’s MAC employ channel sam-
pling techniques to detect pending traffic. B-MAC sends
long preambles which receivers detect with channel sam-
pling.X-MAC senders transmit “packetized preambles”
and listen for a receiver-generated acknowledgment between
packets, which reduces expected channel occupancy. Hui’s
MAC employs a packetized preamble as well but transmits
preamble “chirps” which contain rendezvous time and chan-
nel data. As an optimization, neighbor sleep schedules are
also cached. PTIP, RI-MAC, and Pollcast all employ prob-
ing by transmitting probe packets. Both LPL-based sam-
pling and LPP-based probing are vulnerable to false posi-
tives (waking up when no traffic is pending) or false nega-
tives (prematurely falling asleep when traffic is pending) [4].
A-MAC transmits a probe as well, but uses explicit
hardware-generated acknowledgments as part of its synchro-
nization mechanism. The use of a probe/ack frame exchange
allows A-MAC to determine quickly, robustly, and in con-
stant time whether inbound traffic is pending. This mecha-
nism, called backcast [13], runs over 802.15.4 radios using
O-QPSK modulation [23], but similar schemes have been
shown to work for OFDM modulation as well [10]. A-MAC
also caches neighbor probe times, reducing radio on time.
Finally, A-MAC includes multichannel rendezvous informa-
tion on the first transmitted probe, which reduces congestion
and increases capacity through spectrum reuse. A-MAC es-
sentially integrates several earlier optimizations.
2

Page 3

PA
Sender
Receiver
PA
DATA
DATA
Max data packet
4.256 ms
ACK transmission time 352 µs
RXTX turnaround time: 192 µs
P
PL
L
Figure 1. A-MAC communications timing and flow. A
sender listens (L) for a receiver’s probe (P) which it auto-
acks (A) precisely 192 µs later. The sender subsequently
transmits a data frame (DATA) after a short but random
interval, perhaps on a different channel, which the re-
ceiver acknowledges with a second probe and then listens
briefly for an auto-ack before returning to sleep.
3A-MAC Design Overview
This section presents the design of A-MAC, a receiver-
initiated link layer for low-power wireless networks that
supports several services under a unified architecture. We
ground our discussion in the context of the IEEE 802.15.4
standard. The basic A-MAC design requires a sender to
first listen for a probe frame from the intended receiver, then
acknowledge the frame using the 802.15.4 standard’s sup-
port for hardware automatic acknowledgments (auto-ack or
HACK), then pause for a short, random delay, and finally
transmit the data frame if the channel is clear.
Figure 1 shows the critical time constants of an optimized
A-MAC communication over 802.15.4. In this figure, the
probe, labeled P, is a standard data frame transmitted by
the receiver with the acknowledgment request bit set. The
sender, upon receiving this probe frame, generates an auto-
ack, labeled A. The 802.15.4 standard stipulates that the
auto-ackmustbegeneratedprecisely12symbolperiods(192
µs) after the end of P. The auto-ack frame is 11 bytes long1
and requires 352 µs to transmit. A sender transmits a DATA
frame with a short, random delay after the auto-ack A, poten-
tially on a different channel as stipulated in the probe. A sec-
ond probe acknowledges the data frame. If the second probe
does not trigger an auto-ack, the receiver goes to sleep.
This design choice – to use an auto-ack – departs from
prior work in which receiver-initiated MACs simply send a
data frame in response to a probe [17, 33]. This decision is
motivated by the observation that the most consequential de-
cision that a low-power MAC makes after polling the chan-
nel is whether to stay awake or go back to sleep. Since this
decision must be made on the order of one hundred thousand
times or more per day in a typical low-power MAC, being in-
decisive or incorrect can get very costly very quickly. If the
MAC decides traffic is pending when none exists – a false
positive – then the radio will remain on, wasting energy. If
the MAC decides no traffic is pending when some is – a false
negative – then the sender’s energy is wasted, communica-
tion latency increases, and packet goodput drops.
1A hardware auto-ack or HACK frame includes: preamble (4),
start-of-frame delimiter (1), length (1), frame control (2), sequence
number (1), frame check sequence (2).
P
A
Node 2
(Receiver)
Node 3
(Sender)
P
A
Node 1
(Sender)
Listen
D
D
P
P
L
P
A
Listen
D
P-CW
P
A
Listen
D
P
P
A
D
P-CW
D
CW
D
frame collision
Backcast
Figure 2. A contention-free transfer (left) and a collision
(right). Although the auto-ack frames collide, they do so
non-destructively, so the receiver correctly decodes their
superposition as a valid frame. Hence, the receiver con-
cludes that traffic is pending, so it retransmits a probe
with an explicit contention window, which Node 3 wins.
Clearly, making a good decision about whether to stay
awake or go to sleep is a critical one, but it is not an easy one
for many reasons. First, external interference (e.g., 802.11
network) might be mistaken for legitimate radio activity.
Second, a receiver might overhear a partial packet sent to
a different node, and stay awake until it can conclude that
the packet is destined elsewhere. Third, hidden terminals
might cause packets from multiple senders to collide at the
receiver. Note that it might not be possible for the receiver to
differentiate collisions from interference, forcing the radio to
stay awake for shorter than required or longer than desired.
Our design reliably and efficiently balances these con-
flicting needs by using backcast, a link layer primitive that
allows a node to probe all of its neighbors in parallel and
robustly distinguish the case of zero replies (indicating no
pending traffic) from the case of one or more replies (indi-
cating pending traffic) [13]. In the former case, the MAC can
turn off the radio quickly2and return to a sleep state. In the
latter case, the MAC would leave the radio on to receive the
auto-ack frame and any additional data frames. Note that all
senders with pending traffic for a particular receiver concur-
rently transmit an auto-ack, as Figure 2 shows. Although
these auto-acks collide, they do so non-destructively with
high probability. Therefore, the receiver can decode their
superposition as a valid frame and conclude that traffic is
pending. In the case of a data frame collision, the receiver
retransmits the probe with a larger contention window.
All other link layer services are implemented above the
backcast synchronization primitive using a combination of
hardware auto-acks and judicious frame filtering. Unicast,
in principle, could be implemented by auto-ack-ing frames
based on the probe source address. Broadcast and wakeup
could be implemented by auto-ack-ing all probes which
have the ACK request bit set in the 802.15.4 frame control
field. Pollcast could be implemented by including a pred-
icate in the probe itself, which is quickly evaluated by the
sender and if found true, then auto-acked. Unfortunately, the
needed hardware support is lacking in modern radios, requir-
ing some creative contortions, which we describe next.
2Since the radio would not signal a start-of-frame (SFD) event.
3

Page 4

4Implementation Details
Section 3 presents a conceptual, clean-slate design for the
A-MAC link layer. Unfortunately, modern radios lack the
hardware and software support needed to optimally imple-
ment the A-MAC design. To work around the limitations
of current hardware, we implement a version of A-MAC that
misappropriatesaddresses, violatesstandards, andabusesac-
knowledgments. However, the goal of our work is to demon-
strate the power and performance benefits of the design; the
underlying ideas are more principled than the hacks we em-
ploy to accomplish this goal. We hope this work highlights
new directions for radio hardware and link protocols.
4.1Software, Hardware, and Radio Platform
A-MAC is implemented in TinyOS 2.1 [21] and runs on
the Berkeley TelosB [31] and Epic [14] motes. The back-
cast synchronization primitive of A-MAC also runs on the
Crossbow Iris [7] mote, but we did not implement the rest of
A-MAC on the Iris mote because the radio-processor inter-
face is more limited, due to fewer handshake lines, than the
TelosB and Epic platforms, which offer better A-MAC per-
formance due to a more efficient processor-radio interface.
The TelosB and Epic platforms are based on the TI
CC2420 radio [34] while the Iris uses the Atmel AT86RF230
radio [3].Both the CC2420 and the AT86RF230 radios
are 802.15.4 standards-compliant and they inter-operate at
a 250 kbps data rate. Therefore, they both support back-
cast using offset quadrature phase shift keying (O-QPSK)
modulation with half-sine pulse shaping [19] used in the
802.15.4 standard [23]. This modulation technique employs
continuous-phase frequency shift keying and is also known
as minimum shift keying (MSK) [29].
4.2 Backcast-Based Synchronization
We implement the backcast synchronization primitive us-
ing the hardware automatic acknowledgments (auto-acks)
available in all 802.15.4 standards-compliant radios. The
scheme works as follows on the CC2420 radio. A receiver
transmits a frame to a unicast, multicast, or broadcast ad-
dress. Nodes with pending traffic for the receiver temporar-
ily set their radio’s local hardware address to the particular
destination address transmitted in the probe frame by the re-
ceiver (this address is a special value, specific to the service,
and described later in this section). All nodes that match the
destination address transmitted in the probe frame respond
withidenticalacknowledgmentframesthatareautomatically
generated by their radio hardware. Receiving an auto-ack
signals to the receiver that inbound traffic is pending.
More generally, the 802.15.4 MAC defines a frame con-
trol field (FCF) that includes an acknowledgment request
flag. On the CC2420, when configured for automatic ac-
knowledgments, an auto-ack frame is transmitted after an in-
coming frame meets three conditions: it (i) has the acknowl-
edgment request flag set, (ii) is accepted by the radio’s ad-
dress recognition hardware, and (iii) contains a valid CRC.
Acknowledgments are transmitted without performing clear
channel assessment, so their timing is not delayed due to in-
terference [23, 34].
4.3Unicast Communications
In typical receiver-initiated unicast communications, a
sender first listens for a probe frame and then transmits a data
frame in response to the probe. The sender may jitter the data
transmission with a small, random delay to avoid collisions
when multiple senders are contending. Protocol processing
overhead can introduce additional delays in generating the
data frame (unless it is preloaded into the radio’s transmit
buffer): the sender must receive the probe, copy it from the
radio to the processor memory, signal an interrupt, dispatch
the frame to the link layer, determine if the frame is indeed
a probe from the intended receiver, and if so, then possibly
jitter the transmission, and finally copy the data frame into
the radio’s transmit buffer and issue a transmit command.
Meanwhile, the receiver must wait patiently with its radio
turned on, wasting precious energy and remaining suscepti-
ble to false positives from external interference.
The A-MAC unicast design diverges from traditional
receiver-initiated designs by first acknowledging the probe
with a fast and deterministic radio-generated frame (a back-
cast frame exchange [13]), and only then sending the data
frame.This approach has many benefits.
ceiver only has to wait marginally longer than the radio’s
RX/TX turnaround time before concluding that no inbound
traffic is present, saving considerable energy on every probe.
In the IEEE 802.15.4 standard, a turnaround occurs in
192 µs, nearly 20 times faster than the 3.75 ms beacon-data
turnaround time that RI-MAC requires with its software-
based protocol processing [33]. Second, our approach dis-
tinguishes between collisions and interference, whereas RI-
MAC cannot. In RI-MAC, as with LPL channel samples,
interference leads to extended listening. With a backcast-
based approach, interference is easily distinguished from an
auto-ack superposition since the former appears as just chan-
nel energy while the latter results in a valid frame reception.
Therefore, A-MAC is far less susceptible to interference-
based false alarms than either LPL or RI-MAC.
To implement unicast, we use two key features of
802.15.4-compliant radios: hardware-based address filter-
ing and hardware-generated auto-acks. The critical design
question is what source and destination addresses should be
used in the probe frame? One option is to send the probe
to the broadcast address requesting an auto-ack. Under this
scheme, a node with pending traffic for any destination en-
ables auto-acks for broadcast frames.
However, there are several problems with this approach,
as follows. First, a sender will auto-ack every probe it re-
ceives, including probes from neighbors for which the sender
has no pending traffic. This will cause all but one neighbor
to stay awake unnecessarily and waste energy. We call this
the overreacting problem. Second, the IEEE 802.15.4-2006
standard specifically prohibits this behavior: § 7.5.6.4, “...
any frame that is broadcast shall be sent with its Acknowl-
edgment Request subfield set to zero.” Third, because this
behavior is prohibited, it enjoys somewhat mixed radio sup-
port: while the CC2420 [34] radio and AT86RF230 [3] radio
Rev A silicon both support broadcast auto-acks, the Rev B
silicon “fixes” this standards non-compliance and does not
auto-ack broadcast frames.
First, the re-
4

Page 5

!∀
#∃%&∋(
)∗&+&,−&./
!∀
#∃%&∋2
)0&1%&./
3,45&1
6
56789:∀99;
6<289:999;
6!
!3
=>289:∀99;
56789:∀99;
6<289:999;
>2?89:99;≅
Α<=89:999!
56789:999;
6<289:999!
6ΒΧ89:;≅
Figure 3. Example of an A-MAC unicast communication
showingdynamicaddresschangesandotherframefields.
We avoid the overreacting problem and design a
standards-compliant unicast protocol as follows.
sender S has pending traffic for receiver R, S enables hard-
ware address recognition, enables its hardware auto-acks,
and sets its hardware address to R+0x8000.3
sending a probe to the broadcast address, receiver R sends
its probe to destination address R+0x8000 and requests an
auto-ack. Sender S (as well as any other nodes with pending
traffic to R) respond to the probe. If its probe is acknowl-
edged, R remains awake to receive a frame while sender S
does not succumb to the overreacting problem.
Figure 3, shows a sender (Node 1) with traffic pending
for the receiver (Node 2). The sender turns on its radio, sets
its hardware address to 0x8002, enables hardware auto-acks,
and begins to listen. At some later time, the receiver wakes
up and sends a probe with a source address of 0x0002and a
destination address of 0x8002, and requests an acknowledg-
ment. When the sender receives the probe frame, its radio
generates an auto-ack. Upon detecting the beginning of the
auto-ack, the receiver decides that an auto-ack frame may be
incoming, so it continues to listen for at least 352 µs (or pos-
sibly less if the data appear garbled) before turning off the
radio. If a valid auto-ack is received, the receiver concludes
there is pending traffic for it, and it remains awake to receive
this data. At the same time, the sender transmits a data frame
(after a short random delay comprising the contention win-
dow) with a source address of 0x0001, a destination address
of 0x0002, and a locally-selected sequence number of 0x23,
which is successfully received. The sender does not change
its radio hardware address for this transmission. The receiver
then prepares its next probe which explicitly acknowledges
the preceding data frame by source address (0x0001) and
sequence number (0x23). The sender turns off auto-acks if
it has no further data pending (or repeats this process if it
has more data), letting the receiver’s second probe go un-
acknowledged, which allows the receiver to return to sleep
after a brief wait.4
When
Instead of
3We reserve addresses with the high-order bit set for such use.
4As an optimization, the receiver could acknowledge the
sender’s data frame, which the sender would use as a “hint” that its
transmission was successful (since hardware auto-acks only have
sequence numbers but not source or destination addresses). This
optimization allows the sender to disable auto-acks prior to the re-
4.4Broadcast Communications
Broadcast is a fundamental operation used by a wide
range of higher-layer services and applications. Neighbor
discovery, routing updates, and data dissemination all de-
pend on a robust broadcast service for operation. A-MAC’s
designofthebroadcastserviceisidenticaltounicastcommu-
nications with one important difference. A sender S, simply
disables hardware address recognition altogether but keeps
hardware auto-acks enabled. Of course, this requires that
auto-acks be used exclusively for responding to probes (e.g.,
they cannot be used to acknowledge data).
When a higher-layer service needs to send a broadcast,
it sets the destination address of the frame to the broadcast
address, e.g., 0xFFFF, and submits the frame to A-MAC for
delivery. When this frame is ready for transmission, A-MAC
disables the hardware address recognition function of the ra-
dio for at least as long as the probe period of its neighbors
(or the longest of its neighbors’ probe periods, if different
neighbors are operating with different periods). During this
time, S will auto-ack every probe it receives, regardless of
the probe’s actual destination address, and proceed to send
the data packet like in the unicast case. Although this design
does not violate the 802.15.4 standard, it clearly abuses the
standard in support of physical and link layer primitives that
thestandardwasnotoriginallydesignedtoprovide. Ourgoal
is to show the feasibility of the A-MAC design using existing
hardware, not that it is necessarily standards-compliant (al-
though the latter is preferable, to allow it to be tested using
off-the-shelf, standards-compliant hardware).
A common case that arises with this design is what to do
if, while the broadcaster is listening for neighbors’ probes,
the broadcaster’s own probe timer fires. Should it send the
probe and then return to listening or should it forgo the probe
and continue listening? The A-MAC design chooses the first
approach: a probe is transmitted when the probe timer fires.
Doing so avoids a scenario we call the broadcast standoff in
which two or more nodes that attempt to broadcast a packet
wait patiently for the other(s) to first transmit a probe. The
A-MAC design avoids this situation, but it raises two fur-
ther issues. First, the transmit and receive state machines
within a node become more complex and cross-coupled.
Second, while probing, a broadcaster may miss other neigh-
bors’ probes, thereby reducing broadcast reliability.
A potential issue with our design is that if hardware auto-
acks are used to acknowledge data frames as well as probes,
then a broadcaster would inadvertently acknowledge every
single data frame it received, signaling that the data were
successfully received when in fact it may not have actu-
ally been received. Our unicast implementation avoids this
problem by reserving hardware auto-acks exclusively for ac-
knowledging probes. Data frames are acknowledged by in-
cluding the acknowledgment information in the next probe.
ceiver’s next probe transmission, eliminating a race condition in
which sender has to check the contents of the receiver’s second
probe to decide whether to acknowledge it. The sender still waits
for the receiver’s second probe to verify the hint by checking that
the second probe’s sequence number and source address match
sender’s previous frame. However, this approach is incompatible
with broadcasting, as we describe in § 4.4.
5

Page 6

PA
Node 2
Node 3
Node 4
PA
Node 1
Node 5
Listen
ListenP
A Listen
P
A
Listen
PAListen
Listen
PA
PA
ListenPA
P
P
A
A
Backcast
DST=0xFFFF
SRC=0x0002
DST=0xFFFF
SRC=0x0003
DST=0xFFFF
SRC=0x0004
DST=0xFFFF
SRC=0x0005
Figure 4. Asynchronous network wakeup with A-MAC.
Although Nodes 2, 3, and 4 all ACK Node 5’s query
probe, the ACK collision is non-destructive, and Node 5
remains awake to communicate.
4.5Asynchronous Network Wakeup
Waking up a multihop network of duty cycled nodes is a
fundamentalprobleminsensornetworks. Applicationsasdi-
verse as interactive data collection, exceptional event detec-
tion, and target tracking require nodes to wake up neighbors
or even the entire network in response to an asynchronous
event. In many such applications, nodes will remain asleep
for long periods of time and so they are likely to lose syn-
chronization. Ideally, the nodes would wake up only in re-
sponse to external events or user queries, but would other-
wise remain asleep. In the case of mobile sensors, nodes
may only need to communicate when they have data to up-
load. However, itisstillusefultobeabletowakeupamobile
node to issue it a command or query.
Several techniques have been proposed for asynchronous
network wakeup in a low-power setting including various
forms of flooding and dissemination, but these techniques
have poor channel efficiency, exhibit logistic-like perfor-
mance in that they start and end slowly, or are designed
with the assumption that nodes are synchronized. As a re-
sult, none of these techniques are ideally suited to the low-
power, asynchronous network wakeup problem. In this sec-
tion, we discuss two approaches to designing a backcast-
based wakeup service – one that can work with standards-
compliant radios and one that cannot. They exhibit high
channel efficiency, achieve the lower bound on wakeup time,
and do not assume synchronization.
Figure 4 shows the first approach. In this figure, all nodes
cease periodic communications like routing beacons and in-
stead operateat averylow dutycycle. Thenodes wake upin-
frequently, perhapsonceeverytensecondsoreachminute, to
check if any of their neighbors requires them to stay awake,
by sending a probe to the broadcast address. Node 1 ini-
tiates an asynchronous network wakeup by configuring its
radio to acknowledge all frames. After some time, Node 2
sends a probe. Node 1 auto-acks this probe and Node 2 stays
awake. This process repeats with Node 2 waking up Node 3
and Node 4. However, when Node 5 wakes up, all of its
neighbors – Nodes 2, 3, and 4 – are already awake and they
all simultaneously auto-ack Node 5’s probe, which Node 5
correctly decodes as a valid frame and hence remains awake.
Node 2
(Receiver)
Node 3
(Sender)
Node 1
(Sender)
PredEvent
PredEvent
Pred
Listen
MAC=0x8765
Listen
MAC=0x8765
Listen
PA
PA
PA
Backcast
DST=0xFFFF
SRC=0x0002
PRED=elephant
MAC=0x8765
Event
DST=0x8765
Figure 5. Pollcast implemented using the A-MAC archi-
tecture. All nodes observe an “elephant sighting” event.
Node 2 wishes to corroborate this observation with its
neighbors. It uses backcast to efficiently determine if any
neighbor also observed this event.
Transmitting to the broadcast address with the acknowl-
edgment request bit set does not comply with the 802.15.4
standard (and hence only works with the CC2420). One way
to sidestep the issue is to send the probe to a reserved wakeup
address rather than the broadcast address. This leads to a
wakeup phase, in which a node firstperforms wakeup for one
cycle, and then engages in normal communications. This ap-
proach may be preferred since it also disentangles broadcasts
and floods from wakeup, and is standards-compliant.
One problem common to both designs is that if a
node misses the acknowledgments to its specially-addressed
probes during the network wakeup phase, then the node will
remain asleep after the wakeup phase since its neighbors will
no longer acknowledge specially-addressed probes.
problem, too, can be avoided by using a special wakeup ad-
dress and increasing the length of the probe frame so that
the processor has enough time to: (i) read the address from
the radio’s receive FIFO while the rest of the probe is being
received (i.e., pipelining the read and reception), (ii) check
if the address matches the special wakeup address, and (iii)
instruct the radio to auto-ack the frame within the tight time
window needed to generate a timely auto-ack. We do not ex-
plore this idea any further in the context of network wakeup,
but we do return to it in a more general form in the context
of Pollcast.
This
4.6Pollcast Neighborhood Queries
Demirbas et al. recently proposed pollcast, a two-phase
primitive in which a node broadcasts a poll about the exis-
tenceofanode-levelpredicatePandthenallnodesforwhich
P holds reply simultaneously [9]. The poller detects one or
more positive replies by sampling its radio’s Clear Channel
Assessment (CCA) signal which indicates whether the re-
ceived signal strength exceeds a threshold. While pollcast
offers a novel approach to quickly calculate predicates, the
proposed mechanism has some drawbacks, as their work ac-
knowledges: simultaneous pollcasts within a two-hop neigh-
borhood causes false positives (as would external interfer-
ence). Selecting the CCA threshold presents a tuning chal-
lengesincesettingittoolowcausesfalsepositivesbutsetting
it too high causes false negatives.
6

Page 7

A-MAC provides a more robust architecture for im-
plementing pollcast by mapping the original two-frame,
query/response to a three frame operation. First, a single
frame transmission containing the predicate to be evaluated
is sent to the broadcast address, received by all neighbors,
and evaluated. Next, a short time later, a probe is transmitted
to a special address (contained in the first transmission). Fi-
nally, the probe is acknowledged by all nodes for which the
predicate evaluated true.
Figure 5 illustrates an example in which all nodes ob-
serve an event. Node 2 wishes to corroborate an “elephant
sighting” event with its neighbors so it transmits a predicate
describing the event, including a locally-generate ephemeral
identifier. The destination address of the predicate is 0xFFFF
(broadcast), the source address is 0x0002, the predicate is
’elephant’, andtheephemeralidentifieris0x8765. Node2
then waits for some time to allow Nodes 1 and 3 to receive
and evaluate the predicate. Node 2 then sends a probe des-
tined to the ephemeral identifier 0x8765. Since both Node 1
and Node 3 observed the same event, they both auto-ack the
probe, indicating the predicate was true. Note that although
the predicate is sent to the broadcast address, it does not need
to be automatically acknowledged, so this approach is com-
patible with 802.15.4 and A-MAC’s unicast and broadcast.
A drawback with this approach to pollcast is the need for
two packet transmissions by the receiver: the first packet
sends the predicate and the second packet sends the probe to
the ephemeral identifier. Ideally, the predicate could be pig-
gybacked onto to the probe, eliminating the separate pred-
icate transmission and its associated delay. The two chal-
lenges with this approach include choosing the destination
address of the probe and ensuring that the predicate can be
evaluated quickly enough (by the processor) to generate a
properly timed auto-ack. One option that we explore is to
send the probe to the broadcast address, piggyback the pred-
icate on the probe, and pad the probe with a large payload.
This allows a node to detect the beginning of the probe, read
and evaluate just the predicate while the rest of the packet is
being received, and enable hardware auto-acks before frame
reception completes. The pad bytes provide buffer time to
evaluate the predicate before the 192 µs auto-ack timer fires.
4.7Miscellaneous Details
In the current A-MAC implementation, each node
chooses its own probe schedule without any local or global
coordination. If two nodes pick identical schedules, we rely
on capture and contention for short-term progress, and clock
drift for long-term desynchronization.
plementation could use an explicit desynchronization proto-
col like DESYNC [8]. When two nodes communicate, the
sender caches the receiver’s probe period and phase. This al-
lows the sender to minimize its radio on-time during subse-
quent communications by listening just before the expected
probe transmission. The cache holds four entries and uses an
LRU eviction policy. An alternate policy might consider us-
age frequency. Queued packets are transmitted round-robin
for fairness, but this can result in head-of-line blocking. An
EDF policy that orders pending packets by their receivers’
probe times may be a better option, especially since current
hardware can only auto-ack one receiver’s probes at a time.
An improved im-
5Backcast Evaluation
Backcast is a critical primitive upon which A-MAC rests,
so we evaluate its reliability, efficiency, and performance un-
der a range of conditions including carefully-controlled lab-
oratory settings and more realistic indoor settings. Our re-
sults show that backcast works on two different radios from
two different vendors, has a narrow range of failure cases,
provides high energy- and channel-efficiency, and provides
a strong foundation upon which to build the remaining link
layer services.
5.1Methodology
We use the Moteiv Tmote (Telos B) [31], Berkeley
Epic [14], and Crossbow Iris [7] motes for these experi-
ments. We find that the backcast performance of both radios
is similar, so we only report detailed results for the CC2420
radio. In the experiments that follow, signal strength is mea-
sured by the radio over the first eight symbols of an acknowl-
edgment (ACK) frame and reported as the received signal
strength indicator (RSSI) in dBm. Signal quality (LQI) is
also measured by the radio over the first eight symbols and
is reported as a 7-bit unsigned integer that can be viewed as
the average correlation value or chip error rate (values near
100 indicate an excellent link).
5.2ACK Reception Robustness
We first explore how delay differences in the path length
affect ACK reception rate. Figure 6(a) presents the setup for
this experiment. Two nodes, an initiator and a responder
(both Tmotes) are connected to each other through a pair of
circulators and a wireless channel emulator. A circulator is
essentiallyanRFsplitterthatprovidesalow-lossRFpathbe-
tween some terminals (1-to-2, 2-to-3, and 3-to-1) but a very
high-losspathbetweenotherterminals(1-to-3, 2-to-1, and3-
to-2). Circulators are used to split a single bi-directional RF
path into two unidirectional paths. We use the D3C2060 cir-
culator from DiTom Microwave. A wireless channel emula-
tor allows a complex RF environment, including attenuation,
delay, fading, Doppler shift, and multipath, to be evaluated
in a laboratory setting. We use the Spirent SR5500 wireless
channel emulator in these experiments. The SR5500 allows
each channel to be composed of several independent paths,
each with its own delay and attenuation.
5.2.1 Effect of Path Delay Differences
To evaluate the effect of path delay difference on destruc-
tive intersymbol interference, the ACK channel from the re-
sponder to the initiator (Channel 2) is split into two equal
loss paths inside the channel emulator. The delay in the sec-
ond path is swept from 0 to 1 µs in 10 ns steps. For each
delay step, the initiator transmits 100 packets to the hard-
ware broadcast address, at 125 ms intervals, and logs the
RSSI, LQI, and sequence number of the resulting acknowl-
edgments. The results are shown in Figure 6(b) and indicate
intersymbol interference becomes destructive between 500
and 600 ns, as expected. Note that a delay of 500 ns corre-
sponds to a path delay difference of 150 m. Such path delay
differences are rare in low-power wireless networks; links
are rarely more than tens of meters, so such significant delay
differences would result in different received signal strength
values as well (unless transmission power control is used).
7

Page 8

Wireless Channel Emulator
1
2
2
3
13
Circulator (2)
Channel 1Channel 2
Responder
Faraday Cage
USBRF
Initiator
(a) Experimental Setup
500550 600650700 750800
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Path Delay Difference (ns)
Packet Reception Rate
(b) Intersymbol Interference
0.51 1.522.53 3.5
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Path Power Difference (dB)
Packet Reception Rate
(c) Power Capture
Figure 6. Figure (a) shows the experimental setup. Figure (b) shows the onset of destructive inter-symbol interference.
Packet reception rate falls sharply as the delay difference in two paths exceeds 0.5 µs. Figure (c) shows the effect of
power capture. When two frames collide, the first frame to arrive will be decoded correctly if its receive power is 3 dB
higher than the second frame.
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G
G G
G
GG
G
G
G
G
G
G
G
G
G
G
G
G
GG
G
G
G
G
G
G
G G
G
GG
G
G
GG
G
G
G
G
GGG
G
G
G
G G
G
G
G
G
G
G
G
G
G
G
G
G
G G G
G
G
GG
G
G
G
G
G
G
G
G
G
G
G
G
GGG
G
LQI
HACK
74
78
82
86
90
94
98
102
106
14710 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94
Figure 7. The effect on LQI as the number of concurrent ACKs increases from 0 to 94 in a typical indoor deployment
setting. The median value of LQI falls quickly for the first six nodes and then falls slowly. Beyond approximately 30
nodes, the LQI values stabilize at approximately 100. The data suggest that even in the presence of a large number of
ACK collisions, the receiver can successfully decode the ACK frame. Note the y-axis ranges from 74 to 106.
5.2.2Effect of Path Loss
Power capture occurs when the received signal from one
node is sufficiently stronger than the sum of the received sig-
nals from all other nodes [1]. To explore the effect of power
capture on backcast performance, the second path compo-
nent in Channel 2 is delayed by 8,00 ns (1/2 of the 802.15.4
symbol time). This base configuration ensures intersymbol
interferenceand, assumingequalpathloss, resultsindestruc-
tive interference and complete packet loss.
However, by adjusting the attenuation for the second path,
from 0 to 3.5 dB, in 0.1 dB steps, the effect of power capture
becomes evident. The initiator receives the superposition of
two (identical) frames, delayed by 8,000 ns, over a range
of SINR values. The results show that when the first frame
arrives with 3 dB or higher power, it will be decoded consis-
tently by the radio. The data also show a fairly linear transi-
tion region between approximately 1 dB and 2.5 dB. These
figures establishes that power capture dominates (and ex-
plains) the backcast phenomenon when the strongest ACK’s
power exceeds the sum of the remaining ACKs by more than
approximately 3 dB.
5.2.3Large-Scale Performance
Wenowexplorehowbackcastperformsinamorerealistic
setting – a university testbed. The testbed consists of Telos B
nodes and it is located in an office building with a typical
RF environment. For this experiment, 94 nodes within radio
range of an initiator are programmed to automatically ac-
knowledge all probes. The 94 nodes are turned on, one after
the other, and remain on for the rest of the experiment. After
each node is turned on, 500 frames are transmitted at 125 ms
intervals. This procedure generates a gradual increase in the
number of auto-ack frame collisions. The LQI statistics are
shown in Figure 7. The PRR is 100%.
The results show that the median value of LQI falls
quickly for the first six nodes and then falls slowly. Beyond
approximately 30 nodes, the LQI values stabilize at approxi-
mately 100, although there are outliers. The data suggest that
even in the presence of a large number of ACK collisions, the
receiver can successfully decode ACK frames, even when no
single ACK frame’s power dominates. The ACK reception
rate is nominally 100% (ACKs are received consistently, in-
dependent of the number of concurrent transmissions).
8

Page 9

051015
0
5
10
15
20
Time (ms)
Current (mA)
(a) Probe
0 20 40
Time (ms)
6080
0
5
10
15
20
Current (mA)
(b) Receive (Len=127 bytes)
590600 610 620630
0
5
10
15
20
Time (ms)
Current (mA)
(c) Transmit (Len=127 bytes)
0200400
Time (ms)
600 800
0
5
10
15
20
Current (mA)
(d) Idle (500 ms wait)
Primitive
Probe
TX only
RX only
CCA Check
Cost (µJ)
253
1578
2670
194
(e) Primitive energy costs
0 0.5
Probe period (s)
1 1.52
0
200
400
600
800
1000
Average current (uA)
(f) Probe
10
−1
10
0
10
1
10
2
10
2
10
3
10
4
Data period (s)
Average current (uA)
(g) Receive
10
−1
10
0
10
1
10
2
10
2
10
3
10
4
Data period (s)
Average current (uA)


Asynchronous
Scheduled
(h) Transmit
Figure 8. Link Power Model. Figures (a)-(c) show the Telos B mote’s instantaneous current draw for representative
asynchronous link primitives. Figure (d) shows the current draw when only listening (from 200 ms to 800 ms). Figure
(e) shows the cost of each link primitive. Figures (f), (g), and (h) show the average current for probing, receiving, and
transmitting, respectively, as a function of the probe period Tprobe(f), and data period (g) and (h) with Tprobe= 0.5 s.
The data suggest that both constructive and destructive
interference of the carrier signal occur. This result is not
surprising since the carrier signals are neither synchronized
in phase nor frequency across these 94 nodes. Rather, they
are generated locally by each node from a free-running crys-
tal oscillator. The statistical superposition of an increasing
number of signals does not lead to destructive interference,
making backcast a robust synchronization primitive.
5.3 Energy Microbenchmarks
A-MAC services are built by combining a small set of
link primitives including probe, receive, transmit, and idle
(listening for a probe). Figures 8(a)-8(d) show the traces of
these primitives as well as their energy costs. The vertical
line in Figure 8(c) indicates the point at which the sender’s
radio signals that the probe’s start-of-frame delimiter (SFD)
event has occurred. These data are collected by capturing
the voltage drop across a 10 Ω resistor in series with a 3 V
power supply using a Tektronix TDS3014 digital storage os-
cilloscope. Figure 8(e) summarizes the energy cost of each
basic primitive. In all cases, other than probes, we use the
802.15.4 link MTU frame size (127 byte payload).
Figure 8(f) shows how the average current due to probing
cost scales with the probe period. Figure 8(g) shows how
the receive cost scales with data rate. Figure 8(h) shows
how the transmit cost scales with data rate for both asyn-
chronous communications (when the sender does not know
the receiver’s probe schedule) and synchronous communica-
tions (when the sender knows the receiver’s probe schedule).
Figure 8 shows that A-MAC’s link primitives are more ex-
pensive than in an optimized, commercial-grade LPL imple-
mentationapproach[22], butunderthecriticalassumptionof
no external interference. Section 5.4 explores what happens
when this assumption is false.
Figure 9. A-MAC probe states and their energy con-
sumption. Transitions of the lower line indicate state
changes. Total consumption is 263.56 µJ. Breakdown:
(i) start (40.98 µJ); (ii) load probe (60.60 µJ); (iii) load
done (22.7 µJ); (iv) probe alarm fired (re)send (6.31 µJ);
(v) strobe and transmit (55.71 µJ); (vi) start ACK timer
(29.86 µJ); (vii) send done ACK timeout (25.71 µJ); (viii)
radio stop (8.78 µJ); and (ix) radio stopped (12.91 µJ).
Figure 9 shows the A-MAC probe’s power draw and as-
sociated state transitions. In our implementation, the instru-
mented probe consumes approximately 263 µJ. With radio
hardware support, the following states would be eliminated:
load probe, load done, probe alarm fired (re)send, and send
done ACK timeout. This would save about 115 µJ and re-
duce the cost of the probe to approximately 148 µJ – less
than three times the cost of an optimized LPL check [22].
9

Page 10

05001000 1500
0
5
10
15
20
Time (ms)
Current (mA)
(a) LPL sampling (no interfer-
ence)
051015
0
5
10
15
20
Time (ms)
Current (mA)
(b) LPL sample detail
0 100020003000
0
5
10
15
20
Time (ms)
Current (mA)
(c) LPL sampling (w/ interfer-
ence)
050100150
0
5
10
15
20
Time (ms)
Current (mA)
(d) LPL overhearing detail
Figure 10. LPL preamble sampling techniques leave receivers susceptible to noisy wireless environments, such as those
caused by 802.11 interference. Figures (a) and (b) show the macroscopic and microscopic behavior of the TinyOS 2.1
sampling algorithm when the channel is clear: the receiver immediately returns to sleep. Figures (c) and (d) show the
macroscopic and microscopic behavior while a file transfer is in progress using a nearby 802.11 access point. Of the
seven channel samples visible in this trace, five are unnecessarily lengthened due to channel noise.
5.4 Robustness to External Interference
A basic problem with LPL and LPP systems that employ
RSSI to detect the presence of incoming traffic is that they
suffer from many sources of false alarms including inter-
ference, overhearing, and collisions. Recent research has
demonstrated the cost of external interference on the effec-
tive duty cycle of LPL protocols. The results show that sig-
nificant differences can exist between the expected and ac-
tual duty cycles [4, 16]. We repeat similar experiments to
quantify the effects of interference on MAC layer operation
and energy consumption.
Table 1 shows the results of an experiment in which we
measure the receiver’s idle listening current in an office en-
vironment using three different synchronization schemes,
TinyOS 2.1 LPL, RI-MAC LPP, and A-MAC LPP, under two
different interference workloads (with and without a nearby
802.11 file transfer in progress). Although the TinyOS LPL
technique performs better under ideal conditions, it degrades
dramatically in the presence of interference, increasing av-
erage current draw by a factor of 17.3 compared to the idle
listening case. The RI-MAC LPP technique performs even
worse, exhibiting an increase in idle current by a factor of
54.7. A-MAC, in contrast, exhibits a nearly negligible 1.12×
increase in current draw, demonstrating the backcast’s re-
silience to false positives. For completeness, we also include
reported figures for Hui’s MAC [22] which uses 54 µJ per
sample (54 µJ / (3 V × 0.5 s) = 36 µA). We estimate the
power draw is doubled in the presence of external interfer-
ence (and equals the reported overhearing cost).
Figure 10 illustrates in detail how the preamble sampling
techniques used in LPL protocols leave receivers suscepti-
ble to noisy wireless environments, such as those caused by
802.11 interference during beaconing, file transfers, or au-
dio/video streaming. Figure 10(a) shows the current draw
over time when the channel is clear and Figure 10(b) shows
the detailed current draw of one channel sample.
ure 10(c) shows the current draw of the same system while
a file transfer is in progress using a nearby 802.11 access
point. Of the seven channel samples in this trace, five are of
extended length due to channel noise. Figure 10(d) shows
the details of an extended sample.
Fig-
Primitive
Operation
TinyOS LPL
RI-MAC LPP
A-MAC LPP
Hui LPL
w/o 802.11
interference
175 µA
383 µA
206 µA
36 µA†
w/ 802.11
interference
3,030 µA
12,576 µA
230 µA
72 µA‡
Increase
in Current
17.3×
54.7×
1.12×
2.0ׇ
Table 1. The effect of interference on idle listening cur-
rent. The average current draw of three different syn-
chronization schemes under no-load conditions and a
500 ms check/probe interval. Results are the average of
five samples, each one minute long. Although the LPL ex-
hibits the lowest power under ideal conditions, both the
TinyOS LPL and RI-MAC LPP exhibit dramatic power
increases under interference while A-MAC’s LPP mech-
anism shows a relatively negligible increase which shows
A-MAC’s low-power probing is resilient to false positives.
Hui’s LPL reported figure (†) is included for comparison
and our estimate of its interference current is noted (‡).
The extended channel sample in Figure 10(d), termed
“delay-after-receive-check,” improves communications reli-
ability. Shorter delays work under ideal circumstances but in
noisy or congested environments, they lead to failed commu-
nications [27]. The 100 ms delay-after-receive-check, when-
ever channel energy is detected, substantially reduces LPL
delivery failure (a false negative). A-MAC is largely immune
to this problem because it uses an explicit probe rather than
an implicit channel energy signal. We hypothesize the Hui’s
MAC is also more robust than the default TinyOS LPL due to
its use of an explicit chirp, but lacking access to it, we could
not verify this thesis. An open question is to further explore
the complex relationship between duty cycles, delivery ra-
tios, false positives, false negatives, and latency as channel
sample time is adjusted after a “busy” channel assessment.
These results show the challenge of predicting network life-
time based only on a model of the data workload, but without
a good model of the environmental factors. Although over-
hearing and interference are well-known problems, these re-
sults suggest they deserve further study.
10

Page 11

MAC No. of Senders
Packet Delivery Ratio
Avg Min
99.9%—
97.5%97.3%
95.6%95.0%
90.7%90.3%
99.9%—
99.3%98.2%
99.3%98.3%
98.5%96.7%
Max
—
97.7%
96.8%
90.9%
—
100%
99.5%
99.5%
RI-MAC
1
2
3
4
1
2
3
4
A-MAC
Table 2. Packet delivery ratios for 1 through 4 distinct
senders transmitting to a single receiver. The packet in-
terval on each sender is uniformly drawn from 0.5 to
1.5 s, so on average, it is 1 pkt/s from each sender. The
receiver uses a Tprobe=1 s. Senders attempt to send on
each probe to stress the contention algorithms, for 1,000
packets. The largest difference between the maximum
and minimum success rates for A-MAC is 2.8%, showing
that A-MAC provides fairness under modest contention.
6 Macrobenchmark Evaluation
Our evaluation thus far has focused on microbenchmarks
comparing the time, energy, false positives, and false neg-
atives of TinyOS LPL, RI-MAC, and A-MAC primitives.
We now explore several macrobenchmarks to explore how
low-level power and performance improvements translate to
high-level performance for several link layer services. For
these experiments, we use the standard TinyOS 2.1 distribu-
tion’s default LPL MAC and the RI-MAC [33] source code,
which was provided by its authors.
6.1Multiple Contending Unicast Flows
Itiswellknownthatreceiver-initiatedMACschemeshan-
dle contending flows and hidden terminals much better than
low-power, sender-initiated ones [6, 17, 33]. We now eval-
uate how well A-MAC handles multiple contending flows.
Table 2 shows between one and four senders contending to
transmit to a single receiver for both RI-MAC and A-MAC.
In this experiment, the receiver sends a probe, the senders
may all auto-ack the probe concurrently, and then they con-
tend for the channel. The receiver resends a probe after ei-
ther each successful transmission or after receiving an auto-
ack, but no data. The receiver sends up to a total of five
probes before stopping. Each probe doubles the size of the
contention window. The base contention window size is 20
jiffies (610 µs). Each node transmits 1,000 packets.
The data in Table 2 show that A-MAC matches RI-MAC’s
performance for a single transmitter but performs better than
RI-MAC when additional senders begins to contend. The
largest min-max difference is 2.8%, showing that even when
four nodes are contending, A-MAC is fair.
6.2 Multiple Parallel Unicast Flows
We now evaluate how well A-MAC supports multiple
concurrent flows between distinct pairs of senders and re-
ceivers that are all located in a single collision domain.
This experiment tests A-MAC’s multichannel optimization
in which the probe and acknowledgment are transmitted on
a shared control channel, but data transfer may occur on a
WL
(#)
3
3
3
Tx:Rx
(ratio)
1:1
1:1
2:2
Probe
(ms)
512
128
128
Rate
(pkt/s)
48.4
71.9
78.7
78.9
27.8
58.9
74.7
77.0
88.9
80.7
85.0
22.1
30.8
36.5
Pkts
(#)
8,010
8,056
5,982
5,365
1,895
3,535
5,975
6,007
4,912
4,834
5,098
1,324
1,845
2,219
Time
(s)
164
112
76
68
68
60
80
78
60
60
60
60
60
59
PDR
(%)
100
100
99.2
99.4
98.9
99.9
99.2
99.3
99.3
99.3
99.4
97.0
98.3
99.3
1 2:2 128
62:2128
6 3:3 128
1 3:3128
Table 3. A-MAC performance with multiple parallel uni-
cast flows. Throughput and packet delivery ratio im-
prove with additional channels. Even without the multi-
channel optimization, A-MAC can sustain multiple, par-
allel unicast flows located in the same collision domain.
different channel as stipulated in the probe. Table 3 shows
A-MAC throughput and packet delivery ratio as a function
of the number of different whitelisted channels that are avail-
able for use, the number of sender:receiver pairs transferring
data concurrently, and the receivers’ probe interval.
WL refers to the number of channels in the whitelist
where WL=1 means all traffic happens on the control chan-
nel (25), WL=3 means channels 15, 21, and 24 are in the
whitelist, and WL=6 means channels 11, 15, 20, 21, 24, and
26 are in the whitelist. Tx:Rx identifies the number of inde-
pendent transmitter:receiver pairs concurrently transmitting.
The Probe field specifies the probe interval. The throughput
(Pkt/s), data size (#Pkts), transfer time (Time), and packet
delivery ratio (PDR) are shown. The data show that through-
put improves significantly with additional channels while the
(already high) packet delivery ratio improves slightly with
additional channels.
6.3Asynchronous Network Wakeup
A network wakeup is a special case of flooding or dis-
semination in which the goal is to ensure that every node
in the network receives a wakeup message. Prior work has
shown that LPL-based flooding techniques can cause sig-
nificant contention and can use the radio channel (a scarce
resource) over an extended period of time to complete a
flood [26]. Figure 11 explores how well the TinyOS 2.1 LPL
and A-MAC wakeup implementations of asynchronous net-
work wakeup compare. In this experiment, the LPL wakeup
algorithmisasimpleflood: thesourceofthefloodrepeatedly
resends a wakeup packet for slightly longer than the sleep in-
terval. Every node that receives the packet also retransmits
it, after it detects a clear channel. The A-MAC flooding algo-
rithm is a recursive broadcast without subsequent data packet
transmissions. Since network wakeup is a special case of
flood, this experiment also establishes A-MAC’s broadcast
performance.
11

Page 12

1251000
Probing inverval (ms)
20004000
0
5
10
15
Wakeup Time (s)


LPL
A−MAC
(a) Wakeup Latency
1251000
Probing inverval (ms)
20004000
0
1
2
3
4
5
x 10
4
Sent Packets


LPL
A−MAC
(b) Sent Packets
1251000
Probing inverval (ms)
20004000
0
1
2
x 10
5
Received Packets


LPL
A−MAC
(c) Received Packets
0
1
0.5
1
LPL
0246
0
0.5
Wakeup Latency
A−MAC


125
500
2000
4000
(d) Normalized Wakeup Latency
Figure 11. Wakeup latency for LPL and A-MAC for several sleep periods (0.125 s, 0.5 s, 1 s, 2 s, 4 s). Figure (a)
shows A-MAC wakes up the network in about 38% less time than LPL. Figures (b) and (c) show that far few packet
transmissions and receptions are required by A-MAC. Figure (d) shows the CDF of wakeup latencies normalized by the
probe interval. A-MAC wakes up the network faster, uses far fewer packets, and is far more channel efficient than LPL.
LPL
6.36%
95.1%
7.34
14
A-MAC
4.44%
99.7%
4.85
13
Average Duty Cycle
Average Packet Delivery Ratio
Average Hop Count
Maximum Hop Count
Table 4. CTP performance over LPL and A-MAC. A-
MAC offers higher packet delivery ratio, lower duty cy-
cles, and lower average hop count. A-MAC performance
meets or exceeds the widely-used TinyOS LPL link layer.
Figure 11 shows the wakeup times of 59 nodes in a mul-
tihop testbed across a range of sampling/probing intervals.
Figure 11(a) shows that A-MAC wakes up the network about
38% faster than the default TinyOS LPL. Figures 11(b)-(c)
show A-MAC transmits far fewer packets to do so, hence ex-
hibiting dramatically better channel efficiency. Figure 11(d)
shows the CDF of wakeup latencies. The better relative per-
formance of longer probe intervals seems counter-intuitive,
but it occurs because there is a lower probability of a node
transmitting a probe when a neighbor is otherwise occupied
as the probe interval length is increased.
6.4 Collection Tree Protocol Performance
We next explore how well the Collection Tree Protocol
(CTP) [18] performs over the A-MAC link layer. CTP is the
default collection routing protocol in TinyOS and it repre-
sents a canonical link layer client. Since A-MAC exports
the standard TinyOS ActiveMessage layer, running CTP
over A-MAC is largely a matter of changing configuration
wirings. In the following two experiments, a network of
59 nodes run CTP and are programmed such that each node
generates one data packet every 60 seconds. An experiment
runs for one hour and the default TinyOS LPL implementa-
tion and A-MAC are tested in different experiments. Table 4
summarizes the results of the two experiments, run sequen-
tially, and repeated twice, for a total of three trials.
Figure12(a)showstheCDFofper-nodedutycycleswhen
running CTP over LPL and A-MAC. The A-MAC and LPL
experiments are run sequentially and the experimental pair
is repeated three times. The A-MAC and LPL CDFs are
largely self-correlated across the runs. A substantial frac-
tion (≈ 60%) of the nodes running A-MAC exhibit a strictly
lower duty cycle than any LPL nodes. Every node running
A-MAC exhibits a lower duty cycle, on a percentile basis,
thanthecorrespondingLPLnode, exceptfor5%ofthenodes
in Trace 1. The results show that A-MAC satisfies realistic
workloads, achieves lower duty cycles, offers higher packet
delivery ratios, and provides greater channel efficiency.
6.5Interference Vulnerability
In Section 5.4, we explored the effect of interference on
the idle listening average current of TinyOS LPL, RI-MAC
LPP, and A-MAC LPP in the presence and absence of nearby
802.11 file transfers. We now extend these experiments to
explore the power draw of these three MAC layers on two
different channels (802.15.4 channels 18 and 26) in a typi-
cal computer science department over the course of approxi-
mately 12 hours (noon to midnight). Figure 12(b) shows the
average power draw on channel 18 and Figure 12(c) shows
the average power on channel 26, all for a fixed probe inter-
val. In all cases, A-MAC exhibits both the lowest and most
stable power draw while LPL exhibits the highest power
draw and RI-MAC exhibits the most volatile power draw.
6.6 Effect of Density on Packet Delivery
One major drawback of receiver-initiated protocols is that
because of their periodic probing, their channel utilization
scales with node density and probe frequency rather than
strictly with traffic. We now explore the effect of node den-
sity and probe frequency on packet delivery rate. In this ex-
periment, a single sender transmits 100 packets to a single
receiver with an inter-packet interval of 500 ms. We vary the
number of additional nodes that are in the same collision do-
main (between 0, 1, 2, 3, 8, 13, and 18) who simply transmit
probes with varying probe periods (32 ms, 64 ms, 128 ms,
and 256 ms). Probes are transmitted using clear channel
assessment (CCA) enabled. Figure 12(d) shows the results
of running these 28 experiments. We see that packet deliv-
ery ratio drops with both increasing density and decreasing
probe intervals, as expected. This figure illustrates a ma-
jor weakness of receiver-initiated protocols in general and
A-MAC in particular. However, we do not coordinate the
probe phases in this experiment, so the delivery ratios may
improve if the probe times were better scheduled or evenly
distributed, perhaps using a protocol like DESYNC [8].
12

Page 13

0
1
0.25
0.5
0.75
1
Trace 1
0
1
0.25
0.5
0.75
Trace 2
00.05 0.1
Duty Cycle
0.15 0.20.25 0.3
0
0.25
0.5
0.75
Trace 3


LPL
A−MAC
(a) CDF of CTP Duty Cycles
12 14 16 18 20 22 24
0
Time (hour of day)
5
10
Average Power (mW)


A−MAC
RI−MAC
LPL
(b) Interference Effects (Ch. 18)
12 14 16 18 20 22 24
0
Time (hour of day)
5
10
Average Power (mW)


A−MAC
RI−MAC
LPL
(c) Interference Effects (Ch. 26)
0510 15 20
0
0.2
0.4
0.6
0.8
1
Number of additional nodes in cell
Packet delivery rate


32ms
64ms
128ms
256ms
(d) Effect of Density
Figure 12. Macrobenchmarks and statistics for (a) collection routing duty cycles, (b) and (c) power draw vs. extant
environmental interference (co-channel communications), and (d) effect of neighborhood on packet delivery ratios.
7 Discussion
In this section, we outline how future radio hardware
could improve A-MAC performance and discuss some of the
limitations that are fundamental to this design.
7.1 Future Hardware Support
A handful of radio enhancements could improve the per-
formance and energy efficiency of both the backcast prim-
itive and the link layer services multiplexed above it. The
mainbottlenecksinourcurrentdesignoccurfromthelimited
processor-radio bandwidth. Since backcast-based communi-
cations requires multiple loads and unloads of the transmit
and receive FIFOs, respectively, they are often the critical
path operations that occur over a slow serial bus. If either
hardware support for backcast existed inside the radio, or
A-MAC was implemented in a processor with a memory-
mapped radio [24, 35], the loads and unloads could be made
more efficient. Based on Figure 9, we estimate that simple
hardware support would reduce the probe energy cost from
263 µJ to 148 µJ, reducing idle listening power by 40%.
Under the current unicast design, a sender S sets it lo-
cal address to R+0x8000, where R is the receiver’s hardware
address. As a result, S cannot concurrently acknowledge
probes from a different receiver, R′, for which it also has
pending traffic. Richer support for hardware address recog-
nition in the radio would allow a sender to multiplex listen-
ing for a probe. For example, a radio could filter for multiple
source or destination addresses in parallel. Some radios, like
the TI CC2520 [36], can already filter frames on up to twelve
different source addresses but these frames must be sent to a
unicast destination address (meaning some of the approaches
outlined in this paper will not benefit). More flexible ad-
dress recognition and auto-ack support would greatly reduce
the processor burden and offer better efficiency than modern
LPL protocols.
7.2Limitations
There are two fundamental limitations to A-MAC. First,
since A-MAC is a receiver-initiated protocol, the channel
must be probed periodically.
damentally less channel efficient under no-data conditions
than sender-initiated protocols that listen quietly when no
traffic is present. In other words, A-MAC channel usage
scales with neighbor density and not necessarily with traf-
fic. Hence, A-MAC may be incompatible with networks
that have high node density, short communication latency,
or low-probability of detection requirements.
This makes A-MAC fun-
The first two issues are partly addressed by using a differ-
ent channel for the probes and auto-acks than for the actual
data transmissions, since the initial probes can be sent on a
control or pilot channel. The latter issue is more severe: A-
MAC is fundamentally at odds with stealthy networks since
nodes cannot just listen quietly.
The second fundamental limitation with this approach is
that A-MAC’s primitive operation, a channel probe, is in-
herently more expensive than the channel sample primitive
in sender-initiated protocols. Sending a probe frame and
listening for an acknowledgment will always require more
time than sampling the channel. However, the benefits of us-
ing backcast, namely its fixed energy cost, low false alarm
rate, and efficient multiplexing ability, underscore a famil-
iar theme in systems and networking research: optimal solu-
tions that work well over a narrow range often perform more
poorly over the diversity of workloads observed in practice.
8Conclusion
Optimizing performance for a narrow range of operat-
ing conditions or isolated performance metrics is often rela-
tively straightforward. For example, designing protocols that
achieve low power or high throughput under ideal conditions
is easy. It is more difficult to find general solutions that work
well across a broad spectrum of workloads and externalities.
In this paper, we present A-MAC, the first receiver-
initiated link layer that concurrently supports unicast, broad-
cast, wakeup, andpollcastservices. Despiteitsgenerality, A-
MAC achieves high channel efficiency, is resilient to a wide
range of external interference and noise, offers high packet
delivery ratios across a wide range of workloads including
n-to-1 incast and multiple parallel flows, offers lower power
than prior receiver-initiated protocols, and leverages multi-
channel optimizations. We achieve these results using ex-
isting radios in novel ways, but we note that performance
would improve with even a modicum of hardware support.
This work establishes that there is still plenty of room at
the MAC layer to improve duty cycles, achieve predictable
operation, offer high channel efficiency, and provide better
support for bursty workloads. This work paves the way for
new research in the design of radio hardware, MAC sub-
layer primitives, MAC-layer services, and performance stud-
ies to assess the utility and performance of this approach for
emerging needs, like the 802.15.4(e) working group’s search
for a low-power, channel efficient, asynchronous link layer.
13

Page 14

9 Acknowledgments
Special thanks to R˘ azvan Mus˘ aloiu-E. for help in de-
veloping and evaluating a preliminary A-MAC prototype,
Steven Lanszisera for help with modulation schemes and ra-
dio receiver architectures, Rabin Patra for help with wireless
channel emulation, Intel Labs Berkeley for allowing us ac-
cess to their laboratory, the anonymous reviewers for their
insightful feedback, and Alberto Cerpa for shepherding this
paper. This material is based upon work partially supported
by the National Science Foundation under grants #0964120,
#0435454, #0454432, #0546648, #0834470, and #0627611,
as well as a Microsoft Research Graduate Fellowship.
10
[1] J. Arnbak and W. van Blitterswijk. Capacity of slotted ALOHA in
rayleigh-fading channels. IEEE Journal on Selected Areas in Communications,
5(2):261–269, Feb 1987.
References
[2] Association of Radio Industries and Businesses (ARIB). ARIB STD-T67:
Telemeter, Telecontrol, and Data Transmission Radio Equipment for Specified
Low-Power Radio Station, Version 1.1. ARIB STD-T67, 2005.
[3] Atmel. AT86RF230. Available at: http:
//www.atmel.com/dyn/products/product_card.asp?part_id=3941.
[4] C. A. Boano, T. Voigt, N. Tsiftes, L. Mottola, K. Roemer, and M. A. Zuniga.
Making sensornet MAC protocols robust against interference. In EWSN’10:
Proceedings of the 7thEuropean Conference on Wireless Sensor Networks, Feb.
2010.
[5] M. Buettner, G. Yee, E. Anderson, and R. Han. X-MAC: A short preamble
MAC protocol for duty-cycled wireless sensor networks. In Sensys’06:
Proceedings of the 4thInternational Conference on Embedded Networked
Sensor Systems, Nov. 2006.
[6] D. P. Connors and G. J. Pottie. Response Initiated Multiple Access (RIMA), a
Medium Access Control protocol for satellite channels. In GLOBECOM’00:
Proceedings of the IEEE Global Telecommunications Conference, 2000.
[7] Crossbow. Wireless Module - IRIS 2.4GHz. Available at:
http://www.xbow.com/Products/productdetails.aspx?sid=264.
[8] J. Degesys, I. Rose, A. Patel, and R. Nagpal. DESYNC: self-organizing
desynchronization and tdma on wireless sensor networks. In IPSN ’07:
Proceedings of the 6thInternational Conference on Information Processing in
Sensor Networks, pages 11–20, Apr. 2007.
[9] M. Demirbas, O. Soysal, and M. Hussain. A singlehop collaborative feedback
primitive for wireless sensor networks. In INFOCOM’08: Proceedings of the
27thConference on Computer Communications, Apr. 2008.
[10] A. Dutta, D. Saha, D. Grunwald, and D. Sicker. SMACK: a SMart
ACKnowledgment scheme for broadcast messages in wireless networks. In
SIGCOMM’09: Proceedings of the ACM Conference on Data Communication,
pages 15–26, Aug. 2009.
[11] P. Dutta and D. Culler. Practical asynchronous neighbor discovery and
rendezvous for mobile sensing applications. In SenSys ’08: Proceedings of the
6thInternational Conference on Embedded Networked Sensor Systems, pages
71–84, Nov. 2008.
[12] P. Dutta, D. Culler, and S. Shenker. Procrastination Might Lead to a Longer and
More Useful Life. In HotNets-VI: Proceedings of the 6thWorkshop on Hot
Topics in Networks, Nov. 2007.
[13] P. Dutta, R. Mus˘ aloiu-E., I. Stoica, and A. Terzis. Wireless ACK collisions not
considered harmful. In HotNets-VII: Proceedings of the 7thWorkshop on Hot
Topics in Networks, Oct. 2008.
[14] P. Dutta, J. Taneja, J. Jeong, X. Jiang, and D. E. Culler. A building block
approach to sensornet systems. In SenSys’08: Proceedings of the 6th
International Conference on Embedded Networked Sensor Systems, pages
267–280, Nov. 2008.
[15] A. El-Hoiydi and J.-D. Decotignie. Low power downlink MAC protocols for
infrastructure wireless sensor networks. Mobile Networks and Applications,
10(5):675–690, 2005.
[16] R. Fonseca, P. Dutta, P. Levis, and I. Stoica. Quanto: Tracking energy in
networked embedded systems. In OSDI’08: Proceedings of the 8thUSENIX
Symposium on Operating Systems Design and Implementation, pages 323–338,
Dec. 2008.
[17] J. J. Garcia-Luna-Aceves and A. Tzamaloukas. Reversing the
collision-avoidance handshake in wireless networks. In MobiCom ’99:
Proceedings of the 5thInternational Conference on Mobile Computing and
Networking, pages 120–131, Aug. 1999.
[18] O. Gnawali, R. Fonseca, K. Jamieson, D. Moss, and P. Levis. Collection Tree
Protocol. In SenSys’09: Proceedings of the 7thInternational Conference on
Embedded Networked Sensor Systems, pages 1–14, Nov. 2009.
[19] S. A. Gronomeyer and A. L. McBride. MSK and offset QPSK modulation.
IEEE Transactions on Communications, 24(8), 1976.
[20] J. Hill and D. Culler. Mica: A Wireless Platform for Deeply Embedded
Networks. IEEE Micro, 22(6):12–24, Nov. 2002.
[21] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler, and K. Pister. System
architecture directions for network sensors. In ASPLOS-IX: Proceedings of the
9thInternational Conference on Architectural Support for Programming
Languages and Operating Systems, Nov. 2000.
[22] J. W. Hui and D. E. Culler. IP is dead, long live IP for wireless sensor networks.
In SenSys’08: Proceedings of the 6thInternational Conference on Embedded
Networked Sensor Systems, pages 15–28, Nov. 2008.
[23] IEEE Standard for Information technology – Telecommunications and
information exchange between systems – Local and metropolitan area networks.
Specific requirements – Part 15.4: Wireless Medium Access Control (MAC)
and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area
Networks (LR-WPANs), May 2003.
[24] Jennic. Wireless Microcontrollers: JN5121 and JN513x. Available at
http://www.jennic.com/products/, 2007.
[25] C.-J. M. Liang, R. Musaloiu-E, and A. Terzis. Typhoon: A reliable data
dissemination protocol for wireless sensor networks. In EWSN’08: Proccedings
of the 5thEuropean Conference on Sensor Networks, Jan. 2008.
[26] J. Lu and K. Whitehouse. Flash flooding: Exploiting the capture effect for rapid
flooding in wireless sensor networks. In INFOCOM’09: Proceedings of the
28thConference on Computer Communications, Apr. 2009.
[27] D. Moss. Personal communications, 2010.
[28] R. Mus˘ aloiu-E., C.-J. Liang, and A. Terzis. Koala: Ultra-low power data
retrieval in wireless sensor networks. In IPSN’08: Proceedings of the 7thIntl.
Conference on Information Processing in Sensor Networks, Apr. 2008.
[29] S. Pasupathy. Minimum Shift Keying: A spectrally efficient modulation. IEEE
Communications Magazine, 1979.
[30] J. Polastre, J. Hill, and D. Culler. Versatile low power media access for wireless
sensor networks. In Sensys’04: Proceedings of the 2ndInternational
Conference on Embedded Networked Sensor Systems, Nov. 2004.
[31] J. Polastre, R. Szewczyk, and D. Culler. Telos: Enabling ultra-low power
wireless research. In IPSN’05: Proceedings of the 4thInternational Conference
on Information Processing in Sensor Networks, Apr. 2005.
[32] Y. Sun, O. Gurewitz, S. Du, L. Tang, and D. B. Johnson. ADB: an efficient
multihop broadcast protocol based on asynchronous duty-cycling in wireless
sensor networks. In Sensys’09: Proceedings of the 7thInternational Conference
on Embedded Networked Sensor Systems, pages 43–56, Nov. 2009.
[33] Y. Sun, O. Gurewitz, and D. B. Johnson. RI-MAC: a receiver-initiated
asynchronous duty cycle MAC protocol for dynamic traffic loads in wireless
sensor networks. In SenSys’08: Proceedings of the 6thth International
Conference on Embedded Networked Sensor Systems, pages 1–14, Nov. 2008.
[34] Texas Instruments. CC2420: 2.4 GHz IEEE 802.15.4 / ZigBee-ready RF
Transceiver. Available at
http://www.chipcon.com/files/CC2420_Data_Sheet_1_3.pdf, 2006.
[35] Texas Instruments. CC2430: System-on-Chip Solution for 2.4 GHz IEEE
802.15.4 / ZigBee. Available at http://www.ti.com/lit/gpn/cc2430, 2007.
[36] Texas Instruments. CC2520: Second generation 2.4 GHz IEEE 802.15.4 /
ZigBee-ready RF Transceiver. Available at
http://www.ti.com/lit/gpn/cc2520, 2007.
[37] T. van Dam and K. Langendoen. An adaptive energy-efficient MAC protocol for
wireless sensor networks. In SenSys’03: Proceedings of the 1stIntl. Conference
on Embedded Networked Sensor Systems, pages 171–180, Nov. 2003.
[38] W. Ye, J. Heidemann, and D. Estrin. An energy-efficient MAC protocol for
wireless sensor networks. In INFOCOM’02: Proceedings of the 21st
Conference on Computer Communications, June 2002.
[39] W. Ye, F. Silva, and J. Heidemann. Ultra-low duty cycle MAC with scheduled
channel polling. In SenSys ’06: Proceedings of the 4thInternational Conference
on Embedded Networked Sensor Systems, pages 321–334, Nov. 2006.
14