Pulse Switching: Toward a Packet-Less Protocol Paradigm for Event Sensing

Pulse Switching: Toward a Packet-Less
Protocol Paradigm for Event Sensing
Qiong Huo, Student Member, IEEE, Jayanthi Rao, Member, IEEE, and
Subir Biswas, Senior Member, IEEE
Abstract—This paper presents a novel pulse switching protocol framework for ultra light-weight wireless network applications. The key
idea is to abstract a single Ultra Wide Band (UWB) pulse as the information switching granularity. Pulse switching is shown to be
sufficient for on-off style event monitoring applications for which a monitored parameter can be modeled using a binary variable.
Monitoring such events with conventional packet transport can be prohibitively energy-inefficient due to the communication,
processing, and buffering overheads of the large number of bits within a packet’s data, header, and preambles for synchronization. The
paper presents a joint MAC-routing protocol architecture for pulse switching with a novel hop-angular event localization strategy.
Through analytical modeling and simulation-based experiments it is shown that pulse switching can be an effective means for event
networking, which can potentially replace the traditional packet transport when the information to be transported is binary in nature.
Index Terms—Impulse radio, pulse switching, ultra wide band, sensor network, event monitoring, pulse routing
Ç
1 INTRODUCTION
THE objective of this paper is to develop an ultralight pulse
switching protocol framework for resource-constrained
sensors in on-off style event monitoring applications [1]. The
key idea is to introduce a new abstraction of pulse switching
in order to replace the traditional packet switching for event
monitoring. An example application is intrusion detection in
which while surveying a building, it may be sufficient for a
sensor to generate an event to indicate an intrusion in its
vicinity. Sending an event, indicating an intrusion, to a sink
would require single bit information transport. For that, the
conventional packet paradigm can be energy inefficient
due to the communication, processing, and buffering overheads
of a large number of bits within the payload, header,
and the preamble for synchronization [3] for each packet.
In the proposed pulse switching paradigm, such an event
can be coded as a single pulse, which is then transported
multihop while preserving the event’s localization information.
The resulting operational lightness, leveraged via zero
collision, zero buffering, no addressing, no packet processing,
and ultralow communication and idling energy
budgets makes the protocol applicable for severely resource-
constrained sensor devices such as Radio Frequency
Identifiers (RFIDs) operating with tight energy budgets,
possibly from harvested energy [4].
The primary challenges for pulse networking are: 1) how
to transport localization information using a single pulse,
2) how to route a pulse multihop without being able to
explicitly code any information within the pulse, and 3) how
to cope with pulse loss and false-positive detection errors.
Ultra Wide Band (UWB) Impulse Radio (IR) technology
is used for implementing the abstraction of single pulse
transport. A key architectural novelty in this work is to
integrate a pulses’ (i.e., event’s) location of origin within the
MAC-routing protocol syntaxes. More specifically, by
observing the time of arrival of a pulse with respect to the
MAC-routing frame, a sink can resolve the corresponding
event location with a preset resolution. The problem for
multihop pulse routing is addressed by introducing a novel
wave front routing protocol. Synchronized pulse waves are
created in the network so that a pulse can simply “ride”
synchronized phase waves across different hop-distance nodes
from a sink in order to get delivered to the sink. The paper
explores architectural solutions to address those three
fundamental protocol challenges for pulse switching.
The contributions of the paper are:
1. a new pulse-switching protocol paradigm and its
associated MAC and routing syntaxes for multihop
operations,
2. a hop-angular framework for event localization,
3. an implementation approach using Ultra Wide Band
IR, and finally
4. an analytical and simulation framework for performance
characterization and comparison with packetbased
event monitoring in the presence of various
types of pulse loss and pulse detection errors.
The proposed pulse switching architecture is targeted
mainly to small sensor networks with few tens of sensors
distributed within a restricted geographical area. As a
result, a number of proposed protocols may not scale well
for large networks. However, the protocols can enable
targeted event monitoring applications such as intrusion
detection and certain Structural Health Monitoring (SHM)
[2] for aircraft wings, bridges, and other small structures.
IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 12, NO. 1, JANUARY 2013 35
. Q. Huo and S. Biswas are with the Networked Embedded and Wireless
Systems Laboratory, Electrical and Computer Engineering Department,
Michigan State University, Engineering Building, East Lansing, MI
48824. E-mail: sbiswas@egr.msu.edu, qionghmsu@gmail.com.
. J. Rao is with Ford Research and Advanced Engineering, Ford Motor
Company, 2101 Village Road, Dearborn, MI 48121.
E-mail: jrao1@ford.com.
Manuscript received 27 Aug. 2010; revised 24 Sept. 2011; accepted 6 Oct.
2011; published online 20 Oct. 2011.
For information on obtaining reprints of this article, please send e-mail to:
tmc@computer.org, and reference IEEECS Log Number TMC-2010-08-0404.
Digital Object Identifier no. 10.1109/TMC.2011.234.
1536-1233/13/$31.00  2013 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
The paper is organized as follows: Section 2 provides
background and existing work related to pulse switching.
Section 3 introduces UWB physical layer issues relevant to
this work. Section 4 presents the network and the application
abstraction. Section 5 details the proposed pulse
switching architecture. Section 6 presents the impacts of
physical layer node cooperation and mechanisms to
mitigate its impacts on pulse switching. Section 7 describes
the specific measures adopted for energy conservation.
Section 8 analytically investigates error modeling. Section 9
analyzes delay and energy of pulse switching compared
with packet-based solutions. Section 10 further evaluates
pulse switching performance, and finally Section 11
summarizes the paper.
2 RELATED WORK
Majority of the event monitoring solutions in the literature
use traditional packet switching. Packet aggregation [5] is
a natural and effective approach to reduce synchronization
preamble and header overheads by aggregating the
payloads from multiple short packets into a single large
packet that is routed to a monitoring sink node. While
being able to cut the energy costs, aggregation still
requires the inherent packet overheads at the originating
nodes for the short packets, and then end-to-end for fewer
numbers of the large packets. The objective of our work is
to develop protocols for fully eliminating such overheads
required due to packet abstraction.
The mechanism in [6] partially mitigates the packet
switching overhead by sending two very shorts packets for
a sensed event, so that the arrival delay between those short
packets represent a value corresponding to the event. The
short packets serve only the purpose of start/stop delimitation
and do not carry any data. Although there are a
number of practical challenges as outlined in [6], it is an
innovative approach for partially eliminating the need for
packet PDU related overheads. The short control packets,
including start, stop, and intermediate, however, require
substantial bit overheads contributed by packet headers
with node addresses and the per-packet preambles. The
objective of our work is to develop pulse-based protocols
for fully eliminating such overheads.
Communication energy cost for pulse switching can
be significantly smaller than those for packets due to the
difference in the number of bits to be transported. Also, the
processing and buffering costs of packets can be avoided
using pulses.
The authors in [7] develop models for comparative
energy and delay bounds for bit (i.e., packet-based) and
pulse communications in single hop network scenarios.
The main results in [7] are to demonstrate that the worst
case energy performance of pulse communication can be
substantially better than that of packet-based communication,
although with a possibly worse delay performance. A
notable limitation is that it does not provide mechanisms for
scaling these results for multihop networks. Also, no
protocol details are provided for MAC and routing syntaxes
that would be needed for a practical implementation.
Routing a pulse multihop can be particularly challenging
given that no explicitly coded information can be carried in
a single pulse. The objective of this paper is to design a
MAC-routing framework that can be used for practical
implementations of a pulse-based communication paradigm
working in multihop environments.
The paper in [8] proposes a MAC protocol that utilizes
out-of-band contention pulses for packet collision detection.
Unlike our solution, pulses in [8] are of varying length,
rendering technologies such as UWB-IR unusable. Additionally,
although pulses are used for handling collisions,
traditional packets are still used for sending information.
Therefore, the PDU related overheads of packet switching
are presented in the solution in [8].
Idling energy reduction in synchronous packet-based
MAC protocols such as T-MAC [9] is accomplished via
interface sleeping in appropriately scheduled packet slots.
Idling in asynchronous protocols such as B-MAC [10] is
reduced by relying on low power listening, also called
preamble sampling, to link together a sender to a receiver
that is duty cycling. Hybrid protocols also exist that
combines a synchronized protocol like T-MAC with asynchronous
low-power listening [11]. Distributed TDMA
protocols [12] avoid idling consumption by turning interface
off in all packet slots except when needed for transmissions
and receptions. Joint MAC scheduling and route computation
is proposed in [13] for delay and energy optimization for
event monitoring applications. In the cross-layer approach
in [1] it is shown how reporting delay can be optimized in
the presence of predefined sleep-wake MAC cycles.
Although the above MAC, routing and cross-layer
solutions can improve idling energy expenditure in low
duty-cycle networks, they still use packet switching, thus
suffering from the overheads that a pulse-based system can
avoid. It will be shown in this paper that by sending a single
pulse, instead of a packet, the idling energy expenditure
[14] can be significantly reduced.
3 PHYSICAL LAYER WITH UWB IMPULSE RADIO
UWB-IR is used for implementing the abstraction of single
pulse transport. Ultra narrow pulses (i.e., fine time resolution),
carrier-less transmissions, low transmit power, and
low hardware complexity make UWB-IR an ideal tool for
pulse switching. This section presents the usage details of
UWB-IR for pulse switching.
3.1 UWB Slotting for Pulse Switching
The ability to transmit and receive a single pulse without
per-pulse synchronization overhead is a key physical layer
requirement for supporting the frame structure described
above. UWB-IR [15], [16] technology can be used for
implementing such framing because: 1) it can support
single pulse transmission, and 2) the technology is mature
enough for practical system [17] implementation. The top
graph in Fig. 1 depicts a UWB implementation of the
required slot structure in Fig. 4. A typical UWB pulse width
is 1 ns, and the pulse repetition period Tb is 1,000 ns [15],
which determines the slot size in this case. This large
difference between the pulse width and the slot size
minimizes the overlapping probability between pulses in
adjacent slots in the presence of multipath delay [18].
36 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 12, NO. 1, JANUARY 2013
Multiband Orthogonal Frequency-Division multiplexing
(MB-OFDM) [19] is an alternative to IR for UWB implementation.
Unlike IR which uses very short pulses with
relatively low energy, in MB-OFDM the UWB frequency
spectrum is divided into multiple nonoverlapping bands.
Within each band, OFDM-based transmission is used.
While the MB-OFDM based approach can be also used for
the proposed system, we chose IR because of its implementation
simplicity (i.e., due to the lack of mixers in the IR
RF hardware).
3.2 Modulation and Synchronization
Unlike in packet-based UWB, the pulses in different slots
are not correlated among themselves. Instead, each individual
pulse carries information about an event by itself. This
is why the usage of impulses in this architecture does not
require any modulation scheme such as the UWB Pulse
Position Modulation (PPM) which is commonly used for
packet transport over UWB-IR.
For packet transport, nodes use pseudorandom Time
Hopping Sequence (THS) [16] for implementing PPM. This
requires a large preamble bit sequence for synchronizing
the THS phase sequence between the transmitter-receiver
pairs on a per-packet basis. Such synchronization overhead
can be up to 512 to 600 pulse repletion periods or bit
durations as reported in [20]. For pulse communication in
this architecture, since the synchronization is achieved
through the sink (see Section 4.2), no per-pulse synchronization
preambles are needed.
3.3 Energy Budget
The simple all-digital baseband operation of the UWB
impulse radio enables it to be implemented in low-cost
CMOS logic [17]. For example, with 0:18 m CMOS-based
UWB, power consumptions of 4 nJ for each pulse
transmission (average 4 mW with 1,000 ns pulse repetition
period), 8 nJ for each pulse reception (average 8 mW) and
idling consumption of 8 mWcan be typical [17]. It should be
noted that the transmission of a pulse using the baseband
UWB would have the exact same energy expenditure for a
bit transmission using THS-based PPM modulation during
packet transmissions.
3.4 Sources of Error
The primary sources of errors for UWB-IR are large
multipath delay spread (see the bottom graph in Fig. 1),
noise and interference. Large multipath delay spreads can
cause overlapping pulses, leading to errors in pulse
detection. As shown in Fig. 1, very large slot size
(i.e., 1,000 ns) can be chosen in comparison to the pulse
width (i.e., 1 ns) for minimizing the overlapping by
constraining the multipath delay spread within a slot. Pulse
losses due to the misdetection of overlapped pulses need to
be dealt with in the proposed pulse switching architecture.
The other primary contributors to error are interferences
including multiuser interference (MUI) and narrow-band
interference (NBI). MUI can occur when the pulse switching
system coexists with other spread-spectrum users in a UWB
system, and NBI can occur when pulse switching would
coexist with signals from conventional communication
systems. The effects of NBI can be reduced significantly
by using a rejection filter [21]. Although the effect of MUI in
UWB-IR networks is generally less harmful than in narrowband
networks, MUI can still degrade performance by
creating the near-far effect. The approach in [22] proposes a
receiver to mitigate MUI by a combination of statistical
interference modeling and thresholding. The above factors
can result [23] in false positive and/or pulse loss errors.
Measures to minimize the impacts of both of these errors
are presented in Section 8.
4 APPLICATION AND NETWORK MODEL
This section first introduces the application and network
model, and then describes a hop-angular event localization
which forms the basis for the Pulse Switching Architecture
presented in Section 5.
4.1 Application Model
Pulse networking can cater to on-off style monitoring for
structural health, intrusions, and disasters—all for generating
events when specific parameters of interest cross
predefined thresholds. An event results in a pulse which is
transported multihop to a sink. A pulse is able to represent:
1) the very occurrence of the event, and 2) its location of
origin. As investigated in [6], even with such limited
information, several application level conclusions can be
derived at the sink by correlating multiple event pulses. For
example, while monitoring a bridge, by correlating the time
and approximate location of a fatigue event it is possible to
study the structural failure dynamics. Similarly, by fusing
pulses from different intrusion events, the trajectory and
speed of an intrusion can be inferred. The success of pulse
switching hinges upon event localization using a single
pulse transported to a sink.
HUO ET AL.: PULSE SWITCHING: TOWARD A PACKET-LESS PROTOCOL PARADIGM FOR EVENT SENSING 37
Fig. 1. Pulse switching with unmodulated UWB impulses.
Fig. 2. Network model with hop-angular event localization.
4.2 Network Model
As shown in Fig. 2, a network contains arbitrarily
distributed sensors that send pulses to a sink. Depending
on the node locations and the transmission range (assumed
to be nonuniform), each node resides at a certain hopdistance
from the sink. In Fig. 2, the hop-distance for each
node is marked by the node.
The sink is assumed to be capable of making high-power
transmissions with full network coverage for frame-synchronizing
the sensors. For small networks with few tens of
sensors distributed in a restricted geographic area, such
reach-ability is achievable. This is especially true with highpower
pulse generation techniques described in [24]. Since
the proposed architecture is targeted to small networks (see
Sections 1 and 4), full network coverage by the sink is a
valid assumption.
4.3 Hop-Angular Event Localization
A concept of hop-angular event area is introduced for event
localization. The network is logically divided into a fixed
number of angular sectors. In Fig. 2, for example, there are
16 sectors that are 22.5 degrees wide. With a predefined
sector-width (), the location of a sensor can be represented
by the tuple {sector-id, hop-distance}. For example, the
location of the encircled sensor in Fig. 2 can be represented
as {15, 3}, meaning the node is located in the 15th sector,
with a hop-distance 3 from the sink. While the angle for a
node is preprogrammed at the deployment time, its hopdistance
can be dynamically discovered using the process
outlined in Section 5.2. The {sector-id, hop-distance} tuple
indicates an event-area, whose size determines the event
localization resolution. This tuple for an event’s origin is
carried to the sink by the corresponding pulse.
With known transmission range (between the parameters
Rmin and Rmax) and the sector-width , the sink
can estimate the event-area using the {sector-id, hop-distance}
tuple. Higher angular resolution (i.e., smaller ) and smaller
transmission range result in smaller event areas, leading to
better localization resolution. Note that multiple simultaneous
events from the same event-area will be resolved
using the same {sector-id, hop-distance} tuple.
Consider the example event-area identified by the
tuple {; h} in the top portion of Fig. 3. With a sectorwidth
of , and Rmin, Rmax representing the known
minimum and maximum wireless transmission range, the
most conservative (coarse) localization resolution can be
expressed as the largest possible event area: Aconservative ¼
fh2R2
max  ðh  1Þ2R2
ming=360. The average resolution is
Aavearge ¼ fh2R2  ðh  1Þ2R2g=360;
where R ¼ ðRmin þ RmaxÞ=2. For example, with a transmission
range spanning between 1 to 1.5 m, in a network with a
sector-width (i.e., ) of 10 degrees, the size of an event-area
that is 5 hop-distance away is approximately 3.5 square
meters. For the intrusion detection application, this means
that an intrusion can be localized within an area of approximately
3.5 square meters. For a given  and transmission
range, since this resolution reduces with higher hopdistances,
the maximum network size will have an upper
bound for a desired target resolution for event localization.
5 PULSE SWITCHING ARCHITECTURE
This section presents architectural details and their interworking
for the proposed pulse switching architecture.
5.1 Joint MAC-Routing Frame Structure
Nodes in the proposed system are frame-by-frame time
synchronized by the sink, and they maintain MAC-Routing
frames (see Fig. 4) in which each slot is used for sending a
single pulse. The slot includes a guard time to accommodate
the cumulative clock-drift during a frame, which can
be very small for RF technology such as UWB-IR, as the
frame size itself can be ultra short (s) for UWB. The
downlink subframe contains a synchronization slot in
which the sink transmits a full power pulse to make all
nodes frame-synchronized. The following two downlink
slots and the reconfiguration part of the uplink control
subframe are used for hop-distance discovery. The reconfiguration
area has (H þ 1) slots, where H is the maximum
hop-distance. The H-slot routing area of the control
subframe is for energy management.
The event subframe contains H slot clusters, each
containing 360= slots, where  corresponds to the sectorwidth.
Each slot within a cluster corresponds to a specific
{sector-id, hop-distance} tuple. Meaning, for each event-area,
represented by {sector-id, hop-distance}, there is a dedicated
slot in the event subframe. An event originating node
transmits a pulse during the dedicated event subframe slot
that corresponds to the {sector-id, hop-distance} of the node’s
event area. While routing the pulse toward the sink, at all
intermediate nodes it is transmitted at the same event
subframe slot that corresponds to the {sector-id, hop-distance}
38 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 12, NO. 1, JANUARY 2013
Fig. 4. Joint MAC-Routing frame structure for multihop pulse switching.
Fig. 3. Hop-distance event localization.
of its event-area of origin. In other words, while being
forwarded, the transmission slot for the pulse at all
intermediate nodes does not change with respect to the
frame. This is how information about the location of origin
of an event is preserved during routing. Upon reception, the
sink can infer the event-area of origin from the {sector-id,
hop-distance} value corresponding to the slot at which the
pulse is received.
5.2 Hop-Distance Self-Discovery
The sink initiates a reconfiguration phase by sending a full
power start-reconfig pulse. It then transmits a regular power
(i.e., as used by other nodes) pulse at the first slot in the
reconfiguration area in the frame. Nodes that receive this
pulse conclude that they are 1 hop-distance away from the
sink. All hop-distance 1 nodes send a pulse in the second
slot of the reconfiguration area during the next few frames.
Nodes receiving these pulses conclude that they are in hopdistance
2. This process continues for Hc frames (c > 1) after
which all nodes are done discovering their individual hopdistances.
The general logic is that if a node receives
reconfiguration pulses from nodes with different hopdistances
(inferred from the slots of receptions), then its
own hop-distance is one more than the smallest hopdistance
node from which a pulse was received. After Hc
frames, the sink ends the reconfiguration process by
sending a full power pulse in the stop-reconfig slot. Although
H frames can be sufficient for the hop-discovery, the factor
c is introduced for redundancy to cope with pulse losses.
Note that the above discovery process and the proposed
architecture in general, do not assume any specific shape
(i.e., circular or otherwise) of a node’s transmission coverage
area. It could be of any arbitrary shape as shown in Fig. 2.
Due to fading and changing network conditions, the
transmission coverage of the nodes and the resulting hopdistances
are expected to change over time. To accommodate
such changes, the above hop-distance discovery process
needs to be periodically executed. Although each such
discovery process for Hc frames will incur certain additional
energy overhead, such long term overhead is expected to be
marginal for stable wireless environments.
5.3 Pulse Forwarding Using Wave Front Routing
When a pulse is transmitted by a node at hop-distance h,
only its neighboring nodes at hop-distance (h  1) need to
forward it toward the sink. In the absence of MAC
addressing, a node cannot determine the hop-distance of a
pulse’s transmitter node. We introduce a wave front routing
in which nodes synchronously transition in a frame by
frame Sleep (S)-Listen (L)-Transmit (T) state cycle that enables
pulse forwarding toward the sink. Nodes with the same
hop-distance cycle in-phase, but those with different hopdistances
remain synchronized but out-of-phase so that
when the hop-distance h nodes transmit, the hop-distance
h-1 nodes listen, and the hop-distance h þ 1 nodes sleep.
This synchronized cycling ensures that pulses transmitted
by nodes in hop-distance h are received by those at hopdistance
h-1, but are ignored by nodes at hop-distance h þ 1.
This creates a wave front that carries pulses closer to the sink
on a frame-by-frame basis.
Immediately after the reconfiguration process is terminated,
a node at hop-distance h decides its state phase by
computing h modulo 3. The outcomes 0, 1, or 2 cause the
node’s state to be initialized as L, T, or S, respectively.
During the subsequent frames, the state machine cycles in
the sequence S-L-T. In all states, a node wakes up at the end
of a frame (see Fig. 4) for receiving frame synchronization
pulse from the sink.
The concept is explained in Fig. 5 using the example
wave front routing of a pulse generated at node C {sector-id:
2, hop-distance: 3}. The maximum hop-distance H is 4, and
with an  of 90 degrees, the maximum number of sectors is
360=90 ¼ 4. The routing area of the control subframe, and
the event subframe are shown in Fig. 5b. When C is in
transmit phase and has an event to send, it sends a pulse in
the corresponding slot {sector-id: 2, hop-distance: 3} in the
event subframe. As shown in Fig. 5c, transfer of the pulse
from C to B occurs during Frame-1. In Frame-2, B forwards
it to A, and finally during Frame-3, the pulse is delivered to
the sink during the same {sector-id: 2, hop-distance: 3} slot. All
three frames used by C, B, and A look the same as that
shown in Fig. 5b.
This pulse forwarding is termed as wave front routing,
because the pulses simply “ride” the synchronized phase
waves across different hop-distance nodes, and get delivered
to the sink. No address-based forwarding is needed. The
buffering need is drastically smaller than that of the packetbased
systems with variable queuing. Also, the routing
depends only on a node’s knowledge of its own hopdistance,
and not on the underlying event localization
mechanism (e.g., hop-angular). Therefore, as long as the hopdistance
information is known, the pulse routing can be
implemented with other event-localization mechanisms.
5.4 Exploiting Route Diversity
Wave front routing ensures that a pulse is forwarded only
across nodes with reducing hop distances. As shown in
Fig 6a, a pulse originated from node E is not forwarded by
nodeDwhich has the same hop-distance 3, but both nodes B
and C forward it to node A, which in turn delivers it to the
sink. Since all transmissions take place at the same slot in the
HUO ET AL.: PULSE SWITCHING: TOWARD A PACKET-LESS PROTOCOL PARADIGM FOR EVENT SENSING 39
Fig. 5. Wave front routing using synchronized transitions.
event subframe, the transmissions from B and C get merged
while being delivered to node A. This phenomenon of
multiple intermediate route segments gives rise to route
diversity, which provides tolerance from errors. For example,
the pulse can be delivered to sink in spite of a failure of
node B or C, or a transmission error across E-B or E-C. More
about such errors and their impacts are analyzed in Section 8.
5.5 Collision-Less Pulse Merging and Aggregation
Collisions between pulses may not necessarily lead to
information loss. In Fig. 6, since the pulse is transmitted by
B and C on the same slot in the event subframe, the
receiver A simply detects RF signals for a merged pulse in
that slot. As long as the RF hardware can detect the
presence of this overlapped pulse, the routing continues. In
fact, this pulse merging and route diversity provides
inherent in-network aggregation for events from the same
event area. Note that the pulse stacking, as shown in Fig. 6,
is just a representation of the fact that multiple pulses are
transmitted by different nodes during the same slot.
5.6 Sector-Constrained Routing
Angle-based filtering can be activated so that the forwarding
of a pulse remains constrained within a predefined
number of sectors around that of its origin. While higher
sector-constraints (i.e., stricter filtering) curtail route diversity
and subsequent pulse duplications leading to better energy
economy, they also reduce the error tolerance due to lack of
pulse duplications.
The extent of sector-constraints during wave front routing
can be parameterized using , which represents the ratio of
the angular resolution  and an angle . The quantity  is the
sector-width beyond which a pulse may not be flooded while
routing. For a given , the minimum and the maximum
values of  are  and 180 degrees, respectively. The
corresponding  values are 1 and 180=. When  is 1,
routing is maximally constrained, indicating the minimum
communication energy consumption, and the maximum
susceptibility to errors due to the minimum route diversity .
The impacts of the sector-constraints are extensively evaluated
in Figs. 11, 12, and 13 in Section 10.2.
Note that pulse switching is general in that it can work
with other event localization mechanisms. For example, the
hop-angular localization abstraction can be replaced by a
generic flat area-coded mechanism in which a sensor field is
divided into K event areas, and each node is preprogrammed
with an area code (1 through K) at deployment
time. The event subframe in Fig. 4 will then contain K slots,
each corresponding to an event area. The sink can map the
event-area-id (implicitly derived from the slot of pulse
reception) to a prebound geographical area. Wave front
routing can work as is.
6 IMPACTS OF PHYSICAL LAYER COOPERATION
Pulse merging, as explained in Section 5.5, can give rise to
an undesired effect of node cooperation [27] when multiple
nodes simultaneously transmit pulses in a slot. Such
simultaneous transmissions can increase the effective
transmission power, thus extending RF transmission range
to more than 1 hop distance. Node cooperation can affect
both hop-distance discovery and pulse forwarding.
6.1 Mitigating Cooperation during Hop-Distance
Discovery
During hop-distance discovery (see Section 5.2), after the
nodes at hop-distance 1 have discovered their own hopdistance,
they send simultaneous discovery pulses in the
same slot in the reconfiguration area. Unintended node
cooperation due to the energy aggregation of all such pulses
can cause them to reach nodes that are beyond hopdistance
2. This can give rise to faulty hop-distance recovery,
leading to possible pulse forwarding failures. Such problems
in hop-distance discovery due to node cooperation can
happen at all hop-distances except hop-distance 1. The
following mechanisms are proposed for minimizing impacts
of node cooperation by reducing the chances of overlapping
pulses during hop-distance discovery.
The number of slots in the reconfiguration area (see
Fig. 4) is increased from H þ 1 to HM þ 1. The first slot is
still allocated to the sink, and the nodes in each hop distance
are allocated a cluster of M slots instead of a single slot.
Additionally, a slot is functionally divided into Np pulse
positions or minislots. The hop-discovery process in Section
5.2 is augmented with the following new rule. A node
in the hop distance h generates a pulse after: 1) randomly
selecting one of M slots allocated for the hth hop-distance,
and 2) randomly selecting one of Np pulse positions within
the slot selected in step 1.
With this rule, the probability of nonoverlapping pulses
generated by Nh nodes in hop distance h can be estimated as
Pdif
h ¼ 1 PNh
k¼2 ðNh
k Þð^pÞkð1  ^pÞNhk, in which the second
term represents the probability of having at least two
overlapping pulses among Nh nodes, and ^p represents the
probability of having exactly one pulse in the slot cluster
(i.e., M slots) allocated for the nodes in hop-distance h.
^p ¼ MNppcð1  pcÞM1psð1  psÞNp1, where pc ¼ 1=M and
ps ¼ 1=Np. For a typical UWB pulse width of 1 ns and pulse
repetition period of 1,000 ns [15], typical value of Np is 1,000,
with which the quantity Pdif
h approaches to 1. It means
that with the augmented hop-distance discovery rule, the
probability of nonoverlapping pulses generated by the
nodes in a hop-distance is near 100 percent. Therefore,
the effects of physical layer node cooperation on hop
discovery can be mostly mitigated.
In very rare occasions when a node may still receive
overlapping pulses (a receiver node cannot detect overlapping
pulses), the following additional rule can further
reduce the possibility of faulty hop-distance discovery. If a
node receives multiple pulses in different hop-distance slots
40 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 12, NO. 1, JANUARY 2013
Fig. 6. Route diversity and pulse merging/aggregation.
(i.e., hop-distances 1; . . . ; h  1) in the reconfiguration area,
the node decides its own hop distance as maxf1; . . . ;
h  1g þ 1 ¼ h. For example, consider the rare occasion in
which two nodes at hop-distance 2 send out overlapping
pulses, which can reach a node with hop-distance 4. This
happens due to node cooperation caused by energy
aggregation of the overlapping pulses. Without such
cooperation, the receiver node in question should have
received discovery pulses only from nodes in hop-distance
3 and not hop-distance 2. With cooperation, however, the
receiver node ends up receiving pulses from nodes in hoparea
2 as well as correct discovery pulses from hop-area 3.
According to the max-rule, the node interprets its own hopdistance
as 4 which is one more than the maximum of 2 and
3. This way, impacts of node cooperation is fully mitigated
during the hop-distance discovery.
6.2 Immunity to Node Cooperation during Pulse
Forwarding
Wave front pulse forwarding, as proposed in Section 5.3, is
inherently immune to node cooperation due to its hopdistance
synchronized state transitions. Consider a cooperation
situation in which multiple nodes in hop-distance h þ 1
(at state T) simultaneously forward a pulse for an event that
was generated at a higher hop-distance event area. During
this transmission, nodes in h, h  1, and h  2, hop-distances
are in states L, S, and T, respectively. This means, due to
node cooperation even if the pulse reaches areas with hopdistances
h  1 and h  2, it will be ignored simply because
the nodes in those areas are not in a listen state. Nodes only
in hop-distance h will receive it, which is intended.
7 MEASURES FOR ENERGY CONSERVATION
Following architectural measures are taken for improving
the energy efficiency of the baseline pulse switching
mechanism.
7.1 Interframe Sleep
Due to the cycling through the sleep-listen-transmit states, a
node is forced to sleep one in every three frames, leading to
a default 33 percent conservation of idling energy. Note that
because of a pipeline effect, this synchronized sleep does
not in fact introduce any additional event delivery delay.
Events from an h hop-distance node take exactly h frames to
be transported to the sink.
7.2 Intraframe Sleep
During a listen or a transmit frame, in addition to the
downlink slots, a node remains awake during the uplink
control subframe slots. However, during the event subframe,
a node can sleep except during the slots it transmits
or expects to receive pulses. During a transmit frame, since a
node knows the {sector-id, hop-distance} information about a
pulse that need to be transmitted, it can simply wake up
before the appropriate slot cluster and sleep after the
transmission. In a listen frame, without the knowledge of
the expected pulses such sleeps are not possible. We
incorporate the following scheme to address this.
Whenever a pulse (primary) is transmitted in the event
subframe, an associated pulse (secondary) is also transmitted
in the corresponding slot (with the same hop distance) within
the routing area of the control subframe. As shown in Figs. 5b
and 6b, since the primary pulse is transmitted in the event
subframe slot {sector-id: 2, hop-distance: 3}, the secondary
pulse is transmitted in the third hop-distance slot within the
routing area of the control subframe. This ensures that a listen
state node is informed about an impending primary pulse
arrival in this frame by listening to the secondary pulse
during the slots in the routing area. This way, the node can
now wake up before the appropriate slot clusters and sleep
after the corresponding primary pulse is received. Such sleep
reduces idling consumption significantly.
7.3 Delay-Traded Sleep
A third type of sleep mechanism can be used for reducing
idling energy consumption by inserting additional sleep
frames in between the regular sleep, listen, and transmit
frames. As shown in Fig. 7, two additional sleep frames,
marked as D, are inserted after each S, L, and T frame.
These D frames, which behave same as the S frames, do not
affect the nodes’ ability for synchronized state cycling, thus
allowing the wave front routing to continue. D frames
provide a tunable mechanism for reducing idling consumption
at the expense of additional pulse transportation delay.
Using the D frames, delay can be scaled up by a constant
factor k, which is one more than the number of inserted D
frames. In Fig. 7, k is 3, meaning that the delay is scaled up
by three times and the idling energy is scaled down by 3.
With this arrangement, a pulse may now have to remain
buffered at the origin or at an intermediate node before it
can be routed.
8 ERROR HANDLING MEASURES
Functional and performance impacts of UWB pulse errors,
as outlined in Section 3.4, are analyzed in this section.
8.1 Immunity from Pulse Loss
Pulse losses can manifest in the form of unreported events.
Such effects, however, can be alleviated by exploiting the
route diversity inherent to pulse routing. As shown in Fig. 6
for a linear network and in Fig. 11 for a lattice network, a
pulse initially fans out due to the duplicative nature of the
wave front routing, and then converges as it nears the sink.
The degree of fan out and the resulting route diversity
depend on the network topology and the routing sector
constraint . This route diversity due to pulse fan out can
make the architecture partially tolerant to pulse losses
within the fan-out areas. For example, in Fig. 6, the event
can be reported to the sink in spite of a pulse loss across
hops E-to-B or E-to-C. Since with lower sector constraints
(i.e., lower ) the fan out is higher, the resulting higher route
diversity is expected to make the system more tolerant to
pulse losses. An event is more likely to remain unreported
when pulse losses take place near the origin or the sink,
since the route diversity is the minimum in those locations.
For example, in Fig. 6, the event will not be reported to the
sink if a pulse loss occurs on the A-to-Sink hop.
HUO ET AL.: PULSE SWITCHING: TOWARD A PACKET-LESS PROTOCOL PARADIGM FOR EVENT SENSING 41
Fig. 7. Delay-traded sleep for idling energy conservation.
The Pulse Loss Rate (PLR) is defined as the probability
that a pulse is lost due to multipath, channel noise, or
various types of interferences. According to the frame
structure in Fig. 4, in each hop an event is represented by
one pulse in the control subframe and a corresponding
pulse in the event subframe. Loss of any of these two pulses
will lead to the loss of the corresponding event, which will
remain unreported to the sink. Therefore, the probability
(termed as e) of losing an event on any hop on the route of
the event is the same as the probability of losing at least one
of the following two pulses; one at the control subframe and
the other is at the event subframe. This probability can be
expressed as: e ¼ 1  ð1  PLRÞ2 ¼ 2PLR  PLR2. Note
that an event routing may consist of multiple pulses routing
due to the inherent route diversity as explained in Fig. 6.
The following model expresses the relationship between
PLR and the corresponding Event Loss Rate (ELR). Let ni
represent a node on the route of an event and pi represent
the probability that the node ni fails to receive the event due
to pulse losses along the route (including diversity) from
the event source to node ni. Let