Reliability and Acknowledgements in Low-Power

Wireless Sensor Networks
by Author George William Shaw




Research Project

Submitted to the Department of Electrical Engineering and Computer Sciences,
University of California at Berkeley, in partial satisfaction of the requirements for the
degree of Master of Science, Plan II.


Approval for the Report and Comprehensive Examination:



Committee:



Professor Kristofer S. J. Pister
Research Advisor


(Date)



* * * * * * *



Professor Alberto L. Sangiovanni-Vincentelli
Second Reader


(Date)






Contents 
I.  Introduction ................................................................... 1 
A.  Related Work ........................................................... 1 
B.  Reliability ................................................................ 2 
C.  Metrics ..................................................................... 2 
II.  Terms and Assumptions .......................................... 3 
III.  Basic Acknowledgements ........................................ 4 
A.  No Acknowledgement Protocol ............................... 4 
1)  Reliability-Case Tx-Rx Actions .......................... 4 
2)  Mean Tx-Rx Actions ........................................... 4 
3)  Route Congestion ................................................ 4 
4)  Reliability-Case Tx-Rx Events ............................ 4 
5)  Mean Tx-Rx Events ............................................ 4 
B.  Only End-To-End Acknowledgement ..................... 4 
1)  Reliability-Case Tx-Rx Actions .......................... 5 
2)  Mean Tx-Rx Actions ........................................... 5 
3)  Route Congestion ................................................ 5 
4)  Reliability-Case Tx-Rx Events ............................ 5 
5)  Mean Tx-Rx Events ............................................ 5 
C.  Link-Layer Acknowledgement ................................ 6 
D.  Implicit Acknowledgment ....................................... 7 
1)  Reliability-Case Tx-Rx Actions .......................... 7 
2)  Mean Tx-Rx Actions ........................................... 7 
3)  Route Congestion ................................................ 7 
4)  Reliability-Case Tx-Rx Events ............................ 7 
5)  Mean Tx-Rx Events ............................................ 7 
E.  Immediate Acknowledgment ................................... 8 
1)  Reliability-Case Tx-Rx Actions .......................... 8 
2)  Mean Tx-Rx Actions ........................................... 8 
3)  Route Congestion ................................................ 8 
4)  Reliability-Case Tx-Rx Events ............................ 8 
5)  Mean Tx-Rx Events ............................................ 8 
F.  Separate Acknowledgment ...................................... 8 
1)  Reliability-Case Tx-Rx Actions .......................... 8 
2)  Mean Tx-Rx Actions ........................................... 8 
3)  Route Congestion ................................................ 8 
4)  Reliability-Case Tx-Rx Events ............................ 8 
5)  Mean Tx-Rx Events ............................................ 9 
IV.  Comparison of Basic Acknowledgements Protocols9 
A.  Bounding Values ................................................... 10 
B.  Bandwidth .............................................................. 11 
C.  Node and Network Energy .................................... 11 
D.  Resource Usage ..................................................... 12 
E.  Reliability .............................................................. 12 
F.  Latency .................................................................. 13 
V.  Comparison of End-to-End Acknowledgements ... 14 
A.  Extraordinary Events ............................................. 14 
B.  When to Use End-to-End Acknowledgments ........ 14 
C.  End-To-End Acknowledgement with Link-Layer
Acknowledgements ......................................................... 15 
1)  Calculations ....................................................... 15 
2)  Bandwidth ......................................................... 15 
3)  Node and Network Energy. ............................... 15 
4)  Resource Usage ................................................. 15 
5)  Reliability .......................................................... 16 
6)  Latency .............................................................. 16 
VI.  Cumulative or Selective Acknowledgments .......... 16 
VII.  Conclusions............................................................ 16 
References ........................................................................... 33 




1



Abstract— All wireless sensor networks with greater than
0% packet delivery rate (PDR) can be made 100% reliable—
when given unbounded time to achieve successful packet
delivery. Real systems, however, don't have unbounded time or
resources. Reliability is the probability that a packet will be
successfully delivered within the required time bound. The
delivery reliability requirement and time bound are determined
from the application-dependent delivery MTBF at which
exceptional reliability measures can be tolerated. Reliability is
generally achieved through packet delivery acknowledgements,
and conventional wisdom says that reliable operation requires
these acknowledgements be end-to-end. However, reliability
meeting the application requirement can often be achieved
without the overhead of an end-to-end acknowledgment. This
report contributes a fundamental comparison of the common
underlying mechanisms used to acknowledge packet delivery,
their behaviors under changing conditions, and the
environments in which they perform best. It suggests a method
for evaluating actual packet delivery reliability to estimate the
packet delivery rates used for planning the network, and a
method for determining the time bounds that must be
accommodated based on the estimated packet delivery rates so
as to achieve the desired delivery reliability. It finds that by
changing acknowledgment protocols, the reliability of the
system can be increased by several orders of magnitude (to the
limit of the inherent network reliability), that immediate
acknowledgments perform generally best overall, and that if an
end-to-end acknowledgment must be used, implementing it
with a link-layer acknowledgment is the most efficient.
I. INTRODUCTION
LL multi-hop wireless sensor networks (WSN) with
greater than 0% packet delivery rate (PDR) can be
made 100% reliable—when given an unbounded amount of
time to achieve successful packet delivery. Real systems,
however, don’t have unbounded time or resources, so
achieving a reliability of less than 100% must be tolerated.
Thus, reliability is the probability that a packet will be
successfully delivered within the required time bound.
Acknowledgement protocols can significantly affect delivery
reliability, but a desired level of reliability cannot be
“created” by the protocol; the level of reliability can only be
reached or not. Reliability is determined by the packet
delivery rate and the number of successful packet
transmissions the protocol can achieve within the time
bound. Conventional wisdom says that end-to-end
acknowledgements are required for reliable operation. They


have their uses, but their requirement depends on several
specific factors. Using end-to-end acknowledgements adds
network workload that can lower, rather than raise, the
possible time-bounded delivery reliability for the route and
for the network as a whole.

A. Related Work
Most papers look at acknowledgments as part of some
other function, such as routing algorithms or network layer
protocols. The author did not find any references that simply
analyzed the fundamental characteristics of the basic
available mechanisms. In [5], the authors define the metric
ETX (expected transmissions) that uses live data for p
dutu

and p
ACK
to compute expected transmissions at the next hop
and thus select the best route. They look at only one hop,
however. In [17], the authors improve on ETX with MT,
which looks ahead a hop and also considers the variance of
the packet delivery rate in making its routing decision.
While both of these have the live data available, and while
they both attempt to pick a good route, they do not
determine if the path they chose meets any outside criteria,
such as a specific reliability or time bound. Many papers
looks at various mesh routing systems, including braided
diffusion, directed diffusion, disjoint multipath, braided
multipath and others to some quality metric, but none
determine or evaluate time bounded reliability.
We limit our analysis to the fundamental behavior of the
underlying mechanisms on a linear route. Many papers look
at some metric of energy consumption or efficiency, or time-
bounded access to the wireless medium, but not bounded-
time packet delivery. The exception is in [15], the author
uses collected or live packet delivery rate data to evaluate
the reliability of a route of links given time-varying packet
delivery rate information and determines hard time bounds
for packet delivery. We look at the fundamental operators on
the time-bounded reliability of delivery: the work load of a
protocol and the primary factor to which the work load is
sensitive, the packet delivery rate. The conclusions suggest
that by all measures immediate acknowledgments are
generally the most robust and efficient; they are even built
into the hardware of many radios. However, current work
still uses other protocols and ignores the primary driver of
delivery reliability, the packet delivery rate.
Messina in [10] proposes a network-layer protocol called
PERLA that uses implicit acknowledgements rather than
link-layer acknowledgements with packet caching and
Reliability and Acknowledgements in Low-Power
Wireless Sensor Networks
George William Shaw
A
2

retransmission from alternate local motes, and finds a higher
reliability. However, their gains appear to come entirely
from scheduled retransmissions that avoid collisions
compared to the competing protocols for which they do not
consider scheduling. Using scheduled acknowledgement as
in TSCH might alone eliminate that advantage. Further, their
analysis assumes an ideal channel, i.e. PÐR = 1.u, which is
unrealistic for WSNs. A realistic packet delivery rate causes
implicit acknowledgements to start competitively but to
degrade more quickly as the packet delivery rate drops
compared to immediate acknowledgements.
This report contributes a fundamental comparison of the
common underlying mechanisms used to acknowledge
packet delivery, their behaviors under changing conditions,
and the environments in which they perform best. It suggests
the method for evaluating actual packet delivery reliability
from [15] for estimating the packet delivery rates used for
planning the network, and a method for determining the time
bounds that must be accommodated based on the estimated
packet delivery rates so as to achieve a desired delivery
reliability.

B. Reliability
Achieving high reliability is desirable, but high reliability
cannot guarantee delivery, it only makes successful delivery
more likely. A packet can be acknowledged at every hop, yet
not reach the destination due to a failure of the packet to be
forwarded at any hop. An end-to-end acknowledgement is
the only way to guarantee that a packet, or group of packets,
did arrive at the destination, but it is still impossible to
guarantee that this acknowledgement can be received within
the time bound. Thus, the highest reliability solution for a
particular route is generally an efficient packet
acknowledgement protocol to move packets to the
destination and an efficient end-to-end acknowledgement to
verify that they were received at the destination. However,
the last measure of reliability in guaranteeing packet
delivery, the end-to-end acknowledgment, comes at a high
resource cost, can lower the reliability of the network as a
whole, and is frequently not needed to achieve the required
application-level reliability.
The delivery reliability requirement and time bound are
determined from the application-dependent delivery MTBF
at which exceptional reliability measures, such as retrying
the entire packet sending process to as much as human
intervention, can be tolerated. If the application sends one
packet per minute and can tolerate a delivery MTBF of one
delayed or failed delivery per day, then the required delivery
reliability is

R = 1 -
1
6u minutcs - 24 bours
= 99.9S%.

For non-critical packets the time bound may be the
network timeout, which may be as long as several minutes or
more, while for time-critical packets the time bound may be
only a few hundred milliseconds. The time bound directly
affects the delivery reliability since it limits the amount of
work that the network may perform in attempting to deliver
the packet. The work performed is counted in Tx-Rx actions,
and the temporal character of this activity is counted in Tx-
Rx events. Sufficient Tx-Rx events must be accommodated
to achieve the desired reliability R. Only if the specified time
bound is larger than the latency required by that number of
Tx-Rx events can the protocol achieve this reliability. We
refer to values related to this level of work as reliability-case
values. In general, the less work a protocol must perform to
achieve a given reliability, the more reliable the network can
be for a given time bound. In some cases, the value of Tx-Rx
events and Tx-Rx actions are the same. We will use the term
Tx-Rx events when considering latency and otherwise use
the term Tx-Rx actions to refer to activity involved. Using
one term often implies information about the other.
We assume packets transmissions are Bernoulli trials that
fail or succeed with some probability based on the packet
delivery rates. This creates a distribution of the probability
of a given number of successes in a given number of trials
(Tx-Rx actions). The median of the distribution tells us the
number of Tx-Rx actions at which 50% of the total Tx-Rx
actions are successful, or a reliability R of 0.5. The mean of
the distribution tells us the average number of Tx-Rx actions
expected to achieve the given number of successful
receptions. Determining the reliability-case number of Tx-
Rx actions to achieve reliability R is generally a matter of
finding the number of Tx-Rx actions (the point on the
distribution) that the probability is R that the network can
achieve the given number or more of successful receptions.
Thus if R is 0.999, we want to find the number of Tx-Rx
actions to be accommodated to ensure that 99.9% of the time
we will have at least the required number of successful
receptions. Reliability-case values specify the minimum
resources required to achieve the desired reliability and are
used where worst-cast resources must be evaluated, such as
in packet buffers and delivery latency, where exceeding the
value causes the failure of a packet to be delivered within the
required time bound. The mean value, conversely, is useful
for values which are averaged over time, such as goodput or
energy consumption.

C. Metrics
In this analysis we are concerned with motes that are
power-constrained (battery operated), resource constrained,
and latency sensitive. They have limited battery life,
memory, processing power and radio bandwidth. The longer
a mote is awake or the longer the receiver or transmitter is
on, the more energy it consumes. The longer a mote must
hold a packet for possible retransmission, the more memory
it consumes. The more frequently a mote transmits the more
radio bandwidth it consumes. The transmission and
reception of packets is generally the largest resource
consumer on a mote. Reducing the number of Tx-Rx actions
3

can not only increase the packet delivery reliability, but by
reducing the resources used, it can increase the battery life of
the mote, the battery life of the network and the reliability of
the network as a whole. Tx-Rx Actions are similar to
expected transmissions, ETX, described by [5], and to
minimum transmissions, MT, described by [17] as routing
metrics, except they are calculated to greater detail and
rather than using dynamically measured values per link to
plan routes, we are using predicted or statically measured
values for design and planning.
We will primarily compare per-packet acknowledgements
(ACK) protocols. As a group, we will look at protocols that
transmit no acknowledgement, only end-to-end acknowled-
gement, and three link-layer acknowledgements: implicit
acknowledgement, immediate acknowledgement, and
separate acknowledgement. Then, as a group, we will also
look at the case for end-to-end acknowledgments and review
link-layer protocols that include an end-to-end acknowled-
gement per data packet. These selections describe limiting
behavior for other protocols and set the expected bounds of
the system.

II. TERMS AND ASSUMPTIONS
Originate is used to refer to a link-layer transmission of a
packet from the source mote into the network toward the
destination mote that may entail many transmissions en
route. A single packet sent from the network level to the link
level and into the network may originate the same packet
several times if required by the acknowledgement protocol.
Source is the mote that originates a packet and is thus the
first mote in the sequence of motes on the route that transfers
the packet to the destination.
Destination is the last mote in the sequence of motes that
receives a packet that originates at the source.
Route refers to the entire sequence of motes that transfers
a packet from the source mote to the destination mote.
Hop is a counting measure referring to the number of
wireless links required to transfer a packet from one mote,
through neighboring motes, to another mote.
(Re)Transmission refers to data transfers between motes
on the route to the destination. Transmission implies a
receiver will attempt reception.
Reliability R is defined as the probability that a packet
will arrive at its destination within a specified time bound, or
latency.
Packet delivery rate (PDR) is the probability that a packet
will be successfully transferred from one mote to an
immediate neighbor. For the most accurate estimation of Tx-
Rx events, the packet delivery rate should consider CCA
failures and scheduling delays as packet delivery failures
that lower the packer delivery rate.
Acknowledgements (ACKs) are packets transmitted to
confirm the arrival of a data packet and are only transmitted
after the successful reception of a packet (positive
acknowledgements).
Failure is the lack of or incorrect reception of a packet for
any reason, such that the data in the packet cannot be
considered valid by the available measures, or the arrival of
a packet outside of the required time bound.
Reliability-case values are the minimum amount of an
item to be accommodated to achieve the desired reliability
R.
p
dutu
is the packet delivery rate for a data packet.
p
ACK
is the packet delivery rate for an acknowledgement
packet.
Tx-Rx Actions count the occurrences of the operation of a
transmitter and receiver on a pair of motes.
Tx-Rx Events count the periods of transmitter or receiver
activity, regardless of how many transmitters or receivers are
involved. It is a metric to evaluate latency and can be
translated into seconds.

In most cases, one can expect p
uck
> p
dutu
for several
reasons. When sent as an explicit packet, acknowledgement
packets are generally shorter than data packets, so the
contribution to bit error rate due to Gaussian noise is
reduced. If the acknowledgment is transmitted soon after
the data packet then the channel coherence that made the
data packet successful is maintained while CAA has
suppressed the transmissions from local nodes [16]. While
asymmetric packet delivery rates have been reported by
[9],[4] has showed by physically swapping pairs of motes
that asymmetries can be due to reception sensitivities of the
different motes, i.e., inconsistency in the quality of the
receiver hardware. Thus, we assume all packet delivery rates
are symmetric.
We also make several simplifying assumptions:
p
AB
(¡, t) = p
dutu
or p
ACK
and is a constant for all
routes, times, and frequencies.
p
ACK
¸ p
dutu
, as discussed above.
All probability distributions are independent and
identically distributed.
Each node has a single parent and is routed in a linear
topology. Multiple parents in a mesh topology can improve
the packet delivery rate but complicate evaluation of the
resources used and are not treated in this report.
I
hop
, the delay per successful hop, is a constant in the
network and independent of the success or failure of
previous hops or previous attempts to transmit the packet on
this hop. This is consistent with some of the most popular
MAC layers, such as B-MAC, SCP, TSCH, TSMP, but is
not the case with epoch-based duty-cycling with epochs that
are long compared to the length of a packet.
For the graphs in this report, a reliability of R = 0.999 is
used unless labeled otherwise as this represents greater than
one failed delivery per day using the example above of one
transmission per minute. Several metrics are evaluated at a
packet delivery rate of 1.0, 0.9, 0.6, and 0.3 to characterize
the protocols. Trends for other lower and higher reliability
are discussed in the text and in Table 5 and Table 6.

4

III. BASIC ACKNOWLEDGEMENTS
A. No Acknowledgement Protocol
A packet can be reliably sent with a no-acknowledgement
protocol to achieve some reliability R > PÐR by
unconditionally originating the packet the reliability-case
times to achieve a probability of delivery greater than R. If
R < PÐR then the packet need only be originated once.
With its simplicity, the technique might at first be
considered to reduce resource usage, and is commonly used
in single-hop home sensor networks where high reliability is
not a concern.
A single data packet traveling k hops to its destination
with no retransmissions at any hop will arrive with a
probability of p
dutu
k
. However, at a 90% packet delivery
rate and ten hops, the probability that the single packet sent
from the source mote arrives at the destination mote
is p
AB
= u.9
10
÷ SS%. If we originate the packet some
multiple b times, we can increase the delivery probability.
The probability that all packets are lost is the overall error
rate, e.

e = (1 -p
dutu
k
)
b


The probability of at least one successful reception p
AB
,
which is all that is required, is also the reliability R.


R = p
AB
= 1 -e = 1 -(1 -p
dutu
k
)
b


Thus, the packet originations b for reliability R is

b =
log(1 -R)
log(1 -p
dutu
k
)
.

To achieve a reliability R of 0.999 over ten hops
with p
dutu
= u.9, the packet must be originated sixteen
times. Multiple copies of the same packet are originated and
generally received to ensure reliability R is met and only in
about 1 in 1000 cases will just one packet reach the
destination. It is for the assurance of the 1 in 1000 case that
the sixteen packets must be originated in this scenario.
1) Reliability-Case Tx-Rx Actions
An originated packet fails along its route following a
geometric distribution. The probability of x = k Tx-Rx
actions traversing the route is the same distribution plus the
probability of success reaching the last mote, mote n. Since
these probabilities sum to one, the distribution is inherently
normalized. Thus, the probability of x Tx-Rx actions is

p
#tx
(x) = _
p
x-1
· q, x < n
p
x-1
· q +p
x
, x = n
.

The expectation value for Tx-Rx actions for each packet
sent along the route is

EX
NA
(n) = p
#tx
(x) · x
n
x=1
.

For the no-acknowledgement protocol, the reliability-case
Tx-Rx actions to be accommodated to ensure reliability R
for hops k is the packet originations multiplied by the
expected Tx-Rx actions for each packet.

RCIxRxActions = b · EX
NA
(k)

2) Mean Tx-Rx Actions
While the packet will on average be received at the
destination after some smaller number than packet
originations b, with no acknowledgement to confirm
delivery, all b packets must always be transmitted to assure
reliability R is achieved. As a result, the mean packet
originations is also b, so the reliability-case and mean values
are the same. The majority of the packets are assumed to be
lost somewhere in route.

HconIxRxActions = RCIxRxActions

3) Route Congestion
The route congestion is computed using the distribution
produced by p
#tx
, and accumulated towards the source as if
to create the cumulative distribution function (CDF). Since
the source only transmits and the destination only receives,
the CDF is summed to itself shifted one hop toward the
destination and then renormalized. Multiplying by
HconIxRxActions creates the Tx-Rx actions distributed
across the route.
4) Reliability-Case Tx-Rx Events
Since there is no acknowledgement to wait for, packets
can be originated with no delay. Once all b packets have
been transmitted the resources on the source mote are freed
for other activity. However, the time bound for reliability R
requires that the data packet be delivered, so we must
include the time for the last packet to trickle through the
route to the destination. There are many overlapping Tx-Rx
events as the packets move through the network.

RCIxRxE:cnts = b +(k -1)

5) Mean Tx-Rx Events
As with the mean Tx-Rx actions, with no acknowled-
gment to confirm delivery, all b packets are always
transmitted. The reliability-case and mean Tx-Rx events
have the same value.

HconIxRxE:cnts = RCIxRxE:cnts

B. Only End-To-End Acknowledgement
A packet can be reliably sent with an only end-to-end
acknowledgement protocol to achieve some reliability R by
conditionally or unconditionally resending the packet some
5

zero or more times until delivery is verified by reception of
the acknowledgement packet. Conditionally resending the
packet has the advantage that only the required Tx-Rx
actions occur, but suffers in delivery latency because a
failed-acknowledgement timeout must occur before the
packet is originated again. Unconditionally originating the
packet has the advantage of reducing the delivery latency
because the packet can be transmitted without waiting for a
failed-acknowledgement timeout, but suffers from the
possibility of unneeded transmitting of the packet, unneeded
Tx-Rx actions across the network, and a higher probability
of multiple copies of the packet reaching the destination.
When the sender of a data packet requests an end-to-end
acknowledgement then the destination must create and
transmit an acknowledgement packet back through the
network. The end-to-end acknowledgement packet is
independent in scheduling, routing, delivery probability and
transit time from the initiating data packet. The probability
that a data packet originated from source mote A arrives at
destination mote B is p
dutu
k
, and the probability that the
end-to-end acknowledgement packet transmitted from
destination mote B arrives at source mote A is p
ACK
k
¸
p
dutu
k
, for a packet-acknowledgement round-trip proba-
bility of p
ABA
¸ p
dutu
2k
. Using the prior no-acknowled-
gement protocol scenario of a ten-hop distant route and 90%
packet delivery rate, the total hops for the packet-
acknowledgement round-trip is twenty hops. The probability
that the data packet originated from source mote A arrives at
destination mote B is again about 35%, but the probability
that the acknowledgement packet originated from
destination mote B is also received at source mote A is as
low as about 12%. Even if the data packet is received
correctly at destination mote B, there is still a probability
1 -p
dutu
k
(about 65%) that the data packet will be
originated one or more additional times because the
acknowledgement packet did not complete the trip from
destination mote B to source mote A. In general, whether
conditionally or unconditionally originated, multiple copies
of the data packet are received at the destination and by this
measure more packets will be originated than necessary.
The calculation of the reliability-case packet originations
b that must be accommodated to achieve reliability R is the
same as for the no-acknowledgement protocol, except that
for a k hop destination 2k hops must be successfully
traversed since success requires the acknowledgement
packet to also be received. Thus, the packet originations
b for reliability R is

b =
log(1 -R)
log(1 -p
dutu
2k
)
.

The expectation value for Tx-Rx actions is

EX
LL
(n) = p
#tx
(2x) · 2x
n
x=1
.
1) Reliability-Case Tx-Rx Actions
For the only end-to-end acknowledgement protocol, the
number of reliability-case Tx-Rx actions to be accommo-
dated to ensure reliability R for hops k is the packet
originations b multiplied by the expected Tx-Rx actions for
each packet.

RCIxRxActions = b · EX
LL
(k)

2) Mean Tx-Rx Actions
The mean Tx-Rx actions are the mean originations that
occur multiplied by the expected Tx-Rx actions for each.

HconIxRxActions =
1
p
dutu

2k
· EX
LL
(k)

3) Route Congestion
The route congestion is computed using the distribution
produced by p
#tx
for 2k hops and accumulating towards the
source as if to create the cumulative distribution function
(CDF). Since the source only transmits and the destination
only receives, the CDF is summed to itself shifted one hop
to toward the destination and then renormalized. However,
since the returning acknowledgement is assumed to traverse
the same route, the distribution is folded around the
destination to create the backtracking. Then, multiplying by
HconIxRxActions creates the Tx-Rx actions distributed
across the route.
4) Reliability-Case Tx-Rx Events
The protocol can either conditionally or unconditionally
originate a duplicate packet. If the data packet is conditional
originated, then the source must wait for 2k Tx-Rx events to
determine that the end-to-end acknowledgment has not been
received and all reliability-case packet originations b must
be accommodated. There are no retransmissions en route.

RCIxRxE:cnts = b · 2k

If the data packet is unconditionally originated for all of
the reliability-case packet originations, the minimum Tx-Rx
events is that required to transmit all b packets and for the
acknowledgement from the last one to be received at the
source. There are many overlapping Tx-Rx events as the
packets move through the network and the acknowledgments
return.

RCIxRxE:cnts = b +(2k -1)

This value is plotted in Figure 18.
5) Mean Tx-Rx Events
The protocol can either conditionally or unconditionally
originate a duplicate packet. If the data packet is conditional
originated, then the source must wait for 2k Tx-Rx events to
determine that the end-to-end acknowledgment has not been
received for the mean packet originations needed to achieve
success. There are no retransmissions en route.
6

HconIxRxE:cnts =
2k
p
dctc
2k


If the data packet is unconditionally originated, the
minimum Tx-Rx events is that required to transmit the mean
originated packets and for the acknowledgement from last
packet to be received at the source. There are many
overlapping Tx-Rx events as the packets move through the
network and the acknowledgments return.

HconIxRxE:cnts =
1
p
dutu
2k
+(2k -1)

This value is plotted only in Figure 17.

C. Link-Layer Acknowledgement
A data packet can be reliably sent with link-layer
acknowledgements to achieve some reliability R by
conditionally retransmitting the data packet at each mote in
the route that does not receive, within the acknowledgement
timeout, an acknowledgement packet from its neighbor
indicating successful reception. We assume that the data
packet is retransmitted at each mote as many times as
required to achieve successful delivery, though many
systems have limits. Since only one hop is involved for
acknowledgement, the probability of successful data packet
delivery is p
dutu
and the probability of successful data
packet and acknowledgement packet delivery is p
dutu
· p
ACK.

Link-layer acknowledgements thus have an advantage
over no-acknowledgements or only end-to-end acknowled-
gements because only local motes are involved in the
retransmission and thus fewer data packets are retransmitted
due to failed acknowledgement packets. Because the only
redundant packets ever transmitted are when the data packet
has successfully traversed the one hop but the
acknowledgement packet has not successfully returned the
one hop, Tx-Rx actions are reduced with all of the
subsequent advantages. Thus, even with an application-level
verification mechanism, link-layer acknowledgments are
justified because they improve performance [14].
Retries occur at each hop as the transmission succeeds or
fails, and are thus Bernoulli trials, with the trials ending in
success on the last trial at the destination mote B. The
probability of success of the data packet being forwarded
through the network at each hop is p = p
dutu
and is
considered independent of the success of the acknowled-
gement packet. The probability of exactly retries b to
achieve hops k is a negative binomial distribution given by

NB(b; k, p) =
ncboosck(b +k -1, k -1) · p
k
(1 -p)
b
.

The probability p
AB
of sending a packet from source mote
A to destination mote B and incurring up to retries b to
achieve hops k is the cumulative distribution function (CDF)
of the negative binomial distribution.
R = p
AB
= CÐFNB(b; k, p) = NB(i; k, p)
b
ì=0


The probability p
AB
of success on the route from source
mote A to destination mote B is also the reliability R of that
route. We perform an inverse CDF by computing the sum
until the CÐF >= R to compute retries b to be accommo-
dated.

ForworJIxRxActions = b +k.

While this represents the Tx-Rx actions to forward the
data packet through the network, acknowledgement activity
also consumes resources so this number does not tell the
whole story. The data packet is transmitted some number of
additional times because the acknowledgement packet is not
successfully received.
To ensure that the data packet is delivered and that all
acknowledgement activity is complete within the time
bound, the probability of delivery and the computation of the
retries b to be tolerated to achieve hops k with reliability R
must be that of the successful delivery of the data packet and
the successful delivery of the acknowledgement packet. To
determine the Tx-Rx actions, since both a data and an
acknowledgement Tx-Rx action can occur, we must
compute the number of both. We assume that duplicate data

Figure 1. Order of Tx‐Rx events across hops A, B, C, D for 
basic acknowledgment protocols. 
7

packets received due to unsuccessful acknowledgement
packet reception are dropped or do not interfere with timing.
The approximation of the total Tx-Rx actions varies with the
three link-layer protocols below, but is similar in form to:

RCIxRxActions = (b
dutu+ACK
+k) +(b
ACK
+k).

The term (b
dutu+ACK
+k) accounts for the number of
data packet Tx-Rx actions and the term (b
ACK
+k)
accounts for the number of acknowledgement packet Tx-Rx
actions.
The mean Tx-Rx actions for hops k are generally
computed as

HconIxRxActions =
k
p
dctc
·p
ACK
+
k
p
ACK
.

The first term accounts for the data packet Tx-Rx actions.
The second term accounts for the acknowledgement packet
Tx-Rx actions and considers that some acknowledgment
packets are not successful.
Since failures are distributed throughout the route, the
average activity for all motes is spread evenly across the
route. The source and the destination, however, do not
forward the packet, so they generally incur half the activity.

HconRoutcCongcstion
= |o
0
…o
k
] ·
1
∑ o
ì
k
ì=0
· HconIxRxActions

The vector represents the pattern of Tx-Rx actions across
the route while the scalar summation normalizes the
distribution. Multiplying by the mean allocates the Tx-Rx
actions.
The reliability-case and mean Tx-Rx events are computed
similarly to the Tx-Rx actions, but the components may be
grouped differently to represent temporally separated events.

D. Implicit Acknowledgment
The link-layer implicit-acknowledgement protocol
consists of transmitting a data packet one hop from mote A
to mote B, and mote A implying an acknowledgement of
successful delivery of the packet by overhearing mote B
forwarding the packet to the next mote C. This saves
resources required for transmitting an explicit acknowled-
gement packet, except for the last hop since the packet is not
forwarded. However, mote A must store the data to
recognize the forwarded packet, and will generally need to
listen to a data-packet length transmission rather than an
acknowledgement-packet length transmission. This is also
part of the type of acknowledgement structure performed by
systems that piggyback acknowledgments on returning data
packets. While they do not overhear the packet being
forwarded, they also do not transmit an explicit
acknowledgement. Instead, they wait for a return data packet
to carry an acknowledgment.
Since the forwarded packet used for acknowledgement is
a data packet and is independent in scheduling and delivery
probability, we assume p
ACK
= p
dutu
. Thus, the probability
that the data packet transmitted from mote A is received at
mote B and the packet forwarded by mote B to mote C is
received by mote A is

p
dutu+ACK
= p
dutu
· p
ACK
= p
dutu
2
.

1) Reliability-Case Tx-Rx Actions
Each acknowledgement en route adds only one-half Tx-
Rx action for the overheard forwarded packet.

RCIxRxActions ÷ (b
dutu+ACK
+k) +
b
ACK
+k
2


2) Mean Tx-Rx Actions
The mean Tx-Rx actions count the same types of activity
as the reliability-case values, except using mean values, and
with the addition of a value for the explicit acknowledgment
transmitted at the last hop.

HconIxRxActions ÷
k
p
dutu

2
+
1
2
·
k +1
p
dutu


3) Route Congestion
Since the implicit acknowledgment does not transmit an
explicit acknowledgement when a packet is forwarded but
does receive a packet for acknowledgement, middle motes
incur one-half a Tx-Rx action.

HconRoutcCongcstion
= |1 1.S…1.S 1] ·
1
∑ o
ì
k
ì=0
· HconIxRxActions

4) Reliability-Case Tx-Rx Events
Since the acknowledgment occurs at the same time as the
data packet is forwarded, it does not add latency and counts
as the same event except on the last hop, when an explicit
acknowledgment must be transmitted.

RCIxRxE:cnts ÷ (b
dutu+ACK
+k) +1

Networks that piggyback acknowledgments on return data
packets will need an additional factor to account for the
delay until a return data packet occurs. Many also will
transmit a separate acknowledgment packet should a data
packet not occur soon enough, making these networks
hybrids.
5) Mean Tx-Rx Events
The first data packet Tx-Rx event and last explicit
acknowledgement Tx-Rx event are separate, while all the
Tx-Rx actions in the middle are combined data-packet
forwarding/acknowledgment events.

8

HconIxRxE:cnts ÷
k -1
p
dutu

2
+
2
p
dutu


As above, networks that piggyback acknowledgments on
return data packets will need an additional factor to account
for the delay until a return data packet occurs. Many also
will transmit a separate acknowledgment packet should a
data packet not occur soon enough, making the networks
hybrids.

E. Immediate Acknowledgment
The link-layer immediate-acknowledgement protocol
consists of transmitting a data packet one hop from mote A
to mote B, and then mote B immediately transmitting an
acknowledgement packet to mote A. This is the acknowled-
gment type performed by hardware acknowledgments as
well as the acknowledgment type performed in some TDMA
systems. This saves resources by reducing latency to
schedule an acknowledgement packet and greatly increases
the likelihood of successful acknowledgement packet
delivery due to channel coherence.
As a bounding upper-limit of acknowledgement reliability
for link-layer protocols, we assume p
ACK
= 1.u, i.e. that an
acknowledgement packet is always successful. Thus, the
probability that the data packet transmitted from mote A is
successfully received at mote B and the acknowledgment
packet transmitted from mote B is also received at mote A
is:
p
dutu+ACK
= p
dutu
· p
ACK
= p
dutu
.

1) Reliability-Case Tx-Rx Actions
Since the acknowledgment packet is always successful
there are no retries and only one acknowledgment packet is
transmitted per hop.

RCIxRxActions = (b
dutu+ACK
+k) +(u +k)

While this is an idealistic representation, it is not far from
correct with reasonable packet delivery rates, and is
nonetheless positions correctly as a bounding value with
respect to the other protocols. See Bounding Values, below.
2) Mean Tx-Rx Actions
The mean Tx-Rx actions count the same types of activity
as the reliability-case values, except using mean values.

HconIxRxActions =
k
p
dutu

+k.

3) Route Congestion

HconRoutcCongcstion
= |1 2…2 1] ·
1
∑ o
ì
k
ì=0
· HconIxRxActions.

4) Reliability-Case Tx-Rx Events
Since the acknowledgment occurs immediately after the
data packet is successfully received it does not add latency
and counts as the same event.

RCIxRxE:cnts = (b
dutu+ACK
+k) +u

5) Mean Tx-Rx Events
As above, since the acknowledgment occurs immediately
after the data packet is successfully received it does not add
latency and counts as the same event.

HconIxRxE:cnts =
k
p
dutu

.

F. Separate Acknowledgment
The link-layer separate-acknowledgement protocol
consists of transmitting a data packet one hop from mote A
to mote B, and then mote B scheduling and transmitting a
separate acknowledgement packet to mote A before
forwarding the data packet to the next mote C. This
acknowledgment type also occurs when acknowledgments
are performed at a higher level in software.
As a bounding lower-limit of acknowledgement reliability
for link-layer protocols we assume p
ACK
= p
dutu
. Thus, the
probability that the data packet transmitted from mote A is
received at mote B and the acknowledgment packet
transmitted from mote B is also received at mote A is:

p
dutu+ACK
= p
dutu
· p
ACK
= p
dutu
2
.

1) Reliability-Case Tx-Rx Actions

RCIxRxActions = (b
dutu+ACK
+k) +(b
ACK
+k).

While this represents a pessimistic representation,
certainly in applications where data packets and
acknowledgment packets are of similar length this would be
nearly typical. See Bounding Values, below.
2) Mean Tx-Rx Actions
The mean Tx-Rx actions count the same types of activity
as the reliability-case values, except using mean values.

HconIxRxActions =
k
p
dutu

2
+
k
p
dutu


3) Route Congestion

HconRoutcCongcstion
= |1 2…2 1] ·
1
∑ o
ì
k
ì=0
· HconIxRxActions

4) Reliability-Case Tx-Rx Events
Since all the Tx-Rx actions are separate, they are all also
separate events.
9


Figure 3. The PDR (stability) variations on the link from 
node 24 to node 17 shows daily periodicity corresponding to 
operational days in the factory. [6] 

Figure 2. Overall network PDR (stability) as a function of 
time throughout the sample period with each point 
representing one 15‐minute interval. The solid line is a 5‐hour 
moving average. [6] 
RCIxRxE:cnts = RCIxRxActions.

5) Mean Tx-Rx Events
Since all the Tx-Rx actions in the mean are separate, they
are all also separate events.

HconIxRxE:cnts = HconIxRxActions.

IV. COMPARISON OF BASIC ACKNOWLEDGEMENTS
PROTOCOLS
We first look at five common acknowledgment protocols
to reliably deliver packets: none, only end-to-end, implicit,
immediate, and separate. The basic acknowledgment
protocols reviewed vary in work load imposed on the
network with the number of hops and the available packet
delivery rate. While the number of hops is in part a hardware
architectural decision, it does affect the packet delivery rate
as the physical distance between motes and their location
affects the signal strength and signal multipath [1]. The
resulting packet delivery rate, however, varies greatly over
time. Figure 3 shows the packet delivery rate data for one
link collected over 26 days from an industrial site in
Berkeley, California [6]. The network is time-slotted
channel-hopping (TSCH), and the graph shows the link
packet delivery rate (stability) over all 16 IEEE 802.15.4
channels (0-15 on the graph mapped to 11-26 in the
standard). The baseline is 100% packet delivery rate, and the
detail below the line shows how much the packet delivery
rate has dropped. Note that channels 0, 3, 10, and 14 had
high packet delivery rates for the first 13 days with drops in
packet delivery rates for the days after, channels 5 and 12
experienced drops in packet delivery rates in the first 13
days and high packet delivery rates for the days after, and 2,
7, 8, 9, 11, 13, 15 had constantly varying packet delivery
rates. These packet delivery rate fluctuations are persistent
on the order of days. Since the network is a type of TDMA,
there are no collisions from several nodes trying to share the
same channel. All of the packet delivery rate degradation is
from the environment. Clearly one could not select one
“good” channel and expect a stable packet delivery rate, or
even the best packet delivery rate. All channels show packet
delivery rate variations lasting for days. Even the normally
expected quiet channel 24 (13 on the graph) varies
constantly throughout the test period. The variance of the
packet delivery rate for this link might be improved by
hopping through only the best channels rather than all the
channels.
Figure 2 shows the average packet delivery rate of the
entire network of 44 nodes and 89 links over the same period
[6]. Note that for the first two weeks of the period the 5-hour
average of the packet delivery rate was around 95% most of
the time, but for the remainder of the period dropped
significantly to as low as 83% and then recovered on the last
couple of days. This data is a network average, but
individual links fared much worse. In [15], the author
analyzes this same site data. Figure 6 shows the probability
density function of the packet delivery rates collected in
10% intervals. Note that many links have a significant
probability of lower packet delivery rates, one even a
notable amount in the 0-10% range. Thus while the network-
wide packet
delivery rates
averages in the
mid-80% or above
even in the worst
times, the
individual link data
tells a different
story.[6][18]
We evaluate a
design and choose
Protocol Sensitivity to 
Changes in PDR 
Order  Protocol 
Least  Immediate 
  Implicit 
  Separate 
Most  Only End‐to‐End 
As reliability R increases, 
sensitivity increases. 
Table 1. Protocol Sensitivity 
10

R=0.999  PDR:  0.3  0.6  0.9  1.0 
Reliability‐Case Tx‐Rx Actions for 1 hop 
End‐to‐End    95.2  24.8  7.9  2 
None    19.4  7.5  3  1 
Separate    94  24  8  2 
Implicit    84  20  6.5  1.5 
Immediate    21  9  4  2 
 
Reliability‐Case Tx‐Rx Actions for 4 hops 
End‐to‐End    1.5e5  1000  69.9  8 
None    1200  108  22.3  4 
Separate    180  47  18  8 
Implicit    161  39  14  6 
Immediate    43  20  12  8 
 
Reliability‐Case Tx‐Rx Actions for 10 hops 
End‐to‐End    2.8e11  4.7e5  468  20 
None    1.7e6  2.8e3  105  10 
Separate    314  86  36  20 
Implicit    280  71  28  28 
Immediate    79  40  26  20 

  PDR:  0.3  0.6  0.9  1.0 
Mean Tx‐Rx Actions for 1 hop 
End‐to‐End    14.4  4.4  2.4  2 
None 
(R=0.999) 
  19.4  7.5  3  1 
Separate    14.4  4.4  2.4  2 
Implicit    14.4  4.4  2.4  2 
Immediate    4.3  2.7  2.1  2 
 
Mean Tx‐Rx Actions for 4 hops 
End‐to‐End    2.2e4  146  13.2  8 
None 
(R=0.999) 
  1.2e3  108  22.3  4 
Separate    57.8  17.8  9.4  8 
Implicit    52.8  15.3  7.7  6.5 
Immediate    17.3  10.7  8.4  8 
 
Mean Tx‐Rx Actions for 10 hops 
End‐to‐End    4.1e10  6.8e4  72.3  20 
None 
(R=0.999) 
  1.7e6  2.8e3  105  10 
Separate    141  44.4  23.5  20 
Implicit    129  36.9  18.5  18.5 
Immediate    43.3  26.7  21.1  20 

Table 2. Reliability‐case and mean Tx‐Rx actions for basic 
acknowledgments by PDR for 1, 4, and 10 hops. 
a time-bounded reliability of just 90% and, based on the
early network-wide average packet delivery rate (stability)
of above 90%, assume a fixed link packet delivery rate of
90%. To achieve this design, for all links the probability
must be 90% or more that a packet delivery rate of at least
90% occurs. See the bar graph for link one in Figure 6. Link
one appears to be a reasonable link relative to the mesh
diagram, but note that the 90%-100% packet delivery rate
interval has a probability of less than 90%, so packet
delivery rates in the 80% to 90% interval will need to be
tolerated on this link to achieve 90% reliability. At higher
target reliabilities of 99%, 99.9% or more, some
occurrences of packet delivery rates in the 50% to 60%
interval or even the 40% to 50% interval will need to be
tolerated. This means that in determining the time bound
for a route, these lower intervals must be the design
values. If not, then these lower packet delivery rates will
occur, requiring more Tx-Rx events than we have
allocated time for in the time bound, and since they occur
within the probability of our chosen reliability, cause the
time bound to be exceeded and the time-bounded delivery
to fail more often than the desired reliability R [15].
Not all links on a route may fare so poorly, but because
the number of Tx-Rx events is generally an exponential
function of the packet delivery rate, using the average
packet delivery rate of a network is not sufficient for
design. Lower packet delivery rates will regularly occur
on some links so the protocol must behave well at lower
packet delivery rates for any reliability to be maintained.
A review of the graphs in this report shows that the
protocols have an overall sensitivity to changes in packet
delivery rate as listed in Table 3, an important
characteristic since packet delivery rates are time-varying.
In [15] the author discusses a heuristic to determine the
time bound for a route from the time-varying packet
delivery rate behavior on each link. We use constant
packet delivery rate for all links to simplify evaluation
and believe that reliability-case values from packet
delivery rates in the 60% to 90% range should represent
likely overall behavior of most links. Mean values from
time-weighted mean packet delivery rates are appropriate
for other evaluations.

A. Bounding Values
The assignments of packet delivery rates to protocols
were chosen to give the best range of evaluation for the
available graphs and to simplify calculations. We use two
bounding values for p
ACK
, 1.0 and p
dutu
. Large
differences in the length of data and acknowledgment
packets will show in the bit error rate (BER) and
subsequent packet delivery rate up due to Gaussian noise
effects, however, multipath generally dominates BER in
real environments (particularly indoors), making
differences much less than would be expected due to any
packet length difference [1].
For the only end-to-end acknowledgment protocol and all
end-to-end with link-layer acknowledgments, we use
p
ACKPuckct
= p
dutu
for the returning end-to-end acknowled-
gment packet. In cases where the length of the end-to-end
acknowledgment packet is much less than the length of the
original data packet, then the plotted lines would be the
upper boundary. On the graphs for the basic acknowled-
gments, the range of the only end-to-end acknowledgement
is at the line and just below. On the graphs of the end-to-end
11

Graph Bounding Regions 
ACK Protocol  Bounds  Weight 
 Tx‐Rx Actions 
End‐to‐End and Link‐Layer+EE  Self (as below) and lower  self 
Immediate  Immediate to Separate  Heavily Immediate 
Separate  Separate to Immediate  Separate 
 
 Tx‐Rx Events 
End‐to‐End and Link‐Layer+EE  Self (as below) and lower  Self 
Immediate  Immediate to Implicit  Heavily Immediate 
Separate  Separate and lower  Separate 
Notes: 
No‐acknowledgment and Implicit protocols have no bounding value 
 
Table 3. Graph Bounding Regions
acknowledgments, all plots would
be affected similarly, so while the
lines are also upper bounds, all
would range similarly lower
together.
For the link-layer protocols,
there are two separate bounding
cases. For Tx-Rx actions, the
immediate acknowledgment
protocol is a lower bound, with
p
ACK
= 1.u. This is not far off, as
error rates for the immediate
acknowledgment packet are
reported to be about 1/5 the error
rate for data packets. For a data
packet delivery rate of 90%, 60%
and 30% the acknowledgment packet delivery rate would be
about 98%, 92% and 86%, respectively, though at lower data
packet delivery rates this may not be accurate [12]. The
upper bound for the immediate acknowledgment protocol is
the line for the separate acknowledgement protocol, which is
plotted with p
ACK
= p
dutu
. The immediate acknowledgment
range is thus the area between immediate and separate
acknowledgments lines, heavily weighted toward the
immediate acknowledgment protocol line. The separate
acknowledgment protocol does not have the advantage of a
known-good channel since the acknowledgment packet is
separately scheduled, thus the packet delivery rate is less.
For Tx-Rx events the reasoning is the same but the bounds
are different due to difference in the computation of the
events. See Table 3.

B. Bandwidth
To prevent bandwidth congestion, Tx-Rx actions need to
be spread evenly across motes on a route (Figure 7 and
Figure 8). The relative metric we use is Tx-Rx actions as a
measure the bandwidth consumed by one packet transfer.
Note that in interpreting these graphs for bandwidth
comparison, the Tx-Rx action values for the implicit
acknowledgment protocol are slightly high because the
forward/acknowledgment single transmission is actually one
bandwidth usage but is counted as 1.5 Tx-Rx actions since
there are two receives and one transmitter active.
At PÐR = 1.u, all acknowledgments are redundant and
unneeded, so the no-acknowledgment protocol has the least
bandwidth usage. As the packet delivery rate drops to 0.9 the
saved acknowledgment transmission of the implicit
acknowledgement reveals its efficiency. However, as the
packet delivery rate drops further, the requirement of the
implicit acknowledgment that both the data packet and the
acknowledgment packet be successful, p = p
dutu
2
, causes it
to lose ground to the immediate acknowledgment protocol.
While we consider the probability of delivery at each hop
as independent, Figure 2 shows that effects on packet
delivery rate can be network-wide. As packet delivery rates
drop, more links will require greater retries, toward their
reliability case values (Figure 10). The system designer must
be concerned with having sufficient bandwidth to
accommodate the reliability-case Tx-Rx actions on many
links when necessary.
Since the failure of the transmission of packets in the no-
acknowledgment and only end-to-end acknowledgment
protocol follow a geometric progression, the bandwidth
usage is similarly exponentially loaded toward the source,
increasingly so as the packet delivery rate drops. As can be
seen from Figure 9 with the Tx-Rx actions representing the
relative total bandwidth usage along a route of k hops for a
single packet transfer, these protocols consume an
impractical amount of bandwidth as the packet delivery rate
drops or the hops increase.

C. Node and Network Energy
Transmitting and receiving packets are generally the most
energy consuming activates of a mote. The relative metric
we use is mean Tx-Rx actions. Each unit of the metric
represents one transmitter and receiver in operation for one
packet. Since the implicit acknowledgment protocol
performs the forward/acknowledgment in a single trans-
mission with two receivers, each such activity is counted as
1.5 Tx-Rx actions. Many current radios specify about the
same current for both transmission and reception so the
mean Tx-Rx actions on the graphs can be directly translated
into energy usage by multiplying by an appropriate constant.
To prevent nodes depleting their energy sources unevenly,
Tx-Rx actions need to be spread evenly across motes on a
route (Figure 7 and Figure 8). Nevertheless, application
source and destination requirements will place burdens on
specific motes. At PÐR = 1.u, all acknowledgments are
redundant and unneeded, so the no-acknowledgment
protocol has the least energy usage per mote and for the
route. As the packet delivery rate drops to 0.9, the saved
acknowledgment transmission of the implicit acknowled-
gement protocol again reveals its efficiency. However, as the
packet delivery rate drops further, the requirement for the
implicit acknowledgment protocol that both the data packet
12


Figure 4. Tx‐Rx events by basic acknowledgments for 
reliabilities from 0.9 to 0.999999. 
0.9 0.99 0.999 0.9999 0.99999 0.999999
0
5
10
15
20
25
30
35
40
Reliability
R
e
l
i
a
b
i
l
i
t
y
-
C
a
s
e

T
x
-
R
x

E
v
e
n
t
s
Latency by Reliability of Route, PDR=0.9, 4 hops


End-to-End
None
Separate
Implicit
Immediate
and the acknowledgment packet be successful, p = p
dutu
2
,
causes it again to lose ground to the immediate
acknowledgment protocol.
Figure 9 shows mean Tx-Rx actions for an entire route for
k hops and they behave here as above. Since the failures of
transmissions of packets in the no-acknowledgment and only
end-to-end acknowledgment protocol follow a geometric
progression, the energy usage is similarly exponentially
loaded toward the source, increasingly so as the packet
delivery rate drops. In Figure 9 with the Tx-Rx actions
representing relative total energy for a single packet transfer,
these protocols consume an impractical amount of energy as
the packet delivery rate drops or the hops increase.

D. Resource Usage
Resource usage refers to the duration of usage measured
in Tx-Rx events for two types: resources that are used during
the short duration of a Tx-Rx event and those that are used
for the long duration of a successful or failed delivery,
possibly spanning many Tx-Rx events. The first type include
items such as the availability of CPU, radio, temporary
storage, program state and other items that are used while a
packet is being processed. The second type is primarily
packet buffers of which both mean and reliability-case
values are of interest. The metric for both are units of time
measured in multiples of Tx-Rx events.
Figure 11 shows the total mean Tx-Rx events for a route
of k hops. For the link-layer protocols, the values are k
times the mean individual usage, and represent how many
times, for this packet, the CPU will likely wake up and go to
sleep, the radio will be activated, etc. For the only end-to-
end acknowledgment protocol, the value is usage of
resources on the source mote of long duration since the
source mote must wait for the acknowledgment packet from
the destination before it can release the buffer with the
packet. The values for the other motes on this route are of
short duration distributed geometrically from the source,
similar to the distribution in Figure 7. Further, while the
source mote is waiting for the end-to-end acknowledgment
on one packet, many other packets may also be sent. This
creates packet buffer resource usage accumulating over a
long duration. The longer the route in hops and the lower the
packet delivery rate, the more buffers can need to be in
service concurrently.
As discussed above, the packet delivery rate can degrade
network-wide. As packet delivery rates drop, more links will
require greater retries, toward their reliability case values
(Figure 12). The system designer must be concerned with
having sufficient resources to accommodate the duration of
reliability-case Tx-Rx events on many links when necessary.
Note that the immediate acknowledgment protocol has the
lowest Mean Tx-Rx events. This implies that resources are
tied up the least amount and for the shortest duration. It also
implies that, other things being equal, it can achieve the
highest packet transfer rate.
E. Reliability
Reliability, we have defined, is the probability of
delivering a packet within a given time bound. Given a
proper estimate of the packet delivery rate and an adequate
time bound, all of the basic acknowledgment protocols can
deliver the same reliability R, within a factor for the inherent
reliability of the network o, o ¸ 1, which results from
extraordinary events (discussed below). That is, given a
protocol
ì
that for reliability R requires up to c
ì
Tx-Rx
events to achieve that reliability, and we allow sufficient
time to accommodate c
ì
events, every protocol
ì
will
succeed or fail (late or no delivery) with reliability R or oR.
The no-acknowledgment and link-layer protocols cannot
overcome the inherent reliability of the network o and will
deliver the packet with reliability oR. The only end-to-end
acknowledgment protocol (and other end-to-end protocols)
can overcome the inherent reliability of the network o and
will have delivered the packet with reliability R, by
designing for reliability
R
u
(if o ¸ R) such that the final
reliability R = o ·
R
u
. In general, we assume o = 1, as is
often the case, so o can be neglected. The case of o < R is
discussed in Extraordinary Events, below. Each of these
protocols is a mechanism with differing work load
requirements to perform the basic function of ensuring that
the packet reaches its destination with time-bounded
reliability R. If we misestimate the packet delivery rate, the
effect on the available reliability varies greatly with the
protocol.
13

R=0.999  PDR:  0.3  0.6  0.9  1.0 
Reliability‐Case Tx‐Rx Actions for 1 hop 
End‐to‐End    95.2  24.8  7.9  1 
Separate+EE    126  33  12  2 
Implicit+EE    113  27.5  9.5  4 
Immediate+EE    29  13  7  3 

Reliability‐Case Tx‐Rx Actions for 4 hops 
End‐to‐End    1.5e5  1000  69.9  8 
Separate+EE    271  74  30  16 
Implicit+EE    242  61  23.5  12 
Immediate+EE    67  34  21  16 
 
Reliability‐Case Tx‐Rx Actions for 10 hops 
End‐to‐End    2.8e11  4.7e5  468  20 
Separate+EE    512  144  63  40 
Implicit+EE    456  119  48.5  48.5 
Immediate+EE    133  71  49  40 

  PDR:  0.3  0.6  0.9  1.0 
Mean Tx‐Rx Actions for 1 hop 
End‐to‐End    14.4  4.4  2.4  2 
Separate+EE    28.9  8.9  4.7  4 
Implicit+EE    28.9  8.9  4.7  4 
Immediate+EE    8.7  5.3  4.2  4 

Mean Tx‐Rx Actions for 4 hops 
End‐to‐End    2.18e4  146  13.2  8 
Separate+EE    116  35.6  18.8  16 
Implicit+EE    106  30.6  15.4  13 
Immediate+EE    34.7  21.3  16.9  16 
 
Mean Tx‐Rx Actions for 10 hops 
End‐to‐End    4.1e10  6.8e4  72.3  20 
Separate+EE    289  88.9  46.9  40 
Implicit+EE    259  73.9  36.9  36.9 
Immediate+EE    86.7  53.3  42.2  40 

Table 4. Reliability‐case and mean Tx‐Rx actions for end‐to‐end 
acknowledgments by PDR for 1, 4, and 10 hops. 
Given the low packet delivery rates of wireless networks,
reliability requires repeatedly transmitting a packet until it is
successfully delivered. The ability of the network to do the
least amount of work to deliver a packet means that network
resources are available for retrying the delivery of other
packets. Thus, the more Tx-Rx events that a protocol
requires for a given set of conditions, the less reliable the
network becomes overall. Refer to Figure 4. Note that the
implicit acknowledgment protocol requires 12 reliability-
case Tx-Rx events for a reliability R of 0.999 at PÐR = u.9.
For those same 12 Tx-Rx events, using the immediate
acknowledgment protocol, we can achieve over two orders
of magnitude greater reliability. Refer to Figure 9 and
compare the mean Tx-Rx actions for implicit and immediate
acknowledgements between the PÐR = u.9 and the
PÐR = u.6 graphs, the area where the network will likely
operate. The two protocols change places such that they are
close in value within our operating range. Thus, by
choosing the immediate acknowledgment protocol
over the implicit acknowledgment protocol we can
improve the reliability of the network at the same
average energy, bandwidth, etc. Alternatively, by
keeping the time bound for reliability R at 0.999, we
can accommodate a higher network load and still
have the resources to meet the reliability
requirements.

F. Latency
To simplify the translation of Tx-Rx events to
latency in seconds, the packet delivery rate used for
determining Tx-Rx events (mean or reliability-case)
should include the effect of CCA failures and
scheduling delays. The effects should be considered
delivery failures for this purpose, thus lowering the
effective packet delivery rate. We would expect then
that single-channel and CSMA systems could have
lower effective packet delivery rates than TDMA or
TSCH systems [8][18]. For similar reasons we would
expect less variance in the latency in TDMA or
TSCH systems.
For most CSMA architectures, and simplistically
for TDMA architectures, the total latency is

Iotcncy = IxRxE:cnts · I
hop
.

I
hop
is the time interval between transmission
attempts on a link. In a TDMA system, if at least one
transmission can be attempted within the same frame
at each hop, then latency is

Iotcncy ¸ k · I
]¡umc
+(IxRxE:cnts -k) · I
hop
.

I
]¡umc
is the maximum time between two slots in
a frame. Since I
]¡umc
¸ I
hop
, the latency can be
less in TDMA compared to CSMA systems given
equal I
hop
. Further, for critical routes, multiple TDMA slots
can be allocated to a link, further decreasing the latency.
Many systems are more complex, such as those that
piggyback the acknowledgement if a return data packet is to
be transmitted soon; otherwise the acknowledgement is
transmitted as a separate packet. Appropriate metrics to
combine these models and account for the additional delays
need to be devised for the behavior of each system.
Given equal values of I
hop
, the delivery latency of
systems of similar network architectures can be compared
directly in units of Tx-Rx events. While at high packet
delivery rates several protocols are similar in mean and
reliability-case Tx-Rx events (Figure 11 and Figure 12), as
packet delivery rates drop the immediate acknowledgment
protocol has the clear advantage. By performing the
acknowledgment immediately there are simply fewer events
to schedule.
14


Figure 5. Tx‐Rx events by end‐to‐end acknowledgments 
for reliabilities from 0.9 to 0.999999. 
0.9 0.99 0.999 0.9999 0.99999 0.999999
20
40
60
80
100
120
140
160
180
Reliability
R
e
l
i
a
b
i
l
i
t
y
-
C
a
s
e

T
x
-
R
x

E
v
e
n
t
s
Latency by Reliability of Route, Immediate ACK, 10 hops


PDR 0.3+EE
PDR 0.6+EE
PDR 0.9+EE
PDR 0.3
PDR 0.6
PDR 0.9
0.9 0.99 0.999 0.9999 0.99999 0.999999
0
20
40
60
80
100
120
140
160
180
200
Reliability
R
e
l
i
a
b
i
l
i
t
y
-
C
a
s
e

T
x
-
R
x

E
v
e
n
t
s
Latency by Reliability of Route, Immediate ACK, 4 hops


PDR 0.3+EE
PDR 0.6+EE
PDR 0.9+EE
PDR 0.3
PDR 0.6
PDR 0.9
Only EE PDR 0.9
V. COMPARISON OF END-TO-END ACKNOWLEDGEMENTS
End-to-end acknowledgments are often used as an
absolute verification of delivery and are well suited to that
task. However, the reliability improvement (1 -o) that is
available is typically small, except in systems that drop
packets when delivery is difficult—some systems have no
retries or limit retries to as little as three [18]. The end-to-
end acknowledgment can then supply the mechanism to
inform of the need to retry delivery. Referring to Table 2, we
find that at packet delivery rates of 0.6 and 0.3 for one hop,
for the immediate acknowledgment protocol the mean
number of transmissions is 2.7 and 4.3 respectively.
However, to have a 99.9% probability of success, we need 9
and 21 attempts, respectively. Taken as a whole, the retries
due to a missing end-to-end acknowledgment can supply the
additional attempts to raise reliability, but at the great
expense of an end-to-end protocol rather than a more local
approach.

A. Extraordinary Events
Other than intentionally dropped packets, end-to-end
acknowledgments also add delivery security against events
where a packet might be successfully delivered and
acknowledged at an intermediate mote on the route to the
destination and then might not be forwarded. This can occur
due to hardware or software bugs, node failure, link failures,
unreported buffer overflows, node isolation, or other
transient or permanent failures. We capture the limited
reliability due to these extraordinary events in an inherent-
reliability-of-the-network factor o. Generally, these events
should occur very infrequently, often at much less than the
typical reliability-design of the system, and far, far less
frequently than delays due to normal packet delivery rates
that can readily cause a packet to be delivered late. (Or,
conversely, if they occur more frequently a new sensor mote
vendor should be selected.) For example, in [6], over a 26
day period at an industrial site, only 17 packets out of 3.6
million generated were lost, for an inherent reliability
o = u.99999S. Time bounds were not mentioned, but the
Dust Networks products used at the site essentially never
stop retrying to deliver a packet so the effective time bound
may have been large. From this, it can be said that inherent
reliability can be high and extraordinary events can be rare.
Compared to link-layer acknowledgment protocols, end-
to-end acknowledgment protocols can only add the benefit
of mitigating the inherent reliability o, generally a very
small additional factor of reliability. They cannot create
100% time-bounded reliability because they cannot
guarantee that the acknowledgment packet will arrive within
the time bound. However, in cases where the environmental
or other conditions cause some extraordinary events to be
ordinary, such that the events become a material failure rate
relative to the desired reliability R, per-packet or selective
end-to-end acknowledgments can be necessary.

B. When to Use End-to-End Acknowledgments
Per-packet end-to-end acknowledgments can be very
expensive. An only end-to-end acknowledgment requires
more mean and reliability-case Tx-Rx events than any other
basic protocol reviewed under almost all conditions of
PDR<1.0. Adding an end-to-end acknowledgement to a link-
layer protocol is generally more efficient, but typically
doubles the mean Tx-Rx actions at all packet delivery rates
because twice as many hops are traversed. The effect is not
as direct on reliability-case Tx-Rx actions because twice as
many hops result in less than twice as many trials.
15

If the inherent reliability o is high such that o > R
dcsìgn
,
and reliability R
]ìnuI
= oR
dcsìgn
is suitable, then end-to-end
acknowledgments should not be used. They will lower the
overall network reliability in that resources are unnecessarily
consumed and unavailable. Refer to Figure 5. At
PDR=0.9+EE, four hops, and a design reliability of 0.9, it
appears we can obtain five orders of magnitude more
reliability for the same number of Tx-Rx events by not using
end-to-end acknowledgments. At this point our reliability
would likely be limited by the inherent reliability o. At ten
hops the reliability improvement appears unlimited. If we
start at a more reasonable value at R=0.999 then we have at
least three orders of magnitude before leaving the plot.
If the inherent reliability o is low such that it encroaches
on or is lower than the desired reliability R, then an end-to-
end acknowledgment protocol is need. Since we cannot
compute reliability greater than 1.0, we perform the
computation in two layers. First we compute the reliability-
case values for an end-to-end acknowledgment protocol,
then we compute how many retries of those reliability-case
values must be accommodated to obtain the desired
reliability R. We can use equations for one hop of the only
end-to-end acknowledgment protocol.

p = o · R
LLp¡otocoI

b
LLp¡otocoI
=
log(1 -R)
log(1 -p
2
)
.

The selected end-to-end acknowledgment protocol that
had been designed (using the computations below) to deliver
reliability R
LLp¡otocoI
will need accommodation for retries
b
LLp¡otocoI
to achieve desired reliability R. Since
incremental increases in reliability are exponentially less
expensive, the highest reliability R
LLp¡otocoI
that is practical
should be selected. Since link-layer with end-to-end
acknowledgments are more efficient than the only end-to-
end acknowledgment protocol, we chose one of those for the
internal primary link-layer protocol.

C. End-To-End Acknowledgement with Link-Layer
Acknowledgements
Adding a per-packet end-to-end acknowledgement to a
link-layer protocol simply transfers an a packet the same k
hops back along the route from the destination to the source.
While this creates additional activity, since the packets are
still moved forward with a link-layer acknowledgment
protocol, the average workload grows linearly rather than
exponentially as it does for the only end-to-end acknowled-
gment protocol. The only end-to-end acknowledgment
protocol requires fewer Tx-Rx actions and Tx-Rx events at
extremely high packet delivery rates, as has been found on
wired connections. However, this advantage drops with
packet delivery rate, though there is still an advantage in Tx-
Rx actions up to five hops, and Tx-Rx events against some
link-layer alternatives up to three or so hops at PÐR = u.9.
Similarly, the route congestion in Figure 13 and Figure 14
flips for the only end-to-end acknowledgment protocol being
the best at four hops to the worst at ten hops.
Further, link-layer acknowledgements are far less
sensitive to variations in hop depth or packet delivery rate
and are thus more robust. In cases described above when an
end-to-end acknowledgment is needed, it should be on top of
an efficient link-layer acknowledgment protocol (Figure 17,
Figure 18). While the only end-to-end acknowledgment
fares better against the link-layer protocols with end-to-end
acknowledgment than to the basic acknowledgment
protocols, the advantage quickly dies with increased hops or
decreased packet delivery rate.
Thus, an end-to-end acknowledgment with link-layer
acknowledgement results in better performance [2], so long
as the timing for retries is adequate so that the behaviors of
the two layers do not interact poorly [11]. However, all end-
to-end protocols create significant buffer demands at the
source mote that do not exist with only link-layer
acknowledgments. The latencies are quite long, as described
above, and may entail many buffers simultaneously as
several end-to-end transactions may be in process. Thus,
unless reliability R exceeding the inherently reliability o is
required, end-to-end acknowledgments add significant
stresses to the network that may elicit failures and lower the
inherent reliability o rather than increase overall reliability.
1) Calculations
End-to-end acknowledgments with link-layer acknowled-
gments perform identically to their non-end-to-end
counterparts, except the return acknowledgment packet
doubles the number of hops traversed. For all link-layer
acknowledgment calculations with hops k, apply 2k.
2) Bandwidth
As discussed above, the metric for bandwidth is mean Tx-
Rx Actions (Figure 15). The only end-to-end acknowled-
gment protocol uses less bandwidth up to a few hops, on
average, than the end-to-end acknowledgment with link-
layer acknowledgment protocols, but this advantage is
exponentially eliminated and is unevenly distributed along
the route (Figure 13) as the packet delivery rate drops.
3) Node and Network Energy.
The same patterns already discussed follow here. There
are narrowly bounded conditions where the only end-to-end
acknowledgment consumes less energy than using an end-to-
end acknowledgment with a link-layer-acknowledgment.
4) Resource Usage
Using an end-to-end acknowledgment with a link-layer
acknowledgment now causes all the protocols to have the
same long-term packet buffers on the source mote. However,
the mean and reliability-case durations are significantly less
in most cases for the link-layer acknowledgment alternatives
(Figure 17, Figure 18). This implies that by using the link-
layer alternatives, potentially fewer extraordinary events,
such as unreported packet buffer exhaustion, occur, and thus
the number of end-to-end acknowledgment retries should be
reduced.
16

To a degree, the exponential resource usage duration of
the only end-to-end protocol can be mitigated by
unconditionally retransmitting packets from the source
rather than waiting for the failed end-to-end
acknowledgment. However, while the mean Tx-Rx actions
become comparable to the link-level alternatives at high
packet delivery rates, as the packet delivery rates drop
exponential growth again occurs and any benefit is lost after
a few hops. In the reliability-case, the exponential growth is
slower, but still exponential and exceeds the link-level
alternatives after a few hops.
5) Reliability
As discussed above, there are well defined instances when
an end-to-end acknowledgment is required to meet
application reliability requirements. If the reliability
requirements do not need an end-to-end acknowledgment
protocol, then much higher reliability is available from the
basic link-layer protocol for the same number of Tx-Rx
events (Figure 5). When needed, using an end-to-end
acknowledgment with a link-layer acknowledgment protocol
consumes far fewer Tx-Rx events in almost all cases than
the only end-to-end acknowledgment protocol for reasons
previously discussed (Figure 16). In all cases, the end-to-end
acknowledgment will allow overcoming the inherent
reliability o limitation.
6) Latency
Latency follows the same behavior as Resource Usage
and Latency previously discussed.

VI. CUMULATIVE OR SELECTIVE ACKNOWLEDGMENTS
Either cumulative or selective acknowledgments can
reduce the overhead for the end-to-end acknowledgment
packet by reducing the number that is transmitted. While
these acknowledgments were not specifically modeled for
this report, we believe sufficient data is available to glean
their first order behavior.
Cumulative or selective acknowledgments are often used
as a replacement for per-packet only end-to-end acknowled-
gments. As such, the packet goes out toward the destination
with no link-layer acknowledgments. If the ratio of data
packets to acknowledgments is high, the closest model to the
mean Tx-Rx actions of data packet transmissions is the only
end-to-end acknowledgment protocol at one-half the hop
distance. We use one-half because we want just to evaluate
the Tx-Rx actions to reach the destination but not to return
the end-to-end acknowledgment. We want to determine the
activity to get a packet to the destination, but assume an
infrequent and yet uncounted acknowledgment packet to
indicate which packets will need to be retransmitted.
Refer to Figure 9. For PÐR = u.9 at five hops (ten
equivalent for our comparison), the only end-to-end
acknowledgment protocol is comparable to the link-layer
protocols at ten hops. Thus, at very high packet delivery
rates as seen in the first half of Figure 2, selective or
cumulative acknowledgments may be an advantage.
However, as the packet delivery rate drops toward 0.6 in the
next plot, the number of Tx-Rx actions for our new protocol
is off the plot. At up to two hops, the situation is more
comparable, but there is no definite win for the new
protocols. Clearly, operating in the expected packet delivery
rate range the new protocol is comparable or not efficient
compared to a per-packet end-to-end acknowledgment with
a link-layer acknowledgment.
However, using cumulative or selective acknowledgments
with link-layer acknowledgments would be very efficient, if
an end-to-end acknowledgment were necessary. The
inherent reliability o of the network could be overcome with
fewer resources used compared to per-packet end-to-end
acknowledgments, aside from the packet buffers required
until successful delivery.

VII. CONCLUSIONS
Careful evaluation of each wireless system is necessary to
determine the time bounds required to meet the desired
delivery reliability. While many acknowledgment protocols
are in use, few perform well across a range of network
conditions. The best in each area are summarized in Table 5
and Table 6, for R=0.999, with changes for higher and lower
reliabilities. The no-acknowledgment protocol is suitable for
use when the packet delivery rates are high, or required
reliabilities are low, but otherwise is more expensive than
alternatives. Its simplicity does not save resources outside of
those ranges. The only end-to-end acknowledgment protocol
fares well at very high packet delivery rates and few hops, as
it would in a wired environment. Otherwise, it, too, quickly
becomes too expensive. Of the link-layer protocols, the
implicit acknowledgment protocol fares well briefly at high
packet delivery rates, but, in general, the immediate
acknowledgment protocol performs best overall, even when
the bounded immediate range is reviewed. It also is the most
robust with the lowest sensitivity to changes in packet
delivery rate. As Figure 6 shows, packet delivery rates on
many links drop well below the average for enough of the
time that the lower packet delivery rates will dominate the
packet delivery rate for those links, and thus routes that
include those links.
While link-layer protocols typically operate indepen-
dently from higher-layer protocols, that does not mean that
the higher-layer protocol should operate unaware of the link-
layer behavior. Ignoring actual link-layer behavior while
considering parameters such as delivery timing is not link-
layer agnostic, but the attempted ignorance itself is assuming
some generic model and can lead to unfortunate interactions
between the layers that reduce goodput. Studies that find
layer interaction problems are typically looking at the results
of using the TCP, which has significant mechanisms that
assume the characteristics and reliability of a wired
connection at the lower layers. When not taken into account,
inappropriate timing interactions between the layers then
generates degraded goodput results [7][11]. Inter-layer
17

awareness and cooperation to prevent redundant operations
has been found to significantly improve goodput [3][11], but
this must be carefully abstracted to allow the architecture to
evolve and prevent obsolesce [13] .
Arbitrarily adding end-to-end acknowledgments to packet
deliveries without understanding the underlying link-layer
reliability can reduce overall network reliability by
increasing resource usage and limiting resource availability
with little, if any, gain in reliability. Only if the inherent
reliability o of the network is low should end-to-end
acknowledgments be considered, and then only on top of the
most efficient link-layer acknowledgment, which we find to
be immediate acknowledgments. Even more efficient are
cumulative or selective acknowledgments when resources
are available to hold the packets that may need to be
retransmitted. These appear efficient on a link-layer
acknowledgment protocol, and still should be used only
when necessary, i.e. because inherent reliability o is low.
In general, reliability is obtained by accommodating the
time required for sufficient retransmissions to achieve the
desired reliability. In this report we show how to determine
the required reliability, how to determine the required time
bound given a sufficient estimate of the packet delivery rate
behavior, metrics to select the best type of acknowledgment
for the conditions, and reasoning for when end-to-end
acknowledgments are required and how to improve
reliability up to the inherent limit of the network and above.
18


Figure 6. Time‐varying packet delivery rate probability distributions. To achieve even reasonable reliability, mid‐valued PDRs 
on many links and the corresponding larger number of retransmission must be accommodated in the time bound. 

0
0.5
0.9
0
50
100
0
0.5
1
PDR Interval
PDFs of PDR for All 89 Links
Link Number
P
r
o
b
a
b
i
l
i
t
y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PDFs of PDR of All 89 Links
PDR Interval
P
r
o
b
a
b
i
l
i
t
y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PDF of PDR of Link 1
PDR Interval
P
r
o
b
a
b
i
l
i
t
y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PDF of PDR of Lowest Mean PDR Link
PDR Interval
P
r
o
b
a
b
i
l
i
t
y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PDF of PDR of Link 4
PDR Interval
P
r
o
b
a
b
i
l
i
t
y
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
0
0.2
0.4
0.6
0.8
1
PDF of PDR of Link 59
PDR Interval
P
r
o
b
a
b
i
l
i
t
y
19



Figure 7. Mean Tx‐Rx actions by basic acknowledgment type at each hop on a linear route of 4 hops. Tx‐Rx actions are a 
metric for bandwidth and energy usage. 

0 1 2 3 4
10
0
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End, Sep., Immed.
None (R=0.999)
Implicit
0 1 2 3 4
10
1
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
None (R=0.999)
Separate
Implicit
Immed.
0 1 2 3 4
10
0
10
1
10
2
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
0 1 2 3 4
10
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
0 1 2 3 4
0
0.5
1
1.5
2
2.5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
0 1 2 3 4
0
1
2
3
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
0 1 2 3 4
0
1
2
3
4
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
0 1 2 3 4
0
5
10
15
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
20



Figure 8. Mean Tx‐Rx actions by basic acknowledgment type at each hop on a linear route of 10 hops. Tx‐Rx actions are a 
metric for bandwidth and energy usage. 

0 1 2 3 4 5 6 7 8 9 10
10
0
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End, Sep., Immed.
None (R=0.999)
Implicit
0 1 2 3 4 5 6 7 8 9 10
10
1
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
None (R=0.999)
Separate
Implicit
Immed.
0 1 2 3 4 5 6 7 8 9 10
10
0
10
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
0 1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
0 1 2 3 4 5 6 7 8 9 10
0
0.5
1
1.5
2
2.5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
0 1 2 3 4 5 6 7 8 9 10
0
5
10
15
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
21



Figure 9. Mean Tx‐Rx actions by basic acknowledgment type for 1 to 10 hops. Tx‐Rx actions are a metric for bandwidth and 
energy usage. 

1 2 3 4 5 6 7 8 9 10
10
0
10
1
10
2
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End, Sep., Immed.
None (R=0.999)
Implicit
1 2 3 4 5 6 7 8 9 10
10
2
10
4
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
None (R=0.999)
Separate
Implicit
Immediate
1 2 3 4 5 6 7 8 9 10
10
2
10
4
10
6
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
1 2 3 4 5 6 7 8 9 10
0
50
100
150
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
22



Figure 10. Reliability‐case Tx‐Rx actions by basic acknowledgment type for 1 to 10 hops and R=0.999. Reliability‐case values 
indicate the activity that must be accommodated to meet the desired reliability. 

1 2 3 4 5 6 7 8 9 10
10
0
10
1
10
2
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1, R = 0.999


End-to-End, Sep., Immed.
None
Implicit
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9, R = 0.999


End-to-End
None
Separate
Implicit
Immediate
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
100
200
300
400
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3, R = 0.999
23



Figure 11. Mean Tx‐Rx events by basic acknowledgment type for 1 to 10 hops. Mean Tx‐Rx events are a metric for resource 
usage duration and latency. 

1 2 3 4 5 6 7 8 9 10
10
0
10
1
10
2
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 1, R = 0.999


End-to-End, Separate
Implicit
Immed., None (R=0.999)
1 2 3 4 5 6 7 8 9 10
10
5
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.9, R = 0.999


End-to-End
None (R=0.999)
Separate
Implicit
Immediate
1 2 3 4 5 6 7 8 9 10
10
5
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
100
200
300
400
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.3, R = 0.999
24



Figure 12. Reliability‐case Tx‐Rx events by basic acknowledgment type for 1 to 10 hops and R=0.999. RC Tx‐Rx events are a 
metric for resource usage duration, reliability time‐bound, and latency. 

1 2 3 4 5 6 7 8 9 10
10
0
10
1
10
2
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 1, R = 0.999


End-to-End, Separate
Implicit
Immed., None (R=0.999)
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.9, R = 0.999


End-to-End
None (R=0.999)
Separate
Implicit
Immediate
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
100
200
300
400
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.3, R = 0.999
25



Figure 13. Mean Tx‐Rx actions by end‐to‐end acknowledgment type at each hop on a linear route of 4 hops. Tx‐Rx actions are 
a metric for bandwidth and energy usage. 

 
0 1 2 3 4
10
0
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End
Implicit+EE
Sep.+EE, Immed.+EE
0 1 2 3 4
10
0.3
10
0.5
10
0.7
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
Sep.+EE
Implicit+EE
Immed.+EE
0 1 2 3 4
10
0
10
1
10
2
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
0 1 2 3 4
10
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
0 1 2 3 4
0
1
2
3
4
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
0 1 2 3 4
0
2
4
6
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
0 1 2 3 4
0
2
4
6
8
10
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
0 1 2 3 4
0
10
20
30
40
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
26



Figure 14. Mean Tx‐Rx actions by end‐to‐end acknowledgment type at each hop on a linear route of 10 hops. Tx‐Rx actions are 
a metric for bandwidth and energy usage. 

0 1 2 3 4 5 6 7 8 9 10
10
0
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End
Implicit+EE
Sep.+EE, Immed.+EE
0 1 2 3 4 5 6 7 8 9 10
10
1
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
Sep.+EE
Implicit+EE
Immed.+EE
0 1 2 3 4 5 6 7 8 9 10
10
0
10
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
0 1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
0 1 2 3 4 5 6 7 8 9 10
0
1
2
3
4
5
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
0 1 2 3 4 5 6 7 8 9 10
0
2
4
6
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
0 1 2 3 4 5 6 7 8 9 10
0
2
4
6
8
10
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
0 1 2 3 4 5 6 7 8 9 10
0
10
20
30
Hop Distance from Source
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
27



Figure 15. Mean Tx‐Rx actions by end‐to‐end acknowledgment type for 1 to 10 hops. Tx‐Rx actions are a metric for bandwidth 
and energy usage. 

 
1 2 3 4 5 6 7 8 9 10
10
1
10
2
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 1


End-to-End
Implicit+EE
Immed.+EE, Sep.+EE
1 2 3 4 5 6 7 8 9 10
10
1
10
2
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.9


End-to-End
Sep.+EE
Implicit+EE
Immed.+EE
1 2 3 4 5 6 7 8 9 10
10
2
10
4
10
6
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.6
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s

(
l
o
g
)
PDR = 0.3
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6
1 2 3 4 5 6 7 8 9 10
0
100
200
300
Hops
M
e
a
n

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3
28



Figure 16. Reliability‐case Tx‐Rx actions by end‐to‐end acknowledgment type for 1 to 10 hops and R=0.999. Reliability‐case 
values indicate the activity that must be accommodated to meet the desired reliability. 
 

1 2 3 4 5 6 7 8 9 10
10
1
10
2
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1, R = 0.999


End-to-End
Implicit+EE
Immed.+EE, Sep.+EE
1 2 3 4 5 6 7 8 9 10
10
2
10
4
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9, R = 0.999


End-to-End
Sep.+EE
Implicit+EE
Immed.+EE
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
10
20
30
40
50
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
50
100
150
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
200
400
600
Hops
R
C

T
x
-
R
x

A
c
t
i
o
n
s
PDR = 0.3, R = 0.999
29



Figure 17. Mean Tx‐Rx events by end‐to‐end acknowledgment type for 1 to 10 hops. Mean Tx‐Rx events are a metric for 
resource usage duration and latency. 
 
1 2 3 4 5 6 7 8 9 10
10
1
10
2
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 1, R = 0.999


All other EE&+EE
Separate+EE
Implicit+EE
1 2 3 4 5 6 7 8 9 10
10
5
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.9, R = 0.999


End-to-End
Separate+EE
Implicit+EE
Immed.+EE
EE Unconditional
1 2 3 4 5 6 7 8 9 10
10
5
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
50
100
150
200
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
200
400
600
Hops
M
e
a
n

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.3, R = 0.999
30



Figure 18. Reliability‐case Tx‐Rx events by end‐to‐end acknowledgment type for 1 to 10 hops and R=0.999. RC Tx‐Rx events 
are a metric for resource usage duration, reliability time‐bound, and latency. 
 

1 2 3 4 5 6 7 8 9 10
10
1
10
2
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 1, R = 0.999


All other EE&+EE
Implicit+EE
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.9, R = 0.999


End-to-End
Separate+EE
Implicit+EE
Immed.+EE
EE Unconditional
1 2 3 4 5 6 7 8 9 10
10
5
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
10
5
10
10
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s

(
l
o
g
)
PDR = 0.3, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
5
10
15
20
25
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 1, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
20
40
60
80
100
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.9, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
50
100
150
200
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.6, R = 0.999
1 2 3 4 5 6 7 8 9 10
0
200
400
600
Hops
R
C

T
x
-
R
x

E
v
e
n
t
s
PDR = 0.3, R = 0.999
31



Basic Acknowledgments Summary
  Changes at R=0.9  Compared to R=0.999  Changes at R=0.99999 
   Lowest Mean Tx‐Rx Actions 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same  No ACK  All   
same  0.9  No ACK  1 to ~4  Implicit 
Implicit/all  All   
Lower 
same  Immediate  All   
   
  Lowest Reliability‐Case Tx‐Rx Actions 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same  No ACK  All    same 
0.9 
No ACK  1 to ~6  Implicit/ 
Immediate 
No ACK  1 to ~2  Immediate  Immediate  All   
Lower 
same  Immediate  All    same 
   
  Lowest Mean Route Congestion 
PDR  4 Hops  10 Hops  4 Hops  10 Hops  4 Hops  10 Hops 
~1.0  same 
No ACK  No ACK 
same 
0.9 
No ACK/ 
Implicit/ 
Immediate 
same 
Implicit/ 
Immediate 
Implicit/ 
immediate 
Lower 
same  Immediate  immediate 
   
  Lowest Mean Tx‐Rx Events 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same 
Immediate/
No ACK 
All   
same 
0.9 
Immediate  All   
Lower 
Immediate  All   
 
 
  Lowest Reliability‐Case Tx‐Rx Events 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same 
Immediate  All   
same  0.9 
Immediate  All   
Lower 
Immediate  All   
Notes: Protocol1/Protocol2… means that the two are very close 
 
Table 5. Summary of basic acknowledgments and changes that occur at higher and lower reliabilities. 



32

End‐to‐End Acknowledgments Summary
  Changes at R=0.9  Compared to R=0.999  Changes at R=0.99999 
   Lowest Mean Tx‐Rx Actions 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same 
End‐to‐End  All   
same  0.9 
End‐to‐End  1‐~5  Implicit/All 
Lower 
Immediate  All   
   
  Lowest Reliability‐Case Tx‐Rx Actions 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same  End‐to‐End  All    same 
0.9 
End‐to‐
End 
1 to ~3  Implicit  Immediate/ 
Implicit 
All    Immediate  All   
Lower 
same  Immediate  All    same 
   
  Lowest Mean Route Congestion 
PDR  4 Hops  10 Hops  4 Hops  10 Hops  4 Hops  10 Hops 
~1.0 
same 
End‐to‐End  End‐to‐End 
same 
0.9 
End‐to‐End/ 
Implicit 
Implicit/ 
immediate 
Lower 
Immediate  immediate 
   
  Lowest Mean Tx‐Rx Events 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same 
Any except 
Separate, 
Implicit 
All   
same 
0.9 
Immediate 
/Unc EE 
1 to ~6  Immediate 
Lower 
Immediate  All   
   
  Lowest Reliability‐Case Tx‐Rx Events 
PDR  Which  Hops  Then  Which  Hops  Then  Which  Hops  Then 
~1.0 
same 
Any except 
Implicit 
All   
same 
0.9 
UC‐EE  1 to ~4  Immediate  Immediate  All   
Lower 
Imediate      Immediate  All   
Notes: 
1. Protocol1/Protocol2… means that the two are very close 
2. “UC‐EE” means Unconditional End‐to‐end 
3. Immediate, Separate, Implicit all imply +EE for this table only. 
Table 6. Summary of end‐to‐end acknowledgments and changes that occur at higher and lower reliabilities. 
 

33

REFERENCES
[1] D. Aguayo, J. Bicket, S. Biswas, G. Judd, R. Morris, “Link-Level
Measurements from an 802.11b Mesh Network,” in Proc. SIGCOMM
’04, Portland, Oregon, Aug. 30-Sep. 3, 2004.
[2] E. Ayanoglu, S. Paul, T. F. LaPorta, K. K. Sabnani, R. D. Gitlin,
“AIRMAIL: A link-layer protocol for wireless networks,” ACM
ACM/Balzer Wireless Networks J, vol. 1, pp. 47-60, February 1995.
[3] H. Balakrishnan, V. N. Padmanabhan, S. Seshan, and R. H. Katz, “A
Comparison of Mechanisms for Improving TCP Performance over
Wireless Links,” in Proc. ACM SIGCOMM ’96, Stanford, CA, August
1996.
[4] A. Cerpa, N. Busek, and D. Estrin, “Scale: A tool for simple
connectivity assessment in lossy environments,” Technical Report
0021, Center for Embedded Networked Sensing, Univ. of California,
Los Angeles, Sept. 2003.
[5] D. De Couto, D. Aguayo, J. Bicket, and R. Morris, “A High-
Throughput Path Metric for Multi-Hop Wireless Routing,” in Proc.
MobiCom ’03, San Diego, CA, September 2003.
[6] L. Doherty, W. Lindsay, J. Simon, “Channel-Specific Wireless Sensor
Network Path Data,” in Proc. ICCN 2007, Honolulu, Hawaii, August
2007.
[7] A. DeSimone, M. C. Chuah, and O. C. Yue, “Throughput
Performance of Transport-Layer Protocols over Wireless LANs,” in
Proc. Globecom ’93, Houston, TX, December 1993.
[8] Z. Fu, H. Luo, P. Zerfos, S. Lu, L. Zhang, M. Gerla, “The Impact of
Multihop Wireless Channels on TCP Performance,” IEEE
Transactions on Mobile Computing, vol. 4, no. 2, pp. 209-221,
Mar/Apr, 2005
[9] D. Ganesan, B. Krishnamachari, A. Woo, D. Culler, D. Estrin, and S.
Wicker, “Complex behavior at scale: An experimental study of low
powered wireless sensor networks,” Technical Report UCLA/CSD-
TR 02-0013, 2002.
[10] D. Messina, M. Ortolani, and G. Lo Re, “Achieving Robustness
through Caching and Retransmission in IEEE 802.15.4-based WSNs,”
in Proc. ICCCN 2007, August 2007.
[11] C. Parsa, J. J. Garcia-Luna-Aceves, “TULIP: A Link-Level Protocol
for Improving TCP over Wireless Links,” in Proc. WCNC 1999, New
Orleans, LA, September 1999.
[12] K. Pister, University of California, Berkeley, CA, private
communication, October 2008.
[13] J. Polastre, J. Hui, P. Levis, J. Zhao, D. Culler, S. Shenker, I. Stoica,
“A Unifying Link Abstraction for Wireless Sensor Networks,” in
Proc. SenSys ’05, San Diego, CA, November 2005.
[14] J.H. Saltzer, D.P. Reed, D.D. Clark, “End-to-End Arguments in
System Design,” ACM Transactions in Computer Systems 2, 4,
November, 1984, pp. 277-288.
[15] G. W. Shaw, “Determining Hard Time Bounds for Reliability
Delivery in Wireless Sensor Networks,” unpublished.
[16] K. Srinivasan, P. Dutta, A. Tavakoli, P. Levis, “Some Implications of
Low Power Wireless to IP Networking,” in Proc. HotNets-V, Irvine,
California, November 2006.
[17] Z. Fu, H. Luo, P. Zerfos, S. Lu, L. Zhang, M. Gerla, “The Impact of
Multihop Wireless Channels on TCP Performance,” IEEE
Transactions on Mobile Computing, vol. 4, no. 2, pp. 209-221,
Mar/Apr, 2005
[18] A. Woo, T. Tong, D. Culler “Taming the Underlying Challenges of
Reliable Multihop Routing in Sensor Networks,” in Proc. SenSys ’03,
Los Angeles, CA, November 2003.
[19] J. Zaho, R. Govindan, “Understanding Packet Delivery Performance
In Dense Wireless Sensor Networks,” in Proc. SenSys ’03, Los
Angeles, CA November 2003.

IV.  A.  B.  C.  D.  E.  F.  V.  A.  B. 

Comparison of Basic Acknowledgements Protocols9  Bounding Values ................................................... 10  Bandwidth .............................................................. 11  Node and Network Energy .................................... 11  Resource Usage ..................................................... 12  Reliability .............................................................. 12  Latency .................................................................. 13  Comparison of End-to-End Acknowledgements ... 14  Extraordinary Events ............................................. 14  When to Use End-to-End Acknowledgments ........ 14 

Contents 
I.  Introduction................................................................... 1  A.  B.  C.  II.  III.  A.  1)  2)  3)  4)  5)  B.  1)  2)  3)  4)  5)  C.  D.  1)  2)  3)  4)  5)  E.  1)  2)  3)  4)  5)  F.  1)  2)  3)  4)  5)  Related Work ........................................................... 1  Reliability ................................................................ 2  Metrics ..................................................................... 2  Terms and Assumptions .......................................... 3  Basic Acknowledgements ........................................ 4  No Acknowledgement Protocol ............................... 4  Reliability-Case Tx-Rx Actions .......................... 4  Mean Tx-Rx Actions ........................................... 4  Route Congestion ................................................ 4  Reliability-Case Tx-Rx Events ............................ 4  Mean Tx-Rx Events ............................................ 4  Only End-To-End Acknowledgement ..................... 4  Reliability-Case Tx-Rx Actions .......................... 5  Mean Tx-Rx Actions ........................................... 5  Route Congestion ................................................ 5  Reliability-Case Tx-Rx Events ............................ 5  Mean Tx-Rx Events ............................................ 5  Link-Layer Acknowledgement ................................ 6  Implicit Acknowledgment ....................................... 7  Reliability-Case Tx-Rx Actions .......................... 7  Mean Tx-Rx Actions ........................................... 7  Route Congestion ................................................ 7  Reliability-Case Tx-Rx Events ............................ 7  Mean Tx-Rx Events ............................................ 7  Immediate Acknowledgment ................................... 8  Reliability-Case Tx-Rx Actions .......................... 8  Mean Tx-Rx Actions ........................................... 8  Route Congestion ................................................ 8  Reliability-Case Tx-Rx Events ............................ 8  Mean Tx-Rx Events ............................................ 8  Separate Acknowledgment ...................................... 8  Reliability-Case Tx-Rx Actions .......................... 8  Mean Tx-Rx Actions ........................................... 8  Route Congestion ................................................ 8  Reliability-Case Tx-Rx Events ............................ 8  Mean Tx-Rx Events ............................................ 9 

C.  End-To-End Acknowledgement with Link-Layer Acknowledgements ......................................................... 15  1)  2)  3)  4)  5)  6)  VI.  VII.  Calculations ....................................................... 15  Bandwidth ......................................................... 15  Node and Network Energy. ............................... 15  Resource Usage ................................................. 15  Reliability .......................................................... 16  Latency .............................................................. 16  Cumulative or Selective Acknowledgments .......... 16  Conclusions............................................................ 16 

References........................................................................... 33 

.

the author uses collected or live packet delivery rate data to evaluate the reliability of a route of links given time-varying packet delivery rate information and determines hard time bounds for packet delivery. Many papers look at some metric of energy consumption or efficiency. and a method for determining the time bounds that must be accommodated based on the estimated packet delivery rates so as to achieve the desired delivery reliability. However. Real systems. the level of reliability can only be reached or not. A. Acknowledgement protocols can significantly affect delivery reliability. rather than raise. Thus. It suggests a method for evaluating actual packet delivery reliability to estimate the packet delivery rates used for planning the network. don’t have unbounded time or resources. Real systems. but a desired level of reliability cannot be “created” by the protocol. they are even built into the hardware of many radios. however. that immediate acknowledgments perform generally best overall. The delivery reliability requirement and time bound are determined from the application-dependent delivery MTBF at which exceptional reliability measures can be tolerated. Conventional wisdom says that end-to-end acknowledgements are required for reliable operation. their behaviors under changing conditions. reliability meeting the application requirement can often be achieved without the overhead of an end-to-end acknowledgment. However. Many papers looks at various mesh routing systems. and while they both attempt to pick a good route. Reliability is determined by the packet delivery rate and the number of successful packet transmissions the protocol can achieve within the time bound. and conventional wisdom says that reliable operation requires these acknowledgements be end-to-end. Messina in [10] proposes a network-layer protocol called PERLA that uses implicit acknowledgements rather than link-layer acknowledgements with packet caching and 1 I. and the environments in which they perform best. the possible time-bounded delivery reliability for the route and for the network as a whole. INTRODUCTION A multi-hop wireless sensor networks (WSN) with greater than 0% packet delivery rate (PDR) can be made 100% reliable—when given an unbounded amount of time to achieve successful packet delivery. We limit our analysis to the fundamental behavior of the underlying mechanisms on a linear route. such as routing algorithms or network layer protocols. have their uses. but their requirement depends on several specific factors. It finds that by changing acknowledgment protocols. current work still uses other protocols and ignores the primary driver of delivery reliability. the packet delivery rate. In [17]. Reliability is generally achieved through packet delivery acknowledgements. implementing it with a link-layer acknowledgment is the most efficient.Reliability and Acknowledgements in Low-Power Wireless Sensor Networks George William Shaw Abstract— All wireless sensor networks with greater than 0% packet delivery rate (PDR) can be made 100% reliable— when given unbounded time to achieve successful packet delivery. The conclusions suggest that by all measures immediate acknowledgments are generally the most robust and efficient. reliability is the probability that a packet will be successfully delivered within the required time bound. We look at the fundamental operators on the time-bounded reliability of delivery: the work load of a protocol and the primary factor to which the work load is sensitive. or timebounded access to the wireless medium. braided multipath and others to some quality metric. they do not determine if the path they chose meets any outside criteria. the authors improve on ETX with MT. disjoint multipath. which looks ahead a hop and also considers the variance of the packet delivery rate in making its routing decision. This report contributes a fundamental comparison of the common underlying mechanisms used to acknowledge packet delivery. The exception is in [15]. The author did not find any references that simply analyzed the fundamental characteristics of the basic available mechanisms. however. directed diffusion. but not boundedtime packet delivery. so achieving a reliability of less than 100% must be tolerated. however. the authors define the metric ETX (expected transmissions) that uses live data for and to compute expected transmissions at the next hop and thus select the best route. In [5]. don't have unbounded time or resources. and that if an end-to-end acknowledgment must be used. but none determine or evaluate time bounded reliability. Related Work Most papers look at acknowledgments as part of some other function. such as a specific reliability or time bound. including braided diffusion. the reliability of the system can be increased by several orders of magnitude (to the limit of the inherent network reliability). They look at only one hop. Reliability is the probability that a packet will be successfully delivered within the required time bound. Using end-to-end acknowledgements adds network workload that can lower. They LL . the packet delivery rate. While both of these have the live data available.

In general. We will use the term Tx-Rx events when considering latency and otherwise use the term Tx-Rx actions to refer to activity involved. yet not reach the destination due to a failure of the packet to be forwarded at any hop. A realistic packet delivery rate causes implicit acknowledgements to start competitively but to degrade more quickly as the packet delivery rate drops compared to immediate acknowledgements. but high reliability cannot guarantee delivery. Thus. B. which is unrealistic for WSNs. In some cases. memory. the less work a protocol must perform to achieve a given reliability. and latency sensitive. Only if the specified time bound is larger than the latency required by that number of Tx-Rx events can the protocol achieve this reliability. only a few hundred milliseconds. while for time-critical packets the time bound may be . can be tolerated. did arrive at the destination. comes at a high resource cost. The time bound directly affects the delivery reliability since it limits the amount of work that the network may perform in attempting to deliver the packet. Further. their gains appear to come entirely from scheduled retransmissions that avoid collisions compared to the competing protocols for which they do not consider scheduling.retransmission from alternate local motes. The transmission and reception of packets is generally the largest resource consumer on a mote. is useful for values which are averaged over time. and the temporal character of this activity is counted in TxRx events. the more memory it consumes. the end-to-end acknowledgment. i. then the required delivery reliability is 1 1 60 24 99. the highest reliability solution for a particular route is generally an efficient packet acknowledgement protocol to move packets to the destination and an efficient end-to-end acknowledgement to verify that they were received at the destination. The longer a mote must hold a packet for possible retransmission. resource constrained. can lower the reliability of the network as a whole. We refer to values related to this level of work as reliability-case values. but it is still impossible to guarantee that this acknowledgement can be received within the time bound. An end-to-end acknowledgement is the only way to guarantee that a packet. The mean value. However. C.e. Thus if is 0. The median of the distribution tells us the number of Tx-Rx actions at which 50% of the total Tx-Rx actions are successful. The mean of the distribution tells us the average number of Tx-Rx actions expected to achieve the given number of successful receptions. the more energy it consumes. their behaviors under changing conditions.999. If the application sends one packet per minute and can tolerate a delivery MTBF of one delayed or failed delivery per day. A packet can be acknowledged at every hop. Determining the reliability-case number of TxRx actions to achieve reliability is generally a matter of finding the number of Tx-Rx actions (the point on the distribution) that the probability is that the network can achieve the given number or more of successful receptions. which may be as long as several minutes or more. processing power and radio bandwidth. or group of packets. Metrics In this analysis we are concerned with motes that are power-constrained (battery operated). and a method for determining the time bounds that must be accommodated based on the estimated packet delivery rates so as to achieve a desired delivery reliability. where exceeding the value causes the failure of a packet to be delivered within the required time bound. It suggests the method for evaluating actual packet delivery reliability from [15] for estimating the packet delivery rates used for planning the network. Reliability-case values specify the minimum resources required to achieve the desired reliability and are used where worst-cast resources must be evaluated.5. Using one term often implies information about the other. This report contributes a fundamental comparison of the common underlying mechanisms used to acknowledge packet delivery. the more reliable the network can be for a given time bound.9% of the time we will have at least the required number of successful receptions. They have limited battery life. and finds a higher reliability. such as retrying the entire packet sending process to as much as human intervention.0. However. The delivery reliability requirement and time bound are determined from the application-dependent delivery MTBF at which exceptional reliability measures. We assume packets transmissions are Bernoulli trials that fail or succeed with some probability based on the packet delivery rates. Sufficient Tx-Rx events must be accommodated to achieve the desired reliability . such as in packet buffers and delivery latency. Using scheduled acknowledgement as in TSCH might alone eliminate that advantage. 1. The more frequently a mote transmits the more radio bandwidth it consumes. their analysis assumes an ideal channel. it only makes successful delivery more likely. such as goodput or energy consumption. we want to find the number of Tx-Rx actions to be accommodated to ensure that 99. conversely. or a reliability of 0. The work performed is counted in Tx-Rx actions. Reliability Achieving high reliability is desirable. the last measure of reliability in guaranteeing packet delivery. Reducing the number of Tx-Rx actions 2 For non-critical packets the time bound may be the network timeout.93%. The longer a mote is awake or the longer the receiver or transmitter is on. the value of Tx-Rx events and Tx-Rx actions are the same. This creates a distribution of the probability of a given number of successes in a given number of trials (Tx-Rx actions). and the environments in which they perform best. and is frequently not needed to achieve the required application-level reliability.

or the arrival of a packet outside of the required time bound. is the packet delivery rate for an acknowledgement packet. routes. as a group. such that the data in the packet cannot be considered valid by the available measures. we assume all packet delivery rates are symmetric. Tx-Rx Actions count the occurrences of the operation of a transmitter and receiver on a pair of motes. through neighboring motes. For the most accurate estimation of TxRx events. but is not the case with epoch-based duty-cycling with epochs that are long compared to the length of a packet. . and 0. it can increase the battery life of the mote.[4] has showed by physically swapping pairs of motes that asymmetries can be due to reception sensitivities of the different motes. Reliability is defined as the probability that a packet will arrive at its destination within a specified time bound. Tx-Rx Actions are similar to expected transmissions. . and three link-layer acknowledgements: implicit acknowledgement. 3 . Tx-Rx Events count the periods of transmitter or receiver activity.6. Multiple parents in a mesh topology can improve the packet delivery rate but complicate evaluation of the resources used and are not treated in this report.9. only end-to-end acknowledgement. Trends for other lower and higher reliability are discussed in the text and in Table 5 and Table 6. 0.3 to characterize the protocols. It is a metric to evaluate latency and can be translated into seconds. is a constant in the network and independent of the success or failure of previous hops or previous attempts to transmit the packet on this hop. described by [5]. Transmission implies a receiver will attempt reception. to another mote. 0. or latency. We also make several simplifying assumptions: or and is a constant for all . Then. i.e. such as B-MAC. a reliability of R = 0. and to minimum transmissions. We will primarily compare per-packet acknowledgements (ACK) protocols. While asymmetric packet delivery rates have been reported by [9]. as discussed above. we are using predicted or statically measured values for design and planning. Several metrics are evaluated at a packet delivery rate of 1. Reliability-case values are the minimum amount of an item to be accommodated to achieve the desired reliability . Failure is the lack of or incorrect reception of a packet for any reason. II. This is consistent with some of the most popular MAC layers. and frequencies. For the graphs in this report. one can expect reasons. we will also look at the case for end-to-end acknowledgments and review link-layer protocols that include an end-to-end acknowledgement per data packet. Hop is a counting measure referring to the number of wireless links required to transfer a packet from one mote. A single packet sent from the network level to the link level and into the network may originate the same packet several times if required by the acknowledgement protocol. acknowledgement packets are generally shorter than data packets.. for several In most cases. Destination is the last mote in the sequence of motes that receives a packet that originates at the source. These selections describe limiting behavior for other protocols and set the expected bounds of the system. so the contribution to bit error rate due to Gaussian noise is reduced. SCP. inconsistency in the quality of the receiver hardware. As a group.can not only increase the packet delivery reliability. Acknowledgements (ACKs) are packets transmitted to confirm the arrival of a data packet and are only transmitted after the successful reception of a packet (positive acknowledgements).0. TERMS AND ASSUMPTIONS Originate is used to refer to a link-layer transmission of a packet from the source mote into the network toward the destination mote that may entail many transmissions en route. Packet delivery rate (PDR) is the probability that a packet will be successfully transferred from one mote to an immediate neighbor. the delay per successful hop. MT. we will look at protocols that transmit no acknowledgement. Thus. is the packet delivery rate for a data packet. Each node has a single parent and is routed in a linear topology. the battery life of the network and the reliability of the network as a whole. (Re)Transmission refers to data transfers between motes on the route to the destination. but by reducing the resources used. times. If the acknowledgment is transmitted soon after the data packet then the channel coherence that made the data packet successful is maintained while CAA has suppressed the transmissions from local nodes [16]. described by [17] as routing metrics. When sent as an explicit packet. the packet delivery rate should consider CCA failures and scheduling delays as packet delivery failures that lower the packer delivery rate. Source is the mote that originates a packet and is thus the first mote in the sequence of motes on the route that transfers the packet to the destination.999 is used unless labeled otherwise as this represents greater than one failed delivery per day using the example above of one transmission per minute. ETX. except they are calculated to greater detail and rather than using dynamically measured values per link to plan routes. regardless of how many transmitters or receivers are involved. All probability distributions are independent and identically distributed. Route refers to the entire sequence of motes that transfers a packet from the source mote to the destination mote. immediate acknowledgement. TSMP. TSCH. and separate acknowledgement.

· . However. B.9 35%. No Acknowledgement Protocol A packet can be reliably sent with a no-acknowledgement protocol to achieve some reliability by unconditionally originating the packet the reliability-case times to achieve a probability of delivery greater than . Since these probabilities sum to one. The majority of the packets are assumed to be lost somewhere in route. the technique might at first be considered to reduce resource usage. and accumulated towards the source as if to create the cumulative distribution function (CDF). Only End-To-End Acknowledgement A packet can be reliably sent with an only end-to-end acknowledgement protocol to achieve some reliability by conditionally or unconditionally resending the packet some 4 The expectation value for Tx-Rx actions for each packet sent along the route is . 1 Thus. the probability that the single packet sent from the source mote arrives at the destination mote 0. If then the packet need only be originated once. With its simplicity. and is commonly used in single-hop home sensor networks where high reliability is not a concern. The probability that all packets are lost is the overall error rate.9. Multiplying by creates the Tx-Rx actions distributed across the route. As a result. BASIC ACKNOWLEDGEMENTS A. If we originate the packet some is multiple times. 1 The probability of at least one successful reception which is all that is required. . . 4) Reliability-Case Tx-Rx Events Since there is no acknowledgement to wait for. the reliability-case Tx-Rx actions to be accommodated to ensure reliability for hops is the packet originations multiplied by the expected Tx-Rx actions for each packet. at a 90% packet delivery probability of rate and ten hops. with no acknowledgement to confirm delivery. the mean packet originations is also . For the no-acknowledgement protocol. all packets must always be transmitted to assure reliability is achieved. Multiple copies of the same packet are originated and generally received to ensure reliability is met and only in about 1 in 1000 cases will just one packet reach the destination.III. the probability of Tx-Rx actions is # 3) Route Congestion The route congestion is computed using the distribution produced by # . the packet originations log 1 log 1 1 1 for reliability . 1 5) Mean Tx-Rx Events As with the mean Tx-Rx actions. the CDF is summed to itself shifted one hop toward the destination and then renormalized. The probability of Tx-Rx actions traversing the route is the same distribution plus the probability of success reaching the last mote. However. is . the distribution is inherently normalized. so the reliability-case and mean values are the same. the time bound for reliability requires that the data packet be delivered. mote . 1) Reliability-Case Tx-Rx Actions An originated packet fails along its route following a geometric distribution. the packet must be originated sixteen times. so we must include the time for the last packet to trickle through the route to the destination. of 0. A single data packet traveling hops to its destination with no retransmissions at any hop will arrive with a . packets can be originated with no delay. · 2) Mean Tx-Rx Actions While the packet will on average be received at the destination after some smaller number than packet originations .999 over ten hops To achieve a reliability with 0. Once all packets have been transmitted the resources on the source mote are freed for other activity. Since the source only transmits and the destination only receives. Thus. is also the reliability . It is for the assurance of the 1 in 1000 case that the sixteen packets must be originated in this scenario. we can increase the delivery probability. # · . all packets are always transmitted. There are many overlapping Tx-Rx events as the packets move through the network. The reliability-case and mean Tx-Rx events have the same value. · . with no acknowledgment to confirm delivery.

delivery probability and transit time from the initiating data packet. and a higher probability of multiple copies of the packet reaching the destination. but suffers in delivery latency because a failed-acknowledgement timeout must occur before the packet is originated again. 2 1 The expectation value for Tx-Rx actions is 2 ·2 . Then. · 2) Mean Tx-Rx Actions The mean Tx-Rx actions are the mean originations that occur multiplied by the expected Tx-Rx actions for each. the distribution is folded around the destination to create the backtracking. then the source must wait for 2 Tx-Rx events to determine that the end-to-end acknowledgment has not been received and all reliability-case packet originations must be accommodated. Conditionally resending the packet has the advantage that only the required Tx-Rx actions occur. If the data packet is conditional originated. for a packet-acknowledgement round-trip probability of . Using the prior no-acknowledgement protocol scenario of a ten-hop distant route and 90% packet delivery rate. The end-to-end acknowledgement packet is independent in scheduling. the minimum Tx-Rx events is that required to transmit all packets and for the acknowledgement from the last one to be received at the source. the CDF is summed to itself shifted one hop to toward the destination and then renormalized. The probability that the data packet originated from source mote A arrives at destination mote B is again about 35%. ·2 If the data packet is unconditionally originated for all of the reliability-case packet originations. 5 . then the source must wait for 2 Tx-Rx events to determine that the end-to-end acknowledgment has not been received for the mean packet originations needed to achieve success. # This value is plotted in Figure 18. but the probability that the acknowledgement packet originated from destination mote B is also received at source mote A is as low as about 12%. The probability that a data packet originated from source mote A arrives at . If the data packet is conditional originated. When the sender of a data packet requests an end-to-end acknowledgement then the destination must create and transmit an acknowledgement packet back through the network. the total hops for the packetacknowledgement round-trip is twenty hops. but suffers from the possibility of unneeded transmitting of the packet. The calculation of the reliability-case packet originations that must be accommodated to achieve reliability is the same as for the no-acknowledgement protocol. since the returning acknowledgement is assumed to traverse the same route. except that for a hop destination 2 hops must be successfully traversed since success requires the acknowledgement packet to also be received. There are many overlapping Tx-Rx events as the packets move through the network and the acknowledgments return. there is still a probability (about 65%) that the data packet will be 1 originated one or more additional times because the acknowledgement packet did not complete the trip from destination mote B to source mote A. However. In general. Thus. routing. Unconditionally originating the packet has the advantage of reducing the delivery latency because the packet can be transmitted without waiting for a failed-acknowledgement timeout. 5) Mean Tx-Rx Events The protocol can either conditionally or unconditionally originate a duplicate packet.zero or more times until delivery is verified by reception of the acknowledgement packet. There are no retransmissions en route. 4) Reliability-Case Tx-Rx Events The protocol can either conditionally or unconditionally originate a duplicate packet. the packet originations for reliability is log 1 log 1 . Even if the data packet is received correctly at destination mote B. 1 · 3) Route Congestion The route congestion is computed using the distribution produced by # for 2 hops and accumulating towards the source as if to create the cumulative distribution function (CDF). the number of reliability-case Tx-Rx actions to be accommodated to ensure reliability for hops is the packet originations multiplied by the expected Tx-Rx actions for each packet. 1) Reliability-Case Tx-Rx Actions For the only end-to-end acknowledgement protocol. Since the source only transmits and the destination only receives. whether conditionally or unconditionally originated. multiplying by creates the Tx-Rx actions distributed across the route. multiple copies of the data packet are received at the destination and by this measure more packets will be originated than necessary. There are no retransmissions en route. unneeded Tx-Rx actions across the network. and the probability that the destination mote B is end-to-end acknowledgement packet transmitted from destination mote B arrives at source mote A is .

If the data packet is unconditionally originated. though many systems have limits. . Retries occur at each hop as the transmission succeeds or fails. acknowledgement activity also consumes resources so this number does not tell the whole story. 1. 1 2 1 This value is plotted only in Figure 17. . The data packet is transmitted some number of additional times because the acknowledgement packet is not successfully received. There are many overlapping Tx-Rx events as the packets move through the network and the acknowledgments return. the probability of successful data packet and the probability of successful data delivery is packet and acknowledgement packet delivery is · . To ensure that the data packet is delivered and that all acknowledgement activity is complete within the time bound. Link-Layer Acknowledgement A data packet can be reliably sent with link-layer acknowledgements to achieve some reliability by conditionally retransmitting the data packet at each mote in the route that does not receive. and are thus Bernoulli trials. The probability of exactly retries to achieve hops is a negative binomial distribution given by . Tx-Rx actions are reduced with all of the subsequent advantages. the minimum Tx-Rx events is that required to transmit the mean originated packets and for the acknowledgement from last packet to be received at the source. we must compute the number of both. Thus. B. Order of Tx‐Rx events across hops A. . within the acknowledgement timeout. Because the only redundant packets ever transmitted are when the data packet has successfully traversed the one hop but the acknowledgement packet has not successfully returned the one hop.  . The probability of success on the route from source mote A to destination mote B is also the reliability of that route. . an acknowledgement packet from its neighbor indicating successful reception. C. link-layer acknowledgments are justified because they improve performance [14]. since both a data and an acknowledgement Tx-Rx action can occur. We assume that duplicate data 6 The probability of sending a packet from source mote A to destination mote B and incurring up to retries to achieve hops is the cumulative distribution function (CDF) of the negative binomial distribution. Since only one hop is involved for acknowledgement. the probability of delivery and the computation of the retries to be tolerated to achieve hops with reliability must be that of the successful delivery of the data packet and the successful delivery of the acknowledgement packet. Figure 1. To determine the Tx-Rx actions. C. with the trials ending in success on the last trial at the destination mote B. . . D for  basic acknowledgment protocols. We assume that the data packet is retransmitted at each mote as many times as required to achieve successful delivery. 1 · 1 . The probability of success of the data packet being forwarded and is through the network at each hop is considered independent of the success of the acknowledgement packet. We perform an inverse CDF by computing the sum until the to compute retries to be accommodated. Link-layer acknowledgements thus have an advantage over no-acknowledgements or only end-to-end acknowledgements because only local motes are involved in the retransmission and thus fewer data packets are retransmitted due to failed acknowledgement packets. While this represents the Tx-Rx actions to forward the data packet through the network. even with an application-level verification mechanism.

The mean Tx-Rx actions for hops are generally computed as · Since the forwarded packet used for acknowledgement is a data packet and is independent in scheduling and delivery . This saves resources required for transmitting an explicit acknowledgement packet. the probability probability. The second term accounts for the acknowledgement packet Tx-Rx actions and considers that some acknowledgment packets are not successful. and with the addition of a value for the explicit acknowledgment transmitted at the last hop. 1 ∑ The vector represents the pattern of Tx-Rx actions across the route while the scalar summation normalizes the distribution. Since failures are distributed throughout the route.5 1 · · 4) Reliability-Case Tx-Rx Events Since the acknowledgment occurs at the same time as the data packet is forwarded. and mote A implying an acknowledgement of successful delivery of the packet by overhearing mote B forwarding the packet to the next mote C. D. 1 ∑ … · · 3) Route Congestion Since the implicit acknowledgment does not transmit an explicit acknowledgement when a packet is forwarded but does receive a packet for acknowledgement. However. While they do not overhear the packet being forwarded. they wait for a return data packet to carry an acknowledgment. 7 1 1. but the components may be grouped differently to represent temporally separated events. but is similar in form to: . except for the last hop since the packet is not forwarded. 1 · 2 1 The first term accounts for the data packet Tx-Rx actions. 5) Mean Tx-Rx Events The first data packet Tx-Rx event and last explicit acknowledgement Tx-Rx event are separate. mote A must store the data to recognize the forwarded packet. Instead. when an explicit acknowledgment must be transmitted. The reliability-case and mean Tx-Rx events are computed similarly to the Tx-Rx actions. The source and the destination. Implicit Acknowledgment The link-layer implicit-acknowledgement protocol consists of transmitting a data packet one hop from mote A to mote B. . making these networks hybrids. do not forward the packet.5 … 1. Multiplying by the mean allocates the Tx-Rx actions. it does not add latency and counts as the same event except on the last hop.packets received due to unsuccessful acknowledgement packet reception are dropped or do not interfere with timing. so they generally incur half the activity. This is also part of the type of acknowledgement structure performed by systems that piggyback acknowledgments on returning data packets. while all the Tx-Rx actions in the middle are combined data-packet forwarding/acknowledgment events. . 1) Reliability-Case Tx-Rx Actions Each acknowledgement en route adds only one-half TxRx action for the overheard forwarded packet. however. we assume that the data packet transmitted from mote A is received at mote B and the packet forwarded by mote B to mote C is received by mote A is · . middle motes incur one-half a Tx-Rx action. Thus. 2 2) Mean Tx-Rx Actions The mean Tx-Rx actions count the same types of activity as the reliability-case values. and will generally need to listen to a data-packet length transmission rather than an acknowledgement-packet length transmission. except using mean values. The term accounts for the number of data packet Tx-Rx actions and the term accounts for the number of acknowledgement packet Tx-Rx actions. they also do not transmit an explicit acknowledgement. the average activity for all motes is spread evenly across the route. 1 Networks that piggyback acknowledgments on return data packets will need an additional factor to account for the delay until a return data packet occurs. Many also will transmit a separate acknowledgment packet should a data packet not occur soon enough. The approximation of the total Tx-Rx actions varies with the three link-layer protocols below.

Many also will transmit a separate acknowledgment packet should a data packet not occur soon enough. we assume acknowledgement packet is always successful. except using mean values. making the networks hybrids. 2) Mean Tx-Rx Actions The mean Tx-Rx actions count the same types of activity as the reliability-case values. 2) Mean Tx-Rx Actions The mean Tx-Rx actions count the same types of activity as the reliability-case values. 3) Route Congestion 4) Reliability-Case Tx-Rx Events Since the acknowledgment occurs immediately after the data packet is successfully received it does not add latency and counts as the same event. This acknowledgment type also occurs when acknowledgments are performed at a higher level in software. As a bounding lower-limit of acknowledgement reliability . . See Bounding Values. they are all also separate events. 3) Route Congestion 1 ∑ 1 2…2 1 · 1 2…2 1 · 1 ∑ · . certainly in applications where data packets and acknowledgment packets are of similar length this would be nearly typical. i. As a bounding upper-limit of acknowledgement reliability 1. This saves resources by reducing latency to schedule an acknowledgement packet and greatly increases the likelihood of successful acknowledgement packet delivery due to channel coherence. it is not far from correct with reasonable packet delivery rates. See Bounding Values. .1 2 As above. 0 5) Mean Tx-Rx Events As above. and then mote B scheduling and transmitting a separate acknowledgement packet to mote A before forwarding the data packet to the next mote C. E. and then mote B immediately transmitting an acknowledgement packet to mote A. that an for link-layer protocols. . Thus. This is the acknowledgment type performed by hardware acknowledgments as well as the acknowledgment type performed in some TDMA systems. Thus. While this represents a pessimistic representation. F. · 4) Reliability-Case Tx-Rx Events Since all the Tx-Rx actions are separate. Immediate Acknowledgment The link-layer immediate-acknowledgement protocol consists of transmitting a data packet one hop from mote A to mote B. 0 While this is an idealistic representation. 8 . except using mean values. networks that piggyback acknowledgments on return data packets will need an additional factor to account for the delay until a return data packet occurs. Separate Acknowledgment The link-layer separate-acknowledgement protocol consists of transmitting a data packet one hop from mote A to mote B. since the acknowledgment occurs immediately after the data packet is successfully received it does not add latency and counts as the same event. the probability that the data packet transmitted from mote A is successfully received at mote B and the acknowledgment packet transmitted from mote B is also received at mote A is: · . below. 1) Reliability-Case Tx-Rx Actions Since the acknowledgment packet is always successful there are no retries and only one acknowledgment packet is transmitted per hop. below. the for link-layer protocols we assume probability that the data packet transmitted from mote A is received at mote B and the acknowledgment packet transmitted from mote B is also received at mote A is: · 1) Reliability-Case Tx-Rx Actions .e. and is nonetheless positions correctly as a bounding value with respect to the other protocols.0.

but individual links fared much worse. The solid line is a 5‐hour  moving average. The basic acknowledgment protocols reviewed vary in work load imposed on the network with the number of hops and the available packet delivery rate. immediate.. Clearly one could not select one “good” channel and expect a stable packet delivery rate. Figure 3 shows the packet delivery rate data for one link collected over 26 days from an industrial site in Berkeley. All channels show packet delivery rate variations lasting for days. Note that many links have a significant probability of lower packet delivery rates. and 14 had high packet delivery rates for the first 13 days with drops in packet delivery rates for the days after. there are no collisions from several nodes trying to share the same channel.4 channels (0-15 on the graph mapped to 11-26 in the standard). [6]  on the order of days. and the graph shows the link packet delivery rate (stability) over all 16 IEEE 802.  design and choose Table 1. channels 5 and 12 experienced drops in packet delivery rates in the first 13 days and high packet delivery rates for the days after. Thus while the networkwide packet delivery rates Protocol Sensitivity to  averages in the Changes in PDR  mid-80% or above Order  Protocol  even in the worst Least  Immediate  times. IV. one even a notable amount in the 0-10% range. While the number of hops is in part a hardware architectural decision. 3. Since the network is a type of TDMA. Figure 2 shows the average packet delivery rate of the entire network of 44 nodes and 89 links over the same period [6]. Note that channels 0. . however. These packet delivery rate fluctuations are persistent Figure 2. This data is a network average. 5) Mean Tx-Rx Events Since all the Tx-Rx actions in the mean are separate. Even the normally expected quiet channel 24 (13 on the graph) varies constantly throughout the test period. Note that for the first two weeks of the period the 5-hour average of the packet delivery rate was around 95% most of the time. 7.[6][18] As reliability   increases. COMPARISON OF BASIC ACKNOWLEDGEMENTS PROTOCOLS We first look at five common acknowledgment protocols to reliably deliver packets: none. 9. and the detail below the line shows how much the packet delivery rate has dropped. only end-to-end.  We evaluate a sensitivity increases. 13. or even the best packet delivery rate. and 2. but for the remainder of the period dropped significantly to as low as 83% and then recovered on the last couple of days. California [6]. it does affect the packet delivery rate as the physical distance between motes and their location affects the signal strength and signal multipath [1]. Protocol Sensitivity  9 . Overall network PDR (stability) as a function of  time throughout the sample period with each point  representing one 15‐minute interval. the author analyzes this same site data. Figure 6 shows the probability density function of the packet delivery rates collected in 10% intervals. the   Implicit  individual link data   Separate  tells a different Most  Only End‐to‐End  story. 15 had constantly varying packet delivery rates. implicit. The baseline is 100% packet delivery rate. 8. they are all also separate events. [6]  Figure 3.15. The PDR (stability) variations on the link from  node 24 to node 17 shows daily periodicity corresponding to  operational days in the factory. The resulting packet delivery rate. The variance of the packet delivery rate for this link might be improved by hopping through only the best channels rather than all the channels. 11. All of the packet delivery rate degradation is from the environment. The network is time-slotted channel-hopping (TSCH). 10. In [15]. and separate. varies greatly over time.

1. See the bar graph for link one in Figure 6. for all links the probability must be 90% or more that a packet delivery rate of at least 90% occurs.4  7. then these lower packet delivery rates will occur. Not all links on a route may fare so poorly.9  26.3        57. Large bounding values for differences in the length of data and acknowledgment packets will show in the bit error rate (BER) and subsequent packet delivery rate up due to Gaussian noise effects.4    19. A.4  2.4  7.2  24. however.7e5  468  None    1. Link one appears to be a reasonable link relative to the mesh diagram.7e6  2. and 10 hops.7  9. We use two .9  None    19.5  21.999)  Separate  Implicit  Immediate    Mean Tx‐Rx Actions for 10 hops    4.5  3  Separate    94  24  8  Implicit    84  20  6.4  2.5  2  8  4  8  6  8  20  10  20  28  20  1. We use constant packet delivery rate for all links to simplify evaluation and believe that reliability-case values from packet delivery rates in the 60% to 90% range should represent likely overall behavior of most links.3  17.4  7.4  14. On the graphs for the basic acknowledgments. but note that the 90%-100% packet delivery rate interval has a probability of less than 90%.1  1.8e4  72.5e5  1000  69.1e10  6.4  8  4  8  6. If not. cause the time bound to be exceeded and the time-bounded delivery to fail more often than the desired reliability [15].8  52.4  4.  10 .3  44. some occurrences of packet delivery rates in the 50% to 60% interval or even the 40% to 50% interval will need to be tolerated.7  2. assume a fixed link packet delivery rate of 90%.3  0. In [15] the author discusses a heuristic to determine the time bound for a route from the time-varying packet delivery rate behavior on each link.9% or more. On the graphs of the end-to-end R=0.3  10.4  2. requiring more Tx-Rx events than we have allocated time for in the time bound.8e3  105        141  129  43.3    1.7  23. A review of the graphs in this report shows that the protocols have an overall sensitivity to changes in packet delivery rate as listed in Table 3.6  0.1  20  10  20  18.5  3        14. we use for the returning end-to-end acknowledgment packet. so packet delivery rates in the 80% to 90% interval will need to be tolerated on this link to achieve 90% reliability.7  8. 99. 4. multipath generally dominates BER in real environments (particularly indoors). Lower packet delivery rates will regularly occur on some links so the protocol must behave well at lower packet delivery rates for any reliability to be maintained. making differences much less than would be expected due to any packet length difference [1].8  17. and since they occur within the probability of our chosen reliability. Reliability‐case and mean Tx‐Rx actions for basic  acknowledgments by PDR for 1. Bounding Values The assignments of packet delivery rates to protocols were chosen to give the best range of evaluation for the available graphs and to simplify calculations.2e4  146  13.5  18.9  Reliability‐Case Tx‐Rx Actions for 1 hop  End‐to‐End    95.4  2. based on the early network-wide average packet delivery rate (stability) of above 90%.0 and . Mean values from time-weighted mean packet delivery rates are appropriate for other evaluations.0  2  1  2  2  2  End‐to‐End  None  (R=0.3  4. To achieve this design. but because the number of Tx-Rx events is generally an exponential function of the packet delivery rate. the range of the only end-to-end acknowledgement is at the line and just below.8  15.0  2  1  2  1.9  None    1200  108  22.8  7.999  PDR:  0.2e3  108  22.8e11  4.999)  Separate  Implicit  Immediate    Mean Tx‐Rx Actions for 4 hops    2.a time-bounded reliability of just 90% and.5  20  Table 2.7e6  2.3  0.8e3  105  Separate    314  86  36  Implicit    280  71  28  Immediate    79  40  26    End‐to‐End  None  (R=0.5  Immediate    21  9  4    Reliability‐Case Tx‐Rx Actions for 4 hops  End‐to‐End    1. At higher target reliabilities of 99%.9  Mean Tx‐Rx Actions for 1 hop    14.5  8  End‐to‐End  None  (R=0.2    1. This means that in determining the time bound for a route. In cases where the length of the end-to-end acknowledgment packet is much less than the length of the original data packet.4  36. an important characteristic since packet delivery rates are time-varying.4  4. using the average packet delivery rate of a network is not sufficient for design. these lower intervals must be the design values. then the plotted lines would be the upper boundary.3  Separate    180  47  18  Implicit    161  39  14  Immediate    43  20  12    Reliability‐Case Tx‐Rx Actions for 10 hops  End‐to‐End    2.999)  Separate  Implicit  Immediate  PDR:  0. For the only end-to-end acknowledgment protocol and all end-to-end with link-layer acknowledgments.4  4.6  0.

As can be known-good channel since the acknowledgment packet is seen from Figure 9 with the Tx-Rx actions representing the separately scheduled. As packet delivery rates 11 C. The system designer must packet delivery rates this may not be accurate [12].5 Tx-Rx actions since there are two receives and one transmitter active. though at lower data reliability case values (Figure 10). However.0. The immediate acknowledgment plotted with acknowledgment and only end-to-end acknowledgment range is thus the area between immediate and separate acknowledgments lines. as the packet delivery rate drops further. the requirement for the implicit acknowledgment protocol that both the data packet . relative total bandwidth usage along a route of hops for a For Tx-Rx events the reasoning is the same but the bounds single packet transfer. Note that in interpreting these graphs for bandwidth comparison. Tx-Rx actions need to be spread evenly across motes on a route (Figure 7 and Figure 8). as Immediate  Immediate to Implicit  Heavily Immediate  Separate  Separate and lower  Separate  error rates for the immediate Notes:  acknowledgment packet are No‐acknowledgment and Implicit protocols have no bounding value  reported to be about 1/5 the error   rate for data packets. toward their about 98%. events. As the packet delivery rate drops to 0. these protocols consume an are different due to difference in the computation of the impractical amount of bandwidth as the packet delivery rate drops or the hops increase. Each unit of the metric represents one transmitter and receiver in operation for one packet.0. all plots would Graph Bounding Regions  be affected similarly. The relative metric we use is mean Tx-Rx actions. 60% and 30% the acknowledgment packet delivery rate would be drop. Node and Network Energy Transmitting and receiving packets are generally the most energy consuming activates of a mote. End‐to‐End and Link‐Layer+EE  Self (as below) and lower  self  For the link-layer protocols. more links will require greater retries. The relative metric we use is Tx-Rx actions as a measure the bandwidth consumed by one packet transfer. Bandwidth To prevent bandwidth congestion. the requirement of the implicit acknowledgment that both the data packet and the . causes it acknowledgment packet be successful. Since the failure of the transmission of packets in the no. the bandwidth immediate acknowledgment protocol line. However. application source and destination requirements will place burdens on specific motes. all acknowledgments are redundant and unneeded. each such activity is counted as 1. Many current radios specify about the same current for both transmission and reception so the mean Tx-Rx actions on the graphs can be directly translated into energy usage by multiplying by an appropriate constant. so the no-acknowledgment protocol has the least bandwidth usage. See Table 3. The be concerned with having sufficient bandwidth to upper bound for the immediate acknowledgment protocol is accommodate the reliability-case Tx-Rx actions on many the line for the separate acknowledgement protocol. Tx-Rx actions need to be spread evenly across motes on a route (Figure 7 and Figure 8). To prevent nodes depleting their energy sources unevenly. respectively. the immediate acknowledgment  Tx‐Rx Events  protocol is a lower bound. the saved acknowledgment transmission of the implicit acknowledgement protocol again reveals its efficiency.5 Tx-Rx actions. thus the packet delivery rate is less. so the no-acknowledgment protocol has the least energy usage per mote and for the route. For Tx-Rx actions. For a data Table 3. At 1. This is not far off. Figure 2 shows that effects on packet delivery rate can be network-wide.0.9 the saved acknowledgment transmission of the implicit acknowledgement reveals its efficiency. acknowledgment protocol does not have the advantage of a increasingly so as the packet delivery rate drops. At 1. Graph Bounding Regions packet delivery rate of 90%. as the packet delivery rate drops further.9. As the packet delivery rate drops to 0. to lose ground to the immediate acknowledgment protocol. 92% and 86%. Since the implicit acknowledgment protocol performs the forward/acknowledgment in a single transmission with two receivers. which is links when necessary. Immediate  Immediate to Separate  Heavily Immediate  Separate  Separate to Immediate  Separate  there are two separate bounding   cases. the Tx-Rx action values for the implicit acknowledgment protocol are slightly high because the forward/acknowledgment single transmission is actually one bandwidth usage but is counted as 1. heavily weighted toward the protocol follow a geometric progression. all acknowledgments are redundant and unneeded.acknowledgments. The separate usage is similarly exponentially loaded toward the source. B. While we consider the probability of delivery at each hop as independent. Nevertheless. with End‐to‐End and Link‐Layer+EE  Self (as below) and lower  Self  1. so while the ACK Protocol  Bounds  Weight  lines are also upper bounds. all would range similarly lower  Tx‐Rx Actions  together.

etc. as is often the case. The values for the other motes on this route are of short duration distributed geometrically from the source. the energy usage is similarly exponentially loaded toward the source. The only end-to-end acknowledgment protocol (and other end-to-end protocols) can overcome the inherent reliability of the network and will have delivered the packet with reliability . similar to the distribution in Figure 7.99 0. Since the failures of transmissions of packets in the no-acknowledgment and only end-to-end acknowledgment protocol follow a geometric progression. Note that the immediate acknowledgment protocol has the lowest Mean Tx-Rx events. radio. For the link-layer protocols. That is. The first type include items such as the availability of CPU. all of the basic acknowledgment protocols can deliver the same reliability .9 to 0. Given a proper estimate of the packet delivery rate and an adequate time bound. toward their reliability case values (Figure 12). The no-acknowledgment and link-layer protocols cannot overcome the inherent reliability of the network and will deliver the packet with reliability . below. Further. The longer the route in hops and the lower the packet delivery rate.9. Figure 11 shows the total mean Tx-Rx events for a route hops.9999 Reliability 0. The system designer must be concerned with having sufficient resources to accommodate the duration of reliability-case Tx-Rx events on many links when necessary.  . D. Resource Usage Resource usage refers to the duration of usage measured in Tx-Rx events for two types: resources that are used during the short duration of a Tx-Rx event and those that are used for the long duration of a successful or failed delivery. program state and other items that are used while a packet is being processed.and the acknowledgment packet be successful. within a factor for the inherent reliability of the network . the radio will be activated. is the probability of delivering a packet within a given time bound. more links will require greater retries. Figure 9 shows mean Tx-Rx actions for an entire route for hops and they behave here as above. The case of is discussed in Extraordinary Events. other things being equal. these protocols consume an impractical amount of energy as the packet delivery rate drops or the hops increase. In Figure 9 with the Tx-Rx actions representing relative total energy for a single packet transfer. Tx‐Rx events by basic acknowledgments for  reliabilities from 0. by (if ) such that the final designing for reliability reliability · .999999 Figure 4. It also implies that. so can be neglected. If we misestimate the packet delivery rate.999 0. In general. many other packets may also be sent. As discussed above. while the source mote is waiting for the end-to-end acknowledgment on one packet.9 End-to-End None Separate Implicit Immediate Reliability-Case Tx-Rx Events 0. Reliability Reliability. . for this packet. temporary storage. The metric for both are units of time measured in multiples of Tx-Rx events. we assume 1. the CPU will likely wake up and go to sleep. 1. PDR=0. the effect on the available reliability varies greatly with the protocol. we have defined. This creates packet buffer resource usage accumulating over a long duration. every will succeed or fail (late or no delivery) with reliability or . Latency by Reliability of Route. As packet delivery rates drop. For the only end-toend acknowledgment protocol. the value is usage of resources on the source mote of long duration since the source mote must wait for the acknowledgment packet from the destination before it can release the buffer with the packet.999999. increasingly so as the packet delivery rate drops. which results from extraordinary events (discussed below). and represent how many times. 12 E.99999 0. The second type is primarily packet buffers of which both mean and reliability-case values are of interest. 4 hops 40 35 30 25 20 15 10 5 0 0. given a that for reliability requires up to Tx-Rx events to achieve that reliability. possibly spanning many Tx-Rx events. and we allow sufficient time to accommodate events. the packet delivery rate can degrade network-wide. Each of these protocols is a mechanism with differing work load requirements to perform the basic function of ensuring that the packet reaches its destination with time-bounded reliability . the more buffers can need to be in service concurrently. the values are of times the mean individual usage. causes it again to lose ground to the immediate acknowledgment protocol. This implies that resources are tied up the least amount and for the shortest duration. it can achieve the highest packet transfer rate.

8e4    289  88.5  4  keeping the time bound for reliability at 0. For most CSMA architectures.18e4  146    116  35. the area where the network will likely acknowledgment immediately there are simply fewer events operate. We would expect then that single-channel and CSMA systems could have lower effective packet delivery rates than TDMA or TSCH systems [8][18]. Appropriate metrics to requires for a given set of conditions.4  16. multiple TDMA slots reliability requires repeatedly transmitting a packet until it is can be allocated to a link. the more Tx-Rx events that a protocol transmitted as a separate packet.9  8.9  1  average energy.6 graphs.2          8  16  12  16  20  40  48.5  40  1.7  53.2  20  40  36. thus lowering the effective packet delivery rate. 4. By performing the 0.9  2. While at high packet acknowledgment protocol. as compare the mean Tx-Rx actions for implicit and immediate packet delivery rates drop the immediate acknowledgment acknowledgements between the 0.9  271  74  30  242  61  23.7  4.3  46. Reliability‐case and mean Tx‐Rx actions for end‐to‐end  acknowledgments by PDR for 1. and 10 hops.5  67  34  21    Reliability‐Case Tx‐Rx Actions for 10 hops  End‐to‐End    2. systems of similar network architectures can be compared For those same 12 Tx-Rx events.999  PDR:  0. further decreasing the latency. the less reliable the combine these models and account for the additional delays network becomes overall.6    34.9  40  Table 4.9    8.  13 . The ability of the network to do the Many systems are more complex.3  0.4  4. Since .7e5  468  Separate+EE    512  144  63  Implicit+EE    456  119  48. Alternatively. we can achieve over two orders delivery rates several protocols are similar in mean and of magnitude greater reliability.6  Mean Tx‐Rx Actions for 1 hop    14.3  0. then latency is · · . etc. the delivery latency of Given equal values of case Tx-Rx events for a reliability of 0. by choosing the immediate acknowledgment protocol R=0. Refer to Figure 9 and reliability-case Tx-Rx events (Figure 11 and Figure 12).7  4. Thus.3    Mean Tx‐Rx Actions for 10 hops    4.9    86.3  4  Mean Tx‐Rx Actions for 4 hops    2.9  8  16  13  16  is the maximum time between two slots in a frame.5e5  1000  69.1e10  6.8e11  4.8  7.7  21. In a TDMA system.9    259  73.999 at 0. the packet delivery rate used for determining Tx-Rx events (mean or reliability-case) should include the effect of CCA failures and scheduling delays.9    28.7  5. close in value within our operating range.6    106  30. and simplistically for TDMA architectures. Latency To simplify the translation of Tx-Rx events to latency in seconds. equal .8  15. by Separate+EE    126  33  12  2  Implicit+EE    113  27.5  Immediate+EE    133  71  49    End‐to‐End  Separate+EE  Implicit+EE  Immediate+EE  PDR:  0.0  over the implicit acknowledgment protocol we can Reliability‐Case Tx‐Rx Actions for 1 hop  improve the reliability of the network at the same End‐to‐End    95. otherwise the acknowledgement is packets.9  36. Note that the need to be devised for the behavior of each system.5  9. for critical routes. such as those that least amount of work to deliver a packet means that network piggyback the acknowledgement if a return data packet is to resources are available for retrying the delivery of other be transmitted soon. For similar reasons we would expect less variance in the latency in TDMA or TSCH systems. the latency can be less in TDMA compared to CSMA systems given End‐to‐End  Separate+EE  Implicit+EE  Immediate+EE  72. using the immediate directly in units of Tx-Rx events.9  42.9  1. bandwidth.0  2  4  4  4  is the time interval between transmission attempts on a link.999.Given the low packet delivery rates of wireless networks.4    28. if at least one transmission can be attempted within the same frame at each hop.9.2  18. we Immediate+EE    29  13  7  3  can accommodate a higher network load and still 4  have the resources to meet the reliability Reliability‐Case Tx‐Rx Actions for 4 hops  requirements.9 and the protocol has the clear advantage. F. The two protocols change places such that they are to schedule.4  4.2  24. the total latency is · . Thus. successfully delivered.3  0. The effects should be considered delivery failures for this purpose. Further. Refer to Figure 4. implicit acknowledgment protocol requires 12 reliability. End‐to‐End  Separate+EE  Implicit+EE  Immediate+EE  13.6  0. End‐to‐End  Separate+EE  Implicit+EE  Immediate+EE  1.9  8.

6 and 0.6 PDR 0. except in systems that drop packets when delivery is difficult—some systems have no retries or limit retries to as little as three [18]. An only end-to-end acknowledgment requires more mean and reliability-case Tx-Rx events than any other basic protocol reviewed under almost all conditions of PDR<1.99999 0. and far. This can occur due to hardware or software bugs. We capture the limited reliability due to these extraordinary events in an inherentreliability-of-the-network factor .9 PDR 0. or other transient or permanent failures.99999 0. (Or. node failure. However. endto-end acknowledgment protocols can only add the benefit of mitigating the inherent reliability . A.3 PDR 0. Adding an end-to-end acknowledgement to a linklayer protocol is generally more efficient. in [6]. we need 9 and 21 attempts.9 to 0.9999 Reliability 0. Tx‐Rx events by end‐to‐end acknowledgments  for reliabilities from 0.9+EE PDR 0.6+EE PDR 0. Referring to Table 2.3 for one hop. to have a 99. Compared to link-layer acknowledgment protocols. unreported buffer overflows. for the immediate acknowledgment protocol the mean number of transmissions is 2. Extraordinary Events Other than intentionally dropped packets.6 million generated were lost.3+EE PDR 0. often at much less than the typical reliability-design of the system. the reliability improvement 1 that is available is typically small.999999 Latency by Reliability of Route. but typically doubles the mean Tx-Rx actions at all packet delivery rates because twice as many hops are traversed.99 0. COMPARISON OF END-TO-END ACKNOWLEDGEMENTS End-to-end acknowledgments are often used as an absolute verification of delivery and are well suited to that task.6+EE PDR 0. far less frequently than delays due to normal packet delivery rates that can readily cause a packet to be delivered late. When to Use End-to-End Acknowledgments Per-packet end-to-end acknowledgments can be very expensive. However. node isolation. They cannot create 100% time-bounded reliability because they cannot guarantee that the acknowledgment packet will arrive within the time bound. However.999 0. the retries due to a missing end-to-end acknowledgment can supply the additional attempts to raise reliability. generally a very small additional factor of reliability.9 Figure 5.9 0. per-packet or selective end-to-end acknowledgments can be necessary.9999 Reliability 0.  14 .) For example. Latency by Reliability of Route.99 0. over a 26 day period at an industrial site.3 respectively. Immediate ACK.999999 PDR 0.999995. 4 hops 200 180 160 Reliability-Case Tx-Rx Events 140 120 100 80 60 40 20 0 0. in cases where the environmental or other conditions cause some extraordinary events to be ordinary.999999.V. link failures. conversely. such that the events become a material failure rate relative to the desired reliability . it can be said that inherent reliability can be high and extraordinary events can be rare.9% probability of success.6 PDR 0. Immediate ACK. From this. for an inherent reliability 0. The end-toend acknowledgment can then supply the mechanism to inform of the need to retry delivery.3 PDR 0. only 17 packets out of 3. Time bounds were not mentioned. end-to-end acknowledgments also add delivery security against events where a packet might be successfully delivered and acknowledged at an intermediate mote on the route to the destination and then might not be forwarded. Taken as a whole.7 and 4. 10 hops 180 160 Reliability-Case Tx-Rx Events 140 120 100 80 60 40 20 0. these events should occur very infrequently. B. The effect is not as direct on reliability-case Tx-Rx actions because twice as many hops result in less than twice as many trials.9+EE PDR 0. we find that at packet delivery rates of 0.9 0. but at the great expense of an end-to-end protocol rather than a more local approach.9 Only EE PDR 0.999 0. Generally.0. but the Dust Networks products used at the site essentially never stop retrying to deliver a packet so the effective time bound may have been large. if they occur more frequently a new sensor mote vendor should be selected.3+EE PDR 0. respectively.

In cases described above when an end-to-end acknowledgment is needed. For all link-layer acknowledgment calculations with hops . then we compute how many retries of those reliability-case values must be accommodated to obtain the desired reliability . so long as the timing for retries is adequate so that the behaviors of the two layers do not interact poorly [11]. then an end-toend acknowledgment protocol is need. apply 2 . but this advantage is exponentially eliminated and is unevenly distributed along the route (Figure 13) as the packet delivery rate drops. At this point our reliability would likely be limited by the inherent reliability .9. and may entail many buffers simultaneously as several end-to-end transactions may be in process. · log 1 log 1 . The only end-to-end acknowledgment protocol uses less bandwidth up to a few hops. While the only end-to-end acknowledgment fares better against the link-layer protocols with end-to-end acknowledgment than to the basic acknowledgment protocols. an end-to-end acknowledgment with link-layer acknowledgement results in better performance [2]. At PDR=0. The only end-to-end acknowledgment protocol requires fewer Tx-Rx actions and Tx-Rx events at extremely high packet delivery rates. occur. C. it should be on top of an efficient link-layer acknowledgment protocol (Figure 17. this advantage drops with packet delivery rate. though there is still an advantage in TxRx actions up to five hops. If the inherent reliability is low such that it encroaches on or is lower than the desired reliability . . end-to-end acknowledgments add significant stresses to the network that may elicit failures and lower the inherent reliability rather than increase overall reliability. 1) Calculations End-to-end acknowledgments with link-layer acknowledgments perform identically to their non-end-to-end counterparts. This implies that by using the linklayer alternatives. At ten hops the reliability improvement appears unlimited. Thus. the advantage quickly dies with increased hops or decreased packet delivery rate. However. However. First we compute the reliabilitycase values for an end-to-end acknowledgment protocol. and thus the number of end-to-end acknowledgment retries should be reduced.9+EE. The latencies are quite long. However. four hops. Since we cannot compute reliability greater than 1. the mean and reliability-case durations are significantly less in most cases for the link-layer acknowledgment alternatives (Figure 17. the average workload grows linearly rather than exponentially as it does for the only end-to-end acknowledgment protocol. we chose one of those for the internal primary link-layer protocol. we perform the computation in two layers. the route congestion in Figure 13 and Figure 14 flips for the only end-to-end acknowledgment protocol being the best at four hops to the worst at ten hops. The selected end-to-end acknowledgment protocol that had been designed (using the computations below) to deliver will need accommodation for retries reliability to achieve desired reliability . We can use equations for one hop of the only end-to-end acknowledgment protocol. on average. Figure 18). End-To-End Acknowledgement with Link-Layer Acknowledgements Adding a per-packet end-to-end acknowledgement to a link-layer protocol simply transfers an a packet the same hops back along the route from the destination to the source. 2) Bandwidth As discussed above. as has been found on wired connections.. Thus. They will lower the overall network reliability in that resources are unnecessarily consumed and unavailable. it appears we can obtain five orders of magnitude more reliability for the same number of Tx-Rx events by not using end-to-end acknowledgments. 15 Similarly. than the end-to-end acknowledgment with linklayer acknowledgment protocols. Further. since the packets are still moved forward with a link-layer acknowledgment protocol.9. If we start at a more reasonable value at R=0. 4) Resource Usage Using an end-to-end acknowledgment with a link-layer acknowledgment now causes all the protocols to have the same long-term packet buffers on the source mote. all endto-end protocols create significant buffer demands at the source mote that do not exist with only link-layer acknowledgments. then end-to-end acknowledgments should not be used. and Tx-Rx events against some link-layer alternatives up to three or so hops at 0. There are narrowly bounded conditions where the only end-to-end acknowledgment consumes less energy than using an end-toend acknowledgment with a link-layer-acknowledgment. While this creates additional activity. Refer to Figure 5.0.999 then we have at least three orders of magnitude before leaving the plot. Since link-layer with end-to-end acknowledgments are more efficient than the only end-toend acknowledgment protocol. potentially fewer extraordinary events. the metric for bandwidth is mean TxRx Actions (Figure 15). Figure 18). unless reliability exceeding the inherently reliability is required. 3) Node and Network Energy. except the return acknowledgment packet doubles the number of hops traversed. and a design reliability of 0. the highest reliability that is practical should be selected. If the inherent reliability is high such that and reliability is suitable. such as unreported packet buffer exhaustion. as described above. The same patterns already discussed follow here. Since incremental increases in reliability are exponentially less expensive. link-layer acknowledgements are far less sensitive to variations in hop depth or packet delivery rate and are thus more robust.

If the ratio of data packets to acknowledgments is high. In the reliability-case. too. The best in each area are summarized in Table 5 and Table 6. CUMULATIVE OR SELECTIVE ACKNOWLEDGMENTS Either cumulative or selective acknowledgments can reduce the overhead for the end-to-end acknowledgment packet by reducing the number that is transmitted. We use one-half because we want just to evaluate the Tx-Rx actions to reach the destination but not to return the end-to-end acknowledgment. few perform well across a range of network conditions. the exponential resource usage duration of the only end-to-end protocol can be mitigated by unconditionally retransmitting packets from the source rather than waiting for the failed end-to-end acknowledgment. but still exponential and exceeds the link-level alternatives after a few hops. the situation is more comparable. However. If the reliability requirements do not need an end-to-end acknowledgment protocol. When needed. VII. but there is no definite win for the new protocols. While many acknowledgment protocols are in use. at very high packet delivery rates as seen in the first half of Figure 2. and thus routes that include those links. When not taken into account. 6) Latency Latency follows the same behavior as Resource Usage and Latency previously discussed. packet delivery rates on many links drop well below the average for enough of the time that the lower packet delivery rates will dominate the packet delivery rate for those links. As Figure 6 shows. in general. Of the link-layer protocols. aside from the packet buffers required until successful delivery. However. but assume an infrequent and yet uncounted acknowledgment packet to indicate which packets will need to be retransmitted. which has significant mechanisms that assume the characteristics and reliability of a wired connection at the lower layers. . then much higher reliability is available from the basic link-layer protocol for the same number of Tx-Rx events (Figure 5). the only end-to-end acknowledgment protocol is comparable to the link-layer protocols at ten hops. Clearly. However. Thus. As such. the exponential growth is slower. if an end-to-end acknowledgment were necessary. the packet goes out toward the destination with no link-layer acknowledgments. but the attempted ignorance itself is assuming some generic model and can lead to unfortunate interactions between the layers that reduce goodput. as the packet delivery rate drops toward 0. We want to determine the activity to get a packet to the destination. The no-acknowledgment protocol is suitable for use when the packet delivery rates are high. 5) Reliability As discussed above. For 0. using an end-to-end acknowledgment with a link-layer acknowledgment protocol consumes far fewer Tx-Rx events in almost all cases than the only end-to-end acknowledgment protocol for reasons previously discussed (Figure 16). At up to two hops.6 in the next plot. the implicit acknowledgment protocol fares well briefly at high packet delivery rates. for R=0.9 at five hops (ten equivalent for our comparison). as it would in a wired environment. CONCLUSIONS Careful evaluation of each wireless system is necessary to determine the time bounds required to meet the desired delivery reliability. While these acknowledgments were not specifically modeled for this report. it. but otherwise is more expensive than alternatives. the closest model to the mean Tx-Rx actions of data packet transmissions is the only end-to-end acknowledgment protocol at one-half the hop distance. using cumulative or selective acknowledgments with link-layer acknowledgments would be very efficient. Cumulative or selective acknowledgments are often used as a replacement for per-packet only end-to-end acknowledgments. Refer to Figure 9. Ignoring actual link-layer behavior while considering parameters such as delivery timing is not linklayer agnostic. selective or cumulative acknowledgments may be an advantage.To a degree. In all cases. the number of Tx-Rx actions for our new protocol is off the plot. but. with changes for higher and lower reliabilities. inappropriate timing interactions between the layers then generates degraded goodput results [7][11]. as the packet delivery rates drop exponential growth again occurs and any benefit is lost after a few hops. It also is the most robust with the lowest sensitivity to changes in packet delivery rate. While link-layer protocols typically operate independently from higher-layer protocols. quickly becomes too expensive. while the mean Tx-Rx actions become comparable to the link-level alternatives at high packet delivery rates. Inter-layer 16 VI. that does not mean that the higher-layer protocol should operate unaware of the linklayer behavior. or required reliabilities are low. Studies that find layer interaction problems are typically looking at the results of using the TCP. operating in the expected packet delivery rate range the new protocol is comparable or not efficient compared to a per-packet end-to-end acknowledgment with a link-layer acknowledgment. Its simplicity does not save resources outside of those ranges. the end-to-end acknowledgment will allow overcoming the inherent reliability limitation. the immediate acknowledgment protocol performs best overall. we believe sufficient data is available to glean their first order behavior. The only end-to-end acknowledgment protocol fares well at very high packet delivery rates and few hops. even when the bounded immediate range is reviewed. there are well defined instances when an end-to-end acknowledgment is required to meet application reliability requirements. Otherwise. The inherent reliability of the network could be overcome with fewer resources used compared to per-packet end-to-end acknowledgments.999.

metrics to select the best type of acknowledgment for the conditions. which we find to be immediate acknowledgments. In this report we show how to determine the required reliability. how to determine the required time bound given a sufficient estimate of the packet delivery rate behavior. and reasoning for when end-to-end acknowledgments are required and how to improve reliability up to the inherent limit of the network and above. i. Arbitrarily adding end-to-end acknowledgments to packet deliveries without understanding the underlying link-layer reliability can reduce overall network reliability by increasing resource usage and limiting resource availability with little. because inherent reliability is low. and then only on top of the most efficient link-layer acknowledgment. reliability is obtained by accommodating the time required for sufficient retransmissions to achieve the desired reliability. Only if the inherent reliability of the network is low should end-to-end acknowledgments be considered. 17 . and still should be used only when necessary. gain in reliability.awareness and cooperation to prevent redundant operations has been found to significantly improve goodput [3][11]. if any. These appear efficient on a link-layer acknowledgment protocol. but this must be carefully abstracted to allow the architecture to evolve and prevent obsolesce [13] . Even more efficient are cumulative or selective acknowledgments when resources are available to hold the packets that may need to be retransmitted. In general.e.

6 0.9 PDR Interval 0 0 0.3 0.5 0.5 PDR Interval 0 0 0.9 PDR Interval Figure 6.1 0.6 Probability 1 0.4 0.8 0.2 0.2 0.8 0.3 0.9 50 Link Number 0 0 0.3 0.8 Probability 0.5 0.3 0.5 0.2 0. To achieve even reasonable reliability.8 0.6 0.4 0.4 0.4 0.4 0.2 0.1 0.3 0.9 PDR Interval PDF of PDR of Lowest Mean PDR Link PDF of PDR of Link 1 1 0.6 0.4 0.7 0.  18 .5 0.6 Probability 0 0.2 0.2 0.4 0.4 0.9 PDR Interval PDF of PDR of Link 59 0.5 0.4 0.1 0.8 0.2 0. Time‐varying packet delivery rate probability distributions.4 0.7 0.7 0.7 0.6 0.1 0.9 PDR Interval PDF of PDR of Link 4 1 0.6 1 0.2 0 0 0.7 0.6 PDFs of PDR of All 89 Links 0.5 0. mid‐valued PDRs  on many links and the corresponding larger number of retransmission must be accommodated in the time bound.8 Probability 0.2 0 100 0.PDFs of PDR for All 89 Links 1 1 Probability 0.6 0.1 0.8 0.8 0.6 0.8 0.2 0 0 0 0.8 Probability 0.

999) Implicit 0 1 2 3 Hop Distance from Source PDR = 0. None (R=0.6 1 0 0 1 2 3 Hop Distance from Source PDR = 0.3 Mean Tx-Rx Actions 15 4 4 3 2 1 0 10 1 10 0 0 1 2 3 Hop Distance from Source PDR = 0.5 0 PDR = 1 10 0 End-to-End. Sep.5 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 2 1.3 4 10 Mean Tx-Rx Actions (log) 5 10 5 0 1 2 3 Hop Distance from Source 4 0 0 1 2 3 Hop Distance from Source 4 Figure 7.PDR = 1 2.9 4 10 Mean Tx-Rx Actions (log) 1 3 Mean Tx-Rx Actions 4 2 End-to-End None (R=0.6 4 10 Mean Tx-Rx Actions (log) 2 5 Mean Tx-Rx Actions 0 1 2 3 Hop Distance from Source PDR = 0..9 4 0 1 2 3 Hop Distance from Source PDR = 0.5 1 0. Mean Tx‐Rx actions by basic acknowledgment type at each hop on a linear route of 4 hops.  19 . 0 1 2 3 Hop Distance from Source PDR = 0. Tx‐Rx actions are a  metric for bandwidth and energy usage.999) Separate Implicit Immed. Immed.

 Tx‐Rx actions are a  metric for bandwidth and energy usage. Mean Tx‐Rx actions by basic acknowledgment type at each hop on a linear route of 10 hops. None (R=0.999) Separate Implicit Immed. 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.3 9 10 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.6 1 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.999) Implicit 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0. Sep.9 9 10 3 Mean Tx-Rx Actions (log) 10 1 Mean Tx-Rx Actions 9 10 2 End-to-End None (R=0.5 1 0.9 9 10 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0..PDR = 1 2.5 0 PDR = 1 10 0 End-to-End.6 9 10 10 Mean Tx-Rx Actions (log) 5 5 Mean Tx-Rx Actions 4 3 2 1 0 10 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.3 9 10 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 0 1 2 3 4 5 6 7 8 Hop Distance from Source 9 10 10 10 15 10 10 5 5 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source 9 10 Figure 8.  20 .5 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 2 1. Immed.

. 10 2 PDR = 1 None (R=0.6 7 8 9 10 10 Mean Tx-Rx Actions (log) 6 50 Mean Tx-Rx Actions 1 2 3 4 5 6 Hops PDR = 0.9 7 8 9 10 10 Mean Tx-Rx Actions (log) 4 25 Mean Tx-Rx Actions 7 8 9 10 End-to-End None (R=0. Tx‐Rx actions are a metric for bandwidth and  energy usage.3 7 8 9 10 Mean Tx-Rx Actions (log) 10 Mean Tx-Rx Actions 1 2 3 4 5 6 Hops 7 8 9 10 10 100 10 5 50 0 1 2 3 4 5 6 Hops 7 8 9 10 Figure 9.999) Separate 20 15 10 5 0 10 2 Implicit Immediate 1 2 3 4 5 6 Hops PDR = 0.  21 .3 150 7 8 9 10 40 30 20 10 0 10 4 10 2 1 2 3 4 5 6 Hops PDR = 0. Mean Tx‐Rx actions by basic acknowledgment type for 1 to 10 hops. Sep.9 7 8 9 10 1 2 3 4 5 6 Hops PDR = 0.6 1 2 3 4 5 6 Hops PDR = 0. Immed.PDR = 1 25 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions End-to-End.999) Implicit 20 15 10 5 0 10 1 10 0 1 2 3 4 5 6 Hops PDR = 0.

3.9. R = 0. Reliability‐case values  indicate the activity that must be accommodated to meet the desired reliability.999.3.6.9.999 RC Tx-Rx Actions 5 6 Hops 7 8 9 10 10 End-to-End.999 300 200 100 0 10 5 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 Figure 10. 20 15 10 5 0 10 0 1 2 3 4 1 2 3 4 5 6 Hops 7 8 9 10 10 RC Tx-Rx Actions (log) 5 PDR = 0.999 80 60 40 20 0 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.  22 . R = 0.6.999 400 RC Tx-Rx Actions (log) 10 RC Tx-Rx Actions 10 PDR = 0.. R = 0.PDR = 1.999 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.999 100 RC Tx-Rx Actions (log) 10 RC Tx-Rx Actions 5 PDR = 0. R = 0. Reliability‐case Tx‐Rx actions by basic acknowledgment type for 1 to 10 hops and R=0. Immed. R = 0. R = 0. R = 0. R = 0.999 40 End-to-End None Separate Implicit Immediate RC Tx-Rx Actions 30 20 10 0 PDR = 0.999 25 RC Tx-Rx Actions (log) None Implicit 10 1 2 PDR = 1. Sep.

.3.PDR = 1. Separate Implicit Immed.999 400 Mean Tx-Rx Events (log) 10 10 PDR = 0.6.6.999 80 60 40 20 0 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.999) Separate Implicit Immediate Mean Tx-Rx Events End-to-End 30 20 10 0 PDR = 0.999) 20 15 10 5 0 10 1 10 0 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 10 Mean Tx-Rx Events (log) 5 PDR = 0. R = 0. Mean Tx‐Rx events are a metric for resource  usage duration and latency. None (R=0.999 40 None (R=0.9. R = 0. R = 0.999 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0. R = 0. R = 0. Mean Tx‐Rx events by basic acknowledgment type for 1 to 10 hops.3.  23 .999 100 Mean Tx-Rx Events (log) Mean Tx-Rx Events 10 5 PDR = 0.999 End-to-End.9. R = 0.999 Mean Tx-Rx Events 300 200 100 0 10 5 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 Figure 11.999 25 Mean Tx-Rx Events (log) 10 Mean Tx-Rx Events 2 PDR = 1. R = 0. R = 0.

R = 0.3.999 RC Tx-Rx Events 8 9 10 10 End-to-End. reliability time‐bound.999 100 RC Tx-Rx Events (log) 10 RC Tx-Rx Events 5 PDR = 0.  24 .9. R = 0. R = 0.PDR = 1. R = 0. R = 0. Separate 20 15 10 5 0 10 0 1 2 3 4 5 6 Hops 7 1 2 3 4 5 6 Hops 7 8 9 10 10 RC Tx-Rx Events (log) 5 PDR = 0.9.999 80 60 40 20 0 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.3. RC Tx‐Rx events are a  metric for resource usage duration.999) Separate Implicit Immediate RC Tx-Rx Events 30 20 10 0 PDR = 0. R = 0. None (R=0.999 400 RC Tx-Rx Events (log) 10 RC Tx-Rx Events 10 PDR = 0. R = 0.6.999 40 End-to-End None (R=0..999 25 RC Tx-Rx Events (log) Implicit Immed. and latency.999.6. Reliability‐case Tx‐Rx events by basic acknowledgment type for 1 to 10 hops and R=0.999) 10 1 2 PDR = 1. R = 0.999 300 200 100 0 10 5 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 Figure 12.999 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.

9 6 4 0 1 2 3 Hop Distance from Source PDR = 0.3 40 Mean Tx-Rx Actions 30 20 10 0 4 8 6 4 2 0 10 1 10 0 0 1 2 3 Hop Distance from Source PDR = 0.7 4 10 0.+EE 0 1 2 3 Hop Distance from Source PDR = 0.6 4 10 Mean Tx-Rx Actions (log) 2 10 Mean Tx-Rx Actions 0 1 2 3 Hop Distance from Source PDR = 0.6 4 0 0 1 2 3 Hop Distance from Source PDR = 0. Immed.+EE 0 1 2 3 Hop Distance from Source PDR = 0.5 End-to-End Sep.    25 .PDR = 1 5 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 4 3 2 1 0 PDR = 1 End-to-End Implicit+EE 10 0 Sep.+EE. Mean Tx‐Rx actions by end‐to‐end acknowledgment type at each hop on a linear route of 4 hops.9 Mean Tx-Rx Actions 4 Mean Tx-Rx Actions (log) 10 0.3 4 10 Mean Tx-Rx Actions (log) 5 0 1 2 3 Hop Distance from Source 4 0 1 2 3 Hop Distance from Source 4 Figure 13.+EE 0. Tx‐Rx actions are  a metric for bandwidth and energy usage.3 2 10 Implicit+EE Immed.

9 6 Mean Tx-Rx Actions 9 10 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0. Tx‐Rx actions are  a metric for bandwidth and energy usage.+EE.+EE 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.6 9 10 2 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.6 9 10 10 Mean Tx-Rx Actions (log) 5 10 Mean Tx-Rx Actions 8 6 4 2 0 10 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.9 9 10 10 Mean Tx-Rx Actions (log) 1 4 End-to-End Sep.3 9 10 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0.3 9 10 Mean Tx-Rx Actions (log) 10 Mean Tx-Rx Actions 0 1 2 3 4 5 6 7 8 Hop Distance from Source 9 10 10 30 20 10 5 10 0 0 1 2 3 4 5 6 7 8 Hop Distance from Source 9 10 Figure 14.  26 .+EE Implicit+EE Immed.PDR = 1 5 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 4 3 2 1 0 PDR = 1 End-to-End Implicit+EE 10 0 Sep. Immed.+EE 0 1 2 3 4 5 6 7 8 Hop Distance from Source PDR = 0. Mean Tx‐Rx actions by end‐to‐end acknowledgment type at each hop on a linear route of 10 hops.

+EE 40 30 20 10 0 10 1 1 2 3 4 5 6 Hops PDR = 0. Mean Tx‐Rx actions by end‐to‐end acknowledgment type for 1 to 10 hops.6 1 2 3 4 5 6 Hops PDR = 0. Tx‐Rx actions are a metric for bandwidth  and energy usage.9 7 8 9 10 50 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions 7 8 9 10 End-to-End 10 2 Sep. Sep.+EE.3 300 7 8 9 10 80 60 40 20 0 10 4 10 2 1 2 3 4 5 6 Hops PDR = 0.+EE 40 30 20 10 0 10 1 1 2 3 4 5 6 Hops PDR = 0.    27 .9 7 8 9 10 1 2 3 4 5 6 Hops PDR = 0.3 7 8 9 10 Mean Tx-Rx Actions (log) 10 Mean Tx-Rx Actions 1 2 3 4 5 6 Hops 7 8 9 10 10 200 10 5 100 0 1 2 3 4 5 6 Hops 7 8 9 10 Figure 15.6 7 8 9 10 10 Mean Tx-Rx Actions (log) 6 100 Mean Tx-Rx Actions 1 2 3 4 5 6 Hops PDR = 0.PDR = 1 50 Mean Tx-Rx Actions (log) Mean Tx-Rx Actions End-to-End 10 2 PDR = 1 Implicit+EE Immed.+EE Implicit+EE Immed.

999 600 RC Tx-Rx Actions 10 RC Tx-Rx Actions 10 PDR = 0.+EE RC Tx-Rx Actions PDR = 0. R = 0. R = 0.9.999 400 10 5 200 1 2 3 4 5 6 Hops 7 8 9 10 0 1 2 3 4 5 6 Hops 7 8 9 10 Figure 16. R = 0.999 End-to-End Sep. R = 0.999 60 40 20 0 10 2 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0. Reliability‐case  values indicate the activity that must be accommodated to meet the desired reliability.PDR = 1.3.999.999 End-to-End 40 30 20 10 0 10 1 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 10 RC Tx-Rx Actions 4 PDR = 0.+EE RC Tx-Rx Actions 2 PDR = 1.3.+EE.999 50 RC Tx-Rx Actions 10 Implicit+EE Immed.6.999 RC Tx-Rx Actions 10 RC Tx-Rx Actions 1 2 3 4 5 6 Hops 7 8 9 10 5 100 50 0 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0. R = 0.+EE Implicit+EE Immed.6. Sep. Reliability‐case Tx‐Rx actions by end‐to‐end acknowledgment type for 1 to 10 hops and R=0. R = 0.9.    28 . R = 0.999 150 PDR = 0. R = 0.

999 25 Mean Tx-Rx Events (log) 10 2 PDR = 1. R = 0.999 10 5 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.9.999 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.999 Mean Tx-Rx Events 400 10 5 200 1 2 3 4 5 6 Hops 7 8 9 10 0 1 2 3 4 5 6 Hops 7 8 9 10 Figure 17.3. Mean Tx‐Rx events by end‐to‐end acknowledgment type for 1 to 10 hops. R = 0. R = 0.9.3.999 Mean Tx-Rx Events All other EE&+EE Separate+EE Implicit+EE 20 15 10 5 0 10 1 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 10 Mean Tx-Rx Events (log) 5 PDR = 0. R = 0. R = 0.PDR = 1.999 600 Mean Tx-Rx Events (log) 10 10 PDR = 0.999 100 Mean Tx-Rx Events End-to-End Separate+EE Implicit+EE Immed.    29 . R = 0.6. R = 0. R = 0. Mean Tx‐Rx events are a metric for  resource usage duration and latency.999 200 Mean Tx-Rx Events (log) Mean Tx-Rx Events 150 100 50 0 PDR = 0.6.+EE EE Unconditional 80 60 40 20 0 PDR = 0.

PDR = 1. Reliability‐case Tx‐Rx events by end‐to‐end acknowledgment type for 1 to 10 hops and R=0.999 10 5 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0.3.999 600 RC Tx-Rx Events (log) 10 10 PDR = 0. R = 0.+EE EE Unconditional RC Tx-Rx Events 80 60 40 20 0 PDR = 0. R = 0.6.9.9. reliability time‐bound.999 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 PDR = 0. and latency.999 All other EE&+EE 20 15 10 5 0 10 1 1 2 3 4 5 6 Hops 7 8 9 10 1 2 3 4 5 6 Hops 7 8 9 10 10 RC Tx-Rx Events (log) 5 PDR = 0. R = 0.999 25 RC Tx-Rx Events (log) Implicit+EE RC Tx-Rx Events 10 2 PDR = 1. R = 0. R = 0.6.999.    30 .999 200 RC Tx-Rx Events (log) RC Tx-Rx Events 150 100 50 0 PDR = 0.3. RC Tx‐Rx events  are a metric for resource usage duration. R = 0.999 RC Tx-Rx Events 1 2 3 4 5 6 Hops 7 8 9 10 400 10 5 200 0 1 2 3 4 5 6 Hops 7 8 9 10 Figure 18. R = 0.999 100 End-to-End Separate+EE Implicit+EE Immed. R = 0.

0  0.0  0.Basic Acknowledgments Summary   Changes at R=0.0  0.9  No ACK  Lower      PDR  Which  ~1. Summary of basic acknowledgments and changes that occur at higher and lower reliabilities.9  No ACK  Lower      PDR  ~1.999   Lowest Mean Tx‐Rx Actions  Which  Hops  Then  No ACK  Implicit/all  Immediate  All  All  All        Changes at R=0.9    PDR  Which  ~1.9  Lower  Hops  same  Hops  same  1 to ~4  same  Then  Implicit  Compared to R=0.9  Lower      PDR  Which  ~1.99999  Which  Hops  same  Then  Hops  same  1 to ~6  same  Then  Implicit/  Immediate    Lowest Reliability‐Case Tx‐Rx Actions  Which  Hops  Then  Which  No ACK  No ACK  Immediate  All  1 to ~2  All    Immediate    Immediate  Hops  same  All  same  Then    4 Hops  same  No ACK/  Implicit/  Immediate  same  10 Hops    Lowest Mean Route Congestion  4 Hops  10 Hops  No ACK  No ACK  Implicit/  immediate  immediate  Implicit/  Immediate  Immediate  4 Hops  10 Hops  same  same  Hops  same  Then    Lowest Mean Tx‐Rx Events  Which  Hops  Then  Immediate/ No ACK  Immediate  Immediate  All  All  All          Which  Hops  same  Then  Then  Lowest Reliability‐Case Tx‐Rx Events  Which  Hops  Then  Which  Immediate  Immediate  Immediate  All  All  All        Hops  same  Then  Notes: Protocol1/Protocol2… means that the two are very close    Table 5.0  0.0  0.9  Lower      PDR  Which  ~1.  31 .

999   Lowest Mean Tx‐Rx Actions  Which  Hops  Then  End‐to‐End  End‐to‐End  Immediate  All  1‐~5  All    Implicit/All    Changes at R=0.  Implicit  Immediate  /Unc EE  Immediate  All    Which  Hops  Then  same  1 to ~6  All  Immediate    same  Hops  same  1 to ~4    Then  Immediate      Lowest Reliability‐Case Tx‐Rx Events  Which  Hops  Then  Which  Any except  Implicit  Immediate  Immediate  All  All  All        Hops  same  Then  Notes:  1.0  0.0  0.99999  Which  Hops  same  Then  Which  End‐to‐ End  Hops  same  1 to ~3  same  Then  Implicit    Lowest Reliability‐Case Tx‐Rx Actions  Which  Hops  Then  Which  End‐to‐End  Immediate/  Implicit  Immediate  All  All  All        Immediate  Hops  same  All  same  Then    4 Hops  same  10 Hops    Lowest Mean Route Congestion  4 Hops  10 Hops  End‐to‐End  End‐to‐End/  Implicit  Immediate  End‐to‐End  Implicit/  immediate  immediate  4 Hops  same  10 Hops  Which  Hops  Then    Lowest Mean Tx‐Rx Events  Which  Hops  Then  Any except  Separate. Protocol1/Protocol2… means that the two are very close  2. “UC‐EE” means Unconditional End‐to‐end  3.9  Lower      PDR  ~1.9  Lower      PDR  ~1.    32 . Immediate.9  Which  Hops  same  Then  Compared to R=0.0  0.  Table 6.End‐to‐End Acknowledgments Summary     PDR  ~1. Summary of end‐to‐end acknowledgments and changes that occur at higher and lower reliabilities. Implicit all imply +EE for this table only.0  0.9  Lower      PDR  ~1.9  Lower  Which  UC‐EE  Imediate  Changes at R=0. Separate.0  0.9  Lower      PDR  ~1.

Culler. “End-to-End Arguments in System Design. Busek. February 1995. “The Impact of Multihop Wireless Channels on TCP Performance. Zhang. G. T. “AIRMAIL: A link-layer protocol for wireless networks. Clark. 4. D. 2. 30-Sep. Srinivasan. August 2007. pp. J. Los Angeles. D. November 2003. C. ICCCN 2007. J. Zhang. Balakrishnan.D. Lo Re. Honolulu. CA. R. CA. J. Katz. Levis. vol. “Achieving Robustness through Caching and Retransmission in IEEE 802.” in Proc.” in Proc. M. 1. pp. vol. R.” in Proc. Parsa. Houston. Chuah. California. P. G. and R. Woo. CA November 2003. L. and O. 4. H. C. Bicket. Zerfos. Mar/Apr.” in Proc. no. F. Zhao. September 1999. San Diego. Culler. M. Gerla. Tavakoli. Stanford. 2005 D. and G. Culler “Taming the Underlying Challenges of Reliable Multihop Routing in Sensor Networks. Aug. K. December 1993. S. Oregon.” unpublished. J. ACM SIGCOMM ’96. Krishnamachari. P. D. SenSys ’05. 2. Biswas.” in Proc. November 2005. T. 209-221. “Determining Hard Time Bounds for Reliability Delivery in Wireless Sensor Networks. and R. D. J. Sept. J. 209-221. “Some Implications of Low Power Wireless to IP Networking. Messina. J. C. D. 2005 A. Fu. “Complex behavior at scale: An experimental study of low powered wireless sensor networks. J. D. “Throughput Performance of Transport-Layer Protocols over Wireless LANs. K.H. P. L. Garcia-Luna-Aceves.” IEEE Transactions on Mobile Computing. September 2003. Ortolani. Fu. 2003. SenSys ’03.REFERENCES [1] D. Lindsay. TX. S. Padmanabhan.” in Proc. K. CA. pp. Tong.” in Proc. A. K. University of California. S. “The Impact of Multihop Wireless Channels on TCP Performance. Estrin. Ayanoglu. Aguayo. Shenker. J.” in Proc.15. M. of California. I. H. Wicker. Hui. S.” ACM ACM/Balzer Wireless Networks J. Z.” IEEE Transactions on Mobile Computing. Gitlin. 277-288. Stoica. San Diego. P. SIGCOMM ’04. August 1996. Irvine. private communication. Berkeley. Lu. “A Comparison of Mechanisms for Improving TCP Performance over Wireless Links. Shaw. Portland. “Channel-Specific Wireless Sensor Network Path Data.P. SenSys ’03. WCNC 1999. Zerfos. S. H. Simon. A. MobiCom ’03. D. W. “Link-Level Measurements from an 802. Z. A. “A HighThroughput Path Metric for Multi-Hop Wireless Routing. Morris.” ACM Transactions in Computer Systems 2. Univ. D. 2002. Center for Embedded Networked Sensing. Ganesan. Sabnani. Polastre. De Couto. Woo. Luo. Los Angeles. “Understanding Packet Delivery Performance In Dense Wireless Sensor Networks. 1984. N. CA. D. Cerpa. November. M. Bicket. Lu. Paul. and D. and S. Reed. L. Morris. N. October 2008. 4. “Scale: A tool for simple connectivity assessment in lossy environments. Globecom ’93. J. Hawaii. Aguayo. HotNets-V.” in Proc. V. R. Dutta.” in Proc. LA. Luo.4-based WSNs. A. “TULIP: A Link-Level Protocol for Improving TCP over Wireless Links. Los Angeles.” Technical Report UCLA/CSDTR 02-0013. D. Gerla. August 2007. Estrin. P. ICCN 2007. New Orleans. 3. Levis. Saltzer. CA.11b Mesh Network. pp.” Technical Report 0021. LaPorta. Zaho. “A Unifying Link Abstraction for Wireless Sensor Networks. Seshan. Yue. DeSimone. 2004. Mar/Apr. H. B.” in Proc. W. vol. November 2006. E. S. Govindan. no. [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] 33 . Pister. Judd. 47-60. Doherty.