Clemson University

TigerPrints
All Theses Theses

12-2008

Transmission Error Analysis and Avoidance for

IEEE 802.15.4 Wireless Sensors on Rotating
Structures
Jobin Jacob
Clemson University, [email protected]

Follow this and additional works at: https://tigerprints.clemson.edu/all_theses

Part of the Electrical and Computer Engineering Commons

Recommended Citation
Jacob, Jobin, "Transmission Error Analysis and Avoidance for IEEE 802.15.4 Wireless Sensors on Rotating Structures" (2008). All
Theses. 520.
https://tigerprints.clemson.edu/all_theses/520

This Thesis is brought to you for free and open access by the Theses at TigerPrints. It has been accepted for inclusion in All Theses by an authorized
administrator of TigerPrints. For more information, please contact [email protected].
TRANSMISSION ERROR ANALYSIS AND AVOIDANCE FOR IEEE 802.15.4
WIRELESS SENSORS ON ROTATING STRUCTURES

A Thesis
Presented to
the Graduate School of
Clemson University

In Partial Fulfillment
of the Requirements for the Degree
Master of Science
Electrical Engineering

by
Jobin Jacob
December 2008

Accepted by:
Dr. Kuang-Ching Wang, Committee Chair
Dr. Harlan Russell
Dr. Yong Huang

i
 ABSTRACT

Wireless sensors are increasingly adopted in manufacturing and vehicular systems

for monitoring critical components under continuous operation. Many such components

move rapidly and frequently in metallic containments with challenging radio propagation

characteristics.

For wireless sensors mounted on rotating structures, previous studies identified an

eminent increase in packet transmission errors at higher rotation speeds. Such errors were

found to occur at specific locations around the rotating spindle’s periphery and such

locations depended sensitively on sensor location and surrounding geometry. This thesis

presents a systematic study of the expected packet error rates due to such errors, and

analytically derives the first transmission error rate for a given system. Simulations done

on C++ are used to characterize the error region properties. A transmission error

avoidance approach based on on-line error pattern inference and packet transmission time

control for IEEE 802.15.4 compatible sensor radios is proposed.

The transmission avoidance scheme has two phases: error identification phase to

determine the error characteristics of the system and the operational phase to avoid errors.

Simulation studies showed a 50% error reduction and up to 75% throughput increase for

a rotation system with four symmetric 4º wide error zones with 100% BER inside the

error region and 0% BER outside the error region. Higher throughput gains for higher

rate and larger size transmissions were also noticed for this system. Simulations also

show that the throughput decreases when the packet size duration is greater than the

separation between the error zones

ii
 DEDICATION

This thesis is dedicated to my family.

iii
 ACKNOWLEDGMENTS

I would like to sincerely thank my advisor, Dr. Kuang-Ching Wang for his

guidance, support and motivation without which this thesis would not have been possible.

Every meeting with him has been motivational and a learning experience. I am indebted

to him for his valuable advice throughout the two years of my research.

I would also like to thank Lei Tang and Dr. Huang for their help and support. I am

also grateful to all my friends and colleagues in the wireless department.

iv
 TABLE OF CONTENTS

Page

TITLE PAGE ....................................................................................................................i

ABSTRACT.....................................................................................................................ii

DEDICATION................................................................................................................iii

ACKNOWLEDGMENTS ..............................................................................................iv

LIST OF TABLES.........................................................................................................vii

LIST OF FIGURES ......................................................................................................viii

CHAPTER

1. INTRODUCTION .........................................................................................1

2. BACKGROUND AND RELATED WORK .................................................5

2.1 Current Use of IEEE 802.15.4 Radios .....................................................5

2.2 Constraints Faced by Radios on Rotating Structures............................... 7

3. SYSTEM DESCRIPTION...........................................................................10

3.1 Rotating Wireless Sensor Test Bed........................................................ 10

3.2 IEEE 802.15.4 MAC Protocol ...............................................................14

4. ROTATION ERROR MODEL....................................................................18

4.1 Experiment Results ................................................................................ 18

4.2 Proposed Error Model............................................................................21

4.3 Error Rate Analysis ...............................................................................25

4.3.1 Qualitative Interpretation of PER Dependency on
Rotation Speed ...........................................................................27
4.3.2 Numerical Analysis of First Transmission Error Rate.................. 33

v
 4.4 Simulation Studies ................................................................................. 37
4.4.1 Effects of Different Error Region Widths..................................... 39
4.4.2 Effects of Different Number of Error Regions ............................. 40
4.4.3 Effects of Imprecise Knowledge of Rotation Speed.....................40

5. TRANSMISSION ERROR AVOIDANCE.................................................45

5.1 Overview of Proposed Method ..............................................................45

5.2 Error Identification Phase ......................................................................54

5.3 Operational Phase .................................................................................. 55

5.3.1 Synchronization ............................................................................ 55
5.3.2 Transmission Error Avoidance .................................................... 57

5.4 Simulation Results ................................................................................. 58

6. CONCLUSION AND FUTURE WORK ....................................................67

6.1 Conclusion .............................................................................................67

6.2 Future Work ...........................................................................................69

REFERENCES ..............................................................................................................70

vi
 LIST OF TABLES

Table Page

4.1 First transmission error rate and mean of retransmission count vs

packet size at different packet generation intervals ...............................21

4.2 Maximum, mean, and standard deviation of detected error window sizes
with varied error rate thresholds. ........................................................... 24

5.1 Simulation parameters ................................................................................. 59

vii
 LIST OF FIGURES

Figure Page

3.1 Rotating sensor testbed photograph .............................................................11

3.2 Projected view from the right side of the rotating sensor test bed
(not to scale) ..........................................................................................12

3.3 Sensor communication architecture............................................................. 13

3.4 Unslotted CSMA/CA algorithm .................................................................. 15

3.5 Beacon mode frame structure ...................................................................... 16

4.1 Normalized error burst distance distribution from testbed experiments

at 2054 rpm with 5000 packets sent at 15 msec intervals...................... 20

4.2 Three error region pattern ............................................................................ 22

4.3 Illustration of the time taken in transmitting a correct packet

(Timings not to scale) ............................................................................ 26

4.4 Illustration of the time taken in transmitting an incorrectly

received packet and the corresponding retransmission.......................... 26

4.5 First safe zone illustration considering traversed range during a

packet time only. While not shown, the same rule applies for the
rest of the cycle ...................................................................................... 28

4.6 Safe zone illustration considering traversed range in CCA and

packet time, with and without random backoff. While not shown,
the same rule applies for the rest of the cycle........................................ 29

4.7 Safe zone illustration considering traversed range during CCA,

packet time, turn-around time, and ACK time, without random backoff.
While not shown, the same rule applies for the rest of the cycle...........31

4.8 PER vs rotation speed .with three evenly spaced 4 degree error regions .... 33

viii
List of Figures (Continued)

Figure Page

4.9 Probability of successful packet transmission illustration considering

traversed range during CCA, packet time, and random backoff time
. While not shown the same rule applies for the rest of the cycle............ 34

4.10 Probability of successful packet transmission to miss the error zone,

considering traversed range during CCA, packet time, turn-around time,
and..... random backoff time. While not shown, the same rule applies for
the rest of the cycle ............................................................................... 35

4.11 Probability of ACK success illustration, considering traversed range during

CCA, packet time, turn-around time, and random backoff time. While not
shown, the same rule applies for the rest of the cycle ........................... 36

4.12 First transmission error probability distribution for a three error zone
system centered at 0,120 and 240 degrees with width 4 degrees each
(E1b=-2º, E1e=2º, E2b=118º,E2e=122º,E3b=238º and E3e=242º)...............36

4.13 Normalized error burst distance distribution obtained with error region
widths 5, 10, and 15 degrees at 2000 rpm ............................................. 39

4.14 Normalized error burst distance distribution for one, two, and
three evenly distributed error regions .................................................... 40

4.15 Error burst distance distribution obtained for R=2000 rpm and
(a) C=1950 rpm, (b) C=1900 rpm, (c) C=2050 rpm, (d) C=2100 rpm.. 42

4.16 Error burst distance distribution for large speed inaccuracies with one error
region .....................................................................................................44

4.17 Error burst distance distribution for large speed inaccuracies with two error
region .....................................................................................................44

5.1 Two error region distributions can result in similar error burst distance
distributions: (a) three error regions centered at 0, 90, and 180 degrees;
(b) four error regions centered at 0, 90, 180, 270 degrees. ................... 46

5.2 Location distribution with clock drift for error bursts shown in
Figure 4.1. ............................................................................................. 48

5.3 Calibrated error burst location distribution..................................................48

ix
List of Figures (Continued)

Figure Page

5.4 Error burst location distribution for a three error zone system
centered at 0,90 and 180 degrees, each with a width of 4 degrees .......50

5.5 Percentage error in error location histogram if the rotation speed

is considered as 2054 rpm when the actual rotation
speed is 2000rpm, vs number of probe packets with error ................... 51

5.6 Percentage error in error location histogram vs number of probe

packets with error...................................................................................52

5.7 Synchronization algorithm ..........................................................................56

5.8 Error avoidance algorithm ..........................................................................58

5.9 2000rpm, one 4-degree error region, (a) PER v.s. packet size v.s.
transmission interval (b) throughput v.s. packet size v.s.
packet interval........................................................................................61

5.10 2000rpm, four 4-degree error regions PER v.s. packet size v.s. transmission
interval ...................................................................................................63

5.11 2000rpm, four 4-degree error regions Throughput v.s. packet size v.s.
transmission interval .............................................................................. 63

5.12 2000rpm, four 4-degree error regions, throughput v.s. PER outside error
region with 92 byte packets (BER in error region is irrelevant since
transmissions are entirely avoided)........................................................ 64

5.13 2000rpm, two 4-degree error regions centered at 0 and 60 degrees, throughput
v.s. packet size for different packet generation interval ........................ 66

x
 CHAPTER ONE

INTRODUCTION

Wireless sensors, engineered as miniaturized, low-cost embedded devices with

integrated sensing, processing, and radio communication capabilities, have recently

gained substantial attention for use in different monitoring applications [1]. Wireless

sensors are promising for replacing cables in monitoring systems and enabling flexible

sensing over structures that are difficult to monitor with wired sensors [2]. Rotating

mechanical structures found in a wide range of mechanical, civil, and aerospace systems

are among such hard-to-reach structures with crucial monitoring importance [3-6]. Such

structures are mostly found in metallic enclosures that are harsh for radio

communications, posing to the industry and research community a number of open

challenges such as the frequent transmission errors, low data throughput, unknown

system reliability, and effects of varied sensor deployment strategies [7-10].

Rotating mechanical structures, typically built within complexly shaped metallic

enclosures, create a challengeable condition for wireless data transmission due to their

complex, reflective, and fast changing radio paths. Doppler effect might be introduced

into the data transmission due to the high rotation speed. During high speed rotation, the

stability of radio hardware, such as the antenna connection, is also crucial to reliable

transmission. Altogether, these challenging conditions impose on the sensor radios a

significant challenge that far exceeds conventional radio communication systems. By far,

studies of rotating wireless sensors have not clearly addressed the issue of data

transmission reliability [3-6]. The recent study [6] reported the number of received

1
packets out of all transmitted packets; yet, it did not measure the number of correctly

received packets out of all received packets.

In [11,12] a sensor radio mounted on a rotating spindle in a metal enclosure was

studied. In [11], it was concluded that errors during rotation were primarily due to

multipath and Doppler effects and caused error bursts spanning multiple consecutive bits.

In [12], such errors were found to occur near one specific location around the spindle

periphery; its vicinity was referred to as an error region, and a reliable transmission

method based on automatic repeat request (ARQ) was introduced to recover transmission

errors. The ARQ approach achieved a high retransmission success rate; nevertheless,

new transmissions still faced above 20% packet error rates, severely undermining the

achievable data throughput and wasted substantial energy in retransmissions. As later

shown in this thesis, packet error rates can easily reach as high as 50% under certain

configuration and channel characteristics. In [13], the temporal distribution and size of

error regions were analyzed based on recorded time stamps and bit contents of failed

probing transmissions. Probing transmissions were also used to determine the effects on

the error distribution with the change in rotation speed of the spindle and clock drift of

the sensors.

With multipath and Doppler effects being the primary causes of transmission

errors for a rotating sensor, the error region distribution is expected to depend sensitively

on the surrounding structure, sensor position, and rotating speed. In [8], the PERs were

found to differ substantially by moving the receiver’s location by just 3cm. To avoid

transmissions that would most probably result in errors, a sensor radio must be aware of

2
its error region distribution in order to determine its transmission strategy, while

obtaining accurate sensor position around the periphery during high speed rotation is very

difficult without complex hardware design. To reduce such transmission errors for

enhancing data throughput and energy efficiency, an error-avoidant transmission scheme

utilizing error region distribution needs to be considered.

The purpose of this thesis is to characterize the error region and its avoidance

scheme. An analytical model is developed based upon the protocol of IEEE802.15.4.

Different experiments and simulation results are used to evaluate the effectiveness of the

proposed transmission error avoidance mechanism. A rotating radio simulator is

developed based on the IEEE 802.15.4 protocol to model the radio error model, error

distribution inference, and transmission time control. Simulation results are divided

according to the two phases of the transmission error avoidance method. In the training

phase the sensor utilizes probing packets to generate the error location distribution. This

is used to identify the error locations and their widths. In the operational phase the sensor

sends data packets. Synchronization is done initially to deduce the location of the

transmitter with respect to its error location. Simulation results show a 50% error

reduction and up to 75% throughput increase for a rotation system with four symmetric 4º

wide error zones with 100% BER inside the error region and 0% BER outside the error

region. Higher throughput gains for higher rate and larger size transmissions are also

noticed for this system. The main contribution of this thesis is the transmission error

analysis method and the transmission error avoidance scheme used for a rotating wireless

sensor system.

3
 The rest of the thesis is organized as follows. In chapter 2 the related work and

achievements are described. Chapter 3 describes the experimental setup of the wireless

sensors in the computer numerical control (CNC) lathe and the IEEE 802.15.4 MAC

protocol. The model developed for the rotating error system is described in chapter 4,

and the transmission error avoidance scheme is described in chapter 5. The conclusions

and future work are discussed in chapter 6.

4
 CHAPTER TWO

BACKGROUND AND RELATED WORK

IEEE 802.15.4 radios are used in networks to ensure low cost communication

with low power consumption. The thesis focuses on the communication between an IEEE

802.15.4 radio attached to a rotating device and a stationary radio. In this chapter the

current use of IEEE 802.15.4 radios and the constraints faced by radios on fast rotating

structures is discussed.

2.1 Current Use of IEEE 802.15.4 Radios

The IEEE 802.15.4 standard compliant radios are designed for low-data-rate, low-

power, and short-range transmission [15] in the 868/915 MHz and 2.45 GHz Industrial,

Scientific, and Medical (ISM) radio bands. Existing IEEE 802.15.4-compliant radios do

not implement complex circuitry such as a RAKE receiver which improves reception in

the presence of multipath channels or a carrier frequency tracker for compensating

Doppler shifts, but rely on the limited multipath and Doppler tolerance inherent in its

direct sequence spread spectrum (DSSS) modulation scheme.

Numerous field measurement studies of wireless sensors have been done, e.g.,

[16-19]. Most of these studies included IEEE 802.15.4 radios in addition to other sensor

platforms with their respective radio protocols, frequency bands and power levels.

Measurements were conducted in both indoor and outdoor settings and had the same

conclusions that there exists a grey zone [16,18] at a specific distance from the

5
transmitter, within which the receivers link quality fluctuates with even small

displacements. The studies also conclude that link asymmetry occurs, such that a node

may be able to transmit to another node but vice-versa may not happen, even if the nodes

were set to the same transmit power. In addition to studies on wireless sensor links, work

has been done on the use of IEEE 802.15.4 radios in monitoring the environment [20]

and location estimation [21-22]. [23] discusses the introduction of an IP stack to the IEEE

802.15.4 stack to enable the efficient transmission of IPv6 datagrams over IEEE 802.15.4

links. This opens up the use of such radios for Internet applications.

The majority of applications examined to-date had sensors deployed on the

ground or on structures that are either stationary or slow moving. Very few studies have

examined the use of wireless sensors on structures that move substantially within the

timeframe of a wireless transmission. Rotating structures, specifically, are commonly

found in mechanical and vehicular systems with substantial monitoring importance.

Only few studies have examined wireless sensors operating on a rotating

structure. In [2], wireless sensors were installed inside polymer covered cylindrical rolls

to measure acoustic emission pulse counts and surface temperatures; up to 10k packets

per second data were transmitted over an IEEE 802.11 radio. In [3], a wireless motor

spindle measurement system measured torque, power, and rotation speed with sampling

rates up to 963.9 Hz, and the data were transmitted over a low-power digital radio at

19200 bits per second (bps). Both studies did not mention transmission errors in their

experiments.

6
 2.2 Constraints Faced by Radios on Rotating Structures

Radio communication challenges under moving conditions have been studied

extensively in the context of cellular mobile communications. The conclusions of such

work cannot be directly applied to the rotation of a wireless sensor. Sensor radios differ

greatly from cellular radios in their very low transmit power, simple antenna and circuitry

design. Commercial wireless sensors trade off robustness for low costs

Rotating mechanical structures are usually located within confined enclosures,

with the structures themselves and their surroundings made of metallic materials. To

monitor such rotating structures, a typical wireless measurement system usually consists

of two parts: 1) the measuring and transmitting unit, which is mounted on or within the

rotating structure, and 2) the receiving unit which is mounted on a stationary part near the

rotating structure. Within such an environment, a received radio signal is expected to

have rich multipath radio components, each experiencing different path loss and Doppler

effect, and resulting multipath effects that depend on the relative phases, amplitudes, and

frequency shifts of all transmitted radio components. While wireless sensors have been

demonstrated with stably acceptable performance in amicable environments such as in an

office or out in the field, the performance of rotating wireless sensors may be severely

affected by harsh monitoring environments due to excessive obstruction, multipath

propagation, antenna polarization, Doppler shift, and electromagnetic noise.

Radio signal attenuates along the propagated path. The energy attenuation,

known as the path loss, is proportional to a certain power of the propagated distance, with

additional obstruction and/or refraction loss caused by objects along the path. Signal is

7
radiated from an antenna in a span of directions regardless of their types. When static or

moving objects exist in the surrounding, the radio signal may follow multiple reflected

paths to reach a receiver with different amplitudes and phases due to the different path

lengths. The received signal is the sum of all such multipath signal components, which

generally causes distortion in the signal duration and waveform. Rotating structures also

face the additional challenge of the Doppler effect, which occurs due to the fast change of

the distance between the transmitting and receiving sensor during signal transmission.

The distance change results in frequency shifts and signal duration changes that also

distort the received signal [24]. Electromagnetic noise is typically present around

engineering systems.

In [11], experiments with rotating wireless sensors concluded: (1) the radio

performance was dominated by multipath effects, whereas machine electromagnetic noise

and hardware stability effects were insignificant; (2) when stationary, the radio had

consistently low PERs; (3) when rotating, PERs increased with speed at some locations

but not all; the dependency changed when receiver was moved by 3cm; (4) bit errors

occurred in bursts, when the transmitter rotated past a particular location; (5) path loss,

antenna directive gains and stability, machine noise, and automatic gain control had

insignificant relevance with bit errors.

The experimental setup utilized in this thesis is designed to model the essential

features of a wireless sensor-based monitoring platform for rotating structures in civil,

mechanical, and aerospace systems. The sensor platform can potentially be integrated

with sensors for acoustic emission, strain gauge, thermocouple, accelerometer, etc.

8
Examples of such applications can be found in roll contact monitoring [3], grinding

wheel truing and grinding monitoring [4], and motor shaft torque monitoring [5, 6]. The

evaluation study conducted serves to highlight key performance characteristics that

cannot be overlooked in designing a reliable wireless monitoring system, whereas the

experimental procedure is applicable for channel characterization in a wide range of

systems.

9
 CHAPTER THREE

SYSTEM DESCRIPTION

In this chapter the rotating wireless sensor test bed is explained. The

experimental results on this test bed form the basis for this thesis. The transmission error

avoidance scheme developed utilizes the timing information of the IEEE 802.15.4 MAC

protocol. This chapter also explains the MAC protocol and how its packet scheduling can

be used in the transmission error avoidance scheme, it also explains the reason for

choosing the non-beacon mode version of the protocol for the transmission error

avoidance scheme.

3.1 Rotating Wireless Sensor Test Bed

The testbed is built with two wireless sensors mounted inside a computer

numerical control (CNC) lathe, with one base station placed outside the machine for

issuing commands and retrieving data from the wireless sensors.

Crossbow MicaZ motes [26] are chosen as the wireless sensors for this testbed.

The MicaZ mote adopts an IEEE 802.15.4-compliant radio based on the Chipcon

CC2420 chip, which supports 250 Kbps raw data rate, controllable transmit power range

of -25 dbm to 0 dbm, and 16 channels in the 2.4 GHz ISM band. The radio chip feeds to a

quarter-wavelength monopole antenna (about 1.2 inches long) screw-attached to the

circuit board with an MMCX connector that is designed for a slot-less, snap-fitting, and

minimal-leakage contact. The mote outputs various radio parameters to be used in the

10
experiments. For each received packet, an 8-bit received signal strength indication

(RSSI, linear transformable into dBm unit), an 8-bit link quality indicator (LQI), and the

output of a 16-bit cyclic redundancy check (CRC) error-detecting circuit [27] are

provided. The radio adopts direct sequence spread spectrum (DSSS) and O-QPSK half-

sine modulation, such that each 4-bit data is mapped to a pseudo-random 32-chip binary

spreading sequence, which is transmitted with O-QPSK modulation. The mote is 2.25 by

1.25 by 0.25 inch³ in size.

Figure 3.1. Rotating sensor testbed photograph

11
 14 cm 38cm

40 cm

Transmitter
300° Receiver
0°

240° 45 cm
40 cm

60°
6 .5
cm 180°
120°
70 cm
25 cm 42 cm 40 cm

30 cm
47 cm 10 cm

80 cm

Figure 3.2. Projected view from the right side of the rotating sensor test bed (not to scale)

The machine used in the testbed is the Hardinge Talent 6/45 CNC lathe with a

speed-controllable rotating spindle enclosed in a metallic chamber. The spindle can rotate

at a maximum speed of 6000 rpm and minimum speed of 60 rpm. Figure 3.1 shows a

photograph and Figure 3.2 shows a cross-section view along the spindle axis. The

transmitter mote is tape-bound to the spindle surface, with its antenna perpendicular to

the surface. The receiver mote is placed on the metallic side board next to the spindle,

aligned with the transmitter on the plane perpendicular to the spindle. Transmission

experiments are controlled via a base station composed of a MicaZ mote connected to a

Crossbow mote interface board (MIB510), located outside the lathe. The precise location

of the base station is not relevant, as it only communicates with the sensors for control

and data logging between experiments. As illustrated in Figure 3.3, the base station

communicates with the sensors over the same IEEE 802.15.4 channel, while it interfaces

with a personal computer (PC) for data logging over a universal asynchronous

12
receiver/transmitter (UART) serial cable.

CNC Machine Radio

UART

Receiver

Link of
Base Station
interest
MIB 510 PC
Transmitter

Figure 3.3. Sensor communication architecture.

Wireless transmission experiments are conducted to transmit a sequence of

packets using the ARQ reliable transmission method and to record all transmission errors

seen at the receiver. With the IEEE 802.15.4 acknowledgement option enabled, the

receiver sends a short acknowledgement (ACK) packet to the transmitter immediately

after receiving a packet correctly; otherwise, an ACK is not sent and the transmitter will

retransmit the packet after one IEEE 802.15.4 ACK timeout duration. The correctness of

a packet is determined at the receiver by checking a hardware-generated cyclic

redundancy check (CRC) bit. If the CRC bit is 1, the entire packet is considered to have

been received correctly. If CRC bit is 0, some bits in the packet must have been

corrupted. The bit sequences of all incorrectly received packets are recorded with their

received timestamps for further processing. After each successful transmission

(transmitter receiving an ACK), the transmitter waits for a standard-defined random

backoff duration and proceeds to send the next packet. Effectively, the ARQ approach

13
sent a continuous flow of packets from the transmitter to the receiver, with short intervals

between every two consecutive transmissions. Experimental results are discussed in

chapter 4.

3.2 IEEE 802.15.4 MAC Protocol

The proposed transmission error avoidance scheme is based upon the scheduling

of the packet transmissions by the 802.15.4 MAC protocol; hence the following section

describes the 802.15.4 MAC protocol.

The IEEE 802.15.4 MAC Protocol is essentially a CSMA/CA protocol and can be

operated in two different modes [15], namely the beacon mode (slotted CSMA/CA

algorithm) and the non-beacon mode (unslotted CSMA/CA algorithm) . Figure 3.4 shows

the non-beacon mode of operation. The transmission time of the packet depends on which

mode is chosen.

With the beacon mode, a coordinator radio can assign transmission slots to each

device radio. The coordinator broadcasts periodic beacons at its chosen interval, between

two beacons the coordinator can allocate contention access period (CAP) slots,

guaranteed time slots (GTSs), and inactive (low-power sleep) slots. A device radio can

transmit a data packet during CAP based on slotted CSMA/CA (carrier sense multiple

access with collision avoidance), i.e. waiting a randomly chosen number of idle slots

prior to transmission, or it can transmit data packets in its GTS assigned by the

coordinator.

14
 CSMA-CA

Unslotted

NB = 0
BE = macMinBE

Delay for
random(2BE -1) unit
backoff periods

Perform CCA

Channel Y
Idle?

NB=NB+1,
BE = min(BE+1,macMaxBE)

N NB>
macMaxCSMABackoffs?

Failure Success

Figure 3.4. Unslotted CSMA/CA algorithm [15]

15
 Beacon
Beacon

CAP CFP

GTS GTS
Inactive

Figure 3.5. Beacon mode frame structure [15]

Conceptually, the proposed approach can be easily implemented with the beacon

mode, by configuring the receiving radio as a coordinator, the transmitting radio as a

device, the beacon period equal to the rotation period, and assign the device with GTS

slots aligned with the low-error regions. In reality, however, the standard imposes

various constraints that make it very restrictive, if not impossible, to implement the

concept properly with the beacon mode.

These constraints are: (1) a minimum of aMinCapLength symbol duration must

be allocated for CAP after each beacon for control frame exchange and new device joins,

(2) GTS can only be allocated after CAP, (3) it is not always possible to assign a beacon

period equal to the rotation period, (4) beacon frames must be correctly received every

period for GTS transmissions to proceed. These constraints result in substantial

overheads and scheduling inefficiencies in practice.

16
 With non-beacon mode, there is no distinction between a coordinator and a device

radio for transmission purposes. Each radio attempts to transmit a packet using an

unslotted CSMA/CA algorithm. The radio waits for one fixed clear channel assessment

(CCA) duration and a random number of backoff slots. If the channel remains idle

during the wait, the transmission proceeds, and the receiver responds with an ACK if the

packet is received correctly. The expected time for completing a data transmission is

thus the time from start of CCA till end of ACK reception. The main inefficiency with

this approach lies in the random backoff periods. Fortunately, the standard allows setting

minimum values for the random backoff parameters to limit the minimum backoff

window to 20 backoff periods and the maximum backoff window to 23 backoff periods.

The non-beacon mode approach is considered for the transmission error avoidance with a

maximum back-off duration of 1 period, because of all these advantages.

17
 CHAPTER FOUR

ROTATION ERROR MODEL

This chapter focuses on the study of the error patterns of the rotation system

which is used later on in developing the error avoidance scheme. The first section of this

chapter explains the experimental results which led to the identification of the error

region. The second section provides an analytical model for analyzing the packet error

rate in such a system and the third section describes the simulation model for the system.

4.1 Experiment Results

ARQ experiments were conducted at 2054 rounds per minute (rpm) rotation with

the configuration shown in Figure 3 1. The radio transmitted at -25 dBm power in

channel 26 (2478.5 ~ 2481.5 MHz). In each experiment, 5000 packets of packet size 92

bytes (10 byte header size included) were transmitted continuously; the sender was

always backlogged and sent a new packet whenever the previous packet had been

transmitted/retransmitted successfully or the maximum retry limit (9) was reached.

In each experiment the sender sent probing transmissions of a known pattern (all

1’s) and size (100 bytes) at a chosen interval to the receiver. For each packet received

with errors (failing CRC check), the receiver recorded the packet’s received timestamp

and bit contents. Using a sliding window algorithm, the receiver identified all error

bursts in these packets; an error burst is defined as a set of consecutive bits with at least a

specified fraction H of bits in error. The time interval between the start times of two

18
consecutive error bursts, Ti, was measured to derive the normalized error burst distance

which is defined below.

Definition 1: Error Burst Distance Distribution

Consider N consecutive error bursts obtained from the probe packets. Let the time

interval between the start times of the ith and i+1th error bursts be Ti, the normalized error

burst distance Di for the ith error burst is defined as

Ti modulo TR (1)
Di =
TR
were TR is the spindle rotation period. The normalized error burst distance represents the

distance (in fraction of periphery) between the two positions on the periphery where

consecutive error bursts have occurred.

Figure. 4.1 shows the error burst distance distribution estimated on the rotating

wireless sensor testbed. With the two main modes residing near 0 and 1, respectively, it

meant that the errors had always occurred in the same vicinity, which was defined as an

error region. The error burst distance distribution is considered mainly because of the

clock drift error seen in the sensors. Another parameter called error burst location

distribution which is explained in Section 5.1 is used in determining the error patterns if

the sensor clock drift error is overcome. Further discussion on sensor clock drift is given

in Section 5.1.

19
 Figure 4.1. Normalized error burst distance distribution from testbed experiments at
2054 rpm with 5000 packets sent at 15 msec fixed intervals

ARQ experiments were also conducted with packets of chosen packet size (30, 60

and 92 bytes) transmitted at chosen constant interval (0, 100, 150, and 200 msec) for a

constant rotation speed of 2000rpm. The results of the experiment are shown in Table

4.1. First transmission rate gives the percentage of packets received with error during

their first transmission, while retransmission count gives the number of retransmissions

required for the successful delivery of the packet. The table shows that the first

transmission error rate increases with packet size and the average retransmission count is

nearly one. This shows that almost all retransmissions are successful and indicates that

the retransmissions always tend to happen in a “low- error” region.

20
 Table 4.1. First transmission error rate and mean of retransmission count vs packet size
at different packet generation interval.

Packet Generation Interval = 100

Packet Generation Interval = 0 msec
msec
1st Retrans. 1st Retrans.
Packet trans. count mean Packet trans. count mean
size error /standard size error /standard
rate deviation rate deviation
30 bytes 0.098 1.026/0.159 30 bytes 0.067 1.000/0.000
40 bytes 0.137 1.253/0.435 60 bytes 0.136 1.000/0.000
92 bytes 0.151 1.191/0.424 92 bytes 0.171 1.140/0.365
Packet Generation Interval = 150 Packet Generation Interval = 200
msec msec
30 bytes 0.059 1.000/0.000 30 bytes 0.067 1.000/0.000
60 bytes 0.143 1.000/0.000 60 bytes 0.087 1.046/0.209
92 bytes 0.181 1.042/0.202 92 bytes 0.220 1.112/0.316

4.2 Proposed Error Model

The error burst distribution indicated that at a certain rotation speed one or a few

“high-error regions” can be identified, and among them are interleaving “low-error”

regions. Figure 4.1 indicates that errors tend to occur at the same position which could

mean that the system could have just one error region. If any segment of a packet is

transmitted during the high-error region, those bits have a greater possibility to have

errors; on the other hand, any segment transmitted in a low-error region is expected to see

less, if any, error bits in it. A different system with more multipath and Doppler effects

could result in a different number of error regions; but all systems can be represented by

the same error model.

21
 To determine the error model, probe packets are sent. The bits of the probing

packets are analyzed to determine all the possible error bursts and the error burst distance

distribution is generated using the start times of the error burst.

Let M be the number of distinguishable error regions identified from the error

burst distance distribution. Each error region will have its own width Wi which is

determined by the bit error rate Hi inside the error region. The bit error rate inside the

error region would be higher than the bit error rates in the remaining “low-error regions”.

The position of each error region can be identified by its center Ci which is represented in

degrees. Thus each region of the system can be represented by the parameters { Wi, Hi, Ci

} where i =1,2,…, M.

An example of a three error zone system is shown in Figure 4.2

Figure 4.2. Three error region pattern

22
 To extract the model parameters from actual experimental data a sliding window

algorithm is adopted.

The sliding window algorithm is defined as follows. Given the bit sequence of an

incorrectly received packet of size N bits, {Si, i=1…N}, a subset of consecutive bits in

{Si} is defined as a window W(j, l) where j indicates the sequence number of the first bit

in W(j, l) and l is the number of bits in W(j, l). The algorithm initiates with j=1 (the first

bit), l=2 (two bits in the window), and a specified error rate threshold H. The error rate

for W(j, l), PW, is defined as the number of error bits in W(j, l) divided by the total

number of bits in W(j, l). In each step, if PW is less than H, j increases by one (the

window shifts by one bit position); otherwise, l increases by one. One round is

considered completed when the window encloses SN, and another round is started with

j=1 and the l value as at the end of the previous round. The algorithm repeats until l does

not increase in one round; thus, l is the size of the largest window that exceeds the

specified error rate threshold, and it is used to estimate the size of the error region.

The sliding window algorithm was applied to all error packets in the ARQ

transmission experiments. With different threshold H, the error burst window size for all

error packets are found. The error packets have a size of 92 bytes. The maximum, mean,

and standard deviation of the window sizes are summarized in Table 4.2. Each (H, l) pair

provides a valid model to reproduce the observed error statistics. With l given in bits, the

corresponding error region width can be calculated in terms of the duration or the rotated

angle given the rotation speed. Note that IEEE 802.15.4 radios have a data rate of 250

kbps; thus l bits represent a duration of l/(250k) sec. . The statistics show that the error

23
region size and the transmission error rate are inter-dependent. The table shows that the

mean error width decreases with increase in the error burst threshold. The table also

shows that the mean error width size is between 4 and 23 bits which is equivalent to 0.2

to 1.1 degrees at a rotation speed of 2054 rpm. Thus all simulations consider an error

region width of 4 degrees which is a close approximation to the width calculated using

the sliding window algorithm. Such a small error region can cause an average PER of

11.8% without the ACK option enabled. As will be shown with simulation in Section 5.1,

the PER can increase substantially with even a few more such narrow error regions. With

four error regions (4 degree wide, evenly spaced), the PER can be as high as 50%.

Table 4.2. Maximum, mean, and standard deviation of detected error window sizes with
varied error rate thresholds.

Max l Mean l Std.Dev

H
(bits) (bits) (bits)
0.4 215 23.0 25.0

0.5 149 16.5 18.0

0.6 81 11.0 12.0

0.7 67 7.3 8.0

0.8 52 5.2 5.0

0.9 28 4.0 2.8

It is the highly predictable error region distribution and the high PER that can be

caused by very small error regions that have motivated this thesis to study means to avoid

such errors over existing sensor radios by controlling packet transmission times based on

24
identified error region distribution. Equation 1 on page 20 shows that the error burst

distance distribution is readily obtainable with accurate knowledge of the rotation speed

and times of previously seen error bursts. Obtaining such information and ensuring its

accuracy are by themselves challenging, since (1) accurate rotation speed can be obtained

only with complex sensor design or has to be reported by the rotating machine, (2) the

estimated or reported rotation speed can still be inaccurate, and (3) the sensor clock can

be inaccurate with respect to the clock used for speed measurement. In section 4.3.3, the

impact of inaccurate knowledge of rotation speed is studied.

4.3 Error Rate Analysis

As explained in section 3.2, for error avoidance the IEEE 802.15.4 MAC protocol

version considered for this thesis is the CSMA/CA non-beacon mode version. Our system

comprises just one sensor communicating to the other sensor at a time and hence the

channel is always idle for transmitting data and there is no chance of collisions.

Understanding the MAC protocol packet scheduling time in this system is easy and can

be utilized to determine the error rate in the system. So this section explains the

parameters of the CSMA/CA algorithm involved in scheduling the packet transmission

and comes up with intuitive and numerical analysis to determine the error rate.

As shown in Fig. 4.3, every transmission is preceded by a random back-off time δ

and a clear channel access (CCA) duration. The maximum back-off time is ∆ which

corresponds to 1 random back-off period. There is an additional turnaround time D

involved for switching the radio from receiver mode to transmit mode or vice versa. If the

25
 link layer ACK option is enabled an acknowledgement packet is sent for every correctly

received packet. Our experimental results in Chapter 3 show that the retransmission

success rate is high and indicates the effectiveness of using an ARQ algorithm for

retransmission of incorrectly received packets and hence the ACK option is enabled.

Thus every incorrect packet is retransmitted after a constant acknowledgement time out

duration τ. A picture showing the time spent in transmission of a correct packet and a

packet with erorrs is shown in Fig. 4.3 and 4.4.

D CCA δ Packet time D ACK

Figure 4.3. Illustration of the time taken in transmitting a correct packet (Timings not to
scale)

D CCA δ Packet time τ D CCA δ Packet time

Figure 4.4. Illustration of the time taken in transmitting an incorrectly received packet
and the corresponding retransmission (Timings not to scale)

4.3.1 Qualitative Interpretation of PER Dependency on Rotation Speed

Packet error rate (PER) is defined as the fraction of the transmissions that are

successful with the ACK option enabled. For a three error region distribution with error

zones centered at 0, 120 and 240 degrees and error regions width of 4 degrees each, the

expected PER can be found analytically with respect to the temporal pattern and size of

the transmitted packets. The analysis provided in this section gives an intuitive

26
explanation for the dependence of PER on rotation speed based upon three different

assumptions. The explanation given is only corresponding to the given symmetric error

system.

Assumption 1: Considering only the packet transmission time in the IEEE 802.15.4

MAC Protocol.

Consider a rotation cycle with a given error region distribution as illustrated in

Figure 4.5. During the time a B-byte packet is transmitted at rate R bits per second, a

sensor traverses for 8Bw/R degrees around the circumference of a spindle rotating at

speed w; if the traversed region overlaps the error region in any part, the packet may

contain a number of error bits. To simplify the analysis, a 100% BER within an error

region, 0% BER outside an error region, and no error corrections are considered.
tB
ac

Of
k

Timings not to scale

Figure 4.5. First safe zone illustration considering traversed range during a packet time
only. While not shown, the same rule applies for the rest of the cycle

The transmission correctness depends on the starting location of a packet. Let any

portion around the circumference that a packet can start and end its transmission without

27
encountering an error region be called a safe zone. The wider the safe zone, the more

probable a packet initiated at random time can complete successfully. If the pre-

transmission wait times required by the IEEE 802.15.4 protocol is ignored for a given

error distribution, the safe zone depends on the traversed degrees during one packet time,

which is proportional to the rotation speed. The faster the rotation speed, the more

degrees a sensor traverses during its transmission, and hence the smaller safe zone it has.

The reasoning suggests an increasing chance of packet errors with increasing rotation

speeds if packets are transmitted at arbitrary times without an error avoidance

mechanism; the phenomena was observed in actual experiments .

With random backoff

Without random backoff

ac

Of
k

Timings not to scale

Figure 4.6. Safe zone illustration considering traversed range in CCA and packet time,
with and without random backoff. While not shown, the same rule applies for the rest of
the cycle.

28
Assumption 2: Considering the CCA and random back-off time of the IEEE 802.15.4

MAC Protocol.

In reality, an IEEE 802.15.4 sensor node in the non-beacon mode must wait for a

constant clear-channel-assessment (CCA) duration and a random back-off duration prior

to transmitting a packet. This variable pre-transmission wait changes the start time

analysis slightly as illustrated in Figure.4.6. Here the distance travelled in the CCA

duration is assumed to be less than the error region width. Hence the constant CCA

duration simply shifts the safe zones ahead but does not change the overall safe zone size

and, therefore, does not change the PER. The random back-off duration, however, can

reduce the safe zone size. As the figure shows, the total safe zone decreases when the

wait time increases from CCA to CCA+Δ, where Δ stands for the maximum backoff

duration. This is because the safe zone accounts for the areas where a data transmission

can begin and be sure to complete entirely outside any error region and a transmission

with maximum random back-off duration would need to start at a position earlier than

without random back-off. This would result in pushing the boundary of the safe zone

further back. Outside the safe zone, there are some areas where a transmission can begin

and has some less than 100% probability to complete outside an error region depending

on its random backoff duration.

Assumption 3: Considering the ACK time of the IEEE 802.15.4 MAC Protocol.

The link layer ACK option of IEEE 802.15.4, if enabled, requires a radio A that

receives a transmission correctly to transmit an ACK packet back to the sender, say radio

29
B, of that transmission. If B happens to be in an error region, it will not receive the ACK

correctly and the transmission will be considered a failure. For reliability sensitive

applications, the ACK option is expected to be indispensable and the ACK failure

conditions would affect the resulting PER performance. An ACK is transmitted at a

transmit-receive-turnaround time D after the completion of the data packet transmission.

The impact of ACKs on PER is not as obvious as that of data packets. Given a correct

data transmission taking place outside an error region, its ACK must also start and end

without crossing an error region for the data transmission to successfully complete.

Hence, the safe zone must be redefined as the areas where a data transmission can begin

such that it and its corresponding ACK will both complete outside any error region.

Timings not to scale

Figure 4.7. Safe zone illustration considering traversed range during CCA, packet time,
turn-around time, and ACK time, without random backoff. While not shown, the same
rule applies for the rest of the cycle.

30
 It can be shown that the safe zone with ACKs considered is either less than or

equal to the safe zone estimated without ACKs. As illustrated in Figure 4.7, the safe

zone size depends on the relative sizes of the turn-around time D and the subsequent error

region size after a data transmission. There are two cases shown in Figure 4.7. If the

traversed range during D is shorter than the following error region size, the ACK must

take place before the error region and thereby reduce the available safe zone size. The

range traversed during D, CCA, ACK or packet time depends on the rotation speed and

greater the rotation speed larger would be the range traversed So if the traversed range

during D is wider than the following error region size, ACK can potentially take place

after the error region and gain an additional safe zone that is equal to the range difference

between D and the error region size as shown in Figure 4.7. It is thus clear the safe zone

size changes with rotation speed, since the traversed range during an ACK time changes

accordingly. As the speed increases, the data packet range increases, such that the safe

zone decreases, but the D range increases as well, such that an additional safe zone

appears once D range exceeds the error region range; as the speed decreases, the reverse

trends apply. Thus, the overall safe zone and, therefore, the resulting PER is not

necessarily monotonic with the speed changes. The rotation speed at which the D range

equals the error region width is the speed at which the change can be noticed. This is

speed is calculated from the equation.

D*(rotation speed)*6 = Error region width. (2)

To illustrate the concept, randomized simulations were conducted with our

simulator with 92 byte packets and three 4-degree wide error regions centered at 0, 120,

31
and 240 degrees on the spindle. As shown in Figure. 4.8, the PER generally increased

with speed but had an exception decrease past 3500 rpm, after which the increasing trend

resumed. Equation (2) when solved for this scenario gives the same result of 3500 rpm.

0.9

0.8

0.7

0.6
PER

0.5

0.4

0.3

0.2

0.1

0
1000 1500 2000 2500 3000 3500 4000 4500 5000
Rotation Speed (rpm)

Figure 4.8. PER vs rotation speed .with three evenly spaced 4 degree error regions

4.3.2 Numerical Analysis of First Transmission Error Rate

First transmission error rate represents the percentage of packets received with

errors in their first transmission attempt.

In order to simplify the calculation of the first transmission error rate, the same

error system illustrated in Figure 4.5 is considered and the probability of bit error in the

32
high error region and low error region is considered as 1 and 0 respectively. Also the

rotation speed considered for this system is 2000 rpm. As explained in the previous

section the packet transmission start position determines its probability of success. The

probability of success of a packet transmission can be determined by finding the length of

the safe zone. The safe zone probability of success is 1. But as explained in the previous

section the safe zone size is also determined by the maximum back off duration Δ and the

ACK duration. Hence the probability calculation should also consider these factors.

If a packet transmission starts Δ+CCA+Packet time duration before the error

zone, the transmission would end before the error zone as illustrated in Figure 4.9. The

duration between CCA+Packet time and Δ+CCA+Packet time has a success probability

of ½, this is because a packet transmission starting in this region can choose between 0 or

1 random back-off period and if it chooses 0, it would finish its transmission before the

error zone, while if it chooses 1, it would end the transmission in the error zone.

33
 Figure 4.9. Probability of successful packet transmission illustration considering
traversed range during CCA, packet time, and random backoff time. While not shown,
the same rule applies for the rest of the cycle.

If the sensor position is such that the CCA duration and the random back off

duration ensures that the packet transmission starts only after the error zone, the packet is

successful. Similar to the above explanation there is a probability of ½ due to the

maximum back-off duration ∆. This is shown below in Figure 4.10.

34
 Figure 4.10. Probability of successful packet transmission to miss the error zone,
considering traversed range during CCA, packet time, turn-around time, and random
backoff time. While not shown, the same rule applies for the rest of the cycle.

Similar analysis applies for successful ACK. Figure 4.11 shows the ACK success

regions

Figure 4.11. Probability of ACK success illustration, considering traversed range during
CCA, packet time, turn-around time, and random backoff time. While not shown, the
same rule applies for the rest of the cycle.

35
 Probability Color
of error
1
1/2
0

Figure 4.12. First transmission error probability distribution for a three error zone system
centered at 0,120 and 240 degrees with width 4 degrees each (E1b=-2º, E1e=2º,
E2b=118º,E2e=122º,E3b=238º and E3e=242º).

We derive the probability distribution for a scenario with three error zones

centered at 0, 120 and 240 degree each with width 4 degrees. The corresponding

probability distribution is shown in Figure 4.12.

The first transmission error rate is given by the equation.

1
Area under the Red region in Degrees × 1 + Area under the Grey region in Degrees ×
2
360 360

To verify the equation simulations are done on the simulator and both the results

matched to give the first transmission error rate of 0.37.

36
 The calculation of overall packet error rate which includes the errors in the first,

second and other consecutive transmissions is a tedious and complex process. For

example the probability of error of the second transmission is a conditional probability

depending on the position of the sensor after the end of the first transmission. The overall

packet error rate equation will be hard to derive The error rate calculated using

conditional probability is for all the possible positions of the sensor and hence can be

verified with a simulation which has perfect randomness for all transmissions and

retransmission. Achieving this is difficult with the present simulator, so the calculation of

the packet error rate is not considered in this thesis.

4.4 Simulation Studies

An error region exists due to the multipath and Doppler effects associated with a

specific surrounding, sensor placement and rotation speed. Hence it is expected that the

error region distribution changes in different environments. To study the varied

transmission error characteristics in a wide range of possible surroundings, a software

simulator is developed using C++ to simulate the rotating process, the radio channel

properties, and the ARQ transmission method based on any given sensor rotation speed

R, the number of error regions and their respective widths and location around the

periphery of rotating structure. The simulator keeps track of the instantaneous location of

the radio. When a radio transmits a packet, the simulator computes the transmission

starting and ending times, and it identifies any segment(s) that are transmitted in an error

37
region. For each bit transmitted in the error region, a uniform bit error is generated

according to a specified bit error probability for each error zone.

The simulator also models the packet transmission and random wait durations

precisely according to the IEEE 802.15.4 specification. Each simulation experiment can

specify the number of packets to be transmitted, and all packets transmitted are recorded

including their received bit contents and timestamps. The simulation also models the

ACK transmission time during ARQ transmission.

The simulation is used to study the effect of: i) different error region widths for

one error zone, ii) different number of error regions, and iii) imprecise knowledge of the

actual rotation speed on the error region distributions, and (iv) a transmission error

avoidance scheme. The simulation results shown below are for a constant packet size and

considering a probability of bit error of 1 inside the error region and 0 outside the error

region. Also the packet generation interval is 0msec.

4.4.1 Effect of Different Error Region Widths

Figure 4.13 shows the error burst distance distribution observed with one error

region of different widths. As shown, for one error region, the main modes remained

near positions 0 and 1, indicating the consecutive errors occur either at the same location

or one spindle periphery distance away, which is again the same location. As the error

region width increases, however, a diluting effect to each mode size was observed. This

diluting effect is a direct result of the wider range of potential starting times of an error

38
burst; as a result the width of each mode is proportional to the error region width. The

proportional width dilution is observed clearly in Figure 4.13.

Error burst distribution for 5 degree error zone width

1

Probability 0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for 10 degree error zone width

1
Probability

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for 15 degree error zone width

1
Probability

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.13. Normalized error burst distance distribution obtained with error region
widths 5, 10, and 15 degrees at 2000 rpm.

4.4.2 Effect of Different Number of Error Regions

Figure 4.14 shows the error burst distance distribution for one, two, and three

evenly distributed error regions each of 4 degrees wide. As clearly shown, the error burst

distances correspond to that from an error region to any other regions as well as itself.

Hence, the error burst distance distribution graph can be used to estimate the number of

error regions in a rotating system if the rotation speed is accurately known.

39
 Error burst distribution for an one error zone system
1

Probability
0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for a two error zone system

1

Probability 0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for a three error zone system

1
Probability

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.14. Normalized error burst distance distribution for one, two, and three evenly
distributed error regions.

4.4.3 Effect of Imprecise Knowledge of Rotation Speed

The error burst distance distribution shown in Figure 4.1 is obtained with a known

rotation speed of 2054 rpm. Accurate rotation speed is not, however, always available to

either a sensor node or an external operator. This section studies the impact of an

imprecise knowledge of the rotation speed on the transmission error pattern. The analysis

is conducted by simulating the transmission errors at a rotation speed R rpm while

computing the error burst distances based on a potentially imprecise rotation speed of C

rpm. The following examines the effects of small inaccuracies that may be caused due to

inaccurate speed measurement, as well as the effects of large inaccuracies that may be

caused by a total lack of process knowledge. Insights to the latter case are essential for

deriving adaptive error pattern discovery methods that can adapt to varying rotation

speeds without any knowledge of such speed changes.

40
 Figure 4.15 shows the error burst distance distribution acquired with R=2000 rpm,

while C is (a) 50 and 100 rpm lesser than R, and (b) 50 and 100 rpm higher than R. As

seen, the inaccurate rotation speed adopted in the error burst distance analysis resulted in

various minor modes displaced from the main mode. Such displacements are due to the

incorrect assumption of a rotation cycle period. With C < R, the sampling period is

longer than an accurate period, causing an error burst at the same location to appear

relatively earlier during the sampling period. For an error burst that occurs in exactly one

true period after the previous error burst, the normalized error distance will be displaced

by δ= (1/R – 1/C) / (1/C) = C/R-1. A negative δ indicates a displacement to the left. If

the error burst occurs in exactly k true periods after the previous error burst, the error

distance will be displaced by k(C/R-1). Given the back-to-back transmission pattern in

this study, a consecutive error burst mostly occurs in the immediate next cycle, while the

probability of more cycles (elapsing prior to a next error) gradually decreases. The

finding suggests that a series of equally displaced modes be an indication of a slight

inaccuracy in the adopted rotation speed, and the true rotation speed can be found as

R=C/(1+δ).

41
 Error burst distribution for R=C+d
1

Probability
0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=C+2d

1
Probability
0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=C-d

1
Probability

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=C-2d

1
Probability

0.5

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.15. Error burst distance distribution obtained for R=2000 rpm and (a) C=1950
rpm, (b) C=1900 rpm, (c) C=2050 rpm, (d) C=2100 rpm.

Figure 4.16 shows the error burst distance distribution acquired with R = 2000

rpm and C = R/2 (1000 rpm) and C=2R (4000 rpm). As seen, the mode displacement

expression in the previous section still holds for large speed inaccuracies. With C=R/2,

the displacement was -0.5; with C = 2R, the displacement was 1, which is identical with

the original distribution.

Similar displacement expressions are found for scenarios with multiple error

regions while the derivations are more complex. For evenly distributed error regions and

C integer multiples of the actual rotation speed, the resulting error burst distance

distribution can be with the aliasing effect of a sampling process. Consider two error

42
regions 180 degrees apart. If R=C, the correct distribution is obtained with error burst

distances at 0, 0.5, and 1 (Figure 4.17 (a)). With a sampling speed of C = R/2, each

sampling cycle actually consists of two true cycles; hence, the error burst distribution is

expected to consists of two identical distributions in each half of its sampled cycle, as

seen in Figure 4.17 (b). With a sampling speed of C=2R, the sampling cycle corresponds

to only half of the true cycle; hence, one error region is sampled in each sampling cycle,

the other error region is sampled in the next sampling cycle at exactly one sampling cycle

later, thus resulting in a distribution identical to that one error region scenario (Figure

4.17 (c)). The figures indicate that the error burst distance distribution can be used to

identify errors in the rotation speed and also to determine the error region pattern of the

system.

43
 Error burst distribution for R=C
1

Probability
0.5 (a)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=2C

1
Probability

0.5 (b)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=C/2

1
Probability

0.5
(c)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.16 Error burst distance distribution for large speed inaccuracies with one error
region

Error burst distribution for R=C

1
Probability

0.5 (a)
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=2C

1
Probability

0.5
(b)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Error burst distribution for R=C/2

1
Probability

0.5
(c)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Figure 4.17. Error burst distance distribution for large speed inaccuracies with two error
region

44
 CHAPTER FIVE

TRANSMISSION ERROR AVOIDANCE

The purpose of the transmission error avoidance scheme is to enhance the

achievable data throughput of a sensor radio by avoiding transmission errors due to

rotation.

5.1 Overview of Proposed Method

As explained in previous chapters the knowledge of the error region pattern and

its characteristics is important because it indicates the regions of low error probability. To

enable error avoidance, the receiver records the time stamps and bit contents of all

packets received with errors; the error distribution is then analyzed and the transmitting

radio is informed to control its packet transmission time. The error transmission scheme

should include measures to overcome the effects of imprecise knowledge of the rotation

speed and the sensor clock drift. Another problem is the transmitting sensor has to be

synchronized to the error pattern. Our scheme takes all these factors into consideration

and consists of two different phases: the error identification phase and the operational

phase.

The error identification phase involves determining the error region distribution.

The purpose of obtaining an error region distribution is to identify the low or zero BER

regions within each rotation cycle for data transmission. Given the assumption that a

sensor does not know its position at a point of time, it relies on the transmission success

45
and failure events to infer whether it is inside or outside an error region each time it

transmits.

The error burst distance distribution discussed in Section 4.1 does not convey the

actual number of error regions or the location of such regions. Figure 5.1 shows the close

similarity between the simulated error burst distance distributions for two different error

region distributions. The three error zone system with error regions centered at 0, 90 and

180 degrees has a similar error burst distribution to a four error zone with error regions

centered at 0, 90,180 and 270 degrees. This makes it hard to distinguish between an

asymmetric three error zone system from a symmetric four error zone system by studying

the error burst distribution. This calls for the use of error burst location distribution. The

error burst location distribution is discussed below.

Figure 5.1. Two error region distributions can result in similar error burst distance
distributions: (a) three error regions centered at 0, 90, and 180 degrees; (b) four error
regions centered at 0, 90, 180, 270 degrees.

46
Definition 2: Error Burst Location Distribution

Given the beginning timestamps of a set of N consecutive error bursts, let the

normalized distance between the ith and i+1th error bursts be Di and TR the duration for

one revolution of the spindle, the first error burst’s beginning location as the reference

origin, the normalized error burst location for the ith error burst is defined as

⎛ i ⎞
Li = ⎜ ∑ D j −1 ⎟ mod TR (3)
⎝ j =1 ⎠
Sensor clock drift is the main challenge when obtaining the location distribution.

Clock drift renders the recorded timestamps and calculated error burst locations to drift

from their actual location. In fact, this was the main reason that error burst distance

distribution discussed in Section 4.1 was used initially to find discernible error regions

instead of using location distribution. As the machine does not provide digital clock

output, a visual assessment found the sensor clock to lead approximately 15 seconds

ahead the machine clock every 12 minutes. Fig. 5.2 shows the location distribution

subject to clock drift.

47
Figure 5.2. Location distribution with clock drift for error bursts shown in Figure 4.1.

Figure 5.3. Calibrated error burst location distribution.

48
 To compensate for the drift impact, accumulated error in the estimated error burst

locations must be calibrated periodically. Without external synchronization, the

calibration is done by analyzing the location distribution in small time windows and

aligning the resulting distributions. Fig. 5.3 shows the calibrated location distribution

derived by analyzing and aligning the location distribution for every 10 error bursts. The

calibrated error burst location distribution clearly indicates the error regions in the

system.

The error burst location distribution shown in Figure 5.4 corresponds to a system

with three error zones centered at 0, 90 and 180 degrees, each with a width of 4 degrees.

The error burst location distribution is assumed to be generated with the correct rotation

speed and the effects of the sensor clock drift is not considered. This figure when

compared to Figure 5.1(a) itself shows the effectiveness of using the error location

distribution because it clearly indicate the asymmetric nature of the system, while the

error burst distance distribution indicates a symmetric system The error zones are

identified by determining the bins of the error location distribution with a probability of

bit error greater than a particular threshold. The width of the error zone depends on the

resolution of the bins of the error burst location distribution. Figure 5.4 has a resolution

of 3.6 degrees, so each bar in the histogram corresponds to 3.6 degrees. The position of

the error zone is always found relative to the first error burst is shown in the figure.

Therefore the error burst distribution indicates error locations at 0, 0.25 and 0.5 (in terms

of spindle periphery) distance away from the first error seen.

49
 If the correct rotation speed is not used in generating the error burst location

distribution, it would not be possible to identify the error regions as discussed in Section

4.4. Figure 5.5 shows the percentage difference between the error burst location

distribution generated with a rotation speed of 2054 rpm and the error burst location

distribution generated with a rotation speed of 2000 rpm. The percentage difference is

large even after many probe packets are used in generating the location distribution. Thus

the error identification phase should begin with a phase to identify the correct rotation

speed before generating the error burst location distribution.

Figure 5.4. Error burst location distribution for a three error zone system centered at 0,90
and 180 degrees, each with a width of 4 degrees.

Another factor considered is the number of probe packets to be sent before which

it can be concluded that the error location distribution is an accurate representation of the

error properties of the system. Figure 5.6 shows the percentage error in the error location

50
distribution with the number of probe packets. So with more probe packets the error

location distribution becomes more accurate. The steps involved in the error

identification phase are explained in the next section.

Figure 5.5. Percentage error in error location histogram if the rotation speed is
considered as 2054 rpm when the actual rotations speed is 2000rpm, vs number of probe
packets with error

51
 Figure 5.6. Percentage error in error location histogram vs number of probe packets
with error

In the operational phase, the sensor begins with data transmission. Initially,

transmissions can occur at any time when data are present until a number of

transmissions are received with errors in a predefined time window. The sensor analyzes

the error packets’ timestamps to logically deduce its present location with respect to (i.e.

synchronize with) its error burst location distribution. Once it successfully synchronizes,

it starts to control the times of its data transmissions. Prior to each transmission, the

expected transmission interval is estimated based on the IEEE 802.15.4 medium access

control procedure and parameters; a data transmission is permissible when its expected

duration does not overlap with any error region; otherwise, the transmission will be

delayed until the end of an earliest error region with a following low-error duration that is

sufficient for the transmission to complete.

52
 To determine the exact time an error burst occurs within an arbitrary data packet

carrying real application payload is more difficult than that with probing transmissions.

Since the data contents are not known previously, the location where bit errors have

occurred cannot be directly determined. There are methods to explore this information

without substantial overheads. For example, since ARQ is adopted, packets are

retransmitted until they complete successfully. By caching and comparing packets in

error with their successful retransmissions, the error burst times can be found. For

another example, transmitted data can be encoded with known bit patterns at predictable

intervals in each packet, then error burst times in each packet can be identified at a

granularity proportional to the encoding interval. These approaches are not considered in

this thesis.

The purpose for an error avoidance approach is to improve energy efficiency and

data throughput. It is beneficial when the following conditions are satisfied

1. The PER for unconstrained transmissions is found to be unacceptably high

2. Error region distribution obtained in the error identification phase reveals

distinguishable error regions

3. Intervals between identified error regions have much lower PER and are long

enough to complete at least one data transmission

The proposed error avoidance scheme works by avoiding high error regions and if no

high error regions are identified for a system, then there would be no purpose in

implementing the error avoidance approach as there would not be any improvement in

the throughput. Unless there are low PER regions in between the high error regions there

53
would be no improvement in throughput. The packet size duration should also be lesser

than the interval between identified error regions or else the throughput would be less.

This case is illustrated in the simulation results section.

5.2 Error Identification Phase

In the error identification phase, the sensor uses probing transmissions to collect

sufficient transmission history to establish the error burst location distribution. The sensor

needs to first make sure that the generated error location distribution is correct. For this it

should first make sure that the assumed rotation speed of the spindle is correct. It can

make use of the error burst distribution to find out the error due to the imprecise

knowledge of the rotation speed as discussed in section 4.3.3. This would require a

sampling speed varying algorithm which is not discussed in this thesis and the rotation

speed is assumed to be correct in the simulations. The sensor should then generate the

calibrated error location distribution which is devoid of sensor clock drift errors. Given

the distribution, the center and width of each error region with PER above a chosen

threshold is identified. In section 4.2, it was shown that the error region width depends on

the assumed BER for the error region. The higher the assumed BER, the narrower the

width, while the more chances an error could occur outside the defined error region.

Probing packets are transmitted and analyzed until the percentage change in the error

location distribution is below a specified threshold. Figure 5.6 and section 5.1 discusses

the percentage change in the error location distribution with the number of probing

54
packets, but further discussions on the choice of threshold is not studied in this thesis. At

the end of the error identification phase the number of error regions, the center and width

of each error region is known.

5.3 Operational Phase

The phases involved in the operational phase are given below.

5.3.1 Synchronization

In the operational phase, the receiver first synchronizes with a given error burst

location distribution by observing the timing of multiple error bursts, it then proceeds

with transmission time control.

The pseudo code for synchronization is shown in Figure 5.7.At any given time if

the transmitting sensor wants to synchronize with the error regions, it requests the latest

N consecutive error burst’s burst distance sequence {Di} , the error burst location {Li},

the first error burst’s time stamp t0 and the receivers current time t_nowrx. It also records

its current time t_nowtx. Using the error burst distribution it checks the error burst

location distribution Li which it is closest to below a threshold E as shown in Figure 5.7.

Once synchronization is complete the transmitter can determine its location CTx(t_nowTx)

as Li + (t_nowRx – t0) * rotation speed.

55
 {Di}: error burst distances for N consecutive error bursts
{Lj}: error burst location distribution
M: number of identified error regions
δ: error region width
E: miss rate threshold
C: current location
s: miss count
i: error burst index
START
for j=1 to M
C= Lj, s=0, i=1, k=j;
while (i<N)
if( (C+Di) ∈ Lk+1±δ)
C= Lk+1, k=k+1;
else
C=C+Di, s=s+1;
endif
i=i+1;
end while
if (s/N)<=E
Synchronization Complete;
Break;
endif
end for

Figure 5.7. Synchronization algorithm

Once the synchronization completes successfully, the receiver informs the

transmitting sensor of the last error burst timestamp and its corresponding synchronized

location CTX. If the algorithm ends unsuccessfully, the algorithm must be repeated at a

later time.

56
 5.3.2 Transmission Error Avoidance

Assuming the transmitting sensor and the receiver have synchronized local

clocks, the transmitting sensor can determine at any time, its location and the duration

until beginning of the next error region. It also can determine the end of the nearest

following error region with a low-error region no shorter than the estimated transmission

time of the next packet. The transmitting sensor can then go on to schedule its

transmission to avoid the error regions.

The IEEE 802.15.4 MAC protocol version considered for implementing the

transmission error avoidance scheme is the non-beacon mode version. Before any packet

transmission the radio waits for a clear channel access (CCA) duration and a random

number of back-off slots τ (section 3.2). The standard allows setting minimum values for

the random backoff parameters to limit the minimum backoff window to 20 backoff

period which is considered for the proposed transmission scheme Hence the maximum

number of back-off slots ∆, before packet transmission is 1.

The error avoidance algorithm shown in Figure 5.8 calculates the delay that needs

to be injected before scheduling a packet transmission to avoid the error region. The

algorithm is run before every packet is sent out. Before every transmission the algorithm

checks if the CCA+∆+Packet time duration will result in the transmission ending in the

error zone. If this is the case then the packet transmission is scheduled to start only after

the error region. If the separation between the error zones is not large enough for a packet

transmission the transmission is further delayed to start only after the second error zone.

57
 {Lj}: error region locations
M: number of identified error regions
δ: error region width
P: packet transmission duration
C: current sensor location
CCA: CCA duration
∆: maximum back-off slot
START
for j=1 to M
if( ( (C+P+CCA+∆) ∈ Lk±δ ) || ( ( (C+CCA)< Lk-δ) &&
( (C+CCA+P)> Lk+δ) ) )
if(P>( Lk+1- Lk) )
C= Lk+1+δ-CCA;
else
C= Lk+δ-CCA;
endif
endif
i=i+1;
end for
Figure 5.8. Error avoidance algorithm

5.4 Simulation Results

The rotating wireless sensor simulator explained in section 4.3 is used to verify

the effectiveness of the transmission error avoidance scheme. The error regions are

assumed to be identified and the error identification phase is assumed to be successful.

The radios are also assumed to have synchronized with the error distribution. The IEEE

802.15.4 non-beacon mode data and ACK transmission timing is modeled. Table 5.1

summarizes the simulation parameters.

58
 Table 5.1. Simulation parameters

Parameter Value
Radio link rate 250 kbps
Radio symbol duration 16 usec
CCA duration 8 symbols
Radio turn-around time 12 symbols
ACK frame 22 symbols
ACK timeout 2.4 msec (sensor default [11])
Spindle rotation speed 2000 rpm
Backoff slot duration 20 symbols

The simulations are done to verify the advantages of the transmission error

avoidance scheme mainly the packet error rate and throughput improvement. The effect

of the number of error regions on the performance of the error avoidance scheme requires

testing. To do so, simulations are done on a one error zone symmetric system and a four

error zone symmetric system. The bit error rate inside the error zone is considered as

100% and outside the error zone as 0%. In reality the bit error rates inside the error zone

would be less than 100% and outside the error zone would be greater than 0%. The effect

of such a scenario is also studied. Lastly the performance in an asymmetric two error

zone system is studied to understand the importance of the packet size in the error

avoidance scheme. The results of the simulations are discussed below.

59
100 % BER Inside Error Region, 0% BER Outside Error Region

This is an idealized setting that illustrates the typical benefit of the scheme. Figure

5.9 shows the PER and data throughput achieved when there was one error region (4

degree wide) with and without error avoidance, with different packet sizes and packet

generation intervals. A 0-second interval represents saturating transmission at the link’s

maximum capacity. In Figure. 5.9(a), it is seen that larger packets had higher PERs on

average; PERs remained consistent with different packet generation intervals, reaching

about 10% with 92 byte packets.

60
 0.2
Packet size 30B without EA
0.18 Packet size 60B without EA
Packet size 92B without EA
0.16
Packet size 30B with EA
0.14 Packet size 60B with EA
Packet Error Rate

Packet size 92B with EA

0.12

0.1

0.08

0.06

0.04

0.02

0
0 0.005 0.01 0.015
Packet generation interval(sec)

(a)

200
Packet size 30B without EA
180
Packet size 60B without EA
160 Packet size 92B without EA
Packet size 30B with EA
Throughput (kbps)

140 Packet size 60B with EA

Packet size 92B with EA
120

100

80

60

40

20

0
0 0.005 0.01 0.015
Packet generation interval(sec)

(b)

Figure 5.9. 2000rpm, one 4-degree error region, (a) PER v.s. packet size v.s.
transmission interval (b) throughput v.s. packet size v.s. packet interval.

61
 With error avoidance, errors were completely avoided, while a slight decrease in

throughput was seen.

A more challenging scenario was studied with four error regions each 4 degree

wide and centered at 0, 90, 180, and 270 degrees. Figure 5.10 shows the PER and Figure

5.11 shows the throughput. Interestingly, with scattered error regions, PERs increase as

packets are generated faster, reaching 50% with 92 byte packets. With error avoidance,

throughputs were substantially increased (50 to75% for continuous transmissions). Also

interesting was that with- and without-error-avoidance throughput curves came close as

the packet interval increased; they merged eventually, and the merging point depended on

the packet sizes – the larger the size, the later the merge, the more throughput gain

achievable with error avoidance.

62
 1
Packet size 30B without EA
0.9
Packet size 60B without EA
0.8 Packet size 92B without EA
Packet size 30B with EA
Packet Error Rate

0.7 Packet size 60B with EA

Packet size 92B with EA
0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.005 0.01 0.015
Packet generation interval(sec)

Figure 5.10. 2000rpm, four 4-degree error regions PER v.s. packet size v.s. transmission
interval
120
Packet size 30B without EA
Packet size 60B without EA
100 Packet size 92B without EA
Packet size 30B with EA
Packet size 60B with EA
Throughput (kbps)

80 Packet size 92B with EA

60

40

20

0
0 0.005 0.01 0.015
Packet generation interval(sec)

Figure 5.11. 2000rpm, four 4-degree error regions Throughput v.s. packet size v.s.
transmission interval

63
Less than 100 % BER Inside the Error Region, non Zero BER Outside the Error

Region

In practice, BERs inside and outside error regions may not be 100% and 0%.

Figure 5.12 shows the achieved throughput with different BERs outside the error region.

When BERs were 0.01 or less, error avoidance increased achievable throughput by nearly

40% when data rate is very high. Once the BER approached 0.1, substantial errors

occurred even outside the error region therefore error avoidance was no longer beneficial.

The BERs at different levels of magnitude were chosen to represent a wide range of

typical environments.

120
BER 0 w ithout EA
BER 0.001 w ithout EA
BER 0.01 w ithout EA
100
BER 0.1 w ithout EA
BER 0 w ith EA
Throughput (kbps)

BER 0.001 w ith EA

80 BER 0.01 w ith EA
BER 0.1 w ith EA

60

40

20

0
0 0.005 0.01 0.015
Packet generation interval(sec)

Figure 5.12. 2000rpm, four 4-degree error regions, throughput v.s. PER outside error
region with 92 byte packets (BER in error region is irrelevant since transmissions are
entirely avoided).

64
Packet Size Relationship with Error Region Separation

A two error system with error regions centered at 0 and 60 degree, each with a 4

degree width is considered, the BER inside the error region is 1 while the BER outside

the error region is 0. The throughput for different packet sizes at different packet

generation intervals with error avoidance is analyzed. Figure 5.13 shows the result. The

distance between the error regions corresponds to the transmission duration of a 150 byte

packet. So for any packet size above 150 bytes the two error regions are considered as

one error region by the transmission avoidance scheme and hence the throughput

decreases with the increase in packet size. This result can be used in deciding the packet

size for transmission. Another factor which is seen is that the packet generation interval

doesn’t have an effect on the throughput for larger packet sizes. This is because the

packet transmission interval itself would be smaller than the packet transmission duration

hence resulting in a constant throughput for different packet generation intervals.

65
 200
Packet size 50B
180
Packet size 100B
160
Packet size 150B
Packet size 200B
Throughput (kbps)

140 Packet size 250B

120

100

80

60

40

20

0
0 0.005 0.01 0.015
Packet generation interval(sec)

Figure 5.13. 2000rpm, two 4-degree error regions centered at 0 and 60 degrees,
throughput v.s. packet size for different packet generation interval.

66
 CHAPTER SIX

CONCLUSIONS AND FUTURE WORK

6.1 Conclusions

The thesis utilized a rotating machine testbed as well as a software simulator to

investigate the transmission error patterns experienced by a wireless sensor mounted on a

fast rotating structure. A model for transmission error patterns is developed through

analysis of packet transmissions in a testbed. The model is used for on-line error pattern

identification, communication performance prediction, and development of intelligent

error-avoidant transmission methods.

From the testbed experiments it is found that:

• One specific error region on the spindle periphery is identified by analyzing

the transmitted packets’ normalized error burst distance distribution,

• The width of the error region is estimated using a sliding window algorithm

with a range of different packet error rates, and

• The normalized error burst distance distribution sensitively depends on the

accurate knowledge of the true rotation speed.

From the simulation experiments it is found that:

• The larger width of an error region led to a dilution of each mode in the error

burst distance distribution,

67
 • From an error burst distance distribution the number of distinct error regions

can be estimated,

• Discrepancy between the true rotation speed and the speed adopted by the

error burst distance analysis resulted in a series of aliasing modes, and

displacement between such modes can be used to assess the speed

discrepancy, and

• With multiple error regions, the speed discrepancy can be similarly identified

by analyzing the aliasing effects.

• It was shown that a relatively small region with disadvantaged radio

propagation can cause substantial errors and energy wasted in retransmissions.

In this thesis the PER is analytically shown to have a generally increasing trend

with increasing rotation speeds, while exceptions can occur at certain speeds depending

on the error region distribution if the ACK option is enabled. Numerical analysis was also

able to predict the first transmission rate accurately for a given error system.

This thesis also proposes an online error inference and error-avoidance

transmission method for sensors with low accuracy clocks. With simulation, the method

was shown to be particularly effective when required data transmission rates are high and

error regions are scattered around the periphery. For a one error zone system with an

error region width of four degree, the packet error rate was reduced to 0 with

transmission error avoidance. A four error zone symmetrical system with error region

widths 4 degree shows 50% reduction in packet error rate with a 75% improvement in

throughput. The packet size also affects the data transmission throughput. Larger the

68
packet size, lesser would be the throughput for an asymmetrical two error zone system

with error width 4 degrees.

6.2 Future Work

The first phase of the error avoidance scheme is the error identification phase. The

effects of imprecise knowledge of rotation speed and the effects of clock drift on the error

burst distribution are shown. It was shown in Section 4.4 that by changing the sampling

speed and analyzing the error burst distribution, the correct rotation speed can be

determined. Such a rotation speed aware algorithm needs to be incorporated into the error

identification phase to determine the correct rotation speed.

It was also shown in Section 5.1 that calibration of the error burst location

distribution is required to avoid the clock drift errors. This calibration algorithm needs to

be incorporated and verified for different scenarios. Also the synchronization algorithm

in the presence of clock drift and rotation speed error needs to be verified.

The present simulator can simulate the error characteristics of up to four error

zones. A more generalized simulation to generate the error characteristics for any number

of error zones needs to be developed.

69
 REFERENCES

[1] Akyildiz, I. F., Su, W., Sankarasubramaniam, Y., and Cayirci, E.,
“Wireless Sensor Networks: A Survey,”Computer Networks
38(4), 393-422 (2002)

[2] Straser, E. G., [A Modular, Wireless Damage Monitoring System for Structures],
Ph.D. Thesis, Stanford University, Stanford CA (1998).

[3] Miettien, J., Salemenpera, P., Jarvinen, V., and Hervonen, M.,
“Wireless Operation Monitoring System for Polymer Covered
Cylinders in Rolling Contact,” in Proceedings of the ASME
Engineering Technology Conference on Energy, Petroleum
Division 2, 1019-1023 (2002).

[4] Varghese, B., Pathare, S., Gao, R., and Guo, C., “In-process Monitoring
of Truing Using a Sensor Integrated Diamond Grinding
Wheel,” Transactions of NAMRI/SME 30, 295-302 (2002).

[5] Dzapo, H., Stare, Z., and Bobanac, N., “Portable Wireless Measuring
System for Monitoring Motor Shaft Parameters,” in Proceedings
of the Instrumentation and Measurement Technology Conference,
901-906 (2004).

[6] Sarkimaki, V., Tiaien, R., Lindh, T., and Ahola, J., “Applicability of
ZigBee Technology to Electric Motor Rotor Measurements,”
in Proceedings of the International Symposium on Power
Electronics Electrical Drives Automation and Motion, 1-5 (2006).

[7] Sundararajan, V., Redfern, A., Schneider, M., Wright, P., and Evans, J.,
“Wireless Sensor Networks for Machinery Monitoring,” in
Proceedings of the International Mechanical Engineering
Congress and Exposition, 1-9 (2005).

[8] Werb, J., Newman, M., Berry, V., Lamb, S., Sexton, D., and Lapinski, M.,
“Improved Quality of Service in IEEE 802.15.4 Mesh
Networks,” in Proceedings of the International Workshop on
Wireless and Industrial Automation, 1-6 (2005).

[9] Ota, N., and Wright, P., “Trends in Wireless Sensor Networks
for Manufacturing,” International Journal of Manufacturing
Research 1(1), 3-17 (2006).

70
[10] Tang, L., Wang, K.-C., Huang, Y., and Gu, F., “Channel Characterization
and Link Quality Assessment of IEEE 802.15.4-Compliant
Radio for Factory Environments,” IEEE Transactions on
Industrial Informatics 3(2), 99-110 (2007).

[11] Wang, K.-C., Tang, L., and Huang, Y., “Wireless Sensors on Rotating
Structures: Performance Evaluation and Radio Link
Characterization,” in Proceedings of the WinTECH Workshop
at ACM MobiCom, 3-10 (2007).

[12] Tang, L., Wang, K.-C., and Huang, Y., “Performance Evaluation and
Reliable Implementation of Data Transmission for Wireless
Sensors on Rotating Mechanical Structures,” Structural
Health Monitoring, accepted (2008).

[13] Wang, K.-C., Jacob, J., Tang, L., Huang, Y., and Gu, F., “Error Pattern
Analysis for Data Transmission of Wireless Sensors on Rotating
Industrial Structures,” in Proceedings of SPIE, vol. 6932, 2008.

[14] Wang, K.-C., Jacob, J., Tang, L., and Huang, Y., “Transmission Error
Avoidance for IEEE 802.15.4 Wireless Sensors on Rotating
Structures,” in Proceedings of ICCCN 2008.

[15] IEEE, [IEEE Standard 802 Part 15.4: Wireless Medium Access Control (MAC)
and Physical Layer (PHY) Specifications for Low-Rate Wireless
Personal Area Networks] (2003).

[16] Zhao , J., Govindan, R., “Understanding packet delivery performance

in dense wireless sensor networks,” in Proceedings of 1st
ACM Conf. Embedded Network Sensor Systems, 1–13(2003)

[17] Lal, D., Manjeshwar, A., Herrmann, F., Uysal-Biyikoglu.

E., Keshavarzian, A.,“Measurement and characterization of
link quality metrics in energy constrained wireless sensor
networks,” in Proceedings of Globecom,446–452(2003)
[18] Ganesan, D., Krishnamachari,B., Woo, A., Culler, D., Estrin, D.,
Wicker, S., “Complex Behaviour at Scale: An Experimental
Study of Low-Power Wireless Sensor Networks”,
UCLA, Tech. Rep. CSD-TR 02-0013, 1–11(2003).

[19] Srinivasan, K, Levis, P., “RSSI is under appreciated,” in

Proceedings of. 3rd Workshop on Embedded Networked
Sensors, 1–5(2006).

71
[20] Valente, A., Morais, R., Serodio, C., Mestre, P., Pinto, S., Cabral, M.
“A ZigBee Sensor Element for Distributed Monitoring of
Soil Parameters in Environmental Monitoring,” in Proceedings
of IEEE Sensors, 135-138 (2007)

[21] Cho, H., Jung, Y., Choi, H, Jang, H., Son, S., Baek, Y., “Real Time Locating
System for Wireless Networks using IEEE 802.15.4 Radio,
” in Proceedings of SECON , 578-580 (2008)

[22] Hara, S., Zhao, D., Yanagihara, K.; Taketsugu, J.; Fukui, K.,
Fukunaga, S., Kitayama, K., “Propagation characteristics
of IEEE 802.15.4 radio signal and their application for
location estimation”, in Proceedings of Vehicular Technology
Conference, 97 – 101 (2005).

[23] Hui, Jonathan W., Culler., David E.”Extending IP to Low-Power, Wireless

Personal Area Networks”. Internet Computing 12(4), 37 – 45(2008)

[24] Rappaport, T. S., [Wireless Communication – Principles and Practice],

2nd Ed., Prentice Hall, New Jersey (1996).

[25] Chiasserini, C-F. , and Garetto, M., “An Analytical Model for Wireless
Sensor Networks with Sleeping Nodes,” Transactions on
Mobile Computing 5(12), 1706-1718(2006).

[26] Crossbow Technology, Inc., http://www.xbow.com, accessed April 2007 (2007).

[27] CC2420 data sheet, http://www.chipcon.com/files/CC2420_Data_Sheet_1_3.pdf,

accessed April 2007 (2007).

72