Resource efficient floating-point data compression using mas in wsn

International Journal of Ad hoc, Sensor & Ubiquitous Computing (IJASUC) Vol.4, No.5, October 2013

RESOURCE-EFFICIENT FLOATING-POINT DATA
COMPRESSION USING MAS IN WSN
Maher El Assi1, Alia Ghaddar2, Samar Tawbi3, Ghaddar Fadi4
1, 2, 3
4

Lebanese University, Lebanon
Saint Joseph University, Lebanon

ABSTRACT
In a wide range of applications, large amounts of floating-point data are generated by Wireless Sensor
Networks (WSNs). This data is often transferred between several sensor nodes, in a multi-hop fashion,
before reaching its ultimate destination (the base station). It is well known that data communications is the
most energy-consuming task in sensor nodes [1]. This can be a great concern when the nodes are
constrained in energy. Therefore, the amount of data to be transferred between nodes should be reduced to
save energy. In this paper, we investigate data compression for resource-constraint WSNs; we introduce
MAS as a novel adaptive lossless floating-point data compression algorithm for WSNs. MAS exploits the
disproportionality in energy consumption between data transmission and processing. Simulation results,
obtained from OMNeT++ and Atmel Studio, show that MAS surpasses other tested compression algorithms
in terms of compression ratio, compression speed, memory requirements and most importantly energy
savings.

KEYWORDS
Wireless Sensor Networks, Lossless Compression, Floating-point Data, Energy Efficiency

1. INTRODUCTION
In the last few decades, Wireless Sensor Networks (WSNs) has proven to be an interest grabbing
technology, offering great contributions in several application domains. Wireless Sensor
Networks can provide a low cost solution to a variety of real-world problems including but not
limited to health care, industry process control, object tracking, volcanic and seismic monitoring,
smart parking, home automation, etc. Moreover, WSNs can provide enhanced situation awareness
in responding to today’s public safety situations. For example, Sleep Safe project is designed for
monitoring infants while they sleep. Sleep Safe sensor nodes can prevent sudden infant death
syndrome (SIDS) by autonomously detecting the sleeping position of an infant and alerting the
parents wirelessly in real time when the infant is lying on its stomach [2].
Typical WSNs are composed of a relatively high number of sensor nodes communicating through
an infrastructure-less multi-hop wireless network architecture. These nodes usually perform three
main tasks: data collection, data processing and data communication. The nodes capture data
from their surrounding environment, they process it and finally they transfer it to a base station
where decision makers can make use of it.
Sensor nodes are small, cheap and smart devices that are made up of four basic components [3],
as shown in Figure 1:
DOI : 10.5121/ijasuc.2013.4502

13


1)
2)
3)
4)

A sensing unit which captures a physical quantity from the environment.
A processing unit which processes and analyzes the captured data.
A transceiver which is responsible for data communication.
A power unit which is in most of the cases a battery.

Sensing Unit
Sensor

ADC

Processing
Unit
Processor

Transceiver

Storage

Power Unit
Figure 1. Main Components of a sensor node. [3]

Despite their promising range of applications, most WSNs are constrained in resources; they have
limited amount of energy, limited processing capabilities, short range of communication and
limited memory size. Out of these constraints, energy is considered the primary concern
especially for battery-operated sensors; this is true because when a sensor node is depleted of
energy, it would be useless for the network. This could affect the performance of the whole
network especially if the node is used in critical locations, such as mines, volcanoes, etc.

Instruction Cycles for same
Energy as One Byte
Transmitted

Over the past years, different studies and techniques have been proposed for WSNs to reduce
energy consumption and increase network performance and lifetime. Data compression has been
adopted as a practical technique and reliable solution in terms of energy efficiency in WSNs. The
efficiency of data compression techniques mainly bears on the drastically disproportionate energy
cost between data transmission and processing. This can be seen in Figure 2, which shows how
many compute cycles, on a Texas Instruments MSP430 microcontroller, would be performed for
the same amount of energy required to transmit a single byte over three commonly used radios [7]
(Chipcon CC2420 [4]: short range 125 m, the Chipcon CC1000 [5]: medium range 300 m, and
MaxStream XTend [6]: long range 15 km).

10000000
1000000
100000
10000
1000
100
10
1

CC2420

CC1000

Xtend

Radio
Figure 2. Number of TI MSP430F1611 compute cycles that can be performed for the same amount of
energy as transmitting one byte over three radios. [7]

It is obvious that the most energy consuming part in WSNs is the communication. Approximately
80% of power consumed in each sensor node is used for data transmission [1]. Thus to save
14


energy and maximize network lifetime, data transmissions should be minimized without losing
vital information. The lower the size of the transmitted data the lower the number of required
transmissions. In our work, we study data compression as the technique for minimizing data size.
In this paper, we propose an energy efficient floating-point data compression algorithm for WSNs
called MAS. MAS is a new adaptive streaming lossless compression algorithm that relies on an
accommodative coding technique to achieve compression at low processing costs. MAS offers
great contributions to WSNs, because it is one of the first algorithms to specifically compress
floating-point data. By focusing on floating-point data, it is possible to achieve much better
compression ratios because we can exploit the characteristics and the nature of numbers to build
our algorithm. In fact, floating-point data is generated in a wide range of applications such as
weather monitoring (temperature, pressure…), healthcare (blood pressure, cardiac activity…),
localization and tracking (position, height, coordinates…), industry (temperature, vibrations,
radioactivity…), etc.
The remainder of this paper is organized as follows: Section 2 gives an overview of common
compression schemes specifically designed for WSNs. Section 3 presents our newly proposed
algorithm. Section 4 presents an evaluation of the presented algorithms. Section 5 concludes this
paper while Section 6 presents future works.

2. RELATED WORK
Energy efficiency has been a major concern in the design and development of WSNs. Since radio
communication is known to be the main source of energy consumption, most of the proposed
techniques in the literature, which aim to increase energy savings, have focused on reducing data
communication (transmission/reception). Data compression is such a technique, which is often
used in conjunction with data collection techniques to transmit the collected data in an energy
efficient manner.
Due to the distributed nature of WSN applications, and the resource-constrained nature of sensor
nodes, traditional data compression techniques cannot be easily used. It may not be feasible to run
sophisticated data compression algorithms on sensor nodes. The limited resources available in
these nodes demand the development of specifically designed algorithms. In this section, we
present two famous compression algorithms used in WSNs: SLZW [7] and K-RLE [8]. We also
present our proposed compression algorithm MAS [9] and compare it with these algorithms.

2.1. S-LZW
S-LZW [7] (Sensor-LZW) is an adaptation of the popular lossless data compression algorithm
LZW [10]. S-LZW follows the same procedure used by the LZW algorithm, but with little
restrictions regarding the size of the used data structures. The added restrictions ensure that the
requirements of the algorithm are still within the bounds of the available resources in sensor
nodes.
Before heading into the details of the modification, it is first important to understand why LZW is
not suitable for WSN. LZW is a dictionary-based compression algorithm; it works by converting
strings of symbols into integer codes. LZW does not use a static dictionary; instead, it builds the
dictionary on the fly in a special way to allow both the encoder and the decoder to be able to
generate the same dictionary from the input data. First, both the encoder and the decoder initialize
the dictionary with 256 entries containing the symbols in the ASCII code. Then the dictionary
continues to grow while parsing the input.

15


The dictionary is the main obstacle preventing LZW from being applicable to WSN. Throughout
the compression mechanism, the dictionary keeps growing and can reach sizes much higher than
the available RAM on sensor nodes, and this can clearly disrupt the stability of the system.
Another problem that LZW faces is that it requires a predefined data volume, i.e. in order for it to
start the compression procedure a significant amount of data must already be available. That is
why S-LZW can only be used in delay tolerant networks.
Several modifications were done on LZW to make it portable to WSNs. Most of these
modifications focus on reducing the amount of RAM required for LZW to operate. Here is a list
of modifications that gave the birth to S-LZW:

•

S-LZW uses a 512-entry dictionary. As we mentioned before, this dictionary will be
initialized with 256 ASCII code symbols. With this size, the dictionary may get full while
compressing or decompressing certain datasets. There are two protocols to follow when the
dictionary fills, either fix the dictionary to its state whence it get full, or reset the dictionary to
the 256 entries. The authors of S-LZW [7] proved that using the fixed protocol produces
better results when compressing data of small block sizes (528 bytes).

•

S-LZW divides input data into block sizes of 528 bytes, and then it compresses these blocks
individually. It is important to note that S-LZW requires 528 bytes of data to be available in
order to compress it, if this amount of data is not available, it has to wait until data
accumulate and reach the required size because compressing data of smaller size will be
inefficient as shown in [7]. WSN data sampling rate is relatively low and it may take some
time to collect 528 bytes of data to be able to start the compression. This is why this
algorithm can only be used in delay tolerant networks.

•

The last modification enhances S-LZW by allowing it to benefit from the similarity of data
generated by sensor nodes. This is done by adding a mini-cache, which is a hash-indexed
dictionary of size N, where N is a power of two, which stores recently used and created
dictionary entries. The authors show that it is best to use mini-caches of sizes 32 or 64
dictionary entries.

2.2. K-RLE
K-RLE [8] is a new compression algorithm whose idea is inspired from the lossless data
compression algorithm RLE [11]. RLE stands for Run-Length Encoding, which is a very basic
and simple compression algorithm that works in this way: if a data item d occurs n consecutive
times in the input stream, we replace the n occurrences with a single pair nd.
RLE itself is very simple and can be used in WSN without any major changes, its RAM and
processing requirements are very low. However, there is a major limiting constraint in RLE, for
RLE to achieve good compression ratio, the input data must contain long sequences of repeated
characters, and this rarely occurs in the data generated from sensors. To solve this problem, KRLE algorithm has been proposed; K-RLE means RLE with K precision.
The idea behind this algorithm is: let K be a number, if a data item d, d+K, or d-K occur n
consecutive times in the input stream, we replace the n occurrences with a single pair nd [8]. The
new addition in K-RLE allows it to achieve higher compression ratios than RLE but at an even
cost, which is data loss. In contrary to RLE, K-RLE is a lossy compression algorithm, and the
amount of data loss is strongly related to the K parameter. Higher values for K means better
compression ratio but more data loss, while lower values of K means lower compression ratio but
less data loss.
16


There are two main advantages of K-RLE over S-LZW:
1) K-RLE uses much less amount of RAM than S-LZW, so it can be used in several sensor
platforms where S-LZW cannot be used.
2) K-RLE has the streaming feature, which means it does not need to buffer data before
being able to start the compression process. So K-RLE can be used in networks that
cannot tolerate delay.
The main two disadvantages of K-RLE are:
1) It is a lossy algorithm, so it is not suitable for some applications
2) It requires the input data to contain long sequences of similar characters in order to have a
good compression ratio.

3. MAS COMPRESSION ALGORITHM
MAS stands for Minimalist, Adaptive and Streaming compression algorithm. Minimalist means
that it uses the minimum possible amount of resources. Adaptive means it generates variable-size
output according to the number of digits in the input, and Streaming means that it does not require
buffering of the data before starting the compression process.
MAS is a specialized lossless compression algorithm that only compresses single-precision
floating-point data. MAS’ implementation does not require any correlation or similarity in the
input data, which makes it general and applicable in various domains.
MAS can encode any floating-point number satisfying the following two conditions:
1) The number of significant digits should be at most 7, if a floating-point number exceeds 7
significant digits it would be truncated.
2) The floating-point number when put in scientific notation must have a power of 32 or
less.
Although these conditions mean that some numbers representable by IEEE standard 754 [12] will
not be representable in MAS. However, these numbers almost do not exist in the data generated
by WSN. Numbers that have decimal powers of more than ±32 are almost not found in any
application in WSN.
One of the greatest merits of MAS is that it does not require any floating-point operation
(addition, subtraction, multiplication, division) to compress floating-point numbers. This is very
important because most microcontrollers and processors in sensor platforms are not equipped
with an FPU (Floating point unit). The FPU is responsible for carrying out operations on floatingpoint numbers. In the absence of an FPU, these operations are emulated in software but at the cost
of time and cycles, which could lead to higher energy consumption. Internally, MAS treats
floating-point numbers as strings of characters to carry out the needed operations with a low
number of cycles.

3.1. MAS Encoding Technique
The first step in encoding a floating-point number is to write it in scientific notation. Scientific
notation allows the representation of very small or big numbers with ease. So any number is first
written under the following format (d = digit, e = exponent):
17


± ݀. ݀݀݀݀݀݀ ‫±01 ݔ‬௘
To encode a number, the different parts in the above format must be encoded. MAS encodes them
in five sections detailed below and are shown in Figure 3 from left to right:

•
•
•
•
•

Number of significant digits (n) (number of d’s): represented on 3 bits because its
maximum value is 7.
The exponent (e): represented on 5 bits because its maximum value is 32.
Number sign (ns): represented on 1 bit. (0 for positive, 1 for negative).
Exponent sign (es): represented on 1 bit. (0 for positive, 1 for negative).
The integer formed by the d’s without the decimal point (ddddddd): variable bit size
depending on the number of d’s (maximum 24 bits). Details are in Table 1.

n n n | e e e e e | ns | es | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Figure 3. MAS encoding of a real number.
Table 1.Number of bits needed to represent an integer in binary formed by a certain number of digits.

Number of digits
1
2
3
4
5
6
7

Number of needed bits
4
7
10
14
17
20
24

It is clear that MAS exploits the significant number of digits to represent floating point numbers,
for example: -0.0001 is represented on 14 bits while 92301.1 is represented on 30 bits.
Integers form a large part of real numbers, but the above representation may be unfair for integers
because there is no need for the exponent and its sign when representing integers. Therefore, a
special encoding has been chosen for integers, but there should be a discriminator for the decoder
to know which type of number it is going to decode. We exploit the fact that the number of
significant digits cannot be zero, and use this as a discriminator between the 2 encodings. Integer
representation thus has four parts as shown in Figure 4:

•
•
•
•

3 bits that are all zeroes acting as a discriminator.
Number of significant digits (n) in the integer: 3 bits.
Number sign (ns): 1 bit.
The integer: variable bit size following Table 1 (maximum 24 bits).

0 0 0 | n n n | ns | _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _
Figure 4. MAS encoding of an integer.

We also exploit the fact that in an integer the number of significant digits cannot be zero to make
a special representation for the actual zero. A zero is represented in MAS as 6 zero bits: 000000.
Finally, since our algorithm generates codes of variable size, we need to make sure that the
overall output is aligned to memory; we do this by filling the required number of bits by 1s. The
18


number of alignment bits is between 1 and 7, and they are added only once at the end of the
output.

3.2. MAS Decoding Technique
The decoding technique is straightforward and is done following these steps:
1) Read the first three bits.
2) If the bits are zeroes, the encoded number is an integer.
a. Read the next three bits to extract the number of significant digits.
b. Read the next bit to determine the number’s sign.
c. Read the number of bits required to extract the number referring to Table 1, then
extract the integer.
d. Combine the readings so far to decode the integer.
3) If the bits are non-zeros, the encoded number is a real.
a.
b.
c.
d.
e.
f.

The already read bits represent the number of significant digits.
Read the next five bits to extract the exponent.
Read the next bit to determine the number’s sign.
Read the next bit to determine the exponent’s sign.
Read the number of bits required to extract the number referring to Table 1, then
extract the integer.
Combine the readings so far to decode the real number.

The explanation of MAS’ encoding and decoding techniques clarifies why MAS is considered a
streaming algorithm. MAS can compress or decompress even one single value, and does not
require predefined data volume as S-LZW.

4. EVALUATION
The evaluation metrics of any compression algorithm for WSN are based on the resource
limitations of sensor nodes. Thus, we chose the following metrics to evaluate the presented
algorithms: compression ratio, processing cost, memory requirements, and energy savings.
In the following sub-sections, we will start by describing the chosen platform and the chosen
simulators, and then we will move to present the simulation results and evaluate the algorithms
performance.

4.1. Platform Overview
Our chosen platform is the Waspmote [13] because of its interesting characteristics. A Waspmote
can be connected to 60 sensor types and can support up to 8 different wireless technologies.
A Waspmote uses an 8-bit AVR microcontroller called ATmega1281 [14], which is low power
microcontroller provided by Atmel [15]. It is supplied with a 128 KB flash (program) memory
and an 8 KB RAM. The microcontroller can have frequencies between 0 and 16 MHz at an
operating voltage of 1.8/5.5 V. In the active mode, as shown in Figure 5, the microcontroller
consumes 1 mA at a voltage of 3.3 V and frequency of 1 MHz; this means this microcontroller
consumes 3.3 nJ for one computation cycle in its active mode. This microcontroller does not have
an FPU unit.
19

2.5
5.5V

2

5.0V
4.5V
4.0V

1

3.3V
2.7V

0.5

Icc ( mA)

1.5

1.8V

0
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Frequency (MHz)
Figure 5. Active Supply Current vs. frequency (0.1MHz - 1.0MHz) on ATmega1281 microcontroller. [14]

We chose the wireless technology to be CC2420 RF [16] transceiver, which complies with the
IEEE 802.15.4 standard [17]. This transceiver is designed for low-power and low-voltage
wireless applications. It has low current consumption; for transmission, it consumes 17.4 mA and
for reception, it consumes 18.8 mA. It has an effective data rate of 250 kbps.

4.2. Simulators Overview
In order to calculate the total energy consumed, we need to calculate two different kinds of
energy, the computation energy, and the communication energy. To calculate the computation
energy, we calculate the number of cycles needed to achieve the required computation and this is
done by using the Atmel AVR Studio [18]. To calculate the communication energy, we calculate
the transmission energy only at the node performing the compression using OMNeT++ [19].
Atmel AVR Studio [18] is an Integrated Development Environment (IDE) for writing and
debugging AVR/ARM applications. It supports the complete range of Atmel AVR tools and
devices. The simulator in AVR Studio can simulate the CPU, including all instructions, interrupts,
and most of the on-chip I/O modules. We use Atmel AVR Studio to calculate the required
number of cycles as well as the memory requirements of the algorithms.
OMNeT++ [19] is an object-oriented modular discrete-event network simulation framework.
OMNeT++ itself is not a simulator, but rather provides infrastructure and tools for writing
simulations. It is considered the best simulation framework for WSN as demonstrated in [20]. In
order to simulate WSN, we use MiXiM [21], which is an OMNeT++ modelling framework
created for mobile and fixed wireless networks. We use OMNeT++ to calculate transmission
energy consumption and to model realistic behaviour of nodes in an environment close to reality.

4.3. Datasets
To make our results more realistic, we use real-world datasets from various application domains.
Our datasets include carbon dioxide monthly measurements in ppm above Mauna Loa (CO2),
monthly mean water levels in meters in the lake of the wood at Warroad (Water), the radioactivity
in the ground at one minute intervals over one day (Radio) [22], temperature measurements in a
garden (Temp) [23], average humidity at Limoges (Hum), sea low level pressure (Pressure) [24].
Table 2 shows the sizes in bytes of the used datasets.
20

Table 2. The different used datasets and their sizes in bytes.
CO2
3070

Dataset
Size (bytes)

Hum
264

Radio
49450

Temp
10192

Pressure
2405

Water
5395

4.4. Simulation Scenario
Our OMNeT++ simulation model consists of two sensor nodes that are 100 meters apart. One of
these nodes sends the data in compressed form while the other one only receives them and replies
with MAC acknowledgements. We assume that the data is transmitted in packets having payloads
of 64 bytes. No noise or interference was added to the simulation model so no packets were
dropped. Sensor nodes are equipped with battery having a nominal voltage of 3.3 V and a
nominal capacity of 1000 mAh. Regarding the algorithms, we use S-LZW-MC32 (mini cache of
size 32) and K-RLE with K = 2.

4.5. Simulation Results and Analysis
The following subsections present the simulation results along with their respective analysis.

4.5.1. Compression Ratio
The compression ratio is a very important metric when comparing compression algorithms. In
WSNs, having higher compression ratios means lesser amount of data to be transmitted, which
means more energy savings. The compression ratio is calculated according to the following
equation:

‫− 1 = ݋݅ݐܽݎ ݊݋݅ݏݏ݁ݎ݌݉݋ܥ‬

ܿ‫݁ݖ݅ݏ .݀݁ݏݏ݁ݎ݌݉݋‬
‫݁ݖ݅ݏ .݈ܽ݊݅݃݅ݎ݋‬

Compression Ratio (%)

Compression Ratios on Various Datasets
80%
70%
60%
50%
40%
30%
20%
10%
0%
CO2

Hum

MAS

Radio

S-LZW

Temp

Pressure

Water

K-RLE

Figure 6. Compression ratios on various datasets.

The results shown in Figure 6, show that MAS beats both S-LZW and K-RLE in all the datasets.
MAS’ highest compression ratio is about 68.7% and its lowest compression ratio is 57.5%, SLZW highest compression ratio is 57.7% and its lowest compression ratio is 31.1%, K-RLE does
not perform well in compressing these data sets, its highest compression ratio is 8.3% while its
lowest is 0%. This could be justified by the fact that numbers found in datasets generated by
sensor nodes often do not contain long sequences of repeated symbols.
21


These results show a great advantage of using MAS to compress floating-point data, since it
always achieves better results than the other algorithms.

4.5.2. Computation Time and Energy
After compression ratios have been presented it is important to see how much computation cycles
does the microcontroller run in order to achieve these results. From computation cycles, we can
calculate the compression time required by the microcontroller by assuming that the
microcontroller is operating at a frequency of 1 MHz.
Figure 7 shows that in all cases K-RLE requires the least number of computation cycles. This is
justified by the fact that K-RLE compression ratios are low, thus the algorithm is not performing
all the required procedures. The results for the K-RLE algorithm are not reliable to be used for
comparison with the other algorithms since they do not reflect the actual performance of the KRLE algorithm. For the other algorithms, we notice that S-LZW requires more computation
cycles than MAS in all the datasets. In most cases, MAS requires about one-third the amount
required by S-LZW. The same justification applies to compression time and computation energy
(shown in Figure 8) since they are directly proportional to computation cycles.
Compression time is calculated using the following formula:

‫= ݁݉݅ݐ ݊݋݅ݏݏ݁ݎ݌݉݋ܥ‬

ܰ‫ݏ݈݁ܿݕܿ ݊݋݅ݐܽݐݑ݌݉݋ܿ ݂݋ ݎܾ݁݉ݑ‬
‫ݕܿ݊݁ݑݍ݁ݎܨ‬

In our experimentation, the microcontroller has in its active state a frequency of 1 MHz.
As for the computation energy, it is calculated using the following formula:

‫݈݁ܿݕܿ ݁݊݋ ݂݋ ݕ݃ݎ݁݊ܧ ݔ ݏ݈݁ܿݕܿ ݊݋݅ݐܽݐݑ݌݉݋ܥ = ݕ݃ݎ݁݊݁ ݊݋݅ݐܽݐݑ݌݉݋ܥ‬
The energy of one cycle is calculated in section 4.1, and was found to be 3.3 nJ.

16
14
12
10
8
6
4
2
0

16
14
12
10
8
6
4
2
0
CO2

Hum

Radio

MAS

Temp

S-LZW

Pressure

Compression Time (s)

Cycles (million cycle)

Computation Cycles and Compression Time

Water

K-RLE

Figure 7. Required computation cycles and compression time on ATmega1281 microcontroller.

22


Computation Energy Cost

Energy (mJ)

50
40
30
20
10
0
CO2

Hum

MAS

Radio

S-LZW

Temp

Pressure

Water

K-RLE

Figure 8. Computation energy cost.

4.5.3. Transmission Energy
Computation energy alone is not sufficient to reflect the energy efficiency of an algorithm. The
energy efficiency depends on both computation and transmission energy. Transmission energy is
the energy consumed by the sensor to send the compressed data wirelessly. Figure 9 shows the
energy required to transmit the compressed form of the datasets.

Transmission Energy Cost

Energy (mJ)

500
400
300
200
100
0
CO2

Hum

MAS

Radio

S-LZW

Temp

Pressure

Water

K-RLE

Figure 9. Transmission energy cost

Regarding transmission energy consumption, it is obvious that K-RLE consumes the most energy.
This is because K-RLE compression ratio is low, thus it is transmitting larger amounts of data
than the other algorithms. Again, MAS consumes the least amount of transmission energy and
beats the other algorithm.

4.5.4. Total Consumed Energy
The total energy consumed could better reflect the energy efficiency of the three algorithms.
Despite the fact that K-RLE consumes the least amount of computation energy, results in Figure
10 show that it consumes the most amount of total energy. This is because it sends large amounts
of data over the network. For the other two algorithms, MAS consumes the least energy and thus
proves to be a strong candidate for compression in WSN.

23


Energy (mJ)

Total Energy Cost when Using Compression
450
400
350
300
250
200
150
100
50
0
CO2

Hum

MAS

Radio

S-LZW

Temp

Pressure

Water

K-RLE

Figure 10. Total energy cost when using compression.

To be able to calculate energy savings of each algorithm, we have to calculate the energy
consumed when not using any compression algorithm. The energy consumed when not using a
compression algorithm is the energy required to send the data in uncompressed form, so it
depends greatly upon the data sizes. That is why we see, in Figure 11, that the Radioactivity
dataset is consuming the most energy since it has the largest size.

Energy (mJ)

Total Energy Cost when not Using Compression
500
450
400
350
300
250
200
150
100
50
0
CO2

Hum

Radio

Temp

Pressure

Water

Figure 11. Total energy cost when not using compression.

4.5.5. Energy Savings
Figure 12 shows the percentage of energy saved when using each of the three compression
algorithm. Energy savings is calculated according to the following formula:

ܵܽ‫− 1 = ݕ݃ݎ݁݊݁ ݀݁ݒ‬

‫ݐ݅ݓ ݕ݃ݎ݁݊ܧ‬ℎ ܿ‫݊݋݅ݏݏ݁ݎ݌݉݋‬
‫ݐ݅ݓ ݕ݃ݎ݁݊ܧ‬ℎ‫݊݋݅ݏݏ݁ݎ݌݉݋ܿ ݐݑ݋‬

24


Energy Savings
70%

Saved Energy (%)

60%
50%
40%
30%
20%
10%
0%
-10%

CO2

Hum

MAS

Radio

S-LZW

Temp

Pressure

Water

K-RLE

Figure 12. Energy saved when using compression algorithms.

It is clear that MAS achieves the most energy savings with results better than both S-LZW and KRLE. MAS’ energy savings are more than that of S-LZW by an average of 20%, and better than
that of K-RLE by an average of 48%.
In some datasets, when using the K-RLE algorithm, the saved energy is negative, it means that
the compression algorithm is not saving energy; instead, it is leading to more energy
consumption. This is due to the low compression ratios of K-RLE on some datasets.

4.5.6. Memory Requirements
To complete the evaluation we must calculate the amount of memory consumed by each
algorithm. As a reminder, our platform has a flash memory of 128 KB and a RAM of 8 KB. It is
important to note that the memory results are independent of the datasets used. This is because
these results are obtained just when building the algorithm and before running any operation or
procedure. So these values represent the amount of memory allocated by the algorithm when they
are loaded into RAM and before operating on any dataset. Figures 13 and 14 show the absolute
and relative memory consumption of each algorithm for flash memory and RAM respectively.

Memory (Bytes)

Flash Memory Consumption
4000

[VALUE], 2.8%

3000
[VALUE], 1.5%

2000

[VALUE], 0.9%

1000
0

MAS

S-LZW

K-RLE

Figure 13. Flash memory consumption.

MAS consumes the largest amount in the flash memory. In fact, MAS program code is a little
long since it has two representations for integers and real numbers. Flash memory consumption is
not important and it is not a concern, since flash memory is always of a large size, and MAS is
25


only using 2.8% of that memory. This does not introduce any problems in performance since the
flash memory is reserved for program code and not for random access.

Memory (Bytes)

RAM Consumption
4000
[VALUE], 39.6%

3000
2000
1000
[VALUE], 0.5%

0, 0.0%

0

MAS

S-LZW

K-RLE

Figure 14. RAM consumption.

In terms of RAM usage, which is the important memory concern, MAS consumes only 44 bytes
while K-RLE consumes almost zero bytes. S-LZW turns out to be the most RAM consuming
algorithm, and this is because of the dictionary it uses, S-LZW consumes 3240 bytes, which is
equivalent to about 40% of the RAM.
These results prove that MAS is a strong candidate for compression in WSNs, since it beats the
other algorithms in all the proposed metrics. It achieves compression ratios better than S-LZW by
an average of 13% and better than K-RLE by an average of 59%. MAS saves the most amount of
energy, it saves by an average of 20% more than S-LZW and 54% more than K-RLE. In terms of
memory, MAS and K-RLE use a very little amount of RAM, MAS uses only 0.5% of the total
amount of RAM, while K-RLE consumes almost zero bytes of RAM. S-LZW uses the most
amount of RAM; it consumes 3240 bytes that is equivalent to 39.6% of the total available RAM.

5. CONCLUSIONS
In this paper, we propose MAS, a new lossless floating-point data compression algorithm for
WSNs. MAS is applicable to a variety of sensor hardware and platforms due to its low memory
and processing requirements.
Simulation results show that MAS’ energy savings are on average 54% on all the tested datasets,
while maintaining the highest compression ratios. MAS surpasses the other tested compression
algorithms in terms of compression ratio, compression speed, memory requirements and energy
savings. These results, which are obtained from accurate and trustworthy simulators, present
MAS as a strong and competing candidate for data compression in WSN.

6. FUTURE WORKS
As a short-term step, we would like to improve MAS to exploit the correlation and the similarity
in the data generated by sensor nodes. Such an improvement would allow MAS to achieve higher
compression ratios. We would also like to implement a transformation that aims at reducing the
number of digits in the input. This transformation is expected to increase MAS’ compression ratio
since MAS relies mainly on the number of digits in the data to achieve compression
As a long-term step, we would like to study the efficiency of using MAS with an aggregation
technique. The main challenge here is to prove that introducing MAS to aggregated networks
26


does not lead to more energy consumption at the level of aggregators, which are supposed to
follow this cycle to achieve their job: decompression – aggregation – compression.

REFERENCES
[1]

[2]
[3]
[4]
[5]
[6]
[7]

[8]

[9]

[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]

[27]

N. Kimura and S. Latifi. “A survey on data compression in wireless sensor networks,” In
Information Technology: Coding and Computing, 2005. ITCC 2005. International Conference on,
volume 2, pages 8–13 Vol. 2, 2005.
Jennifer Yick, Biswanath Mukherjee and Dipak Ghosal, "Wireless sensor network survey”,
Department of Computer Science, University of California, Davis, CA 95616, United States, 2008.
I. Akyildiz, W. Su, Y. Sankarasubramaniam and E. Cayirci, “A survey on sensor networks,” in IEEE
Communications Magazine, Vol. 40, No. 8, August 2002, pp. 102-114.
Chipcon AS. Chipcon SmartRF CC2420 Datasheet rev. 1.3. http://www.ti.com/product/cc2420,
October. 2005.
Chipcon AS. Chipcon SmartRF CC1000 Datasheet rev. 2.3. http://www.ti.com/product/cc1000,
August. 2005.
MaxStream, Inc. XTend OEM RF Module: Product Manual v1.2.4. http://www.maxstream.net/,
October. 2005.
C. M. Sadler and M. Martonosi, “Data compression algorithms for energy-constrained devices in
delay tolerant networks,” in Proceedings of the 4th International Conference on Embedded
Networked Sensor Systems (SenSys), 2006.
Capo-chichi, E. P., Guyennet, H. and Friedt, J, “K-RLE a new data compression algorithm for
wireless sensor network,” in Proceedings of the 2009 Third International Conference on Sensor
Technologies and Applications.
Maher Assi, Alia Ghaddar, Samar Tawbi, Ali Jaber, Rami Tawil. MAS: A New Floating Point
Compression Algorithm for Wireless Sensor Networks. In Ocean & Coastal Observation: Sensors
and observing systems, numerical models & information Systems (OCOSS 2013), October (28-31),
Nice, France, in press.
Mark Nelson, “LZW revisited”, in Dr. Dobb's Journal, Volume 15 Issue 6, June 1990, Pages 126 –
127, CMP Media, Inc., USA.
Heinola, Finland, "RLE compression", Baltic Olympiad in Informatics, BOI 2006, DAY-2.
IEEE 754: Standard for Binary Floating-Point Arithmetic, http://grouper.ieee.org/groups/754.
Libelium Company, Waspmote wireless sensor platform,
http://www.libelium.com/products/waspmote.
ATmega1281, Atmel 8-bit AVR RISC-based microcontroller, ATmega640/1280/1281/2560/2561
Datasheet, revision P, 10/2012.
Atmel Corporation, http://www.atmel.com.
Texas Instruments, CC2420: Single-Chip 2.4 GHz IEEE 802.15.4 Compliant and ZigBee™ Ready
RF Transceiver, Datasheet Rev. C, http://www.ti.com/product/cc2420, 07 Mar 2013.
802.15.4e-2012 - IEEE Standard for Local and metropolitan area networks--Part 15.4: Low-Rate
Wireless Personal Area Networks, http://standards.ieee.org/about/get/802/802.15.html.
Atmel AVR Studio, version 6.1, http://www.atmel.com/tools/ATMELSTUDIO.aspx.
OMNeT++, version 4.3, http://www.omnetpp.org
Xiaodong Xian, Weiren Shi and He Huang, “Comparison of OMNeT++ and other simulator for
WSN simulation,” college of automation, Chongqing university, Chongqing, 400044, China, 2008.
MiXiM an OMNeT++ modeling framework, version 2.2.1, http://mixim.sourceforge.net
Hyndman, R.J. Time Series Data Library, http://data.is/TSDLdemo
Mischa Dohler Jialiang Lu, Fabrice Valois, « Optimized data aggregation in WSNs using adaptive
arma”. SensorComm, July 2010.
Weather Forecast and Reports, http://www.wunderground.com
Sameer Tilak, Nael B. Abu-Ghazaleh, and Wendi Heinzelman, “Taxonomy of wireless micro-sensor
network models”. Mobile Computing and Communication Review, 6, April 2002.
Subhankar Mishra, Sudhansu Mohan Satpathy and Abhipsa Mishra, “Energy efficiency in ad hoc
networks”, in International Journal of Ad hoc, Sensor & Ubiquitous Computing (IJASUC) Vol.2,
No.1, March 2011.
Stephen Wolfram. A New Kind of Science, Wolfram Media, Inc. Champaign IL 2002, United States.
IBSN: I-57955-088-8.
27

[28]

[29]
[30]

[31]

[32]
[33]

Zeeshan Ali Khan and Mustafa Shakir, “Interplay of communication and computation energy
consumption for low power sensor network design”, in International Journal of Ad hoc, Sensor &
Ubiquitous Computing (IJASUC) Vol.3, No.4, August 2012.
D. Salomon, Data Compression: The Complete Reference, Second edition, 2004.
Alia Ghaddar, Tahiry Razafindralambo, Isabelle Simplot-Ryl, David Simplot-Ryl, Samar Tawbi and
Abbas Hijazi. “Investigating Data Similarity and Estimation through Spatio-Temporal Correlation to
enhance Energy Efficiency in WSNs”. In International Journal of Ad Hoc & Sensor Wireless
Networks, Vol. 16, pp. 273-295, 2012.
K.Ramanan and E.Baburaj, “Data gathering algorithms for wireless sensor networks: a survey”, in
International Journal of Ad hoc, Sensor & Ubiquitous Computing (IJASUC) Vol.1, No.4, December
2010.
C.Y. Chong and S. Kumar, “Sensor networks: evolution, opportunities, and challenges”, in
Proceedings of the IEEE, Vol. 39, No. 8, pp. 1247-1256, 2003.
Eduardo F. Nakamura, Fabiola G. Nakamura, Carlos M. S. Figueiredo, and Antonio A. F.Loureiro.
“Using information fusion to assist data dissemination in wireless sensor networks”.
Telecommunication Systems, pages 237–254, November 2005.

AUTHORS
EL ASSI Maher is a PhD student starting from October 2013. His PhD is arranged by joint supervision
between the University of Franche Comte, Besançon – France, and the Lebanese University, Beirut –
Lebanon. He received M.S. degree in the Lebanese University in 2012. He also graduated as a Civil
Engineer in 2012 from the Lebanese University. His research interests include Internet of Things (IoT),
Wireless Sensor Networks (WSN), energy efficiency in embedded devices, network modelling and
simulation and lightweight data analysis techniques.
GHADDAR Alia has PhD in Computer Science with more than 6 years of teaching experience and 4 years
of web development. Obtained her Master degree in Computer Science from the Lebanese University. She
started her PhD at the University of Lille-1, Science and technology in France. During that time, she was
member in POPS-project team; A joint project of INRIA, University of Lille-1 and CNRS. Her interests
now lie in the Internet of things, mobile sensors, data communication and knowledge discovery in the
wireless sensor networks.
TAWBI Samar, PhD in Computer Science. She is an associate professor in the Computer Sciences
department in the Lebanese University since 2005. She received the M.S. degree in “Mathematical
Modelling and Scientific Software Engineering” by the Lebanese University and the universities of Rennes,
Reims (France) and EPFL (Switzerland). She obtained her PhD in 2004 at ‘Paul Sabatier’ University,
France. Her research interests include data aggregation and communication in wireless sensors networks,
auto configurable WSN, Internet of things and Cloud computing.
GHADDAR Fadi, MS in Network and System Security. He has more than 15 years of experience in
software analysis and development, network and system security and team leading. He obtained recently
his Master degree from Saint Joseph University. His research interests include information security, data
aggregation and security in wireless sensors networks.

28

Resource efficient floating-point data compression using mas in wsn

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Resource efficient floating-point data compression using mas in wsn (20)

Recently uploaded (20)

Resource efficient floating-point data compression using mas in wsn