Efﬁcient Power Consumption Strategies for Stationary Sensors

Connected to GSM Network

abor Paller, P

eter Sz

armes and G

abor

echenyi Istv

an University, Information Society Research & Education Group, Egyetem t

er 1. Gy

or, Hungary

Keywords:

Agriculture, Sensors, Power Efﬁciency.

Abstract:

The number of large sensor systems are rapidly growing nowadays in many ﬁelds. Well-designed Big Data

solutions are able to manage the enormous data ﬂow and create real business beneﬁts. One dynamically grow-

ing application area is precision farming. It requires robust and energy-efﬁcient sensors, because the devices

are placed outdoors, often in harsh conditions, and there is no power outlet in the middle of a corn ﬁeld.

Power efﬁciency is one of the major themes of the Internet of Things (IoT). According to the IoT vision, em-

bedded sensors send their data to processing units (either located near to the sensor or on some intermediate

gateway device or in the cloud) using heterogeneous transport networks. Some sensors employ short-range

network like Bluetooth and some gateway device like a tablet. Other sensors directly connect to wide-area

networks like cellular networks.

This paper will analyse different communication patterns accomplished over GSM network from the view-

point of the energy consumption of the sensor device with the assumption that the sensor is stationary. The

measurements were done using two different GSM modems designed for embedded systems to ensure that

the results represent a wider picture and not some implementation property of a particular GSM modem.

Recommendations are given about the strategies applications should follow in order to minimize the energy

consumption of their GSM subsystems.

1 INTRODUCTION

Internet of Things is often considered a recent trend

but the vision was presented ﬁrst in 1991 (Weiser,

1991). Weiser envisioned computers that ”disappear

into the background” and are connected with wired or

wireless links. One key element of Weiser’s ubiqui-

tous computing was the low-power nature of the com-

puting elements that are able to function for an ex-

tended period of time without recharging otherwise

battery issues would prevent the devices from ”disap-

pearing into the background”. Low-power and ultra

low-power energy consumption has been a key IoT

research theme ever since (Sundmaeker et al., 2010),

(Zorzi et al., 2010).

IoT systems employ heterogeneous networks to

connect the sensors to the data processing units. Some

solutions are based on short-range networks (e.g. Zig-

Bee, Bluetooth), the data is collected by some ”gate-

way” device (e.g. smartphone, tablet, set-top box)

which then connects to a wide-area network. Isolated

sensors that are rarely visited by humans and are far

from any other elements of the ubiquitous network

cannot adopt this solution, these sensors have to con-

nect to the wide-area network directly. The most com-

mon wide-area network with low connectivity cost

and large coverage is the public cellular network.

2 THE AgroDat PROJECT

Today sensors and sensor networks gain more and

more importance in many application areas. Ma-

chines (including cameras, sensors, satellites, imag-

ing devices, etc.) are already generating more data

than we, humans and business processes (Figure 1).

These devices often operate in a harsh environment

without access to electric networks, where robustness

and energy efﬁciency are very important characteris-

tics.

One such application ﬁeld is agriculture. Preci-

sion agriculture is an integrated agricultural manage-

ment system incorporating several technologies. This

technology can reduce the cost of producing crops

and the risk of environmental pollution (Earl et al.,

1997). The AgroDat R&D project with notable in-

Paller G., Szármes P. and Élõ G..

Efﬁcient Power Consumption Strategies for Stationary Sensors Connected to GSM Network.

DOI: 10.5220/0005262500630068

In Proceedings of the 4th International Conference on Sensor Networks (SENSORNETS-2015), pages 63-68

ISBN: 978-989-758-086-4

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

dustrial and scientiﬁc partners aims to build an agri-

cultural information system in Hungary. The system

relies on collecting and analyzing high-volume data

about crops and environmental conditions, like soil

moisture and temperature, air temperature, precipita-

tion, solar radiation, etc.

Soil sensors (see Figure 2) can measure water po-

tential, electric conductivity, volumetric water con-

tent, soil temperature etc. Electric conductivity corre-

lates with salt content, inﬂuencing plant growth. Wa-

ter potential refers to the water available for plants.

This data can be used for planning irrigation, fore-

casting plant diseases, and analyzing soil aspiration.

Light sensors (see Figure 2) can measure the inten-

sity of photosynthetically active radiation, or the spec-

trum of incoming and reﬂected light in certain bands,

which can then be used to calculate the Normalized

Difference Vegetation Index and Photochemical Re-

ﬂectance Index (Garrity et al., 2010). These indexes

correlate closely with vegetation and photosynthetic

activity respectively, and they are good indicators

of biomass and plant stress. Sensors can measure

relative humidity, air temperature or vapor pressure.

These values are linked with plant evaporation. Leaf

wetness sensors are designed to detect wetness (pres-

ence and duration) and ice formation on leaf surfaces.

The data is useful for forecasting plant diseases and

determining spraying actions.

Combining different sensors into a sensor group

creates synergies, and during the design of such a

sensor unit, low energy consumption and ability to

withstand harsh environmental conditions are impor-

tant objectives. Much of the data needed for the agri-

cultural information system can be collected by these

sensor units, which can make measurements even on

a minute-rate. Sensors are very different in terms of

their data communication requirements. The current

batch of agricultural sensors being developed by our

project have the following properties.

• These sensors are stationary. Once installed, they

move very rarely.

• Their environment changes only slowly. For ex-

ample sudden changes in ground temperature or

ground moisture are rare. This means that sensor

values can be sampled with quite long sampling

periods (multiple hours or even daily).

• The quantity of the data to be transmitted is rel-

atively small. One measured quantity is a nu-

merical value and the sensor equipment measures

about 10-20 such quantities.

• These sensors are installed on locations that are

rarely accessed and are far from the usual net-

work infrastructure endpoints. For example one

Figure 1: Sources of the data growth.

Figure 2: Decagon soil and light sensors (source: Decagon).

of our sensors are meant to be installed on large

corn ﬁelds. Long, unassisted operation is an im-

portant requirement.

These requirements have led to the following

high-level design decisions.

• The sensors will be connected using ordinary

GSM network directly, without the help of some

”gateway” node. Each sensor will be a GSM end-

point.

• Low-bandwidth data bearers like SMS or GPRS

satisfy the transfer requirements.

• Low energy consumption/long operating time

without on-site service is crucial.

• Remote manageability of the sensor is a must.

The conceptual architecture of the system can be

seen in Figure 3.

3 EVALUATED GSM MODEMS

In order to ensure that we are really evaluating the

communication alternatives, we chose to run our mea-

surements from two different GSM modem vendors.

GL865-QUAD is a variant of Telit’s extremely

popular GE865 product family

. The module has

2.5G network support which means that it can ac-

cess GSM (voice call and SMS) and GPRS network

services. The modules can be used in GSM modem

mode when the application code is executed by some

GL865-QUAD, http://www.telit.com/products/product-

service-selector/product-service-selector/show/product/

gl865-quad/

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

Figure 3: Conceptual architecture.

external CPU (e.g. a microcontroller) but a stan-

dalone mode is also available when the application

code is executed by the on-chip Python interpreter.

SIMCOM’s SIM900 module

was selected to cross-

check the power consumption measurement results of

certain communication scenarios on a different GSM

modem implementation. It is a more traditional unit

in the sense that SIM900 needs an external CPU to

execute the application logic.

Power consumption measurements were accom-

plished with about 3 Hz sampling the ﬁlter capacitors

on the power lines ﬁlter out higher frequencies. The

samples were further analysed using the R/R Studio

mathematical suite

4 COMMUNICATION

SCENARIOS

4.1 Network Registration

This is seemingly the simplest scenario but it comes

with the most complications. Registering on the net-

work and staying registered involves network regis-

tration and location update procedures but more im-

portantly, it requires that the GSM module is active

and listens to network events. As we assume station-

ary operation, procedures relevant to mobility man-

agement e.g. cell handover do not occur but in order

to stay registered on the network, periodic location

update has to be executed. Figure 4 shows the power

consumption of the Telit GL865 executing this sce-

nario. The two spikes of power consumption are re-

lated to the network registration and location update

procedures. Location update occurs on the network

SIM900 Speciﬁcation, http://wm.sim.com/

downloaden.aspx?id=2972

http://www.rstudio.com/

Figure 4: Telit GL865 power consumption (initial registra-

tion and location update).

used during the measurements (Telenor Hungary) in

about every 55 minutes. This is a conﬁguration value

chosen by the network operator and can be expected

to be between 30 minutes and 2 hours. Typically this

value is constant for the same network. It is more

important to note, however, that the idle power con-

sumption of the module is about 7mA. This means

that while the actual network procedures consume

720 mAs ( milliamper-second) for the network reg-

istration and 400 mAs for one location update (with

this network, there are about 26 location updates per

day which means about 10400 mAs or 2.89 mAh cost

for location updates), keeping the module operational

costs about 170 mAh for a day. Note that these val-

ues are relatively unaffected by the received signal

strength. The measurements were done with RSSI=5,

RSSI=4 and RSSI=2 signal strengths and the results

were very similar. The reason of this similarity is that

actual network transmission is very short in these sce-

narios.

The results are very similar with the SIM900 mod-

ule. Short power consumption spikes related to the

network registration and location update procedures

can be observed but it is more important to note the

idle current consumption of the module which is close

to 20 mA. While the network registration procedure

costs only 834 mAs and 26 location updates cost 2.71

mAh, keeping the module operational costs 456 mAh

for a day.

Both modules offer custom power saving modes.

The idea behind these modes is that only the func-

tional units executing GSM procedures remain opera-

tional, units communicating with the application CPU

(and in case of the Telit module, units executing the

application logic) are switched off. With these power

saving modes, the idle consumption of the devices de-

creases quite dramatically. For both the GL865 and

the SIM900, the idle power consumption falls below 1

mA. Speciﬁcally, for the GL865 the power consump-

tion needed to keep the module operational for a day

is about 11 mAh while network procedures cost ad-

ditional 2.9 mAh, resulting a total of 14 mAh for a

day. For the SIM900, the idle power consumption for

EfficientPowerConsumptionStrategiesforStationarySensorsConnectedtoGSMNetwork

a day is about 23.3 mAh and network procedures add

2.71 mAh, resulting a total of about 26 mAh for a day.

These measurements show the importance of

implementation-speciﬁc power-save modes and high-

light the fact that the Telit GL865 is about twice more

efﬁcient than the SIM900 when it comes to low-power

operation. It is a much more important observation,

however, that even with power-save modes active, the

continuous operation of the module has by far the

highest cost. For the Telit GL865, only 20% of the

power budget is spent on actual network procedures,

the remaining 80% is the cost of keeping the module

operational. The difference is more dramatic for the

less power-efﬁcient SIM900: only 10% of the daily

power budget is spent on network operations, the re-

maining 90% is needed to simply keep the module

active.

4.2 Data Communication

So far only the cost of being registered on the net-

work was calculated. Data communication comes

with additional costs. Our sensors send relatively

small amount of data (10-20 numerical values) rel-

atively rarely (1-2 times a day) so network bearers

with lower bandwidth were analysed. A wide vari-

ety of data encodings have been proposed for IoT ap-

plications but the area is far from settled. XML-based

formats (Su et al., 2014) and publish-subscribe frame-

works are being proposed for IoT (Hakiri et al., 2013).

Our intention was to keep the amount of data

transmitted, the power needed for data transmission

and the CPU cycles needed to encode/decode pack-

ets low so we adopted a size-efﬁcient data encoding

based on ASN.1 and Basic Encoding Rules (BER)

(Kaliski and Jr., 1993). These BER data structures

were then sent to the server using HTTP implemented

on top of the modules’ native TCP support.

The power consumption was measured with in-

creasing amount of data items (16 bit values) using

the BER encoding mentioned earlier. With regards to

PDP context handling, two different approaches were

implemented. The ﬁrst approach activates the PDP

context, sends the packet then deactivates the context.

This is closer to our data communication scenarios

when we send data packets only rarely. In order to

evaluate the cost of PDP context activation, the sec-

ond approach activates the PDP context once, sends

all the test packets then deactivates the context after

all the packets are sent. Table 1 shows the results for

the ﬁrst approach while Table 2 shows the results for

the second approach using the GL865 module. It can

be observed that PDP context activation adds a con-

stant but not too signiﬁcant power cost to the commu-

Table 1: Power consumption of data communication, PDP

context activated/deactivated for each packet.

Data items Packet size

(bytes)

Power con-

sumption

(mAs)

16 287 2370

32 511 2595

64 1981 2945

128 4157 3307

256 8509 3951

Table 2: Power consumption of data communication, PDP

context activated only once.

Data items Packet size

(bytes)

Power con-

sumption

(mAs)

16 287 1987

32 511 2180

64 1981 2590

128 4157 3270

256 8509 3570

nication scenario.

Conclusion is that data size/data format optimiza-

tion does matter when trying to lower power con-

sumption. To signiﬁcantly increase power consump-

tion, however, data sizes must be several times larger

than the baseline data size. Optimization of data sizes

may be more relevant for ensuring data transfer in

case the radio path between the base station and the

sensor is not very optimal.

4.3 SMS Bearer

Data may also be sent using short messsages, pop-

ularly called SMS. Binary SMS is often ﬁltered by

operators so we employed Base64 encoding and sent

the ASN.1 BER content in textual format. Figure 5

shows the power consumption using the SMS bearer

with a 112 byte long data packet (which is actually

154 character long after Base64 encoding) and Fig-

ure 6 shows the sending of the same packet using

GPRS. Intuitively, it seems that SMS requires much

less power and it is indeed the case: GPRS requires

2347 mAs power while SMS needs only 247 mAs

power. The large difference is caused by the fact that

SMS uses signalling radio channels that are already

allocated when the module registered with the net-

work while GPRS has to allocate (and deallocate) ad-

ditional radio channel for the data transfer. SMS is

therefore attractive due to its much lower power con-

sumption requirement but quite frequently the pricing

of the subscription prevents using SMS extensively

for data transfer.

SENSORNETS2015-4thInternationalConferenceonSensorNetworks

Figure 5: Power consumption of the data transfer with SMS.

Figure 6: Power consumption of the data transfer with

GPRS.

4.4 Push Bearer

One strong requirement for our remotely placed sen-

sors is manageability because physically accessing

the sensors’ location is not always feasible. Manage-

ment operations are usually initated by the manage-

ment server operator asynchronously, independently

of the sensor’s scheduled operations. This requires a

push bearer that can be used to instruct the sensor to

contact the management server.

If the sensor is not registered to the network, such

an operation is impossible. The management server

operator may have to wait for the sensor to contact the

server when the sensor sends in its scheduled batch of

data and may send its management commands in the

context of the sensor data sending session. In case

of doubt (e.g. when an accident damaging the sensor

is suspected), the sensor’s health cannot be veriﬁed

immediately which may prevent timely maintenance

operations. Management requirements create a strong

incentive to register the sensor to the mobile network

continuously.

If the GSM module is registered continuously,

short message service (SMS) may be used to send

alert to the sensor to connect to the management

server for management operations. As we have seen,

SMS is very power-efﬁcient and management opera-

tions are infrequent enough so that SMS pricing is not

so much of an issue. Another option is to simulate the

push bearer using TCP.

TCP-based push bearer simulation relies on the

sensor to maintain a TCP connection to the manage-

ment server. When the server wants to send a man-

agement packet, it may use the duplex nature of TCP

streams to send the packet to the sensor. Timeout

issues, however, make this solution tricky to imple-

ment. TCP timeouts on the sensor and on the server-

side can be controlled by the implementation but mo-

bile and backbone networks between the mobile net-

work gateways to the servers often employ Network

Address Translators (NATs) that remove IP address

associations for TCP streams that look idle. The

problem was demonstrated with a test program im-

plemented on both the GL865 and SIM900 modules

and a test server application deployed on a cloud-

based Windows Server. The GSM modules attached

to the mobile network (Telenor Hungary), opened a

TCP connection to the server and left the connection

idle. After a timeout expired, a packet was sent from

the server to the GSM module. It was found that the

maximum safe timeout period was 2 hours which is

consistent with the recommendations in (Guha et al.,

2008). Longer timeout resulted in the server and the

GSM module to be silently disconnected by some

NAT on the network without either of the commu-

nicating parties being aware of the disconnection.

The results were consistent with both GSM modules,

demonstrating that this behaviour is the property of

the network between the GSM module and the cloud-

based server. Without deeper investigation of the full

network topology, it is hard to say where the NAT was

located that terminated the connection.

Reliable implementation of the TCP-based push

bearer must use a heuristic algorithm (Price and Tino,

2010), (Haverinen et al., 2007) to estimate the time-

out between the GSM module and the server by send-

ing test packets with different timeouts. The heuris-

tic algorithm must also be prepared for the fact that

this timeout may also change dynamically, due to

changes in the network topology. Once that timeout is

known, a keepalive packet must be sent in any direc-

tion over the TCP stream to prevent any NAT that may

be present between the GSM module and the server

to terminate the connection. This keepalive operation

has a power consumption cost.

Both modules are able to wake up from power-

save mode when an incoming data packet arrives on a

TCP connection that has been opened previously. For

the GL865, the reception of one such packet costs 942

mAs. Using 2 hour timeout (hence 12 such packets

per day), the daily power consumption cost is about

3.1 mAh. The SIM900 performs better in this test,

the cost of one keepalive packet was 570 mAs which

means 1.9 mAh for a day. This means that the power

consumption cost of maintaining one TCP connection

is comparable to the cost of the location update oper-

EfficientPowerConsumptionStrategiesforStationarySensorsConnectedtoGSMNetwork

ations that keep the module registered on the mobile

network. For the Telit GL865, such keepalive proce-

dure increases the daily power consumption by 22%.

For the SIM900, the increase is only 7% due to the

higher baseline power consumption of the module and

the better TCP packet reception power cost. It must

be noted that the GL865 also executed the application

logic for this test but the SIM900 acted only as a mo-

dem. TCP-based push bearer comes with other prob-

lems on the server-side like keeping a large amount

of TCP connections open at the same time but these

issues are not discussed in this paper.

5 CONCLUSIONS

Directly connecting a remotely located, battery-

powered sensor to the GSM network comes with a set

of compromises. In our case, the power consumption

and manageability requirements were in direct con-

ﬂict with each other. From the power consumption

point of view, the best solution would be to attach

the sensor to the mobile network only for the dura-

tion of sending the scheduled measurement data pack-

age. This would also decrease the load on the mobile

network infrastructure in case of a large number of

sensors. This approach would make the sensors more

complicated to manage, however. In order to send a

management operation, the management server oper-

ator should wait until the sensor connects back to the

server for the scheduled data sending operation.

The compromise may be the power-saving mode

of the GSM modules. Both GSM modules we eval-

uated have such mode even though these features are

non-standard and are speciﬁc to the particular GSM

module. A daily consumption of 15-30 mAh means

80-160 days of operation with a low-cost 2400 mAh

battery pack. As special, high capacity batteries are

now commercially available, this operational time

may be increased dramatically.

Push bearer is required for asynchronous manage-

ment operations. SMS offers an attractive alternative.

TCP-based push bearer is possible to implement with

relatively minor increase of the power consumption

but is problematic to make reliable due to NAT issues

and limitations of the number of the TCP streams on

the server side.

REFERENCES

Earl, R., Wheeler, P. N., Blackmore, B. S., and Godwin,

R. J. (1997). Precision farming the management of

variability. In Landwards. Winter, 1997, pp. 18-23.

Garrity, R. S., Vierling, L. A., and Bickford, K. (2010). A

simple ﬁltered photodiode instrument for continuous

measurement of narrowband ndvi and pri over vege-

tated canopies. In Agricultural and Forest Meteorol-

ogy, Vol. 150, Issue 3, 15 March 2010, pp. 489-496.

Guha, S., Biswas, K., Ford, B., Sivakumar, S., and

Srisuresh, P. (2008). Nat behavioral requirements for

tcp. In RFC 5382.

Hakiri, A., Berthou, P., Gokhale, A., Schmidt, D., and

Gayraud, T. (2013). Supporting end-to-end scalability

and real-time event dissemination in the omg data dis-

tribution service over wide area networks. In Elsevier

Journal of Systems Software (JSS), 86(10):25742593,

Oct. 2013.

Haverinen, H., Siren, J., and Eronen, P. (2007). En-

ergy consumption of always-on applications in wcdma

networks. In Vehicular Technology Conference,

2007. VTC2007-Spring. IEEE 65th, 22-25 April 2007,

Dublin.

Kaliski, B. S. and Jr. (1993). A layman’s guide to a subset

of asn.1, ber, and der.

Price, R. and Tino, P. (2010). Adapting to nat timeout val-

ues in p2p overlay networks. In 2010 IEEE Interna-

tional Symposium on Parallel & Distributed Process-

ing, Workshops and Phd Forum (IPDPSW), Atlanta,

GA.

Su, X., Riekki, J., Nurminen, J. K., Nieminen, J., and

Koskimies, M. (2014). Adding semantics to internet

of things. In Concurrency Computat.: Pract. Exper..

doi: 10.1002/cpe.3203.

Sundmaeker, H., Guillemin, P., Friess, P., and Woelfﬂe, S.

(2010). Vision and challenges for realising the internet

of things. In Cluster of European Research Projects

on the Internet of Things-CERP IoT, 2010.

Weiser, M. (1991). The computer for the 21st century. In

Scientiﬁc American 265(3):66–75 (January 1991).

Zorzi, M., Gluhak, A., Lange, S., and Bassi, A. (2010).

From today’s intranet of things to a future internet of

things: a wireless- and mobility-related view. In Wire-

less Communications, IEEE (Volume:17 , Issue: 6 ),

December 2010.

SENSORNETS2015-4thInternationalConferenceonSensorNetworks