CSMA

The PLR results for the CSMA system are in Figure 5.9. As can be observed, the PLR linearly increases with the transmitted power, with the exception ofPt=−20dBm. This can be explained by the hidden node effect. The hidden node effect results in either of two situations: 1) a sensor wrongly senses the channel idle and transmits, causing a collision on another sensor that it did not detect or 2) a sensor detects the channel is busy and refrains from transmitting when it would not have caused any collision.

These two situations are called a false negative and false positive, respectively. A sensor can detect transmitters that are within its range, r(i.e. the received power is larger than the sensitivity,P_det), and can suffer collisions from transmitters that are within its minimum interfering distance⁴, dImin. For the case of P_t = −20 dBm, the range is approximatelyr = 27.0 m, while d_Imin = 52.6 m for the

4Note that the minimum interfering distance corresponds to the situation of only one active interferer. If more than one interferer are active, the interfering powers add, so the minimum interfering distance would be different.

Figure 5.8: PLR of a DC system withDC= 0.1%as a function of the number of gateways.

worst case scenario (the sensor is as far as possible from the gateway). The range is significantly smaller than the interfering distance, creating the conditions for false negatives to occur – there are potential interferers within collision range that the sensor cannot detect. This effect can be seen in Figure 5.10, where the collision rate for this case is much higher. For the case of Pt = −10 dBm, however, the distances become closer in value: r = 54.2 m (andd_Imin = 52.6m as previously mentioned, since it does not depend on the transmitted power). In this case, most nodes would be able to successfully detect all the interferers within collision range. If we consider higher values of P_t, on the other hand, false positives become a problem. The range is much larger than the minimum interference distance (e.g., for P_t = 0dBm, r = 109.0m), which means that sensors can detect far-away transmitters that they would not collide with, thus wrongly refraining from transmitting. This can be observed in Figure 5.11, where the transmission rate is shown as a function ofPt. The transmission rate is defined as the average number of transmissions (successful or not) per cycle. The higher the power, the lower the transmission rate, which indicates the presence of false positives. Furthermore, the effect of the false negatives for P_t=−20dBm is also clear, since collisions increase the number of retransmissions. In conclusion, the transmitted power that provides the best PLR results is the one that corresponds to the case where the minimum interference distance,d_Imin, approaches the range,r.

For the rest of the simulations in the present section, the transmitting power ofPt=−10dBm is used, since it represents the case with the best PLR, among the ones that were simulated.

Figure 5.9: PLR of a CSMA system as a function of the transmitted power.

Figure 5.10: Packet collision rate of a CSMA system as a function of the transmitted power.

Figure 5.11: Transmission rate of a CSMA system as a function of the transmitted power.

5.3.3 DC and CSMA Coexistence

This section analyses the effect of the coexistence of two different systems – CSMA and DC – on their correspondent performances. This study considers a transmitted power ofP_t = −10 dBm, since it is (approximately) the one that shows the best performance in terms of PLR, as can be concluded from Figure 5.9. The remaining CSMA parameters are the same as the ones applied for the CSMA analysis in Section 5.3.2. For the DC system, the samePtwas used and aDC value of0.1%was applied. The packet size,T_p = 15ms, is assumed to be the same for both systems.

In Figure 5.12, the PLR of the CSMA system can be observed as a function of the number of interfer- ing DC sensors, N_DC. It can be concluded that the presence of the DC system does not significantly impair the CSMA PLR. Furthermore, the maximum number of CSMA sensors in the network,Nmax, corresponding to the maximum admitted PLR (1%), almost does not vary, as can be seen in Figure 5.13.

Although the PLR is not significantly higher, the effect of the interference of the DC can be observed in Figures 5.14 and 5.15: both the collision rates and the ACK loss rate, respectively, are significantly affected by the interference of the DC system, for lower values of N_{CSM A}. Therefore, it can be con- cluded that the presence of a DC system, with DC = 0.1%, does not significantly impair the CSMA PLR, although it causes an increase in the number of retransmissions due to collisions and loss of ACK packets, which proportionally increases the energy spent by the sensors. In the case of systems that are highly energy-constrained, this effect could mean that the two systems cannot coexist.

On the other hand, the DC system suffers considerable degradation due to the presence of the CSMA system. In Figure 5.16, the PLR of the DC system increases significantly with each added CSMA sensor.

Figure 5.12: PLR of the CSMA system as a function of the number of CSMA and DC sensors.

Figure 5.13: Maximum number of CSMA sensors (corresponding to a maximum PLR of 1%) as a function of the number of DC sensors.

Figure 5.14: Packet collision rate of the CSMA system as a function of the number of CSMA and DC sensors.

Figure 5.15: ACK loss rate of the CSMA system as a function of the number of CSMA and DC sensors.

Figure 5.16: PLR of the DC system as a function of the number of CSMA and DC sensors.

This can be explained by the fact that the CSMA system has aDCvalue that is ten times higher, which means that, in practice, adding one CSMA sensor creates up to as much traffic as 10 DC sensors, and thus having a considerate impact. In Figure 5.17, the maximum number of DC sensors (for a maximum PLR of 1%) as a function of CSMA sensors is represented. As it can be observed, the maximum number of DC sensors reaches zero for only 7 CSMA sensors, which means that the DC system cannot coexist with a CSMA system with 7 or more sensors. In practice – and since CSMA networks can work with far more sensors than this – it can be concluded that DC systems have very low coexistence capabilities.

Finally, the conclusion to be taken from the present analysis is that a CSMA system can mitigate the interference caused by a DC system by increasing the number of retransmissions and, as a consequence,

Figure 5.17: Maximum number of DC sensors (for a maximum PLR of 1%) as a function of the number of CSMA sensors.

the energy consumed. A DC system, on the other hand, suffers such a degradation from the CSMA interference that it makes the coexistence not feasible.

5.3.4 SRD and LTE Coexistence

The purpose of the present study is to determine, in general terms, if the interference of an LTE sys- tem can significantly impair the SRD system, and vice-versa. Appendix B contains all the auxiliary calculations needed to compute the transmitted power by the LTE system on the SRD frequency in study.

As proven by (5.5), the closer the sensor and gateway are, the closer that an interferer can be without caus- ing collisions – more specifically, this equation determines the minimum distance between the interferer and the victim (sensor/gateway) for collision-free communication as a function of the sensor-gateway distance.

This analysis, on the other hand, was carried out by determining the maximum distance between the sen- sor and the gateway,d_max, as a function of the distance between the UE and the victim (sensor/gateway), d_{U E}, in order for the UE not to cause collisions (by itself) in the information received at this node. This means that, for a UE interferer with a known position, a circle of radiusd_maxcan be drawn around each gateway – and every sensor within this area (with d < dmax) can send its information to the gateway without collisions directly caused by the UE; if, on the other hand, the sensor is outside the area (with d < dmax), the corresponding gateway is vulnerable to the effects of the UE interference. This circular area around the gateways is called an inclusion zone, while the area outside the circle is the exclusion zone.

Similarly, d_max and the correspondent inclusion zone can be determined for sensors. In this case, the gateway needs to be inside the sensor’s inclusion zone in order for the ACKs to be correctly received at the sensor without being affected by the UE’s interference.

In short, an inclusion area can be defined around either the gateway or the sensor. The sensor-gateway pair can only be completely immune to the UE’s interference if they are both inside one another’s inclu- sion areas, meaning that neither the sensor packets or the ACKs suffer from the UE’s interference. The maximum sensor-gateway distance can indicate, in general terms, if an SRD network could coexist with LTE UEs: ifd_maxis too small, it means that coexistence is not feasible, and vice-versa.

In this analysis, only the UE interference at the gateways is considered, since it is this situation that directly affects the SRD system’s PLR (UEs could also cause collision on ACKs received by the sensors, although that would only increase the energy spent for additional retransmissions, since the information was effectively received at the gateway – it was only the ACK that was lost).

Figure 5.18: Inclusion area around gateways for one UE interferer.

Similarly to (5.5) (and also considering that Ilim ≈ Pr −SIN Rmin), the dmax expression can be computed from (4.7) and (4.10), resulting in:

dmax =dU E10^Pt

−PtUE−SIN Rmin

10a , (5.6)

wheredU Eis the distance between the UE and the gateway andPtU E =−20dBm is the power transmit- ted by the UE in the SRD frequency band in use (as computed in Annex B). Since the UE is considered to be inside the building and in the same floor as the SRDs, the floor penetration factor is zero:Lf = 0.

As previously mentioned, only the inclusion zone around the gateways is considered in the present anal- ysis, since it is the interference at the gateway that directly impacts the SRD system’s PLR. Moreover, the purpose of the analysis is to evaluate the general feasibility of the coexistence without going into too much detail.

Let us consider the UE is inside the building and at its center. The inclusion and exclusion zones for the gateways within the building are depicted in Figure 5.18, for the lowest transmission power ofP_t=

−20dBm. As it can be observed, there is an exclusion area of significant size, meaning that a UE could hardly work inside the building without impairing the SRD system. Increasing the SRD system’s power mitigates this problem. By increasingPt, the inclusion area expands – and it can expand so as to include the UE itself, which means in practice that the UE power P_{tU E} is not enough to cause interference on the gateway-sensor pair. Having said this, the limit for the inclusion of the UE in the inclusion area corresponds to d_{U E} = d_max, which, from (5.6), results in P_t ≈ −13 dBm. This means that for a transmitted power above−13dBm, the UE cannot cause cause collisions on the SRD system. Since this power value is possible for SRDs, it can be concluded that SRDs can coexist with at least one LTE UE.

Finally, when considering the possible interference of SRDs on LTE, it can be easily concluded that the impact would be insignificant: considering, similarly to LTE’s case, that the frequency spill from SRDs into the UE frequency band is around -30 dB below the transmitted power, as determined in Annex B, a maximum interfering power density of -18 dBm/MHz⁵would be present in the UE upper channel. This value is not significant when compared to the UE’s maximum power density (23 dBm/MHz).

5This value results from a -30 dB attenuation of the highest transmitted power of 5 dBm, which corresponds to a 200 kHz channel.

Chapter 6

Conclusion

The objective of the present dissertation was to study two basic channel access schemes for SRDs – DC and CSMA – in terms of their performance and coexistence. Additionally, the coexistence of SRDs with LTE in the UHF band was analysed. A program was developed in order to simulate a sensor network communicating using DC and/or CSMA.

The results show that the performance of a DC system in terms of PLR is independent of the transmitted power, but it is influenced by how close sensors are to gateways – hence a higher density of gateways would mean an increase in system performance. CSMA’s PLR, on the other hand, is influenced by the transmitted power, due to the hidden node effect: a low transmission power implies a smaller detection range, and therefore the sensor’s transmissions suffer collisions from other sensors that it could not detect (i.e. a false negative occurs); on the other hand, a high transmission power results in a large detection range, which in turn means that the sensor repeatedly detects ongoing transmissions and refrains from transmitting, when in fact they would not have caused a collision (a false positive takes place). Interferers need to be at a minimum distance from other receivers to cause collisions, and the best PLR-performing transmission power is the one that results in a detection range that is close in value to this minimum distance for collision; in these conditions, the sensor can detect all on-going transmissions with which it could collide – and no more than those.

The main conclusion concerning the coexistence of DC and CSMA is that while DC has almost no effect on the CSMA system (since it can mitigate its effects by retransmitting lost packets), the opposite is not true, since DC has no way of defending itself against CSMA’s contention for the channel. DC is a less polite mechanism, in the way it transmits regardless of the presence of other transmitters on the channel, but it is also considerably more vulnerable to collisions. The significant difference in terms of performance makes CSMA a clear candidate for improving the efficiency of utilization of radio resources

in SRD frequency bands. The fact that it is effectively more robust in the presence of other devices competing for the channel means that CSMA would be a much more effective choice than DC, which is specially relevant in places where a high density of sensors is expected (such as cities or homes, for instance). DC has the advantage of a higher energy efficiency, increasing the network’s lifetime. In places with an expected low density of SRDs (like rural areas, where sensors could be used for environmental monitoring for agriculture), DC is the most adequate channel access scheme.

As far as the coexistence of SRDs with LTE goes, the main finding is that LTE could in fact degrade SRD performance for lower transmission powers. Increasing the power also increases the SINR, which can completely compensate for LTE’s interference, for SRD’s transmitted powers higher than a certain value.

Nevertheless, the effects of LTE on SRDs are a theme to be considered, if the minimum performance of the latter is to be protected. On the other hand, the frequency spill of SRD’s emissions onto LTE’s closest channel is not significant, when compared to LTE’s transmission power.

Bibliography

[1] ECC. Report 181: Improving Spectrum Efficiency in the SRD Bands. Technical report, 2012.

[2] Abdelmalik Bachir, Mischa Dohler, Thomas Watteyne, and Kin K Leung. MAC essentials for wireless sensor networks.IEEE Communications Surveys & Tutorials, 12(2):222–248, 2010.

[3] IMST GmbH. Channel Access Rules for SRDs. Technical report, 2012.

[4] ECC. ECC Report 207: Adjacent Band Co-existence of SRDs in the Band 863-870 MHz in Light of the LTE Usage Below 862 MHz. Technical report, ECC, 2014.

[5] David J. Wetherall Andrew S. Tanenbaum. Computer Networks. Prentice-Hall, 5th edition, 1994.

[6] Leonard Kleinrock and Fouad Tobagi. Packet switching in radio channels: Part I–carrier sense multiple-access modes and their throughput-delay characteristics. IEEE transactions on Commu- nications, 23(12):1400–1416, 1975.

[7] ECC. CEPT Report 014: Report from CEPT to the European Commission in response to the mandate to develop a strategy to improve the effectiveness and flexibility of spectrum availability for Short Range Devices (SRDs). Technical report, 2006.

[8] European Commission. Radio Spectrum Policy Group Report on Strategic Sectoral Spectrum Needs. Technical report, 2013.

[9] Ericsson. Cellular networks for massive IoT. Technical report, 2016.

[10] ECC ETSI. The European Regulatory Environment for Radio Equipment and Spectrum: an Intro- duction. Technical report, 2016.

[11] ANACOM. Quadro Nacional de Atribuic¸˜ao de Frequˆencias. https://www.anacom.pt/

eqnaf/content/freqPortalSearch.do. Accessed: 2017-02-01.

[12] ANACOM. Plano Estrat´egico do Espetro. Technical report, 2016.

[13] ECC. ERC Recommendation 70-03. Technical report, ECC, 2017.

[14] ECC. Report 189: Future Spectrum Demand for Short Range Devices in the UHF Frequency Bands. Technical report, 2011.

[15] ECC. ERC Report 182: Survey about the use of the frequency band 863-870 MHz. Technical report, 2012.

[16] OFCOM. 900 MHz SRD Spectrum: 870-876 MHz and 915-921 MHz, New SRD Bands for Europe . https://docbox.etsi.org/Workshop/2015/201503_

LPRA_RadioSolutions2015/RS2015_SESSION03/900_MGZ_SRD_Spectrum_

DONOGHUE_Robin_OFCOM.pdf. Accessed: 2017-02-01.

[17] ETSI. ETSI TR 102 886: Spectrum Requirements for Smart Metering . Technical report, 2011.

[18] ETSI. TR 103 055: Spectrum Requirements for Short Range Device, Metropolitan Mesh Machine Networks (M3N) and Smart Metering (SM) applications . Technical report, 2011.

[19] Introduction to Monte Carlo Methods. D. J. C. Mackay. https://www2.stat.duke.edu/

courses/Spring12/sta270/lec/mcmcnotes.pdf, 2012. Accessed: 2016-09-01.

[20] UAH College of Science. The Central Limit Theorem. http://www.math.uah.edu/stat/

sample/CLT.html. Accessed: 2016-09-01.

[21] Michael D Byrne. How many times should a stochastic model be run? An approach based on confidence intervals. InProceedings of the 12th International conference on cognitive modeling, Ottawa, 2013.

[22] Martocci, Jerry and Vermeylen, Wouter and Riou, Nicolas and Mil, Pieter De. Building automation routing requirements in low power and lossy networks. Technical report, IETF, 2010.

[23] International Electrotechnical Commision (IEC). Internet of Things: Wireless Sensor Networks.

Technical report, 2014.

[24] ITU. Recommendation ITU-R P.1238-8 (07/2015). Technical report, 2015.

[25] Ian Poole. What is noise figure. http://www.radio-electronics.com/info/

rf-technology-design/rf-noise-sensitivity/noise-figure-factor.

php. Accessed: 2017-04-12.

[26] Crossbow Technology. Mica2 Datasheet. https://www.eol.ucar.edu/isf/

facilities/isa/internal/CrossBow/DataSheets/mica2.pdf. Accessed: 2016- 09-01.

[27] Fouad Tobagi and Leonard Kleinrock. Packet switching in radio channels: Part II–the hidden terminal problem in carrier sense multiple-access and the busy-tone solution. IEEE Transactions on communications, 23(12):1417–1433, 1975.

[28] ETSI. TS 136 101: LTE; Evolved Universal Terrestrial Radio Access (E-UTRA); User Equipment (UE) radio transmission and reception. Technical report, 2016.

[29] Anacom. Operadores em atividade - Direitos de utilizac¸˜ao de frequˆencias atribu´ıdos. http:

//www.anacom.pt/render.jsp?categoryId=360615#.WHDkvFz4Lt0, 2014. Ac- cessed: 2017-01-05.

Appendix A

Detailed Flowcharts of

Interference-related Functions

The present Annex details the functions from Figure 4.12, including the respective complete flowcharts.

The general approach towards managing interference-related values comprises two main parts: 1) updat- ing theinterferenceSumvalue and 2) checking the effects of the previous changes.

The algorithm in the Add Interference function (in Figure A.1 and Figure A.2), for the three blocks is explained as follows:

• Interference is first added to all active receivers; afterwards, all active transmitters’ situation is evaluated – if their correspondent receiver registers an interference above the threshold, they are

“marked” as having collided;

• On the second block, a similar process takes place, where the “target” is now the current transmitter- receiver pair – interference from all active transmitters is added and collision is checked;

• Finally, in the third block interference is added to each listening node, and the signal detection is evaluated – the node only registers the signal detection if 1) the signal power is above its sensitivity and 2) it has not yet started detecting any activity (if otherwise activity has been already detected, no changes are made).

Upon the end of a packet transmission the Subtract Interference function (shown in Figure A.3 and Figure A.4) is executed, where:

• Firstly, interference is removed from other currently active receivers; note that in this case there is no “checking the consequences” phase, since there are none (only whenaddinginterference to ongoing transmissions might there be collisions);