Inter national J our nal of Electrical and Computer Engineering (IJECE) V ol. 9, No. 4, August 2019, pp. 2281 2295 ISSN: 2088-8708, DOI: 10.11591/ijece.v9i4.pp2281-2295 r 2281 Modified timed efficient str eam loss-tolerant authentication to secur e po wer line communication Boyce Sigweni, Mmoloki Mangwala, J oseph Chuma F aculty of Engineering and T echnology , Botsw ana International Uni v ersity of Science and T echnology Department of Electrical Computer and T elecommunications Engineering, Botsw ana Article Inf o Article history: Recei v ed Sep 24, 2018 Re vised Mar 8, 2019 Accepted Mar 12, 2019 K eyw ords: Load management Po wer line communication Smart meters security TESLA ABSTRA CT This paper in v estig ates the feasibility of T imed Ef ficient Stream Loss-tolerant Authen- tication to serv e sec urity needs of Po wer Line Communication (PLC) system. PLC netw ork has been identified as the ideal choice to function as the las t mile netw ork, deli v er load management messages to smart meters. Ho we v er , there is a need to ad- dress the security concerns for load management messages deli v ered o v er po wer line communications. The ubiquitous nature of the po wer line communicat ion infrastruc- ture e xposes load management systems (LMS) deplo yed o v er it to a security risk. Or - dinarily , PLC net w ork does not emplo y security measures on which the smart meters and data concentrators can depend on. Therefore, the need to pro vide a secure mech- anism for communication of load management system messages o v er a PLC netw ork. In LMS, source authentication is of highest priority because we need to respond only to messages from an authenticated source. This is achi e v ed by in v estig ating suitable rob ust authentication protocols. In this paper we present modifications to T imed Ef fi- cient Stream Loss-tolerant Authentication for secure aut hentication to secure messages for load management o v er PLC. W e sho w that PLC may be us ed to securely and ef fec- ti v ely deli v er Load Management messages to smart meters, with minimal o v erhead. Copyright c 2019 Institute of Advanced Engineering and Science . All rights r eserved. Corresponding A uthor: Bo yce Sigweni, Botsw ana International Uni v ersity of Science and T echnology , Pri v ate Bag 16, Botsw ana. Email: sigwenib@biust.ac.bw 1. INTR ODUCTION The introduction of smart meters enables electricity suppliers to manage electricity demand ef fici ently , by implementing load management systems (LMS), thus coping with electricity demand. These LMS systems forecast the demand [1, 2, 3] therefore advising on mitig ati ng steps. This demand w ould be in terms of quantity and quality –which is still increasing by the escalation of ne w and more electronic de vices in homes as popula- tion gro ws. Prior to Smart grids, po wer suppliers could not suf ficiently e xploit the adv ances in communication and information technol o gy to impro v e the electri city grid’ s ef ficienc y , reliability , security , and qualit y of ser - vice (QoS). Smart grid addresses all these desired features by modernizing the electricity grid by incorporating of communication technologies [4, 5]. The term “smart grid” has been e xpanded from just smart meters, to more focused on adv anced metering infrastructure (AMI) [6]. Successful implementation of electrical load management system via smart meters requires a s ecure communication channel which must also be rob ust to deli v er load management commands such as load redistri- b ution, dimming of lights and switching of f of hot w ater ge ysers. While, the ef fects of transferring data at high bit rate through the mains netw ork generates acceptable radiated emission re gulated by internat ional standards. The increment in s peed for Ne w Generation PLC may cause higher le v els of emissions that could be mitig ated J ournal homepage: http://iaescor e .com/journals/inde x.php/IJECE Evaluation Warning : The document was created with Spire.PDF for Python.
2282 r ISSN: 2088-8708 through the use T ime Re v ersal (TR) technique [7]. In the load management system source authentication is of highest priority because we need to respond only to messages from an authenticated source. Pri v ac y is not a priority for load management messages because the y are broadcast to e v eryone on the netw ork to manage the load. Therefore, there is no need to mak e load management messages pri v at e through encryption, b ut there is a need to respond only to commands from authentic sources because of possible attacks, such as denial of service (DOS) [8] elaborated in title-24 [9]. F or e xample —an attack er could f alsify data thereby transmitting wrong commands to smart meters –such as “electricity demand lo w” therefore users may switch on non-essential g adgets. This could cause o v erloading that may lead to grid instability or e v en po wer outages, thus defeating the sole intended purpose of load management. The scheme we present can be used by an y application em- plo yed on PLC netw ork to authenticate mess ages b ut it is hea vi ly biased to w ards PLC based load management systems. These are systems that emplo y data concentrators and smart meter s as the tw o primary components. Figure 1 sho ws a typical po wer line communication netw ork for adv anced metering infrastructure (AMI).   Figure 1. T ypical PLC Netw ork [10] The rest of this paper is or g anised as follo ws; In section 2. we discuss PLC channel characteris tics, follo wed by its security threats, risk management methods and mitig ation techniques. (Both crypto and non- crypto). T imed Ef ficient Stream Loss-tolerant Authentication (TESLA) scheme is presented in section 3., follo wed by research methodology in section 4. Modification to TESLA scheme are outlined in section 4.2. Finally , performance analysis and results are presented in section 5. 2. B A CKGR OUND 2.1. PLC channel characteristics In PLC systems, a transmit signal propag ating from one location to another suf fers from reflections at impedance discontinuities along its path. Branching and impedance appearing at the termination points are the main source of impedance discontinuity in po wer line netw orks (PLNs) gi vi ng rise to reflections. These mechanisms are illustrated in Figure 2. Figure 2. Propag ation mechanism for PLC channels [4] Due to the propag ation mechanisms ef fecti v e in both en vironments, when a signal is emitted by a transmitter , the signal recei v ed at the recei v er consists of attenuated, delayed, and phase-shifted replicas of the transmit signal leading to time dispersion. In communications community , significance of time dispersion is quantified by a parameter call ed root-mean-squared (RMS) delay spread. RMS delay spread for both commu- nication mediums is to be discussed in a more detailed w ay in the subsequent sections. Besides time dispersion characteristic, both wireless and PLC channels are time selecti v e. Mobility (or relati v e motion between trans- mitter and recei v er from a broader perspecti v e) is the main reason behind time selecti vity of wireless channels, whereas the reason for time selecti vity in PLC channels is r elated to the v arying impedance conditions in the Int J Elec & Comp Eng, V ol. 9, No. 4, August 2019 : 2281 – 2295 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 2283 PLN especially at the termination points. T ime selecti vity is a no t her aspect that the study will focus on . F or digital communication systems, the most common figure of merit is the bit error rate (BER) which is directly related to signal-to-noise ratio (SNR). Being a function of SNR, BER can be computed by only ha ving infor - mation re g arding amplitude statistics of the recei v ed signal and the noise characteristics in the communication channel. In this respect, amplitude statistics and the noise characteristics of wireless and PLC channels are among issues that the study has focused on. Po wer line communication characteristics such as, frequenc y-distance-dependent attenuation in lo w v oltage (L V), based on e xtensi v e measurements is defined as: A ( f ; d ) = exp(( a 0 a 1 f k ) d ) (1) where: f correspond to frequenc y of the signal, d is the distance co v ered by the signal while a 0 ; a 1 and k are all cable-dependent parameters e xtracted by empirical measurements [4, 11] 2.1.1. Multipath characteristics A complete characterization of the PLC channel can be gi v en by its channel frequenc y response (CFR) as follo ws:[4, 11] H ( f ) = N X i =0 " K Y k =1 ik M Y m =1 im # A ( f ; d i ) exp( j 2 f i ) (2) gi v en that the t otal number of replicas recei v ed at the recei v er is considered to be limited to N [4, 11] . where: K and M represent the number of reflection and transmission coef ficients correspond to the reflection coef ficient along the propag ation path, is the transmission coef ficient along the propag ation path while A ( f ; di ) corresponds to the frequenc y and distance-dependent attenuation deri v ed from the ph ysical characteristics of the cable, and exp( j 2 f i ) refers to the phase of the i th component due to the time delay . Finally , it is w orth mentioning that multiplication of ’ s and ’ s in (2) is referred as the reflection f actor ( j r i j e j i ) of a particular propag ation path. Note that i , the time delay , is related to the speed of propag ation within the communication medium, po wer line cables in our consideration as follo ws: i = d i p r c 0 (3) where: r is the dielectric constant of the insulation material c 0 is the speed of light in v acuum. The time-and frequenc y-v arying beha viour of a po wer -line netw ork is the result of v ariable impedance loads connected to its terminal points. An y signal transmitted through such a netw ork is subject to time-v arying multipath f ading [12]. In addition to this basic frequenc y domain-based PLC multipath model, there are other characterization approaches, such as —A matrix-based approach for the calculation of multipath components based upon the presented model in PLC netw orks is gi v en in [12, 13]. PLC channel models that are based on treating the transmission line as a tw o-port netw ork are gi v en in [14, 15, 16]. Besides these deterministic models, some statistical PLC channel characterization ef forts re g arding attenuation, mult ipath-related parameters, and so forth, that consider the PLN as a black box without dealing with its attrib utes s uch as cable characteristics, netw ork topology , and so forth are presented in [17]. Each of these channel modeling approaches has some adv antages and disadv antages. F or instance, all attrib utes of the PLN such as the netw ork topology , cable distance-frequenc y-dependent attenuation characteristics, and termination impedance conditions must all be kno wn prior to computation if a frequenc y or transmission line theory-based approach is to be adopted. Statistical models can be emplo yed if an y information re g arding the netw ork attrib utes cannot be acquired a priori. Ho we v er , an e xtensi v e measurement campaign may be required in order to dra w statistically meaningful conclusions from the data sets obtained from v arious netw orks with dif ferent topologies. Modified timed ef ficient str eam loss-toler ant... (Boyce Sigweni) Evaluation Warning : The document was created with Spire.PDF for Python.
2284 r ISSN: 2088-8708 2.2. Security thr eats A threat to a Po wer Line communication system is an y malicious occurrence that w ould ha v e an unde- sirable ef fect on the assets and resources associated with the po wer line communication. Netw ork threats tak e adv antage of the distrib uted aspects of information transmission [18, 19] and [20]. Amoroso [21] cate gorized threats to a communication system as follo ws: (a) Denial of Service (DoS) threat: The DoS threat arises when access to the po wer line communication chan- nel is intentionally block ed as a result of malicious actions tak en by an attack er . F or e xample, someone could flood the data concentrators with junk commands —therefore pre v enting load management mes- sages to be deli v ered to smart meters. (b) Inte grity threat: The inte grity thre at in v olv es unauthorized change to information stored for e xample on a smart meter (meter reading for billing purposes) or in transit between the data concentrator and smart meter . (c) Disclosure Threat: This in v olv es the dissemination of pri v ate information. Protection of po wer line com- munication system ag ainst unintended disclosure. 2.3. Risk management methods Security risk for PLC based load management programs can be assessed using a risk management approach [9]. This is whereby assets that need protection are identified and their sensiti vity to attack analysed. There is a need to identify a possible source, strength and intent of threats, as well as enumerating vulnera- bilities and finally determining appropriate mitig ation methods. Hence, the need for anomaly detection and monitoring [22]. 2.3.1. P otential attacks Se v eral attack scenarios were considered to determine vulnerabilities, assets and threats. The follo w- ing are some of the attacks on a load management system [18]: (a) An attack er could block load reduction commands, therefore pre v enting the required reduction percentage. Therefore, resulting in forced load shedding or black outs. (b) An attack er could broadcast incorrect synchronisation time, which can cause e v ents to occur at wrong times, either earlier or later than scheduled. (c) An attack er could modify the softw are set-point for air -conditioning unit in the smart meter so that it appears to be dra wing l ess or no po wer . This action results in command for load reduction being ignored by the smart meter , therefore the unit is not switched of f nor ha v e its po wer reduced. (d) An attack er could switch ON all the appliances (heaters, air cons) controlled through the smart meter for load management, causing an une xpected and e xcessi v e load, leading to possible black outs or e v en grid instability . (e) In order to anno y the public, an attack er could switch of f the air conditioning units or set temperature thermostat to uncomfortable le v els. (f) By flooding the netw ork with multiple requests for time synchronisation, the attack er can cause Denial- of-Service. In the ne xt subsection we look at non-cryptographic and cryptographic mitig ation techniques, so that we can e xplore w ays as to ho w these potential attacks could be mitig ated in PLC load management system. 2.4. Mitigation techniques The focus tends to be cryptograph y as the primary defence ag ainst attacks when the security of infor - mation systems is in question. Due to the unique characteristics, constraints and design of the PLC based load management system, it presents an opportunity to consider se v eral non-cryptographic methods. 2.4.1. Non-cryptographic mitigation techniques The ideologies in v olv ed in non-cryptographic mitig ation techniques methods are outlined in [9] and these include: (a) Depending on pre v ention (ph ysical barrier around smart meters) as well as detection (temper alert ) mech- anisms as deterre nts. Intrusion detection system that could be emplo yed on the netw ork; consists of recei v ers placed on strate gic locations where the y w ould compare transmitted data with data the y are recei ving to identify bogus transmitters Int J Elec & Comp Eng, V ol. 9, No. 4, August 2019 : 2281 – 2295 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 2285 (b) Reducing the capability of an attack er by making the system a v ailable only at certain times or responds to e xternal commands after some random time. Therefore, if an attack er does something, it will only tak e ef fect after some time. Essentially , by that time the breach w ould ha v e been detected (as discussed in the pre vious point) and acted upon. The capability of an intruder to do damage could be further reduced by setting the safe set-point for de vices if one is changing settings remotely via commands. F or e xample one cannot set the temperature to unsafe le v els (too lo w or too high) remotely . (c) Pre v enting messages that will result in load increase to be sent remotely , that is, the system must not be able to send commands to smart meters to switch appliances on. If appliances ha v e been remotely switched of f, the customer could manually switch them on, or ha v e appliances fitted with a de vice that is set to check the smart meters’ mode. If the demand is lo w , the y could automatically s witch on. Note that this w ould be a one w ay communication as the smart meter w ould not communicate or control these de vices, the smart meter can only switch of f the appliance b ut cannot turn it on or instruct the de vice to turn the appliance on. 2.4.2. Cryptographic mitigation techniques There are man y cryptographic mitig ation techniques a v ailable to secure PLC for smart meters. These include Distrib uted Netw ork Protocol (DNP3) [23], [24], X.509 [25], RSA [26], and TESLA [27]. All these techniques ha v e dif ferent capabilities and limitations. F or e xample; digitally signing each pack et using X.509 pro vides proficient data source authentication. Unfortunately , it incurs a high o v erhead in terms of time needed to sign and v erify and also in terms of required bandwidth. Signature v erification through X.509 is compu- tationally costly . Therefore, smart meters with their modest computation capabilities w ould be o v erwhelmed trying to v erify the signatures. F or e xample, if an attack er floods the netw ork with f ak e pack ets containing theoretically a rob ust signature. These are some of the reasons X.509 may not ne suitable for the system. Security pro vided by X.509 is also not completely inf allible. Some researchers ha v e e xploited some of its weaknesses, e.g. [26] demonstrated that tw o certificates containing identical signatures can be constructed using a collision attack on the MD5 hash function. Distrib uted Netw ork Protocol (DNP) secure authentication may not be suitable for securing commu- nication o v er PLC for smart meters due to its k e y management and specialisation e v en though it may be used on smart grid [28]. Orte g a et al proposed for the DNP3 o v er TCP/IP for smart grid application. This is not fea- sible for PLC due the follo wing: DNP Session K e y is periodical ly changed and used to calculate the HMA Cs. The Update k e y occupies the second le v el and is used for encryption of the Session k e y before it is sent to the remote de vice. F or load management on PLC, DNP w ould place a lar ge processing o v erhead[29]. Another dra wback of DNP Secure Authentication if it is used on PLC netw ork to authenticate load management mes- sages between smart meters and data concentrators, is that when Update k e ys are compromised or corrupted, or if the custodian of the k e y lea v es the or g anisation, the po wer supplier has no choice b ut to dispatch personnel to the remote de vices to change the Update k e y . Thousands or e v en millions of smart meters are connected on the grid, therefore pending remote do wnload of Update K e ys, practical systems are restricted to perhaps hundreds of de vices. DNP Secure Authentication utilises 16-bit v alues for addresses and user numbers, thus presenting a scalability challenge. Challenge-Handshak e Authentication Protocol (CHAP) in a smart grid system that includes smart meters is not feasible [30]. In the ne xt section we therefore present TESLA as the most ef fecti v e scheme that may be emplo yed to ef ficiently secure PLC for load management. TESLA in its modified form can authenticate pack ets immediately and due to its lo w computational and per -pack et communication o v erhead. 3. TIMED EFFICIENT STREAM LOSS-T OLERANT A UTHENTICA TION (TESLA) TESLA is widely used to authenticate broadcast messages [31, 32], such as DoS attack-tolerant TESLA-based broadcast authent ication protocol in Internet of Things [33]. W e first present an o v ervie w of TESLA by outlining properties that mak e TESLA suitable for securing PLC for load management systems. W e then discuss threat model and security guarantee and the modification to TESLA needed to secure load man- agement through PLC. These modifications include, using indirect time synchronisation for loose time syn- chronisation to combat the DoS threat and instantaneous authentication to pre v ent delay . W e selected TESLA for securing PLC for load management based on its follo wing properties: Low per -pac k et communication o verhead : The calculation of MA C utilises the n m parameter [27], which is the length of the truncated output of the function. The n m v alues depend on the MA C function Modified timed ef ficient str eam loss-toler ant... (Boyce Sigweni) Evaluation Warning : The document was created with Spire.PDF for Python.
2286 r ISSN: 2088-8708 selected, hence, per -pack et communication o v erhead can be as lo w as 80 bits. Low computation o verhead : The primary reason for the use of smart meters is to sa v e electricity . Smart meters ha v e limited or lo w processing po wer which sa v es electricity . Hence, TESLA is ideal because of its authentication protocol, which is not po wer hungry . It in v olv es one hash computation done on the message and one MA C function computation done on the k e y and message per pack et. Therefore, TESLA requires minimal computational ef fort, therefore can be managed by smart meters and data concentrators. No r eceiver -side b uf fering : Ev ery pack et will be authenticated as soon as it arri v es at the recei v er; therefore, there is no need for pack et b uf fering at the recei v er . P ack et loss tolerance: All pack ets recei v ed within their time interv al will be authenticated e v en if the preceding pack et w as lost. Superior ass ur ance of authenti city : Pr o vidi ng the cryptographic and timi ng assumptions are enforced as the recei v er has a high pledge of authenticity , therefore, the system pro vides a formidable authenticity . Scalability : There are no ackno wledgement s after the initial set-up connection has been es tablished, therefore, during normal communication data flo ws only from the sender to the recei v er . This entails that the sender’ s authentication o v erhead is not dependent on the number of recei v ers; making the scheme v ery scalable. F or instance it will allo w one data concentrator to communicate with man y smart meters as per the current set-up for load management were one data concentrator can ha v e o v er 1000 smart meters connected to it [34]. 3.1. Thr eat model and security assurance Smart meters are installed in customer homes, therefore, the o wners ha v e unlimited access to smart meter in the pri v ac y of their homes. In addition, customers also ha v e unrestricted access to the PLC channel through po wer points in their houses where the y plug their appliance s. W e present a modified TESLA that is secure ag ainst a formidable adv ersary who by virtue of being able to access the channel and de vice has the follo wing capabilities: (a) The challenger has a right to use to a f ast netw ork with insignificant delay . (b) The challenger can listen in, capture, retransmit, drop, hold-up, and modify pack ets thereby ha ving full control o v er the PLC channel. (c) The challenger’ s computational resources may be v ery formidable, b ut not unbounded. In particular , this means that the adv ersary can perform ef ficient computations, such as computing a reasonable number of pseudo-random function applications and MA Cs with ne gligible delay . Nonetheless, the adv ersary cannot in v ert a pseudo-random function (or distinguish it from a random function) with non-ne gligible probability . 3.1.1. Security assurance The security assurance with this modified TESLA scheme is that the recei v er should not accept an y message M j as authentic e xcept for when M j w as sent by the alle ged sender . This security assurance includes protection ag ainst message duplication through message numbering and time-stamping and we also address denial-of-service (DoS) attacks. 4. RESEARCH METHODOLOGY 4.1. Repeated measur es design W e used repeated design measures for this study because of —Reduction in the v ariance of results. This allo ws statistical inference to be made with fe wer runs and man y e xperiments can be completed more quickly , as fe wer cases need to be trained to complete an entire e xperiment. This enables us to monitor ho w message size change o v er time for both requests and response messages. Stra w-man reference design for demand response information e xchange [35] i s used to present a guide to ho w security is pro vided through implementation of the proposed authentication protocol, in the enabling services layer of the load management infrastructure. The message is sent do wn the stack to the security layer which performs a hash computation on the message and k e y and then sends t he hashed message o v er the PLC netw ork [36]. When the security layer at the recei v er recei v es the hashed message from the PLC and authenticates it using disclosed k e y or MA C ( i.e. HMA C-MD5). If authentication is successful the message is sent up to the application layer otherwise it is discarded. The ne xt subsection sho w who modification are made on TESLA for PLC security . Int J Elec & Comp Eng, V ol. 9, No. 4, August 2019 : 2281 – 2295 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 2287 4.2. TESLA modification f or PLC The original TESLA is modified in se v eral w ays to mak e it ef ficient and practically suitable for PLC netw ork for Load Management via use of Data concentrators and smart meters. Smart meters are connected to the data concentrator from dif ferent distances because some houses are close to the distrib ution transformer while others are quite a distance a w ay . Therefore, the first modification is the use of the authentication chains with dif ferent disclosure delays to cater for the dif ferent distances of the smart meters from the data concen- trator . Secondly , we present the technique to support Instantaneous Authentication, implying that the recei v er w ould be able to authenticate a pack et immediately upon arri v al without delay . A data concentrator can be con- nected to man y smart me ters. F or e xample, Echelon NES data concentrator [34] can connect o v er 1000 smart meters, and o v er 4000 other de vices. Therefore, there is the a need for modifications to address the scalability issue and vulnerability , both due to time synchronisation protocol. In the ne xt sub-sections, the issue of smart meters being at dif f erent distances a w ay from the data concentrator resulting in dif ferent netw ork delays is addressed by emplo ying a space optimisation method whereby the data concentrator uses se v eral TESLA instances for one stream. T o successfully address this issue we ha v e to look into time synchronisation and attend to the k e y management techniques as well as address the vulnerability that could rise from use of these methods and techniques and ho w to eradicate or minimise them. 4.2.1. Optimal Disclosur e Delay and T ime Inter v al P arameters The follo wing parameters must be determined by the sender for optimal performance as per the re- quirements of PLC based load management. These parameters are ( T int ), the interv al duration which usually ranges from 100 milliseconds to 1 second e xpressed in milliseconds and the k e y disclosure delay ( d d ) which is the w aiting time before the k e y is disclosed. A good choice of T int and d d is essential for the ef ficienc y of the scheme. F or e xample, if the product of T int and d d is too lar ge, it causes an e xcessi v e delay in the process of authentication, and when it is too lo w , it will den y most recei v ers the opportunity to v erify pack ets. The parameters T int and d d must not be altered throughout the durat ion of a session to pre v ent introduction of vulnerabilities. 4.2.2. Optimal T ime Inter v al T o determine the optimal time interv al duration, the sender w ould di vide the time into st andardised interv als of duration T int . The numbering for the time interv al s tarts at 0 and incremented successi v ely . An unsigned 32-bit inte ger is used to store the interv al inde x. Therefore, the wrapping to 0 can only tak e place after 2 32 interv als thus making the system to be v ery scalable. F or e xample, if: T int = 0 : 5 seconds, then the wrapping will only happen after 0 : 5 2 32 = 2147483648 s , which translates to approximately just o v er 68 years before wrapping to 0 can tak e place [27]. 4.2.3. Optimal Disclosur e Delay T o determine the optimal disclosure delay in v olv es a trade-of f. This is because smart meters that are close to the data concentrator ha v e lo w netw ork delay , hence, demand short k e y disclosure delays because it results in short authentication delays. Unfortunately , using a short k e y disclosure delay means that smart meters that are f ar from the data concentrator (with long netw ork delay) will not be accommodated because most of their pack ets will arri v e outside the set period hence violating the set security condition. Therefore, the y will be discarded without authent ication. Emplo ying a long k e y disclosure delay will resul t in unneces sary delay in authentication for smart meters close to the data concentrator . It is important to note that the security aspect of the system is not af fected whether long or short k e y discl osure delay is used. This is mainly a performance f actor , and performance is v ery important for ef fecti v e Load Management. Ho w the system will perform depends hea vily on the c h oi ce of the k e y disclosure delay . W e illustrate ho w to determine a k e y disclosure delay ( d d ) for a system using indirect time synchronisation. W e do that by pro ving that if the round trip time ( R tt ) is a suf ficient upper bound time between the smart meter and data concentrator , then the optimal choice for d d is as follo ws; d d = D S R + " T int + 1 (4) where: T int is the duration of the interv al, D S R is a suf ficient upper bound on netw ork delay for pack ets tra v ersing from sender Modified timed ef ficient str eam loss-toler ant... (Boyce Sigweni) Evaluation Warning : The document was created with Spire.PDF for Python.
2288 r ISSN: 2088-8708 to recei v er and " T ime synchronisation error sum for both sender to recei v er T o deri v e the disclosure delay we first ha v e to mak e sure it does not mak e pack ets to violate the security conditions. W e tak e into account a pack et P j created in the time interv al I i and the k e y will be disclosed d d time interv als later , when the pack et P j at the recei v er its local time is gi v en as equal to l T R , thus the security condition is that: d d > l T R + T n T int I i (5) where: T int is the duration of the interv al, T n is the be ginning of the n th time interv al and T ime synchronisation error sum (full round-trip time). W e use the assumption the pack et P j w as sent when the senders’ local time w as l T S , hence: l T S < T int = ( I i T int ) + T n + T int , therefore the round trip time R tt = D S R + D R S , with D R S denoting the netw ork delay from the recei v er to the sender . Using the deri v ation from Perrig et. al. [37] referring to Figure 3, Resulting in eqn D S R = l S R + l T S . Finally we ha v e a tight bound for d d satisfying equation 4 and this d d af fords most pack ets the opportunity to meet the set security condition and the recei v er w ould not ha v e to w ait longer than necessary before authenticating the pack ets. The optimal d d does not solv e the issue that smart meters are at dif ferent distances a w ay from the data concentrator . It is just the best time for meters at one particular distance. T o address this issue, one approach w ould be to use multiple TESLA instances and treat them independently each with its o wn k e y , hence d d . Unfortunately this approach results in unmanageable communication o v erhead because of this multiple k e ys for each instance. In the ne xt section we present an optimisation that reduces the space o v erhead of multiple instances by using the same k e y chain with a dif ferent k e y schedule for all instances [27].   R t 3 t 2 t 1 t S t ∆ δ Figure 3. Recei v er and Sender delays [37] 4.2.4. Multiple concurr ent TESLA instances The core idea for this technique is to mak e use of the same k e y b ut a dif ferent schedule for all instances as an alternati v e to utilising one self-determining k e y chain for each instance. It w orks as follo ws; all instances for a stream share the same k e y chain and the same time interv al period. That is each time interv al I i , is associated with the corresponding k e y K i , in the pro vided k e y chain. Therefore, we can e xpect K i to be re v ealed in the time interv al I i . Figure 4 depicts an e xample of ho w multiple instances could be arranged to be used for concurrent TESLA instances. In this case there are tw o TESLA instances, ha ving a k e y disclosure time of one interv al and the other fi v e interv als [27]. In Figure 4 the bottom ro w of k e ys sho ws the k e y re v ealing plan. It sho ws which k e y is re v ealed at which time interv al. The top and middle ro ws of k e y sho w the k e y schedule of the tw o instances, the latter being the first instance while the former being the second instance. F ollo wing this method, the sender needs only to disclose Int J Elec & Comp Eng, V ol. 9, No. 4, August 2019 : 2281 – 2295 Evaluation Warning : The document was created with Spire.PDF for Python.
Int J Elec & Comp Eng ISSN: 2088-8708 r 2289 one k e y chain inspite of ho w man y instances are used concurrently . This technique al lo ws space sa ving. F or e xample, if each k e y is 16 bytes long, then for a stream with n concurrent instances, this method will conserv e 16( n - 1) bytes per pack et and for small pack ets such as the ones used for PLC Load Management. This is a significant sa ving. Using concurrent instances also helps in achie ving scalability . One issue to consider is the vulnerability of the TESLA due to the mechanism emplo yed for the k e y chain reconstruction at the recei v er . First, the recei v er must check if the k e y chain arri v ed within the stipulated time interv al. If that time has e xpired then the pack et is discarded els e the recei v er will try to v erify the k e y re v ealed in the pack et by putting into operation the pseudo-random function until the v ery last committed k e y chain v alue. This operation can be e xploited by an attack er who w ould timestamp their pack et with a time f ar in the future. Therefore, when the recei v er checks if the time has e xpired it will find that the time is still v alid and therefore attempts to v erify the k e y , pre v enting it from v erifying the le gitimate pack ets. That results in denial of service for deserving pack ets. A no v el approach to deal with this is to ha v e lo wer and upper time limits for pack ets so that if a pack et is sent with future timestamp it is dropped [27].   Figure 4. Recei v er and Sender k e y delays [37] 4.3. PLC TESLA instantaneous authentication Basic TESLA requires the recei v er to b uf fer pack ets before the y can be authenticated. This is becaus e the sender sends the k e y required for authentication at a later stage. This delayed authentication is not suitable for Load Management because monitoring and control command actions need to be carried out in real-time. F or e xample, if the grid is e xperiencing some instability , the information must be relayed immediately to the control centre without delay . Also, if the load e xceeds supply and needs to switch of f non-critical b ut high po wer consuming de vices such as heaters, that action must happen immediately without delay or there will be the risk of po wer outages while w aiting for the command to switch of f de vices to be authenticated. This delayed authentication also causes storage problems, requiring data conce n t rators and smart m e- ters to ha v e lar ge memories to store these pack ets while the y are w aiting to be authenti cated. The other disad- v antage of this delayed authentication is that it mak es the system to be vulnerable to Denial-of-Service attack. It is because of the reasons abo v e that modific ations to the original TESLA are required so that pack ets can be authenticated instantaneously upon arri v al with no delay . Therefore, this eliminates the need for b uf fering at the recei v er side, thus reducing the risk of DoS attack where the attack er floods the recei v er with spurious pack ets. As it w ould be seen later in this section, this modification comes at a cost of at least one e xtra hash per pack et and the need for b uf fering at the sender side. This is acceptable since it does not induce the risk of DoS (by flooding), or introduce significant delay . In this method, sender b uf fering replaces recei v er b uf fering. The sender b uf fers pack ets during one disclosure delay so that it can put the hash v alue of the data of the ne xt pack et in an earlier pack et. Therefore, the instant the earlier pack et is authenticated the ne xt pack et will be authenticated as soon as it arri v es at the recei v er through its hash v alue that w as contained in the earlier pack et thus achie ving instant authentication with no more delays. T o simplify the illustration of ho w this is achie v ed, we assume that the sender will send out a constant number n of pack ets per time interv al. Figure 5 sho ws ho w a pack et for the message se gment M j in the interv al T j is constructed. The hash v alue of the ne xt message M j + v d is appended to the current message, that is H ( M j + v d ) is appended to M j . The sender then calculates the MA C v alue o v er the k e y K i together with H ( M j + v d ) to get M AC ( K i ; D j ) where D j = H ( M j + v d ) jj M j (note that jj means that messages are concatenated). Modified timed ef ficient str eam loss-toler ant... (Boyce Sigweni) Evaluation Warning : The document was created with Spire.PDF for Python.
2290 r ISSN: 2088-8708 j M ) ( v d j M H + ) , ( ' j i D K M A C d i K − v d j M + ) ( 2 v d j M H + ) , ( ' v d j d i D K M A C + + i K j P v d j P + j D v d j M 2 + ) ( 3 v d j M H + ) , ( 2 ' 2 v d j d i D K M A C + + d i K + v d j P 2 +   Figure 5. Instantaneous P ack et Authentication[36] W ith reference to Figure 5, the technique for instantaneous authentication for the pack et P j + v d is as follo ws; P j incorporates a hash of the data M j + v d and this data is in P j + v d and if P j has been authenticated it implies that H ( M j + v d ) is also authentic. Therefore , the message M j + v d is authenticated immediately , hence using the same technique. The ne xt pack et P j +2 v d w ould also be authenticated immediately , so will the ne xt pack et. If a pack et is lost or discarded then the ne xt pack et w ould not be authenticated immediately b ut w ould be authenticated later through its MA C v alue. F or e xample, if P j w as lost or discarded, then P j + v d w ould not be authenticated immediately b ut will be authenticated as soon as the ne xt pack et P j +2 v d arri v es. . It will be authenticated t hrough its MA C v alue because upon arri v al pack ets disclose the k e y of the pre vious pack et, therefore P j +2 v d w ould disclose the k e y K i + d which w as used for P j + v d MA C v alue, therefore P j + v d w ould then be authenticated. Delayed authentication can be easily be o v ercome by incorporating hashes of multiple future messages. This can easily be done in PLC Load Management because all the messages and their sequence of transmission is kno wn. This is a technique similar to Ef ficient Multi-chained Stream Signature (EMSS) [38], and the introduced message o v erhead is ne gligible. Using multiple hashes eliminat es the need to send pack ets at a constant rate which is dif ficult in a hostile en vironment lik e PLC. 4.3.1. Indir ect time synchr onisation f or load management via PLC Complicated time synchronisation protocols are a v ailable b ut the y require considerable m anagement o v erhead, these are protocols such as the Netw ork T ime Protocol (NTP) [39], whi ch ha v e a high comple xity and attain properties electrical load management via PLC do not in v olv e. Loose time synchronisation is an essential component in TES LA b ut also a security Achilles’ heel, due to the mechanism for time synchronisation which mak es the system vulnerable to DoS through netw ork flooding wi th requests for synchronisation. It is for this reason t hat we present a modified TESLA time synchronisation protocol that is simple and yet secure, that will meet the modest requirements of Load Management via smart metering through a PLC channel. The sender (data concentrator) and each recei v er ( smart meter) must synchronise independently se- curely through an e xternal time reference, when Indirec t T ime Synchronisation (ITS) is used. T o achie v e this synchronisation se v eral options are a v ailable: (a) Senders and recei v ers could synchronise via NTPv3, NTPv4 (Netw ork T ime Protocol v ersion3 /4) [39] or SNTPv4 (Simple Netw ork T i me Protocol v ersion 4) hierarch y of serv ers [40]. Unfortunately , this cannot be adopted for synchronisation of smart meters and data concentrators because for load management via PLC the g ate w ay for smart meters is the data concentrator; therefore, smart meters cannot ha v e an independent path direct to the serv ers. (b) The s econd option which w ould guarantee direct access for both sender and recei v er to e xternal time reference w ould be for the sender and recei v er to synchronise via a GPS system or an y simil ar de vice that can pro vide a high precision time reference. Unfortunately , spoofing attacks on the GPS system ha v e been reported [41] therefore the le v el of security required for PLC load management cannot be guaranteed when synchronisation is achie v ed through GPS. (c) The other option, we adopt for PLC based load management system is whereby a dedicated hardw are is embedded in each recei v er and the sender that pro vides a clock that has a time-drift that is ne gligible in-terms of the time accurac y requirement for TESLA. T o deal with this insignificant clock drift an yw ay , the de vice mak es it possible for the sender and recei v er to ha v e their embedded clock to be synchronised with the of ficial time reference periodically . This can be done during equipment servicing interv al or after a period of kno wn maximum allo wed clock drift and thereafter left to be autonomous. That is, the de vice w ould continuously consult its internal clock which has minimal clock drift. Int J Elec & Comp Eng, V ol. 9, No. 4, August 2019 : 2281 – 2295 Evaluation Warning : The document was created with Spire.PDF for Python.