Distributed Discovery Clients for Spectrum Allocation
DDistributed Discovery Clients for SpectrumAllocation
Torleiv Maseng , Øivind Kure and Magnus Skjegstad
1. Norwegian Defence Research Establishment (FFI)Postbox 25, 2027 Kjeller, Norway2. ITEM, NTNU3. Cambridge University
Abstract —By using a distributed P2P system where the agentsreside in in microprocessors already present in most radio nodeslike Wi-Fi access points, base stations, TVs connected to Internetetc, these agents can discover other agents over the back-haulnetwork. As a result, client node lists (similar to neighbor listsused in 3GPP) are created. An alternative is to use a centralizeddatabase. Why this best is done by distributed agents is discussedin this paper along with security considerations. Examples ofother applications which will benefit from this system is alsopresented.
I. I
NTRODUCTION
In 2010 it was 5 billion connected devices and in 2020 therewill be 50 billions of devices connected to Internet [1] by radio.This enormous increase is partly caused by introduction of IoTand M2M devices which communicate over short distances,e.g. inside a house. Since databases do not scale with numberof nodes and since there is little willingness for many cheapdevices to pay for centralized database assistance, we claimthat the use of a distributed architecture is better suited. Adatabase may however be used as a supplement.Once the Distributed Agents are universially implemented,they will simplify coordination among the various device typeslike Wi-Fi, power meters, TVs etc communicating via IPpackets over the back haul network. These agents can bethe same for all these devices, but the Dynamic SpectrumAllocation (DSA) will most likely depend on the radio systemand tailored specifically to each system. This DSA applicationwill enable the different radio devices to share the samefrequency band while maximizing capacity of the radio linksby minimizing interference between nodes operating in thesame band.Cognitive Radio (CR) or DSA requires control channelsto set up the radio channel. Since these are not standardizedfor all type of radios and require different radio protocols, wepropose to use the back haul network or the Internet for controlchannels since this is available to most access points, hubs,base stations etc. We have previously proposed a decentralizedDSA architecture and P2P protocol [2]. In this paper, we arguethat decentralized DSA improves the radio capacity for IoT andM2M communication in the open spectrum. We also discusssecurity implications and future directions of research.DSA is different for IoT and M2M devices compared tothe process for a mobile cellular operator. The local IoT andM2M nets are mainly owned by individuals and produced bycompanies whose prime business is to produce components and not run network management. Therefore the structure ofthese nets is distributed as opposed to the mobile networkswhich is centrally controlled. In a mobile network all basestations and their location are known and the coordinationbetween the base stations is therefore easier. In addition,they are all connected through the back haul network. Thecoordination proposed in this paper is therefore best suitedfor IoT and M2M nets unless the mobile operators need tocarry out coordination between different mobile operators orbetween mobile networks and IoT and M2M nets.In the case of CR for White space, the IEEE 1900 DYSPANgroup defines how the procedures for secondary users to getaccess to the TV part of the spectrum. ECMA 392 is designedfor shorter distances and personal devices operating in theWhite Space. IEEE 802.11af was formed for the purpose ofwireless local area network (WLAN) operating in TV whitespace spectrum in the VHF and UHF bands between 54 and790 MHz which increases the possible range and data ratecompared to 802.11 a/b/g/n/ac. The upcoming IEEE 802.11ahstandard which will use the 900 MHz band is expected to beapproved by January 2016. That makes 2014-2015 the timeframe for development of new silicon for end nodes and accesspoints. The standard is expected to cover many modest homeuses at 10-20 Mbits/s.The IEEE 802.19 is the Wireless Coexistence TechnicalAdvisory Group deals with coexistence between unlicensedwireless networks. Reusing existing spectra assignments hasbeen announced by the Radio Spectrum Policy Group assistingthe European on radio spectrum policy issues in EU. Thereforethe Norwegian Post and Telecommunications Authority and theSwedish Post and Telecoms Authority argue that DSA and CRshould be allowed secondary users as long as they maintainknowledge about other users in their vicinity and do not disturbthe primary users. This decision is expected to apply for allfuture spectrum assignments [3].Operation of a centralized database requires an organizationthat is willing to fund and coordinate the location database forall involved receivers and transmitters. For Wi-Fi access andIoT and M2M devices such an authority is not immediatelyidentifiable. An alternative is a more organic method based onvoluntary participation in peer to peer(P2P) networks. In [2] aDiscovery system architecture is described which enable radionodes to be discovered using a distributed P2P agent networkcommunicating over the back-haul network. This system isbased upon that all the nodes know their own position. a r X i v : . [ c s . N I] N ov ow positions are obtained for IoT and M2M networks,is discussed in [4] [5] and it can be done in the same wayas for UMTS (3G) and LTE . For networks with very shortrange, note that in order to carry out DSA among them, it it isthe radio connectivity matrix with path-losses that is neededand their relative locations are less important. The connectivitymatrix may be found by coordinated transmit and receivesessions and from the resulting measurements.Assuming that the Discovery Agents (DA) reside insidethe existing microprocessors in all participating radio nodes,additional features may be offered to the users of radio nodeslike Proximity Services in Figure 1. Examples of these servicesare context-aware services for roaming nodes and for nodesbelonging to different organizations, handover and a networkconsisting of many small base stations.II. A DISTRIBUTED
P2P
DISCOVERY SYSTEM
In [2], the P2P protocol for discovering radio devices wasfirst published in detail. The objective of the protocol is todiscover radio devices that are operating on the same fre-quencies in overlapping geographical areas. The P2P networkis established over the back-haul network to enable nodes todiscover each other without having radio contact. In order tomake the the Discovery Agents interoperable, the format of the”IP Discovery messages” and their content, need to be defined.The P2P protocol creates an unstructured overlay thatconnects nodes that are likely to be interfered or interferewith each other. In addition, nodes maintain connections toa few nodes that are far away, so that they can help reduce thediscovery time for new nodes that enter the network. The goalof the protocol is to produce a list of network addresses toother nodes one needs to coordinate frequency use with. Thisis accomplished using two main mechanisms that run on eachnode.The first mechanism is a random peer sampling mechanism[6] that aims to produce a random set of nodes from the overalltopology. The random set of nodes is mainly used as a startingpoint for the discovery mechanism. The random set is alsouseful to new nodes that have yet to find other nodes fromtheir area yet.The second mechanism maintains a set of nodes it hasdiscovered so far that have the highest utility according to autility function, using an adapted version of the T-Man protocol[7]. The utility function in [2] is based on degree of overlapbetween the frequency coordination areas of the nodes, as wedescribe later in this section. The more likely two nodes areto interfere each other, the more important they should beconsidered to be to each other. The mechanism contacts nodeswith high utility to exchange information in the hope that theymay know about even more important nodes that it is interestedin. Initially, when a node joins the P2P network, the nodes itknows about so far are in the random set of nodes producedby the first mechanism. However, as the second mechanismbegins to improve its results by contacting nodes with highutility, we gradually learn more about nodes that are moreimportant. Eventually, we discover the nodes that we need tocoordinate frequency use with. These nodes will also have
Field Length Description
Identifier 8 bytes Overlay node IDLocation 16 bytes Geographical locationCoordination radius 8 bytes Radius of coordination areaIPv4 or 4 bytes Source IPIPv6 address 16 bytesTimestamp 8 bytes When the news item was created
TABLE I. F
IELDS INCLUDED IN ”IP D
ISCOVERY MESSAGES ”. T
OTALLENGTH WITH IP V IS BYTES [2]. gathered information that we are interested in if they reside inthe same area, reducing the overall discovery time.The protocol has been extensively evaluated though simu-lation, and its ability to converge and find all nodes even if thestarting point (seed) was far away from the location of interest.This was demonstrated in a network with 2.3 million nodesrepresenting Wi-Fi routers corresponding to every householdin Norway. The protocol was implemented in real Wi-Fi routersrunning the open source Wi-Fi router firmware OpenWRT andmade available for free use as long as the origin of the sourcescode producer is retained in the source code [8].Table I lists the data fields distributed with the P2P pro-tocol in [2]. The fields should contain enough informationto enable the nodes to contact each other and maintain theP2P overlay. As the amount of information included in thefields distributed by the discovery protocol affects the overallbandwidth requirements, it should be kept as small as possible.The total size of the fields in Table I is 56 bytes, but couldin many cases be reduced further. After the nodes have foundeach other additional information can be exchanged (such asradio parameters, sensing information etc) directly using theIP address provided by the discovery protocol.In order to make the Dsicovery Agents interoperable, theformat of the IP Discovery Messages and their content shouldbe defined.Each P2P agent uses the P2P network to gather the networkaddresses of nodes operating in its surroundings. This isaccomplished by periodically contacting other P2P agents overthe back-haul network and requesting information about nodesthey have discovered. It is based upon a Gossip protocol [7]and communicates with only one other node at the time asopposed to multicast. Therefore the overhead bandwidth usageis limited. With a periodic interval of 15 seconds, the averageamount of data sent and received from a single client wouldbe approximately 0.5 kilobytes per second.An illustration of how coordination areas are used bythe P2P protocol with omnidirectional antennas is shown inFigure 2 (from [2]). Each node defines a coordination areaaround itself based on its transmission power and where ithas clients. Each node is then interested in discovering othernodes that have coordination areas that overlap their own. Themost important nodes (according to the utility function) willbe nodes that have the largest overlap between coordinationareas. In the illustration, node D will for example be of interestto all nodes as it has a large coordination area that overlapswith them. C on the other hand, may only be of interest toD as it can interfere with its users. Similarly, C may be veryinterested in getting in contact with D to negotiate how to avoidinterference. A and B are mainly interested in each other andD. ig. 1. Two radio nodes A and B are are connected via the Internet. They both share the same frequency band even if they may use different radio protocols.Once the Discovery Agent is in place, this enables the establishment of Neighbor Relation Tables in LTE and called Client Node Lists in this paper. This listis needed to execute DSA and proximity services application programs which may be different for A and B since they use different radio protocols. In order tomake the the Discovery Agents interoperable, the format of the ”IP Discovery messages” and their content, need to be defined.
DAB C
Fig. 2. Nodes A, B, C and D with coordination areas [2].
The results are ranked according to relevance and themost relevant nodes are then contacted in the next periodicexchange. Gradually all nodes will discover the nodes that aremost relevant to them. We refer to the most relevant nodes as candidate nodes , as they are candidates for additional resourcenegotiation.In the simulations, the number of exchanges required for2.3. million nodes to find all their candidate nodes after anew node has joined was 20 on average, corresponding to 5minutes. III. I
MPROVING SPECTRUM CAPACITY
By choosing operating frequency and transmit powerwisely, the capacity of radio systems can be improved byminimizing the disturbance between the radio nodes. This isnormally done by using a database in which knowledge ofwhere the TV transmitters are located (PAWS). This spectrumcan be further increased if knowledge where some of the TVsare located [9] [10]. In order to reduce interference, it is useful to change the operating frequency of wireless access points.This feature is already present in IEEE Std 802.11-2012, butfor another purpose.A commonly used criterion for Dynamic Spectrum Al-location systems is to maximize joint capacity. This is botha tough theoretical challenge (NP hard) and an impracticaloptimization criterion since the allocation may result in someusers getting too little and some getting too much capacity. Be-sides, the parameters needed like path loss and transmit powerare hard to measure accurately . There are few parameterswhich can be measured accurately in most communicationssystems considering that traffic load, propagation conditionsand user location is changing over time. There are many smartalgorithms for assigning power and frequency between radioterminals which share the same band [11] [12] .IV. P
RACTICAL CONSIDERATIONS
The framework has been described with a one to onemapping between IP address and access points location. In anactual deployment there will have to be agents that representmultiple access points. Most enterprises protect their accesspoints behind firewalls. To avoid opening the firewall and usepublic addresses for each access point, it is better to haveone agent that represents all the access points behind thefirewall. To enable incoming messages from other agents toenter the firewall, it will be necessary to open a port, usingport forwarding. The operation of the common protocol willbe executed on one machine rather than on many. Severaldiscovery messages will originate from the same agent.Similarly, IP address will not be used as the actual iden-tifier. Since a large fraction of access points are in privateaddress spaces, their actual IP addresses are not globallyunique. A possible solution is either to use the agent IP addressconcatenated with a local unique identifier, or the mac addressof the access point or using port forwarding. In these cases, theinformation fields in the protocol must be extended with theIP address of the agents, so a mapping between mac addressr port number and agent exists. In order to be self-organizingin private address spaces, an additional discovery protocolbetween the agent and the access point must be deployed.However, there are no special requirements, and a simplebroadcast response protocol is sufficient.V. S
ECURITY I SSUES AND P RIVACY I SSUES
In a system using a licensed spectrum, e.g. UMTS, thespectrum owner has the right to enforce an authenticationprocedure and has the means to do so, using the SIM card. Thespectrum owner has the right to deny to user from participationunlike a user of an unlicensed band like the ISM bands. Fora user of an unlicensed band, only the equipment need toadhere to some specification and the user is free to operate.To enforce authorization is therefore difficult. To encouragecoopertion only with devices authorized by a third party likeGoogle, would however be a possibility in an attempt to isolatenonauthorized users.A distributed system has a slight edge against a centralizedsystem regarding the ability to detect misbehaving participantsbecause the participants can compare the information againsttheir own measurement locally and thereby judge the validityof the information. Such a system can be augmented with adistributed reputation system, adding some improvements inthe validity checking of the information. In a centralized archi-tecture it is more complex to marshal and collect the relevantmeasurements. This is because anomalies must me checkedfor all participants remotely and more information must betransferred to a central point instead of being processed locally.The vulnerability and possible defense methods of a cen-tralized system against Distributed Denial of Service attacksare well documented in the last decade. A distributed systemis more resilient. A substantial attack can be viewed asan increase in transaction volume. In our simulations, thesystem scaled well up to millions of nodes. An attack throughincreasing the number of nodes should therefore have minimaleffect. It may stop individual nodes, and make allocation worsethat optimal, but not stop the overall system from working.A possible attack vector is adding and withdrawing accessnodes, called churn in the following. An analysis of churn[2] showed a reasonable stability for our proposed system.However, large volumes of adding and deletion of nodes willaffect the convergence and result in that some nodes will befalse and som not yet detected. The allocation may thereforenot be correct. Essentially, the lack of convergence will resultin hints that potentially will not be worth following andtherefore neglected by the access points. As such excessivechurn can be used to neutralize the system, but then the scalewill be sufficient to be detected by the traditional mechanismsthat are already implemented in the back-haul network. How-ever, individual agents and access points may be attacked andoverwhelmed by churn and volume of messages. This is a riskall nodes in the internet are already running.
A. Privacy Issues
The proposed system raises some privacy issues. In prin-ciple, it maps, an IP address range to a particular location.This information is available to anyone pretending to be at the
Fig. 3. Node 1 in the figure has obtained a Neighbor List consisting of nodeA,B, C and D. In his Neighbor List the IP address of node X is visible atlocation B. It is node X who runs the agent on behalf of B. This secret isonly known to X and B. location and participating in the system. To hide this infor-mation, the system need to create a mapping between accesspoints and agents that can do the negotiations. This can bedone if all users who ”opt for privacy”, called ”privacy nodes”in the following, must offer to run agents on behalf of othernodes for a certain period of time. Each privacy nodes select apartner from this pool of nodes randomly. This is illustrated inin Figure 3. This selection can work asynchronously among theprivacy nodes and is selected among privacy nodes with sparecapacity to run the agent remotely. An agent can be selected torepresent several nodes and must announce it’s current capacityto run a agent on behalf of others nodes at random locations.The discovery, convergence and negotiations will be the same.However, the IP address of the access point cannot be inferredby the IP address of the agent. The detailed design of such asystem has not been implemented.
B. Susceptibility to malignant manipulation without any cen-tral authority
Some users may invent additional access points with lo-cation close to their actual location using the same channelas the actual point. To external viewers, the channels willappear as overloaded, while in reality it is used by onlyone access point. Such an attack will not differ from thelegitimate use of an agent announcing multiple access points.Our recommended solution is to treat the proposed allocationas a hint. If the system is not manipulated, the outcomeis a better solution. However, the hints may be the resultof manipulations. In additions, there might also be accesspoints that do not participate in the protocol or refuse tofollow the hints. The user must therefore implement their owncontrol strategy for the validity of the hint. Such a strategycan be implemented without the information exchange withneighbors.As an example, the users of the access point can selectto follow the hint from the resource allocation. If the Qualityof Experience (QoE) is improved, they continue to follow thehint. Otherwise they revert back to their original allocation orandomly select a new channel. Overall the users gain fromparticipating. If the system is sabotages, they will temporarilyselect a channel that reduces their QoE. However, they willrapidly revert back to an expected QoE they would havewithout participating in the system. If the system is notsabotaged, the QoE is improved. As part of a cyber-attack,external force can inject massive location messages in orderto try to disrupt the system. In the simulations, we haveshown that the system can handle large volumes of traffic.In order to be successful, order of magnitude larger volumeswill potentially be needed in order to disrupt the system.VI. O
THER PROXIMITY SERVICE APPLICATIONS
The main purpose of this paper is to introduce how aDSA system can be implemented for IoT and M2M devicesas discussed in the previous sections. There may be otherapplications too presented in the following, but these have notbeen evaluated thoroughly.
A. Context-Aware Services for roaming nodes
To adapt the services to the context of the user consider-ing his environment, is very popular and involves using theknowledge about the location of the user to offer services. Asan example, once the location is known, geospatial databasesare often used to answer queries from mobile computer usersfor the nearest post office, best realtor, directions to the airport,and so forth. Considering where the mobile user has been andprojecting in which direction he is going, could offer yet, moreservices. For this purpose the Discovery Agent presented inthis paper will be most useful. A common way to implementpublish/subscribe M2M services is by using Web services. TheOASIS standard WS-Notification can be leveraged for thispurpose, as it describes how topics can be used for subscriptionusing XML schemas. How to extend this scheme to locationsas a topic is not straight forward especially for mobile nodesfor which the topics must be organized in areas and new topicsneed to be addressed following the route. An easier way isto restrict the search for topics among the candidates on thecandidate node list provided by the Discovery Agent.
B. Contextual Services for IoT beween different organizations
As long as the devices are registred and managed by oneorganization, there exist lists of devices and the need forcoordination is less. Between devices belonging to differentgroups like that which is found e.g. in a smart city, thereis already many devices which may benefit from cooperationto make better judgments which make the life easier for theoccupants. Motion detectors used for burglar alarms can beshared for switching on the light in rooms and shut it off asthere is no movement. Smoke detectors for fire alarm maybe shared with systems for illumination by switching on thelight when an alarm is triggered. Water leakage triggered by adishwasher may also result in that the light is switched on etc.All these sensors and more, can be monitored and managedremotely by mobile devices. To make the information of thesensors available to all other in a local network, the can beconnected by wire or by radio to Internet. To make them awareof each other and provide a IP connectivity matrix a Discoveryprotocol is must useful.
C. Handover
Once the nodes move withing the coverage of fixed accesspoints, there is a need of hand-over to ensure uninterruptedservices. For mobiles,3GPP compliant mobile networks (3GPPin Release 8) [13] uses an system architecture called AccessNetwork Discovery and Selection Function (ANDSF) used bythe evolved packet core (EPC) network. This enables not onlyuser equipment like 3GPP access networks (such as HSPAor LTE) , but also non-3GPP access networks such as Wi-Fi or WIMAX to be discovered by radio and connect toeach other. A similar functionality for Wi-Fi is IEEE 802.11u.802.11u was developed to effectively automate how devicesconnect to available Wi-Fi networks. 802.11u enables Wi-Fihotspots to advertise their capabilities and then allows devicesto connect to them automatically rather than requiring the enduser to manually select an SSID. One of the challenges whenincluding Wi-Fi traffic into the 3GPP network is that Wi-Fi traffic must be connected to the EPC network via secureIP tunnels (GRE, IPSec. This makes this interconnectionchallenging to implement.In IEEE 802.21 Information Service is specified to allownetwork entities to discover information that influences theselection of appropriate networks during handovers. Since thisstructure is based upon a centralized concept, this informationmay supplement the Candidate Node List for a given geo-graphical area and shared directly with surrounding nodes toreduce traffic to the Information Service server.The ANR [14] function relies on cells broadcasting theiridentity. In order to minimize the time of no connection duringhandover, it is very useful to be able to configure the operatingfrequency prior to the new link establishment.In the FP7 EU project MOTO, Wi-Fi technology is consid-ered for offloading mobile networks [15] . A similar effort is touse Heterogeneous Network (Hetnet) based on vendor 3GPP-standardized and coordinated radio network with integratedWi-Fi and traffic management. In spite of this, there is a strongunwillingness among operators to introduce features whichmay result in loss of paid services. To enable handover froma LTE network to a Wi-Fi system may therefore be difficultunless the operators are the owners of the Wi-Fi system as apart of the 3GPP architecture. This reluctance exists even if ahandover to a Wi-Fi net is in the interest of the operator torelieve the operator network which is fully loaded.
D. Many small base stations
When femtocells or Hotspots are deployed in large quan-tities , it will be practically impossible to do careful radio-planning. For this purpose, the concept of Self OrganizingNetworks (SON) has been introduced in 3GPP Release 8,to make planning, configuration, management, optimizationand healing of mobile radio access networks simpler andfaster. This initiative can enjoy the automatic DistributedDiscovery System proposed. Since these are normally ownedand operated by an operator, the members are authenticatedand the members can trusted.To enable the mobile network to offer more capacity tomany users even if the spectrum is limited, it is necessary tomake the number of users per base station small. Ultimatelyhere will only be one user per base station and this user doesnot need to share the bandwidth with other users. This basestation is called a femtocell and a network of very small andlarge base station is called HETerogenous Networks (HET-NET) [16] in LTE and Hotspot in Wi-Fi. Hotspot 2.0 is anattempt to automate network discovery, registration, provision-ing, and network connectivity, which are manual steps todaywhen a user connects to a given Wi-Fi hotspot.A problem with the introduction of pico-cells in LTE comesfrom cell association protocols. Traditionally the user connectsto the cell with the highest signal power to achieve maximumuser performance, but in the case of Hetnet s this can degradethe overall throughput because the pico-cells have much lowertransmit power than the macro cells.This means that the macrocell serves more UEs than it should and the pico-cell becomesunder utilized because of small coverage area. This problemcan be solved by creating Neighbor Relation Tables producedfrom information obtained through the back haul networkrather then radio, using the results of the Discovery systemarchitecture proposed in this paper.VII. D
ISCUSSION AND C ONCLUSION
A Distributed Discovery System has been described whichis believed to be well suited for IoT and M2M devices andparticular those owned and operated by individuals. It is ourgoal with this effort to initiate further modules like candidateResource Allocation components to be implemented at testeda shared and finally standardized. Another example is to spurnew ideas related the shared frequency bands and in particularthe White Space.The advantage of this gradual implementation approach isthat no operator need to be chosen and no common infrastruc-ture is needed- it may be implemented by two neighbors andgrow organically as the usefulness of the system is accepted.R ((2012-09)) (Release 11)[6] Jelasity, M., Voulgaris, S., Guerraoui, R., Kermarrec, A.M., van Steen,M.: Gossip-based peer sampling. ACM Trans. Comput. Syst. (3)(2007). DOI 10.1145/1275517.1275520. URL http://doi.acm.org/10.1145/1275517.1275520[7] Jelasity, M., Montresor, A., Babaoglu, O.: T-man: Gossip-based fastoverlay topology construction. Computer Networks (13), 2321–2339(2009)[8] p2p-dprd. https://github.com/MagnusS/p2p-dprd. [GitHub November2013] [9] Wang, J., Baykas, T., Filin, S., Rahman, M., Song, C., Harada, H.:Coexistence protocol design for autonomous decision-making systemsin tv white space. In: Wireless Communications and NetworkingConference (WCNC), 2012 IEEE, pp. 3249 –3254 (2012). DOI10.1109/WCNC.2012.6214368[10] Probasco, S., Bajko, G., B. Patil, B.: Protocol to access white spacedatabase: Use cases and requirements draft-probasco-paws-overview-usecases-01. Internet Engineering Task Force (IETF) (2011). Nokia[11] Brik, V., Rozner, E., Banerjee, S., Bahl, P.: Dsap: a protocol for coordi-nated spectrum access. In: New Frontiers in Dynamic Spectrum AccessNetworks, 2005. DySPAN 2005. 2005 First IEEE International Sympo-sium on, pp. 611 –614 (2005). DOI 10.1109/DYSPAN.2005.1542680[12] Ellingsæter, B., Skjegstad, M., Maseng, T.: A potential game for powerand frequency allocation in large-scale wireless networks. CoRR abs/1212.0724 (2012)[13] Sachs, J., Olsson, M.: Access network discovery and selection in theevolved 3gpp multi-access system architecture. European Transactionson Telecommunications (6), 544–557 (2010)[14] Sesia, S., Toufik, I., Baker, M.: LTE: the UMTS long term evolution.Wiley Online Library (2009)[15] Reggani, A., Whitbeck, J., Dias de Amorim, M., Fonseca, M., Conan,V., Fdida, S.: Mobility trace breeding. In: Wireless Days (WD), 2013IFIP, pp. 1–3 (2013). DOI 10.1109/WD.2013.6686527[16] Hong, M., Garcia, A.: Mechanism design for base station associationand resource allocation in downlink ofdma network. Selected Areas inCommunications, IEEE Journal on30