36th Telecommunications Policy Research Conference, Sept Quantifying the Costs of a Nationwide Broadband Public Safety Wireless Network

Transcription

1 36th Telecommuncatons Polcy Research Conference, Sept Quantfyng the Costs of a Natonwde Broadband Publc Safety Wreless Network Ryan Hallahan and Jon M. Peha Carnege Mellon Unversty Abstract The problems facng the publc safety wreless communcaton systems n the US could be sgnfcantly reduced or elmnated through the deployment of a sngle natonwde network that serves all publc safety personnel. Two major efforts towards such a natonwde network are the Integrated Wreless Network (IWN), a program only for federal emergency responders, and an effort by the FCC to create a publc-prvate partnershp n the 700MHz band that serves state and local emergency responders; the future of both projects s uncertan due n part to concerns surroundng cost. To nform these concerns, ths paper presents the frst verson of a fully transparent model to estmate cost for two fundamental approaches: a publc-safety-only network and a publc-prvate partnershp whch serves both publc safety and commercal subscrbers. We apply ths general model to four scenaros: 1. a publc-safety-only network that only serves all publc safety personnel (.e. local, state, and federal) on 10MHz of spectrum n the 700MHz band, 2. a publc-prvate partnershp that serves all publc safety personnel and commercal subscrbers on 20MHz of spectrum n the 700MHz band, and 3&4. a network that only serves all publc safety personnel n ether of the two bands that may be used for the federal-only IWN project (168MHz & 414MHz). In each of these scenaros, we consder networks that carry voce only, data only, and both voce and data. We demonstrate the neffcences of the exstng publc safety nfrastructure by showng that a sngle natonwde network could be bult n ts place wth a small fracton of the tower stes and spectrum. In fact, the cost of buldng an entre natonwde system s comparable to what s lkely to be spent n just a few years on the exstng nfrastructure. More specfcally, for the publc-prvate partnershp carryng voce and data, we found deployment costs on the order of $10 bllon whch s less than the $15 20 bllon prevously estmated. For the publc-safety-only network carryng voce and data at 168MHz, we found deployment costs on the order of $6 bllon. Thus, f suffcent spectrum can be dentfed, the current IWN system could be extended to nclude state and local responders and provde broadband data wthout a sgnfcant ncrease n cost. In addton, these cost estmates are hghly dependent on some key parameters, such as those related to capacty and coverage relablty, over whch there has been lttle serous debate. If a publc-prvate partnershp s to be successful, values must be establshed for such parameters before bds are sought. Otherwse, potental bdders cannot even roughly estmate ther costs. Addtonally, we fnd that 83% of US area s currently covered by exstng publc safety wreless systems, whereas some clam the populaton buld-out requrement establshed by the FCC would cover just 63% of the US, and the actual coverage s more lkely to be roughly 50%. We also show that the estmated cost savngs from relaxng the exstng buld-out requrements are overstated. The authors gratefully acknowledge the fnancal support of the MacArthur Foundaton. Ryan Hallahan, Ph.D. Student n the Department of Engneerng and Publc Polcy, Carnege Mellon Unversty, hallahan@cmu.edu Jon M. Peha, Professor of Electrcal Engneerng and Publc Polcy, Carnege Mellon Unversty, peha@cmu.edu,

2 1 Introducton Consderng the mportant role that they play n ensurng the safety of the publc, the exstng publc safety wreless communcaton systems n the Unted States are far from adequate. Ths can be attrbuted to several factors, but chef among them s that the many publc safety agences across the country deploy networks ndependently wth lmted coordnaton or standardzaton wth neghborng agences [1]. The potental deployment of a natonwde publc safety wreless system presents an opportunty to solve some of these current problems [2]. A major mpedment to the deployment of a natonwde publc safety wreless network s the substantal estmated cost of such a project as well as the uncertanty surroundng ths cost. By understandng these costs better and understandng what factors mpact them, polcymakers may be better able to determne f any of the proposals currently beng consdered are even feasble and f so whether or not they present a cost savngs when compared to supportng the exstng nfrastructure. In addton to cost, there are some fundamental dfferences between the current proposals for deployng a natonwde publc safety network and polcymakers must understand and be able to wegh the tradeoffs between these proposals. In ths paper, we provde a frst-cut analyss of the costs of varous network proposals ncludng a natonwde publc-prvate partnershp and a natonwde network that serves only publc safety whle nvestgatng the tradeoffs between each of them, wth more detal comng n future work. In secton 1.1, we provde some background on the exstng publc safety wreless communcatons nfrastructure n the US, dentfy the factors whch have contrbuted to ts unfortunate state, and dscuss why a natonwde network s a possble soluton. In secton 1.2, we present and compare the recent proposals for a natonwde publc safety network n the US, and present some alternatves that have not receved wde attenton. In secton 1.3, we dscuss the research questons ths work hopes to address and hghlght the mportant concepts at the root of these questons. Secton 1.4 dscusses the outlne of ths paper. 1.1 Background Instead of a natonwde network, n the past, state and local publc safety agences have deployed ther own wreless communcatons systems. There are currently more than 50,000 state and local publc safety agences usng moble rado systems for wreless communcaton n the Unted States [3]. These agences employ approxmately 1.1 mllon frst responders [4]. However, the majorty of these agences are relatvely small, havng fewer than 50 users [3]. A US spectrum polcy of allocatng spectrum ndvdually to these agences has led to that spectrum beng substantally fragmented and allocated across 10 bands 1 rangng from 20MHz to 4900MHz [5] [6] [7] [8]. Havng many small agences deploy ther own systems has substantally ncreased the cost of the exstng nfrastructure whle smultaneously makng neffcent use of the spectrum [1]. In addton, the lmted coordnaton between agences when deployng communcaton systems and the lack of a wdely adopted techncal standard [9] [10] has led to exstng systems that are prone to falure when needed most [1]. Unfortunately, several recent tragedes 2 have been 1 Appendx A summarzes the federal and non-federal publc safety spectrum allocatons. 2 Perhaps the most tragc example s the loss of lfe durng the rescue efforts n the aftermath of the World Trade Center attacks on 9/11. The lves of 121 frefghters were lost because they were unable to receve an evacuaton - 1 -

3 drectly attrbuted to falures n the exstng publc safety communcatons nfrastructure [11] [12]. Interoperablty s the term gven to the ablty to communcate across agences, and a lack of whch s a commonalty shared by those recent tragedes. However, n addton to the largescale dsasters, a lack of nteroperablty s an ssue that publc safety agences must deal wth on a routne bass [13]. Whle nteroperablty problems have receved consderable study [14] [15] [16] [17] and recent efforts have been focused on reducng these nteroperablty ssues [18] [19] [20], even f these ssues were solved entrely, the majorty of exstng publc safety communcaton systems would stll be lmted n functonalty to narrowband voce, use more spectrum and cost more to mantan than a sngle network shared by all publc safety agences [1]. A natonwde publc safety wreless network avods many of the shortcomngs of the prevous polcy [2]. Instead of plannng thousands of systems ndependently, there s a sngle network to be desgned and deployed. By combnng these users nto a sngle pool, spectrum can be allocated and used much more effcently and techncal nteroperablty ssues are nherently solved by the use of a sngle technology on the network. Addtonally, buldng a new natonwde network presents an opportunty to deploy a broadband system whch can ntroduce data capabltes such as streamng vdeo and nternet access to users who prevously had to rely on voce-only systems. 1.2 Proposals There a two fundamentally dfferent proposals for the creaton of a natonwde publc safety wreless network, a proposal for a system that would serve only publc safety users and a publcprvate partnershp that would serve both commercal and publc safety users on the same network [2]. An example of a proposal for a publc-safety-only network s the Integrated Wreless Network (IWN). Ths s a proposal by the US Departments of Justce, Treasury, and Homeland Securty for a natonwde wreless system that would serve up to 80,000 federal publc safety users [21]. As t currently stands, ths network wll provde msson crtcal voce servce across the naton [22] but the technology 3 used wll not support broadband data applcatons [23] [24]. It s expected that the network wll use spectrum from the federal allocatons at 160MHz and/or 400MHz as most of the agences that wll use ths network have ther exstng land moble rado (LMR) operatons concentrated n these two bands [25]. It s possble that by expandng ths system to support broadband data applcatons as well as serve local and state publc safety users there are potental cost savngs and spectral effcency gans as compared to ndependently buldng two natonwde networks to support these user groups separately, as dscussed n [26]. Currently, extendng IWN does not appear to be a proposal that polcymakers are serously consderng and ths paper studes whether or not t deserves ncreased consderaton. In addton to proposals for a publc-safety-only network lke IWN, there have been proposals for a publc-prvate partnershp network that would serve both publc safety and commercal users call ssued over polce rado [11]. However, ths s not an solated problem: Hurrcane Katrna n 2005 [12], the Oklahoma Cty bombng n 1995 and the frst attack on the World Trade Center n 1993 [6] all demonstrate the shortcomngs of the exstng publc safety wreless communcaton systems. 3 It s planned that IWN wll be based on the Project 25 (P25) technology standard. Phase I of ths standard operates on 12.5 khz channels enablng voce or 9.6kbps data servce [23] [24]

4 [27]. A key motvaton for a publc-prvate partnershp s the observaton that a majorty of the tme, publc safety users do not use all of the avalable capacty on ther wreless systems [28] [29] [2]. Ths realty s due to the fact that publc safety communcaton systems are desgned for worst-case capacty demand scenaros but, thankfully, most of the tme these large-scale emergences are not takng place. Ths mples that f a commercal and publc safety entty were to share spectrum, a majorty of the tme the commercal partner could use some of the publc safety spectrum to serve commercal subscrbers whle allowng the publc safety partner access to both the publc safety and commercal spectrum n the rare emergences when t s needed. In August 2007, n response to an nnovatve proposal [30] n the US for a publc-prvate partnershp between a commercal wreless carrer and state and local publc safety agences, the FCC desgnated a 10MHz porton of the 700MHz spectrum band specfcally for publc safety broadband use. Most notably, ths spectrum was lcensed natonwde to a sngle representatve of publc safety [27] as opposed to ndvdually to each state and local agency. Addtonally, the FCC created a 10MHz commercal lcense for the spectrum adjacent to the publc safety allocaton, whch was auctoned n February 2008 [31]. The wnner of that aucton would have been oblgated to buld a natonwde publc-safety-grade network on the 20MHz of combned spectrum to be shared by publc safety and commercal users [27]. Ths was done n an attempt to have a commercal entty fund and buld out a publc-safety-grade network n exchange for dscounted access to spectrum. Ths aucton concluded wthout a wnnng bdder emergng, a fact that has been wdely attrbuted to the consderable uncertanty about the requrements that would be placed on the network 4 [32] [33]. Wth no wnnng bdder emergng from the aucton, the FCC s reexamnng the rules that were attached to the commercal block of spectrum and s consderng changes before t s reauctoned. In ths paper we consder both a potental publcprvate partnershp on 20MHz of spectrum n the 700MHz band and a publc-safety-only network on 10MHz of spectrum n the same band. 1.3 Research Questons Whle these new proposals represent large steps n a new drecton for publc safety communcatons, many questons reman unanswered. Ths work hopes to nform the current debate by addressng the followng fundamental questons about the cost of a natonwde network. Frst, what wll a natonwde wreless communcaton system for publc safety cost? Prevous cost estmates vary consderably, and ther results are hard to assess because they tend to be unclear about the methods used and assumptons made [34] [35] [36] [37]. Ths paper presents an extensble model that s used to compare the number of cells requred by current proposals for a natonwde network. Although the model s not statc n that further refnements are expected over tme, the fact that all assumptons are transparent makes ths a valuable tool for polcy assessment. Ths paper focuses on the number of cell stes requred because the deployment and operatng costs of a network are roughly proportonal to the number of cell stes requred. 4 The FCC left many of these requrements to negotatons between the commercal partner and publc safety lcensee. Pror to the aucton, the publc safety lcensee dd release a document of desred system desgn requrements [33] however these requrements were stll negotable

5 When consderng a publc-safety-only scenaro, we compare the cost of each proposal to the cost savngs assocated wth mantanng and upgradng exstng publc safety nfrastructure. When consderng a publc-prvate partnershp scenaro, we compare the cost predcted by our model to ndustry estmates of the cost of a publc-prvate partnershp. We also compare the predcted number of cell stes requred for a publc-prvate partnershp to the number of cell stes requred for a publc-safety-only network and estmates of the number of cell stes deployed n exstng commercal systems. The fndngs n ths paper represent the ntal results of our model and future work wll examne addtonal proposals and network desgn requrements. Second, what system characterstcs have the largest mpact on the number of cell stes requred for each of the current proposals for a natonwde network? Many factors mpact the requred number of cells for a network and these factors often dffer between typcal publc safety and commercal wreless systems. In partcular, the followng system factors are studed n ths paper: the amount of area covered, the frequency and bandwdth of the spectrum allocaton, the amount of communcatons capacty requred on the network, and the techncal requrements of a publc-safety-grade network. When consderng the coverage area of a network, commercal cellular systems are bult so that most areas where payng customers resde and travel are served whle publc safety users need coverage wherever emergences can occur. Ths s mportant because n a wreless system, the number of cells requred s hghly dependent on the total area covered by the system. For ths reason, we determne the area currently covered by exstng publc safety systems and nvestgate the mpact on the number of cell stes requred as the fracton of US populaton and land area covered by a new natonwde system s vared. We also examne how costs change and how the relatve merts of a publc-safety-only network and a publc-prvate partnershp may dffer dependng on whch and how much spectrum the FCC and/or NTIA make avalable. The number of cells requred n a wreless network s dependent on the sze of the cells n the network and the sze of a cell s dependent on the frequency and bandwdth of the spectrum used. As the frequency ncreases, the sze of a cell tends to decrease. Meanwhle, as the bandwdth s ncreased, the capacty that a cell can support ncreases. For the proposals studed n ths paper, we wll be concerned wth frequences between 160MHz and 700MHz wth bandwdth allocatons between 7.5MHz and 20MHz. The capacty requred n a cell by commercal cellular users and frst responders s dfferent. These capacty dfferences can be seen at the ndvdual user level where frst responders may requre hgher data rate applcatons lke vdeo, and n the aggregate, when many frst responders must respond to the same emergency where they are concentrated wthn a sngle cell. Therefore, the number of cell stes requred for each proposal s studed for a varety of capacty requrement scenaros. When desgnng a wreless network, commercal cellular users and frst responders typcally have dfferent requrements (e.g. the avalablty of a wreless sgnal wthn buldngs or the relablty of a sgnal n the coverage area), so n a shared publc-prvate network t may be necessary to determne where on that contnuum a network planner should desgn. However, t s mportant that the desgn s not a compromse that s unnecessarly expensve for commercal - 4 -

6 users and nadequate for publc safety. Thus, a publc-prvate partnershp s a tradeoff (.e. wth the advantage of sharng capacty and the dsadvantage of dssmlar requrements) and we examne the mpact of these dssmlar requrements. 1.4 Paper Layout In secton 2 of ths paper, we ntroduce the framework of the model we developed to calculate the number of cell stes requred by each proposal consdered. Ths model takes several varables as nputs and secton 3 dscusses the approprate numercal values to use for the nputs when desgnng for publc-safety-only and publc-prvate partnershp network. In Secton 4, a Geographc Informaton System (GIS) model s ntroduced and then used to calculate the area currently covered by the exstng publc safety wreless nfrastructure. Ths result s then used as an estmate of the area a natonwde, broadband, publc safety wreless network should serve. Secton 5 provdes a revew of the proposals studed, the numercal value chosen for the nputs that are specfc to each proposal, and a summary of all the numerc values used as nputs to the model. Secton 6 provdes the results of the model wth an estmate of the cost for each proposal studed and nvestgates how results change as the nput values are vared. Fnally, secton 7 provdes a dscusson of our conclusons. 2 Model Development In ths secton, we ntroduce the framework of the extensble model we developed to calculate the number of cell stes requred by a publc-safety-grade network under a varety of condtons. Ths secton begns wth an overvew of the model layout n secton 2.1, whch ntroduces the mportant concepts used and dscusses the major assumptons made. Sectons descrbe the major components of the model n detal: estmatng capacty requred n a cell, estmatng the mnmum sgnal power requred at a recever, estmatng factors whch affect the receved sgnal power between transmtter and recever, and estmatng the sze of a cell based on sgnal power lost between transmtter and recever. Secton 2.6 summarzes how all four components relate to form a system of equatons that are solved for the nput values dscussed n secton Overvew The most cost-effectve desgn for the proposals we study of a natonwde publc-safety-grade broadband wreless system s based on a cellular archtecture. Costs n a cellular archtecture are hghly dependent on the number of cells. Therefore, to estmate the cost of a network, we frst develop a model to predct the requred number of cells for that network. We calculate the expected number of cell stes per regon (.e. zp code) as follows. Let C be the expected area per cell f populaton densty were unform, and equal to the populaton densty n regon. Let A be the area of regon. We assume that the expected number of cell stes n regon = A / C. The populaton densty n regon s determned usng natonwde zp code level 5 populaton statstcs [38]. Expected cell sze depends on populaton densty for several reasons ncludng the fact that the capacty requred n a cell and the approprate propagaton model for a cell are dependent on populaton densty. The model calculates the expected area per cell n each regon n 4 steps: by calculatng the capacty requred n a cell, then by 5 The U.S. Census Bureau records the populaton, populaton densty, and geographc area data at several levels of granularty ncludng Census Blocks (8+ mllon cover the U.S.), Zp Code Tabulaton Areas (30+ thousand cover the U.S.) and Countes (3+ thousand cover the U.S.)

7 calculatng the mnmum receved sgnal power requred for the capacty requred, then by calculatng the maxmum amount of sgnal power that can be lost n the path between transmtter and recever, and fnally by calculatng the radus of a cell based on the maxmum amount of power that can be lost n the path. At a hgh level, these calculatons requre that we frst defne the capacty requred n a cell as a functon of frst responder densty. We show n secton 3 that frst responder densty s a lnear functon of populaton densty. Next, we show that the mnmum power receved from each moble devce at the base staton n each cell (.e. the recever senstvty) s a functon of the capacty requred n that cell. Then, we use a lnk budget to determne the maxmum amount of strength that the sgnal s allowed to lose as t travels from the handset to the basestaton (.e. the maxmum allowable path loss or PL). A lnk budget takes nto account the power of the transmtted sgnal, the mnmum sgnal power requred at the basestaton, ncreases n sgnal power due to antennas, and decreases n sgnal power due to factors such as outdoor obstacles n the sgnal path and the sgnal havng to penetrate walls. Fnally, we use a propagaton model that dfferentates urban, suburban, and rural regons to calculate the radus of a cell. Ths model accounts for the maxmum allowable path loss, frequency of operaton, and heght of the basestaton antenna. In practce, every sngle cell wll have a unque coverage area determned by a number of localzed factors; however, we avod layng out exactly where every cell tower n the country must st by assumng that the dstrbuton of populaton denstes of cells s the same as the dstrbuton of populaton densty wthn occuped zp codes. Zp code level granularty appears to be reasonable gven that, as we wll see n Secton 6, the number of cells natonwde s comparable to the number zp codes. In ths work, when technology dependent numercal values are requred for analyss, we use numbers that are consstent wth the CDMA2000 evoluton broadband wreless standard. CDMA2000 s a well developed and wdely deployed 3G standard, whch would be one potental canddate technology for a natonwde broadband wreless system. Although the technology actually used n a next-generaton system may or may not be CDMA2000, t s a reasonable bass for cost estmates as t has been consdered n other analyss of a broadband publc safety system [37] [39] [40] and has recently been chosen as the technology standard for a cty-wde publc safety broadband deployment [41]. Consstent wth a CDMA2000 network, we have assumed that the bandwdth allocated under each proposal s dvded nto 1.25MHz channels wth traffc dstrbuted equally across the channels [42]. As s typcally done for CDMA systems, our model consders a network wth a unversal frequency reuse or a frequency reuse factor of 1, meanng that every cell can operate on each channel. Ths paper focuses on the uplnk as t s assumed to be the lmtng lnk n determnng the sze of a cell as s usually the case n a CDMA system where the moble devces have lower transmt power lmts than basestatons and cochannel nterference from other mobles operatng on the same channel s present at the basestaton [42]. To lmt cochannel nterference, among other reasons, cells are typcally sectorzed n a CDMA system. We have assumed that all cells n our network have 3 sectors per cell. We further assume that the uplnk s perfectly power controlled as s typcally assumed when analyzng CDMA systems [42] [43]

8 Perfect power control means that the power a moble devce transmts at s controlled so that the power receved at the basestaton from mobles communcatng wth that basestaton s no greater than that necessary for adequate communcatons. Some of these parameters wll be vared n future work. When calculatng cell area as a functon of radus, one must account for the fact that cells overlap, typcally by 10 30% [42] [44]. We have assumed an overlap of 17%, whch would be consstent wth cells that are hexagonal as opposed to crcular. Further, we assume no fault tolerance n the desgn of ths publc safety network. Ths means the network s desgned knowng that the loss of any cell ste means a loss of servce n some area. Ths desgn s no worse than what publc safety has today, but the creaton of a natonwde publc safety network presents an opportunty to add fault tolerance [2]. Fault tolerance could make a bg dfference n the response efforts for dsasters smlar to Hurrcane Katrna where the communcatons nfrastructure s partally destroyed durng the dsaster. Future work may consder the tradeoff between the cost of a network and the addton of fault tolerance. 2.2 Capacty Model For each cell, t s necessary to calculate the capacty requred as ths capacty can have a sgnfcant mpact on the cell s sze. In commercal networks the capacty requred n a cell s well understood and network planners often have hstorcal usage data to consder when desgnng a network. Unfortunately, wth the lmted deployment of broadband data networks for publc safety use n the US, there s a lack of emprcal evdence for capacty requred n a publc-safety-only or publc-prvate partnershp cell. There s no wdely accepted model of capacty requrements for publc safety, whch s a serous problem for cost estmates, and more generally for polcy formulaton. Further work s needed n ths area. So, n the absence of an establshed model, we wll suggest one vable possblty and use t as the bass for analyss. Ths secton develops a general model for the capacty requred n a cell by publc safety users on a broadband wreless network whle the actual numercal values used n our smulatons are dscussed n secton 3. We desgn the system to accommodate two sources of publc safety traffc present durng a large-scale emergency. One source of traffc wll come from publc safety personnel who are respondng to the large-scale emergency. The second source s routne traffc whch s due to routne communcatons actvty and s not part of the emergency response. Because the largescale emergency may occur when routne traffc s at ts peak, the total publc safety capacty requred n a cell s the sum of the peak capacty requred by each of these two traffc sources. Consderng frst the emergency traffc, we desgn the system such that capacty wll be suffcent for a large-scale emergency that s localzed, even n the worst-case for a localzed emergency, whch would occur f the emergency response takes place entrely wthn a sngle cell and n the worst part of that cell (.e. the edge). We are desgnng for a large-scale emergency that s relatvely localzed n nature (e.g. a plane crash, a large buldng fre, a terrorst bombng, etc.). By desgnng the system such that every cell can accommodate ths knd of localzed dsaster, t seems lkely that a dsaster such as an earthquake or hurrcane that s spread across many cells wll also have suffcent capacty, but ths s beyond the scope of our model

9 Ths approach s smlar to the exstng work on capacty requrements for emergency response whch also focuses on localzed events [45] [46]. The magntude and nature of a large-scale emergency as well as the response to that emergency may vary dependng upon where n the US the event occurs. More specfcally, the capacty requred n a cell due to an emergency response may be dependent on the number of frst responders n the vcnty of that cell who are avalable to respond. Smlarly, the nature of the emergency may be such that areas wth hgher populaton densty may need even more frst responders to protect the affected populaton. As wll be dscussed n greater detal n secton 3, we have observed a lnear relatonshp between populaton densty and the densty of frst responders n the US. The followng fgure provdes a graphcal representaton of the model developed of the capacty requred n a cell as a functon of the populaton densty n that cell. C A P A C I T Y R E Q U I R E D Emergency Capacty Requred n a Cell: a functon of Populaton Densty Maxmum Capacty Requred by any Emergency Response Capacty Requred by an Emergency Response Mnmum Capacty Requred by any Emergency Response Populaton Densty Fgure 2.1a: A plot of the capacty requred n a cell as a functon of the populaton densty n that cell. In the fgure above, we have desgned the system such that even the most rural cells (.e. cells wth a populaton densty of zero) requre a non-zero capacty. Ths appears reasonable consderng the fact that large-scale emergences can occur anywhere (e.g. a plane crash). In ths case, the response s composed of publc safety personnel based outsde of the cell, respondng to the emergency. Therefore, we establsh a baselne value whch represents a mnmum amount of emergency response capacty requred n every cell and we assume t s equal to 10% of the capacty requred n the most urban cell. Ths appears reasonable gven that the response to large-scale emergences that occur n extremely rural areas, such as the response to the crash of Flght 93 on 9/11, have an ntal response approxmately 10 20% the sze of the ntal response to large-scale emergences n urban areas, such as the 9/11 terrorst attack on the Pentagon [47] [48] [49]. As the populaton densty of a cell ncreases from zero, there wll be a nonzero number of frst responders predcted n that cell. These frst responders wll respond to the emergency n addton to the responders that wll come from outsde of the cell. In ths porton of our model, the capacty requred n a cell ncreases lnearly wth populaton densty from the baselne mnmum value up to a threshold maxmum value. The threshold value s the upper bound on the amount of capacty requred n any cell. Ths bound s due to an expected - 8 -

10 lmt on the number of frst responders who could effectvely partcpate n an emergency response effort that s localzed wthn a cell. The value of ths threshold capacty s dscussed n secton 3. The second source of publc safety traffc s routne traffc. We assume that routne traffc s the same whether or not a large-scale emergency s occurrng. We propose that the routne capacty requred n a cell vares lnearly wth populaton covered by the cell (whch we calculate as the product of the populaton densty n the cell and the sze of the cell). In addton to publc safety traffc, any publc-prvate partnershp wll need to accommodate traffc from commercal users. We assume that the system wll be desgned so that there s suffcent capacty to meet the expected needs of commercal users and routne publc safety traffc when there s no large-scale emergency gong on, and capacty s suffcent to meet the needs of publc safety durng large-scale emergences. Durng a large-scale emergency, any capacty not needed by publc safety wll be avalable to commercal users, but there s no guarantee that there wll be suffcent capacty to carry all ther traffc. We assume that the commercal capacty requred n a cell vares lnearly wth the populaton covered by the cell. We assume that the market penetraton (.e. the fracton of the populaton covered by a cell that subscrbes to the servce) s constant across the naton. The fgure below shows the relatonshp between the capacty requred n a cell by commercal subscrbers and routne traffc and the populaton covered by the cell. C A P A C I T Y Routne and Commercal Capacty Requred n a Cell: a functon of Populaton Covered Capacty Requred by Commercal Subscrbers R E Q U I R E D Capacty Requred for Publc Safety Routne Traffc Populaton Covered Fgure 2.1b: A plot of the capacty requred n a cell as a functon of the populaton covered n that cell. 2.3 Recever Senstvty In a wreless system, the strength of a sgnal decreases as t travels. If the sgnal s too weak when t reaches the recever, the transmtter and recever cannot sustan communcatons. Ths threshold s often called the recever senstvty, the exact value of whch has a large mpact on the sze of the cell. More specfcally, the recever senstvty n a wreless system s defned as - 9 -

11 the mnmum acceptable sgnal level at the recever whch wll support sutable operaton at a gven datarate. In a CDMA system, the recever senstvty s dependent upon the nstantaneous nose and nterference envronment (whch s related to the capacty requred) and the rato of bt energy to nose that s requred to acheve the transmt datarate desred by the user. In lnk budget analyss, recever senstvty s often stated as a constant value ndependent of desred user datarate and the nterference envronment. Ths may be reasonable n some systems where capacty requred s constant across cells and desred user datarate s constant. But ths s not true n our model, so we derve the followng equatons for recever senstvty n a CDMA envronment based on prevous work by [42] [43]. Secton defnes the mnmum acceptable receved sgnal power as a functon of nterference. Secton then determnes the recever senstvty the system must be desgned for and expresses the recever senstvty as a functon of capacty requred. Secton dentfes how ths recever senstvty relates to frst responder densty and populaton densty Receved Sgnal Power and Interference The ablty of the th user to operate at a desred datarate s dependent on the strength of the transmtted sgnal when t s receved at the basestaton, the bt energy to nose rato requred for adequate operaton at that datarate, and the nose and nterference envronment at the basestaton. In a CDMA system, we can express the mnmum acceptable receved sgnal power, s, for the th user to support the desred datarate, R, n the followng form [42]: E b R s ( + ISC, + IOC, ) N η (2.3-1) o W Where: s = mnmum acceptable receved sgnal power of the th user n the sector (W) W = channel bandwdth (Hz) R = datarate or nformaton bt rate desred by the th user n the sector 6 (bps) (E b /N o ) = bt energy to nose rato requred for operaton at datarate R η = envronmental nose (predomnantly thermal nose) power at the recever (W) I SC, = cochannel nterference power to the th user due to users from the same sector (W) I OC, = cochannel nterference power to the th user due to users from other sectors (W) As dscussed n secton 2.1, the equaton above assumes that the spectrum allocated s dvded nto channels of bandwdth W, wth traffc dstrbuted equally across these channels, and each cell dvded up nto equal szed sectors Recever Senstvty and Capacty For network plannng purposes, the recever senstvty of nterest s that of the most demandng user n the sector. Snce equaton (2.3-1) gves the mnmum acceptable receved sgnal power, 6 The rato of W/R s often referred to as the processng gan

12 s, n the sector for the th user, we defne the recever senstvty of a sector as the largest value of s n the sector. Snce we know the datarate desred by the most demandng user n the sector, the only unknowns n the recever senstvty equaton are the nterference terms, I SC, and I OC,. In a CDMA system, there s a relatonshp between the total capacty requred on the channel and the nterference on the channel. An ncrease n capacty s due to ether the number of users that operate on the channel ncreasng or the datarate of users on the channel ncreasng. In ether case, the power transmtted on the channel and receved at the basestaton ncreases. Snce t s a shared channel, a user consders all of the power receved at the basestaton due to other users as nterference. Thus, the measure of capacty requred also serves as a measure of nterference. In a publc-safety-only system, we express same-sector nterference, I SC,, as the sum of all the sgnals receved per channel at the base staton due to the two sources of publc safety traffc 7. Addtonally, t s typcal n a CDMA system to state the nterference due to other sectors, I OC,, as a fracton of same sector nterference, I SC, [42]. However, we have desgned for an emergency response that s localzed wthn a cell, and we assume that all other cells are only carryng routne publc safety traffc. Thus, I OC,, s assumed to be a fracton of nterference due to routne traffc and we calculate the recever senstvty, s, as follows: βmax η s n m 1 βk + (1 + fract) βl k = 1 l = 1 (2.3-2) Where: E b R Eb R β + 1 and s a measure of the capacty that user requres. No W No W β ncreases wth user s need for greater data rates, and wth the energy to nose rato requred to sustan any gven data rate. n k = 1 m l= 1 β Is the sum of the β terms for all n users that are respondng to a localzed emergency k l wthn a sector per channel. Ths term s a measure of the capacty requred by the emergency response per channel n a sector. β Is a measure of the capacty requred by routne traffc per channel n a sector. fract Is a fracton of nterference due to routne traffc whch represents the other sector nterference. β MAX Is the largest β value of any actve user. 7 A more detaled dervaton s provded n Appendx B

13 In a smlar manner as above, we derve the followng equaton for the recever senstvty of a user n a publc-prvate partnershp network: βmax η s (2.3-3) p m 1 (1 + fract) βq + βl q= 1 l = 1 Where: p q= 1 β Is a measure of the capacty requred by commercal subscrbers per channel n a sector. q As dscussed n secton 2.2, the capacty requred by routne traffc s a functon of frst responders served and the capacty requred by commercal subscrbers s a functon of populaton served; both of whch vary by cell. Consderng that the capacty requred s dstrbuted across the avalable channels n a sector, we defne the followng equatons for the jth cell: n β k = 1 m l = 1 p q= 1 k βsum / Num ( A Sect) (2.3-4) β l hexagon, j / ρfr, j ρβrt / Num (2.3-5) ( A Sect) β q hexagon, j Pen ρpop, j ρ SUB / Num (2.3-6) / β Where: Num = the number of uplnk channels avalable n the sector β = the capacty requred for the response to a localzed emergency per sector SUM A, = the area of the jth cell hexagon j Sect = the number of sectors n the jth cell ρ = the frst responder densty n the jth cell FR, j ρ βrt = a measure of the capacty requred per frst responder due to routne traffc. ρ POP, j = the populaton densty n the jth cell Pen = the market penetraton of the provder as a fracton of populaton covered ρ = a measure of the capacty requred per commercal subscrber on the network. βsub As dscussed n secton 2.2, a publc-prvate network wll be desgned such that the capacty s suffcent to meet expected needs of commercal users when there s no large-scale emergency, and such that the capacty s suffcent to meet the needs of publc safety durng large-scale emergences. Thus, the recever senstvty n the jth cell becomes the larger of the followng: s PUBLIC _ SAFETY, j β η (2.3-7) MAX ( β / Num + (1 + fract) ( A / Sect ρ )/ Num) 1 SUM hexagon, j FR, j ρ β RT

14 s COMM, j β η (2.3-8) MAX [( A / Sect (1 + fract) )( Pen ρ ρ ) + ( ρ )/ Num] 1 hexagon, j βsub Pop, j FR, j ρ β RT 2.4 Lnk Budget In a wreless channel, a lnk budget can be used to account for all of the factors that ncrease or decrease the strength of a transmtted sgnal at the recever. These ncreases are typcally referred to as gans whle the decreases are referred to as losses. These gans and losses can greatly mpact the sze of a cell. Ths secton wll dentfy and brefly dscuss the gans and losses whch should appear n a lnk budget approprate for a wreless network whle the actual numercal values of these terms approprate for a publc-safety-grade network are dscussed n secton 3. For a publc-safety-grade cellular system, the power receved at a recever, S, s equal to the ntal transmt power of the sgnal, EIRP, plus the gan of the recevng antenna, G RX, mnus the summaton of any losses, L n the channel (wth all terms expressed n decbels or db 8 ). S = EIRP + GRX L {n db} (2.4-1) We can decompose the summaton of losses nto two man components PL and LM. PL represents the path loss, whch s the dstance (between transmtter and recever) dependent component of loss due to the sgnal beng attenuated as t propagates through space. Path loss s typcally the largest loss n the lnk budget and wll be studed n more detal n secton 2.5 where an approprate propagaton model s dscussed. LM represents all of the loss margns to account for dstance-ndependent components of loss ncludng the margns to ensure relable coverage ndoors and outdoors and mscellaneous margns to account for losses due to mplementaton ssues lke cablng and connector losses and scenaro losses due to recever orentaton. The lnk budget used for ths model s based on the related work n [5] [42] [50] [51] [52] and gven by the followng equaton whch has been solved for path loss, PL: PL = EIRP + G L L L L S {n db} (2.4-2) RX IMPLEMENT SCENARIO RELIABLE BUILD Where: EIRP Effectve Isotropc Radated Power (dbm) S Recever Senstvty (dbm) G RX Recever Antenna Gan (db) L IMPLEMENT Recever Implementaton Losses (db) L RELIABLE Shadowng + Fast Fadng Margn (db) L BUILD Buldng Penetraton Margn (db) L SCENARIO Scenaro Loss Margn (db) Expressng transmt power as EIRP n the lnk budget above s a common way of combnng any gans or losses nternal to the transmtter, such as transmtter antenna gan, wth the power wth whch sgnals are transmtted. Snce the transmttng antenna gan s ncluded n ths term, the 8 X n db = 10*log 10 (X) n absolute unts

15 only gan above s that of the recevng antenna. The recevng antenna gan s a measure of how effectvely the antenna captures more power n certan drectons than n others. The relablty margn determnes how relable communcatons are wthn the outdoor coverage area of the cell. Ths margn s necessary to account for a sgnal beng shadowed by an obstructon n the path from transmtter to recever. Addtonally, ths margn accounts for the possble fast fadng of a sgnal due to multpath effects wheren a sgnal nterferes destructvely wth tself as t takes multple paths to the recever. Smlarly, the buldng margn determnes how relable communcatons are wthn ndoor envronments. Ths margn s necessary to account for a sgnal beng attenuated as t penetrates buldng walls. The mplementaton margn ncludes any losses due to the sgnal beng attenuated as t travels through cablng between the recevng antenna and basestaton. Ths margn also ncludes any losses due to msmatches and connectors at the basestaton. The scenaro margn estmates losses due to recever orentaton and polarzaton msmatches as well as sgnal obstructon due to the body of the user. 2.5 Propagaton Model Path loss s the reducton n strength of a wreless sgnal as t travels through space. Path loss depends on many factors ncludng frequency, antenna heght, termnal locaton relatve to obstacles and reflectors, and lnk dstance, among other factors. Most mportantly, t s ths dependence on lnk dstance that has a consderable effect on the sze of a cell. Ths secton presents the propagaton model that wll be used to relate path loss to cell radus whle the actual numercal values used n the model are dscussed n secton 3. A propagaton model s typcally used when estmatng the medan path loss n a wreless network. For ths paper, the Hata propagaton model [53], a model based on Okumura s emprcal measurements of path loss [54], was chosen as t s arguably the most commonly used model n the wreless ndustry for large-scale network plannng. Ths model makes relatvely accurate predctons whle requrng only mnmal envronment specfc nformaton [5] [43]. Whle propagaton models exst [55] [56] that are more precse at predctng the path loss between a specfc transmtter-recever par by takng nto account locaton-specfc factors whch affect a rado sgnal our goal s not to determne the cell sze for a specfc locaton. Rather, our work s nterested n the sze of a cell averaged over many smlar locatons and the Hata model s suffcent snce t s assumed that the locaton-specfc devatons wll tend to average out. The equatons used n the Hata model are dfferent for urban, suburban, or rural regons. Ths classfcaton s commonly made based on populaton densty [57]. There s no unversally accepted populaton densty threshold whch separates these categores. The dvdng lne between rural and urban vares from fewer than 50 to 400 people per square klometer and the dvdng lne between suburban and urban vares from fewer than 1,000 to 10,000 people per square klometer [37] [33] [57] [58]. We have defned rural as havng less than 100 people per square klometer and urban as havng more than 1900 people per square klometer as these values are nlne wth the values used n smlar analyss [57]

16 The Hata model, gven n the followng equaton [5], predcts medan path loss (PL) based upon the frequency of the wreless sgnal (f), the heght of the base staton (h b ), the heght of the moble rado (h m ), and the dstance of separaton between the transmtter and recever (r): PL = log 10 ( f ) log 10 ( hb ) a( hm ) + ( log 10 ( hb )) log ( r) K (2.5-1) Where: Moble Adjustment: a(h m ) = 1.1 log ( ) 0.7) h (1.56 log ( f ) 0.8) ( 10 f m (log10 ( )) log10 ( f ) (log10 ( / 28)) + 5. f ; Rural Urban Adjustment: K = f 4 ; Suburban 0; Urban Ths model s vald for the followng ranges of nput values: Path Loss: PL n db Frequency: f = MHz Radus: r = 1 20 km Moble Heght: h m = 1 10 m Base Heght: h b = m 2.6 Solvng the System of Equatons From the equatons establshed n the prevous three sectons, t s possble to predct the average radus of a cell n each regon. Pluggng the expresson for recever senstvty, equaton (2.3-9), and propagaton loss, equaton (2.5-1), nto the lnk budget developed n secton 2.4 yelds: K 1 K r = K K + K K K 5 log 10( ) + 10 log ( 2 + 3) {n db} (2.6-1) Where: K = EIRP + G L L L L 0 RX RELIABLE BUILD IMPLEMENT SCENARIO K 1 = β MAX η K = β Num for Emergency Traffc K K K 2 SUM / ( Ahexagon, j / Sect (1 + fract) )( Pen ρ β SUB ρpop, j )/ Num fract) ( A / Sect ρ ρ ) Num for Commercal Traffc 3 = ( 1+ hexagon, j FR, j βrt / 4 = log10( f ) log10( hb ) a( hm 5 = ( log10( hb )) (log10 ( )) log10 ( f ) (log10 ( / 28)) + 5. f Rural K = f 4 Suburban 0 Urban From equaton (2.6-1), t can be observed that the radus of a cell s dependent only on the frequency of operaton, heght of the base staton, effectve sotropc transmtted sgnal power, recevng antenna gan, mplementaton and scenaro losses, shadowng and fadng margns, buldng penetraton margns, as well as the datarate and requred bt energy to nose rato for the )

17 most demandng user n a cell n addton to the requred capacty and bt energy to nose rato for all actve users n a cell. In secton 3, the approprate values for each of these parameters wll be determned and dscussed n more detal. 3 Model Inputs In secton 2, we presented the equatons that our model s based on and each of these equatons s dependent on several varables. Ths secton wll dentfy the proper numercal values whch should be used when evaluatng these equatons for a publc-safety-grade wreless network. For many of the nputs studed, there s consderable uncertanty n the numercal value they should take. For these uncertan nputs, we present base case estmates whch are then vared n secton 6. Gven the range of values that may be approprate, the values chosen n the base case and the results they produce are not meant to be the fnal word on ths topc. Instead, we present the base case values to enable exploraton of the proposals presented n secton 1 but future work wll nclude addtonal analyss. Secton 3.1 presents the nput values for the equaton of recever senstvty presented n secton 2.3. Ths ncludes a regresson model to relate populaton densty to frst responder densty, base case values for the varous measures of capacty dscussed, and estmates of nose n the channel. Secton 3.2 dscusses the values to be used n the lnk budget presented n secton 2.4 whch ncludes values for transmtter power and recever antenna gan as well as the several losses present n the channel. Fnally, secton 3.3 consders the values used n the Hata propagaton model whch was descrbed n secton 2.5 and ncludes a dscusson of moble heght and basestaton antenna heght. 3.1 Recever Senstvty Inputs In secton 2.3, the expressons for recever senstvty n both publc safety and publc-prvate partnershp networks are gven as equaton (2.3-7) and (2.3-8). These expressons are dependent on the densty of users n the cell, the capacty requred by users n the cell and the nose envronment n the cell. In secton 3.1.1, we present a regresson model that relates populaton densty to frst responder densty. In sectons and 3.1.3, we present estmates of the several measures of capacty requred to calculate the recever senstvty for publc safety and commercal users respectvely. In secton 3.1.4, we present an estmate of other sector nterference as a fracton of same sector nterference. Fnally n secton 3.1.5, we present a model of nose power at the recever Regresson Model: Populaton Densty vs. Densty of Frst Responders In secton 2.2 we presented a model of capacty demanded n a cell that s a functon of the densty of frst responders n a cell. However, the zp code level dataset used n our analyss only provdes populaton densty statstcs. Ths secton descrbes the regresson model we developed to relate populaton densty to densty of frst responders. Our regresson analyss shows that the number of frst responders per area s roughly proportonal to populaton densty. The followng three equatons represent the lnear equatons that best ft a Metropoltan Statstcal Area (MSA) s populaton densty, ρ, versus that Populaton MSA s polce densty, ρ Polce, frefghter densty, ρ Fre, and emergency medcal personnel (EMS)

18 densty, ρ EMS, respectvely, based on 2005 employment [59], [60], [61] and census [62] data. Equatons (3.1-1), (3.1-2), and (3.1-3) have R 2 values of 0.85, 0.62 and 0.76 respectvely, suggestng that they all ft the data reasonably well. Addtonally, each of the parameter estmates s statstcally sgnfcant as summarzed n Appendx C. ρ Polce ρ = Populaton ρ Fre ρ = Populaton ρ EMS ρ = Populaton (3.1-1) (3.1-2) (3.1-3) Addtonally, we consder federal publc safety users on the network by defnng the followng equaton to calculate the total frst responder densty, ρ, n an area: ( + )( ρ + ρ ) ρ = 1 + (3.1-4) FR fed Polce Fre ρ EMS Where: fed = s the percentage of all publc safety users n the response to a large-scale emergency that work for a federal agency We chose a value of 8% for fed whch appears reasonable gven that there are about 80,000 federal publc safety personnel [22] as compared to the 1.1 mllon frst responders n the US [4] Publc Safety Capacty As dscussed n secton 2, we have desgned the network to accommodate the capacty requred by publc safety when respondng to a large-scale emergency. Ths publc safety traffc determnes the value of β SUM, β MAX, and ρ βrt used n the calculaton of recever senstvty n secton 2. The value of β SUM s determned by the traffc from publc safety personnel who are respondng to a large-scale emergency. The capacty requred by the user who operates at the hghest datarate determnes the value of β. The capacty requred to support routne traffc determnes the value of ρ. βrt MAX We consder three traffc scenaros for publc safety, wheren the natonwde network carres: 1. (Voce-Only) all publc safety voce traffc and nothng else, 2. (Data-Only) all publc safety data traffc, and no voce traffc, or 3. (Data and Voce) all publc safety traffc ncludng voce and data. The data-only scenaro would be approprate f publc safety agences contnue to rely on ther exstng systems, whle the voce and data scenaro would (eventually) allow publc safety to phase out ther exstng systems. Numerc values for β SUM, β MAX, and ρβrt must be estmated for each of these three traffc scenaros. None of these values are well known and we therefore consder a wde range of uncertanty n the estmates for each of the capacty parameters n our analyss. FR

19 In secton we estmate the numerc values for β SUM, β MAX, and ρ βrt n the Data-Only traffc scenaro. In secton we estmate the numerc values for β SUM, β MAX, and ρ βrt n the Voce-Only traffc scenaro. Secton estmates the numerc values for β SUM, β MAX, and ρ n the Data and Voce traffc scenaro. βrt β MAX, β SUM, and ρ βrt : Data-Only Frst responder usage of broadband systems s not well understood and prevous work has focused on estmatng the capacty necessary for an emergency response to hypothetcal emergency scenaros (.e. buldng fres, tran accdents, terrorst attacks, etc.) [45] [46] [63] [64] wth only lmted nformaton avalable about traffc on exstng networks [65]. As a base case estmate, we assume that the worst-case scenaro for data traffc s the hypothetcal scenaro consdered by the Spectrum Coalton [45]. They consdered a large-scale emergency n Washngton D.C. wth an emergency response that ncluded federal, state, and local publc safety personnel usng vdeo and other data applcatons to montor and coordnate the response to a bologcal and chemcal terrorst attack. Ths work concluded that the followng data communcatons must be supported n the uplnk of the busest sector n the network: two-way vdeo, mappng/locaton trackng, sensor nformaton, web access, emal access. For ths scenaro, the applcatons used, estmates for the peak number of actve users that use each applcaton, the fracton of these users that are communcatng at the same tme, and the datarate each of these applcatons requre are summarzed n the table below. In ths paper, we make no judgment on the accuracy of the numbers presented by the Spectrum Coalton. We chose to use these values because the Spectrum Coalton work s among the most detaled work to date on the data capacty requrements of publc safety. Further research and dscusson on these numbers s requred. Applcaton Requred Datarate [kbps] # of Users Fracton of Users Actve Actve Users Capacty Requred [kbps] Sensors Vdeo Locaton Web Emal Table 3.1: A summary of factors consdered by the Spectrum Coalton n ther estmate of the capacty requred durng the response to a large-scale emergency n the busest sector of the network [45]. Usng the capacty values from ths table and the correspondng value of requred bt energy to nose ratos n a CDMA system [42] [66] we calculate the βsum and β MAX values as 1.6 and 0.34 respectvely based on the equatons gven n secton 2.3. For routne traffc n the base case, we assume the mean amount of data uploaded per frst responder per hour worked = 2 MB. Because usage vares consderably from N Up _ MB / HourWorked

20 one hour to the next, we assume that busy perod traffc rate s K tmes the mean traffc rate. A value of K= 4 would be approprate f busy perods occurred 20% of the tme, and carred 80% of the traffc, and f all of the busy perods were roughly the same. Thus, by the equaton below, TP DATA _ RT = 4.2 kbps/frst responder and ρ βrt = 5.2E-3 per frst responder. TP = K N 40 / ( Up _ MB / HourWorked HoursWorked / Week ) ( Hours / Week Sec / Hour ) bts / Byte DATA _ RT β MAX, β SUM, and ρ βrt : Voce-Only In the base case, we assume that n the worst-case Voce-Only scenaro, the same number of responders that were predcted to respond to the emergency n the Data-Only scenaro would be present. These responders use only voce communcatons, and at the busest tme, 5% are actve. Ths means that there 32 voce streams actve smultaneously n the worst-case and we calculate the β SUM and β MAX values as 0.94 and 0.03 respectvely. For routne traffc n the base case, we assume the mean percentage of tme that a frst responder spends talkng whle on duty = 1%. Because usage vares consderably from one hour to S % Talkng the next, we assume that busy perod traffc rate s K tmes the mean traffc rate, as we assumed wth data traffc. A value of K= 4 would be approprate f busy perods occurred 20% of the tme, and carred 80% of the traffc, and f all of the busy perods were roughly the same. Thus, by the equaton below, TP VOICE _ RT = 0.1 kbps/frst responder and ρ βrt = 2.8E-4 per frst responder. TP VOICE _ RT = K ( S% Talkng 40 HoursWorked / Week 9.6kbps )/( 168Hours / Week ) β MAX, β SUM, and ρ βrt : Data and Voce When both data and voce traffc are supported on the network, we have desgned for the possblty that peak hour for data concdes wth peak hour for voce. Thus the Data and Voce capacty requred s equal to the capacty requred when only data traffc s carred plus the capacty requred when only voce traffc s carred. The value of β SUM for Voce and Data s the sum of ts value n the Data-Only and Voce-Only cases whch s 2.5 n the base case. Smlarly the value of ρ n the Voce and Data case s equal to the sum of ts value n the Data-Only and βrt Voce-Only cases whch s 5.5E-3 n the base case. Meanwhle, the value of β MAX n the Voce and Data case s equal to the larger of the two values for β MAX n the Voce-Only and Data-Only cases whch s 0.34 n the base case Commercal Capacty In addton to publc safety traffc, a publc-prvate partnershp wll need to accommodate traffc from commercal users. The system wll be desgned so that there s suffcent capacty to meet the expected needs of commercal users when there s no large-scale emergency gong on, and capacty s suffcent to meet the needs of publc safety durng large-scale emergences. The fracton of populaton covered that subscrbes to the commercal servce determnes the value of Pen. The capacty requred to support commercal subscrbers on the network determnes the value of ρ. βsub

21 For commercal traffc n the base case, we assume the mean amount of data uploaded per commercal subscrber per month = 200 MB. Ths value appears reasonable N Up _ MB / Month consderng the average commercal cellular subscrber talks about 800 mnutes each month (whch s approxmately 50MB) and the usage of wreless data s ncreasng rapdly [67] [68]. Several major wreless provders have capped usage per subscrber at 5 GB per month [69] [70] [71], so we have assumed mean usage per subscrber s well under these caps. Because usage vares consderably from one hour to the next, we assume that busy perod traffc rate s K tmes the mean traffc rate. A value of K= 4 would be approprate f busy perods occurred 20% of the tme, and carred 80% of the traffc, and f all of the busy perods were roughly the same. Thus, by the equaton below, TP SUB = 2.4 kbps/subscrber and ρ βsub = 2.9E-3 per subscrber. There s consderable uncertanty n ths value and a large range of values are consdered n secton 6. TPSUB = K ( N Up _ MB / Month 8 bts / Byte )/( 720hours / month 3600sec/ hour ) The commercal capacty s a functon of the number of subscrbers on the network n addton to the capacty requred per subscrber. We have defned Pen as the market penetraton of the commercal provder as a fracton of populaton covered by the network. Exstng natonwde wreless servce provders have voce market penetratons of roughly 5% 25% [67]. In the base case, we desgn the network so that t can support 10% of the populaton covered as subscrbers (Pen = 10%). As wth the rest of the capacty nputs, a range of values for Pen are studed n secton fract: Other Cell Interference as a Fracton of Same Cell Interference In a CDMA system, the amount of cochannel nterference present at a basestaton due to other cells n the system s typcally treated as a fracton, fract, of the cochannel nterference due to users n the same cell. The value of fract has been studed extensvely for commercal systems and values tend to range from and we have chosen a value of 0.6 n the base case [42] Nose The nose power present at a recever can have a sgnfcant mpact on the recever senstvty. Snce envronmental nose power can depend on the frequency of operaton, to ensure an extensble model that can be used to study proposals n a varety of frequency bands, we present two equatons for nose: one vald at frequences above 400MHz and one vald at frequences below 400MHz. At frequences above 400MHz, the domnant envronmental nose s thermal [72] and we calculate total nose power usng the followng equaton [73]: N TOT = N P + NF = 10 log10 ( ktw ) + N F {n db} Where: N F = the nose fgure of the recever (db) N P = the thermal nose power (dbm) k = Boltzmann s constant (1.38E-23 J/K) T = the temperature at the recever (K) W = the bandwdth of the receved sgnal (Hz)

22 Below 400MHz, the equaton for envronmental nose power s modfed to nclude an adjustment factor as shown below [72]: N = log ( ktw ) log ( f ) K + N TOT MHz Where: f MHz = the frequency of the sgnal (MHz) K = s a constant: 15dB for rural areas, 18 db for suburban and 25 db for urban. adj adj F It s standard to assume a fxed value equal to room temperature for all recevers (290K) [73]. We use a 1.25MHz channel wdth whch s approprate for the CDMA system we have consdered. Values for nose fgures for a base staton are typcally n the range of 3 8 db and we use a value of 4dB n the base case [42] [50] [74]. 3.2 Lnk Budget Inputs As defned n secton 2.3, a lnk budget can be used to account for all of the gans and losses present n a wreless channel. Sectons wll dentfy numercal values for each of the followng gans and losses, respectvely: transmt power, antenna gan, loss margns for coverage relablty and n-buldng coverage, and mplementaton and scenaro losses approprate for a publc-safety-grade system Transmt Power We have chosen to defne the transmt power n terms of EIRP. EIRP represents the effectve power radated from the transmtter whch means that any losses nternal to the transmtter (e.g. due to cablng) and gans from the transmtter antenna are all ncluded n ths term. We have chosen a maxmum transmt power equal to the power lmt currently mposed on the 700MHz band n the US by the FCC. In the downlnk ths lmt s 1kW ERP (62.15 dbm EIRP) n urban areas and 2kW ERP (65.15 dbm EIRP) n rural areas [75]. In the uplnk, the power lmt s 30W ERP (46.9 dbm EIRP) for moble devces and 3W ERP (36.9 dbm EIRP) for portable devces [76]. These power lmts are lkely comparable to the transmt powers used n systems n the 168MHz and 414MHz federal bands as well. Snce FCC-regulated power lmts are strcter n the uplnk than n the downlnk, cell radus s typcally determned based on upstream communcatons (the downlnk has a 30 db transmt power advantage over portable devces operatng on the uplnk). Whle t s possble that users devces would operate at lower power (e.g. to save battery power), there s no techncal reason a new network cannot be desgned for devces that operate at the band power lmt. We assume n the base case for all proposals that publc safety equpment wll adopt a value of EIRP equal to the max allowed n the uplnk: 37 dbm. By comparson, a typcal commercal handset transmts at about 24dBm. In the publc-prvate partnershp, commercal handsets may be desgned to transmt at a lower power than 37 dbm, thereby choosng to accept a sgnal relablty that s below what publc safety would requre, but ganng the advantages of longer battery lfe and/or smaller and lghter moble devces. Such a decson would not change the cost of the nfrastructure, and therefore falls outsde the scope of our model. However, f the devces used

23 by publc safety operated at lower power, as some analysts have assumed [57], ths would have a sgnfcant mpact on nfrastructure cost. Ths effect wll be examned n secton Antenna Gan Snce we are only consderng the uplnk, the antenna gan of nterest s that of the base staton. Antenna gans at the basestaton are usually the most sgnfcant gans on a rado lnk and result from capturng more power n certan drectons than n others. In our analyss, we assume a standard, 3-sector cell, whch typcally use panel antennas that range from 9 18 db n gan [42] [44] [50] [51] [52]. In the base case, we chose a value of 18dB for antenna gan. There may be factors such as antenna cost and weght whch could lead to an antenna wth lower gan beng selected; however, these consderatons are outsde the scope of ths paper. Snce plannng a network wth reduced antenna gan can have a sgnfcant mpact on the number of cell stes requred, we examne a range of values n secton Coverage Relablty Margns The strength of a sgnal at any locaton wthn a cell s uncertan due to shadowng of the sgnal by obstructons n the path from transmtter to recever or fast fadng of the sgnal due to multpath effects n the channel. To account for ths uncertanty, loss margns are ncluded n the lnk budget to ensure suffcent sgnal power s avalable throughout the coverage area. Increasng ths margn ncreases the relablty of communcatons wthn the coverage area of the cell but reduces the cell s sze. Therefore, there s a tradeoff between the relablty of communcatons and the overall cost of the wreless network. Communcatons relablty n wreless system plannng s typcally expressed as ether a coverage relablty or cell-edge sgnal relablty. Coverage relablty s defned as the probablty that receved sgnal power wll be suffcent at any pont wthn the outdoor coverage area of a cell. Cell-edge sgnal relablty s defned as the probablty that receved sgnal power wll be suffcent at any pont along the outdoor, cell-edge contour. The FCC left the detals of coverage relablty requrements n the publc-prvate partnershp to later actons; only gvng the gudelne that the system be desgned consstent wth typcal publc safety communcaton systems [27] [32]. The publc safety lcensee, the PSST, has suggested that the system should be desgned for 95% coverage relablty [33]. However, best practces n the ndustry recommend that a publc safety system should be bult to 97% coverage relablty [72]. We desgned the system for 97% coverage relablty n the base case as approprate for a publc-safety-grade system. Expressng a value of coverage area relablty n a cell as a cell-edge sgnal relablty makes the calculaton of the approprate margn easer. We have used the method presented n [72] to convert the 97% coverage area relablty value to a value of 89% cell-edge sgnal relablty. The equaton to calculate the relablty margn gven a cell-edge sgnal relablty s provded n [50]. Ths equaton depends on the value chosen for the standard devaton of shadowng, σ L, whch typcally ranges from 4 8dB [50] [72] [77] [78]. We have used a value of σ L = 5.6dB as recommended n [72] whch yelds a margn for 97% coverage relablty of 12.6 db. In contrast, 95% coverage relablty requres only 83% cell-edge sgnal relablty and a margn of 10.3 db In-Buldng Coverage Margn

24 The coverage relablty margns ncluded n the lnk budget wll ensure a level of outdoor coverage relablty as dscussed n secton However, users who wsh to communcate wthn buldngs wll experence unrelable servce due to the attenuaton of sgnals havng to penetrate buldng walls. Whle t stll may be possble to communcate from wthn buldngs that are near the base staton, communcatons from ndoors near the cell edge could be hghly unrelable. To account for ths, a buldng penetraton margn s ncluded n the lnk budget. Ths margn s dependent upon the type of materal used to construct the walls of the buldngs n whch users want to operate. Smlar to the coverage relablty margn, there s a tradeoff between the relablty of communcatons ndoors and the overall cost of the wreless network. The FCC left the determnaton of n-buldng margns n the publc-prvate partnershp to future actons. However the PSST proposed 9 that the n-buldng margn be dependent upon the type of envronment beng served. In the PSST s proposal, the n-buldng penetraton margn s the same for areas classfed as rural as t s for open hghways: 6 db [33]. As dscussed n a submsson to the FCC [32], a 6 db margn should be suffcent for relable servce n a vehcle, but s lkely to be nsuffcent to penetrate the walls of many buldngs. By the PSST s assumptons, 92.3% of the area served s classfed as rural and as such, much of the coverage area of the US could have nadequate n-door coverage n many buldngs. We chose a margn of 13 db for all classfcatons of area (.e. rural, suburban, and urban) n the base case. Ths margn should be suffcent for relable sgnal penetraton of a sngle-walled - concrete buldng 10 [5] [79]. However, ths level of margn s stll not suffcent for relable penetraton through many types of structures. Snce ths desgn choce can sgnfcantly mpact the number of cell stes requred, a range of values s consdered n secton Implementaton and Scenaro Losses The mplementaton loss margn ncludes any losses due to the sgnal beng attenuated as t travels through cablng between the recevng antenna and basestaton as well as any losses due to msmatches and connectons at the basestaton. Typcal values for mplementaton losses at a cellular base staton range from 2 5 db and we have chosen a value of 4dB n the base case [42] [44] [50] [51] [52]. The scenaro loss margn estmates losses due to recever orentaton and polarzaton msmatches as well as sgnal obstructon due to the body of the user. Typcal values for scenaro losses range from 2 5 db and we have chosen a value of 4dB n the base case [42] [44] [50] [51] [52]. 3.3 Propagaton Model Inputs As dscussed n secton 2.5, the path loss predcted by the Hata model depends on frequency, base staton antenna heght, moble devce heght, and cell radus. Sectons and present the approprate values of moble devce and base staton antenna heght, respectvely. 9 [33] (4.) specfes the followng buldng penetraton margns dependng on the area covered: Dense Urban = 22 db; Urban = 19 db; Suburban = 13 db; Rural = 6 db; Hghway = 6 db 10 Appendx D ncludes a table whch summarzes the emprcal results of buldng penetraton loss measurements for a varety of materals

25 Secton 5 wll dscuss frequency as t s dependent on the proposal beng studed whle the model we developed n secton 2 solves for the remanng varables: radus and path loss Moble Devce Heght The moble devce heght s the heght at whch users hold the handset whle operatng the devce. The value for moble devce heght used n analyss of cellular networks typcally ranges from 1 2 meters and n the base case we chose the most common used value of 1.5m [78] Base Staton Antenna Heght We assumed that tower heght n a new system would be comparable to tower heghts avalable today, n part because many antennas are lkely to be placed on exstng towers. Thus, we analyzed the commercal tower heghts for a major US tower company: Amercan Towers. Ths company operates approxmately 23 thousand towers across 49 states n the US. Our analyss of the company s tower portfolo [80] revealed that the mean tower heght s approxmately 60 meters. We therefore assume a basestaton antenna heght of 60 meters n the base case. 4 Buld-out Requrements for a Natonwde Wreless Network In secton 4.1, we frst ntroduce the concept of a buld-out requrement for a natonwde wreless network (.e. what fracton of the US must be covered by the system) and then present a method of translatng between fracton of area covered and fracton of populaton covered. In secton 4.2, we dscuss the buld-out requrements that were ncluded n the network proposals hghlghted n secton 1. Secton 4.3 concludes wth a geographc nformaton system (GIS) model of the area covered by exstng publc safety wreless systems. 4.1 Fracton of Area vs. Fracton of Populaton A buld-out requrement can be expressed ether as a fracton of the US geographc area that s covered by the system or as a fracton of the US populaton covered. Requrements expressed as a fracton of populaton covered wll be consdered to mean the fracton of populaton whose homes are covered by the system. (Ths mplctly mposes no requrements to serve hghways, health care facltes, and other places that are not resdences.) Our analyss of the number of cell stes requred n a network calls for the buld-out requrement to be expressed as a fracton of US area covered, where the regons that are not covered are those wth the lowest populaton densty. Thus, to support comparsons of results based on buld-out requrements expressed n dfferent ways, we must be able to convert a fracton of US populaton covered to a fracton of US geographc area covered. If populaton were unformly dstrbuted across the US, there would be a straghtforward lnear relatonshp between the fracton of area covered and the fracton of populaton covered; however, that s not the case and t s unclear how to best relate these two fractons. We propose the followng method for convertng between the two fractons based on analyss of Census Bureau populaton and area statstcs at the zp code and county level. The followng table shows the fracton of total populaton contaned n all zp codes/countes for whch populaton densty s greater than X and the fracton of total area contaned n these zp codes/countes. Inherent n ths analyss s the assumpton that all zp codes/countes wth a populaton densty greater than X wll be completely covered and that no area n any zp code/county wth populaton densty less than X wll be covered. It s possble that n an actual deployment,

26 dependng on how the boundares of zp codes/countes are drawn, that all populaton n the area could be covered wthout coverng all of the area. County Level ZCTA Level %POP %CONUS AREA %US AREA %CONUS AREA %US AREA 50% 3.5% 3.0% 1.4% 1.2% 75% 13.1% 11.0% 7.1% 5.9% 90% 31.3% 26.5% 21.5% 18.0% 95% 44.3% 37.6% 33.1% 27.8% 96% 48.0% 40.8% 36.6% 30.9% 97% 52.6% 44.7% 41.0% 34.5% 98% 58.7% 50.1% 46.7% 39.4% 99% 68.6% 58.7% 55.3% 46.8% 99.3% 73.0% 62.9% 59.4% 50.3% 99.5% 77.1% 66.8% 63.1% 53.6% 99.9% 90.9% 80.0% 76.6% 65.6% 99.99% 98.0% 91.4% 87.9% 76.3% Table 4.1: A summary of the converson from populaton to area buld-out requrements. For varous percentages of populaton coverage, the percentage of area (CONUS and US) that would need to be covered s gven. Ths table shows that a majorty of populaton s concentrated n a small amount of the area whle a large amount of the US contans lttle populaton. Consderng both zp code and county level results, coverng the last 1% of populaton requres coverng an addtonal 41% to 53% of US area. By comparson, coverng the second to last 1% of populaton (gong from 98% to 99%) requres coverng an addtonal 7% to 9% of US area dependng on dataset consdered. The granularty of data used (zp code level vs. county level) sgnfcantly mpacts the converson from fracton of populaton covered to fracton of area covered. Ths llustrates the uncertanty n how much of the area wll be covered when buld-out requrements are specfed as a fracton of populaton. Ths uncertanty stems from the fact that the actual area covered depends on the placement and sze of the cells used, whch polcymakers wll not know before the network s deployed. Dependng on how polcymakers estmate the fracton of area that wll be covered when a populaton buld-out requrement s establshed, the wreless provder could meet the populaton coverage oblgaton whle coverng less area than antcpated by polcymakers. As wll be shown n secton 6, the number of cell stes requred n a system coverng the US s closer to the number of zp codes n the US than to the number of countes so we conjecture that conversons based on zp code level data are lkely to be more accurate than based on county level data. 4.2 Buld-out Requrements n Exstng Proposals When the FCC establshed the publc-prvate partnershp n the 700MHz band, they specfed a smlar buld-out requrement such that after 10 years 99.3% of the populaton must be covered [27]. Based on the FCC establshed buld-out requrements, the PSST (.e. the publc safety lcensee) generated the followng map whch estmates the area that would be covered and shows that the commercal partner s not expected to cover a thrd of the US wth the new system [81]

27 Ther analyss predcted that to cover 99.3% of the populaton t would be necessary to cover 73% percent of the geographc area of CONUS or about 63% of the entre US. Comparng these conversons to the results n secton 4.1, t appears the PSST s estmates are consstent wth usng county level data to predct coverage area from populaton buld-out requrements. As dscussed n secton 4.1, we beleve that performng the converson n ths manner nstead of usng zp code level data wll overestmate the actual area that wll be covered by a wreless system. We beleve the actual area that wll be covered by a wreless system that covers 99.3% of populaton s closer to 59% of CONUS area (50% of US area). Fgure 4.1: A map of CONUS showng the area (n green) that the PSST estmates wll have terrestral coverage from a publc-prvate partnershp system that covers 99.3% of populaton. Source of fgure: [81]. 4.3 GIS Model of Exstng Publc Safety Wreless Systems Ths secton presents an analyss of exstng publc safety wreless nfrastructure whch accomplshes two goals: (1) determnes how many towers the exstng nfrastructure requres, enablng comparsons wth a new natonwde network to understand potental cost savngs, and (2) determnes how much of the country s currently covered by the exstng nfrastructure enablng comparsons of coverage of a new natonwde system wth the coverage of today s many systems. Based on prevous research [1], we expect to see that a natonwde system would requre fewer tower stes than the exstng nfrastructure deployed by many ndependent agences requred. In secton 4.3.1, the dataset of transmtter stes used n ths analyss s dscussed. Secton presents the method used to calculate the coverage area of each transmtter. Secton contans the results of ths analyss and a bref dscusson Dataset of Transmtter Stes

28 The source dataset for ths project was the Prvate Land Moble Rado (PLMR) database. Ths data was obtaned February 2008 from the Unversal Lcensng System (ULS) database mantaned by the FCC [82]. The PLMR database contans lcense detals for more than just publc safety agences and therefore we fltered the dataset by Rado Servce Code (RSC) to ensure only publc safety agences were ncluded n the analyss 11. The lattude and longtude coordnates for each transmtter ste were extracted along wth the frequency of operaton, base staton antenna heght, base staton antenna gan, and lne losses. For a neglgble number of stes, ether the lattude/longtude or frequency felds were empty; n ths case these records were dropped from the dataset. Smlarly, for a small number of records, base staton antenna heght, base staton antenna gan or lne losses were ncomplete; n ths case the feld at ssue was set to the average value for the dataset. After flterng, 136,322 records reman as publc safety LMR transmtter stes. It s possble that more than one record corresponds to the same tower gven that a tower can be shared by several agences. We calculate the number of unque tower stes by flterng by geographc coordnates and elmnatng any duplcate lattude/longtude pars n the dataset. However, the resoluton n the geographc coordnates s such that duplcate coordnates may not necessarly correspond to a sngle tower but rather two towers sted very near each other. Thus our analyss provdes the followng bounds on the exstng publc safety nfrastructure: Upper Bound: 136,322 [Towers] Lower Bound: 97,660 [Towers] Lnk Budget and Propagaton Model Due to the greater power avalable on the downlnk, we have assumed that the uplnk wll be the lmtng case n two-way voce communcaton for publc safety. Thus, smlar to the lnk budget gven n secton 2, the followng s the lnk budget used for the exstng publc safety narrowband voce systems: PL = EIRP + GRX LIMPLEMENT LRELIABLE LBUILD S {n db} Where: EIRP = 37 dbm (for a typcal 5W EIRP portable land moble rado) L RELIABLE = 12.6 db (for 97% coverage area relablty w/ σ L = 5.6 db) L BUILD = 13 db (for coverage wthn sngle-walled concrete buldngs) S = 119 dbm (for a typcal nose-lmted base staton w/ N f =5 db) [50] [78] And the values for the followng varables are gven n the dataset for each transmtter G RX Recevng Antenna Gan (db) L IMPLEMENT Implementaton Losses at the Recever (db) The coverage radus for each transmtter was predcted based on the path loss calculated n the lnk budget and a modfed verson of the Hata model presented n secton 2; the Hata-Davdson 11 We ncluded records correspondng to the followng Publc Safety RSCs: GE, GF, GP, PW, QM, SG, SL, SY, YE, YF, YP, and YW

29 model. The Hata-Davdson model provdes an extenson of the basc Hata model up to base staton heghts of 2500 meters and the full equatons are avalable n [72] Results: Geographc Area Covered by Exstng Infrastructure Below are the plots of the CONUS and Alaska+Hawa area covered by at least one publc safety wreless transmtter ste. In the CONUS map, area covered s represented by the green colored porton of the map whle n the Alaska+Hawa map, area covered s represented by the purple colored porton of the map. Fgure 4.2: A map of CONUS showng the area (n green) that we calculated to have terrestral coverage n February 2008 from one or more publc safety wreless systems

30 Fgure 4.3: A map of Alaska and Hawa showng the area (n purple) that we calculated to have terrestral coverage n February 2008 from one or more publc safety wreless systems The followng table summarzes the coverage statstcs calculated n ths analyss: Regon of Interest Total Area [km²] Area Covered [km²] Percent Covered Alaska 1,717, , % CONUS 8,080,464 7,757, % US 9,826,630 8,175, % Table 4.2: A summary of the total sze of Alaska, CONUS and the US [83] as well as the area covered and the fracton of area covered by one or more publc safety wreless systems n each regon. As mentoned n secton 4.2, the PSST predcted that, under FCC populaton buld-out requrements, 73% percent of CONUS area or about 63% of US area wll be covered; although we beleve ths s an overestmate and the actual coverage would be closer to 59% and 50% of CONUS and US area respectvely. Comparng our results n the table above to these area estmates correspondng to the FCC buld-out requrements, t appears that consderably more area s currently served by exstng publc safety communcaton systems than would be served by the proposed publc-prvate partnershp. Ths means that 20 or 33% of the US area that s currently covered by exstng systems would not have access to the new network and would thus have to mantan the exstng nfrastructure. In the base case, we consder a publc-safety-grade network that serves the same fracton of the US currently beng served by exstng publc safety wreless systems (.e. 83% of US area)