1: \documentclass[twocolumn]{IEEEtran}
2:
3: \usepackage{epsfig}
4:
5: \usepackage{}
6:
7: \def\BibTeX{{\rm B\kern-.05em{\sc i\kern-.025em b}\kern-.08em
8: T\kern-.1667em\lower.7ex\hbox{E}\kern-.125emX}}
9:
10: \newcommand{\E}{\mbox{E}}
11: \newcommand{\Pb}{\mbox{P}}
12: \newtheorem{theorem}{Theorem}
13:
14: %\renewcommand\floatpagefraction{.9}
15: %\renewcommand\topfraction{.9}
16: %\renewcommand\bottomfraction{.9}
17: %\renewcommand\textfraction{.1}
18: %\setcounter{totalnumber}{50}
19: %\setcounter{topnumber}{50}
20: %\setcounter{bottomnumber}{50}
21:
22: \setcounter{page}{1}
23: \begin{document}
24: \title{Uplink Throughput in a Single-Macrocell/Single-Microcell CDMA System, with Application to Data Access Points}
25: \author{Shalinee Kishore, Stuart C. Schwartz, Larry J. Greenstein, H. Vincent Poor\thanks{S. Kishore is with the Department of Electrical
26: and Computer Engineering, Lehigh University, Bethlehem, PA, USA.
27: H.V. Poor and S.C. Schwartz are with the Department of Electrical
28: Engineering, Princeton University, Princeton, NJ, USA.
29: L.J. Greenstein is with WINLAB, Rutgers University, Piscataway, NJ, USA. This research was jointly supported
30: by the New Jersey Commission on Science and Technology, the
31: National Science Foundation under Grant CCR-00-86017, and the
32: AT\&T Labs Fellowship Program. Contact:
33: skishore@eecs.lehigh.edu.}}
34:
35: \maketitle
36:
37: \begin{abstract}
38: This paper studies a two-tier CDMA system in which the microcell base is converted into a {\em data access point} (DAP), i.e., a limited-range base station that provides high-speed access to
39: one user at a time. The microcell (or DAP) user operates on the same
40: frequency as the macrocell users and has the same chip rate. However,
41: it adapts its spreading factor, and thus its data rate, in
42: accordance with interference conditions. By contrast, the
43: macrocell serves multiple simultaneous data users, each with the same
44: fixed rate. The achieveable throughput for individual microcell
45: users is examined and a simple, accurate approximation for its
46: probability distribution is presented. Computations for average throughputs, both
47: per-user and total, are also presented. The numerical results highlight the impact of a {\em desensitivity}
48: parameter used in the base-selection process.
49:
50: \end{abstract}
51: \begin{keywords}
52: CDMA, throughput, macrocell, microcell, data access points
53: \end{keywords}
54:
55: \section{Introduction}
56: \label{dap_sec}
57: In \cite{skishore1}, we studied the uplink user capacity in a two-tier CDMA
58: system composed of a single macrocell within which a single microcell is embedded. This system assumed that all users transmit over
59: the same set of frequencies, with handoff between tiers.
60: Analytical techniques were developed to both exactly compute and
61: accurately approximate the user capacity supported in such a
62: two-tier system. Here, we implement these techniques to study a
63: particular kind of single-macrocell/single-microcell system. Specifically, we
64: consider a wireless data system where the microcell is designed to attract only a
65: small number of users $(n)$, while the macrocell attracts a
66: larger number of users $(N_M)$. The users are uniformly
67: distributed over the entire system coverage area, all having the same chip rate, $1/W$, where $W$ is the system bandwidth.
68: The macrocell users have the same fixed data rate, $R_M$, and
69: the same output signal-to-interference-plus-noise ratio (SINR)
70: requirement, $\Gamma_M$. The $n$ microcell users, on the other
71: hand, can be high-speed and are given access to the
72: microcell one-at-a-time. The
73: microcell resembles a {\em data access point} (DAP), i.e., a base
74: station with limited coverage area that provides high data rates
75: to a small number of users in sequence. Some examples of DAP-like applications are in \cite{Kohno}-\cite{Goodman}.
76: The uplink data rate, $R_\mu$, of any given microcell user has a
77: maximum achieveable value, $R_\mu^*$, determined by existing
78: interference conditions. We will quantify $R_\mu^*$ and related
79: quantities for a given a microcell
80: output SINR requirement. We focus here on the uplink, which we have
81: shown in \cite{sk_downlink} to be the limiting direction in this kind
82: of architecture.
83: \\
84: \\
85: In Section \ref{sysarch}, we elaborate on the system architecture,
86: the uplink SINR's, and the path gain model used. In the process, we
87: identify a normalized desensitivity factor, $\zeta$, that is a key design
88: variable. In Section \ref{datarates}, we derive the uplink transmission
89: rate for a given user and propose (and confirm via simulation) a simple approximation for its
90: probability distribution. In Section \ref{throughputs}, we compute, as functions of
91: $\zeta$, the average per-user and per-DAP throughputs and show that the choice of $\zeta$ involves a tradeoff between user
92: high-speed capability and overall DAP utilization. We also show the tradeoff between DAP performance and total user population.
93: \section{System and Channel Assumptions}
94: \label{sysarch}
95: \subsection{Architecture}
96: We assume a system in which $N$ data users in some region $\mathcal R$
97: communicate with either a low-speed macrocell base or a microcell base
98: acting as a DAP, as shown in Fig. \ref{sysfig}. In our computations, we will
99: further assume that the region $\mathcal R$ is a square of side $L$, with
100: the macrocell base at its center, the microcell base at some distance
101: $D$ from the center, and users located with uniform randomness over
102: the region.
103: \\
104: \\
105: Each data user communicates with the macrocell at a fixed rate
106: $R_M=W/G$, where $G$ is the spreading factor of the CDMA code.
107: However, when a user's path gain to the microcell base (DAP) exceeds
108: some threshold, it becomes a candidate for communication, at higher
109: speed, with the DAP. The base selection criterion we use is that the
110: macrocell base is selected whenever the user's path gain to it
111: exceeds that to the DAP by some fraction $\delta$, called the {\em
112: desensitivity factor}, or just {\em desensitivity}. Clearly, the
113: smaller $\delta$ is, the smaller the number of users eligible for DAP
114: access.
115: \\
116: \\
117: We denote by $n$ the number of eligible DAP users at any time, and
118: specify that they access the DAP sequentially. Each gets a timeslot
119: whose duty cycle is $1/n$, and a data rate during the timeslot that
120: depends on factors discussed below. The remaining $N_M=N-n$ users
121: simultaneously access the macrocell base at data rate $R_M$. These users require a
122: minimum SINR of $\Gamma_M$ while each of the $n$ DAP users requires a
123: minimum SINR of $\Gamma_\mu$.
124: \subsection{Uplink SINR's}
125: Each of the $n$ microcells users accesses the microcell one-at-a-time so that, at any one time, there are exactly $N_M+1$ system users. The base stations
126: employ matched filter (RAKE) receivers to detect these
127: active users. Each base controls the transmit powers of its own
128: users so that each user meets the minimum output SINR requirement at
129: that base. For such a system, the output SINR's at the macrocell and
130: microcell are
131: \begin{equation}
132: \mbox{SINR}_M = \frac{\frac{W}{R_M}S_M}{(N_M-1)S_M + S_\mu I_M +
133: \eta W} \label{sinr1}
134: \end{equation}
135: and
136: \begin{equation} \mbox{SINR}_\mu = \frac{\frac{W}{R_\mu}S_\mu}{S_M I_\mu +
137: \eta W},\label{sinr2}
138: \end{equation}
139: respectively, where $S_M$ and $S_\mu$ are the received power from each macrocell
140: user at the macrocell base and the received power from the
141: microcell user at the microcell base, respectively; $\eta W$ is
142: the received noise power, at each base, in bandwidth $W$; $R_M$
143: and $R_\mu$ are the bit rates of macrocell and microcell users,
144: respectively; and $I_M$ and $I_\mu$ are the normalized cross-tier
145: interferences. Specifically,
146: $I_\mu$ is the normalized interference from the $N_M$ macrocell
147: users into the microcell base, and $I_M$ is the normalized
148: interference from the one active microcell user into the macrocell
149: base. As shown in \cite{skishore1},
150: $I_\mu$ and $I_M$ depend solely on the set of path gains from the
151: active users to both bases, and on $\delta$. They are
152: \begin{equation}
153: I_M=\frac{T_{M}}{T_{\mu}}
154: \label{im}
155: \end{equation}
156: and
157: \begin{equation}
158: I_\mu = \sum_{k \in M} \frac{T_{\mu k}}{T_{Mk}},
159: \label{imu}
160: \end{equation}
161: where $T_M$ ($T_{\mu}$) is the path gain from the microcell user to
162: the macrocell (microcell) base; $T_{\mu k}$ ($T_{Mk}$) is the path
163: gain from the $k$-th macrocell user to the microcell (macrocell) base;
164: and $M$ denotes the set of all macrocell users. The denominators in
165: (\ref{im}) and in each term of (\ref{imu}) are a consequence of using
166: power control to each user from its serving base. Adequate
167: performance requires that $\mbox{SINR}_M \geq \Gamma_M$ and
168: $\mbox{SINR}_\mu \geq \Gamma_\mu$.
169: \subsection{Path Gain Model}
170: The path gain, $T$, between either base and a user at a distance $d$
171: is assumed to be
172: \begin{eqnarray}
173: T=\left\{ \begin{array}{cc} H\left( \frac{b}{d} \right)^2
174: 10^{\chi/10}, & d \leq b \\ H \left( \frac{b}{d} \right)^4
175: 10^{\chi/10}, & d>b \end{array} \right.,
176: \end{eqnarray}
177: where $b$ is the ``breakpoint distance'' (in the same units as $d$),
178: at which the slope of the dB path gain versus distance changes;
179: $\chi$ is a zero-mean Gaussian random variable for each user
180: position, with standard deviation $\sigma$; and $H$ is a proportionality
181: constant that depends on wavelength, antenna heights and antenna gains.
182: Note that $T$ is a local spatial average, so that multipath effects
183: are averaged out. There can be different values of $b$ for the
184: microcell and macrocell, and similarly for $\sigma$ and $H$. The
185: factor $10^{\chi/10}$ is often referred to as lognormal shadow fading,
186: which varies slowly over the terrain. Both $\chi$ and $d$ are random
187: variables for a randomly selected user.
188: \\
189: \\
190: Due to the greater height
191: and gain of the typical macrocell antenna, we can assume $H_M>H_\mu$. As mentioned above, the
192: system we examine here assumes path-gain-based selection, i.e., a user
193: selects the macrocell base if the path gain to it exceeds the path
194: gain to the microcell base by some specified fraction $\delta$. In addition, the path gain of a user to the
195: macrocell exceeds its path gain to the microcell by a value
196: $h=H_M/H_\mu$ for the same distance and shadow fading. Therefore,
197: base selection
198: depends fundamentally on the ratio $\delta/h$, as opposed to $\delta$ alone.
199: (We chose not to normalize $\delta$ by $h$ in our study in \cite{skishore1}.) The factor $\delta/h$ is called the {\em normalized
200: desensitivity} of the microcell, and is hereafter denoted by
201: $\zeta$. Smaller values of $\zeta$ correspond to higher path gain
202: requirements to the microcell. This means that outlying users are
203: generally excluded from access to the microcell, and a smaller microcell
204: coverage area results.
205: \section{Data Rate for a DAP User}
206: \label{datarates}
207: For a given number of DAP users $n$, we have
208: $N_M=N-n$ macrocell users, and there is a probability $p_n$ (derivable from the
209: results in \cite{skishore1}) that $n$ of the $N$ users will choose
210: the microcell and $N-n$ the macrocell. We can show from (1) and (2) that the requirements $\mbox{SINR}_M \geq \Gamma_M$ and $\mbox{SINR}_\mu \geq \Gamma_\mu$ yield
211: positive solutions for $S_\mu$ and $S_M$ if and only if
212: \begin{equation}
213: \frac{W}{R_\mu}(K-N_M) - \Gamma_\mu I_M I_\mu \geq 0, \label{feas1}
214: \end{equation}
215: where $K=W/(R_M \Gamma_M)+1$ is the {\em single-cell pole capacity} of the
216: macrocell \cite{gilhousen}. Thus, we can find $R_\mu^*(n)$, the maximum achieveable
217: $R_\mu$ when $n$ users choose the microcell: Using (\ref{feas1})
218: with $N_M = N-n$, and noting that the spreading factor
219: $W/R_\mu$ cannot be less than 1, we have
220: \begin{equation}
221: R_\mu^*(n)=\min \left( \frac{W(K-N+n)}{\Gamma_\mu I_M I_\mu},W
222: \right) , \mbox{~~~}1 \leq n \leq N.\label{rmueq}\footnote{We consider all integer
223: spreading factors from 1 to $W/R$. For analytical
224: ease, we further assume a {\em continuum} of values over that range, so
225: that $R^*_\mu$ is the value of $R_\mu$ that meets (6) with equality,
226: up to a maximum of $W$.}
227: \end{equation}
228: We define $r=R_\mu^*/W$ and, henceforth, we examine this normalized data rate. Note that $r$ will be different as each of the $n$ users
229: takes its turn, because $I_M I_\mu$ depends on the set of all (active)
230: user path gains.\footnote{Since this set depends on $\zeta$, the
231: product $I_MI_\mu$ in (7) does, as well. The denominator $T_\mu$ in
232: (3), which decreases with increasing microcell coverage area (i.e.,
233: with increasing $\zeta$), has the dominant impact on $I_MI_\mu$ and,
234: thus, $R^*_\mu$.} Also, for any given user, we see that $I_M$ and
235: $I_\mu$ are random variables because of the random locations and
236: shadow fadings of all users. Thus, $r$ is itself a random
237: variable. To facilitate analysis, we assume (and will show) that
238: $I_M$ and $I_\mu$ can be treated as lognormal variates, whose first
239: and second moments are
240: obtainable using the results in
241: \cite{skishore2}. Since $1/(I_MI_\mu)$ is lognormal under this
242: assumption, we conclude that $r$, as given above,
243: is a truncated lognormal random variable. The Appendix shows how the
244: lognormal parameters are obtained from the moments
245: of $I_M$ and $I_\mu$.
246: \\
247: \\
248: Let $F(r|n)$ denote the cumulative distribution function
249: (CDF) of $r$ for a given $n$. The CDF of $r$ with the condition on
250: $n$ removed, but subject to $n$ exceeding 0, is\footnote{If $n=0$
251: (which happens with probability $p_0$), we have no interest in the
252: data rate of a non-existent user! Thus, we seek the CDF of $r$ when
253: there is at least one DAP user.}
254: \begin{equation}
255: F(r)=\frac{\sum_{n=1}^N p_n F(r|n) \label{cdfsum}}{1-p_0}.
256: \label{cdfRmu}
257: \end{equation}
258: Assuming that $F(r|n)$ can be approximated as the CDF of a truncated lognormal random variable, $F(r)$ can approximated by a weighted
259: sum of CDF's of truncated lognormals.
260: \\
261: \\
262: To test the reliability of the lognormal approximation, we
263: performed a series of simulation trials. We assumed a square
264: region $\mathcal R$ with sides of length $L$ over which users are
265: uniformly distributed, with the two bases separated by $D$. For the system and propagation
266: parameters listed in Table 1, we determined the achieveable
267: data rates of microcell users over 10,000 trials, for
268: various values of $\zeta$ and with $N=26$.\footnote{The case $N=26$
269: corresponds, for the parameters of Table 1, to a highly stressed
270: macrocell, i.e., $N \approx K$, the macrocell pole capacity. Later we show the general tradeoff between $N$ and DAP
271: throughput.} The resulting CDF of
272: $r$ is plotted in Fig. \ref{cdfs_singleuser}.
273: Along with simulation results, this figure contains CDFs obtained from
274: analysis, (\ref{cdfRmu}), assuming lognormal
275: cross-tier interferences. These results show that the
276: lognormal approximation is reliable over a wide range of $\zeta$.
277: As $\zeta$ gets smaller, the data rates of the microcell users
278: increase, suggesting that smaller $\zeta$ is desirable. The problem with {\em extremely}
279: small $\zeta$, however, is that it shrinks the population
280: of possible microcell users. The result is under-utilization of the
281: DAP, as discussed below.
282: \section{Results}
283: \label{throughputs}
284: \subsection{Per-User and Total DAP Throughputs}
285: We define our throughput variable as the time-averaged data rate normalized by $W$.
286: The per-user throughput, $\tau_u$, takes into account the time-limited
287: access that data users must accept if there is more than one microcell
288: user in the system. This division of time effectively reduces data
289: rates $(r)$ in a system with $n$ microcell users by a factor of $n$.
290: The CDF of $\tau_u$ is thus
291: \begin{equation}
292: F_{\tau_u}(\tau_u)=\frac{\sum_{n=1}^N p_n F(n \tau_u|n)}{1-p_0}, \label{cdfsum2}
293: \end{equation}
294: where $F(\cdot|n)$ is as defined earlier. The CDF of $\tau_u$ can be approximated using the truncated-lognormal
295: assumption for $R_\mu^*$, (\ref{rmueq}). Using both simulation
296: and this approximation, we again considered the
297: single-macrocell/single-microcell system described above, with the parameters in
298: Table 1. The results are plotted in Fig.
299: \ref{peruser_cdfs} for three different values
300: of $\zeta$. We see that the approximate distribution follows the
301: simulation distribution very closely for all three values. The jumps
302: in Fig. 3 are related to the fact that per-user data rates are
303: confined to the discrete values $W/n$, $n=1,2,\ldots$ for the high
304: fraction of cases where $R^*_\mu=W$.
305: \\
306: \\
307: The {\em total} throughput for the DAP, for given a $n \geq 0$, is the sum of the
308: throughputs for the $n$ users. We denote the total throughput (i.e., DAP utilization) by
309: $\tau_d$. Thus, $\tau_d$ represents performance from the network operator's point-of-view, while $\tau_u$ represents performance from the user's point-of-view. We quantify the average per-user and per-DAP throughputs, as
310: functions of $\zeta$, where the averaging is over random locations
311: and shadow fadings of the users. $\E \{ \tau_u \}$ is obtained
312: via the formula for its CDF, (\ref{cdfsum2}). $\E \{ \tau_d \} $
313: is obtained by noting that $\E \{ \tau_d | n \}$ is just $\E \{ r|n \}$. Removing the condition on $n$, we get
314: \begin{equation}
315: \E \{ \tau_d \} = \sum_{n=0}^N p_n \E \{r|n \}.
316: \end{equation}
317: The results for $\E \{ \tau_u \}$ and $\E \{ \tau_d \}$ are shown in Fig.
318: \ref{avg_thruputs} for $N=26$. They are given for both simulation and the
319: approximation method, and we see very strong agreement. For this system, we see that, when
320: $\zeta \approx 0.007$, $\mbox{E}\{ \tau_u \} = \mbox{E} \{\tau_d \} \equiv \tau^*$. This
321: value of $\zeta$ (which we call $\zeta^*$) is desirable,
322: since it balances the throughput for both individual users
323: and the overall DAP. When $\zeta > \zeta^*$, individual user throughputs are
324: compromised for the sake of higher DAP utilization; when $\zeta < \zeta^*$, higher per-user throughputs are
325: obtained, but for an under-utilized DAP.
326:
327:
328: \subsection{Effect of $N$ on $\tau^*$}
329: The results reported thus far assumed $N=26$, i.e., $N \approx K$. It is instructive to repeat the above exercise for a range of $N$, to show the tradeoff between total user population and DAP throughput. Thus, we have computed $\E \{ \tau_u \}$ and $\E \{ \tau_d \}$ as functions of $\zeta$ for various $N$, still assuming a uniform distribution of users. For each value of $N$, $\E \{ \tau_u \}$ decreases and $\E \{ \tau_d \}$ increases with increasing $\zeta$, just as in Fig. 4; however, the values of $\tau^*$ and $\zeta^*$ change with $N$. This is shown in Fig. 5.
330: \\
331: \\
332: For a given $\zeta$, decreasing $N$ implies a shrinking number of DAP users, leading to a less utilized DAP. Thus, $\zeta$ must be increased to maintain high DAP utility, i.e., $\zeta^*$ must increase as $N$ decreases below $K$. At the same time, the cross-tier interference at the DAP shrinks, leading to increases in $\tau^*$. In the region $N \leq K$, the average $n$ at $\zeta=\zeta^*$ is close to 1, so that $N_M \approx N-1$. In the region $N>K$, meeting the feasibility condition $N_M < K$, (6), requires that $n>1$. The result is a sharp decline in $\tau^*$ as $N$ increases from $K$, as foretold by the steepness of the $\tau^*$-curve at $N=26$.
333:
334: \subsection{Effect of Dense User Distribution Around the DAP}
335: Using the methods presented here, throughput results can be
336: obtained for a system with a denser user distribution around the
337: DAP. We found that, as the density around the
338: DAP increases, both $\zeta^*$ and $\tau^*$ decrease. Smaller values of $\zeta^*$ are desirable because they help reduce $n$, the number of users that must share the DAP. Despite this, as the user density around the DAP increases, the most probable value of $n$ increases, resulting in smaller $\tau_u$ and a net (modest) reduction in $\tau^*$.
339: \section{Conclusion}
340: We have demonstrated that by controlling the desensitivity factor, a
341: microcell can be converted to a data access point. The data rate of
342: the single DAP user was analyzed and shown to be well-represented as a
343: truncated-lognormal variate. A method was then developed to compute
344: the throughput statistics, and we found values of desensitivity, over the practical range of user population, for which both per-user and total DAP throughputs are reasonably high.
345:
346: \appendix
347: \section{Lognormal Representation of the User Data Rate}
348: The user data rate, $r=R_{\mu}^*/W$ is seen from (\ref{rmueq}) to be
349: the lesser of 1 and $Z$, where $Z=(K-N-n)/(\Gamma_\mu I_M I_\mu)$. Defining $z = \ln Z$, we have $z = \ln ((K-N-n)/\Gamma_\mu) - \ln I_M - \ln I_\mu$.
350: Clearly, if $I_M$ and $I_\mu$ are lognormal, $z$ is a Gaussian random
351: variable whose mean and standard deviation are obtainable from the
352: means and standard deviations of $\ln I_M$ and $\ln I_\mu$.
353: \\
354: \\
355: In \cite{skishore2}, the first and second moments of $I_M$ and $I_\mu$
356: are derived for the two-cell system. The means and standard
357: deviations of $\ln I_M$ and $\ln I_\mu$ are then obtainable as
358: follows: Let the mean and standard deviation of $\ln I_M$ be $m$ and
359: $\sigma$, respectively. From lognormal statistics \cite{andrealarry},
360: \cite{schwartzye},
361: \begin{eqnarray*}
362: \E \{ I_M \} = e^{m}e^{\sigma^2/2} \mbox{~~~and~~~~}
363: \E\{ I_M^2 \} = e^{2m}e^{2 \sigma^2}.
364: \end{eqnarray*}
365: Solving for $m$ and $\sigma$,
366: \begin{eqnarray*}
367: m= \frac{1}{2} \ln \left(\frac{(\E \{ I_M \})^4}{\E \{ I_M^2 \} } \right)\mbox{~~;~~~}
368: \sigma = \sqrt{\ln \left( \frac{ \E \{ I_M^2 \} }{(\E \{ I_M \})^2} \right)}.
369: \end{eqnarray*}
370: Similar results apply to $\ln
371: I_\mu$. Thus, $r$ can be fully described as a truncated lognormal
372: variate given the first two moments of $I_M$ and $I_\mu$, and assuming
373: both are lognormal.
374:
375: \begin{thebibliography}{1}
376:
377: \bibitem{skishore1}
378: S.~Kishore, et al.,
379: \newblock ``Uplink Capacity in a CDMA Macrocell with a Hotspot Microcell: Exact
380: and Approximate Analyses,"
381: \newblock {\em IEEE Transactions on Wireless Communications}, Vol. 2, No. 2, pp. 364-374, March 2003.
382: \bibitem{Kohno}
383: R. Kohno,
384: \newblock ``ITS and Mobile Multi-Media Communication in Japan,''
385: \newblock {\em Proc. of Telecommunication Technique Workshop for
386: ITS}, Seoul, Korea, pp. 9-33, May 2000.
387: \bibitem{LeeKim}
388: B.S. Lee, et al.,
389: \newblock ``Performance Evaluation of the Physical Layer of the
390: DSRC Operating in 5.8 GHz Frequency Band''
391: \newblock {\em ETRI Journal}, Vol. 23, No. 3, pp. 121-128, September 2001.
392: \bibitem{Frenkiel1}
393: R. Frenkiel and T. Imielinski,
394: \newblock ``Infostations: The Joy of Many-Time, Many-Where
395: Communications,''
396: \newblock {\em WINLAB Technical Report}, TR-119, Rutgers University, 1996.
397: \bibitem{Goodman}
398: D. Goodman, et al.,
399: \newblock ``Infostations: A New System Model for Data and
400: Messaging Services,''
401: \newblock {\em Proceedings of the 1997 IEEE Vehicular Techn. Conf.}, Phoenix, AZ, pp. 969-973, May 1997.
402: \bibitem{sk_downlink}
403: S. Kishore, et al.,
404: \newblock ``Downlink User Capacity in a CDMA Macrocell with a Hotspot Microcell,'' in {\em Proceedings of IEEE Globecom}, vol. 3, pp. 1573-1577, December 2003.
405: \bibitem{gilhousen}
406: K.S. Gilhousen, et al.,
407: \newblock ``On the Capacity of a Cellular CDMA System,''
408: \newblock {\em IEEE Transactions on Vehicular Technology,} Vol. 40, pp. 303-312, May 1991.
409: \bibitem{skishore2}
410: S.~Kishore,
411: \newblock ``Capacity and Coverage of Two-Tier Cellular CDMA Networks,"
412: \newblock {\em Ph.D. Thesis}, Department of Electrical Engineering,
413: Princeton University, January 2003, Appendix B.
414: \bibitem{andrealarry}
415: A. J. Goldsmith, et al.,
416: \newblock ``Error Statistics of Real-Time Power Measurements in Cellular Channels with Multipath and Shadowing,''
417: \newblock {\em IEEE Transactions on Vehicular Technology}, Vol. 43, No. 3, pp. 439-446, August 1994.
418: \bibitem{schwartzye}
419: S.C. Schwartz and Y.S. Yeh,
420: \newblock ``On the Distribution Function and Moments of Power Sums
421: with Lognormal Components,''
422: \newblock {\em Bell System Technical Journal}, Vol. 61, No. 7,
423: September 1982, pp. 1441-1462.
424:
425: \end{thebibliography}
426: \vspace{.2in}
427: \begin{table}[h]
428: \begin{center}
429: \begin{tabular}{|c|c||c|c|}
430: \hline $W/R_M$ & 128 & $L$ & 1 km
431: \\ $\Gamma_M$ & 7 dB & $\Gamma_\mu$ & 8.45 dB
432: \\ $b_M$ & 100 m & $b_\mu$ & 100 m
433: \\ $H_M$ & $10 H_\mu$ & $D$ & 300 m
434: \\ $\sigma_M$ & 8 dB & $\sigma_\mu$ & 4 dB
435: \\
436: \hline
437: \end{tabular}
438: \\
439: \caption{System Parameters Used}
440: \end{center}
441: \label{sys_param}
442: \end{table}
443: \begin{figure}[htp]
444: \begin{center}
445: \epsfig{figure=generalR.eps,width=3.5in}
446: \caption{An example of a region $\mathcal R$, with macrocell and microcell
447: base station. Here M BS and $\mu$ BS denote the macrocell and
448: microcell base stations, respectively.}\label{sysfig}
449: \end{center}
450: \end{figure}
451: \begin{figure}[htp]
452: \begin{center}
453: \epsfig{figure=rmu_cdfs.eps,width=3.5in}
454: \caption{CDF's of $r$ using both simulation and the approximation
455: method, for $\zeta = 0.001$, $0.005$ and $0.05$. Results are for $N=26$.} \label{cdfs_singleuser}
456: \end{center}
457: \end{figure}
458: \begin{figure}[htp]
459: \begin{center}
460: \epsfig{figure=peruser_CDFS_th.eps,width=3.5in}
461: \caption{CDF of $\tau_u$ using both simulation and the approximation
462: method, for $\zeta = 0.001$, $0.005$ and $0.05$. Results are for $N=26$.} \label{peruser_cdfs}
463: \end{center}
464: \end{figure}
465: \begin{figure}[htp]
466: \begin{center}
467: \epsfig{figure=avg_throughput_new_2.eps,width=3.5in}
468: \caption{$\mbox{E}\{ \tau_u \}$ and $\mbox{E}\{ \tau_d \}$ as functions of
469: $\zeta$ using both simulation and the approximation method. Results
470: are for $N=26$.}
471: \label{avg_thruputs}
472: \end{center}
473: \end{figure}
474: \begin{figure}[htp]
475: \begin{center}
476: \epsfig{figure=dap_n_results.eps,width=3.5in}
477: \caption{$\zeta^*$ and $\tau^*$ as functions of $N$, for $10 \leq N \leq K$.}\label{densepop}
478: \end{center}
479: \end{figure}
480: \end{document}
481:
482:
483:
484:
485:
486:
487:
488:
489:
490:
491:
492: