Wireless Networks 3 (1997) 325–331 325

Spectrum sharing under the asynchronous UPCS etiquette:The performance of collocated systems under heavy load

Ivan Vukovic and John McKown

Motorola, MIMS Radio Research Lab, 50 E. Commerce Drive, Schaumburg, IL 60173, USA

Recently the FCC opened three 10 MHz bands for unlicensed use. In order to operate in UPCS bands, devices must comply withrules known as the UPCS etiquette (United States Code of Federal Regulations, Title 47, Part 15(d)). In this paper we study channelsharing between two or more collocated systems under the asynchronous UPCS etiquette. In particular we show that under heavy loadindividual systems have a tendency to hold the channel for hundreds of milliseconds, thus blocking all traffic in other, competing systems.We have calculated the distribution of the blocking time for two versions (or interpretations) of the UPCS etiquette. The impact of theaverage blocking time on delay sensitive traffic is discussed and possible improvements achieved through a tradeoff between systemcapacity and average blocking time are investigated.

1. Introduction

The case of several collocated but non-interworking net-works is of high importance for the UPCS industry. One ofthe parameters of interest is the distribution of the blockingtime for a system. In the case of two neighboring networks,channel usage alternates between the two systems. If onesystem is transporting time-sensitive traffic it is importantto know the distribution of the blocking period length. Inthe following analysis we derive this distribution for thecase of the multiple systems, assuming that each one isunder heavy load, i.e., a new burst of 10 msec is alwayswaiting to be transmitted. The analysis is carried out forthe two interpretations of the etiquette, namely the “non-persistent” and the “1-persistent” versions. To the best ofour knowledge the nonpersistent version has been adoptedas an official interpretation of [4].

2. Model

The asynchronous UPCS etiquette belongs to the CarrierSensing (CS) or Listen-before-talk (LBT) family of multi-access schemes [3]. The basic assumptions are that thechannel assessment is not instantaneous and that the maxi-mum period of time a user or a cooperating group of users(referred to as a system in the sequel) can transmit un-interrupted is 10 msec. Before each transmission, a usermonitors the channel for 50 µsec. If, during this interval,the power on the channel never exceeded a predefined def-erence threshold, then the node has permission to transmit.It is assumed here that after successful monitoring, the nodecan start transmission immediately although, in practice, aturnaround time needs to be taken into account. If thepower exceeded the deference threshold the node can fol-low one of two courses, both of which have been promotedas desirable interpretations of Part 15(d). In the first, whichwe have labeled as the nonpersistent case, the node defers

a random amount of time, called the deference interval, andrepeats the monitoring procedure. In the other case, whichwe call 1-persistent with collision avoidance, the node con-tinuously monitors the channel (i.e., repeats the monitoringprocedure) until it detects channel idle (i.e., power belowthreshold for 25 consecutive µsec), upon which it can starta random deference interval. In the following text we shallrefer to the latter scheme as simply 1-persistent having inmind that a transmission is delayed a random time intervalafter the channel becomes clear.

The range from which the uniformly distributed defer-ence interval is chosen is doubled each time a node detectschannel busy. The lower limit is fixed at 0.050 msec whilethe upper limit is initially 0.75 msec (before the first at-tempt) and reaches the upper limit of 12 msec after fourunsuccessful tries. After reaching this upper limit the defer-ence interval is kept constant indefinitely. This is in contrastto Ethernet’s (IEEE 802.3) retransmission scheme where re-transmissions are aborted after 16 tries [1]. After complet-ing a transmission, a node waits additional random intervalbefore it can transmit any new packets.

The etiquette does not specify the way acknowledgmentsare delivered. We have assumed that the ACKs are veryshort packets sent after a short gap by the destination nodewithout the LBT procedure. A system or a cooperatinggroup of users can keep the channel up to 10 msec with-out further LBT provided that the interpacket gap is lessthan 25 µsec.

In the analysis presented below we consider initially twoand later on multiple systems competing for the channel un-der heavy load. Each system monitors the channel after arandom deference interval (as mentioned above) followingthe previous end of transmission (1-persistent case) or anunsuccessful 50 µsec monitoring event (nonpersistent case).If the channel is sensed clear a series of packets is trans-mitted by units within the winning system for 10 msec.If the first packet sent by system A (figure 1) is detected

J.C. Baltzer AG, Science Publishers

326 I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette

Figure 1. A sequence of transmission cycles in a case of two collocatedsystems.

Figure 2. A sample path of the deference renewal process in the nonper-sistent etiquette.

by system B then all subsequent packets sent in the same10 msec window will be detected. We have assumed thatpacket transmission within each system is controlled by aMAC of choice, the details of which are of no importanceto us as long as the interpacket gap is less than 25 µsec. Asequence of transmission cycles, each consisting of a busyperiod followed by an idle period, is shown in figure 1.

In practical situations a collision can occur due to eitherpartial connectivity between the systems or a non-zero radioturnaround time. In the analysis we assume that collisionprobability is negligible. We shall discuss this in section 4.Collisions tend to decrease the blocking time since the def-erence time limits are reset after every transmission, bothsuccessful and unsuccessful.

Assume that a system, A, following the nonpersistent eti-quette has just lost contention as shown in figure 2. A se-ries of channel monitoring periods will follow separatedby random time intervals. The first delay is chosen froma uniform distribution U (0.05, 1.5), times being expressedin msec. In defining the intervals of random length thefollowing notation is used:

XT ,YT ∼ U (0.05,T ), T ∈ 0.75, 1.5, 3, 6, 12.

After four unsuccessful monitoring periods, each time dou-bling the upper limit, the system starts choosing the delayfrom the uniform distribution U (0.05, 12) msec. Thus, afterthe interval X6 (figure 2) the deference process is a renewalprocess. We assume that the renewal process starts duringthe first transmission period and that the renewal processis in “steady-state” at the end of the first busy period ofthe blocking time. It will also be assumed that the end of10 msec busy periods are independent of the deference re-newal process. We justify these assumptions by simulation.

3. Analysis

In this section we derive the probability mass functionand the average length of the blocking period for the non-persistent UPCS etiquette. The 1-persistent case is analyzedin the appendix.

Let us assume that at time t = 0 one system, B, winscontention after being blocked for one or more transmis-sion cycles. At time t = 10 system B stops transmission

and draws a deference delay from a uniform distributionU (0.05, 0.75). We need the distribution of the time intervalfrom the time the channel becomes clear (i.e., t = 10) untilthe blocked system A is scheduled to monitor the chan-nel. Denote this interval as a random variable Z (figure 2).From our assumptions in section 2, random variable Z rep-resents the excess life of the deference renewal process [2].The steady state distribution of Z is given by [2]

fZ (z) =

1− FX12 (z)E[X12]

, 0 6 z 6 12,

0, otherwise,(1)

where FX12 (z) is the cumulative distribution function ofX12. As before all times are expressed in msec. The prob-ability that the blocked system, A, will win the next con-tention round is given by

P [X0.75 > Z] =

∫∫Ω

fX0.75Z (x, z) dx dz, (2)

where Ω = (x, z): x > z. Since X0.75 and Z are inde-pendent we have that

P [X0.75 > Z] =

∫ 0.75

z=0

∫ 0.75

x=z

fX0.75 (x)fZ(z) dx dz. (3)

After performing the integration we obtain P [X0.75 > Z] =p = 0.06525, where p is the probability that the channelwill “change hands” at the end of each transmission cycle.Thus, the probability that one system will be blocked forNb = k transmission cycles is given by

P [Nb = k] = p(1− p)k−1 for k > 1, (4)

and the average length of the blocking period in the caseof nonpersistent UPCS expressed in the number of trans-mission cycles is given by

E[Nb] =1p

= 15.324. (5)

In order to find the average blocking period expressed inmilliseconds we need to obtain the average cycle period.The average idle period E[I] is equal to the expected valueof the random variable X0.75 conditioned on the event thatX0.75 < Z, except in the case of the last idle period Lwhose average idle period is equal to E[Z | X0.75 > Z].Thus,

E[L] =E[Z | X0.75 > Z]

=

∫ 0.75x=0

∫ xz=0 zfX0.75(x)fZ(z) dx dz

P [X0.75 > Z]= 0.248452 (6)

and

E[I] =E[X0.75 | X0.75 < Z]

=

∫ 0.75x=0

∫ 12z=x xfX0.75 (x)fZ(z) dx dz

P [X0.75 < Z]= 0.392962. (7)

I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette 327

It is interesting to note that the average value of the lastidle period is significantly smaller than the preceding idleperiods. The reason for this is that the random variable Zis more concentrated around the origin and that Z is con-ditionally bounded by a smaller value, X0.75. Finally, theaverage length of the blocking period is given by

E[Tb] =(E[Nb]− 1

)E[I] +E[Nb]10 +E[L]

= 159.121. (8)

The analysis above assumes that two systems with zeroturnaround time radios are competing for the channel. Oneextension of the above analysis would be to increase thenumber of competing systems to M > 2. In that case(M−1) blocked systems are trying to take over the channelfrom the “incumbent” system. The blocked system with theshortest deference time Z will compete with the currentlytransmitting system. The distribution in (1) needs to bemodified to represent the distribution of the minimum of(M − 1) iid random variables each having the distributionas in (1). Let us denote the random variable in (1) byZ (1). Thus, assuming that the jth system currently holdsthe channel, the distribution of Z (M−1) is given by

P[Z (M−1) > z

]= P

[mini6=j

(Zi) > z]

= P

[⋂i6=j

Zi > z

]=∏i6=j

P [Zi > z] =(P[Z (1) > z

])M−1(9)

where Zi’s are iid random variables having the distributionas in (1), i.e., the same as Z (1). Thus, the “incumbent”system j will lose contention in the next transmission cyclewith probability

P[X0.75 > Z (M−1)

]=

∫ 0.75

x=0

∫ x

z=0fX0.75 (x)fZ(M−1) (z) dx dz

=

∫ 0.75

x=0fX0.75 (x)

[1−

(P[Z (1) > x

])M−1]dx. (10)

The above expression represents the probability that the in-cumbent system will lose the contention in the next trans-mission cycle. It is clear that any of the (M − 1) blockedsystems can win the contention with equal probability.Thus, the probability that a particular blocked system willtransmit in the next transmission cycle is given by

p(M−1) =1

M − 1P[X0.75 > Z (M−1)

](11)

and therefore, the average number of transmission cyclesan arbitrary system is blocked is given as in (5) by

E[Nb] =1

p(M−1). (12)

In our notation p(1) is equal to the expression in (3). Theexpressions in (6) and (7) need to be updated to cover the

case M > 2, i.e., the distribution of Z (M−1) should beused instead of the distribution of Z = Z (1). The averagelength of the blocking period for an arbitrary system is thencalculated using the expression in (8). Numerical results arepresented in section 4.

The details of the calculation of the average blockingperiod for the 1-persistent version in the case of two com-peting systems is presented in the appendix.

4. Simulation results

It is assumed in the simulation that each 10 msec burstconsists of a large number of short packets exchangedwithin each system. If the first packet is not received theburst is interrupted and the system defers a random inter-val drawn from the uniform distribution U (0.05, 0.75). Inthis way each system can detect collision early on throughthe lack of the acknowledgment packet and backoff. Thetransmission time of the first packet is longer than the radioturnaround time, which guarantees collision detection. Theradio turnaround time refers to the time interval it takes toswitch the radio from the receive (monitor) to the transmitmode. It is also assumed that the collocated systems arefully connected, i.e., the propagation loss and the radio cap-ture effects are not taken into account. The blocking periodincludes the waiting period until the start of the successfulburst.

Figure 3 shows the average blocking time as a functionof the number of collocated systems under the assumptionof zero turnaround time. Analytical results are obtainedusing expressions in (12) and (8). Although each systemholds the channel for a shorter period of time, as the num-ber of systems increases, the average blocking time for anindividual system increases. The assumption of a renewalprocess in section 3 yields numerical results that are lessthan 5% different from the simulation results.

Figure 3. Average value of the blocking time as a function of the numberof collocated systems for the nonpersistent etiquette. Analytical results

match well the simulation results.

328 I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette

Figure 4. Average value of the blocking time as a function of the ra-dio turnaround time with two competing systems. Analytical results are

included for the case of zero turnaround time.

Figure 4 shows the average blocking time as a func-tion of the radio receive-to-transmit turnaround time in atwo-system UPCS scenario. Our analytical results corre-spond to the case of zero turnaround time. It can be notedthat the analytical results match well the simulation, es-pecially for the 1-persistent case. Longer turnaround timesmean that the monitoring procedure provides more outdatedinformation about the channel status. The increase in theturnaround time increases the probability of collision whichforces the nodes to reset their deference time limits. Notethat a system continues to be blocked until it has success-fully acquired the channel. It would seem that collisionsshould increase the probability that the blocked system winscontention. However, the effect of increased collision rateappears to have little or no impact on the average blockingperiod up to 50 µsec turnaround time. If the blocking timeis defined as the interval until the blocked system transmitsfor the first time, the curves in figure 3 would have a slightnegative slope. We have chosen 50 µsec in figure 3, sincelonger radio turnaround times were shown to decrease thechannel capacity significantly [5].

The effect of the maximum transmission time on theaverage blocking time and the throughput is shown in fig-ure 5. The etiquette allows a maximum channel occupancyof 10 msec. In the simulations we have retained all thedeference rules the same as prescribed by [4]. The ra-dio turnaround time was fixed at 50 µsec. As the max-imum transmission time is decreased from 10 msec theblocking time decreases sharply while the throughput de-creases only slightly. At the maximum transmission timeof 4 msec the throughput is 0.89 and the average blockingtime is 62 msec. As we decrease the maximum transmis-sion time to 1 msec the throughput starts decreasing morerapidly while the relative decrease of the average block-ing time continues to be the same. For the maximumtransmission time of 1 msec the average blocking time is13 msec.

Figure 5. The average blocking time as a function of the normalizedchannel throughput for a two system scenario obtained by simulation.Successive points are generated by changing the maximum transmissiontime in 1 msec steps. The radio turnaround time is fixed at 50 µsec.Over long time periods the throughput is shared equally between the two

competing systems.

It was shown by simulation in [5] that for a large numberof competing nodes the two versions of the etiquette do notdiffer in terms of throughput-delay performance. Since thenonpersistent version allows users to spend more time in thesleep mode it is considered more favorable. In figure 5 weshow that the decrease in blocking time comes with a price,namely in decreased channel throughput, which both sys-tems equally share. Even for transmission time of 1 msec,when each system has approximately 32% of the maximumchannel throughput, the probability that the blocking timeis greater than 50 msec is approximately 5% (equation (4)).This shows that it is questionable if the asynchronous UPCSetiquette can guarantee delivery of time sensitive traffic,such as voice traffic.

It should be noted that in the analysis and the simulationwe have assumed that the two systems contend in the sameway as two users would contend for the channel in UPCS.If, however, more users from each of the two systems wereallowed to contend for the channel we would expect theblocking time per system to decrease.

5. Discussion

In this paper we have presented analytical and simulationresults of the blocking time for two or more collocated sys-tems sharing the spectrum under the asynchronous UPCSetiquette. We have shown that presently the UPCS eti-quette allows one system to be blocked for approximately160 msec on the average. This is clearly unacceptable forvoice transmission and other delay sensitive traffic. Withsome simple changes, such as reducing the maximum trans-mission time from 10 msec to 1 msec, the average blockingtime is decreased as well; however, long blocking timesstill occur with non-negligible probability. Some of the

I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette 329

lessons learned from the design and analysis of the UPCSetiquette will be useful for the design of the future rules ofspectrum sharing. Clearly, more work is needed for betterunderstanding of the design of fair etiquettes that guaranteebounded blocking time.

Acknowledgements

The authors would like to thank Tom Freeburg, RileyJackson and Paul Odlyzko of Motorola, MIMS Radio Re-search Lab for their support in the course of this project.

Appendix

We now analyze the 1-persistent UPCS etiquette. A sam-ple path of the protocol is shown in figure 6. As we havealready noted, if the channel was detected busy during themonitoring interval the blocked system (B) continues tomonitor the channel until it becomes clear. At this time arandom deference period is scheduled. Each time a systemdetects a busy channel the upper limit of the deference timeis doubled starting from 0.75 up to the value of 12 (msec).

Assume that shortly before time t = 0 a busy periodends and that in the next transmission cycle (following theidle period at t = 0) system B wins contention. Shortlybefore t = 0 system A has drawn its deference intervalfrom the distribution U (0.05, 0.75). Since busy periods last10 msec, system A must have started monitoring duringthe first busy period. Thus, the probability, p1, that sys-tem A schedules monitoring during the first busy period isequal to 1. At the end of the first busy period (B’s trans-mission) system A draws a new deference interval fromthe distribution U (0.05, 1.5). The probability that systemA starts monitoring during the second busy period is equalto the probability that it will lose contention after the firsttransmission cycle, i.e.,

p2 = p1P [X1.5 > Y0.75]

= p1

∫ 0.75

y=0

∫ 1.5

x=y

fY0.75(y)fX0.75(x) dx dy

= 0.758621. (13)

Similarly, we have the probabilities of system A schedulingmonitoring during the 3rd, 4th and the 5th busy period givenby

p3 = p1p2P [X3 > Y0.75] = 0.668615,

p4 =

(3∏i=1

pi

)P [X6 > Y4] = 0.629285, (14)

p5 =

(4∏i=1

pi

)P [Y0.75 < X12 < Y0.75 + 10] = 0.526598.

At the end of the kth (k > 4) busy period system Astarts drawing deference periods from the distributionU (0.05, 12). It is possible for system A to schedule the

Figure 6. A sample path of the 1-persistent version of the etiquette. Sys-tem A observes the channel busy and continues to monitor until system Bstops transmitting. At that time system A defers for a random period X1.5.

next monitoring during the (k + 1)st and (k + 2)nd busyperiod, where k > 4. Thus, we can express the probabilityof monitoring during the (k + 2)nd interval as follows:

pk+2 = bpk + apk+1 (15)

for k > 4, and where

a= P [Y0.75 < X12 < Y0.75 + 10] = 0.83682,b= P

[X12 > Y0.75 + 10 + Y ′0.75] = 0.100418.

(16)

The solution of the difference equation (11), which uses theinitial conditions p4 and p5, is given by

pk = Aqk−41 +Bqk−4

2 , (17)

where k > 4. The constants are A ∼= 0.565457, B ∼=0.0638178, q1

∼= 0.943277 and q2∼= −0.106457. Note

that∑∞k=1 pk > 1 due to the way the probabilities pk are

defined, e.g., the event that monitoring is scheduled duringthe kth transmission cycle does not exclude the event thatmonitoring is scheduled during the (k + 1)st transmissioncycle.

Assume that system A monitors the channel during thekth busy period. Then the probability that it wins con-tention in the (k + 1)st transmission cycle is given by

c = P [X12 < Y0.75] = 0.0292887. (18)

Similarly, the probability that system A wins contention inthe (k+ 2)nd transmission cycle after the latest monitoringin the kth busy period is given by

d= P[Y0.75 + 10 < X12 < Y0.75 + 10 + Y ′0.75

]= 0.0334728. (19)

The probability that the blocking period lasts k transmissioncycles is given by

rk = dpk−1 + cpk, (20)

where k > 5. For k < 5 we have the following expressions:

r1 = p1P [X1.5 < Y0.75]

= p1

∫ 0.75

y=0

∫ y

x=0fX0.75 (x)fY0.75(y) dx dy = 0.241379,

r2 = p2P [X3 < Y0.75] = 0.09, (21)

r3 = p3P [X6 < Y0.75] = 0.03933,

r4 = p4P [X12 < Y0.75] = 0.01843.

The probability that blocking will last for one transmissioncycle is four times higher than in the case of nonpersistentetiquette (equation (3)). However, ri’s decrease faster thanthe probabilities in (4) (the PMF of the ri’s has a longer

330 I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette

tail). The average number of transmission cycles spent inthe blocked state is given by

E[Nb] =∞∑k=1

krk =4∑k=1

krk +∞∑k=5

k(cpk + dpk−1)

=3∑k=1

krk +∞∑k=4

(k + 1)dpk +∞∑k=4

kcpk

=3∑k=1

krk + (c+ d)∞∑k=4

kpk + d∞∑k=4

pk

= 13.8175. (22)

We denote the idle period after the kth busy period as Ik,except in the case when the blocked system wins contentionin the (k + 1)st transmission cycle. In the latter case wedenote the last idle period as Lk. Thus, the total blockingtime can be expressed as follows:

Tb = 10Nb +

Nb−1∑i=1

Ii + LNb . (23)

Now we calculate the expected values of Ik and Lk fromthe equation (19). The Lk is the last idle period during theblocking period while the Ik’s are regular idle periods.

The average values of Ik can be calculated as

E[I1] =E[Y0.75 | Y0.75 < X1.5]

=1

P [Y0.75 < X1.5]

∫ 0.75

0

∫ 1.5

y

yfY0.75(y)fX1.5(x) dx dy

= 0.362879. (24)

The idle periods I2, I3 and I4 are calculated in a similarway as in (21). Thus, we obtain

E[I2] =E[Y0.75 | Y0.75 < X3] = 0.384295,

E[I3] =E[Y0.75 | Y0.75 < X6] = 0.392703, (25)

E[I4] =E[Y0.75 | Y0.75 < X12] = 0.39648.

For k > 5 a slightly different expression is used,

E[Ik]|k>5 =a

a+ bE[Y0.75 | Y0.75 < X12 < Y0.75 + 10]

+b

a+ bE[Y ′0.75 | Y0.75 + 10 + Y ′0.75 < X12

]=

a

a+ b

1P [Y0.75 < X12 < Y0.75 + 10]

×∫ 0.75

0

∫ y+10

y

yfX12(x)fY0.75(y) dx dy

+b

a+ b

1P [Y0.75 + 10 + Y ′0.75 < X12]

×∫ 0.75

0

∫ 0.75

0

∫ 12

y+10+z

yfX12(x)fY0.75(y)

× fY0.75(z) dx dy dz

= 0.393211, (26)

where a and b are defined in (12). Similarly, we calculatethe expected values of Lk,

E[L1] =E[X1.5 | Y0.75 > X1.5]

=1

P [Y0.75 > X1.5]

∫ 0.75

0

∫ y

0xfX1.5 (x)fX0.75 (y) dx dy

= 0.2833 (27)

and similarly for the case of L2, L3 and L4. Finally, fork > 5 we obtain

E[Lk]|k>5 =a

a+ bE[X12 | Y0.75 > X12]

+b

a+ bE[X12 − Y0.75 − 10 | Y0.75 + 10

< X12 < Y0.75 + 10 + Y ′0.75]

=a

a+ b

1P [Y0.75 > X12]

×∫ 0.75

0

∫ y

0xfX12 (x)fY0.75(y) dx dy

+b

a+ b

× 1P [Y0.75 + 10 < X12 < Y0.75 + 10 + Y ′0.75]

×∫ 0.75

0

∫ 0.75

0

∫ y+10+z

y+10(x− 10− y)fX12(x)

× fY0.75(y)fY ′0.75(z) dx dy dz

= 0.2833. (28)

Thus, it turns out that E[Lk] = 0.2833 for all k = 1, 2, . . . .This should not be surprising since Lk’s are positive uni-form random variables conditionally smaller than a uniformvariable with a smaller range (X0.75).

The average length of the blocking period is given by

E[Tb] = 10E[Nb] +E

[Nb−1∑i=1

]+E[LNb]

= 10E[Nk] +∞∑k=1

(k−1∑i=1

E[Ii]

)rk +

∞∑k=1

E[Lk]rk

= 143.391. (29)

References

[1] IEEE85b, The Institute of Electrical and Electronics Engineers, Car-rier Sense Multiple Access with Collision Detection (CSMA/CD) Ac-cess Method and Physical Layer Specifications, American NationalStandard ANSI/IEEE Std 802.3 (1985).

[2] S.M. Ross, Stochastic Processes (Wiley, New York, 1982).[3] D.G. Steer, Coexistence and access etiquette in the United States un-

licensed PCS band, IEEE Personal Communications (Fourth Quarter1994).

[4] United States Code of Federal Regulations, Title 47, Part 15(d).[5] I. Vukovic, The impact of Receive-to-Transmit (RT) turnaround time

on network performance in UPCS, presented at ANSI C63.7 (July1995).

I. Vukovic, J. McKown / Spectrum sharing under the UPCS etiquette 331

Ivan Vukovic obtained his Dipl. Ing. in electri-cal engineering from the University of Belgrade,Yugoslavia, and the Ph.D. in electrical engineeringfrom Rensselaer Polytechnic Institute, Troy, NewYork. He joined Motorola, Wireless Data Group,in 1994 where he worked on design, performanceanalysis and simulation of high speed wireless dataand multimedia networks. He is currently a mem-ber of MIMS Radio Research Lab, where he is alsoinvolved in standardization activities for wireless

systems both in the US and in Europe.E-mail: w11421@email.mot.com

John McKown has worked on wireless networks for Motorola since 1989,first in the Wireless Data Group and currently in the Space and SystemsTechnology Group.E-mail: p30014@email.mot.com