A novel co-existence algorithm for unlicensed variable power devices

In the unlicensed spectrum, any device is free to transmit without a license. Such a spectrum has significant benefits, but serious challenges must first be overcome. Foremost is the risk of drastic performance degradation due to a lack of incentive to conserve shared resources. Previous work has identified this problem for devices that transmit for longer duration than necessary. This paper demonstrates this problem for devices that always transmit at maximum power to maximize throughput. For devices that can vary transmission power, the problem is solved if devices reduce transmission power when received interference exceeds defined thresholds. We propose a co-existence algorithm to optimize system throughput when each of two such devices can transmit up to the maximum power allowed on a given channel. We show the device performance with current unlicensed band regulations is rarely optimal, and that the proposed algorithm is better.


INTRODUCTION
In unlicensed spectrum, any device is free to transmit without a license that implies exclusive access. Although most spectrum has traditionally been licensed [I], the Federal Communications Commission (FCC) has created several unlicensed bands, such as the Industry, Science and Medicine (ISM) bands, the Unlicensed Personal Communication Services (UPCS) band [2], the Unlicensed National Information Infrastructure (UNII) band [3], and the Millimeter Wave band [4]. The UPCS band is govemed by a Spectrum Etiquette (known as the UPCS etiquette) [2,5], which is a set of rules regulating access to spectrum and its usage. Unlicensed spectrum has several benefits. It promotes spectrum sharing, and furthers experimentation and innovation. It facilitates mobility of wireless applications, as no licenses are needed for new locations. It is also suitable for smart environments [6], wherein intelligent devices interact with each others and users, sending user needs and acting accordingly. Three challenges must be overcome to realize such benefits. First, there may be mutual interference, as devices can transmit at will. Second, enforcing efficient utilization is difficult as applications using unlicensed bands may vary greatly. Third, there is little incentive to conserve shared spectrum. Thus, designers may adopt a greedy approach, where the more a device wastes shared spectrum to improve its performance, the more it is greedy. If this is common, the shared resource will be of little use. This phenomenon, referred to as a Tragedy of the Commons [7], Jon M. Peha Professor, Camegie Mellon University peha@stanfordalumni.org; http://www.ece.cmu.edu/-peha/ made the Citizen Band radio service unusable in crowded regions, where users wasted spectrum with high-power transmitters. As the resources consumed by a device depend on transmission duration, bandwidth, and power, it may be greedy in any of these dimensions.
Previous work [8-101 has shown that greed in transmission duration can cause poor spectrum utilization. This paper demonstrates the same problem due to greed in the power dimension. Although all unlicensed bands enforce power limits to reduce interference, without any incentive to reduce power below the limit, greedy devices may transmit at maximum power. Given information (such as power, offered load, and distance) about other devices sharing spectrum, parameters that maximize system throughput can be determined.
Without such explicit information, these parameters can only be chosen by an etiquette using available information, e.g. local noise and received power. Etiquette design is complicated by the diversity of devices. Some devices can vary transmission power, and some cannot. Also, power limits can vary from device to device. For devices that can vary transmission power, we propose the Sharing etiquette that avoids a Tragedy of the Commons. This etiquette optimizes system throughput when each of two such devices can transmit at the maximum power allowed on a channel, and optimizes individual throughput for isolated devices as well.
We use the following approach for performance comparison: We assume devices transmit at powers maximizing individual device throughput. We identify powers at which devices reach equilibrium, and compare system and individual throughputs at each equilibrium with the optimal throughput, throughput with UPCS etiquette, and with no etiquette. We show performance with current regulations in unlicensed bands is rarely optimal, and that the proposed etiquette performs better.
Section 2 presents our model to analyze greed in transmission power. Section 3 covers performance in unlicensed bands without an etiquette. Section 4 defines optimal performance of two devices sharing spectrum. Sections 5 and 6 discuss the UPCS and the Sharing etiquette respectively. Section 7 compares performance of existing and proposed etiquettes. Section 8 presents our conclusions. 11  and ~~~i~~ j is given by the propagation factor a,. we assume symmetry in propagation loss from one network to another, i.e. a,=a,=a. Since there will be devices that between a device and its basestation is large relative to the path loss between its basestation and the interfering device, device throughput can degrade drastically. For devices with unequal power limits, it is optimal for the device with higher power limit to transmit at a power greater message error probability E, = exp(-c@,) where c is a constant, as is appropriate for DPSK (Differential Phase Shift Keying) or non-coherent FSK (Frequency Shift Keying) modulation. Device i has offered load G, (the sum of loads from amving and retransmitted messages) and throughput than that of the other device [l I]. Thus, there is an inherent tradeoff between maximizing throughput and fairness for devices with power limits.

V. THE UPCS ETIQUETTE
The UPCS etiquette specifies a power limit and system throughuut as devices vam these parameters: etiquette enforces a "Listen Before Talk" (LBT) rule, ( We now determine equilibria with the Sharing etiquette. Without loss of generality, devices are numbered such that y12y2 . For any a , there can be either one or two stable equilibria. For a<a,/y,, each device transmits at maximum power, and this is the sole equilibrium. For asly15cx<a,ly2 , there is only one equilibrium: Device 1 transmits at maximum power. For a>a, /y2 , either Device 1 or Device 2 has maximum power at equilibrium. For a2asly2, equilibrium is also possible with both devices below maximum power, but this is unlikely. (The instant either device stops transmission, the other transmits at maximum power.) Device performance is characterized by up to three regions of a (see Figure 1): For a: a*s I yl , both devices transmit at maximum power.
For a: a, I y, lala, I y2 , Device 1 transmits at maximum power and Device 2 below its maximum power Ps , which decreases as a increases.
Although system throughput is optimized, the etiquette is not fair to the devices in this region. For a:a,ly,<all, it is equally likely for either device to transmit at maximum power and the other below maximum power. Device 2 can have better throughput only when a>as I y2 . However, with Device 2 at maximum power, Device 1 throughput falls sharply as a increases. Although the etiquette is fair, system performance can be far from optimal. This region does not exist if either device has y<a, , as the other always receives power below its threshold. As devices with low power limits have poor performance relative to those with high power limits, multiple unlicensed bands (each with a smaller power limit range) might be better than a single band for devices with a wide range of power limits.  Figure 2 shows these results for yt =y2 =1 . UPCS, has p = 1 . 2 6~1 0 -~, UPCS2 has b=10-6andUPCS3 has p = l .
For devices with unequal power limits, the equilibria with Device 1 at maximum power and the equilibria with Device 2 at maximum power are equally likely with the Sharing and the UPCS etiquette. Figure 3 compares performance by averaging the throughputs at the two equilibria for yl=l and y2=0.6. The Sharing etiquette is better than others over a wide range of a , except for a: as ly2<all where the device with higher power limit can transmit below maximum power. We conclude that the Sharing etiquette provides the best overall performance.

VIII. CONCLUSION
Unlicensed spectrum has several advantages. However, with little inherent incentive to conserve spectrum, designers may adopt greedy strategies, where the more a device wastes shared spectrum to improve performance, the more it is greedy. Devices may display greed in transmission duration, bandwidth, or transmission power. Previous work [8-IO] has shown severe performance degradation due to greed in transmission duration, and suggested solutions. This paper explores greed in transmission power. We show that in bands with power limits only, devices would maximize throughput by always transmitting at maximum power, which also optimizes system throughput when devices are far apart. When devices are near, transmitting at maximum power leads to suboptimal performance. The solution lies in a properly designed etiquette (i.e. a set of rules regulating spectrum usage). This paper proposes the Sharing etiquette, which decreases the maximum power allowed as received power increases beyond a threshold. This etiquette optimizes throughput of isolated devices, and system throughput as well when each of two devices can transmit at the maximum power allowed on a given channel. We demonstrate that this etiquette performs better than current regulations. We show system performance can be improved by discriminating between devices based on transmission power, and by creating multiple unlicensed bands, each band catering to devices with a small range of power limits.  We now show T;=P2=Pn=KI(N+aP,) is a stable equilibrium.
If Device 1 reduces power to &<cl, R l = N +~n = K I P n < K l~.
If y2Pmax<Pn<ylPmax, then there exists one stable equilibrium P2=y2Pmax ,S=KI(N+aP2) . This occurs when either device is already transmitting at maximum power before the other gains access. If J =yl PmaX , Device 2 has maximum allowed power P2=K/(N+@,) .
Device 1 then receives powerRI=N+aKI(N+olP,). Device 1 can continue transmitting only if its received power is less than its threshold T,=KIT; , i.e., if aPl +NPl-K<Q, which is not true as Fi=y,P,,,>P,.
Device 1 will therefore reduce power as long as e > P n , and Device 2 will increase power. As y2Pmax<Pn, devices reach equilibrium with P2=y2Pmax and Device 1 at P,=KI(N+aP2).