Ad Hoc Networks: to Spread Or Not to Spread?

ANDREWS LAYOUT 11/16/07 1:37 PM Page 84

AD HOC AND SENSOR NETWORKS

Ad Hoc Networks: To Spread or Not to Spread?

Jeffrey G. Andrews, University of Texas at Austin Steven Weber, Drexel University Martin Haenggi, University of Notre Dame

ABSTRACT such networks will often have at least some con- nections to wired infrastructure, but we neglect Spread spectrum communication — often this for simplicity of discussion. called code-division multiple access — has been Spread spectrum transmission has long been widely adopted over the years for many types of considered attractive for ad hoc networking for a interference-challenged wireless communication number of reasons, including security and inter- systems including cellular and cordless tele- ference robustness [2, 3]. In this article we care- phones, wireless LANs and PANs, military appli- fully scrutinize the supposed advantages of using cations, and global positioning systems. In this spread spectrum — also known as code-division article we explore whether CDMA, in either its multiple access (CDMA) — in ad hoc networks. frequency hopping (FH) or direct sequence (DS) Similar to the highly contentious CDMA vs. time- form, is an appropriate design approach for division multiple access (TDMA) debate for cel- wireless ad hoc, or mesh, networks. One goal of lular systems, the considerations for ad hoc this article is to help provoke a debate by networks are also laden with subtleties. In cellular explaining the main advantages and disadvan- networks, despite CDMA’s apparent inferiority tages of CDMA in the context of ad hoc net- due to intentional self-interference, the exploita- works as exposed by recent research. We argue tion of voice activity, frequency reuse, and fast that CDMA does not inherently improve the power control were central to the ultimate success spectral efficiency of ad hoc networks; on the of CDMA. Analogously for ad hoc networks, it is contrary, its valued interference averaging effect crucial to adopt a network-level point of view that is not appreciable in ad hoc networks due to the includes considerations such as network capacity, irregular distribution of both the transmitters end-to-end delay, energy efficiency, channel and receivers. On the positive side, both types access, and routing. However, the key traits of (FH and DS) of spread spectrum allow for CDMA in an ad hoc network are very different longer hop distances and a reversal of the usual than in cellular networks with centralized trans- relationship where the desired transmitter must mitters (downlink) and receivers (uplink). We be closer to the receiver than interfering trans- now summarize the key traits of CDMA in ad hoc mitters. These two facts allow for significant networks in terms of pros and cons, which are jus- advantages over narrowband systems in terms of tified in detail in the body of the article. energy efficiency and end-to-end delay. THE ADVANTAGES OF CDMA IN INTRODUCTION AD HOC NETWORKS Applications of wireless ad hoc networks have The advantages of CDMA in ad hoc networks expanded in recent years to include not only are quite different than in cellular networks, and numerous military applications and emerging can also be distinct for the two different types of wireless sensor networks, but also many other CDMA, frequency hopping (FH) and direct exciting and commercially viable applications sequence (DS). FH and DS are described in including wireless community broadband access, more detail later. backhaul for wireless LAN access points, and Longer hops. DS and FH both allow for range extension for cell-based networks [1]. longer hops to be undertaken for a given net- This research was sup- Despite this high level of interest and commer- work density. This allows more direct routing, ported by NSF grant no. cial potential, many basic ad hoc network design reduced end-to-end delay, and perhaps counter- 0635003 (Weber), no. principles are still not well understood, and one intuitively, reduced energy consumption. 0634979 (Andrews), and important design question is the focus of this Capacity enhancements. Neither FH nor DS the DARPA IT-MANET article: Does it make sense to use spread spec- increases the overall ad hoc network capacity on its program, grant no. trum in ad hoc networks? In this article an ad own, in fact the opposite is true for DS. However, W911NF-07-1-0028 (all hoc network implies communication without the DS allows for the possibility of capacity-increasing authors). assistance of wired infrastructure. Naturally, interference canceling receivers, which are ineffec-

Original signal Received signal Spread spectrum Interference P/No w/prob. (M – 1)/M uses noise-like code B SINR = P/(No + Z) w/prob. 1/M P sequences to effectively increase Z Randomly the bandwidth to be hop

er Freq. hopping between far greater than the w

er M freq. o signal bandwidth. w P slots o P When spread spectrum is used to support multiple Frequency Slot 1 Slot 2 Slot m Slot M users, it is called Frequency CDMA.

I Figure 1. Frequency hopping works by randomly picking one of M frequency slots. A narrowband signal of similar bandwidth is usually avoided as M increases.

tive in ad hoc networks unless DS is used. Both FH CDMA’s viability in ad hoc networks with mobili- and DS also provide considerable frequency diversi- ty or bursty traffic, since these scenarios require ty, which helps overcome narrowband fading. frequent code acquisition and synchronization. Network efficiency. The ability of DS-CDMA systems to successfully operate under low signal- CDMA: A MODERN OVERVIEW to-interference-plus-noise ratio (SINR) permits communication with a larger number of potential INTERFERENCE-LIMITED NETWORKS AND THE interferers, which simplifies network coordination. RECENT SUCCESS OF CDMA Security. Spread spectrum radios have innate security features: they are harder to jam, they Dense wireless networks are by nature interference- make eavesdropping more difficult, and their limited, which means that increasing the transmit presence is more difficult to detect. Although power of all nodes in the network simultaneously important for some applications, these topics are will not substantially increase the overall through- beyond the scope of this article. put of the network. Ad hoc networks pose a particularly challenging interference environment because THE DISADVANTAGES OF CDMA IN the lack of agreed upon centralized transceivers AD HOC NETWORKS means that each receiver in the network must bound the level of interference in its vicinity to suc- A common drawback of CDMA in cellular and cessfully receive the desired transmission. ad hoc networks is that the system bandwidth Spread spectrum uses noise-like code needs to be considerably larger than the (per sequences to effectively increase the bandwidth to user) symbol rates. In ad hoc networks CDMA be far greater than the signal bandwidth. When has two other important drawbacks: spread spectrum is used to support multiple users, Interference averaging is ineffective. Interfer- it is called CDMA. The central tenet of CDMA is ence averaging, the hallmark of both DS- and that designing for time or frequency orthogonality FH-CDMA in cellular networks, does not pay (as in TDMA or FDMA) is not appropriate, since off in ad hoc networks. The key reason is the neighboring (i.e., intercell) interference and other lack of a centralized receiver and the associated imperfections would compromise the orthogonali- power control to that receiver. Global power ty anyway. On the other hand, CDMA tolerates control is impossible in ad hoc networks; instead, all sources of interference within bounds deter- usually just one or perhaps two interfering nodes mined by the spreading gain. Due to its robust- dominate the interference power, which makes ness, system capacity, and other implementation interference avoidance (via scheduling or slow and political/economic factors, CDMA overcame FH) far more effective than interference averag- extreme levels of early skepticism to become the ing (using DS or fast FH to proportionally underlying physical layer technology and multiple reduce the interference level). access scheme for all three important third-gener- Considerable setup costs. CDMA requires ation cellular standards: cdma2000, wideband that both the transmitter and receiver have CDMA (WCDMA), and TD-SCDMA. Based on knowledge of: this success, it is natural to seriously consider the • Agreed upon spreading (DS) or hopping viability of CDMA for the emerging class of ad (FH) sequences hoc and mesh networks. • The current position in the sequence (i.e., time synchronization) FREQUENCY HOPPING AND Acquisition of these is a nontrivial resource-con- DIRECT SEQUENCE CDMA suming process, and unless the cost of code acquisition and synchronization is amortized over time, CDMA techniques have historically been divided the above “pros” of CDMA may not justify this into two very different types of modulation: fre- overhead. In practice this raises questions about quency hopping and direct sequence. In this arti-

IEEE Communications Magazine • December 2007 85 ANDREWS LAYOUT 11/16/07 1:37 PM Page 86

Original signal Received signal Processed received signal B P Interference P SINR = P/No + Z/M um

r Z

ct Spreading Despreading Interference pe s er w

o W = M B P P/M Z/M

Frequency Frequency Frequency

I Figure 2. Direct sequence works by spreading the signal over a larger bandwidth. After processing, the desired narrowband signal re- emerges while other interference is attenuated by a factor of M.

cle CDMA without further qualification refers col- is its reliance on accurate power control. In ad lectively to both of these techniques. FH is depict- hoc networks, however, equal received powers ed in Fig. 1. The total bandwidth W is divided into are impossible due to the random node positions M frequency bands of bandwidth B = W/M. At and the distributed nature of the networks. each hop, the transmitter chooses one of the M Despite their significant differences, DS- and bands based on a pseudorandom code sequence FH-CDMA have many similar properties in that is also known to the receiver. Assuming the power-controlled cellular systems and achieve transmitter and receiver are synchronized, they comparable SINRs for the same system load. Both both hop in unison and are able to successfully effectively “average” interference so that a system communicate. If there are other users in the net- can be designed for the average, rather than worst work, there are occasional collisions when two case, interference. DS is generally preferred for transmitters pick the same frequency band, but by multiuser cellular systems since it has smoother coding over time, it is possible to recover from a interference averaging properties (the received moderate number of collisions. Fast hopping SINR does not fluctuate as much), easily allows refers to hopping on the order of a symbol time, for coherent modulation that gives a 3 dB gain, whereas slow hopping refers to hopping on the and, perhaps most important, allows strong error order of a packet time. Examples of well-known correction coding without sacrificing spreading systems that use FH include Bluetooth, which has gain. In contrast to cellular systems, though, FH and 80 frequency bands of 1 MHz width (M = 80, W DS have very different characteristics when used in = 80 MHz), and a hop interval of 625 µs, and an ad hoc network, and the trade-offs between FH, GSM (which is also TDMA), which has a variable DS, and narrowband signaling are quite different. number of possible frequency bands of width B = 200 kHz and a hop interval of 4.617 ms. Direct sequence, shown in Fig. 2, also involves THE KEY FEATURES OF synchronized pseudorandom codes, but in this CDMA AD HOC NETWORKS case a code sequence with bandwidth W = B ⋅ M is multiplied with the user’s data sequence of The fundamental metrics of interest in an ad hoc bandwidth B, creating a transmitted sequence of network are capacity, end-to-end delay, and bandwidth W. M is called the spreading factor. energy efficiency; these often compete with each By correlating the same code with the received other. In this and subsequent sections we discuss signal, the desired signal is converted back to a how the distinctions between spread spectrum narrowband signal (i.e., bandwidth B), while the and narrowband ad hoc networks affect these noise and interference stay at bandwidth W and performance metrics. hence are attenuated by a factor of approximate- ly M at detection. CAPACITY OR THROUGHPUT In summary, FH systems avoid interference “Capacity” is a suspect metric in an ad hoc net- with increased probability as the number of fre- work, since it is interdependent with delay, trans- quency slots M grows, whereas DS systems sup- mit distance, mobility, scheduling, and press interference by a factor of M. Both FH higher-layer network functionality. To ground the and DS entail some important design considera- discussion, we consider the transmission capacity, tions relative to narrowband transmissions. First, which is the average number of reliable simulta- the transmitter and receiver need to be synchro- neous links that can be active in a unit area under nized and aware of each other’s code sequences. a specified typical outage probability. Transmis- Bluetooth provides a useful practical example of sion capacity is a single-hop capacity metric that how an FH-CDMA ad hoc network can achieve permits precise throughput characterization, both time and code synchronization using a whereas the more commonly used metric of trans- straightforward paging and inquiry procedure. In port capacity [4] is a multihop metric that permits a mobile ad hoc network where this procedure only an asymptotic scaling law of network must be done frequently, this overhead may throughput. It is reasonable to expect that spread nonetheless be considerable. A second distin- spectrum would allow a higher transmission guishing factor of DS-CDMA in cellular systems capacity C to be tightly upper bounded as [5]

86 IEEE Communications Magazine • December 2007 ANDREWS LAYOUT 11/16/07 1:37 PM Page 87

2 ε M ε  M α Prohibited CFH < , C DS < Allowed 2 2 2   (1) transmission transmission πr α πr  β  β D D for a transmit distance of r, outage probability ε, path C C M loss exponent α > 2, spreading factor , and target A B A B SINR β. Here we have assumed that the hopping speed is equal to the packet length (i.e., the duration r r s‘ = s/M1/α over which outage is determined), the frequency- s time slots are orthogonal, signal power decays with distance as d–α, and all active nodes are separated by at least a few wavelengths (this prevents anomalous behavior of the d–α expression for small d). Narrowband system DS-CDMA system Insights. Transmission capacity allows for a remarkably simple expression, which indicates that FH is better than DS by a factor of M1–2/α I Figure 3. The interference range (s) and region (shaded) for narrowband and (e.g., by a factor of √M for the typical value α = DS-CDMA. For NB, if A → B is scheduled, C → D is not allowed since it will 4). Similarly, one could ask if the bandwidth spent cause an outage at B. In CDMA with sufficient spreading factor M, C → D is 1/α on spread spectrum is justified by the interference allowed since the interference range shrinks as 1/M . savings. For DS, the answer appears to be no: the capacity grows sublinearly with M (as M2/α) while the consumed bandwidth grows as M. For FH, where β is the required signal-to-interference there appears to be no fundamental bandwidth ratio (SIR) and r is the distance separating the penalty since both the capacity and bandwidth receiver from its associated transmitter. Note that increase linearly with M. The basic conclusion is the effective SIR requirement under DS is β/M, that interference avoidance (by hopping) is prefer- while the interference radius is independent of M able to interference suppression (by despreading) for FH. However, in FH dominant interferers are in an ad hoc network. This is a byproduct of the removed with probability (M – 1)/M. near-far problem, which is why the gain from When nodes are uniformly spread out in interference avoidance increases for large α. space, it is possible to compute the probability of Caveats. There are a few caveats that should be outage due to the presence of one or more dom- noted before concluding that DS-CDMA has inferi- inant nodes. Setting this outage probability for or capacity in an ad hoc network. First, the above both DS and FH equal to ε and solving for the results implicitly assumed ALOHA-type medium radius r we obtain: access control (MAC), that is, the transmitter loca- 1 1 tions are random and independent of one another. − −log(1−ε) rDS = cM α , rFH = c M , c = β α , (3) A better MAC for DS-CDMA, as seen below, πλ deliberately clusters transmitters and receivers. Sec- ond, a matched-filter receiver was assumed, which where λ is proportional to the number of nodes in achieves an SINR gain of M, which is known to be the network. This expression can be interpreted as highly suboptimal (in theory) relative to a multiuser saying that the maximum distance separating suc- interference-canceling receiver. Finally, bandwidth cessful transmitter-receiver pairs scales in the efficiency may not be the key concern in all applica- spreading factor like M1/α under DS and like √M tions. In ultra wideband (UWB) communication, under FH. Given α > 2, this result demonstrates robustness to interference may be more important that FH allows longer hops to be taken than DS or than absolute bandwidth efficiency. Nevertheless, it narrowband, all else being equal. That said, the should be conceded that this initial evidence sug- expression for sDS demonstrates that sDS < r for gests DS-CDMA is not as promising for ad hoc sufficiently large M, that is, successful transmission networks as it was for cellular. under DS requires coordination only with nodes that are closer to the receiver than the transmitter. THE INTERFERENCE RANGE A key difference between spread spectrum and The interference range is a common concept in narrowband transmission can be observed from the understanding ad hoc networks; it is defined as the above relations: with sufficient spreading gain, the minimum distance s such that any interfering node transmission range can be greater than the inter- within s of a receiver will by itself generate sufficient ference range. In contrast, the interference range is interference power to cause an outage at that receiv- larger than the communication range in both nar- er. Conceptually, it is simplest to think of the inter- rowband and FH, assuming β > 1, which is invari- ference range as a disk around the desired receiver, ably true for all but the lowest conceivable data as shown in Fig. 3, but in reality it is an irregular rates. In short, spread spectrum transmission has contour due to random channel effects. We call the interesting property of changing the usual rela- interferers within the interference range dominant; it tionship of s > r to s < r, which has a number of can be shown that, to a first order, many perfor- important implications to be highlighted shortly. mance metrics of interest may be obtained through consideration of dominant interferers alone. The MEDIUM ACCESS CONTROL FOR SPREAD interference radii under FH and DS are SPECTRUM NETWORKS 1 1 DS  β α FH Given these unique traits of spread spectrum — s = r  , s = rβα , (2)  M  its superior allowable transmission density and increased interference-to-communication-range

IEEE Communications Magazine • December 2007 87 ANDREWS LAYOUT 11/16/07 1:37 PM Page 88

Narrowband FH-CDMA network(2 subchannels are shown) DS-CDMA

Scheduled TX Scheduled TX Scheduled TX Scheduled RX Scheduled RX Scheduled RX

I Figure 4. Sample transmit/receiver pairs with near-optimal scheduling in narrowband, FH and DS-CDMA networks, from [7]. Nar- rowband systems require isolated Tx-Rx links. FH (center) allows some collocated links:disks are shown around the users in subband 1, no disks are around users in subband 2. DS throughput is maximized when transmitters and receivers cluster together, as seen at right.

ratio — it is clear that spread spectrum MAC since the use of quasi-orthogonal code sequences should be designed differently than for a nar- is insufficient to suppress nearby interferers. rowband (NB) system. On the negative side, The most popular contention resolving MAC spread spectrum systems are burdened with con- is carrier sense multiple access (CSMA), but its siderable overhead in terms of exchanging and popular implementation in wireless networks is synchronizing their code sequences. highly spectrally inefficient, especially for DS, since it inhibits nodes around both the transmit- Frequency Hopping MAC Design — By its ter (which does the sensing) and the receiver, very nature, FH provides a high probability of which responds to a request to send (RTS) mes- interference avoidance, which is why it achieves sage with a clear to send (CTS). An efficient high capacity even in dense networks. Thus, once MAC would only inhibit transmissions near the the transmitter and receiver have acquired each receiver. In fact, as shown in Fig. 4, clustered other and are synchronized, the channel access transmitters and receivers are highly desirable in part of the FH MAC is extremely simple since it a CDMA ad hoc network, and an optimal DS need not perform contention resolution among MAC will result in a large degree of clustering collocated concurrent transmitters. One way to [7]. Therefore, a better approach for a DS MAC further improve the spectral efficiency of FH is to is for the receiver to instruct neighbors in its use adaptive spectrum sensing (similar to cogni- vicinity to suppress their transmissions. In con- tive radio) or adaptive FH, where the hopping trast to NB systems, this explicit coordination is sequences of nearby nodes are acquired to pre- straightforward in DS systems, since the commu- emptively avoid collisions. The trade-off incurred nication range is longer than the interference for these benefits is that before any communica- range. Therefore, a DS receiver can communicate tion can take place, the channel hopping sequence with all its potential strong interferers. This large and present state must be agreed on, which con- advantage again must be traded off with the sumes significant time and bandwidth resources. overhead inherent in code acquisition and synchronization, which, as in FH, is particularly bur- Direct Sequence MAC Design — Because densome in a mobile network. Typically, DS code nearby interferers cause very strong interference, acquisition is achieved with progressively more simply attenuating this interference by M is not wideband pilot signals, although another option particularly effective, as observed in Eq. 1. In is to use a separate NB control channel for this cellular networks this problem can be mitigated purpose. One considerable drawback of the NB using centralized power control, but in ad hoc control channel is that it is subject to NB fading, networks power control is impractical since: interference, and all the other impediments that • There is no centralized authority to coordi- motivate spread spectrum in the first place. nate the required power levels. • Even perfect power control would not pre- HE DVANTAGES OF ONGER OPS clude excessive interference at some receiv- T A L H ing nodes due to the network geometry. It is commonly assumed that capacity is maxi- This necessitates contention resolution in both mized and energy consumption minimized by DS and NB ad hoc networks. This irreconcilable routing through the nearest neighbors. The argu- strong interference problem is a frequent criti- ment is that shorter-range transmissions cause cism against DS-CDMA in ad hoc networks, but less interference, permitting better spatial reuse. this is a misconception since NB systems suffer For example, if the transmission range for each even more drastically from nearby interferers hop in a route were reduced by half, the effective [6]. The key point is that both DS and NB sys- area of interference would decrease by 22 = 4, tems require scheduling or contention resolution while the number of hops would only double.

88 IEEE Communications Magazine • December 2007 ANDREWS LAYOUT 11/16/07 1:37 PM Page 89

Hence, the overall network capacity would • Increases the chance of packet loss increase if all nodes reduced their transmit range. • Makes bottlenecks or traffic jams more like- In summary, since A further supposed advantage of nearest-neigh- ly to occur bor routing is its improvement in energy efficien- • Increases the delay variance, making delay spreading increases cy. For a path loss of dα, where d is the link guarantees difficult the allowable hop distance, halving the transmission range would Longer hops also reduce the time for route distance for the reduce the per-hop energy consumption by 2α discoveries; this advantage becomes more signifi- and the total energy over the two hops by 2α–1. cant as node mobility increases. same network While in some cases it may indeed be beneficial In summary, since spreading increases the density, FH and DS to route through the nearest possible neighbors, allowable hop distance for the same network the above two arguments are simplistic in the face density, FH and DS both appear to be very both appear to be of important considerations regarding capacity, attractive means of increasing energy efficiency very attractive means energy consumption, and end-to-end delay. As was and reducing end-to-end delay. Even if nearest of increasing energy argued in the previous section, DS spread spec- neighbor routing may turn out to maximize the trum allows longer hop ranges by a factor of M1/α, throughput in many configurations, there may be efficiency and and FH by a factor of √M. Since an understanding many scenarios where delay and energy con- reducing of the effect of hop and route lengths is central to sumption are more pressing requirements. a debate on CDMA’s merits, we now apply spread CDMA ad hoc networks thus provide an addi- end-to-end delay. spectrum to some of the arguments of [8] and tional means of adaptation; for a modest sacrifice explain the key reasons that long hops are often in spectral efficiency, FH and DS allow longer preferable, which is a potentially significant advan- hops, so with the need for decreased power or tage of spread spectrum over NB transmission. delay (i.e., low batteries, real-time applications), spreading can be of pivotal assistance. Capacity — Although increased capacity is ostensibly the largest advantage of short hops, it is not at all clear that nearest-neighbor routing IMPROVING AD HOC actually results in the best observed network ETWORK APACITY throughput. Shorter hops at low power result in N C more transmissions, and hence lower interference A fundamental difference between DS-CDMA levels, but for longer time periods. On the other and both FH and NB is that the latter techniques hand, if all active transmitters increase their are not designed to tolerate any co-channel trans- transmit power by a constant factor, the link missions, but instead avoid them by either SINRs can only increase, since the noise term scheduling (NB) or hopping (FH). DS-CDMA becomes negligible. Hence, it is not obvious that can also benefit substantially from scheduling, as many low-power transmissions are always superi- well as from sophisticated receivers that cancel or to fewer high-power transmissions. In order to co-channel interference. In this section we explore maintain a given end-to-end data rate (spectral the possible gain from each of these techniques, efficiency), the per-hop rate needs to increase as and see that in both cases DS systems are better the number of hops increases. This leads to an suited to take advantage of their gains. optimum number of hops that is considerably less than the maximum number of hops [9]. ADVANCED RECEIVERS Multiuser receivers have been widely studied in Energy Efficiency — We give three reasons that academia, and the principal approaches are well long hops are preferable from an energy perspec- summarized in [11]. However, these receivers have tive. First, the logic that shorter hops require less never really caught on in industry, primarily due to transmit power and hence reduce power con- their complexity for large numbers of users, incom- sumption is suspect, since this assumes that reduc- patibility in actual wireless channels, and adversari- ing the distance by a factor x reduces power al relationship with error correction codes [12]. consumption at the transmitter by xα. This is an Multiuser detection is actually more attractive in oversimplified energy model; from a power ampli- ad hoc networks than in cellular systems due to the fier efficiency standpoint it is far preferable to large benefit attainable from canceling just a few send at the maximum linear operating point and interfering nodes. Interestingly, only DS-CDMA route as far as possible [10]. Backing off from this systems stand to benefit from most achievable mul- power level does not significantly reduce the cur- tiuser detection techniques, since only interferers rent drawn by the power amplifier. Also, reducing stronger than the desired signal can typically be the transmit power implies additional overhead cancelled. In DS systems, this is sufficient since all due to power control and variable power trans- interferers (in particular those farther away than mission. Second, nearest neighbor routing the desired transmitter) are attenuated by the requires many nodes to perform routing when spreading factor, and all those strong enough to they could otherwise go into sleep mode (which cause an outage are within communication range, consumes 1 percent or less the power of being as discussed earlier. However, in NB or FH sys- “awake”). Third, nearest neighbor routing reduces tems, these further interferers can still cause an the network’s ability of load balancing, since most outage, but since they are outside of communica- nodes need to participate in the routing process. tion range, it is nearly impossible to cancel them. Successive interference cancellation (SIC) is Delay — The fact that short-hop routing incurs an appropriate type of multiuser detection for ad more delay than long-hop routing is not disput- hoc networks, given its theoretical optimality and ed. It is typically assumed that delay is propor- amenability to implementation when the number tional to the number of hops. Even this is of nodes to be cancelled is small. The transmis- optimistic, since employing many short hops: sion capacity of imperfect SIC is shown in Fig. 5

IEEE Communications Magazine • December 2007 89 ANDREWS LAYOUT 11/16/07 1:37 PM Page 90

zone that should exist around each receiver; that 1 is, the region around an active receiver that Perfect SIC, UB should be cleared of interfering transmissions by Imperfect SIC, UB ζ = 1/100 Imperfect SIC, UB ζ = 1/10 a higher layer protocol. There is a trade-off between protecting active receivers from outages (larger guard zone) and maximizing the number 0.1 ) of concurrent transmissions (smaller guard c

( zone). Scheduling in ad hoc networks is a multi- ty faceted and rich research topic [14, references aci p therein], but a simple way to observe the gain 0.01 possible due to scheduling in ad hoc networks is to consider the optimum guard zone’s effect on capacity [15]. Although guard zone scheduling is not optimal, it does require only local coordina- ransmission ca T tion among nodes and provides a clear connec- 0.001 tion with the network geometry concepts that are the focus of this article. Modeling a guard zone around a receiver is conceptually similar to perfect interference can- 0.0001 cellation; that is, preventing nodes from transmit- 0 5 10 15 20 ting is like canceling them perfectly beforehand, Maximum number of cancellable nodes (K) with the important distinction that since nearby nodes are prohibited from transmitting, the num- I Figure 5. SIC's effect on transmission capacity. As the number of cancellable ber of concurrent transmissions is less than in perfect SIC. In Fig. 6 the transmission capacities nodes K increases, the additional gain from SIC is essentially zero, even with are compared for two systems. The first has a very accurate interference cancellation, that is, small ζ, where ζ is the fraction of interference power left for each user after cancellation. scheduling algorithm that bans nodes from transmitting within a guard zone radius RGZ around each desired receiver. The second system instead has SIC with varying levels of cancellation accura- vs. the number of cancelled nodes [13]. Some key cy. Guard zones are as effective as 90 percent conclusions, which apply to any type of imperfect accurate SIC when the outage constraint is ε = .1, interference suppression in ad hoc networks, are: and much more effective than SIC at low spread- • Similar to cellular, perfect SIC increases the ing gains or severe outage constraints. The former capacity by perhaps an order of magnitude. is because at low spreading gains, the dominant Good news: large potential throughput gain. interferer is likely to be outside the communica- • Unlike cellular, most of the gain is achieved by tion range and thus impossible to cancel via SIC. just canceling the one or two dominant interferers. Good news: low complexity and latency. ONCLUSIONS • Unlike cellular, the interference cancella- C tion is exceptionally sensitive to the amount In the next decade, applications for autonomous of residual interference. Bad news: channel wireless networks are likely to increase dramatical- estimation must be extraordinarily accurate. ly, and ad hoc networks — as well as their close The key fact is that the residual interference relatives, mesh and sensor networks — will become of the strongest interferer is usually more impor- significant components of the wireless ecosystem. tant than the full interference of the other inter- Superficially, spread spectrum transmission will ferers. Residual interference is inevitable, and appear very attractive for many of these applica- results primarily from imperfect estimates of the tions due to its well-known interference robustness channel amplitude and phase, which is required and security features. This article is intended to for reconstructing the signal to cancel [12]. This deepen understanding of spread spectrum’s advan- is why even for relatively accurate interference tages and disadvantages in the context of ad hoc cancellation (in practice 90–95 percent of the networks. As we have seen, many of the design interference cancelled for each node would be issues are quite distinct from cellular networks. considered a success, i.e., ζ = 0.05–0.1), there is We have argued that fundamentally, inter- no discernible benefit to canceling more than 1 ference averaging is not nearly as profitable in node. Therefore, although the large potential ad hoc networks as it is in cellular networks benefits of multiuser detection are unique to DS due to the very different geometric properties ad hoc networks, designers should approach of the transmit/receive positions. Therefore, claims of huge gains with skepticism. frequency hopping — interference avoidance — should generally be preferred to direct SCHEDULING sequence spread spectrum — interference aver- MAC scheduling is critical to the efficient opera- aging. We have also noted that both FH and tion of DS-CDMA ad hoc networks. The key DS incur considerable overhead in code acqui- difference between DS scheduling and NB and sition and synchronization, and this overhead FH scheduling is that due to DS’s interference needs to be amortized to make spread spec- suppression margin, receivers should be clus- trum competitive. Unless new efficient schemes tered close to each other, as should transmitters, can be developed, this trait discourages the use as discussed earlier. Although we previously dis- of spread spectrum in ad hoc networks with cussed the interference range of DS ad hoc net- high levels of mobility or bursty traffic. On the works, we did not quantify the optimum guard positive side, both FH and DS provide consid-

90 IEEE Communications Magazine • December 2007 ANDREWS LAYOUT 11/16/07 1:37 PM Page 91

ε = .1 ε = .01 100 100 Guard zone Guard zone SIC w/RA SIC w/RA

ζ = 0 φ φ e, e, us us re re

l ζ = 0 l ia ia t t

pa 10 pa 10 s s ζ = .01 ζ = .01 of of o o i i t t a a R R

ζ = .1

1 1 0 10 20 30 40 50 0 10 20 30 40 50

Spreading gain, M Spreading gain, M

I Figure 6. Guard zone performance vs. SIC as measured by the efficiency of spatial reuse φ, for moderate (ε = .1) and severe (ε = 0.01) outage constraints. The baseline case of no spreading, scheduling, or SIC, is φ = 1.

erable flexibility to the network by allowing [11] S. Verdu, Multiuser Detection, Cambridge Univ. Press, 1998. longer hop lengths and, in the case of DS, a [12] J. G. Andrews, “Interference Cancellation for Cellular reduced interference range. These aspects sim- Systems: A Contemporary Overview,” IEEE Wireless plify some protocol design aspects and, perhaps Commun., vol. 12, no. 2, Apr. 2005, pp. 19–29. most important, allow end-to-end delay and [13] S. Weber et al., “Transmission Capacity of Wireless Ad Hoc Networks with Successive Interference Cancella- energy consumption to be reduced, perhaps tion,” IEEE Trans. Info. Theory, vol. 53, no. 8, Aug. substantially. 2007, pp. 2799–2814. [14] L. Georgiadis, M. J. Neely, and L. Tassiulas, “Resource ACKNOWLEDGEMENTS Allocation and Cross-Layer Control in Wireless Net- works,” Foundations and Trends in Networking, vol. 1, The authors extend thanks to Drs. G. de no. 1, 2006, pp. 1–144. Veciana, A. Hasan, and X. Yang for their contri- [15] A. Hasan and J. G. Andrews, “The Guard Zone in Wire- butions to several of the ideas in this article. less Ad Hoc Networks,” IEEE Trans. Wireless Commun., vol. 6, no. 3, Mar. 2007, pp. 897–906. REFERENCES BIOGRAPHIES [1] I. F. Akyildiz, X. Wang, and W. Wang, “Wireless Mesh Networks: A Survey,” Comp. Networks, vol. 47, no. 5, JEFFREY ANDREWS ([email protected]) is an assistant Mar. 2005, pp. 445–87. professor of electrical and computer engineering at the Uni- [2] M. B. Pursley, “The Role of Spread Spectrum in Packet versity of Texas at Austin. He received a B.S. with high dis- Radio Networks,” Proc. IEEE, vol. 75, no. 1, Jan. 1987, tinction from Harvey Mudd College in 1995, and M.S.E.E. pp. 116–34. and Ph.D. degrees from Stanford University in 1999 and [3] T. J. Shepard, “A Channel Access Scheme for Large 2002. He serves as an editor for IEEE Transactions on Wire- Dense Packet Radio Networks,” ACM SIGCOMM, Stan- less Communications, and has industry experience at Qual- ford Univ., Aug. 1996, pp. 219–13. comm, Intel, Palm, NASA, and Microsoft. He received the [4] P. Gupta and P. Kumar, “The Capacity of Wireless Net- NSF CAREER award in 2007 and is the Principal Investigator works,” IEEE Trans. Info. Theory, vol. 46, no. 2, Mar. of an eight-university team of 13 faculty in DARPA’s Infor- 2000, pp. 388–404. mation Theory for Mobile Ad Hoc Networks program. [5] S. Weber et al., “Transmission Capacity of Wireless Ad Hoc Networks with Outage Constraints,” IEEE Trans. STEVEN WEBER ([email protected]) is an assistant pro- Info. Theory, vol. 51, no. 12, Dec. 2005, pp. fessor in the Department of Electrical and Computer Engi- 4091–4102. neering at Drexel University, Philadelphia, Pennsylvania. He [6] A. Muqattash and M. Krunz, “CDMA-Based MAC Proto- received a B.S. from Marquette University in 1995, and M.S. col for Wireless Ad Hoc Networks,” ACM SIGMOBILE, and Ph.D. degrees from the University of Texas at Austin in June 2003, pp. 153–64. 1999 and 2003, respectively, all in electrical engineering. [7] X. Yang and G. de Veciana, “Inducing Spatial Clustering His research interests are focused on wireless networking, in MAC Contention for Spread Spectrum Ad Hoc Net- stochastic geometry, and ad hoc network capacity. works,” ACM MobiHoc, May 2005. [8] M. Haenggi and D. Puccinelli, “Routing in Ad Hoc Net- MARTIN HAENGGI ([email protected]) is an associate profes- works: A Case for Long Hops,” IEEE Commun. Mag., sor of electrical engineering at the University of Notre vol. 43, no. 10, Oct. 2005, pp. 93–101. Dame. He received Dipl. Ing. (M.Sc.) and Ph.D. degrees [9] M. Sikora et al., ”Bandwidth- and Power-Efficient Rout- from the Swiss Federal Institute of Technology in Zurich ing in Linear Wireless Networks,” Joint Special Issue of (ETHZ), Switzerland, in 1995 and 1999, respectively. In IEEE Trans. Info. Theory and IEEE Trans. Networking, 1999–2000 he spent a postdoctoral year at the University June 2006, pp. 2624–33. of California at Berkeley. He received an NSF CAREER award [10] M. Haenggi, “The Impact of Power Amplifier Charac- in 2005 and is a member of the Editorial Board of Elsevi- teristics on Routing in Random Wireless Networks,” er’s Journal of Ad Hoc Networks. His scientific interests Proc. IEEE GLOBECOM, San Francisco, CA, Dec. 2003, include networking and wireless communications, with an pp. 513–17. emphasis on ad hoc and sensor networks.

IEEE Communications Magazine • December 2007 91