An Efficient UWB Radio Architecture for Busy Signal MAC Protocols

Transcription

1 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA An Effcent UWB Rado Archtecture for Busy Sgnal MAC Protocols Nathanel J. August and Dong Sam Ha Vrgna Tech VLSI for Telecommuncatons (VTVT) Lab Bradley Department of Electrcal and Computer Engneerng Vrgna Tech, Blacksburg, VA USA Phone: ; Emal: {nateaugu, Abstract Large wreless ad hoc and sensor networks mpose tght constrants on cost and power dsspaton, so nodes usually adopt a sngle transcever approach. Snce a sngle transcever cannot assess the status of exstng transmssons, t wastes valuable tme and energy on handshakng packets and corrupted packets. To avod such overhead, we propose a sngle transcever approach based on ultra wdeband (UWB) and a companon medum access control (MAC) layer based on busy tone multple access (BTMA). BTMA reduces the tme and energy spent on collsons as compared to handshakng protocols. It s well suted for ad hoc and sensor networks snce t permts random, dstrbuted medum access wth no central pont of falure. The sngle transcever leverages the nherently low duty cycle of mpulse-based UWB (I-UWB) to assess the status of a packet durng ts transmsson. Ths paper descrbes the I-UWB system archtecture, and smulatons show that t effectvely detects a busy sgnal wthout affectng data transmsson. Index Terms Ultra Wdeband, Ad Hoc Networks, Sensor Networks, Medum Access Control, Busy Tone Multple Access I. INTRODUCTION Recent developments n wreless technology have spawned extensve research n wreless ad hoc and sensor networks. Such networks enable varous applcatons such as nventory trackng, home networkng, or structural ntegrty montorng. These applcatons demand low power dsspaton and low cost. Nodes rely on lmted energy resources such as a battery, and the useful lfetme of the network depends on each node s ablty to conserve energy. Snce these networks may contan a large number of nodes, low node cost s essental to contan the overall network cost. The medum access control (MAC) protocol and the rado play a crucal role n determnng energy dsspaton and cost [1]. The rado dctates the energy effcency and hardware complexty of the physcal layer, and the MAC protocol mplements the collson avodance strategy. Collsons and corrupt packets waste energy when packets must be retransmtted. Collsons result from hdden termnals and correlated bursts of traffc. Harsh channel condtons, such as the nsde of a shp s metal hull, corrupt packets. Several research efforts have nvestgated rados and MAC protocols for ad hoc and sensor networks, and they consder both narrowband and wdeband rados [2]-[7]. However, lttle has been publshed to explot the unque capabltes of ultra wdeband (UWB) rados or ther apposte MAC protocols for large-scale ad hoc and sensor networks. UWB s a promsng rado nterface for ad hoc and sensor networks [8], [9]. Impulse based UWB (I-UWB) s partcularly attractve due to ts reslence to Raylegh fadng from multpath nterference, smple transcever crcutry, accurate rangng ablty, and low transmsson power [1]-[12]. Reslence to multpath nterference permts placement of UWB n areas nhosptable to narrowband systems such as nsde metal shp hulls. The carrerless nature of I-UWB results n smple, low power transcever crcutry, whch does not requre ntermedate mxers and oscllators. The rangng capablty allows nodes to accurately (under a centmeter) dscern locaton. Further, UWB s unque n that ts radated power s nherently ultra low as mandated by the FCC maxmum of 5 µw, whch s at least an order of magntude less than the radated power of the systems n [2]-[7]. Large ad hoc and sensor networks mpose tght constrants on cost and power, so nodes usually adopt a sngle transcever approach. In narrowband or wdeband systems, a sngle transcever cannot assess the status of exstng transmssons, so nodes rely on protocols such as carrer sense multple access wth collson avodance (CSMA/CA) that use handshakng packets to avod collsons. In ths paper, we propose an I- UWB rado nterface wth a sngle transcever and a companon MAC layer that avods such overhead. Our proposed sngle transcever leverages the nherently low duty cycle 1 of an I-UWB pulse tran to detect collsons and corrupted packets through transmsson of a busy sgnal. The busy sgnal elmnates handshakng packets such as request-tosend (RTS), clear-to-send (CTS), and acknowledgment (ACK), whch can add sgnfcant overhead from headers and swtchng tme [1]. Further, snce I-UWB systems dsspate far less power transmttng than recevng, the transmsson of a busy sgnal has nsgnfcant mpact on the power dsspated n a transacton. The proposed busy sgnal MAC permts random, dstrbuted medum access wth no central pont of falure, so t s approprate for any large ad hoc or sensor network. Also, note that a busy sgnal MAC provdes performance smlar to CSMA wth collson detecton (CSMA/CD), whch s more 1 In ths case, the duty cycle refers to the amount of tme that the transmtter apples a sgnal to the antenna durng transmsson. Duty cycle s also used to descrbe the cycle of the entre rado through sleep and awake modes.

2 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA effcent than CSMA/CA n throughput, delay, and energy per successful transmsson [14]. The paper s organzed as follows. Secton II presents our I- UWB rado desgn and revews prevous MAC protocols for I- UWB. Secton III explans the proposed system archtecture and the MAC protocol. Secton IV presents smulaton results for the proposed system, and Secton V concludes the paper. Data Control Block Pulse Generator... C 1 C 2 C n Fg. 1. I-UWB Transmtter BPF PA II. PRELIMINARIES Frequency Doman Sampler Energy Harvester A. I-UWB Rado Archtecture Ths secton revews our pror work on an I-UWB recever and transmtter for the proposed system. I-UWB systems communcate wth a tran of pulses that have a pulse wdth on the order of hundreds pcoseconds and a bandwdth on the order of ggahertz. The pulse repetton nterval (PRI) s generally much longer than the pulse wdth. Our transmtter s based on the energy effcent pulse generator n Fg. 1 [11], whch can generate varous pulse shapes and data rates through a programmable control block. Dependng on the desred data rate and pulse shape, the control block adjusts the control voltages C 1, C 2, C n to control the drecton of current flow and output voltage. The bandpass flter lmts the output to the FCC mandated -41. dbm/mhz over ts bandwdth from.1 GHz to 1. GHz. The total power from the crcutry s less than 1 µw [11], and the FCC lmts the radated power to 5 µw. Our recever employs the frequency doman approach n Fg. 2 [1]. At the front end, a low-nose amplfer (LNA) feeds typcal narrowband resonator flters realzng the second order 2 2 transfer functon 1 ( s + ( kωo ). A flter captures an n-band spectral component of the receved sgnal at frequency f, where f = kf for an nteger k. The fundamental frequency F s determned by an observaton perod T p such that F =1/T p. Next, the ADC captures spectral samples at the pulse repetton rate, whch s much lower than the Nyqust over-samplng rate to save power as well as crcut complexty. The energy harvester performs baseband sgnal processng such as the correlaton and rake operatons n the frequency doman, whch results n effcent hardware. The clear channel assessment (CCA) block n Fg. 2 detects a busy medum n the presence of I-UWB traffc just as carrer sense detects sgnals n a certan frequency band [15]. The CCA block examnes spectral power components to avod searchng for a narrow pulse wthn a long PRI. When there s pulse actvty at the pulse repetton frequency (PRF), the flters oscllate and each energy detector receves a frequency component, f, whch s an nteger multple of the PRF and T p. The energy detectors contan an envelope detector followed by an ntegrator. The outputs of the energy detectors are compared to a threshold and combned to detect the presence of I-UWB pulses, whle rejectng narrowband sgnals. The transcever supports data rates from 1 Kbps to 1 Gbps, and t can acheve a bt error rate (BER) of approxmately for a lnk dstance of 1 meters n extreme non lne-ofsght channel condtons at a data rate of 1 Mbps. Lowerng the data rate can ncrease lnk dstance or mprove the BER. The transmtter dsspates less than µw of power [11], whch s sgnfcantly less than the 2 mw from the recever. LNA Flter f Flter f 1 Flter f n-1 ADCs Energy Detector Energy Detector 1 Energy Detector n-1 Multpath Resolvng CCA Block Threshold and Combne Frequency Doman Correlator Template Constructor Decson Block CCA Fg. 2. I-UWB Recever and Assessment (CCA) Block B. MAC Protocols for I-UWB Centralzed protocols perform well n terms of throughput, delay, and Qualty of Servce (QoS), snce they collect nformaton about the state of the network. Examples of centralzed protocols for I-UWB nclude the TDMA approach of IEEE a or tme hoppng [1],[17]. However, centralzed protocols add control traffc overhead, have a central pont of falure, and requre more complex hardware and software. Dstrbuted protocols are less complex and they scale to large ad hoc or sensor networks. As narrowband systems use carrer sensng to mplement random, dstrbuted medum access, I-UWB systems use pulse sense [15]. Pulse sense provdes the CCA functon for I-UWB. For the recever n Fg. 2, the pulse sense block performs two mportant CCA roles. One role s to detect an ncomng packet and the other s to ensure that the channel s free before transmttng. To mtgate the hdden termnal problem, basc PSMA s augmented wth collson avodance (CA) n the form of CTS and RTS handshakng packets. The RTS and CTS packets warn nodes wthn range of the source and destnaton to delay future transmssons untl the current transacton fnshes. Snce I- UWB requres a relatvely long acquston tme, the RTS and CTS packets can create sgnfcant overhead [1]. Amortzng the cost of the RTS/CTS preambles over longer data packets may not resolve ths problem, snce larger data packets ncur more bt errors. III. PROPOSED RADIO AND MAC To manage collsons, a random access protocol tmemultplexes handshakng packets wth data packets. A more effcent approach n terms of channel utlzaton s to provde

3 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA feedback durng data transmsson. Busy sgnal MAC protocols use ths feature to reduce overhead, to ncrease throughput, and to effcently manage collsons. Snce a transmtter mmedately knows of a collson, t wastes less energy transmttng corrupted packets. Ths secton descrbes our sngle I-UWB transcever archtecture that enables a busy sgnal protocol. It also revews applcable busy tone protocols and develops crtera for system desgn. A. System Archtecture To mplement a busy sgnal, a transcever must be capable of full duplex operaton [18]. Narrowband rados mplement full duplex operaton wth two transcevers n dfferent frequency bands. The proposed I-UWB system requres only a sngle transcever, but acheves nearly full duplex through a fne-graned half duplex. Ths s possble because an I-UWB sgnal s not contnuously transmtted lke a narrowband sgnal. The proposed I-UWB rado explots the dle tme between pulses to assess the state of ts transmsson. Note that both the data sgnal and the busy sgnal are n the same band and share the same RF crcutry. Fg. compares tme dvson duplex (TDD), frequency dvson duplex (FDD), and fne-graned half duplex wth I- UWB. The TDD system n Fg. (a) cannot smultaneously transmt and receve, so t ncurs penaltes n latency and data rate. The TDD system also adds delay from the propagaton tme and from the turnaround tme, or the tme to swtch from transmt mode to receve mode. The FDD system n Fg. (b) can transmt and receve smultaneously, but t requres an addtonal frequency band for the feedback channel. The I- UWB system n Fg. (c) acheves full duplex wthout the latency or speed penalty of the TDD system and wthout the addtonal frequency band of the FDD system. An I-UWB system performs the followng operatons for the fne-graned half duplex. Startng n the receve mode, t receves a pulse. Then t swtches from receve mode to transmt mode and transmts a pulse. After transmttng, t swtches back to receve mode, and t s ready to receve the next pulse. Fg. 4 compares an FDD archtecture to the proposed fnegraned half duplex I-UWB archtecture. Fg. 4 (a) shows an archtecture for narrowband FDD full duplex n ad hoc networks. It requres two transcevers and crculators for the two dfferent frequency bands. An ad hoc network has no base staton to translate between frequency bands for nter-node communcaton. Therefore, each node must be able to receve and transmt n ether band, dependng f t s a source node or a destnaton node. In addton, the feedback channel requres an addtonal frequency band to degrade spectral effcency. Fg. 4 (b) shows the proposed archtecture for I-UWB wth fne-graned half duplex. A low duty allows a sngle transcever to access a feedback channel n the same frequency band as the transmtted data. The swtchng tme between transmt and receve modes determnes the mnmum pulse repetton nterval. Instead of usng a typcal T/R swtch, we swtch the dsable nputs to the PA and the LNA. Ths scheme mproves the swtchng tme to 25 ps and results n no addtonal nose fgure lke a T/R swtch. It also provdes the necessary solaton, snce there s very lttle leakage nto the PA and LNA when they are dsabled. The fne-graned half duplex I-UWB transcever sgnfcantly reduces crcut cost and ncreases spectral effcency as compared to a narrowband FDD transcever. Crculator Crculator Echo Canceller - Echo Canceller - PA LNA PA LNA Transmtter Recever Transmtter Recever (a) System Archtecture for Frequency Doman Dvson Full Duplex Transmtter Channel 1 Channel 2 Enable/ Dsable PA Recever Ch 2 Ch 1 (a) Dvson Duplex Turn Around + Round Trp Propagaton Round Trp Propagaton (b) Frequency Dvson Duplex Channel 1 Channel 2.5*PRI PRI (c) Fne-Graned Dvson Half Duplex wth I-UWB Fg.. Types of Duplexng (b) Fne-Graned Doman Half Duplex wth I-UWB Fg. 4. Full Duplex System Archtectures for Ad Hoc Rados B. Busy Sgnal Protocol A busy sgnal provdes two servces: () to nform the source node of a successful (or unsuccessful) transmsson and () to prevent nodes wthn rado range of the destnaton node from ntatng a transmsson. The busy sgnal elmnates control packets such as RTS, CTS, and ACK, so they do not ncur synchronzaton overhead or requre fast turnaround tme between transmt and receve mode [14]. Ths effcently manages collsons and ncreases throughput as compared to a handshakng protocol. Further, the busy sgnal mmedately alerts the source node to a dropped packet or a collson, thus resultng n few wasted transmssons. Fnally, a busy sgnal LNA

4 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA prevents hdden termnals and can also prevent exposed termnals [14]. We brefly revew some varetes of busy-tone multple access (BTMA) that our I-UWB rados can support. In basc BTMA, any node that detects a transmsson emts a busy sgnal to prevent nodes wthn 2R range (where R s the transmt range of a node) of the source node from transmttng [19]. Ths elmnates hdden nodes but ncreases the number of exposed nodes. Recever ntated BTMA (RI-BTMA) requres the destnaton node to emt a busy sgnal after t decodes ts address [2]. Therefore, nodes wthn radus R of the destnaton are prevented from transmttng, resultng n fewer exposed nodes. However, ths results n a long vulnerable perod before the address s decoded. Fnally, wreless collson detect (WCD) combnes RI- BTMA and BTMA [14]. Durng the preamble, all nodes wthn radus 2R of the source node emt a busy sgnal. After the destnaton node decodes ts address, the other nodes termnate ther busy sgnals, and only the destnaton node emts a busy sgnal. WCD allows exposed nodes to transmt after the preamble as n RI-BTMA, but wthout the long vulnerable perod. The above BTMA protocols can leverage the low transmt power of I-UWB by transmttng the busy sgnal for the duraton of the transacton, whle the source node only perodcally checks for a busy sgnal. C. Desgn Goals Ths secton develops two mportant desgn goals to support a busy sgnal. The system should: () not degrade data recepton at the destnaton node and () be easly detectable [14]. Self-nterference complcates the above desgn goals. After transmttng a busy sgnal, the multpath channel causes a long rng down tme, and some of the busy sgnal multpaths could nterfere wth data recepton. Further, when multple recevers emt a busy sgnal, they may nterfere wth data recepton at the destnaton and busy sgnal detecton at the source. The phenomenon of overlap may also degrade performance [18]. Dependng on the flght tme, a busy sgnal pulse may overlap a data pulse at ether the source or the destnaton. For clarty, we assume the overlap occurs at the source node, and the destnaton node s free to transmt the busy sgnal so as to avod overlap. Thus, dependng on the lnk dstance, a porton of the busy sgnal (ncludng multpaths) may arrve whle the source node transmts a data pulse. Fg. 5 llustrates the occurrence of overlap as lnk dstance changes. At 1, the source node transmts a pulse, whch arrves one propagaton tme, T prop, later at the destnaton node at 2. At, the destnaton node sends a busy sgnal pulse exactly.5 PRI after the arrval of the frst data pulse. Fnally, at 4, the source node receves the busy sgnal pulse from the destnaton node. In Fg. 5 (a), the destnaton node s dstance d = c PRI meters (e.g., for a PRI of 1 ns, ths dstance s meters) from the source node, so the round trp propagaton tme s 2 PRI. Therefore, a busy sgnal pulse arrves exactly 2.5 PRI after the correspondng data pulse. In Fg. 5 (b), the busy sgnal pulse overlaps the data transmsson. The destnaton node s dstance d =.75 c PRI meters from the source node; the round trp propagaton tme s 1.5 PRI; and the busy sgnal pulse arrves exactly 2 PRI after the correspondng data pulse. Snce the source node s transmttng at the tme, t loses some energy from the busy sgnal. Note that Fg. 5 shows only the frst multpath of a receved pulse. In real-world stuatons, a recever could detect a sgnfcant porton of the multpath energy. Sgnal at Antenna Sgnal at Antenna 1 T prop = PRI 2.5*PRI PRI (a) Dstance = c*pri 4 Data Sgnal Busy Tone Sgnal at Antenna Sgnal at Antenna 1 T prop =.75*PRI 2.5*PRI Fg. 5. Overlap Effect PRI 4 Data Sgnal Busy Tone (b) Dstance =.75*c*PRI 1) Source Node: The goal of transcever desgn from the source node perspectve s to detect a busy sgnal whle rejectng nose, other data transmssons, and ts own selfnterference from the receved sgnal R. R () t = xˆ() t (, ), () (, ), () ( ) bj t t j hj t + sk t tk hk t + n t j k + x () t s ( t t ) (1) where b j (t) h j, (t) x (t) s (t) t d t j, t k, s k (t) h k, (t) n(t) d busy sgnal from node j, ncludng destnaton node, j source node, j k channel response node j to the source node mpulse response of the swtch n state (Rx, Tx) ncludng rngng data sgnal from the source node delay to the swtch propagaton delay from node j to the source node propagaton delay from node k to the source node data transmsson from node k, ncludng source node, k destnaton node, k j channel response from node k to the source node the nose at the destnaton node recever The source performs a sldng correlaton wth result Cm = R () t b( t t m ) dt (2) where t m [,PRI+T p ] and s changng unformly each PRI by (PRI+T p ) / k, wth k the number of sldng correlatons. To optmally detect a busy sgnal n nose, condtons (), (4), and (5) should be true. x () t s ( t t ) b( t t ) dt = d m ()

5 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA xˆ () t sk () t hk, () t b( t tm ) dt = x k () t bj () t hj, () t b( t tm dt Cmax ˆ ) = j Condton () s satsfed f ether: () the data transmsson does not overlap n tme wth the recever operaton or () the swtch can perfectly separate the transmtted pulse from the recever chan. Snce I-UWB sgnals are not contnuous, they can satsfy condton () wth one transcever. Condton (4) requres data from other transmtters and multpaths from the source node to arrve at a dfferent tme than the busy sgnal at the source node. Snce ths s dffcult to control, the busy sgnal should be separated from the data sgnal. Ths can be acheved wth an orthogonal pulse shape or through spreadng technques such as drect sequence UWB (DS-UWB). Condton (5) means that the transcever should capture as much busy sgnal energy as possble. Further, a busy sgnal should not combne destructvely wth other busy sgnals. 2) Destnaton Node: The goal of transcever desgn from the destnaton node perspectve s to demodulate the data sgnal, whle rejectng nose, other busy sgnal transmssons, and selfnterference from the receved sgnal R. R () t x () t s t t ) h () t + b ( t t ) h () t = ( ˆ,, j j, j, j + x () t b ( t t ) () d (4) (5) + n( t) where b j (t) busy sgnal from node j, ncludng destnaton node, j source node, wthn range of destnaton node b (t) busy sgnal from the destnaton node h j, (t) channel response from node j to the destnaton node x (t) mpulse response of the swtch n state (Rx, Tx) ncludng rngng s (t) data sgnal from the source node t d delay to the swtch t, propagaton delay from the source node to the destnaton node t j, propagaton delay from node j to the destnaton node h, (t) channel response from the source node to the destnaton node n(t) the nose at the destnaton node recever Assumng coherent detecton, the destnaton node performs the followng correlaton on the receved sgnal, where s n s the n th bass functon of the sgnal set. Cn = R () t sn( t) dt (7) To optmally detect the data n nose, condtons (8), (9), and (1) should be true. x () t b ( t t ) s ( t) dt = d n (8) xˆ () t b j () t h j, () t sn ( t) dt = x j () t s () t h, () t sn( t dt Cmax ˆ ) = (9) (1) To satsfy condton (8), the destnaton node s busy sgnal should not nterfere wth recepton. Condton (9) requres the busy sgnal pulses from other nodes to not nterfere wth recepton, so busy sgnal pulses should be separated from data pulses as for the source node. Condton (1) means that the transcever should capture as much of the receved data sgnal energy as possble. Fnally, a node that s nether the source nor the destnaton must accurately detect a busy sgnal. Ths stuaton s smlar to that of a source node detectng a busy sgnal wthout overlap. IV. SIMULATION RESULTS We compare the performance of the followng methods to meet the crtera n Secton III. 1) Use dfferent PRIs for the data sgnal and the busy sgnal. If PRI data sgnal s n*pri busy sgnal, for an nteger n, then the source node can detect ether n or n-1 busy sgnal pulses, each havng a power of 1/n of the data sgnal. Alternately, f PRI busy sgnal s slghtly less than the PRI data sgnal, then the pulses wll only overlap at some small beat frequency [18]. 2) Use dfferent waveforms for the data sgnal and the busy sgnal, e.g. DS-UWB and I-UWB. Also, orthogonal pulse shapes can dfferentate the busy sgnal from the data, e.g. a Gaussan monopulse and the frst dervatve of a Gaussan monopulse are orthogonal. ) Rely on multpath effects to detect the busy sgnal. The multpath spread of an I-UWB sgnal can be qute sgnfcant compared to the pulse length. The source recever s dsabled durng transmsson, but energy from busy sgnal multpaths arrves for a perod much longer than the data pulse wdth. 4) Estmate and equalze the channel. A destnaton node can estmate and subtract ts own busy sgnal reflectons and the busy sgnal of other nodes from the receved data. A source node can also estmate the reflectons from ts own data, but ths s more complcated n hardware due to the modulaton. 5) Use a PRI at least twce the maxmum propagaton tme plus twce the multpath delay spread. After recevng a data pulse, the destnaton node wats for a perod of tme equal to the multpath delay spread before transmttng a busy sgnal pulse. Lkewse, after detectng a busy sgnal pulse, the source node wats for a multpath delay spread before transmttng data. If the PRI s such that PRI 2 (R max /c + D multpath ), then the system avods nterference. For a range R max of m and multpath delay spread D multpath of 2 ns, the mnmum PRI s ns. The maxmum pulse rate becomes 1.7 Mpps, whch s suffcent for most low data rate networks. To show the feasblty of the proposed I-UWB transcever for a busy sgnal protocol, we compare the above methods from the perspectve of a source node and a destnaton node. For

6 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA smulaton, the network topology s random wth multhop connectons. Each node has an average of about neghbors wth a maxmum lnk dstance of 1 m. The channel model s based on the Cassol channel model for ndoor UWB propagaton [21], and t consders statstcal varatons n small scale and large scale fadng and shadowng. Fg. dsplays the average power delay profle. s relatve to the frst arrvng multpath, and the ampltude of each vertcal lne represents the energy gan durng a 2 ns delay bn. A multpath des out f ts power s less than db above the nose floor. On average, 11% of total energy arrves n the frst multpath, 57% arrves after ns, and 92% arrves wthn 1 ns. Each transacton results n a dfferent, random nstance of the channel model and the topology. The average of 1 smulatons results n each data pont. Normalzed Power Name Data Excess Delay (ns) Fg.. Average Power Delay Profle of the Channel TABLE I. SIGNALS USED IN THE SIMULATIONS Descrpton The data sgnal. BPSK modulaton. Uncoded. Busy Sgnal wth the same PRI and energy as the data sgnal. Dvded PRI DS-UWB 1 DS-UWB 2 DS-UWB Busy sgnal. Each pulse has 1/n th the energy and 1/n th PRI of a data pulse. Dvded PRI busy sgnal wth n=2. Dvded PRI busy sgnal wth n=4. Busy Sgnal. Each pulse has the same energy as a data pulse, but the PRI s 99.5% of the data sgnal PRI to result n a beat frequency of 2/PRI. Busy sgnal wth spreadng factor = SF. Each pulse has 1/SF the energy of a data pulse. The chps are transmtted back to back and each bt s separated by the PRI of the data sgnal. DS-UWB wth SF=8. DS-UWB wth SF=1. DS-UWB wth SF=2. The source node performance s measured by the probablty of false alarm (P FA ) vs. the probablty of detectng the busy sgnal (P D ). The destnaton node performance s measured by the nose that the busy sgnal adds to the receved data sgnal. Table 1 descrbes the sgnals used n smulaton. A. Source Node Results To show the accuracy of the proposed system n detectng a busy sgnal wthout constranng the PRI (as n Method 5), we plot the probablty of detecton vs. the probablty of false alarm for a data rate of 1 Mbps under varous condtons of nterference. Assumng an 11 db nose fgure at the recever, we adjust the busy sgnal power such that the strongest receved multpath has a maxmum R of 4 db, whch just meets the FCC lmts at a 1 m lnk dstance. Better performance may be acheved wth multple looks, lower data rates, and shorter dstances. The fgures compare the relatve performance of the dfferent methods. For the deal case of Method 5, the top graph of Fg. 7 (a) shows the performance of the busy sgnal detecton wth the lne labeled, whch serves as a reference lne n the remanng graphs. The other lnes show the performance for a busy sgnal that arrves wth db less power than the data sgnal reflectons. Addtonally, there s no overlap n Fg. 7. Ths stuaton also corresponds to that of an dle node. Fg. 7 (a) does not nclude the busy sgnal, snce t s guaranteed to overlap the transmsson perodcally. The best performng busy sgnal s the 4 Pulse, snce t spreads four pulses evenly over the PRI. In the worst case, only one pulse out of four experences severe nterference. The 2 Pulse busy sgnal performs next best, as t also spreads ts pulses over the PRI; but now half the pulses may experence severe nterference. Next, the DS-UWB sgnals (all spreadng factors performed smlarly) perform almost as well as the 2 Pulse and 4 Pulse sgnals, snce the spreadng code combats nterference. However, snce the chps are sent consecutvely, the data sgnal nterferes wth all the chps, and hence performance s not as good a 2 Pulse or 4 Pulse. Fnally, the busy sgnal performs the worst, snce t has no method to combat nterference. All busy sgnals result n lttle performance degradaton as compared to the deal case, snce the recever only needs to detect the presence of a busy sgnal. Nonlnear effects from the antenna, channel, and RF crcutry may cause the busy sgnal and the data sgnal to lose orthogonalty. Fg. 7 (b) shows that performance degradaton s stll mld even when the data sgnal and the busy sgnal have dentcal pulse shapes. Fnally, we smulated the performance of each busy sgnal when the recever estmates ts self-nterference from the data sgnal and equalzes t. The hardware complexty ncreases but the performance s ndstngushable from the deal Clear Channel case as shown n Fg. 7 (c). Fg. 8 shows the performance of each busy sgnal when the strongest multpaths overlap the data sgnal transmsson. The overlap conssts of a 1 ns transmsson, two.25 ns T/R swtches, and an addtonal.5 ns of settlng tme for a total of 2 ns. The busy sgnal arrves wth db less power than the data sgnal reflectons, and the overlap causes an average loss of about 1% of the total busy sgnal energy.

7 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA 1% 1% Probablty of Detecton (Pd) 9% 8% DS-UWB 7% Probablty of False Alarm (Pfa) (a) Data Sgnal Orthogonal to Busy Sgnal, No Equalzaton Probablty of Detecton (Pd) 9% 8% 1 DS-UWB 2 DS-UWB 7% Probablty of False Alarm (Pfa) (a) No Equalzaton 1% 1% Probablty of Detecton (Pd) 9% 8% DS-UWB 7% Probablty of False Alarm (Pfa) (b) Data Sgnal Non-Orthogonal to Busy Sgnal, No Equalzaton Probablty of Detecton (Pd) 9% 8% 1 DS-UWB 2 DS-UWB 7% Probablty of False Alarm (Pfa) (b) Equalzaton Fg. 8. P D vs. P FA for Orthogonal Busy Sgnals wth Interference and Overlap Probablty of Detecton (Pd) 1% 9% 8% DS-UWB 7% Probablty of False Alarm (Pfa) (c) Data Sgnal Orthogonal to Busy Sgnal, Equalzaton Fg. 7. P D vs. P FA for Busy Sgnals wth Interference Fg. 8 (a) shows the performance wthout channel estmaton and equalzaton. The busy sgnal performs best n these condtons, snce t overlaps the data sgnal transmsson only perodcally. The next best sgnal s the 2 DS-UWB, snce the spreadng code s long, and the source node recever loses a smaller porton of energy due to overlap. The 1 DS-UWB sgnal loses twce as much energy as compared to the 2 DS-UWB sgnal, and the sgnal loses four tmes as much energy. Hence, hgher spreadng gan mproves performance under the condtons of overlap. The 4 Pulse busy sgnal performs smlarly to 2 DS-UWB snce t spreads four pulses evenly over the data PRI. The recever loses energy from only one the pulses due to overlap. The 2 Pulse busy sgnal suffers more degradaton than the 4 Pulse case, snce the overlapped pulse loses twce as much energy. Fnally, the 1 Pulse busy sgnal performs the worst, snce t has no method to offset the effects of overlap. The, Dvded PRI wth larger n, and the DS-UWB wth larger SF result n mld degradaton of the performance as compared to the deal case. They all combat the effects of overlap. Fg. 8 (b) dsplays the performance of each busy sgnal when the recever estmates ts self-reflected channel and equalzes the nterference. As compared to Fg. 8 (a), the performance mproves slghtly at the cost of ncreased hardware complexty and power. The performance gan s lmted, snce the busy sgnal that overlaps wth data transmsson s unrecoverable. Fg. 9 shows the performance when multple nodes emt busy sgnals. Sx neghbors emt a busy sgnal, and they are

8 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA unformly dstrbuted around the source node. The busy sgnals nterfere wth each other and may combne destructvely. The receved busy sgnal also overlaps the data transmsson and s corrupted by the data sgnal multpaths. The total nterference results n a busy sgnal to nterference rato of 7 db, and the overlap loss averages about 1% of busy sgnal energy. The performance of the busy sgnal s worse than the sngle busy sgnal case, snce all the busy sgnals experence overlap and nterference perodcally. The 4 Pulse and 2 Pulse busy sgnals mprove performance as compared to the sngle busy sgnal. Addtonal energy from one of the other busy tones may experence more favorable channel condtons. The busy sgnal mproves performance the most. The pulses appear n a narrower tme wndow and are less lkely to combne destructvely. The best performng sgnal s the 2 DS-UWB busy sgnal, snce the spreadng code s long, and t combats both overlap and destructve combnaton. The, Dvded PRI wth larger n, and the DS-UWB wth larger SF all combat the nterference from multple busy sgnals. For DS-UWB, the spreadng code should be chosen to mnmze autocorrelaton, snce the busy sgnals may overlap and combne destructvely. Fg. 9 shows smulaton results n whch an m-sequence code s used. Fg. 9 also shows the performance degradaton for 2 DS-UWB wth a random, rather than an m-sequence code. Degradaton ncreases for longer codes, as busy tones are more lkely to overlap. 1% occur close n tme to the begnnng of a receve cycle. The 1 Pulse busy sgnal results n the least addtonal nose snce t transmts all busy sgnal energy mmedately after a receve cycle, thus allowng the most rng down tme before the next receve cycle. The 2 Pulse busy sgnal performs slghtly worse than the case for short PRI, snce the reduced power of the second pulse does not offset the fact that t appears closer to the begnnng of the next receve cycle. Fnally performs smlarly to, snce t s also transmtted mmedately followng a receve cycle. All busy sgnals add nose to the receved sgnal at short PRIs, so a desgner may adjust the PRI to meet lnk budget constrants. Busy Sgnal Nose (db) Hgh Interference Medum Interference Low Interference Pulse Repetton Interval ( ) Fg. 1. Busy Sgnal Nose for Dfferent Levels of Interference Probablty of Detecton (Pd) 9% 8% 2 DS-UWB 2 DS-UWB w/o m-sequence 7% Probablty of False Alarm (Pfa) Fg. 9. P D vs. P FA for Multple Busy Sgnals B. Destnaton Node Results We smulate the busy sgnals n Table I to determne ther mpact on data recepton as the PRI decreases. For each busy sgnal, we smulate wth three representatve topologes that result n SINRs of db, db, and - db as the lnk dstance changes. We present the results n terms of nose fgures, and the effect on BER wll depend on varous factors ncludng lnk dstance, recever type, and codng. For reference, n the worst smulaton case, 1 db of nose ncreases the BER from to 8 1-4, and.5 db of nose ncreases the BER to 1-4. Fg. 1 shows the nose that each busy sgnal adds to the data sgnal for the three topologes. The busy sgnal performs worst, snce ts pulses are guaranteed to perodcally Fg. 11 shows the effect of ncreasng the spreadng factor for both Dvded PRI and DS-UWB. The Dvded PRI sgnal adds nose as the number of pulses n ncreases. Although the sgnal power decreases wth n, more busy sgnal energy occurs near the begnnng of the next receve cycle. The DS-UWB sgnal does not ncur any penalty as compared to the case for any spreadng factor. Ths s because the destnaton node transmts the DS-UWB busy sgnal mmedately after the receve cycle, so t contrbutes less nterference to the next receve cycle. Busy Sgnal Nose (db) Dvded PRI DS-UWB 1 DS-UWB 2 DS-UWB Pulse Repetton Interval (ns) Fg. 11. Busy Sgnal Nose for Dfferent Spreadng Factors

9 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA Busy Sgnal Nose (db) Orthogonal Not Orthogonal Pulse Repetton Interval (ns) Fg. 12. Busy Sgnal Nose for Dfferent Pulse Shapes Busy Sgnal Nose (db) Mult-Sgnal, Not Orthogonal Mult-Sgnal, Orthogonal Sngle Sgnal, Orthogonal Pulse Repetton Interval (ns) Fg. 1. Busy Sgnal Nose for Dfferent Busy Sgnal Protocols Snce the antenna, channel, and recever front-end may ntroduce non-lnear effects nto the receved sgnal, the busy sgnal and data sgnal pulses may lose orthogonalty. Fg. 12 shows the smulaton results for a busy sgnal pulse shape that s dentcal to the data pulse shape under the hghest level of self-nterference. The busy sgnal adds from.5 db to 1 db of nose. The busy sgnal adds nose at shorter PRIs, snce longer PRIs allow the multpath reflectons to rng down. The busy sgnal adds slghtly more nose than the busy sgnal due to the closer proxmty of the second busy sgnal pulse to the begnnng of the next receve cycle. Fnally, the sgnal ncurs neglgble addtonal nterference due to the spreadng gan. In all cases, the effects of non-orthogonal sgnals are moderate, and the total energy n the channel has a greater mpact on performance. Fg. 1 shows the effects of multple nodes emttng a busy sgnal. Sx neghbors wth a unform, random dstrbuton of dstances emt a busy sgnal. Each graph compares a sngle busy sgnal wth low self-nterference to multple busy sgnals wth low self-nterference. For all cases, the multple busy sgnals add sgnfcantly more nose than a sngle busy sgnal. Ths s because the source node cannot control the tme at whch t receves the busy sgnals from other nodes. For the three cases of,, and, the loss of orthoganalty s more sgnfcant than the case wth just one node. However, the DS-UWB busy sgnal stll performs as well as the orthogonal case. Fnally, we smulate the effects of channel estmaton and equalzaton on data recepton. The estmaton and equalzaton process s relatvely smple, snce the busy sgnal s known and t never changes. The smulaton uses 8-bt quantzaton of seven frequency doman samples. Equalzaton dramatcally reduces nose at shorter PRIs but adds small quantzaton nose at longer PRIs. Fg. 14 shows that the quantzaton nose averages about.2 db for both sngle and multple busy tones. The channel estmaton scheme mproves performance at the expense of addtonal hardware and power. Busy Sgnal Nose (db) Hgh Interference, Sngle Sgnal Hgh Interference, Multple Sgnals Pulse Repetton Interval (ns) Fg. 14. Busy Sgnal Nose wth Equalzaton Scheme V. CONCLUSION For ad hoc and sensor networks, low node cost and low power dsspaton are essental desgn consderatons. The MAC protocol and the rado play a crucal role n determnng both the node cost and the amount of energy spent on a successful transmsson. I-UWB s a partcularly attractve rado for ad hoc and sensor networks due to ts reslence to multpath nterference, smple transcever crcutry, accurate rangng ablty, and low transmsson power. Further I-UWB enables a sngle transcever archtecture for a MAC based on busy sgnals. Such a MAC provdes performance smlar to CSMA/CD, whch s more effcent than CSMA/CA n throughput, delay, and energy per successful transmsson [14]. It also permts random, dstrbuted medum access wth no central pont of falure, so t s approprate for any large ad hoc or sensor network. The sngle transcever leverages the nherently low duty cycle of an I-UWB pulse tran to detect collsons and corrupted packets through a busy sgnal. The busy sgnal avods the overhead of tme-duplexed control packets. The

10 IEEE Conference on Sensor and Ad Hoc Communcatons and Networks, October 24, Santa Clara, USA system fnely multplexes the busy sgnal channel wth the data channel to re-use the same frequency band and rado front end. Further, I-UWB systems dsspate far less power transmttng than recevng, so the busy sgnal does not sgnfcantly ncrease power. At the cost of lower data rate, the smplest and most effectve method of mplementng the busy sgnal s to set the PRI long enough to avod any nterference. At shorter PRIs, the smulatons fnd that channel estmaton at both the source node and the destnaton node provdes comparable performance at the cost of addtonal hardware. To mplement the busy sgnal wthout ncreased hardware complexty, an orthogonal DS- UWB sgnal wth a hgh spreadng gan ncurs only a slght degradaton of BER at the destnaton node, and t s easly detected at the source node. REFERENCES [1] A. Woo and D. E. Culler, A transmsson control scheme for meda access n sensor networks, Proceedngs of the 7th annual nternatonal conference on Moble computng and networkng, pp , July 21, Rome, Italy. [2] J.-P. Hubaux, J.-Y. Le Boudec, S. Gordano, M. Hamd, L. Blazevc, L. Buttyan, M. Vojnovc, Towards moble ad-hoc WANs: termnodes, Proceedngs Wreless Communcatons and Networkng Conference, 2, vol., pp , Sep. 2. [] The Insttute of Electrcal and Electroncs Engneers, Inc., IEEE Std Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons, 1999 edton. [4] M. Leopold, M.B. Dydensborg, and Phlppe Bonnet, Bluetooth and Sensor Networks: A Realty Check, Proceedngs of the frst ACM Internatonal Conference on Embedded Networked Sensor Systems (SenSys), pp. 1-11, November 2. [5] O. Kasten and M. Lamghenrech, Frst Experences wth Bluetooth n the Smart-ITS dstrbuted Sensor Network, In Workshop on Ubqutous Computng and Communcatons, PACT, 21. [] Macro Motes of Smart Dust Project avalable at [7] Y. We, J. Hedemann, and D. Estrn, An energy-effcent MAC protocol for wreless sensor networks, Proceedngs of INFOCOM 22. Twenty-Frst Annual Jont Conference of the IEEE Computer and Communcatons Socetes. pp , vol., June 22. [8] I.F. Akyldz et al., A Survey on Sensor Networks, IEEE Communcatons Magazne, pp , August 22. [9] F. Legrand, Motvaton n 15.4a, IEEE a Task Group presentaton, ftp://eee:wreless@ftp.82wrelessworld.com/15/4/ a-thales-coms-motvaton-n-eee a.ppt, March 24. [1] H.-J. Lee, D.S. Ha, and H.-S. Lee, A Frequency-Doman Approach for All-Dgtal CMOS Ultra Wdeband Recevers, IEEE Conference on Ultra Wdeband Systems and Technologes, pp. 8-9, November 2. [11] K. Marsden, H.-J. Lee, D.S. Ha, and H.-S. Lee, Low Power CMOS Reprogrammable Pulse Generator for UWB Systems, IEEE Conference on Ultra Wdeband Systems and Technologes, pp. 44-7, November 2. [12] W. C. Chung and D. S. Ha, An accurate ultra wdeband (UWB) rangng for precson asset locaton, 2 IEEE Conference on Ultra Wdeband Systems and Technologes Conference Proceedngs, pp. 89-9, Nov. 2. [1] J. Dng, L. Zhao, S. Medd, and K. Svalngam, MAC Protocols for Ultra-Wde-Band (UWB) Wreless Networks: Impact and Channel Acquston, Proc. SPIE ITCOM2, July 22. [14] A.C.V. Gummalla and J.O. Lmb, Desgn of an access mechansm for a hgh speed dstrbuted wreless LAN, IEEE Journal on Selected Areas n Communcatons, vol. 18, no. 9, pp , Sep. 2. [15] N.J August, H.J. Lee, and D.S. Ha, Pulse Sense: A method to detect a busy medum n ultra wdeband (UWB) networks, Dgest of Papers. 24 Jont IEEE Conference on Ultra Wdeband Systems and Technologes and Internatonal Workshop on UWB Systems, pp , May 24. [1] R. Scholtz, "Multple access wth tme hoppng mpulse modulaton," Conference Record of the IEEE Mltary Communcatons Conference, vol. 2 pp , Oct 199. [17] The Insttute of Electrcal and Electroncs Engneers, Inc., Draft Standard for Telecommuncatons and Informaton Exchange Between Systems - LAN/MAN Specfc Requrements - Part 15: Wreless Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons for Hgh Rate Wreless Personal Area Networks, Draft P82.15./D17, Feb. 2. [18] L.W. Fullerton, Full duplex ultrawde-band communcaton system and method, U.S. Patent , Nov. 11, [19] F. Tobag and L. Klenrock, Packet swtchng n rado channels: part II-- the hdden termnal problem n carrer sense multple-access and the busy-tone soluton, IEEE Transactons on Communcatons, vol. 2, no. 12, pp , Dec [2] C. Wu and V. L, Recever-ntated busy-tone multple access n packet rado networks, Proceedngs of the ACM Workshop on Fronters n Computer Communcatons Technology, p.-42, August 11-1, 1987, Stowe, Vermont, Unted States. [21] D.Cassol, M.Z. Wn, and A.F. Molsch, The ultra-wde bandwdth ndoor channel: from statstcal model to smulatons, IEEE Journal on Selected Areas n Communcatons, Vol. 2, No., pp , Aug. 22.