Design and Evaluation of Localization Protocols and Algorithms in Wireless Sensor Networks Using UWB

Transcription

1 Design and valuation of Localization Protocols and Algorithms in Wireless Sensor Networks Using UWB Di Wu, Lichun Bao, Min Du, Renfa Li Donald Bren School of ICS, University of California, Irvine, USA School of Comuter and Communication, Hunan University, Changsha, China Abstract Localization has many imortant alications in wireless sensor networks (WSNs). A variety of technologies, such as acoustic, infrared, and UWB (ultra-wide band) media have been utilized for localization uroses. In this aer, we roose a holistic, bottomu design of a UWB-based communication architecture and related rotocols for localization in WSNs. A new UWB coding method, called U-BOTH (UWB based on Orthogonal Variable Sreading Factor and Time Hoing), is utilized for minimum interference communication, and an ALOHA-tye channel access method and a message exchange rotocol are used to collect distance information in WSNs. We derive the corresonding UWB ath loss model in order to aly the maximum likelihood estimation (ML) method to comute the distances between neighbor nodes using the RSSI information. Then, we roose NMDS- ML (Non-metric Multidimensional Scaling and Maximum Likelihood stimation) localization algorithms based on the two tyes of distance information: estimated distance and uclidean distance. The erformance of the system is validated using theoretic analysis and simulations. I. INTRODUCTION Large-scale economic wireless sensor networks (WSNs) are widely deloyed for environmental monitoring and control oerations. Object tracking and localization are two imortant caabilities in many WSN alications []. The basic aroach to a WSN localization is to infer distances to anchor locations, then to derive the location of a node by trilateration or other estimation algorithms. The first ste is called ranging, and the second ste is called localization. So far, various ranging solutions have been roosed based on two major ranging techniques: ) time of arrival (ToA) [], time difference of arrival (TDOA) and angle of arrival (AOA) based ranging techniques, as used by GPS systems, ) the ath loss model based on radio RSSI signal strength [] or acoustic signal strength [] attenuation models. Other range-free techniques were also roosed for localization uroses, such as ho count or centroid methods []. We adot the ath loss model to derive range information because it is an efficient method in low-cost WSNs, in contrast to exensive synchronization requirements in the former aroach []. Ranging algorithms based on ath loss model deend on the wireless medium and signal transmission methods. In order to rovide recision ranging, we utilize the UWB (ultra-wide band) transmission and coding technologies in both indoor and outdoor environments. Beside roviding high data bandwidth, UWB exhibits excellent resistance to co-channel interference. I..a has aeared as the de facto standard to rovide low ower long distant low datarate service for real-time communication and recise ranging and localization alications [], []. Of the different UWB transmission techniques, Imulse Radio Ultra-wideband (IR-UWB) is most attractive for localization uroses in WSNs []. However, existing coding algorithms for IR-UWB communication systems, such as DS-UWB (Direct Sequence UWB) and TH-UWB (Time Hoing UWB) [] have failed to guarantee high quality localization due to multiath and multi-user interference. This work was sonsored in arts by the National Natural Science Foundation of China under Grant No. and the Raytheon Comany under Grant No. RC-. In this aer, we aly the Orthogonal Variable Sread Factor (OVSF) coding algorithm in IR-UWB networks to solve the multiuser interference roblem in data transmissions. Once the aroximate distances between a node and a subset of anchor oints are derived in the WSNs, the coordinates of the node can be derive by localization algorithms. Savarese et al. resented a trilateration algorithm based on least squares (LS) method in largescale WSNs []. Calun et al. [] roosed a GPS-free ositioning system for mobile ad hoc networks, by first establishing the local coordinates of two-ho neighbors with each node as the origin, then tuning these local coordinates to the global coordinates of the entire system. The DV-coordinate algorithm [] used similar idea. Different from trilateration algorithms, the MDS (Multidimensional Scaling) method uses two tyes of mas the relative ma and the absolute ma to derive locations using statistical techniques []. The relative ma reflects artial and relative inter-nodal relationshis in lower dimension sace, whereas the absolute ma is generated relative to the anchor nodes using the relative ma. MDS requires less information and configuration overhead than other localization algorithms in WSNs, and rovides strong resilience to measurement errors. Several variants of MDS were roosed so far. MDS-MAP uses connectivity information (whether or not two devices are in range) for localization []. MDS-MAP(P) imroved the basic MDS-MAP on anisotroic toologies [] by building a local relative ma of a small sub-network for each node using MDS, then merging them to form a global relative ma. However, most of MDS algorithms were based on the assumtion that roximity data between objects should be roortional to uclidean distances by underlying quantitative transformation function, which is not flexible or robust. In this aer, we resent the NMDS-ML (Non-metric MDS and Maximum Likelihood stimation) localization algorithm, based on IR-UWB model and RSSI (Received Signal Strength Indication) information. Non-metric MDS is different from revious MDS variants in that the roximity data are only assumed to be related to uclidean distances according to same ordinal level by some monotone transformation. Overall, the contribution of this work is the following: ) A new UWB coding method, called U-BOTH (UWB based on Orthogonal Variable Sreading Factor and Time Hoing), is roosed for minimum interference communication. ) An ALOHA-tye channel access rotocol and a message exchange rotocol are used to collect distance information in WSNs. ) The UWB ath loss model in U-BOTH is derived and alied in the maximum likelihood estimation (ML) method to comute the distances between neighbor nodes using the RSSI information. ) The NMDS-ML (Non-metric Multidimensional Scaling and Maximum Likelihood stimation) localization algorithm is roosed using two tyes of distance information: estimated distance and uclidean distance.

2 The rest of the aer is organized as follows. Section II describes the basic assumtions of the localization system, and the notation used in this aer. Section III resents a new IR-UWB coding method, called U-BOTH (UWB based on Orthogonal Variable Sreading Factor and Time Hoing), and rovides the signal rocessing model. Section IV secified a WSN communication rotocol to localization using U-BOTH. According to the ath loss model and the RSSI information gathered by the target nodes, Section V and Section VI resent the ranging and localization algorithms using the ML and NMDS methods, resectively. Section VII evaluates the system using simulations. Section VIII concludes the aer. II. ASSUMPTIONS AND NOTATION Our ranging and localization in mainly based on the RSSI information. In order to collect the distance information as quickly as ossible and avoid interference of multi-users, we assume that each node in the WSN is able to communicate through U-BOTH, roosed in this aer. For convenience, the notation in Table I is used in this aer. TABL I NOTATION AND MANING Notation Meaning The frame time. T c The chi time. T b The bit time. N s The number of ulses for every bit. N c The number of chis for every frame. d n j The OVSF code of transmitter n. SF The sreading factor of OVSF codes. N s The eriod of OVSF code. T n X The transmission energy of transmitter n. n The received energy of transmitter n. (t) The energy normalized ulse waveform. c n j The time-hoing code with eriod N s. a n j/n s The indication of information bit b. r u(t) The inut useful signal of the receiver. r mui (t) The inut multile users interference signal of the receiver. n(t) The inut additive white Gaussian noise of the receiver. m(t) The correlation temlate of the receiver. Z u The outut useful signal of the receiver. Z mui (t) The outut multile users interference of the receiver. Z n The outut additive white Gaussian noise of the receiver. N The noise sectral density. τ The delay of the other transmitter s interfering ulse. µ x The mean value of variable x. σ x The standard deviation of a random variable x. erfc(x) The comlementary error function of value x. P r b The bit error rate (BR). III. PHYSICAL LAYR MODL A. UWB Signal Sreading and Modulation In order to achieve accurate localization, we need a reliable hysical layer communication technique that reduces bit error rate (BR), while mitigating the multi-users-interference (MUI) and Gaussian noise interference. Our hysical layer is a UWB system based on time-hoing (TH) signal transmission as well as OVSF (orthogonal variable sread factor) for sreading out the symbols. OVSF (Orthogonal Variable Sread Factor) was extensively used in CDMA systems to rovide variable sreading codes []. Shorter OVSF code lengths are usually otimized for short-distance and high-data-rate transmission in less crowed environments due to its smaller sreading factor. TH (time hoing) is one of many signal modulation methods used by UWB. We describe a system, called U-BOTH (UWB modulation Based on OVSF and Time Hoing), which alies the time-hoing ulse osition modulation (TH-PPM) algorithm to encode UWB ulse streams, and OVSF direct sequence to sread the user data bit stream. Bit Stream TH Code TH-UWB Waveform OVSF Chi Sequence U-BOTH Waveform T c Fig.. U-BOTH: Interference Resistant UWB Modulation Using Time Hoing and OVSF. Fig. illustrates the utilization of time hoing (TH) ulse osition modulation and OVSF sreading to encode a single bit in the user data stream. First, U-BOTH sends each bit in the bit time, denoted by T b. Then it modulates the bit using a TH code,, in which each digit denotes a chi slot osition within a frame time,, to send a broadband radio ulse. The number of ulses is denoted by N s. Therefore, each bit duration is T b = N s. ach chi slot lasts for T c, sufficient to send a short UWB ulse signal. After the initial ulse osition modulation using UWB signals, the ulse sequence is again alied with OVSF code so that the hases are shifted by π to rovide orthogonality between multile users. The length of the OVSF code is called the sread factor SF, which is equal to N S. In U-BOTH, the TH code is a seudo-random sequence generated from foreknown seeds, such as node IDs. While the OVSF codes are selected from a well-defined set of orthogonal sreading codes. To formally analyze the system in this aer, we reresent the transmitted signal by the nth transmitter in q. (): s n (t) = + X j= Tb time d n j a n j/n s n T X (t j c n j T c), () in which, d n j = ± is the OVSF code with the eriod N s, T n X is the energy of the nth transmitter, (t) is the energy normalized ulse waveform, c n j [, N c ] is the TH code with eriod N s and a n j/n s indicates the data stream bit. If the data bit is, a n j/n s = +. Otherwise, a n j/n s =. At the receiver side, the received signal consists three source of information: r(t) = r u(t) + r mui(t) + n(t), in which, r u(t) is the desired user signal, r mui(t) is co-channel interference from multile users, and n(t) is the additive white Gaussian noise (AWGN). Denote the ulse energy of the n-th transmitter as n. Without loss of generality, we assume that the first user s transmission is the desired signal at the receiver for simlicity, then q. () rovides the desired signal function at the receiver: r u(t) = + X j= q d ja j/n s (t j c jt c). ()

3 We define the correlation temlate of the receiver: m(t) = (i+)n X s j=in s d j (t j c jt c); i (, + ). () B. Single User System Analysis As the first ste, we assume that the channel is AWGN multiathfree channel, and that the transmitter and the receiver are synchronized. In a single user signal rocessing system, the inut of the receiver has two arts: r u(t) and n(t), and the outut of the receiver in time interval [, T b ] is reresented by: Z = Z u + Z n = Z Tb In q. (), the useful outut signal is: Z u = s j= Z jtf +c j Tc+Tc j +c j Tc (r u(t) + n(t))m(t)dt. () q d jd ja j/n s ω(t)dt, where ω(t) = (t j c jt c) (t j c jt c). Because d jd j =, (t) is the energy normalized ulse waveform, we have R Tc Z u = P N s j= a j/n s (t) (t)dt R = N sa j/n s Tc (t) (t)dt = a j/n s N s In q. (), the outut noise signal is: Z n = s j= Z Tc d j (t)n(t)dt = s j= d jn j, where n j is Gaussian random variable with mean and variance N /. Because d j is not a random variable, the variance of Z n is: s D(Z n) = D( d N jn j) = N s, j= Z n N(, N N s/). Suose that the statistical robabilities of data bit b = and b = are equal, we obtain the BR (bit error rate) of the single user system in AWGN channel as follows: P r b = P (Z > b = ) + P (Z < b = ) = P (Z > b = ). Because a j/n s = if b =, then the useful outut is Z u = a j/n s N s = N s. Using q. (), the BR become: P r b = P (Z > b = ) = P ( N s + Z n > ) = P (Z n > N s ) It can be rewritten by comlementary error function erfc(x) as follow: s P r b = N s A. N R Where erfc(x) = Π x ex( t )dt. Because U-BOTH is a rate variable system using OVSF, we analyze the relation between BR and the bit rate. Suose the system s OVSF code is a code tree of layers [], and the sreading factor is,,,,,, resectively. Further suose the basic rate of our system is R, then the corresonding bit rate of U-BOTH is R b = ir (i=,,,,,, resectively). Denote the bit rate as R b, where R b = ir, i =,,,, we can get the relation between BR and the bit rate: q «P r b = erfc SF N = erfc q R R b N «() q. () shows that the BR decrease when the sreading factor SF increases or when the bit rate decreases. Therefore, we can adjust SF to adat different environments with various noise levels while maintaining the same bandwidth of the signal. This is the main reason we adjust OVSF codes in our system. C. Multi-User Interference Analysis In multi-user communication system, the received signal includes multi-user interference Z mui and noises. The Z u + Z n art is the same as q. (), but the multi-user interference Z mui is additional. Because the hase and delay τ of interfering ulses is random as shown in Fig., we have to comute the interference s variance. Fig.. The Interference to User by The n-th User. Suose that τ n is uniformly distributed over [, ), then the interference variance of the desired signal, i.e. the signal from the st user, caused by transmitter n is []: σ bit = Ns Z Tf Z Tc n d jd n i (t τ n ) (t)dt)«dτ n. Therefore, the total interference variance σmui from all other transmitters is: XN u n= N s n Z Tf Z Tc «! d jd n i (t τ n ) (t)dt dτ n. Because the delay τ for all transmitters has the same distribution, we get the following formula: σ mui = N s in which, P Nu n= n = σ M Ns P Nu n= n σ M = R R Tf ( R T c d jd n i (t τ n ) (t)dt) dτ n R Tc d jd n i (t τ n ) (t)dt dτ = R R (τ)dτ. According to [], and noticing that R b = N sn f and N s = SF =

4 R R b, q. () gives the BR in multi-user interference environments. P r b = v u erfc t v erfc t Ns N + N s R R b N + σ M T P Nu f n= n σ M R P Nu b n= n IV. NTWORK PROTOCOL OPRATIONS!! C A = «! Our localization algorithms deend on a two-ste rocess the first ste is for the target node to acquire the signal strength information from neighbor nodes in the network using U-BOTH based communication rotocols, and the second ste is for the target node to calculate the distances to the neighbor nodes, and infer its own coordinate. In ad hoc networks, code assignments are categorized into transmitter-oriented, receiver-oriented or a er-link-oriented code assignment schemes (also known as TOCA, ROCA and POCA, resectively) [], []. Deending on the ways of assigning the OVSF- TH codes and encoding the MAC data frames for transmissions, we roose two different ways to imlement multile access rotocols using U-BOTH. a) ROCA-Based Protocol Oerations: The first aroach is based on the receiver-oriented code assignment (ROCA), in which case the data acket transmissions are encoded using the unique OVSF-TH code assigned to the receiver. Beside ROCA, there is a common OVSF-TH code for bootstraing and coordination uroses. In ROCA scheme, when a target node needs to find out its coordinate, it sends a location request message using the common OVSF-TH code to the neighbor nodes. The request message includes the request command, and the receiver s OVSF-TH code. Uon receiving the request message, neighbor nodes sends back a resonse message using the receiver s OVSF-TH code using a random backoff mechanism. The distance information could be derived from the resonse message. b) TOCA-Based Protocol Oerations: The second aroach is based on transmitter-oriented code assignment (TOCA), in which case each acket transmission is encoded using two OVSF-TH codes one is a common OVSF-TH code to encode the common hysical layer frame header, and the other transmitter-secific code is to encode the hysical layer frame ayload. The frame head includes the transmitter-oriented OVSF-TH code for encoding the frame ayload. Because the hysical layer headers are sent on a common OVSF- TH code, the hysical layer header transmissions resemble those of ALOHA networks with regard to acket collision. Because the headers are usually short, the collision robability is low. On the other hand, because the data frame ayload is transmitted on unique OVSF-TH codes, the interference between the ayload and other frame headers and ayloads is dramatically reduced. In both ROCA- and TOCA-based systems, ackets from the neighbor nodes can be lost. However, this does not affect the overall erformance of our localization algorithms because they tolerate such losses. After getting the resective signal strength information from neighbor nodes, a target node calculates its coordinate in two stes ranging and localization. A. () V. RANGING ALGORITHM As mentioned before, ranging is to estimate the aroximate distance between the target node and neighbor nodes. We use the ML (maximum likelihood estimation) method for such calculations. First of all, we need to establish the ath loss model of the UWB channel in order to inversely derive the distance information from received signal qualities. A. The Path Loss Model It is well-known that the ath loss model can be exressed by the log-distance ath loss law in many indoor or outdoor environments, as shown by q. (). P L(d) = P L + γ log ( d «) + S; d d, () d in which d is the reference distance (e.g. meter in UWB medium), P L means the ath loss in db at d, d is the distance between the transmitter (T x) and receiver (Rx), γ refers to the ath loss exonent which deends on channel and environment, S is the log-normal shadow fading in db. Usually, S is a Gaussian-distributed random variable with zero mean and standard deviation σ S. q. () could construct a statistical ath loss model for UWB roagation in different environments. The ath loss P L(d) can be exressed as a Gaussian-distributed random variable with: S N(, σ S), P L(d) N(P L + γ log d, σ S). The robability density function (df) of ath loss P L(d) is: (P L) = e [P L (P L +γ log d)] σ S πσ S. () I..a Task Grou rovided Channel Model - by taking limited real measurements to determine the values of γ, σ S and other variables in different situations. When deloying real UWB networks, eole could aroximately choose the corresonding channel model with the arameters secified in I..a. B. Ranging Algorithm based on Maximum Likelihood stimation The distance between the transmitter T x and the receiver Rx in q. () can be calculated by the general ranging method between two nodes using the RSSI information: ˆd = P L(d) P L S γ. Receiver comutes the distance between the transmitter T x and the receiver Rx using random values S. However, in above single random ranging, the random variables S selected by the sensor nodes are not exactly those in the real time-variant channel. In order to avoid the ranging errors caused by the large deviation between the estimated S values and the real S values in each round of ranging estimation, we roose an iterative ranging based on ML (maximum likelihood estimation) in UWB wireless sensor networks. Suose P L i is the ith observation value, we get the joint conditional df (P L d) using q. (). (P L d) = NY i= e [P L i (P L +γ log d)] σ S πσ S. ()

5 The necessary condition to comute the ML of d is: P N i= P Li P L γ log d ln (P L d) = Nγ d σ S d ln =. N We solve q. () and have: log d d = Nγ i= P L i P L γ Therefore, the ML based RSSI UWB ranging is: () ˆd = Nγ P Ni= P L i P L γ. () VI. LOCALIZATION ALGORITHM A. Multi-Dimensional Scaling (MDS) MDS (Multidimensional Scaling) is a statistical technique for exloratory data analysis or information visualization. MDS collects the roximity data between each air of satial objects as reference. Then it visualizes objects as oints in a low dimensional uclidean sace and reresents these roximity data as distances between oints. In order to derive accurate results, MDS has to find some solutions that relate distance information to roximity information as closely as ossible. Suose that n denotes the number of different objects, and the roximity for objects i and j is denoted by ij. Thus, we derive a roximity matrix P n n = ij. The coordinates of maing oints are reresented by a matrix X n m, where m is the dimensions of the solution, e.g. D or D. Now, let d ij(x) be the uclidean distance between oints i and j with coordinates in X n m, resectively. The objective of MDS is to find a matrix X so that d ij(x) roortionally matches ij as closely as ossible, which is resented by f( ij) d ij(x). The closeness is measured by metric STRSS as follows: ST RSS = X [f( ij) d ij(x)]. MDS algorithms are taxonomized into several tyes, deending on whether the similarity data is quantitative or qualitative, and are called metric MDS and non-metric MDS, resectively. Classical metric MDS formulates the relationshi between roximity data of objects and distances in the uclidean sace by transformation functions. In order to find a erfect fitness between roximity data and uclidean distance, the transformation formula d ij(x) = f( ij) is ursued, such as a linear model: d ij(x) = a + b ij. Because d ij(x) reresents the uclidean distance between oints i and j in coordinate matrix X, MDS rests on the fact that the coordinate matrix X can be derived by double centering and eigenvalue decomosition from the roximity matrix P with the least error. The relationshi between the roximity of objects and the uclidean distances of oints in Non-metric MDS is not as strict as metric MDS. Non-metric MDS only requires a monotonic relationshi between them. When Non-metric MDS takes roximity data of different objects to construct corresonding satial coordinates, it only requires that the rank order of the roximity ij have to kee the same ordinal level as the distances d ij. That is, i, j, k, l : ij < kl d ij(x) < d kl (X). Comared with metric MDS, the monotonic assumtion that the data is measured at the ordinal level in Non-metric MDS makes it more flexible and alicable for localization in wireless sensor networks. B. The NMDS-ML Localization Algorithm NMDS-ML localization algorithm combines the ranging and localization rocesses. Ranging is based on the iterative RSSI information collected by above U-BOTH UWB system and refined by the ML method. Localization is based on the NMDS algorithm. As a whole, NMDS-ML localization consist of stes: Gather iterative RSSI from neighbors by U-BOTH system in the network, and form a sarse matrix R, which is derived from the estimated distances denoted by r ij. r ij is estimated by iterative ranging based on RSSI information and ML method. For the nodes that is out of the communication range, r ij is zero. Construct the roximity data matrix P based on sarse matrix R. The estimated distance ij between every air of nodes in the network is comuted by the shortest ath algorithm, such as Dijkstra s or Floyd s algorithm. Construct the coordinate system to lot the objects in the uclidean sace and obtained the distance matrix D comosed by the uclidean distance d ij. Comare the ordinal level between aforementioned two tyes of distance information: estimated distance ij and uclidean distance d ij, and refine the relative coordinate X of nodes in Non-metric MDS. Transform relative coordinate into global absolute coordinate by the anchor nodes in the network. In Ste and Ste, localization is executed by NMDS-ML as Algorithm. Algorithm : NMDS-ML Inut: node set N, initial coordinate matrix X (), roximity data matrix P, threshold ε, iteration number k Outut: relative coordinate X (n) for each i, j q N do d k ij (x k i xk j ) + (yi k yk j ) construct the uclidean distance matrix D (k) end while ST RSS ε do for each i, j, u, v N do if ij < uv and d ij > d uv then ˆd k ij (d k ij + d k uv)/ ˆd k uv (d k ij + d k uv)/ else if ij < uv and d ij d uv then ˆd k ij d k ij ˆd k uv d k uv end end k k + udate the coordinate matrix X (k) udate the distance matrix D (k) end In Algorithm, a monotonic transformation between roximity data and uclidean distance is calculated in line to, which yields an intermediate distance value ˆd ij. By erforming a monotone regression with the current distances d ij as targets and roximity ij as inuts, NMDS-ML generates ˆd ij to reflect the ordinal level of ij in each iteration, where ˆd ij should be subjected to: i, j, k, l : ij < kl ˆ d ij(x) < ˆ d kl (X). Because of above relation between ij and ˆd ij, NMDS-ML takes following STRSS alied in line to evaluate the accuracy of the

6 fitting: s X ST RSS = ij,i j ( d ˆ ij d ij) / X ij,i j d ij. () A small STRSS indicates a good fit, whereas a high value indicates a bad fit. Kruskal [] rovide some guide lines of stress value with resect to the goodness of fit of the solution, shown in Table II. TABL II STRSS AND GOODNSS OF FIT Stress Goodness of fit >. oor. fair. good. excellent. erfect Note in line and line in Algorithm, NMDS-ML udates satial coordinate matrix X k to X k according to d k k ij and ˆd ij, and then obtains a new uclidean distance d ij(x) k. The satial coordinate (x k i, yi k ) is udated as follows: x k i = x k i + α n Pj M,j i y k i = y k i + α n Pj M,j i k ˆd ij ( d k ij k ˆd ij ( d k ij )(x k j )(y k j x k i ), y k i ). () Where n is the number of target nodes, α is the iteration ste length, which is set to be. in the aer. In Ste, the estimated location matrix X reresents the relative coordinates of nodes, which have a different orientation and scaling than the original coordinates. And in Ste, the transformation from relative coordinate X into absolute coordinates usually includes shift, rotation, scaling, and reflection of coordinates, which are imlemented by some transformation to minimize the errors between the absolute coordinates of anchor nodes and their relative locations in the NMDS ma. Suose there are m anchor nodes whose relative locations are X R = (X R, X R,, X Rm ), and real locations are X T = (X T, X T,, X Tm ). We need firstly derive otimal transformation function Q, and then transfer all the relative coordinates of nodes to the absolute coordinates by the otimal transformation function Q. VII. SIMULATION VALUATIONS In order to verify our localization algorithms based on U-BOTH system for WSNs, we simulated the following scenarios: ) With regard to the BR (bit error rate), we evaluate U-BOTH system erformance in single and multi-user scenarios. ) Using NMDS-ML localization algorithm, we evaluate our localization model both in random network and grid network. A. U-BOTH System Performance We assume the channel is AWGN multiath-free single user channel, the transmitter and the receiver are synchronized erfectly. Then we randomly generate bits, every bit uses ulses to reeat coding (N s = ). Fig. illustrates the BR of the received signal using U-BOTH system, in contrast to DS-UWB that only uses direct sequence sreading, and TH-UWB that uses time-hoing ulse osition modulation alone for UWB transmissions. We can see that the BR of U-BOTH and the DS-UWB system which use the π-hase shift keying modulation BR BR U-BOTH DS-UWB TH-UWB - - b/no (db) Fig.. Bit rror Rate in A Single User System with Additive White Gaussian Noise (AWGN). are lower than TH-UWB. This is because the distance of two signals in binary hase shift keying (BPSK) modulation is ulse, but ulse in TH-UWB []. - - U-BOTH DS-UWB - TH-UWB Numbers of Interfering Users Variance of rror Bits U-BOTH DS-UWB TH-UWB Numbers of Interfering Users Fig.. Bit rror Rate and The Variance of The Number of rror Bits of Generated Bits. Secondly, we let b /N = db, N s = and generated bits randomly. Fig. shows the relative erformance of U-BOTH, TH-UWB and DS-UWB systems in multile access scenarios. In this case, the received signal includes by noise and co-channel interference. In Fig., although both the BR and the variance of error bits increase as the number of users increases, the erformance of our U-BOTH system is still better than DS-UWB and TH-UWB, roving that the UWB coding based OVSF-TH effectively handle the burst errors. B. valuation of the Localization Algorithms We evaluate the erformance of localization algorithms with mean estimation error, which is widely used in revious research works: P n i=m+ error = Xi est Xreal i % () (n m) R where n and m are the total number of sensors and the number of anchor nodes in the WNS, resectively, R reresents communication range. Based on the data in [], we adot values of UWB ath loss model in outdoor NLOS environments for simulations as shown in Fig.. ) Random Deloyment: nodes are deloyed randomly in a m m square area as shown in Fig. (a), in which oints reresent nodes and edges reresent the connections between neighbor nodes. The communication range is m and the average connectivity is..

7 Fig.. Notation Meaning LOS Value NLOS d The reference distance m m PL The ath loss at reference distance. db db γ The ath loss exonent.. σ S The standard deviation of shadow fading. Portion of The Simulation Parameters. Fig. (b) reflects the relative coordinate of every node generated by NMDS-ML. It shows that the relative coordinates have a different orientation and scaling than the original network in Fig. (a). This is because that relative coordinate is derived only based on the distance relationshi between every air of nodes in the network. Fig. (c) derives the absolute coordinates of all the nodes. Their relative coordinates in Fig. (b) are transformed based on the location information rovided by random anchor nodes denoted by. The dots reresent the real locations of the nodes, and the lines with arrows indicate the errors of the estimated locations from the real locations, the average localization error is about.%. The MDS- MAP algorithm is also alied in the case and the average localization error is about.%. ) Grid Deloyment: nodes are deloyed in a m m square area with grid deloyment in Fig. (a). The communication range is m and the average connectivity is.. With the same symbol meaning in the figures, Fig. (b) reresents the relative coordinate ma using NMDS-ML algorithm and Fig. (c) deicts the absolute coordinate ma by transformation based on random anchor nodes. The average localization error in the grid case is about.%. For MDS-MAP algorithm, it is about.%. ) Performance analysis: The localization erformance of NMDS- ML in different scenarios under different degrees of connectivity is analyzed by Fig., comared with the MDS-MAP by the same exerimental settings. From the figure, we can see that localization error of NMDS-ML algorithm is much lower and more stable than MDS-MAP in different scenarios. Furthermore, when NMDS- RSSI and MDS-MAP are alied in grid deloyment with varies of connectivity. It shows that NMDS-RSSI obtain higher localization accuracy in the grid layout than in the random layout for the same connectivity level. Fig.. Localization rror (%) NMDS-ML in Grid MDS-MAP in Grid NMDS-ML in Random MDS-MAP in Random Connectivity Relation between The Connectivity and The Localization rror. Fig. resents the relation between localization error and the number of iteration N in NMDS-ML algorithm. Because the accuracy of ranging is imroved by ML method based on the RSSI information rovided by our U-BOTH system, it is obvious that the localization error decreases dramatically when the number of iterations in ranging increases in both random and grid deloyment. Fig.. rror. Localization rror (%) NMDS-ML in Random NMDS-ML in Grid The Number of Iterations N Relation between The Number of Iteration N and The Localization VIII. CONCLUSION In order to rovide a localization algorithm using the NMDS- ML methods, we have roosed the communication rotocols based on a new UWB coding method, called U-BOTH (UWB based on Orthogonal Variable Sreading Factor and Time Hoing), and an ALOHA-tye channel access method and a message exchange rotocol to collect distance information in WSNs. Then we secified the NMDS-ML algorithms using the UWB ath loss model for ranging and localization uroses. The erformance of NMDS-ML algorithms in the U-BOTH based communication system are analyzed using communication theories and simulations. Results show that U- BOTH transmission technique can effectively reduce the bit error rate under the ath loss model, and the corresonding ranging and localization algorithms can achieve comarable or better results than revious localization methods. ACKNOWLDGMNT We would like to exress our sincere areciation of Fanzi Zeng, Ling Xiao, and Juan Luo for their insightful feedbacks during the rearation of this manuscrit, and of the anonymous reviewers for their helful comments. This work has been generously sonsored in arts by the National Natural Science Foundation of China under Grant No. and the Raytheon Comany under Grant No. RC-. RFRNCS [] I Std..a. Part.a: Low Rate Alternative PHY Task Grou (TGa) for Wireless Personal Area Networks (WPANs). Technical reort, I, Jun.. [] F. Adachi, M. Sawahashi, and K. Okawa. Tree-structured generation of orthogonal sreading codes with dierent lengths for the forward link of DS-CDMA mobile radio. I lectronics Letters, ():, Jan.. [] T.A. Alhmiedat and S.H. Yang. A Survey: Localization and Tracking Mobile Targets through Wireless Sensors Network. In The ighth Annual PostGraduate Symosium on the Convergence of Telecommunications, Networking and Broadcasting (PGNT),. [] P. Bahl and V. Padmanabhan. RADAR: An In-Building RF-based User Location and Tracking System. In INFOCOM,. [] I. Borg and P. Groener. Modern Multidimensional Scaling: Theroy and Alications. Sringer-Verlag, New York,. [] Hasan Cam. Nonblocking OVSF Codes and nhancing Network Caacity for G Wireless and Beyond Systemsn. Secial Issue of Comuter Communications on G Wireless and Beyond for Comuter Communications, ():,.

8 Fig (a) Original Coordinates (b) Relative Coordinates (c) Absolute Coordinates Random Deloyment. Fig (a) Original Coordinates (b) Relative Coordinates (c) Absolute Coordinates Grid Deloyment. [] S. Calun, M. Hamdi, and J.P. Hubaus. GPS-free Positioning in Mobile Ad Hoc Networks. In Proc. of the th Annu. Hawaii Int. Conf. System Sciences, ages,. [] T. Gigl and G. J.M. Janssen. Analysis of a UWB Indoor Positioning System Based on Received Signal Strength. In th Worksho on Positioning, Navigation and Communication (WPNC), ages,. [] T. He, J. A. Stankovic, C. Huang, T. Abdelzaher, and B. M. Blum. Range-Free Localization Schemes for Large Scale Sensor Networks. In Proceedings of the Annual International Conference on Mobile Comuting and Networking (MOBICOM), ages,. [] Joa-Ng, M., and I.T. Lu. Sread sectrum medium access rotocol with collision avoidance in mobile ad-hoc wireless network. In Proc. of I Conference on Comuter Communications (INFOCOM), ages, New York, NY, USA, Mar. -. [] J. Kruskal. Multidimensional Scaling by Otimizing Goodness of Fit to a Nonmetric Hyothesis. Psychometrika, :,. [] T. Makansi. Trasmitter-Oriented Code Assignment for Multiho Radio Net-works. I Transactions on Communications, ():, Dec.. [] Maria-Gabriella, D. Benedetto, and G. Giancola. Understanding Ultra Wide Band Radio Fundamentals. Prentice Hall, New Jersey,. [] A.F. Molisch, D. Cassioli, and C.-C. Chong. A Comrehensive Standardized Model for Ultrawideband Proagation Channels. I Transactions on Antennas and Proagation, ():,. [] D. Niculescu and B. Nath. DV based Positioning in Ad Hoc Networks. Journal of Telecommunnication Systems, (/):,. [] N. Patwari, A O Hero III, M. Perkins, N. S. Correal, and R. J. O Dea. Relative location estimation in wireless sensor networks. I Transactions on Signal Processing, :,. [] John G. roakis. Digital Communications: Fourth dition. McGraw- Hill, Columbus,. [] Ghassemzadeh S. S., Jana R., Rice C. W., Turin W., and Tarokh V. Measurement and Modeling of an Ultra-Wide Bandwidth Indoor Channel. I Trans. on Comm., ():,. [] C. Savarese, J.M. Rabaey, and J. Beutel. Locationing in Distributed Ad- Hoc Wireless Sensor Networks. In Proc. I International Conference on Acoustics, Seech, and Signal, volume, ages,. [] Y. Shang and W. Ruml. Imroved MDS-Based Localization. In Proc. of the I Infocom, ages,. [] Y. Shang, W. Ruml, Y. Zhang, and M. Fromherz. Localization from mere connectivity. In Proc. of the th ACM Int l Sym. on Mobile Ad Hoc Networking Comuting, ages,. [] K. Whitehouse and D. Culler. Macro-calibration in sensor/actuator networks. In Mobile Networks and Alications (MONT), volume, ages,. [] M.Z. Win and R.A. Scholtz. Ultra-Wide Bandwidth Time-Hoing Sread-Sectrum Imulse Radio for Wireless Multile-Access Communication. I Transaction on Communication, ():,.