Experimental Demonstration of Nanosecond- Accuracy Wireless

#1570035263



Abstract— Accurate wireless timing synchronization has been

an extremely important topic in wireless sensor networks,

required in applications ranging from distributed beam forming

to precision localization and navigation. However, it is very

challenging to realize, in particular when the required accuracy

should be better than the runtime between the nodes. This work

presents, to our knowledge for the first time, an experimental

timing synchronization scheme that achieves a timing accuracy

better than 5-ns rms in a network with 4 nodes. The experimental

hardware is built from commercially available components and

based on software-defined ultra-wideband transceivers. The

protocol for establishing the synchronization is based on our

recently developed “blink” protocol that can scale from the small

network demonstrated here to larger networks of hundreds or

thousands of nodes.

Index Terms—cooperative synchronization, network timing,

ultra wide band, software defined radio.

I. INTRODUCTION

Time synchronization in distributed wireless networks has

been studied for the past two decades, as it is an essential

component of many applications. The required accuracy, and

the resulting algorithms, differ according to the application. In

many cases, such as multiple-access protocols, timing

synchronization within less than one packet duration (on the

order of a millisecond) is sufficient. In other applications such

as diversity transmission, timing synchronization within one

symbol duration (on the order of microseconds) is desired. For

these applications, the IEEE 1588 standard is widely accepted

as a solution [1]. Comparable algorithms that have been

explored include Reference Broadcast Synchronization (RBS)

[2], Timing Sync Protocol for Sensor Networks (TPSN) [3]

and Flooding Time Synchronization Protocols (FTSP) [4]. In

general, these timing synchronization algorithms are based on

packet exchange, where the accuracy is related to the symbol

or preamble length and the propagation delay between the

nodes is neglected [5].

This work was supported in part by the Office of Naval Research

Electronic Warfare Science and Technology program, and the Ming Hsieh

Institute at the University of Southern California.

M. Segura, H. Hossein and A. F. Molisch are with the Department of

Electrical Engineering, University of Southern California, Los Angeles, CA

90089, USA. e-mail: ({mjsegura ; hosseinh ; molisch}@ usc.edu).

S. Niranjayan, was with University of Southern California, Los Angeles,

CA 90089, USA, now with Amazon (e-mail: sniranjayan@gmail.com).

There are, however, also a variety of applications that

require much better timing synchronization accuracy, in

particular better than the runtime of the signal between the

nodes. Such applications, such as coordinated jamming,

distributed beam forming, fine grain localization, tracking, and

navigation, require timing synchronization with nanosecond

precision or even better. In the absence of differential GPS

(which is often the case in many military and civilian

scenarios), maintaining such accurate timing synchronization

in large wireless networks is challenging; because, the timer in

each node is derived from an independent oscillator that is

affected by random drifts and jitter [6]. To maintain

nanosecond-level timing synchronization in a network (1)

timing deviation between pairs of nodes must be measured

accurately with nanosecond precision, and (2) fast correction

algorithms must be applied across the entire network [9, 16].

This paper offers solutions and experimental demonstrations

for both of these components.

Ultra-Wide Band (UWB) signals are used to precisely

extract the timing information between the nodes due to their

accurate distance measurement capability as well as superior

resiliency to multi-path effects [7]. UWB systems have been

used previously to demonstrate high-precision timing

synchronization between two nodes. In [8], a commercial

IEEE 802.14.5 radio was employed to obtain an accurate Time

of Arrival (TOA) detection algorithm; the authors also

proposed an Adder-Based Clock (ABC) approach for timing

synchronization and implemented a prototype to show the

accuracy of the synchronization block with nanoseconds

precision. However, no network experiments were carried out.

As a matter of fact, to the best of our knowledge, there are no

previous prototypes that demonstrate network synchronization

with nanosecond accuracy.

The challenge in achieving timing synchronization in a

wireless network lies in the required high speed of the node-

to-node timing inaccurate measurements and fast corrections

across the entire network. In this work, a recently developed

algorithm [9, 16], called "Blink" that enables network timing

synchronization without requiring an external broadcast signal

from a coordinator is implemented. The blink algorithm uses a

consensus approach such that timing information propagates

through the network, while timing errors are averaged, and

exploits the path diversity present in the network. Simulations

demonstrate excellent scalability, such that the obtained

timing precision in large (hundreds of nodes) networks is only

Experimental Demonstration of Nanosecond-

Accuracy Wireless Network Synchronization

Marcelo Segura, S. Niranjayan, Hossein Hashemi, Senior Member, IEEE and Andreas F. Molisch,

Fellow, IEEE

#1570035263

Slave

SYNC

SYNC_R

fix delay

Slave

Tier 0

Tier 1

Tier 2

Tier 3

RX node TX node

Fig. 1. a) Simplified node-node time exchange scheme; b) TVN topology

after Phase I.

marginally worse than those in two-node setups [9].

The current paper demonstrates and validates through a set

of experiments the very accurate network synchronization that

was predicted in simulations and shows that it is possible to

achieve this on a prototype using commercial components,

i.e., without requiring custom Integrated Circuits (IC). The

main components of this work are (1) a software defined ultra

wideband transceiver node implemented with Commercial-

Off-The-Shelf (COTS) components, capable of high accurate

synchronization, (2) a fast re-timing algorithm implementation

on a Field Programmable Gate Array (FPGA), and (3) a test

bed network consisting of three slave nodes and one master

that facilitates the validation of this algorithm and future

improved versions.

The remainder of the paper is organized as follows. Section

II presents a formulation for network timing synchronization.

Section III presents summary of the previously-published [9]

blink algorithm. Sections IV and V describe the implemented

node prototype and the hardware implementation on FPGA of

the blink algorithm, respectively. Section VI presents the

simulation and experimental results that prove the stability and

accuracy of the network timing synchronization. Finally, some

conclusions are presented.

II. PROBLEM OVERVIEW

Consider a large network with 



slaves nodes with

inaccurate internal oscillators, and 



master nodes with

accurate internal oscillators, which will serve as timing

references and initiators in the distributed algorithm. Let









 denote the current local timer value of a particular node

k at a given absolute time instance 



. The network timing

problem consists of maintaining, at all times, the offset

between nodes below a predefined threshold 





























  





,   . (1)

A major challenge lies in the fact that due to the random

deployment, the network topology is unknown. Furthermore,

in order to be scalable and robust to changes, the solution

should be distributed. Finally, in a large network, broadcast

transmission of timing information from a master to all other

nodes in the network is not possible. Therefore, the timing

information propagates through the network aggregating

errors.

III. BLINK ALGORITHM DESCRIPTION

A brief review of the blink algorithm is presented in this

section. More details and comparison with other distributed

timing synchronization schemes are given in [9] and

implementation is discussed in [16].

Inside the physical network, a virtual network called Timing

Virtual Network (TVN) is defined. The TVN consist of nodes,

links and the fast re-sync algorithm that maintains the timing

in the network. Different physical layers may be used for

timing and communications. The current work focuses on the

timing physical layer; the implementation of the

communication layer with a certain packet structure (e.g.,

using commercial IEEE 802.15.4 transceivers) is beyond the

scope of this paper, but has been considered in the blink

protocol definition. The topology of the network is defined by

the link matrix L as follows:





  

  



 







 







, (2)

where 



defines the Signal to Noise Ratio (SNR) between

nodes and 



is the minimum SNR needed to establish a

connection. Each node is assigned to a tier 



(Fig. 1.) and by

default masters are in tier 0. The blink algorithm consists of

two phases. In Phase I, the TVN is created and initialized and

in Phase II, the network timing is maintained through

continuous consensus-based corrections.

A. Phase I

All nodes (slaves and masters) learn the propagation delays

(pseudo-ranges) from their neighbors and record the values. In

a fresh deployment scenario, the network topology and the tier

structure is unknown. Therefore, propagation delays are

acquired in a distributed fashion, using a modified version of

Carrier Sense Multiple Access Collision Avoidance

(CSMA/CA). In this phase, the slave nodes also get assigned

to different tiers. Tier 1 consist of nodes that can derive timing

directly from the master; the size of this tier can be limited by

the number of slave nodes that have an acceptable SNR to the

master or the number of slave nodes that can communicate

with the master with acceptable time constraints (this is

important for scalability of the system). Other tiers are also

defined consecutively in a similar fashion (Fig. 1).

In this paper, the main objective is to experimentally

demonstrate the stability and accuracy of the network timing

synchronization. Therefore, it is assumed that nodes in tier 1

have already acquired timing from the master, and the focus is

on demonstration of slave-slave timing propagation and

synchronization.

B. Phase II

The master initializes the blink cycles transmitting a timing

#1570035263

Fig. 2. The (a) architecture block diagram and (b) photo of the prototype

hardware for each node.

signal; in this implementation, a length 31 m-sequence is used

for the timing signal. Slave nodes receive this sequence,

perform correlation and peak detection to extract the timing

information, use the result to correct their internal timers, and

transmit the same sequence according to their tier associations.

The blink algorithm proposes that nodes on odd (or even) tiers

transmit simultaneously. In the current hardware

implementation, in order to guarantee orthogonality of the

nodes’ signals, a TDMA approach was implemented. In the

case of large networks, and to preserve the distributed

characteristics of the algorithm, random slots can be assigned

to different tiers. In this work, each node transmits on the slot

time associated with its tier. If the number of adjacent nodes

with established connection between tiers is denoted by 



and the slot time is 



, then the blinking cycle will take a

total time of 







During the blinking cycles the slave nodes, after receiving

the timing signals from their neighbors, measure the TOA

relative to their own local timers,













, where 



is the

measured TOA of node k. Using the learned propagations

delays from phase I,















, each node compares the two

values and compute its own offset related to all its neighbors



















. The nodes then use a consensus algorithm

where each neighbor node can be weighted with a different

factor to provide a timing correction value as































, (3)

where 



, represent the weighting factor for each link. The

blinking cycles continue indefinitely. Reference signals from

the master “pull” the timing in the network to agree with the

master timing. Timers on tier one nodes are influenced by

master reference signals as well as neighbors and the timing

information from there propagates through the network.

Figure 1.b depicts the working of the algorithm. At the

beginning, in TDMA slot 1, nodes on Tier 1 are in

transmitting mode and nodes on Tier 2 and Tier 3 are on

receiving mode. Due to 



restrictions, the nodes on Tier 3 do

not detect the signals from Tier1 and therefore do not correct

their timers. During the next timeslot, the nodes on Tier 2

transmit and the others nodes listen. The nodes on Tier 2, after

receiving all the diversity information from their neighbors,

correct their internal timers based on the timing correction

value calculated from (3). This process is repeated every

blinking cycle on each tier accordingly with the TDMA slots.

IV. PROTOTYPE NODE HARDWARE

As a proof of concept, to demonstrate the algorithm

capabilities in real time, custom nodes using UWB as physical

layer signals are implemented. UWB signals are selected for

timing purpose due to the inverse relation between the TOA

estimation error and signal bandwidth. The hardware nodes

are software programmable to provide flexibility and rapid

prototyping given that the timing synchronization algorithm is

implemented completely on digital hardware. The downside of

this approach is that timing accuracy will be limited to the

performance of available commercial hardware (to be

discussed later). Future custom realization of the node

hardware enables more accurate network timing

synchronization.

Each node is composed of two principal elements: a high

speed Analog to Digital Converter (ADC) board from Texas

Instruments and a Xilinx Kintex Field Programmable Gate

Array (FPGA) evaluation board (Fig. 2).

The receiver is implemented using the ADC07D1520 board

that can run at a maximum sampling rate of 3 GSps with 7 bits

of resolution. The available FPGA Virtex 4 in this ADC board

is used as a simple pass-through buffer (no computation). In

order to reduce the frequency of the digital signals, the ADC

output provides four interleaved buses running at a quarter of

the sampling frequency. In the current implementation, due to

internal FPGA clock manager restrictions, the sampling

frequency is set to 2.5 GSps. This reference clock is provided

by the Kintex board to the ADC thought a PCI-SMA

connection. The ADC samples the incoming signals at RF,

after amplification with a Low Noise Amplifier (LNA),

without any frequency down-conversion. While flexible and

consistent with the vision of a true Software Defined Radio

(SDR), the limited speed and resolution of available

commercial ADCs dictates the achievable timing accuracy.

The transmitter is implemented digitally using the Gigabit

Transceivers (GTX) embedded in the FPGA without using

frequency up-converters. A simple Binary Phase Shift Keying

(BPSK) modulation is implemented using two GTXs with

opposite polarity that will feed a power combiner. The GTX

provide great flexibility to design monocycle signals changing

the pulse width accordingly with the clock reference. In order

to meet the Nyquist theorem and relaxing the sampling rate,

the implemented monocycle pulses have a 1.2 ns width,

providing approximately 800 MHz signal bandwidth. After the

power combiner, the BPSK signal passes through a Power

Amplifier (PA) and connects to a discone antenna [10],

#1570035263

through an RF switch.

The developed node is clocked from a low-cost 25 MHz

crystal oscillator with 50 ppm (part per million) accuracy that

is located on the KC705 FPGA board. All the clocks that are

internally used in the FPGA and the reference clock that feeds

the ADC are created from the same crystal reference. The

implemented UWB node, shown on Fig. 2, is fully digital,

modular and easy to modify since all transmitter/receiver

blocks are controlled by the Kintex FPGA.

V. FPGA IMPLEMENTATION OF THE BLINK ALGORITHM

The accuracy of the blink algorithm is intrinsically

related with the TOA estimation error. A well-known and

simple method to estimate the signal TOA consists of peak

detection after cross correlation [11]. It is also known that

peak detection will not provide the right time under non-Line-

Of-Sight (LOS) conditions; but, can be used as a starting point

for further processing [12, 13]; however this is not

implemented in our hardware yet. The blink algorithm

proposes that nodes learn the channel multipath signatures,

and use that information to create proper reference templates.

In this implementation, in order to simplify the algorithm, the

same template for all nodes in the network is used. Better

results can be obtained if the nodes continuously update their

reference templates as a function of propagation channels

variations.

The timing signal used for the TOA estimation must

provide a high ratio for the autocorrelation peak to the side-

lobe peak. An m-sequence of length 31 was selected for the

timing signals since it meets the minimum requirements. The

length of the sequence is constrained by the correlator size

feasible with the existing hardware. Longer sequences give

higher peaks; but, increase the complexity of logic hardware.

The sequence transmission is controlled by the TX Logic

block as shown in Fig. 3. As previously explained, a TDMA

scheme was implemented where each node transmit on

different slots according to the node identification and tier

association to achieve better peak detection and reduce the

interference between nodes on the same tier.

The most logic-consuming and timing-constrained

algorithm to be implemented in the FPGA is the parallel

correlator, because it has to run in real time and its size

increases with the sequence length and the ADC resolution.

Hence, to meet the timing constraints imposed by the FPGA

logic, a dual data rate input is implemented, which transforms

the 4 buses coming from the ADC at 625 MHz into 8 buses

running at 312.5 MHz. Furthermore, to reduce complexity and

satisfy real time operation, 4 bits (sign plus 3 bits magnitude)

are taken from each bus. The correlator was implemented in

parallel following an architecture called PTT (Parallel

samples, Parallel Coefficients, Time division multiplexing).

This architecture is highly efficient for an FPGA

implementation and works as follows. On each clock cycle, a

 array, where  is the number of samples in parallel and 

is the number of correlation points calculated simultaneously,

is processed by the PPT correlator. On the next clock, another

array is processed and after  cycles, where  is the

template length, the correlation is completed [14]. In the

current implementation, considering the sampling rate of 2.5

GHz and the sequence length of 62 ns, the template length will

be =160; therefore, the correlator has 20 arrays as those

described in Fig. 4. In the implemented design, s=8, k=8, and

each sample and coefficient has a 4-bit width; therefore, in

each clock cycle, 64 multiplication have to run in parallel for

every array. The major challenge in this design is satisfying

the zero latency restriction for samples propagation (vertical

arrows in Fig. 4) between multipliers in the same array given

the high sampling rate. Finally, the outputs of the parallel

correlator feed the peak detector block.

The peak detector compares each branch at the correlator

output against the 



threshold. In the current proof-of-

concept experiments, static or quasi-stationary channel

conditions are assumed; but, this detector can be easily

updated to handle channel variation implementing a dynamic

threshold algorithm like that one proposed in [15]. The peak

detector determines the time and corresponding branch at

which a correlation peak happens. This information offers two

resolution levels corresponding to (1) the coarse correction,

that will be applied to the master timer and (2) the fine

correction that will be used to adjust the transmission time.

The resolution of coarse correction is limited by the system

clock at which the correlator is running, in our case 312 MHz

or 3.2ns. The resolution of the fine correction is limited by the

rate of the GTX that is 0.2 ns in this design. The current

Fig. 3. Blink algorithm block diagram. The dash line shows the boundary

between the analog hardware and the digital algorithm implementation

inside the FPGA.

Fig.4. Parallel correlator array.

#1570035263

experiments were conducted without controlling the fine error;

this will be included in further implementation.

During the learning process in Phase I, the nodes estimate

the propagation delay,















, between their neighbors. This

phase of the algorithm is implemented by averaging the

roundtrip measurements between the nodes. The roundtrip

measurement works as follow: (1) node A resets its timer and

transmit a sequence, (2) node B uses the correlator and peak

detectors to determine the TOA and immediately resets its

timer and triggers the TX Logic, (3) node A captures its timer

value after peak detection. The process is repeated 16 times to

reduce errors in the TOA estimation and the average value is

stored at the "Diversity Mem" block. The hardware

implementation of the roundtrip procedure is shown in Fig. 3

as "RT Logic".

During the blinking cycles, when a peak is detected, the

master timer captures its current value that represent the

estimated 



. Depending on their tier location, the node waits

for the timing diversity update from its neighbors, as shown on

Fig. 1, and then (3) is computed using the previously learned

values 





, that are stored on the diversity memory block. As a

result, the offset coarse error, 



, is obtained and the master

timer is corrected. Lastly, the fine error will be applied to the

"TX Logic" as a shift in the sequence to be transmitted (not

implemented yet). The node will transmit again, once the

offset correction is done and following the TDMA scheme.

Finally, the "Broad Sync Logic" block implements the

initialization procedure of blinking cycles described in Section

III B, and the "Control State Machine" block controls all the

phases involved in the blink algorithm.

VI. SIMULATION AND EXPERIMENT RESULTS

The blink algorithm was fully implemented using the

Matlab System Generator, a tool that allows running real-time

hardware simulations in a Simulink environment. Simulations

indicate the anticipated performance prior to actual hardware

experiments. Current setup consists of three slave nodes and

one master node. In order to consider the channel multipath

signature, a received signal was recorded and used in the

simulations. The distance between the nodes is simulated by

the channel propagation delays. In the presented simulations,

Node 0 belongs to tier 1, Node 2 belongs to tier 2 and Node 1

belongs to tier 3. Simulations show that synchronization

convergence time is very fast (25.6 ). This time is measured

from the beginning of Phase II of the blink algorithm until the

time that offset error between nodes reach the best system

resolution, which in our implementation is one period of the

master timer clock. The fast convergence is reasonable as the

described network is simple with only one node per tier and

one master node.

Actual hardware experiments were conducted in three

different environments, indoor office with LOS and NLOS,

and outdoor environments. For the NLOS experiment, node 0

has direct LOS link with node 2 while the direct LOS link

between nodes 1 and 2 is obstructed by a metal shelf (Fig. 6).

In order to compare the simulations results against experiment,

the network topology was the same as described in the

simulation setup and the distance between the nodes are

expressed in Table 1.

TABLE I

DISTANCE BETWEEN NODES IN METERS

Master-N0

N0-N2

N2-N1

Indoor (LOS)

1.8

2.4

Indoor (NLOS)

0.9

1.6

1.7

Outdoor

1.9

3.7

4.5

Simulation

5.4

7.5

In order to measure the relative time offset between the

synchronized nodes, an additional timeslot was introduced in

our TDMA scheme where all the nodes transmit

simultaneously accordingly with their internal timer. On this

additional slot, each node transmits the timing sequence and a

short pulse that will be used for measurement. If the slave

nodes are synchronized, all these pulses should be aligned.

This extra slot does not affect the blink algorithm since is can

be consider as an additional tier in the network. The three

pulses coming from the slave nodes are measured by

connecting the nodes through identical coaxial cables to a

high-speed four-channel real-time oscilloscope, Tektronix

DPO71254B (Fig. 5.b). The oscilloscope is equipped with a

statistical analysis tool that allows measuring the mean (offset)

and standard deviation (jitter) of the signals received at its

different channels. Node 2 is considered as reference to trigger

the scope since it is the node that has the most timing

diversity. Therefore, the statistics measured are carried out

considering node 2 as the time reference. In Fig.5.a, the pulses

captured buy the scope are shown with their time offsets.

The results on Table 2, show the mean values and standard

deviation over four thousand captured pulses confirming that

timing jitter values agree with the simulation predictions and

equal to approximately one clock period (3.2 ns).

TABLE II

STATISTICAL ANALYSIS

Indoor(LOS)

Indoor(NLOS)

Outdoor

Node

Node 1

Node

Mean (ns)

0.118

0.886

0.820

2.812

1.554

2.822

Std. dev. (ns)

3.316

3.358

3.412

3.211

3.011

3.002

a) b)

Fig. 5. a) Measured timing offset of the synchronized network.

b) Experimental setup.

#1570035263

Evaluating the mean value results, indoor LOS experiment

has the best performance due to high SNR, achieving sub-

nanosecond offset accuracy. In the case of indoor NLOS

experiment, the mean error of node 0 is better than node 1,

since the NLOS condition reduces the TOA estimation

accuracy.

For outdoor experiments, the timing offset increases due to

degradation in TOA estimation directly related to lower SNR.

The standard deviation in this case, representing the timing

synchronization jitter, is in line with the implemented coarse

timing accuracy of the implemented system clock (3.2 ns).

Therefore, it is expected that implementing fine timing

correction and increasing the master timer clock will push the

jitter under the sub-nanosecond range.

Finally, the self-synchronized wireless sensor network is

setup to provide coherent waveforms for distributed beam-

forming applications. An additional receiving antenna was

positioned at an equal distance from all nodes and connected

to the scope. Coherent addition of signals wirelessly received

from different nodes (Fig. 7) is a nice demonstration of the

benefits of the nanosecond network timing synchronization

scheme.

VII. CONCLUSION AND FUTURE WORK

This work demonstrates experimentally that wireless

network nodes can all synchronize in a cooperative and

distributed manner with a timing accuracy that is markedly

lower than the propagation delay between them. Under good

SNR conditions, the network can attain sub-nanosecond

accuracy as was demonstrated in the indoor experiments. The

timing jitter in the synchronized network is limited primarily

by the coarse correction applied in the master timer. The jitter

can be reduced in future work by adopting optimal correlation

architectures, increasing the frequency of the master timer and

applying post signal processing. In the current

implementation, the timing offset in the synchronized network

is constrained by the resolution of the TOA estimation

algorithm. In the future, offline signal processing to improve

the detection capabilities and improve the synchronization

under non-LOS conditions can be implemented. Finally, future

hardware realizations can benefit from monolithic low-cost

low-energy realization of the sensor nodes.

REFERENCES

[1] IEEE Std. for a Precision Clock Synchronization Protocol for Networked

Measurement and Control Systems, IEEE Std. 1588–2008, 2008.

[2] J. Elson, et al., “Fine-grained network time synchronization using

reference broadcasts,” in Proc. 5th Symp. OSDI, pp. 147-163, 2000.

[3] S. Ganeriwal, et al., “Timing-sync protocol for sensor networks,” in

Proc. Int. Conf. Embedded Network Sensor System, pp. 138-149, 2003.

[4] M. Maroti, et al., “The flooding time synchronization protocol,” in Proc.

2nd Int. Conf. Embedded Net. Sensor Syst., pp. 39-49, 2004.

[5] S. Bholane, D. Thakore, “Time synchronization in wireless sensor

networks,” Int. Journal of Scientific & Engineering Research, vol. 3, no.

7, July 2012.

[6] J. Esterline, “Oscillator phase noise: theory vs practicality,” Greenray

industries Inc. report, Apr. 2008.

[7] Y. Shen and M. Z. Win, “Fundamental limits of wideband localization

part I: a general framework”, IEEE Transactions on Information Theory,

vol. 56, no. 10, pp. 4956-4980, Oct. 2010.

[8] P. Carbone, et al., “Low complexity UWB radios for precise wireless

sensor network synchronization,” IEEE Transactions on Instrumentation

and Measurement, vol. 62, no. 9, pp. 2538-2548, Sept. 2013.

[9] S. Niranjayan and A. F. Molisch, “Ultra-wide bandwidth timing

networks”, in Proc. IEEE Int. Conf. on UWB, pp. 51-56, Sept. 2012.

[10] Y. Zhang and A. K. Brown, “The discone antenna in a BPSK direct-

sequence indoor UWB communication system,” IEEE Transactions on

Microwave Theory and Techniques, vol. 54, no. 4, pp.1675-1680, April

2006.

[11] A. F. Molisch, Wireless Communications, 2

edition, IEEE Press –

Wiley, 2011.

[12] S. Gezici, et al., “Localization via ultra-wideband radios: a look at

positioning aspects for future sensor networks,” IEEE Signal Processing

Magazine, vol. 22, no. 4, pp. 70-84, 2005.

[13] I. Guvenc, at al., “TOA estimation for IR-UWB systems with different

transceiver types,” IEEE Transactions on Microwave Theory and

Techniques, vol. 54, no. 4, pp. 1876-1886, April 2006.

[14] Altera, white paper: “High performance, low cost FPGA correlator for

wideband CDMA and other wireless applications”, ver. 1.0, May 2003.

[15] M. Segura, et al., “Wavelet correlation TOA estimation with dynamic

threshold setting for IR-UWB localization system,” in Proc. IEEE Latin-

America Conf. on Communication, Sept. 2009.

[16] S. Niranjayan, A. F. Molisch, Timing synchronization of wireless

networks, US patent pub. no. US2013/0322426 A1.

Fig. 7. Measured beam forming experimental results.

Fig. 6. Experiments setup: Top Left, indoor with LOS. Top Right, indoor

with NLOS. Bottom, outdoor with LOS.