Wireless Body Area Network Case Study: UWB based Human Motion Tracking 1.

Wireless Body Area Network Case Study:UWB based Human Motion Tracking

1

Outline

2

• Localization via UWB

• ToA estimation of UWB signal

• TOA based human motion tracking System setup and problem formulation Self-calibration method Localization method Summary and outlook

Localization via UWB

based on: H. Arslan, Z. N. Chen, M.G. di Benedetto:

Ultra Wideband Wireless Communication, Wiley 2006

3

4

Localization via UWB

• UWB IR is good candidate for short-range and low-rate communication networks– Nodes operating on battery (autonomy)

– High precision ranging capability

• Location information is derived from radio signals between the target node (agent) and reference nodes (anchors)

• Common radio based positioning systems can be categorized into– angle of arrival (direction of arrival),

– signal-strength, and

– time-based approaches

5

Angle of Arrival (AOA)

• Measure the AOA of target node’s signal at different reference nodes– Antenna arrays required

– AoA measurement based on phase (time) difference of received wave-front at different antennas

• In 2 dimensional space, 2 reference nodes are enough

6

Cramer-Rao Lower Bound (CRLB)

• Lower bound on the variance of an unbiased estimator• For a uniform linear antenna array with N antennas, with a

separation distance d

• For θ = nπ, n {0,±1,…}, the CRLB diverges – Two dimensional antenna grid necessary

• Not suited for UWB positioning– More antennas => higher complexity and increased costs

– Large number of multipaths => multidimensional search for maximum likelihood AoA estimation

2 2 2

constˆvarSNR 1 sind N N

7

Received Signal Strength

• Path loss model relates distance between two nodes to energy loss

• In 2D: Distance estimates from 3 reference nodes are required for triangulation

8

Cramer-Rao Lower Bound (CRLB)

• Dependence on channel characteristics• Very sensitive to the estimation of channel parameters

– E.g. path loss exponent (np), variance of log-normal shadowing (σsh

2), i.e.

• UWB signal characteristic (huge bandwidth) is not exploited

2

ln10ˆvar10

sh

p

dd

n

9

Time-Based Approaches (1)

• Measurements of the propagation delay between nodes– Two nodes (A and B) with common clock

• Node A sends time-stamped signal• Node B receives delayed version and can estimate time of arrival (ToA) and

also the distance by correlation with a template signal• For single path and additive white Gaussian noise (AWGN) channel,

the CRLB of is given by

– Beff is the root mean square signal bandwidth for signal s(t) with Fourier transform S(f)

• UWB very beneficial here! • However, node synchronization is an important assumption

– Accuracy of clocks plays an important role

d

10

Time-Based Approaches (2)

• N reference nodes (at positions ri) with ToA estimates τi do positioning via least squares minimization

– The weights wi reflect reliability of ToA estimates

– Method becomes optimal if ToA measurements are modeled as true ToAs plus independent Gaussian noise samples

• Main sources of error in realistic environment– Multipath propagation– Non-line-of-sight propagation (direct path is blocked)– Interference from other nodes or coexisting systems

2

1ˆ argmin

N i

i iiw

ct

t rt

ToA Estimation of UWB Signal

11

12

Signal Model (1)

• The ultra-wideband channel between the transmitter (agent) and the receiver (anchor)

• The received signal at the anchor

where is the transmit signal and is an AWGN with PSD

1

L

l ll

h t t

1

( ),L

l ll

r t s t z t

s t z t

0 2N

13

Signal Model (2)

• The Anchor is located at and the agent is

located at

• If the agent and the anchor are in a LOS

• The goal of ToA estimators is to estimate given

, ,T

x y zt t t t

1

-

c

t r

, ,T

x y zr r r r

1 r t

14

ToA Estimator: Matched Filter

• Assumes that the first arriving path is the strongest path

• ToA estimate

Low complexity

ˣ In harsh propagation environments, the strongest received echo may not coincide with the first path

MFˆ arg max dr t s t t

15

ToA Estimator: De-convolution

• Extracts all the paths by de-convolving the transmitted signal

from

• Algorithms: WRELAX [1], CLEAN [2]

• Let are the detected paths

• ToA estimate

Can work even under harsh propagation environment

× Computational complexity

r t

[1] J. Li and R. Wu, “An Efficient Algorithm for Time Delay Estimation”, IEEE Tran. on Sig. Proc., Aug. 1998.

[2] J. A. Högbom, “Aperture Synthesis with a Non-Regular Distribution of Interferometer Baselines”, Astron. and Astrophys. Suppl. 15, 417-426.

16

ToA Estimator: Search-Back Window and Thresholding

• Trade-off between matched filter and de-convolution approaches

MF based ToA estimator

Search back for the first cross-correlation crossing witin the window

w MF MFˆ ˆT W r t MF̂

SBWT̂

• ToA estimate

• The search window size (W) and the threshold (VT) has to

be adapted to the channel characteristics

SBWT Wˆ min : d TT r t s t t V

TOA based Human Motion Tracking

17

• Wide area of applications– Rehabilitation– Animation– Sports

Applications of Human Motion Tracking

18

Image copyrights apply

Body Motion Tracking: State of the Art

• Optical systems

Provide reliable tracking

× Dedicated infrastructure and skilled operators required

• Inertial systems

No LOS restriction and high sampling rate

× Prone to drift errors

• Magnetic systems

Accurate and no LOS restriction

× Prone to interference from nearby ferromagnetic materials

19

UWB based Human Motion Tracking Systems

• Advantages

– Low complexity

– Possible reuse of existing communication centric UWB BAN

• Challenges

– Multipath propagation and potential of NLOS conditions

– Limited choice of fixed anchor node positions

– Anchor location uncertainty

• Considering geometric constraints promises performance improvement

20

• The system consists 3 types of nodes

• Agent: extremely low-cost and low-power transmit-only asynchronous node

• Anchor: Low-complexity node which forwards the received signal to the cluster head

• Cluster head: a computationally capable device that implements the motion tracking algorithm. (E.g. a smart phone)

21

System Setup

Cluster head

: Anchor: Agent

Synchronization Requirements

• Agents are transmit-only asynchronous nodes– Transmit beacon signals

• Clocks of the anchors are frequency synchronized– E.g.: connected via e-fiber, exchange pilot sequences

• In here, the anchors are connected to the cluster head via cable

• The communication link between the anchors and the agents is not considered

22

TOA based Range Measurements (1)

• Error free distance measurement between anchor n and agent m

23

TOA based Range Measurements (2)

• We consider only LOS measurements• Range measurement between anchor n and agent m

• Ranging error model [Qi’03]

• Before calibration– Uknowns:

• Goal of calibration: estimate

24

[Qi’03]: Y. Qi, “Wireless geolocation in a non-line-of-sight environment,” Ph.D. dissertation, Princeton University, Dec. 2003.

Formulation of the Self-Calibration Problem

• The agent locations are not surveyed

• The range measurements can be gathered from a single moving agent

• Vector of NrNt range measurements

: covariance matrix of the range measurements

map the offsets to the right range measurement

25

ML Solution of the Self-Calibration Problem

• Jointly estimates all the unknown parameters• Maximum likelihood (ML) estimator

• A non-convex optimization problem• Relaxation to a SDP problem (convex problem)

– Serves as a good initialization for the ML estimator

26

• Defines reference coordinate system and central clock

• Accounts body imposed constraints

Performance of the Self-Calibration Method

• Number of anchor (Nr)= 6, Number of agents (Nt)= 30, σrange= 2 cm

27

Localization Phase

• Ranging offsets of the anchors are calibrated out• Range measurement between anchor n and agent m

• Anchor locations are known up to some uncertainty– Fixed on the torso (mobile); Calibration error

• Localization method estimates the unknown location and offsets of the agents– Accounts the anchor location uncertainties

28

Localization Method: ML Solution

• The ML solution of the localization problem

• A non-convex optimization problem• Relaxed to a SDP problem• Refinement with ML estimator

29

Accounts body imposed constraints

Performance of the Localization Method

• Nr = 6, Nt = 30, σanchor_error= 5 cm

30

• Agent nodes transmit PN sequence

31

Demonstrator System

Agents

Anchors

1-bit ADC receiver board

LANs

Summary

• TOA based localization systems benefit from the very wide bandwidth of UWB signals

• TOA based human motion tracking system is presented

• Low complexity requirements (Asynchronous agents, Frequency synchronous anchors)

• Calibration phase: estimate clock offsets and locations of the anchors

• Localization phase: estimate the locations of asynchronous agents

32

Outlook

• Real-time demonstration of the developed algorithms

• Localization accounting NLOS conditions

– Propagation effect of the body

• Utilizing the captured movements to detect activity (e.g. walking, sitting and standing)

• Communication link between the anchors and the cluster head

33