24-Aug-04 doc.: IEEE 802.15-04-0447-00-004a Project: IEEE

24-Aug-04

Ulrich G. Schuster, ETH Zurich1

doc.: IEEE 802.15-04-0447-00-004a

Submission

Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (Project: IEEE P802.15 Working Group for Wireless Personal Area Networks (WPANsWPANs))

Submission Title: [Indoor UWB Channel Measurements from 2 GHz to 8 GHz]Date Submitted: [24 August, 2004]Source: [Ulrich G. Schuster] Company [Communication Technology Laboratory, ETH Zurich]Address [Sternwartstr. 7, ETH Zentrum, 8092 Zürich, Switzerland]Voice:[+41 (44) 632 5287], FAX: [+41 (44) 632 1209], E-Mail:[[email protected]]Re: [IEEE 802.15.4a Channel Modeling Subcommittee Call for Contributions]

Abstract: [This presentation describes UWB channel measurements from 2 to 8 GHz, conducted in twooffice buildings at ETH Zurich, Switzerland. Measurements were taken for LOS, OLOS and NLOSsettings in a corridor and a large entrance lobby, with transmitter-receiver separations ranging from 8 m to28 m]

Purpose: [To provide additional data for the proposed generic 802.15.4a channel model and discuss someof the modeling aspects used in the generic model]Notice: This document has been prepared to assist the IEEE P802.15. It is offered as a basis fordiscussion and is not binding on the contributing individual(s) or organization(s). The material in thisdocument is subject to change in form and content after further study. The contributor(s) reserve(s) theright to add, amend or withdraw material contained herein.Release: The contributor acknowledges and accepts that this contribution becomes the property of IEEEand may be made publicly available by P802.15.

Indoor UWB Channel Measurementsfrom 2 GHz to 8 GHz

Ulrich Schuster and Helmut Bolcskei,

Swiss Federal Institute of Technology (ETH), Zurich

August 24, 2004

© 2004 Communication Theory Group 1

Objectives

Main goal: establish and verify UWB channel models suitable fortheoretical analysis, concerning

• Tap statistics

• Stochastic degrees of freedom

• Uncorrelated scattering assumption

Genuine focus was not IEEE 802.15.4a channel modeling work, hence notall 802.15.4a channel model parameters could be extracted.


Measurement Setup — Schematic

Minicircuits ZVE 8G

Diadrive positioner

Skycross UWB antenna

Skycross UWB antenna

custom RF amp

HP 8722D VNA

2 m 2 m

2 m

25 mH&S Sucoflex 104 cables

Measurement Control


Measurement Setup — DetailsMeasurements were taken in the frequency domain

• HP 8722D vector network analyzer (VNA), 50 MHz – 40 GHz

• Minicircuits ZVE 8G power amplifier, 2GHz – 8 GHz, 30 dB gain

• Skycross SMT-3TO10M UWB antannas (prototype), Omni

• Custom RF amplifier, 20 dB gain up to 10 GHz, NF < 6

• H&S Sucoflex 104 cables

• Custom modified Diadrive 2000 positioning table

• Control via Matlab (Instrument Control & Data Acquisition Toolboxes)


VNA SettingsThe VNA was equipped with option 12, “direct sampler access”, forimproved dynamic range

• Frequency range 2–8 GHz, divided into two bands

• 1601 points per band, for a total of 3201 points

• 1.875 MHz point spacing

• Max. resolvable delay of 533 ns, equivalent to 160 m path length

• IF bandwidth 300 Hz

• Total sweep time 19 s

• Calibration included all equipment except antennas


Environments

We measured two different environments at the premises of ETH Zurich,Switzerland, in typical European style office buildings.

• Corridor, e.g. for sensing applications; brick walls, windows, concretefloor and ceiling

• Entrance lobby, typical public space; tiled floor, large glass windows,concrete walls

All measurements were taken at night on weekends to ensure a staticchannel.


Corridor Environment

Tx ArrayN

LOS

LOS

12.4 m

12.6 m5.1 m

10.3 m

8.4 m

0.95 m

RF Lab

0 5 10 15 20 m

2.7 m


Lobby Environment

N

0 10 20 30 m

Tx ArrayLO

S

27.2m

9.9m

16.3m

9m

1 2

34 Rx LOS A

B

Tx ArrayN

/QLO

S5

67

Rx OLO

S

827.3m

Rx NLO

S

9


Measurement Methodology

• Virtual array approach to obtain enough small scale data

• Array spacing is 7 cm, 5×9 grid

• measured two arrays per Tx–Rx separation for 90 points total

• recorded one frequency response per array point

• measured several scenarios (LOS, OLOS, NLOS), with various Tx–Rxseparations in each scenario


Sample Impulse Response Magnitude — Lobby LOS

0 20 40 60 80 100 120 140 160-170

-160

-150

-140

-130

-120

-110

-100

-90

-8027.2 m

47 m59.9 m

119.8 m139.6 m

156.8 m

delay-equivalent distance (m)

atte

nuat

ion

(dB)


Average Impulse Response Magnitude — Lobby LOS

0 20 40 60 80 100 120 140 160

-140

-135

-130

-125

-120

-115

-110

-105

-100

-95


atte

nuat

ion

(dB)

27.2 m

47.0 m

59.9m

119.8 m


Average Impulse Response Magnitude — Lobby OLOS

0 20 40 60 80 100 120 140 160

-140

-135

-130

-125

-120

-115

-110

-105

delay-equivalent distance(m)

atte

nuat

ion

(dB)

27.4 m

29.6 m 31.7 m


Average Impulse Response Magnitude — Lobby NLOS

0 20 40 60 80 100 120 140 160-145

-140

-135

-130

-125

-120

-115

-110

-105


atte

nuat

ion

(dB)

22.2 m

34.9 m


Modeling Considerations

The standard wideband fading model

h(t, τ) =N(τ)−1∑

k=0

ak(t)δ(τ − τk(t))ejθk(t)

assumes specular reflections — there are distinct, frequencyindependent propagation paths.

Assumption might not hold for UWB Channels!

Instead, we simply consider the discrete time LTI system h[n]. There is nonotion of resolvable paths; the model is still sufficient for system analysisand design.


Fading Tap StatisticsWe use the small scale spatial variations of the received amplitudeacross the virtual array for statistical analysis.

Goal: find the best fitting distribution within a set of candidate models:

• Rayleigh

• Rice

• Nakagami

• Lognormal

• Weibull

This is a model selection problem. Hypothesis testing is not ameaningful approach.


Model Selection using Akaike’s Information Criterion (AIC)

A measure for the difference of distributions is relative entropy. AIC is anunbiased estimate of the relative entropy difference between acandidate model and the true distribution, given as

−2 log qi(y | Θ(y)) + 2K

with qi(· |Θ) the parameterized PDF of candidate model i, i.i.d. datavector y and ML parameter estimate Θ(y).

From this, it is possible to compute the probability for each candidatemodel of providing the best fit, called Akaike weight wi.

Advantage: no ambiguities due to confidence level selection, test powerand binning.


Akaike Weights, Averaged Impulse Response — Lobby LOS

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Akai

ke W

eigh

t

0 200 400 600 800 1000 1200 1400 1600 1800 2000 2200-130

-120

-110

-100

-90

-80

time domain sample index

atte

nuat

ion

(dB)

RayleighRiceNakagamiLognormalWeibull


Fading Model Selection

We computed Akaike Weights for all scenarios

• Rayleigh provides on average the best fit

• Rayleigh is not good at the start of a cluster

• LOS component is often Weibull distributed

• Lognormal is almost always the worst model

Conclusion: there is not single true model, but Rayleigh still seems to bean appropriate and convenient choice for UWB, except for the first tapsin each cluster.


Small Scale Parameters - Time Dispersion

Mean delay and delay spread often used to characterize timedispersiveness of the channel. They are not the most general description.

Estimates can be computed as

τ =L∑

l=1

h[l]l mean delay

s =

√√√√ L∑l=1

(l − τ)2h[l] delay spread

where the h is the magnitude of the power normalized impulseresponse. Mean and standard deviation can now be computed over allsmall scale positions on the virtual array.


Mean Delay and Delay Spread Statistics — Lobby LOS

Mean Delay Delay SpreadDistance µτ στ µs σs

27 m 27.13 ns 1.74 ns 49.5 ns 2.08 ns24 m 27.15 ns 2.86 ns 49.23 ns 3.37 ns21 m 30.99 ns 2.30 ns 53.62 ns 2.25 ns18 m 29.86 ns 2.11 ns 52.23 ns 1.64 ns15 m 27.26 ns 1.75 ns 49.20 ns 1.63 ns


Mean Delay and Delay Spread Statistics — Lobby OLOS


27 m 49.82 ns 7.78 ns 74.08 ns 7.04 ns24 m 46.86 ns 6.33 ns 71.07 ns 5.91 ns21 m 45.61 ns 5.70 ns 71.23 ns 4.43 ns


Mean Delay and Delay Spread Statistics — Corridor

LOS Setting


12.5 m 7.55 ns 0.88 ns 21.08 ns 1.65 ns10.5 m 10.68 ns 1.69 ns 24.70 ns 2.19 ns8.5 m 9.93 ns 2.15 ns 23.74 ns 2.86 ns

NLOS Setting

Mean Delay Delay Spreadµτ στ µs σs

24.44 ns 1.16 ns 31.11 ns 1.87 ns


Small Scale Parameters - The Saleh-Valenzuela ModelThe proposed 802.15.4a channel model is continuous time and specular:

h(t) =L−1∑l=0

K−1∑k=0

ak,lδ(t− Tl − τk,l,)

with L clusters and K rays per cluster. Ray and cluster arrivals aredescribed by Poisson processes with interarrival probabilities

P(Tl |Tl−1) = Λ exp{−Λ(Tl − Tl−1)}.

Ray and cluster power decay are exponential

E[|ak,l|2

]= E

[|a0,0|2

]exp

{−Tl

Γ

}exp

{−τk,l

γ

}.


Saleh-Valenzuela Model Parameter Extraction

Our discrete time model does not fit this framework⇒ cannot extract allparameters since there are no rays. Using the methodology presented byBalakrishnan in doc. 802.15-04-0342-00-004a, we computed

• Cluster decay coefficient Γ

• Inter-cluster decay coefficient γ

• Cluster interarrival time Λ

The model fit is not always satisfactory, as can be seen in the followingplots. We only extracted S-V parameters for the LOS scenarios, whereclusters were observable.


Cluster Decay — Corridor LOS

0 50 100 150 200 250 300 350 400-18

-16

-14

-12

-10

-8

-6

-4

-2

0

2

Cluster Delay (ns)

Powe

r ( lo

g|a|

2 )

Cluster PowerLinear Least Squares FitΓ = 33.0ns


Intra-Cluster Decay — Corridor LOS

0 10 20 30 40 50 60 70 80 90

-15

-10

-5

0

5

Intra-Cluster Excess Delay (ns)

Tap P

ower

( log

|a|2 )

Tap PowerLinear Least Squares Fit

γ = 16.2 ns


Cluster Interarrival Times — Corridor LOS

0 10 20 30 40 50 60 70 80 900

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Interarrival Time (ns)

Cum

ulat

ive D

istrib

ution

Cluster Interarrival TimesExponential CDF Fit

1

Λ= 29.4ns


Cluster Decay — Lobby LOS

0 50 100 150 200 250 300 350 400 450-15

-10

-5

0

Cluster Delay (ns)

Powe

r ( lo

g|a|

2 ) Cluster Power

Least Squares Fit

Γ = 50.8 ns


Intra-Cluster Decay — Lobby LOS

0 50 100 150 200 250

-20

- 15

- 10

-5

0

5

Intra Cluster Excess Delay (ns)

Tap P

ower

( log

|a|2 )

Tap PowerLinear Least Squares Fit

γ = 59.5 ns


Cluster Interarrival Times — Lobby LOS

0 50 100 150 200 2500

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Cluster Interarrival Time (ns)

Cum

ulat

ive Pr

obab

ility

Cluster Interarrival TimesExponential CDF Fit

1

Λ= 67.5 ns


Large Scale Parameters — Path Loss

The simplest pathloss model consists of a single slope with exponentialdecay

10 log P (d) = G0 + 10ν logd

d0, d ≥ d0

with d0 = 1m, an arbitrarily chosen reference distance, and G0 thereference loss at d0.

Our measurements are not targeted at pathloss extraction; only in threesettings enough large scale data points are available to yield crudeestimates, as can be observed from the following scatter plots.


Path Loss Fit — Lobby LOS

12.6 14.1 15.8 17.8 20.0 22.4 25.1 28.2 31.6 35.5-74

-73

-72

-71

-70

-69

-68

-67

-66

distance (m, log10 scale)

rece

ived

pow

er (d

B)

Impulse Response PowerLinear Fit


Path Loss Fit — Lobby OLOS

20.9 21.9 22.9 24.0 25.1 26.3 27.5 28.8�-79

�-78

�-77

�-76

�-75

�-74

-�73

-�72


rece

ived

pow

er (d

B) Impulse Response Power

Linear Fit


Path Loss Fit — Corridor LOS

7.9 8.9 10 11.2 12.6 14.1 15.8-75

-70

-65

-60


rece

ived

pow

er (d

B)

Impulse Response PowerLinear Fit


Pathloss Coefficients

Setting ν G0

Lobby LOS 1.6 -49 dBLobby OLOS 2.2 -45 dBCorridor LOS 1.2 -51 dB


Documents

24-Aug-04 doc.: IEEE 802.15-04-0447-00-004a Project: IEEE