Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the...

12
Sep 2009 Hart (Cisco Systems) Slide 1 doc.: IEEE 802.11-09-0943r1 Submission Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01 N am e A ffiliation A ddress Phone Em ail Brian H art Cisco System s 170 W estTasm an D rive, San Jose, CA 95134 +1-408-526-3346 [email protected] Hajim e Suzuki(source ofm uch data butnot editorially responsible) CSIRO PO Box 76 Epping, N SW 1710, A ustralia Hajime.Suzuki@ csiro.au G raham Daniels (source ofm uch data butnoteditorially responsible) CSIRO PO Box 76 Epping, N SW 1710, A ustralia Graham.Daniels@ csiro.au

Transcript of Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the...

Page 1: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 1

doc.: IEEE 802.11-09-0943r1

Submission

Impact of the Fluorescent Light EffectUpdate from 09/816r2

Date: 2009-09-01

Name Affiliation Address Phone Email Brian Hart Cisco Systems 170 West Tasman Drive,

San Jose, CA 95134 +1-408-526-3346 [email protected]

Hajime Suzuki (source of much data but not editorially responsible)

CSIRO PO Box 76 Epping, NSW 1710, Australia

[email protected]

Graham Daniels (source of much data but not editorially responsible)

CSIRO PO Box 76 Epping, NSW 1710, Australia

[email protected]

Page 2: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 2

doc.: IEEE 802.11-09-0943r1

Submission

Impact of the Fluorescent Light EffectOverview

Situation

• The Functional Requirements (09/451) assume no fluorescent light effect because recent measurements have not replicated this effect− Likely because fluorescent lights using electronic ballasts do not create the fluorescent

light effect

Complication

• Buildings with fluorescent lights using magnetic ballasts do create the fluorescent light effect.

• Although magnetic ballast sales are now much reduced and magnetic ballasts have a finite lifetime, it takes time for the installed base to taper to zero

• We estimate that in 2012, 35% of educational buildings, tapering to 0% by 2018, are likely to suffer from the fluorescent light effect.

• We show that the measured fluorescent light effect was significant for 10% of locations, and has a greater effect on DL-MU-MIMO than on SDM, TXBF or UL-MU-MIMO

Key Lines

• The fluorescent light effect will not be predominant yet it will have an appreciable effect in about 10% of locations in 35% of educational buildings (tapering to 0% over 6 years)

• This should be kept in mind by implementers but the effect may not be significant enough to warrant its own channel model or simulation scenario

Summary

Page 3: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 3

doc.: IEEE 802.11-09-0943r1

Submission

Technical Summary of Magnetic and Electronic ballasts in fluorescent lights*

• Since fluorescent tubes cannot effectively regulate their current use, ballasts are connected in series to appropriately limit current

• Fluorescent lights have traditionally used a large inductor as a ballast: hence “magnetic ballast”

− The applied voltage, varying at 50 or 60 Hz, flows through the inductor and tube, and creates oscillating periods of reflection and absorption^

− Results presented in 09/816r2 were of magnetic ballasts

• Modern fixtures use electronic ballasts that rectify mains power and modulate it, usually to 20 kHz or higher

− Above 20kHz, tubes are about 9% more efficient, and flicker is eliminated

− The period of any variation is 6.25 OFDM symbols

Page 4: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 4

doc.: IEEE 802.11-09-0943r1

Submission

USA Market Data, Regulations and LifetimesMarket Data+ (see figure)

Regulations *

• After 2006, manufacturers cannot make new fixtures with magnetic ballasts

• “For replacement only” magnetic ballasts are to be phased out after 2010

Ballast Lifetimes

• 50000 hours is the common age for ballasts^, leading to various lifetimes

− Average^: 3400 hours/year: 15 years

− Homesx (kitchen&, garage): 3 hours/day, 7 days/week: 46 years

− Education+: 10 hours/day, 5 days/week: 19 years

− Enterprise+: 11.5 hours/day, 6 days/week: 14 years

− Hospitals+: 16 hours/day, 7 days/week: 9 years

− Always: 24 hours/day, 7 days/week: 6 years

&For aesthetic reasons, these are likely hidden above cupboards, with reduced contribution to multipath

Page 5: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 5

doc.: IEEE 802.11-09-0943r1

Submission

Evolution of Installed Base (simulated): Mostly OK, except for education

• Assume 0% installed base in 1986

• Assume 1/Mean#Years ballasts fail every year

• Assume replacements are in proportion to sales ratio (using previous slide)

• This is a model; reality is likely to differ (worse/better?)

− Passes a basic check: 2/3 non-problematic buildings in 2009

• For the average building after 2012, 11ac is mostly OK and getting better

• For education (a big wireless user), fluorescent lights with magnetic ballasts are a concern for many buildings until 2018

&For aesthetic reasons, these are likely hidden above cupboards, with reduced contribution to multipath

Page 6: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 6

doc.: IEEE 802.11-09-0943r1

Submission

Subcarrier Capacity Impact from the Fluorescent Light Effect (1)• The large amplitude or phase changes in 09/816r2 don’t tell us about the

impact on capacity

• Therefore let’s analyze capacity as per Perahia and Stacey’s book, for SDM, TXBF and DL-MU-MIMO

• SDM: Y[2x1] = H[2x2].X[2x1] + N[2x1]− 2x2 MMSE equalizer− For the measurement set, SDM is also the best model for UL-MU-MIMO

• TXBF: Y[2x1] = H[2x2] Ve[2x2].X[2x1] + N[2x1]− Use the first channel as an estimate, He, for the next 130 ms.− X is precoded with Ve, from SVD He = Ue.Se.Ve’− 2x2 MMSE equalizer

• DL-MU-MIMO: y[1x1] = H[1x2] De[2x2].X[2x1] + n[1x1]− Considering 1 user w/l/o/g, then doubling capacity− Use first channel as an estimate, He, for the next 130 ms. − X is precoded with De, the channel’s “MMSE inverse” with a power constraint− De0 = (rho/NTX.He‘.He+PhiZ)-1*He‘ // MMSE inverse,

– Where rho = signal power, PhiZ = noise autocorr, normalized to 1− De = De0 .* √(NTX ./ mean(abs(De0(:)).^2)) // Power constraint− 1x1 MMSE equalizer

Page 7: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 7

doc.: IEEE 802.11-09-0943r1

Submission

Subcarrier Capacity Impact from the Fluorescent Light Effect (2)• For SDM, TXBF, e[2x1] = MMSE estimate of X, minus X[2x1]

• For DL-MU-MIMO, e[1x1] = MMSE estimate of x, minus x[1x1]

• J = E(ee’), so diagonal entries equal the error power per stream

• SNRi = (1-J(i,i))/J(i,i)

• C = Σi log2(1+SNRi)

• A capacity is claimed if it is available to 90% subcarriers (from the pool of all subcarriers and all locations)

− i.e. we’re reporting the 10% worst subcarriers

• SDM and TXBF calculations of equalizer coefficient W and error autocorrelation J as per Perahia and Stacey’s book

• DL-MU-MIMO calculations of MMSE equalizer coefficient and error autocorrelation:

− w= √(rho/NTX).[1,0].(H.De)'.(rho/NTX.(H.De)(H.De)'+Phiz)-1 − J = rho/NTX.w.(H.De)(H.De)'.w' + w.Phiz.w' -

2√(rho/NTX).Re(w.H.De.[1; 0]) + 1− A ML design likely could improve upon the performance of this

Page 8: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 8

doc.: IEEE 802.11-09-0943r1

Submission

0 20 40 60 80 100 120 1400

1

2

3

4

5

6

7

8

9

10

Time (ms)

Cap

acity

(b/

sym

/sub

carr

ier)

TXBF without lightsTXBF with lights

DL-MU-MIMO without lights

DL-MU-MIMO with lights

SDM without lightsSDM with lights

TXBF

SDM

DL-MU-MIMO

The DL-MU-MIMO 90% subcarrier capacity is sensitive to the fluorescent light effect

• With lights− TXBF and DL-MU-MIMO variation every 10 ms is clearly visible

−The variation is large for DL-MU-MIMO, and capacity can drop slightly below simple SDM performance

−The variation for TXBF is much less significant

−As expected since null steering is much more sensitive to a poor channel estimate than TXBF.

−SDM shows no variation since there is no channel prediction

• Without lights, −No variation −TXBF > DL-MU-MIMO > SDM−This is as expected since SVD > “MMSE inverse” > I

Page 9: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 9

doc.: IEEE 802.11-09-0943r1

Submission

But it’s not so bad looking across a 20MHz channel and considering all the deciles

• This time:−Average the capacity across 20 MHz so isolated bad subcarriers get averaged out−Divide the data into deciles and report the mean within each decile −Only plot DL-MU-MIMO with lights on

• The worst 10% of channels remain bad, but the other deciles only show small degradations

0 20 40 60 80 100 120 1400

2

4

6

8

10

12

14

Time (ms)

Mea

n C

apac

ity a

cros

s 20

MH

z by

dec

ile (

b/sy

m/s

ubca

rrie

r)

0-10%10-20%

20-30%

30-40%

40-50%50-60%

60-70%

70-80%

80-90%90-100%

Page 10: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 10

doc.: IEEE 802.11-09-0943r1

Submission

Summary

• By 802.11ac’s expected ratification, the fluorescent light effect will not be predominant, yet it will have an appreciable effect in about 10% of locations in an estimated 35% of educational buildings (tapering to 0% of educational buildings by 2018)

• DL-MU-MIMO is a valuable technology yet is more sensitive to channel variation than SDM, TXBF, or UL-MU-MIMO since a 1x1 receiver cannot perform spatial processing and so multi-user-interference with a MMSE MIMO equalizer appears as additional noise.

−Likely a ML equalizer could improve upon this

• The fluorescent light effect should be kept in mind by implementers yet the effect may not be significant enough to warrant its own channel model or simulation scenario

Page 11: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 11

doc.: IEEE 802.11-09-0943r1

Submission

Backup Slides

Page 12: Doc.: IEEE 802.11-09-0943r1 Submission Sep 2009 Hart (Cisco Systems) Slide 1 Impact of the Fluorescent Light Effect Update from 09/816r2 Date: 2009-09-01.

Sep 2009

Hart (Cisco Systems)Slide 12

doc.: IEEE 802.11-09-0943r1

Submission

Effect of Patch Orientation

• Some concern was expressed that the ceiling-directed client patch had unrealistic vertical gain and this exacerbated the measured effect

• Let’s try to assess this.

• (1) LOS was 8deg above the horizon

• (2) The ceiling reflection was 22deg above the horizon; here the patch had ~1.5dB greater gain than LOS.

− 1.5 dB is not a great difference, and clients should have consistent gain across 8-22deg above horizon, so this suggests the measurement was not unduly affected by the patch orientation

• (3,4) However, the lights immediately above TX and RX had angles of 45 and 51deg above the horizon; here the patch had ~5.5dB greater gain than LOS.

− This may be significant, yet there is no obvious path from TX to RX via these lights.

• At most, there is some weak evidence for concern.

12

34