12-13 Dec. 2005COST 286 1 «Power Line Communication: Application to Indoor and In-Vehicle Data Transmission» Virginie Degardin, Pierre Laly, Marc Olivas.

12-13 Dec. 2005 COST 286 1

«Power Line Communication: Application to Indoor and In-Vehicle Data Transmission»

Virginie Degardin, Pierre Laly, Marc Olivas Carrion,Martine Liénard and Pierre Degauque University of Lille, IEMN/TeliceFrance

COST 286 212-13 Dec. 2005

Why PLC for indoor or in-vehicle communication ?

Most of the in-house electronic equipment are supplied by the LV power line (220V).

Why putting an additional cable between two equipments for exchanging data since there are already connected to the same the line (Power line)?

In a car, the number of “intelligent” sensors, computers.. is continuously increasing. Development of X by wire technique (Replacing mechanical transmission by data transmission)

Increase the number of dedicated wires, cables, shielded cables.. Weight, cost .. and reliability (connectors). Use the DC PL as a physical support for the transmission

COST 286 312-13 Dec. 2005

Transfer function PTx→PRX(f) Propagation on interconnected multiwire transmission lines Propagation model (Theory/experiments)

Impulsive noise characteristics Measurements→Noise model

Optimization of the modulation scheme (Telecom. aspects) EM Propagation model + noise model

+ simulation of the link (channel coding, .)

Radiated emission (EMC aspects)

Outline

COST 286 412-13 Dec. 2005

Transfer Function Indoor

Within a room “Simple” network architecture. Main variable: loads

connected to the PL. Propagation model: 2-3 wire line + distributed/random

loads (not necessary needed) Measurements: easy (not too many variables)

Inside a building (between different rooms) “Complicated” network architecture, known (new

buildings) or unknown Combine model + measurements

COST 286 512-13 Dec. 2005

In-VehicleComplicated geometry of the cable harness

Complexity >> indoor Extensive measurements : time consuming +

difficulty to have access points Propagation modeling is desirable for a statistical

analysis Elaborate a statistical channel model Extract the channel properties, check with results

deduced from few measurements

COST 286 612-13 Dec. 2005

•Conclusion for determining the channel

properties

Indoor: inside a roompresentation of few experimental results +

channel characteristics Indoor (in a building) and in car

presentation of the propagation modelexample of application: in-car

channel characteristics and channel model

Comparison room/vehicle

COST 286 712-13 Dec. 2005

Preliminary comments on the definition of the transfer function

Let us define H(f) as V/E

Comments: ”Impedance mismatching occurs during the measurements and thus leading to incorrect measurement

results” “Trying to measure path loss without knowing the

impedance at the emission port is non cense”..Suggestion: “Insert a wideband impedance matching..”

OK BUT with such a definition of H(f), the “real word” is modeled. Why?

NetworkE

R (50

R (50) V

COST 286 812-13 Dec. 2005

For LV/MV, the structure of the network does not change and the loads are more or less constant. Passive “equalizer” to match impedances (adapter – line): Enhancement of the performances!

We will see later the architecture of a car harness! A lot of time-varying loads !

An adaptive time varying matching device would be necessary !

Practically: choose a constant value for the input/output impedance of the modem. On the order of the average characteristic impedance of the line (for example 60

COST 286 912-13 Dec. 2005

Taking the terminal loads into account, one can expect that the input impedance of the network will be smaller (few Ohms – 100 Ohms)

Usual impedance of commercially available adapter? Have a look on the data sheet: usually nothing concerning the RF part

It is TRUE that H(f) does NOT correspond to the path loss of the network, alone, BUT to the TRANSFER between the transmitter and the receiver in presence of the network

NetworkE

R (50

R (50) V

COST 286 1012-13 Dec. 2005

What is the physical meaning of H(f) = Vr/Ve? Why not measuring S21?

If ZL is matched to the transmission line between ZL and network output: a2 = 0.

S21 = b2/a1

Definition of the injected power : Power delivered by the source on a matched impedance (a1)

Applying this definition leads to (If Z0 = Zl = R0)

S21 = 2 H(f), whatever R0. Calculating H(f) equivalent to S21 (factor 2)!

Ve

V2V1

a1 a2

b2b1

ZL=50

Z0=50

Vr

COST 286 1112-13 Dec. 2005

Additional comments

Other obvious interpretation of S21 (or H(f)) If Z0 = Zl = R0

Ve

V2V1

a1 a2

b2b1

ZL=50

Z0=50

Vr

22

2 2 0221 2 2

0

/4

/ 4r

i

V RV PS

E E R P

If the source is any generator:Pi corresponds to selected power one can read on the

screen of the generator !

COST 286 1212-13 Dec. 2005

Conclusion The Tx adapter, the line, the Rx adapter .. are considered as a

whole. The transfer function or S21 does NOT correspond to path loss BUT to what happens in a practical case.

If needed, for indoor or in-vehicle PLC, the “intrinsic” path loss: combining the various S parameters BUT still depending on the terminal load

S21 for any load configuration can be deduced from the S50 matrix

Software “equalization” on the data to cope with the frequency selectivity of the PLC channel

In the following, transfer function characterized for an impedance of 50 presented by the modem (same as network analyzer)

For optimizing the modulation scheme, “path loss” is not needed. (only related to average SNR). Channel impulse response !

COST 286 1312-13 Dec. 2005

Transfer function inside a roomTransfer function: ratio between Vout/Vi, (complex number, f(frequency))Various loads are connected at points Pi

COST 286 1412-13 Dec. 2005

Transfer function inside a roomFrequency domain H(f) – Amplitude and phase

COST 286 1512-13 Dec. 2005

Useful statistical parameters

Coherence bandwidth Bc() Absolute value of the autocorrelation of H(f) Bc: frequency shift to get a given value of Typical example: =0.7 or 0.9 →Bc(0.7 or 0.9)

Within Bc, H(f) does not vary appreciably

If transmitted bandwidth<<Bc, flat channel, no signal distortion

Indoor inside a room: Bc=few MHz

COST 286 1612-13 Dec. 2005

Channel characteristics in time domain: Channel impulse response (Multiple reflections ↔ Multipath propagation)

Power delay profile

Mean delay:1

1 n

m i iit

PP

Delay spread

1

22 2

1

n

i i mi

rmst

P

P

Maximum excess delay

COST 286 1712-13 Dec. 2005

If the duration of 1 bit (or symbol) <<, multiple reflections “of the same bit or symbol” arrive nearly at the same time.

No “mixing” of the successive bits: No Inter Symbol Interference (No ISI)

Application to PLC: Usually OFDM modulation scheme → send successive frames.

Avoid interference between frames→ Guard interval between frames >

COST 286 1812-13 Dec. 2005

Impulse response

rms delay spread < 0.2s for a probability <10-3

COST 286 1912-13 Dec. 2005

Transfer function for more complex networks Theoretical modeling of the propagation

Multiple interconnected transmission lines“user-friendly” software tool is needed

Possibility to easy change part of the network configuration

Model based on the “topological” approach proposed by Baum, Liu, Tesche (“BLT” eq.) and developed by ONERA (code Cripte)

COST 286 2012-13 Dec. 2005

Channel transfer function : Deterministic Model, cont.

The harness is divided into a succession of uniform multi conductor (N) transmission lines (N “Tubes”). Along each tube, waves W, combining current and voltages are defined by (matrix form):

Relation between the waves at the ends of the tube ( length l)

Ws : source terms at the end of the tube, propagation constant

Compact form considering all tubes: [W(l)] = [W(0)] + [Ws]

W(z)=V(z)+Zc I(z)

W(l) = W(0) +Ws

COST 286 2112-13 Dec. 2005

Channel transfer function : Deterministic Model, cont. Connection between tubes: junctions. At each junction (including at the ends of the harness), a scattering

matrix S relates incoming and outgoing waves:

[W(0)] = [S] [W(l)]

Combining the various equations leads to:

( [I] - [S] [W(0)] = [S] [Ws]

[I] : identity matrix

Inversion of [I] - [S] [determination of [W(0)] and thus V and I at the ends of each tube.

Advantage: high flexibility for modifying the network architecture, the load impedances ..

COST 286 2212-13 Dec. 2005

Application to in-vehicle PLC

Measurement with a network analyzer (S21), inserting a coupling device

COST 286 2312-13 Dec. 2005

Coupling device

5 Ω 5 Ω 1 MΩ

2 nF

5 Ω 5 Ω 1 MΩ

2 nF140Ω

1 : 1

VNA Port 150 Ω

-10 dB

Filter cut off frequency : 500 kHz

Z seen from the network: about 50 Ohm . Check by measuring S11 up to 40 MHz.

Z seen from the VNA: 20 – 150 Ohm (depending Z network)

COST 286 2412-13 Dec. 2005

Path classification Preliminary measurements: different behavior of H in 2 cases:

Tx Rx

No branching on DC line between Tx and Rx: called “direct path”

Tx Rx

Branching between Tx and Rx: called “indirect path”

COST 286 2512-13 Dec. 2005

Experimental analysis on a vehicle

Computer

Engine computer __ : 12 V__ : ground

+ AB

cigar lighter

Power plug 12 V

CD

E

F

Engine Passenger cell boot(trunk)

« Direct » paths:

A B: 6m

D E: 2m

Indirect paths:

A C

A E

A F

COST 286 2612-13 Dec. 2005

Experimental approach : long direct path (AB, 6m)

Transfer functions

H1 H2 H3 H4

K1 OFF X X

ON X X

K2 OFF X X

ON X X

-0.5 dB / MHz

Port 1 50Ω

Port 2 50Ω

12 V

K1 K2Bundle x, car y wire B100

Computer boot (PSF2)AB

COST 286 2712-13 Dec. 2005

Direct path: Short (AB, 2m) / long (DE, 6m)

•Path n°1 – long ≈6 m• Path n°2 – short ≈1 m

Bc0.9 ≈ 2 MHz

Computer trunk (PSF2)

Port 1 50Ω

Port 2 50Ω

12 V

Harness xxx - AB

Port 1 50Ω

Port 2 50Ω

12 V

harness xxx - DE interior light at 40 cm from port 2

AB

ED

S21 ≥ -30 dB

Δf = 43 kHz

COST 286 2812-13 Dec. 2005

Indirect paths

Bc0.9 ≈ 600 kHz

S21 ≤ -30 dB

Computercoffre (PSF2)

n°1 – A C •n°2 – A E•n°3 – A F

Port 1 50Ω

Port 2 50Ω

12 V

Network car xxx

Cigar lighter

Port 1 50Ω

Prise 12V

C

Port 1 50Ω

BSIA

E

F

COST 286 2912-13 Dec. 2005

Influence of the load configuration

indirect path n°3 – A F between cigar lighter and the computer in the boot (trunk?)

Measurement while driving + activating electric and electronic equipment

]),([.]),([

),(),(

22

*

jfHifH

jfHifHf ij

Port 2 50Ω

12 V

Faisceau xxx – fil B100Allume cigare

Calculateur coffre (PSF2)Port 1

50ΩF A

Correlation coefficient between successive values of the transfer function

COST 286 3012-13 Dec. 2005

Propagation modeling

Z2

3 fils 100 cm

3 fils 30 cm

16 fils 10 cm

1 fils 1 mm

16 fils 50 cm

20 fils 80 cm

10 fils 50 cm

10 fils 30 cm

3 fils 50 cm

3 fils 50 cm

1 fils 10 cm

1 fils 10 cm

2 fils 50 cm

Z1

Batt.

Z6

D1F

1 fils 1 mm

1 fils 10 cm

1 fils 40 cm

3 fils 15 cm

14 fils 50 cm

14 fils 100 cm

20fils 50 cm

11fils 100 cm

16 fils 50 cm

30 fils 50 cm

Z15

Z33 fils 50 cm

1 fils 10cm

11 fils 100 cm

Z11

Z14

D3

30 fils 50 cm

20 fils 150 cm

1 fils 10 cmM

M

5 fils 10 cmM

5 fils 40 cm 10 fils

50 cm

3 fils 50 cm

Z185 fils 100 cm

Z16

3 fils 10 cm

Engine

Dashboard

Passenger cell5 fils 10 cm

Z4

M

Z9

10 fils 100 cm

Z8

1 fils 40 cm

D2

Z7

CC

Z10

Z17

Z12

Z13Z5

1 fils 25 cm

11 fils 100 cm

5 fils 100 cm

9 fils 50 cm

10 fils 50 cm

5 fils 1 m

15 fils 30 cm

Z35 fils 50 cm

M

M

3 fils 50 cm

CC

D1 D3 : 5.75 m

D2 D3 : 7.55 m

Total length of the cables = 205 m

COST 286 3112-13 Dec. 2005

50 load combinations

Example for 3 load config.

D3 50Ω

Config. N°2 (5.75 m)

D1 50Ω

Example: S21between D1 and D3 (about 6m)

S21 > -30 dB

Bc0.9 ≈ 700 kHz

COST 286 3212-13 Dec. 2005

Another example

Bc0.9 ≈ 600 kHzS21 ≤ -30 dB

COST 286 3312-13 Dec. 2005

Statistical results deduced from 50 configurations

Statistical parameters

Experiments Deterministic model

Direct paths

Bc0.9 / Hz * 2 MHz 1.5 MHz

Rms Delay Spread / nS *

60 nS 61 nS

Indirect

paths

Bc0.9 / Hz * 700 kHz 780 kHz

Rms Delay Spread / nS *

84 nS 108 nS

COST 286 3412-13 Dec. 2005

Distribution of the amplitude of H(f) around its mean value versus freq.

Try to fit exp distribution with known analytical distribution

COST 286 3512-13 Dec. 2005

Conclusion on transfer function : indoor or in-

vehicle Use the average statistical values of the channel parameter (transfer

function, Bc, delay spread) for a first optimization of the transmission scheme

Build a statistical channel model (knowing the probability distribution of the discretized channel impulse response from meas. + deterministic modeling)

Insert this model in a software simulating the communication link to deduce system performance ..but also in presence of noise !

Next step: Noise characterization

COST 286 3612-13 Dec. 2005

Noise in indoor environment

COST 286 3712-13 Dec. 2005

Power Spectrum Density, Narrow band noise measured on indoor power lines

Indoor network connected to an overhead outdoor power line

Indoor network connected to a buried power line

Broadcast transmitters

Conclusion: Useful transmissionbandwidth above 500 kHz

COST 286 3812-13 Dec. 2005

Impulsive Noise : conducted emissions due to electrical devices connected to the network.

Single transient: Damped sinusoid

Burst: Succession of heavy damped

sinusoids

Measurements in a house during 40 h 2 classes of pulses (on 1644 pulses) : single transient and burst

I. Impulsive Noise Classification / Noise model

COST 286 3912-13 Dec. 2005

I. Impulsive Noise Classification / Noise model

(b) Burst Model

(a) Single transient model

Parameters of single transient :

- peak amplitude - pseudo frequency f0

=1/T0

- damping factor- duration- InterArrival Time IAT

COST 286 4012-13 Dec. 2005

I. Impulsive Noise Classification / Noise characterization

1644 pulsesfo<500

kHz0.5 MHz < fo <

3MHzfo>3 MHz

Single Transient

Class 1 Class 2Pb = 48 % Pb = 20 %

Burst Class 3 Class 4 Class 5Pb = 3 % Pb = 11 % Pb = 18 %

Bandwidth of

PLT system

1.Classification in time and frequency domain :

5 classes are introduced, depending on the pseudo frequency f0

Pb: Probability of occurence

COST 286 4112-13 Dec. 2005

2. Statistical analysis: Noise Parameters are approximated by well-known analytical distributions to build a noise model

Pseudo Frequency :

Weibull distributionbaxb eabxxf 1)(

I. Impulsive Noise Classification / Noise characterization

COST 286 4212-13 Dec. 2005

2. Statistical analysis: Careful examination of long bursts Pseudo-frequency of the elementary pulse varies with time(calculated with a running time window)

The pseudo-frequency distribution around its mean value follows a normal distribution :

)²

)²(

2

1exp(

2

1)(

s

µx

sxf

and s2 are the meanand the variance of x Agreement: =1, s=0.17

COST 286 4312-13 Dec. 2005

I. Impulsive Noise Classification / Model validation

Model validation : Comparison of the spectral densities of measured pulses and generated pulses :

Good agreement between measurement and model !

COST 286 4412-13 Dec. 2005

Noise on DC line inside a car

COST 286 4512-13 Dec. 2005

System parameters : mobile platform Sampling rate = 100 MHz (Sampling period : 10ns) Observation window : 650 µs Peak limiting 15V Trigger : 300 mV

Noise Model : Experimental setting

Noise acquisition

acquisitionIAT

PC

CH AExt trigger

Port //Trig out

coupler

COST 286 4612-13 Dec. 2005

Typical pulses

Single transient, burst and “atypical pulse”

COST 286 4712-13 Dec. 2005

3 MHz fo < 7 MHz 7 MHz fo < 15 MHz 30 MHz fo < 35 MHz

Single pulse

Class 1 Class 2 Class 3

67.2 % 7.2 % 4.9 %

Burst Class 4 Class 5 Class 6

19.7 % 0.9 % 0.1 %

Objective : For each class, a mathematical function is found to fit the distribution of the characteristic parameter of the pulse

The same approach is followed to model all classes and the others statistical distributions of the pulse characteristics.

Noise Model : Statistical Analysis

COST 286 4812-13 Dec. 2005

Classification of the pulses : Frequency/amplitude and Frequency/duration

COST 286 4912-13 Dec. 2005

Amplitude and Pseudo frequency distribution of bursts during cruising phase

COST 286 5012-13 Dec. 2005

Cumulative probability distribution of IAT normalized in OFDM frames (6.4s in our application . see later)

COST 286 5112-13 Dec. 2005

Time or Frequency domain : The Power Spectral Densities are calculated from measurement and compared with the generated model.

Measurement Model

Noise Model : Stochastic Model

COST 286 5212-13 Dec. 2005

Noise model

From the knowledge of known distribution functions fitting exp. results Noise model . Generation of single transients and bursts satisfying the

same probability in terms of amplitude, IAT, frequency content..

Combine statistical (noise + propagation) model: statistical channel model

Performances of the link and optimization of the modulation scheme

COST 286 5312-13 Dec. 2005

Simulation of the communication link

Frequency selective channel: few frequency bands are strongly attenuated (multiple reflections)

Wide band communication leads to important distortion of the signal, interference inter symbol, ..

Rather than using a given large bandwidth: divide them into a number (64 or 128 or 256) of equivalent parallel channels, each one with a small bandwidth

In each equivalent channel, no frequency selectivity. Flat channel

COST 286 5412-13 Dec. 2005

N sub channels : N sub carriers OFDM

fk

Bf

f f

(b) (a)

(a)Spectrum of a sub carrier (b) Spectrum of an OFDM signal OFDM

N oscillators? not realistic. Use properties of FFT.

Important data: Statistical behavior of H(f) If few frequency bands are strongly attenuated: do not use them!Maximize and optimize bit rate on channels having a good SNR!

Periodically test the channel, detect change in the channel state (variation of H(f) when the loads vary), new channel equalization

COST 286 5512-13 Dec. 2005

Principle of multicarrier-based transmission : Transmission on N orthogonal subcarriers owing to an IFFT/FFT.

TransferFunction (H)

Noise

Analog/digital

Interface

Channel decoding

ChannelCoding

Digital/analog

Interface + Filter CHANNEL

RECEIVER

FFT

Prefixe

removal

S

/

P

EQUALIZER

P

/

S

IFFTPrefix

Add.

P/S

S/P

EMITTER

COST 286 5612-13 Dec. 2005

2. Example of simple channel coding2. Example of simple channel codingReed-Solomon code : RS(N,K) Word of K effective symbols Word of N symb. by adding redundancy (N-K symbols) ADSL normalization: Symbol: byte and N = 255

This code can correct up t = (N-K)/2 bytes. if K=239, t = 8 bytes.

Important data: duration of a pulse (statistical approach)

word of K bytes

Reed-Solomon

code code word of 255 bytes

bytes

COST 286 5712-13 Dec. 2005

Interleaving

Long burst: RS code cannot correct errors. Is it possible to avoid a long disturbance on the same word?

Interleaving: An interleaving matrix of 256 rows by D columns, D interleaving depth, varying from 2 to 64.

Bytes introduced in lines and sent in columns The disturbance is “distributed” on successive words and RS coding

may thus be efficient The interleaving depth depends on the statistics of transient duration

Any other problem?

COST 286 5812-13 Dec. 2005

YES What happens when two successive pulses (burst or single

transient) occur?

Other important parameter: statistics of the IAT

When 2 pulses occur during the time of an interleaved matrix, these two pulses disturb the same matrix and, may be, the RS code will no more efficient. (Problem when the time interval between two successive transients is small)

Other signal processing techniques are needed

COST 286 5912-13 Dec. 2005

Optimisation in presence of impulsive noise (Indoor)

Contribution of channel coding and noise processing on the Bit Error Rate (BER), assuming for all pulses a pseudo frequency f0 within the signal bandwidth and a PSD of -50 dBm/Hz

Pb (BER<10-3) = 77% if D=16

Pb (BER<10-3) = 96 % if D=64

Choice of D depends on acceptable BER

BER

Cumulative probability distributionof the mean BER for three differentvalues of the interleaving depth D

COST 286 6012-13 Dec. 2005

PLC emission

COST 286 6112-13 Dec. 2005

Testing room description

Computer

Receiver

S1

S2

SocketsMagnetic loop

Balun

C.W. source

Data bus Three wires bundle 23 m length

Switch220 V – 50 Hz supply line

Plaster walls

S3

COST 286 6212-13 Dec. 2005

Radiated field but normalized to a given injection. Ratio between the differential voltage at the PL input and the

electric field measured at a given distance (1m, 3m). At low frequency, H is measured. Convert H into E considering the wave impedance in free space (definition, only)

Other possibility: Normalize to the maximum power which could be injected in the line (matched impedances). Expressed in dBm/Hz

Signal generator

Coupling device

Coupling device

50Ω

Spectrum Analyser

Active probe

PLC Line

COST 286 6312-13 Dec. 2005

Signal generator

Coupling Coupling 50Ω

Spectrum Analyzer

LOOP Antenna

Preliminary measurement of the “ambiant noise”

COST 286 6412-13 Dec. 2005

Example: (same differential voltage) Car/Indoor

COST 286 6512-13 Dec. 2005

Field variations in the room

100 1012 3 4 5 6 7 8 9 2

Fréquence MHz

0.0

20.0

40.0

60.0

Cham

p H dB

uA/m

Mesure du champ H

dBµA / m

1 MHz 30 MHz

D = 10 cm

D = 20 cm

D = 3 m

D: distance of the antenna from the wallMagnetic field

10 MHz

COST 286 6612-13 Dec. 2005

Standards?

Another issue!