Transcript Optical Wireless Communication using Digital Pulse Interval

Signal Detection and Adaptive Equalization
Using Discrete Wavelet Transform - Artificial
Neural Network for OOK Indoor Optical
Wireless Links
Z. Ghassemlooy, S Rajbhandari and M Angelova
School of Computing, Engineering & Information Sciences, University of
Northumbria,
Newcastle upon Tyne, UK
http://soe.unn.ac.uk/ocr/
ICEE08, Tehran, Iran
Outline





Optical Wireless – Key issues
Digital Signal Detection
Equalization
Wavelet ANN Based Receiver
Results and Conclusion
ICEE08, Tehran, Iran
Indoor Optical Wireless Links
 The key issues:
- Eye safety
- shift from 900 nm to 1550 nm - eye retina is less
sensitive to optical radiation
- power efficient modulation techniques
- Mobility and blocking
- diffuse configuration instead of line of sight, but at
cost of:
- reduced data rate
- increased path loss
- multipath induced inter-symbol-interference (ISI)
ICEE08, Tehran, Iran
Digital Signal Detection - The Classical
Approach
Input
bits, ai Transmitter
filter
p(t)
Receiver
Channel
Transmitter
X(t)
Multipath
channel
h(t)
2Pavg
S(t)
Z(t)
n(t)
Unit energy
filter r(t)
(matched to
p(t))
Zj
sample
R
 The discrete-time impulse response of the cascaded
system
c
k
 p (t )  h(t )  r (t )
optical channel (ceiling bounce)
h (t ) 
6(0.1D
rms
(t  0.1D
t  kT
b
)6
rms
ICEE08, Tehran, Iran
)7
u (t )
Output
Bits, âi
Digital Signal Detection - The Classical
Approach
 OOK - the average probability of error:
m
2
1
P

 
e, bit, OOK
i
m
2 i 1
the probability of error for the penultimate bit in ai:
.
  y 

opt 
Q i
 if ai  1
 
0
.
5
  (0.5 N 0 ) 

 
i
  
y 
  opt i 
 if ai  0
Q
0
.
5
  (0.5 N 0 ) 
ICEE08, Tehran, Iran
where opt is the
optimum
threshold
level, set to the midway
value of RPave (Tb)0.5
Digital Signal Detection - The Classical
Approach
 Matched filter is difficult to realized when channel is time
varying.
 Maximising the SNR based on the assumption that noise
statistics is known.
 SNR is sensitive to the sampling instants.
- In non-dispersive channel, the optimum sampling point
is at the end of each bit period.
- In dispersive channel, the optimum sampling point
changes as the severity of ISI changes.
ICEE08, Tehran, Iran
Digital Signal Detection - The Classical
Approach
 For higher values of normalized delay spread (> 0.52)
- bit error rate cannot be improved simply by increasing the
transmitter power
 To mitigate the ISI, optimum solutions are:
- Maximum likelihood sequence detector
- Equalizers1-3 - A practical solution
(i) Inverse filter problem
- The frequency response of the equalizing filter is the
inverse of the channel response.
- Adaptive equalization is preferred if the channel
conditions are not known in advance.
- Two classes : linear and decision feedback equalizer.
(ii) Classification problem
1- J. M. Kahn and J. R. Barry, Proceedings of IEEE, 85 (2), pp. 265-298, 1997
2- G. W. Marsh and J. M. Kahn, IEEE Photonics Technology letters, 6(10), pp. 1268 - 1270, 1994
3- D. C. Lee and J. M. Kahn, IEEE Transaction on Communication, 47(2), pp. 255-260, 1999
ICEE08, Tehran, Iran
Equalization - A Classification Problem
 Dispersion induced by channel is nonlinear in nature
 Received signal at each sampling instant may be
considered as a nonlinear function of the past
values of the transmitted symbols
 Channel is non-stationary
- overall channel response becomes a nonlinear
dynamic mapping
ICEE08, Tehran, Iran
Equalization: A Classification Problem
 Classification capability of FIR filter equalizer is limited to a linear
decision boundary (a non-optimum classification1)
 FIR bases equalizers suffer from severe performance degradation in
time varying and non-linear channels2
 The optimum strategy - to have a nonlinear decision boundary for
classification
- ANN
- with capability to form complex nonlinear decision regions
- In fact both the linear and DFE are a class of ANN3 .
- Wavelet4
1- L.Hanzo, et al, Adaptive wireless transceivers: Wiley-IEEE Press, 2002, pp. 299-383.
2- C. Ching-Haur, et al , Signal Processing,vol. 47, no. 2, pp. 145 - 158 1995.
3- S. Haykin, Communications Magazine, IEEE , vol.38, no.12, pp. 106-114, Dec. 2000
4- D. Cariolaro et al, IEEE Intern. Conf. on Communications, New York, NY, USA, pp. 74-78, 2000.
ICEE08, Tehran, Iran
Receiver - Classification Based
Optical Optical
Signal Receiver
Feature
Extraction
Pattern
Classification
Wavelet
Transform
Neural
Network
PostProcessing
Modular based receiver:


Feature extraction (wavelet transform) - for efficient classification
Pattern classification (ANN).
WT-ANN based receiver outperforms the traditional equalizers1.
1- R. J. Dickenson and Z. Ghassemlooy, International Journal of Communications Systems, Vol. 18, No. 3, pp. 247-266, 2005.
ICEE08, Tehran, Iran
Feature Extraction Tools
Time-Frequencies Mapping
Fourier
Transform
No timefrequency
localization
Short-time Fourier
Transform
Fixed time-frequency
resolution:
Uncertainty problem
ICEE08, Tehran, Iran
Wavelet
Transform
No resolution
problem: ultimate
transform
CWT vs. DWT
 CWT
- Infinite scale
- but with redundant coefficients
 DWT
- no redundancy as in CWT
- easier to implement using filter banks (high pass and low
pass)
- reduced computational time
- possibility signal denoising by thresholding the wavelet
coefficient
ICEE08, Tehran, Iran
Discrete Wavelet Transform
Filtering
h[n]
Level 1
DWT
Down- coefficients
sampling
Level 2
DWT
coefficients
cD1
2
cD2
Signal
h[n]
x[n]
g[n]
2
2
cA1
cA2
g[n]
2
 DWT coefficient - obtained by successive filtering and down sampling
cD : yl [ k ]   X ng[ 2k  n]
cA : y h [ k ]   X nh[ 2k  n]
n
n
 Signal is decomposed:
- using high pass h[n] and a low pass g[n] filters
• filters are related to each other and are known as the quadrature mirror filter.
- down sampling by 2
ICEE08, Tehran, Iran
...
WT- ANN Based Receiver Model





8-sample per bit
Signal is decimated into W-bit discrete
sliding window. (i.e. each window contains
a total of 8W-bit discrete samples )
Information content of the window is
changed by one bit
3-level DWT for each window is
determined
DWT coefficients are denoised by:
i) Thresholding : A threshold is set and
‘soft’ or ‘hard’ thresholding are used for detail
coefficients
ii) Discarding coefficients: detail
coefficients are completely discarded


Zj
Z(t)
Tb/n
Feature extractor
& pattern classifier
DWT
Ẑ j Threshold b̂ j
detector
ANN
Bit to decode
3 bit window
Denoised coefficient are applied to ANN
ANN is trained to classify signal into two binary classed based on DWT
coefficients
ICEE08, Tehran, Iran
Denoising Signal using DWT
 Hard thresholding
(k ) HT
 Soft thresholding
0

1
if k  
if k  
(k ) ST  sgn( k )( k  )
The threshold level for universal threshold scheme:
: variance of the wavelet coefficient
 Denoised signal
X d [n]   1 (( X [n])
where -1 is the inverse WT
ICEE08, Tehran, Iran
  2 log N 
16
Simulation Parameters
Parameters
Data rate Rb
155 Mbps
Channel RMS delay spread Drms 10 ns
Value
No. of samples per bit
Mother wavelet
ANN type
No. of neural layers
No. of neurons in 1st layer
No. of neurons in 2nd layer
ANN activation function
ANN training algorithm
8
Discrete Meyer
Feedforward back propagation
2
4
1
log-sigmoid, tan-sigmoid
Scaled conjugate gradient algorithm
ANN training sequence
Minimum error
Minimum gradient
DWT levels
400 bits
1-30
1-30
3
ICEE08, Tehran, Iran
Results – BER for OOK @ 150 Mb/s
 Maximum performance of ~6 dB
compared to linear equalizer.
 Performance depends on the
mother wavelets.
 Discrete Meyer gives the best
performance and Haar the
worst performance among
studied mother wavelet.
Figure: The Performance of OOK at 150Mbps for diffused
channel with Drms of 10ns
ICEE08, Tehran, Iran
Results - BER for OOK @ 150 & 200 Mb/s
0
10
Unequalized 155Mbps
-1
10
ANN(155Mbps, W=3)
-2
ANN(155Mbps, W=5)
BER
10
-3
10
-4
10
-5
10
0
5
10
15
20
25
 The DWT-ANN based
receiver showed a
significant improvement
compared to linear
equalizer
 SNR gain of ~6 dB at BER
of 10-5 for W = 3
 3-bit window is the optimum
 Reduced complexity
compared to CWT based
receiver without any
degradation in performance
SNR (dB)
Figure: The BER performance of OOK linear and DWTANN base receiver at 155 and 200 Mbps for diffused
channel with Drms of 10ns
ICEE08, Tehran, Iran
Conclusions
 The traditional tool for signal detection and equalization is
inadequate in time-varying non-linear channel.
 Digital signal detection can be reformulated as feature
extraction and pattern classification.
 Both discrete and continuous wavelet transform is used for
feature extraction.
 Artificial Neural Network is trained for classify received signal
into binary classes.
 3-bit window size is adequate for feature extraction.
 Enhance performance compared to the traditional FIR equalizer
( a gain of ~ 6dB at BER of 10-5.
 Reduced complexity using DWT compared to CWT based
receiver with identical perfromance.
Thank you!
Questions?
ICEE08, Tehran, Iran