Research in PRML for High Speed Optical Communications

Research in PRML for High Speed Optical Communications Romain Rousseau Tutor: Zaki Ahmed Master of Science (MSc) Communication Engineering and Signal Processing Department of Communication and Electronic Engineering 1 University of Plymouth, Faculty of Technology September 2007

2 Picture taken at Science Museum, London (UK)

Extracted from: Introduction to Fibre Optics and Optical Networks (2007) Available http: //www. engineeringlab. com/fiberoptics. html#demo 3

Outline 1. Background a) b) 2. Project a) b) 3. Realisation Results Conclusion a) 4 ISI PRML technology Future works

1. a) ISI from Fibre optics 5

1. a) Inter. Symbol Interference signal pulse of a bit Respectively amplitude of electric and magnetic fields Angular frequency Propagation constant 6 Normalised Frequency Refractive Index Difference

1. a) Inter. Symbol Interference Birefringence-Polarization Mode Dispersion effects From a polarisation point of view : Fast axis: Mode propagation along the xaxis has the smaller effective mode index Slow axis: Mode propagation along the yaxis has the smaller effective mode index Phase birefringence : Modal birefringence 7 Beat length

1. a) Inter. Symbol Interference Total Dispersion or Chromatic Dispersion velocity of the light becomes wavelength dependent 8

1. a) Inter. Symbol Interference Waveguide Dispersion 9 Material dispersion Total dispersion

1. a) Inter. Symbol Interference Material Dispersion Fibre will have material dispersion if Material dispersion Is the core refractive index 10

1. a) Inter. Symbol Interference Waveguide Dispersion “A waveguide is a device such as coaxial cable or glass fibre, designed to confine and direct the propagation of electromagnetic waves. In optical fibers the confinement is achieved by having a region with a larger refractive index” Österberg Ulf L fibre will have waveguide dispersion if waveguide dispersion 11 is constant propagation

1. a) Inter. Symbol Interference Total Dispersion total dispersion parameter rms pulse broadening data rate limitation 12

1. a) Inter. Symbol Interference Line Coding Line coding consists of modulating binary data in order to map symbols into specific waveform, before sending them to the receiver. 3 points that need to be opposed when you want to choose your line coding: i. 13 The coded spectrum goes to zero at the frequency approaches 0 (DC energy can not be transmitted) ii. A clock can be recovered from the coded data stream (necessary for synchronous systems and detection) iii. They can detect errors or even correct errors

1. a) Inter. Symbol Interference Line Coding Different modulations for data sequence 010111001 14

1. a) Inter. Symbol Interference Line Coding Non Return to Zero (NRZ) format also known as NRZ-OOK (On Off Keying). This modulation is composed of two “sub-modulation” • Unipolar NRZ: values interpreted by the photodiode is +v for a logical ‘ 1’ and 0 v for a logical ‘ 0’. • Polar NRZ: values interpreted by the photodiode is +v for a logical ‘ 1’ and –v for a logical ‘ 0’. NRZ format needs all the bandwidth to transmit information Can not be used for synchronous systems because 15

1. a) Inter. Symbol Interference Line Coding Return to Zero (RZ) format is composed of two “sub-modulation” • Unipolar RZ: values interpreted by the photodiode for a logical ‘ 1’ is +v for and 0 v for next In the case of a logical ‘ 0’, current delivered by photodiode is a 0 v for a full period of • Polar RZ: values interpreted by the photodiode for a logical ‘ 1’ is the same than Unipolar RZ logical ‘ 1’. For a logical ‘ 0’, voltage is –v for a period of and then 0 v for next 16 Both RZ format require 2 times bandwidth than a NRZ format increases lifetime of laser. RZ Polar format enables to extract timing for synchronous-systems.

1. a) Inter. Symbol Interference Line Coding Summary 17

1. b) PRML Technology PRML is a combination of a PR target (created with an equalizer) associated to a Maximum Likelihood Sequence Decoder (MLSD). 18

1. b) PRML Technology In 1962 Adam Lender presents PR targets in his paper “The Duobinary Technique for High-Speed Data Transmission” [5] It was first introduced in magnetic recording system by Kobayashi and Tang, two researchers of IBM, in 1970 Recently, many scientists have proved that duobinary signalling allows high bandwidth frequency and is a very efficient way to counter dispersion effects of fibre optics PR targets require less power and complexity than other Multi level schemes. The aim of duobinary signalling is to accept and to transform ISI into a specific pattern. Then the receiver is built depending on this pattern and it makes use of the nature of the ISI instead of cancelling it. 19

1. b) PRML Technology In 1966 Kretzmer shorted Partial response channels in different class specifying the “partial response polynomial” H(D), where D is a delay For a minimum bandwidth system must have in H(D), but for k>>0 number of levels is increase so error performance is degraded factor reduces low-frequency component in spectrum needed 20

1. b) PRML Technology The characteristic polynomial of widely used channel is a composition of This partial response is called Modified Duobinary or PR Class IV (PR 4) For channel models are usually referred to as “extended Class-4” models and denoted by 21

1. b) PRML Technology 22

1. b) PRML Technology Preset Equalizer Basically an equalizer analyses channel distortions by receiving a specific impulse response. Differences between original pulse and received pulse by equaliser correspond to the channel deformations By adjusting parameters {Cn} of equalizer we can correct attenuations 23

1. b) PRML Technology Preset Equalizer Output of equalizer Ch(t) is channel’s pulse response Example for a EPR 4 target (sequence number 1, 2, 1) and =T 24

1. b) PRML Technology Maximum Likelihood Sequence Decoder Recursive convolutional codes which have both feed forward and feedback nodes. Non-recursive convolutional codes which have only feed forward nodes PR 4 system is non-recursive convolutional code. Convolutional codes are codes with memory which means that the output depends on previous inputs. The code rate of convolutional codes is a ratio between numbers of input symbols k over numbers of output symbols n. PR code rate considered as 1 -rate. The length of input block K is often called constraint length. 25

1. b) PRML Technology Maximum Likelihood Sequence Decoder Trellis Creation: 26

1. b) PRML Technology Most famous MLSD: Viterbi Algorithm Decoding Process: 1. Estimate the new path metric, by adding the path metric with the survivor path from state k-1 respectively for each states available on the trellis. 2. For each path, select the survivor path with the best metric. 3. Store the survivor path and metric for each vertex at time k, increment k and repeat the three steps above until end of data. 4. Choose the minimum value from all possible vertexes and trace back from end to first state and output the bit information using the path matrix saved. 27

2. a) Investigations 28

2. a) My Realisation Single mode fibre link 1 st block: Attenuation generated with a 7 th order elliptic Low Pass filter using a normalised cutoff frequency of 0. 21. The ripple in the passband is 0. 3 d. B and 50 d. B for stopband. 2 nd block: Preset equalizer with EPR 4 and E 3 PR 4 targets. 3 rd block: AWGN noise. It gives details about strength of system especially the decoder part. 4 th block: Detector which consists of a Soft Input Hard Output (SIHO) Viterbi Algorithm 29

2. a) My Realisation 1 st Block: Low Pass Filter On the left, this is the pulse used to estimate the coefficient for equalizer On the right is a typical attenuation for single mode fibre 30

2. a) My Realisation 2 nd Block: Equalizer EPR 4 E 3 PR 4 31

2. a) My Realisation 2 nd Block: Equalizer 32

2. a) My Realisation 3 rd Block: AWGN noise is to observe the strength of our system especially the decoder part. this noise can be assumed as a noise created by photodiode when converting optical impulse into electric voltage. We know that the best photodiode systems to convert data from optical to electric are avalanche photodiode, and they produce a significant noise 33

2. a) My Realisation 4 th Block: Viterbi Algorithm EPR 4 E 3 PR 4 34

2. a) My Realisation 4 th Block: Viterbi Algorithm EPR 4 E 3 PR 4 35

2. a) My Realisation 4 th Block: Viterbi Algorithm E 3 PR 4 36

2. b) Results 37

2. b) Results Number of taps for equalizer EPR 4 = 11 taps 38 E 3 PR 4=17 taps

2. b) Results Number of taps for equalizer EPR 4 = 11 taps 39 E 3 PR 4=17 taps

2. b) Results Result of equalization NRZ-OOK E 3 PR 4= 17 taps 40 EPR 4=11 taps

2. b) Results Strength of PR target 41

2. b) Results CPU time/delay created by Viterbi Algorithm The processor used was an Intel Centrino Duo T 2500@2 GHz 42

2. b) Results Global view 43

2. b) Results Ideal Coefficients {Cn} 44 Issues

3. a) Future works • MLSD is not optimised and deeper searches must be done. Bonek, et al. developed a method to simplify 16 -state trellis to 8 states by considering a butterfly architecture. It reduces 47% wiring and has generalised the algorithm for n-states trellis. Farkas-Weiss and Kalet (1989) “Simulation of a Trellis Code with Duobinary Signaling” IEEE The Sixteenth Conference of Electrical & Electronics Engineers in Israel Pages: 1 – 4 Ozarow and Calderbank have also proposed appropriate trellis to duobinary channel resulting in a rate ½ code. The code rate is also a part of MLSD that can be investigated for next projects Weiss S. F. and Russell A. (? ? ), “Simulation of a trellis code with Duobinary Signaling”, Tel-Aviv University • Code is essential PR associated to serial turbo scheme (A turbo coding much interesting for data storage) and LDPC codes obtained 5 d. B gain compare to an uncoded PRML channel. (Song et. al. ) Song H. , Liu J. , Kumar V. and Kurtas E. (2001) “Iterative soft decoded partial response channels for hybridmagnetooptical recording”, Magnetics, IEEE Transactions on, Volume 37, Issue 2 Page(s): 676 – 681 • PR target GPRML provides 4. 5 d. B gain compare to EPR 4 ML according to Sun’s paper Sun D. , Xotta A. , and Abidi. A. A. , (2005) “A 1 GHz CMOS Analog Front-End for a Generalized PRML Read Channel. ” IEEE Journal of Solid-State Circuits, vol. 40, no. 11, pp. 2275 -2285. 45

Thanks you for your attention Questions ? ? ? • Elijah Wu (2006), “Measuring Chromatic Dispersion of Single-Mode Optical Fibres using White Light Interferometry “, University of Auckland, Ms. C Thesis • Senior J. (1992), Optical Fiber Communication: Principles and Practice Prentice Hall; 2 nd edition • Reeves, 2007, Optical Fibres, University of Plymouth Lecture • Lender A. (1963) “The Duobinary Technique for High-Speed Data Transmission” IEEE Transactions on Communication and Electronics, Vol. 82, pp. 214 -218 • Kobayashi H. and Tang D. (1970) "Application of Partial-Response Channel Coding to Magnetic Recording Systems", IBM Journal. of Res. & Dev. , vol. 14, no. 4, p. 368 • Bosco G. , Carena A. , Curri V. and Poggiolini P. (2006) “Best Optical Filtering for Duobinary Transmission from Optical Communication Theory and Techniques“ Springer US • Lyubomirsky I. (2006) “Coherent detection for optical duobinary communication systems”, Photonics Technology Letters, IEEE, Vol 18, Issue 7 pp. 868 – 870 • Kretzmer E. R. (1966) "Generalization of a Technique for Binary Data Communication, “ IEEE Transactions on Communication Technology, February 1966, pp. 67 -68 • Kabal, P. and Pasupathy S. (1975) "Partial-Response Signaling", IEEE Transactions On Communications, vol. Com-23, No. 9, pp. 921 -934. 46

2. b) Issues 47

2. b) Issues 48

2. b) Issues conclusion 49