You are on page 1of 94

University College London

Department of Electronic and Electrical


Engineering

Enhancing the capacity of multimode


fibre links using modal diversity
multiplexing and receiver based
MIMO digital signal processing

By
William Hughes

Degree: Technologies for Broadband Communications MSc


Supervisor: Dr. Benn Thomsen

September 2011
Abstract
Through simulation and experiment, this work demonstrates the feasibility of mode division
multiplexing as a method for increasing the capacity of multimode optical fibre. Receiver DSP
performance was analysed under various simulated channel effects. These simulations showed that
a channel condition number less than 4.2dB was required to provide sufficient performance. The
transmitter and receiver coupling were optimised for a 2x2 COMIMO system. This was achieved
using a 4x1 Multimode Coupler to 1x4 Multimode Coupler arrangement at the transmitter and
receiver, respectively. This provided the preparation required to achieve a state of the art 2x2
polarisation and mode multiplexed COMIMO multimode fibre system.

1
Acknowledgements
I would like to thank my supervisor Dr. Benn Thomsen for the project proposal, his encouragement
and invaluable support during the project. I am also grateful to Joel Carpenter and Dr. Tim Wilkinson
of the CMMPE group of Cambridge University for the use of their test equipment.

I would like to thank my parents for their support throughout my education. Also thanks to my
friend Nicholas Orton and sister Erin Hughes.

2
Contents
1 Introduction .................................................................................................................................... 6
2 Section I: MIMO System Simulation ............................................................................................... 8
2.1 Theory ..................................................................................................................................... 8
2.1.1 Mode Multiplexed Channel Model ................................................................................. 9
2.1.2 Polarisation Multiplexed Channel Model ..................................................................... 10
2.1.3 Channel Equalisation..................................................................................................... 12
2.2 MIMO Simulation Setup........................................................................................................ 17
2.3 Simulation Results ................................................................................................................. 23
2.3.1 CMA Step Size Parameter ............................................................................................. 23
2.3.2 Equaliser Convergence Criteria ..................................................................................... 25
2.3.3 Polarisation Dependent Effects .................................................................................... 26
2.3.4 4x4 MIMO System: Polarisation and Mode Multiplexed .............................................. 30
2.4 Summary: Simulation ............................................................................................................ 34
3 Section II: Lab Test Results ............................................................................................................ 35
3.1 Theory ................................................................................................................................... 35
3.2 1x2 Multimode Coupler Characterisation............................................................................. 39
3.2.1 MMC Characterisation: Test a) ..................................................................................... 42
3.2.2 MMC Characterisation: Test b) ..................................................................................... 44
3.2.3 MMC Characterisation: Test c) ..................................................................................... 45
3.3 Insertion Loss and Modal Diversity Definitions .................................................................... 47
3.4 2x1 – 1x2 Multimode Coupler: 2x2 MIMO System ............................................................... 49
3.5 Optimisation of System Tx/Rx Coupling................................................................................ 53
3.5.1 2x1 – 1x4 Multimode Coupler 2x4 System ................................................................... 53
3.5.2 4x1 – 1x4 Multimode Coupler 2x4 System ................................................................... 57
3.5.3 Free Space Receiver Coupling Method ......................................................................... 60
3.6 Spatial Light Modulator Transmitter Coupling Technique .................................................... 65
3.6.1 1x2 MMC Characterisation using SLM Technique ........................................................ 65
3.6.2 1x4 MMC Characterisation using SLM Launch .............................................................. 68
3.7 Summary: Tx/Rx Coupling Optimisation ............................................................................... 72
4 Conclusion ..................................................................................................................................... 74
5 Further Work................................................................................................................................. 75
6 References .................................................................................................................................... 76

3
7 Appendix ....................................................................................................................................... 80
7.1 Derivations ............................................................................................................................ 80
7.1.1 Mode solutions for Step Index Multimode Fibre .......................................................... 80
7.1.2 Mode Coupling in terms of Condition Number ............................................................ 81
7.2 MATLAB Code ....................................................................................................................... 82
7.2.1 Example Script to plot 3D Convergence Time Graphs .................................................. 82
7.2.2 Multimode Fibre Channel Model .................................................................................. 85
7.2.3 Code modelling Coherent Detection of the Optical Signal ........................................... 87
7.2.4 Scaled-UP CMA Function .............................................................................................. 88
7.2.5 Subplots of the data ...................................................................................................... 89
7.2.6 Power Meter Reader ..................................................................................................... 90
7.3 Project Management ............................................................................................................ 92

4
Abbreviations:
ADC Analogue to Digital Converter

CM Constant Modulus

CMA Constant Modulus Algorithm

DSP Digital Signal Processing

FIR Finite Impulse Response

FSE Fractionally Spaced Equaliser

IL Insertion Loss

ISI Intersymbol Interference

JCM Constant Modulus Cost Function

JMSE Mean Squared Error Cost Function

LTI Linear Time Invariant

MCF Multicore Fibre

MIMO Multiple Input Multiple Output

MMC Multimode Coupler

MMF Multimode Fibre

MMSE Minimum Mean Squared Error

MZM Mach-Zehnder Modulator

PBE Perfect Blind Equalisation

PMD Polarisation Mode Dispersion

PRBS Pseudo Random Bit Sequence

Rx Receiver

SE Spectral Efficiency

SISO Single Input Single Output

SMF Single Mode Fibre

SLM Spatial Light Modulator

Tx Transmitter

5
1 Introduction
The high bandwidth offered by optical fibre makes it a key component in today’s high-speed
telecommunication networks. However, with internet traffic forecast to grow at a rate of 34% year-
on-year [1], the optical bandwidth of current technologies is being rapidly exhausted. This has led to
research efforts in the capacity enhancement of existing optical communication systems. In addition,
the energy consumption and system cost are important factors. Therefore, unless energy and cost
effective capacity improvements are found the system’s cost per bit will increase as the demand out-
strips the supply.

Research efforts include multilevel modulation and multi-core optical fibre as a means of increasing
the capacity of current optical systems. Coherent detection of multilevel data formats over single
mode fibre (SMF) has resulted in record spectral efficiency (SE) of 4b/s/Hz per polarisation [2].
However, recent theoretical results have shown that the SE of SMF is limited to approximately
9/b/s/Hz by fibre non linearities [3]. A proof of principle multi-core fibre (MCF) with 7 cores has been
demonstrated at 70 GB/s transmission over 550m [4] along with investigations into the limits of this
method. Namely inter-core crosstalk which is strongly affected by fibre bending and structure [5].

This report studies the method of Mode Division Multiplexing which is the other main research field
aiming to increase the capacity of current optical systems. This method uses spatial multiplexing to
independently transmit different data streams on different spatial modes within a multimode fibre
(MMF). This method of capacity increase is also receiving significant interest in 10Gb/s Local Area
Networking (LAN) and Data Storage Centres [6] where the majority of installed fibre is multimode.
MMF is limited in capacity and transmission distance by modal dispersion. Mode division
multiplexing aims to increase the bandwidth-distance product thus increase the bandwidth of such
fixed length systems.

Significant MMF capacity enhancement was carried out in [7] where a 7.5Gbit/s/km bandwidth-
distance product was demonstrated using an offset launch technique. Recently, 107Gbit/s over MMF
using OFDM has been achieved in [8]. Though these approaches increase the capacity of MMF, they
do not fully exploit the modal diversity inherent in MMF. Since controlled transmitter centre
launching and receiver mode filtering aim to excite and extract only the fundamental mode from the
MMF. Hence the higher order modes are not excited, resulting in an underutilised channel.

Mode division multiplexing was first demonstrated using two subcarriers to multiplex two 50Mbit/s
data streams over two spatial modes through 1km MMF [9]. Where the concept of multipath
propagation in a wireless multiple antenna system, to increase capacity, was used in an optical

6
system. This concept is known as multiple-input multiple-output (MIMO) communication. Recent
developments in optical MIMO include a 20Gbit/s per channel 2x2 system over 1.1km using IM-DD
[10] and an 8Gbit/s 2x4 system over 5km using IM-DD [11]. These both utilised MIMO digital signal
processing (DSP) at the receiver to reverse the effects of the MMF channel. This was made possible
through high speed analogue to digital converters (ADC) which can sample the analogue electrical
signal at the required rate allowing DSP algorithms to recover the transmitted data. A benefit of DSP
equalisation is its ability to track changes in the time varying channel to maintain consistent
performance.

Other mode division multiplexing methods have used Few Mode Fibres (FMF) to reduce the mode
coupling effects in the multimode channel [12], [13], [14]. However FMF does not provide the
number of spatial modes required for high speed future state of the art applications that MMF can
offer. To fully exploit numerous spatial modes in MMF requires complex DSP which is limited by ADC
speeds. As ADC speeds increase, so will the ability to multiplex more spatial channels in a MMF.

This report consists of two sections. Section I describes the simulation of an optical MIMO system in
MATLAB, modelling channel effects and analysing the DSP behaviour. Section II explains lab tests
aimed at increasing the spatial diversity at the transmitter and receiver in a 2x2 mode division
multiplexed MIMO system.

7
2 Section I: MIMO System Simulation
This section describes the MATLAB simulation of a coherent optical MIMO (COMIMO) system
utilising a digital coherent receiver with emphasis on the adaptive equaliser in the receiver DSP. The
equaliser’s response to polarisation and mode dependent channel effects will be investigated
leading to the analysis of a simulated 4x4 MIMO system.

2.1 Theory

A Multiple Input Multiple Output (MIMO) system uses multiple transmitters to send data through a
channel to multiple receivers. Independent data streams can be sent from each transmitter in order
to increase the system capacity or the same data stream can be sent from each transmitter to
improve the system resilience to channel fading etc. In an optical MIMO system, multiple
transmitters and receivers are used to exploit the spatial diversity of Multimode Fibre (MMF) by
transmitting independent data streams on different optical modes. A MIMO system is shown in
Figure 1 where the channel, H, represents the optical MMF channel.

Figure 1 High Level 2x2 MIMO System Diagram [15]

The output of a linear time invariant (LTI) channel, y(t), is defined as the convolution between the
input signal, x(t), with the impulse response, h(t).

(1)

(2)

Where, v(t) is complex additive white Gaussian noise.

Assuming narrowband communication, the impulse response can be defined as a delayed and
weighted delta function. Hence the output is given as a delayed and weighted version of the input
signal plus noise.

(3)

8
(4)

Making this assumption simplifies the model allowing each modal path in a multimode fibre to be
modelled as:

(5)

Where, τg and h0 are the group delay and scalar fading parameter associated with the propagating
mode respectively.

2.1.1 Mode Multiplexed Channel Model

Each mode experiences a different group delay in multimode fibre. Therefore different excited
modes are received at different times causing pulse spreading, resulting in intersymbol interference
(ISI). Equation (6) represents a Single Input Single Output (SISO) multimode fibre communication
channel that accounts for this effect. This assumes a frequency independent channel such that the
variation in phase velocity for each mode is negligible resulting in zero chromatic dispersion.

(6)

Where P is the number of guiding modes in the MMF and τgk is the group delay associated with the
kth guiding mode. hk represents the channel distortion effect associated with the kth guiding mode.

(7)

Where αk , τpk and Kk represent the attenuation, phase delay and coupled power into the kth guiding
mode, respectively.

The above model extends to a Multiple Input Multiple Output (MIMO) system with M transmitters
and N receivers, given by:

(8)

Where,

(9)

Ki,j,k describes the coupling from the jth transmitter to the ith receiver through the kth mode.

The group delay spread (Δτg) is given by the difference between the maximum and minimum modal
delays. When Δτg is small compared to the symbol period it can be assumed that all modes are
received at the same time (Δτg = 0). Therefore τgk in Equation (8) becomes τg for all k. Assuming ideal

9
down conversion of y(t) and sampling at a rate of 1/Ts, this gives the discrete received baseband
signal y[n] = y(nTs). Hence the LTI flat fading narrowband communication model is expressed in
matrix form as:
(10)

Where, L is the total number of samples. y[n] and x[n] are the M x L received and transmitted
sampled signals respectively. H is the N x M channel matrix defining the power coupling between
the propagating paths in the channel. Where the ith row and jth column element of H is:

(11)

An example 2x2 MIMO system expressed in matrix form is given by:

(12)

Modelling the channel as LTI and frequency independent simplifies the problem. Since the channel
matrix, H, only describes the transfer of power between the two paths during propagation through
the channel. x1[n] and x2[n] represent two transmitted spatial signals either on two modes or two
polarisation states in the MMF channel. The system model is the same in each case.

2.1.2 Polarisation Multiplexed Channel Model

The above channel model, Hi,j, described the coupling effects between modes in a modal multiplexed
system. A 2x2 MIMO system utilising 2 mode multiplexing was given as an example. A 2x2 MIMO
system can also be realised by multiplexing data on to the two polarisation states of a given mode.
The instantaneous electric field of a light ray is given by:

(13)

Where and represent the X and Y polarisation state vectors, respectively, and k is the
propagation constant. and represent the peak oscillation values of the X and Y polarisation
states, respectively. is the complex amplitude of the electric field vector given by the Jones vector:

(14)

10
45° linear polarised light is given by and such that both orthogonal
polarisation states are in phase with each other and of equal amplitude. Giving the transmitted
electric field vector:

(15)

Assuming 45° linearly polarised optical transmission, channel effects causing a deviation from this
initial polarisation state can be monitored. Two channel effects which result in a change in
polarisation state will be considered. These are polarisation rotation and polarisation phase shift. A
polarisation rotation is modelled by the Jones (rotation) matrix, R:

(16)

R rotates the horizontal ( ) and vertical ( ) states of polarisation by an angle to a new pair of
orthogonal polarisation states.

A polarisation phase shift is modelled by the matrix:

(17)

This results in a phase shift of between the two polarisation states. This corresponds to

in Equation (14) where .

(18)

Assuming ideal down conversion, sampling and demodulation of the modulated data on the optical
carrier, the flat narrowband LTI discrete system is:

(19)

Where,

(20)

The polarisation diverse coherent receiver resolves the received polarisation states in the and
axes. Therefore a polarisation phase shift and rotation results in a received signal that is a ratio of
the two polarisation state powers.

11
The narrowband LTI channel model for the 2x2 modal multiplexed MIMO system and the 2x2
Polarisation multiplexed MIMO system is the same. The difference being the Channel Matrix, H, that
describes the coupling between the two spatial paths.

2.1.3 Channel Equalisation

At the receiver, an equaliser is used to reverse the effect the channel has on the transmitted signal.
Equalisation of the above N x M channel matrix, H, requires an equaliser coefficient matrix, C, which
is the matrix inverse of H:

(21)

Applying the M x N equaliser matrix, C, to the channel output, y[n], gives:

(22)

Where, I is the identity matrix. M and N denote the number of transmitters and receivers into and
out of the channel, respectively. is the equaliser output at time n.

Hence the output of the equaliser is the transmitted data plus a modified channel noise. This
method is termed ‘zero forcing’ and results in perfect ISI equalisation in a noiseless channel [16].
However, when noise is present, the inverse property of the zero forcing equalisation amplifies the
smaller noise terms thus degrading the system SNR.

Equalisation of the 2x2 MIMO example is shown by:

(23)

In the above example, C is a 2x2 matrix that recovers the transmitted data after propagating through
the frequency independent channel H and each C element is a scalar value. If the channel is
frequency dependent this causes current samples to be dependent on past and future samples.
Therefore each element of C requires an FIR filter, of length L, to undo this dispersive (or channel
memory) effect:

(24)

At time n, L samples are taken from each received stream, yM, and multiplied by C to give the
equaliser output, . The matrix representation of a 2x2 MIMO example using an FIR equaliser is:

12
(25)

The FIR filters aim to equalise the ISI effects caused by the frequency dependent channel. For
example, in a polarisation multiplexed system, PMD is a frequency dependent effect that causes a
spreading of power between symbols resulting in ISI. This effect is analogous to modal dispersion in
a modal multiplexed system where different mode groups travel with different velocities through
the MMF. L is chosen such that the pre and post cursors of the channel impulse response with
significant energy are equalised [17]. This improves the SNR and ultimately the BER of the system.

Using adaptive FIR filters in the coefficient matrix, C, allows the equaliser taps to be updated.
Therefore time varying channel effects such as polarisation rotation and polarisation carrier phase
shift can be compensated for.

The above example uses a Baud Spaced Equaliser where the coefficient tap spacing is an integer
multiple of the symbol period. More than one tap per symbol period can be used to equalise the
channel output signal. This equaliser is termed a Fractionally Spaced Equaliser. Where,

(26)

M is the number of taps per symbol. 2L is the total number of pre and post cursor symbols of the
channel impulse response that are equalised. Hence, the desired symbol sample that is equalised is
aligned with the centre of the equaliser tap vector. Each iteration, the tap vector moves M taps
through the sampled signal to equalise the following symbol. FSEs provide a more robust channel
equalisation method than Baud Spaced Equalisers. A Baud Spaced Equaliser would require infinite
taps to perfectly equalise an FIR channel whereas a FSE can achieve perfect equalisation using a
finite number of taps, under certain channel constraints [16],[17] (see Section 2.3.1).

Under noisy conditions an alternative equalisation method to the zero forcing approach is needed. A
method that reduces the noise amplification, inherent in the zero forcing method, whilst minimising
the ISI is required. A compromise between these two criteria is provided by an equaliser that aims to
minimise the mean squared error in a minimum mean-squared error (MMSE) sense. When the
transmit sequence, x[n], is known at the receiver the equaliser error, e[n], is defined as:

13
(27)

For example, this is the case when a known training sequence is transmitted. The equaliser tap
coefficient matrix is updated to minimise the mean square error (MSE) criterion using an adaptation
algorithm. This algorithm aims to minimise a scalar function that is dependent on the MSE criteria.
This scalar function is termed the cost function. The cost function based on the MSE criterion (JMSE) is
given by:

(28)

Minimising the MSE cost function results in the optimal Weiner solution [18]. This solution can be
achieved using a stochastic gradient algorithm such as the Least Mean Squares (LMS) algorithm. The
LMS algorithm calculates the instantaneous gradient of the cost function and uses it to update the
equaliser tap coefficient matrix in order to minimise the cost on the next iteration. The MSE cost
function is a quadratic scalar function having one global minimum. Therefore there exists one unique
tap weight solution resulting in a minimum equaliser error. A system using an adaptive equaliser to
restore the transmitted signal after time varying channel propagation is shown in Figure
2.

Equaliser
System Input System Output
Channel Adaptive FIR
Matrix (H) Filters ( )

Adaptation Error
Algorithm Measurement

Figure 2 Baseband system using an adaptive equaliser to recover the transmitted data from the channel output
When the transmitted signal is unknown at the receiver, this changes the error measurement
resulting in a different error signal definition. Therefore the MSE criteria used to control the LMS
algorithm can no longer be calculated due to a lack of signal information. Because the MSE criteria
were calculated using a training sequence where all the signal properties were known at the receiver.
Equalising a signal where there is a lack of signal knowledge at the receiver is termed Blind
Equalisation. Blind equalisation saves link capacity since training sequences are not required to
recover the data. A common blind equalisation technique uses the Constant Modulus Algorithm
(CMA). The CMA exploits the constant modulus property of certain modulation schemes. This known
signal property at the receiver is used to formulate a cost function. For example, in QPSK modulation,

14
deviations in the squared modulus of the equalised symbol value from a predefined value result in
an error. Defined by:

(29)

Where A is the modulus of the transmitted QPSK symbol, given by from the Godard

Algorithm where = corresponding to the Constant Modulus Algorithm [19]. A is commonly


normalised to 1.

The Mean Squared CMA Error gives the Constant Modulus (CM) cost function:

(30)

The CMA is a stochastic gradient algorithm that uses the instantaneous CM cost gradient to update
the equaliser tap weights to minimise the CMA error, eCMA. Similar to how the LMS algorithm
minimises JMSE. The stochastic gradient algorithm is defined by:

(31)

µ is the step size parameter that controls the size of the incremental correction towards the cost

function minima. for the CMA is given by:

(32)

Where,

Therefore the CMA algorithm is defined as:

(33)

The CM cost function is non-convex unlike the convex MSE cost function which exhibited one global
minimum. Hence the CM cost function exhibits multiple equaliser tap coefficient solutions for a
given input. This means that the equaliser can converge to the incorrect cost minima [20]. In a 2x2
MIMO system this could mean both outputs converge to the same input resulting in a loss of half the
data. This problem is strongly dependent on the initial equaliser tap weights. Setting the initial tap

15
weights to the identity matrix provides an initial guess that assumes no coupling between the spatial
paths. In an ideal case the initialised tap weights would result in the optimal solution giving
minimum equaliser error. In reality, the channel will induce a coupling between the spatial paths
resulting in equaliser error. This error is then used to update the tap weight matrix to minimise the
recovery error for the next iteration.

16
2.2 MIMO Simulation Setup

The MIMO system was simulated in MATLAB. The simulation diagram is shown in Figure 3. M
denotes the number of multiplexed spatial modes.

Transmitter
PRBS
Sequence

Upsample
x16
1 1 1 1 1 MMF
PRBS Low Pass Set Optical
Triple MZM Channel
Cyclic Shift Filter Power Amplifier
M M M M M (H)

1 1 (Re + Im)Xpol 1
1
(Re + Im)Ypol
Downsample Low Pass 90° Optical
Normalise (Re + Im)Xpol
÷8 Filter Hybrid
M
M M (Re + Im)Ypol
M
1 1

BER
CMA Carrier Phase
BER Analysis
Equaliser Recovery Receiver
M M

Figure 3 MIMO System Simulation Diagram, MZM: Mach Zehnder Modulator

Pseudo Random Bit Sequence:

For each mode, random QPSK modulated data was transmitted on both polarisation states. A PRBS
sequence of length 214-1 was generated to achieve this. Figure 4 shows the first 50 bits of an
example PRBS:

17
Figure 4 First 50 bits of an example PRBS

Upsample and Cyclic Shift:

The PRBS was upsampled to 16 samples/bit to represent a continuous digital waveform, Figure 5.
Different cyclic shifts were applied to the upsampled PRBS to give a set of uncorrelated bit streams.
Where each pair of uncorrelated bit streams represented the real and imaginary parts of each QPSK
data stream. And each pair of QPSK data streams represented the data on each mode’s X and Y
polarisation states, respectively.

Figure 5 First 31 bits of an upsampled PRBS (16 samples/bit)

18
For the 2x2 MIMO system simulation exploiting 2 mode multiplexing, 8 copies of the initial PRBS

were created and cyclically shifted such that each was separated by of the bit sequence length.

Giving 8 uncorrelated PRBSs corresponding to the Real and Imaginary data on each of the X and Y
polarisation states for each of the two modes.

Low Pass Filter and Modulate:

The electrical PRBSs were fed through a low pass filter function. This modelled the bandwidth
limited driver circuitry of the Mach-Zehnder Modulator. Figure 6 shows the waveform of Figure 5
after low pass filtering. The uncorrelated PRBS electrical waveforms, as in Figure 6, were then
applied to the Mach-Zehnder Modulator structures resulting in two Dual Polarisation QPSK (DP-QPSK)
modulated optical data streams. The MZM structure for one mode is shown in Figure 7.

Figure 6 Low Pass Filtered MZM drive waveform

19
Re
Electrical PRBS
MZM
Drive Bit Streams

Im

MZM

Optical
Optical
PBS PBC Signal Out
Signal In
MZM DP-QPSK

Im
MZM

Re

Figure 7 MZM Structure resulting in DP-QPSK Modulated Data for 1 Optical Mode
PBS: Polarisation Beam Splitter, PBC: Polarisation Beam Combiner

Set Power and Optical Amplifier:

The optical power was set to a specified level suitable for the optical amplifier. Noiseless
amplification was assumed.

Multimode Fibre Channel:

Polarisation dependent effects and modal coupling were modelled by the MMF channel. Zero
Differential Mode Delay (DMD) and zero Polarisation Mode Dispersion (PMD) were assumed.

Coherent Receiver:

The 90° Optical Hybrid Coherent Receiver linearly mapped the optical signal into the electrical
domain by mixing the optical channel output with an optical Local Oscillator (LO). Maintaining the
polarisation, phase and amplitude information of the 8 transmitted PRBSs.

This function was scaled up to deal with ‘M’ modes, See Appendix 7.2.4.

Filter and Downsample:

The electrical signals were passed through a low pass filter. This provided matched filtering to the
low pass filter function at the transmitter which modelled the bandwidth limited drive circuitry. The

20
electrical signals were then downsampled to two samples/symbol. Such that a T/2 Fractionally
Spaced Equaliser could be applied to the signal.

Normalise Signal and Equalise:

The downsampled signal was normalised to unity magnitude. This meant that the initial equaliser
tap coefficient matrix could be set to the identity matrix to provide an optimal initial ‘best guess’.
This removed the need to calculate the initial equaliser taps from the input signal properties. The
CMA equaliser output signal constellation diagram is shown in Figure 8. Note the relatively constant
modulus of the CMA equaliser output. The circular property of the equalised output is due to a
severe LO phase noise modelled in the coherent receiver. A similar effect would be observed with a
LO frequency offset.

Figure 8 CMA Equaliser Output with modelled LO phase noise

Carrier Phase Recovery:

To undo the phase variations on the received constellation, carrier phase recovery (CPE) was
required. The frequency offset was assumed to be negligible. CPE maps the QPSK constellation
points to a single point by raising the signal to the fourth power. This removes the modulated data
from the signal, leaving the phase offset. This result is summed over a number of samples, providing
a time average. The CPE algorithm minimises the estimated phase error between two consecutive

21
samples by subtracting an accumulated phase offset from each symbol, as shown in [21]. This
resulted in the constellation diagram of Figure 9.

Figure 9 Signal Constellation after Carrier Phase Estimation has been applied

BER Analysis:

Finally, threshold decisions were applied to the CPE output constellation resulting in a set of data
values {1+j, 1-j, -1+j, -1-j}. These data values were cross correlated with the initial PRBS data stream.
This gave the amount of shift that was applied to the PRBS to give that transmitted sequence,
corresponding to the cross correlation maximum. The PRBS and the received sequences were then
aligned accordingly and an XOR function was applied. This gave a BER value for each bit stream. The
8 bit streams were averaged and a system BER was outputted.

Note: For all the simulation experiments, a BER ≤ 10-3 was assumed. Therefore, assuming Forward
Error Correction (FEC) is used, a desired BER of 10-9 can be achieved.

The MIMO simulation model assumed the following:

 Zero Chromatic Dispersion


 Noiseless Optical Amplification
 Channel variations occurred at a much slower rate than the equaliser convergence time
allowing the equaliser to track the channel
 Zero propagation delay incurred in the MMF channel
 Perfect 90° Hybrid Coherent Receiver giving perfect signal orthogonalisation

22
 Perfect homodyne coherent detection (no frequency estimation required, just carrier phase
estimation to reverse the effects of LO phase noise)
 Perfect balanced differential coherent detection assumed with ideal photodiode
responsivities
 Polarisation and Mode dependent loss assumed negligible
 Assumed ideal transmitter and receiver coupling efficiencies into and out of the MMF
 Non ideal MZM drive circuitry, modelled by a Low Pass Transfer Function

2.3 Simulation Results

The equaliser convergence behaviour was assessed under different simulated channel effects and
CMA parameter settings. A fractionally spaced equaliser (FSE) with two taps per symbol was used in
the simulations. The FSE length was fixed at 51 taps thus modal delays of up to 25 symbols could be
compensated. A symbol rate of 10 GBaud was simulated.

2.3.1 CMA Step Size Parameter

A zero ISI 2x2 MIMO channel with 20dB OSNR for each transmit path was simulated. No coupling
between transmit paths was assumed. The four conditions which must be satisfied in order to
achieve Perfect Blind Equalisation (PBE) of the Constant Modulus criterion are [22], [23], [24], [17]:

1. Full column rank Channel Matrix (H)


2. Noiseless Channel
3. Circularly symmetric i.i.d source
4. Sub-Gaussian source

Modelling channel noise violated the second of the above conditions. This allowed the analysis of
non-ideal channel equalisation resulting in imperfect symbol recovery. Figure 10 shows the effect
that changing the CMA step size parameter had on the CMA error convergence curve. The curves
were smoothed using a brick wall low pass filter, length 300 samples, to clearly show the effect of
varying μ.

23
.

Figure 10 CMA Convergence Curves 20dB OSNR:


-5 -4 -3 -3
(Blue: +) µ=5x10 (Red: o) µ=1x10 (Black: x) µ=1x10 (Green: ·) µ=5x10

CMA Step Size Parameter Steady State Error (Abs)


-5
5x10 0.0332
1x10-4 0.0359
1x10-3 0.0364
5x10-3 0.0511

Table 1 Steady State CMA Error values for different Step Size Parameter Values

The steady state error values were calculated by averaging the CMA error over the final 10,000
equaliser iterations where the equaliser had reached steady state i.e. converged.

Increasing the CMA step size parameter, µ, reduced the time it took the equaliser to reach steady
state, shown by Figure 10. However this increased the steady state error of the equaliser,

Table 1. Therefore the choice of step size value offered a trade-off between the convergence rate
and steady state error of the equaliser. The convergence rate of the equaliser shows its ability to
adapt to time varying changes in the channel. If the step size is too small, the equaliser may not track
the channel variations sufficiently. Resulting in tracking lag which increases the CMA error as the
equaliser is unable to keep within the vicinity of the time varying CM cost function minimum[22].

24
Violating the above PBE conditions altered the CM cost function resulting in a CMA solution that did
not match the desired optimal Weiner solution [22]. This gave a non-zero steady state error due to
the residual CMA error present after each equaliser iteration. The step size parameter had a
multiplicative effect on the CMA error term, shown by Equation (33). This explains the increase in
steady state error with increasing µ.

A step size parameter of 1x10-3 was chosen based on the speed of convergence and the relatively
low steady state error value associated with it. This value was used for all the subsequent
simulations.

2.3.2 Equaliser Convergence Criteria

Once the optimal step size value was chosen, a method for defining the convergence rate of the
equaliser was required. The convergence time is defined as the number of equaliser iterations
needed to reach steady state behaviour [22]. This provides a measurement which defines the
equaliser performance. Better performance is indicated by a quicker convergence time for a given
step size parameter.

The following simulations do not model channel noise or ISI. Channel noise was only modelled to
find the optimum step size value in the previous section. The convergence time was measured using
the cumulative sum of the CMA Error2. The squared error gave a positive result therefore the
cumulative sum grew positively with each iteration. Since channel noise was not present, the
cumulative sum gave a flat response as the equaliser converged due to the small steady state error,
Figure 11.

25
2
Figure 11 Cumulative Sum of the CMA Error

The cumulative sum provided a smooth curve on which to make a convergence time measurement.
This eliminated the problem of directly measuring the error curve which exhibited large variations
causing measurement inaccuracy. The convergence time was defined as the number of equaliser
iterations required to reach 80% of the final value of the cumulative sum curve. Assuming that once
the error had reached 80% of its final value the equaliser solution was in the vicinity of the CM Cost
Function minima i.e. had converged.

2.3.3 Polarisation Dependent Effects

The above 2x2 MIMO simulation was extended to account for the following polarisation dependent
channel effects:

 Polarisation Rotation,
 Polarisation Phase Shift,

N.B. Polarisation dependent loss was considered negligible.

This was achieved by creating a MATLAB function that modelled the optical channel in accordance
with the unitary matrix of Equation (20). Figure 12 shows the mean square CMA error, plotted
against polarisation rotation, , and polarisation phase shift, , for a fixed equaliser tap matrix. The
equaliser tap matrix was set to the identity matrix such that the central taps of the c11 and c22 FIR
filters were unity with all other matrix elements set to zero:

26
(34)

The tap matrix, C, was fixed by setting the CMA step size to zero such that the equaliser did not
update each iteration. and were incremented from to in steps of resulting in 16,641

simulation points.

Figure 12 Constant Modulus Cost Function

Figure 12 shows the constant modulus cost function, JCM, with respect to and for one equaliser
output channel. The fixed identity tap matrix, C, provided the optimal solution to the ideal channel
( = =0), i.e. minimum error. The equaliser error showed a strong dependence on . Moving away
from the ideal 0 value generated error, until reached where minimum error was achieved

again. This rotation of corresponded to the orthogonal polarisation states being perfectly

resolved on the X and Y axes, hence zero error.

The maximum equaliser error occurred at a polarisation rotation of = and a polarisation

phase shift of . These maxima corresponded to circularly and linearly polarised light for

27
odd and even n and m respectively. These maxima occurred due to all the channel matrix elements
of H being of equal magnitude:

(35)

This meant that equal power was resolved on to the X and Y polarisation axes from both X and Y
polarisation states. Hence the equaliser’s initialised identity tap matrix gave a worst case
approximation to equalise the received signal resulting in maximum error. At this point on JCM the
CMA is equally probable to converge to a minimum either side of the maximum singularity. This can
cause the equaliser outputs to converge to the same source. A problem inherent in systems
employing independent channel blind equalisation techniques due to their multimodal cost
functions [20]. Methods exist to overcome this problem. For example, in [25] a cross correlation
term between the equaliser outputs is used in the formulation of JCM to ensure correct equaliser
convergence.

The minimum values in Figure 12 were not zero due to the non-ideal components in the simulated
system inducing recovery error. Namely the Low Pass Filter function which modelled bandwidth
limited driver circuitry.

and vary on a time scale of the order of milliseconds due to mechanical perturbations[26]. The
equaliser convergence time is of the order of 2000 symbols (at 10GBaud = 200ns), Figure 13.
Therefore it can be assumed that the polarisation dependent effects, and , are fixed over the
equaliser convergence period.

28
Figure 13 shows the convergence time of the equaliser as a function of and . This shows whether
the equaliser converged and, if so, how many equaliser iterations it took.

Figure 13 Equaliser Convergence Time for different Polarisation Rotation and Polarisation Phase Shift Values

The convergence time was given by the convergence criteria in section 2.3.2. To ensure that the
equaliser had converged to the correct data streams, the real and imaginary components of the
equaliser outputs were cross correlated with their four transmitted PRBS data sequence
counterparts. The PRBS sequences were offset by a quarter sequence length from each other.
Therefore if the equaliser had converged correctly the cross correlation product gave four maxima
each at a distance of a quarter sequence length apart. If this occurred, the convergence time was
plotted for the corresponding and value. Otherwise the convergence time was not plotted and
and were incremented to their next simulation values. This explains the omitted values in Figure
13.

Figure 13 exhibits similar properties to Figure 12 showing that a larger initial CMA error increases the

equaliser convergence time. The longest convergence time occurred in the vicinity of

corresponding to the region of maximum initial CMA error. The convergence time in Figure 13
peaked where the saddle points occurred in Figure 12. Since the CMA experienced a slow change in
gradient as it traversed a saddle point on the cost function curve.

29
The equaliser did not converge correctly when = . This was due to the problem discussed

above where the equaliser outputs locked to the same input source due to the starting point on JCM
being at the singularity.

2.3.4 4x4 MIMO System: Polarisation and Mode Multiplexed

Coupling between modes occurs when light propagates in MMF. This is due to fibre bends, refractive
index imperfections and fibre core ellipticity and eccentricity [27]. Mode coupling effects need to be
equalised at the receiver to reduce errors in the received data. This is achieved using the CMA
equaliser to reverse the modal coupling effects as in the polarisation multiplexed case where the
power coupled between polarisations was equalised. The Mode Coupling between the two modes
was modelled in matrix form by:

(36)

For a 2x2 mode multiplexed MIMO system the received power on each mode, y, from each
transmitted mode, x, is given by:

(37)

In the ideal case of no mode coupling; M is the identity matrix i.e. all the received power on a
particular mode is due to the power transmitted on that mode. Modal coupling was achieved by
introducing coupling terms into the zero elements of the ideal identity M matrix. The mode coupling
was expressed in terms of the Channel Matrix condition number. The channel matrix can be defined
as:

(38)

Where U and V are unitary matrices and S is a diagonal matrix whose diagonal values, Si,i, are the
singular values of M. The Channel Matrix Condition number is given by:

(39)

Equations (40) and (41) show a mode coupling example with a channel matrix condition number of
CNdB = 3dB:

30
(40)

The columns of M were normalised by dividing each column by the column sum. This ensured that
the total received power per mode was unity, representing lossless propagation. Modal dependent
loss was assumed negligible such that the coupling between modes was of an equal ratio.

Note: This assumption allowed the formulation of M using ‘X’ in Equation (40). See Appendix 7.1.2
for the derivation of X.

Normalising M gave:

(41)

The 2x2 MIMO system was extended to a 4x4 MIMO system utilising two modes with both
polarisation states on each mode being used to multiplex data. Resulting in a doubling of capacity
compared to the 2x2 MIMO system. The modelled 4x4 MIMO system was given in matrix form by:

(42)

Where and represent the transmitted and received digital data sequences on the ath
mode (1 or 2) and bth polarisation state (X or Y). The 4x4 channel matrix, H, now accounts for
polarisation and mode coupling.

The Channel Matrix condition number, CNdB, was increased in steps of 0.2dB and the equaliser
convergence time was measured at each point. This was repeated for different polarisation rotation
and phase shift values ranging from = =0° to = =45° in 5° increments. The results are shown in
Figure 14 and Figure 15.

31
Increasing φ
and θ values

Figure 14 Modal Coupling Effect on Equaliser Convergence Time


(Blue: x) φ=θ=0° (Red: o) φ=θ=10° (Black: -) φ=θ=20° (Green: +) φ=θ=30° (Orange: ·) φ=θ=40°

Increasing φ
and θ values

Figure 15 Modal Coupling Effect on Equaliser Convergence Time


(Blue: x) φ=θ=5° (Red: o) φ=θ=15° (Black -) φ=θ=25° (Green: +) φ=θ=35° (Orange: ·) φ=θ=45°

32
The results show that as the channel condition number increased, the convergence time of the
equaliser increased. Greater diversity between the two spatial modes corresponded to a lower
condition number. For lower condition numbers there was less unwanted coupled power on each
received signal. Therefore the initial error of the CMA equaliser was lower so the equaliser
converged more quickly. As seen in the polarisation multiplexed case, Figure 13. The simulation
points that did not converge correctly were not plotted. Explaining the missing data points for the
= =45° curve.

For condition numbers below 4.2dB, the equaliser converged for all and values, assuming
= =45° to be worst case from the previous results.

The CMA equaliser MATLAB function was made scalable to deal with any NxN MIMO system. (See
Appendix 7.2.4) Allowing user entry of the number multiplexed modes to be equalised. Figure 16
and Figure 17 show the converged real and imaginary components of the equaliser tap vectors for
an ideal 4x4 and 6x6 MIMO system, respectively. These have been included to show the MATLAB
code scalability. Note how the central real tap weights of the diagonal matrix elements are
approximately unity with all other matrix elements ≈ 0. This shows the equalisation of the ideal
identity channel matrix case.

Figure 16 Real (Blue) and Imaginary (Green) parts of the Equaliser FIR Filters for a 4x4 MIMO System

33
Figure 17 Real (Blue) and Imaginary (Green) parts of the Equaliser FIR Filters for a 6x6 MIMO System

2.4 Summary: Simulation

A polarisation multiplexed 2x2 MIMO system was simulated. The optimal CMA step size was chosen
and the CMA convergence criteria were outlined. The effects of polarisation rotation and
polarisation phase shift on the signal were analysed in terms of the equaliser convergence time. A
4x4 MIMO system was then simulated which multiplexed two spatial modes each utilising two
polarisation states in an optical channel. The effects of mode coupling on the equaliser convergence
time were analysed. It was shown that for a channel matrix condition number of less than 4.2dB the
equaliser converged for all values of polarisation rotation and polarisation phase shift. This suggests
that a CMA based equaliser is able to adaptively track polarisation changes in a MMF channel given
necessary modal diversity. Under the stated assumptions.

34
3 Section II: Lab Test Results
This section describes the optimisation of the transmitter and receiver coupling in a Multimode Fibre
COMIMO system exploiting 2 mode multiplexing. Various methods are investigated to maximise the
modal diversity between the two modes multiplexed in the MMF channel whilst maintaining a
constant insertion loss for each output.

Firstly, different MMC arrangements are compared. Secondly, a free space receiver coupling method
is analysed. Finally, a receiver based modal coupling method using a spatial light modulator (SLM) is
analysed. This aims to maximise the coupling into the desired mode groups at the transmitter,
increasing the modal diversity. A summary of the results for each test setup is shown provided in
Section 3.7 where the optimal method is highlighted based on system trade-offs.

3.1 Theory

An optical mode represents a unique path of light propagation through an optical waveguide.
Multimode Fibre can allow a number of modes to propagate. Light is a form of electromagnetic
radiation. The time varying electric and magnetic fields can be described in circular coordinates by:

(43)

(44)

Where, and are the electric and magnetic fields, respectively. and are the fibre cylindrical
coordinates. is the propagation constant and is the unit vector parallel to the z-axis. The electric
and magnetic fields are decomposed into their longitudinal ( , ) and transverse ( , )
components. and , Figure 18.

Figure 18 Cylindrical polar coordinate system used to describe the spatial dependence of the electric and magnetic fields
of the propagating light wave. [28]

35
The modes in a multimode fibre with a step refractive index profile are the solutions to the scalar
wave equation [29],[30],[31]:

(45)

Where represents the longitudinal or field. is the core refractive index. k is the
propagation constant of light in a vacuum. and are the fibre cylindrical coordinates.

Equation (45) is obtained from Maxwell’s Equations (See Appendix 7.1.1) where the step index fibre
is assumed to be weakly guiding such that . This leads to simpler solutions, giving
rise to degenerate modes [28].

Note: Graded Index (GI) MMF was used in the following lab tests. The above mode solutions of the
step index fibre were used to approximate the modes in the GI fibre for ease of calculation.

There are a finite number of ‘guided’ modes and a continuum of ‘radiated’ modes supported by a
multimode optical fibre [32],[28]. These modes are given by:

(46)

Where the guided modes are denoted by the two finite summations and the integral indicates the
continuous distribution of radiated modes. c and d are the modal amplitudes for the azimuthal and
radial components, respectively. This shows a radial and azimuthal dependency for each mode.

The number of guided modes supported by a multimode fibre is determined by:

 Core Diameter (2 )
 Transmission Wavelength ( )
 Numerical Aperture (NA)

A measure of this is given by the normalised frequency, otherwise known as the V-parameter:

(47)

And the number of supported modes is:

(48)

Hence for 62.5µm (OM1) step index fibre operated at 1550nm and assuming a NA of 0.2, V = 25.36.
This gives rise to approximately 322 modes.

36
The guided modes are a combination of Transverse and Hybrid modes. Namely :TE, TM, EH, HE
modes. The transverse modes TE and TM correspond to zero and field components,
respectively. The hybrid modes HE and EH correspond to non-zero and field components.
Visually depicted as a meridional (Transverse) and skew rays (Hybrid) in a ray model [30].

As a result of the weakly guiding fibre approximation, some transverse and hybrid modes
demonstrate approximately the same propagation constants, . Such modes are said to be
degenerate. These transverse and hybrid modes are grouped in terms of their matching values.
These groups are labelled (Linearly Polarised) modes. Where ‘ ’ denotes the number of field
maxima around the circumference of the fibre core and ’ ’ denotes the number of maxima along
the core radius vector. A selection of LP electric field distributions are shown in Figure 19.

Figure 19 A selection of LP Modal Distributions. The different colours correspond to different signs of the electric field
values [33]

37
The mode is the lowest order (or fundamental) mode. The order of modes supported increases
as the V-number increases, Figure 20. Therefore a multimode fibre such as that above with V=25.36
supports many higher order modes.

Figure 20 Supported Modes depending on the V-parameter, Vc denotes a mode’s normalised cut-off frequency

Mode Coupling:

Fibre bends, core diameter variations and refractive index variations cause coupling between the
modes in multimode fibre. Mode coupling between guided modes causes signal distortion, reducing
the modal diversity. Mode coupling from guided modes to radiation modes causes signal loss as
power is dissipated away from the MMF core.

When two parallel waveguides are brought close together, the modes in each waveguide couple
between each other as a result of evanescent wave coupling. Modes couple from one waveguide to
the other if they have similar propagation constants, . Consider the mode electric fields of
waveguide ‘r’ and ‘s’ given by ER(x,y) and ES(x,y), respectively. Defining the complex mode electric
field amplitudes as R(z) and S(z), the coupling between the two waveguides with propagation
distance, z, is approximated by the first order differential equation.

(49)

Where, is the mode coupling coefficient from waveguide R to S and is the mode
coupling coefficient from waveguide S to R, for a given mode ‘m’.

38
Launching mode ER with an amplitude of 1 with the initial power in ES = 0 gives the initial conditions:
R(0) = 1 and S(0) = 0. Giving the solutions to Equation (52) as:

(50)

Suggesting a sinusoidal energy exchange between coupled waveguide modes, dependent on and .
Where and represent the waveguide spacing and interaction length, respectively. Hence for a
fixed waveguide spacing, the interaction length is chosen to maximise the coupling from one
waveguide into the other and minimise the coupling in the other direction giving: R(z) = 0, S(z) =1.
Such that, ideally, the total power in mode ‘m’ couples from waveguide R to waveguide S over an
interaction length = z.

The above coupler theory explains the coupling properties in a Multimode Coupler (MMC). A 1x2
MMC is shown in Figure 21.

Coupling Port 3
Region (Through)

Port 1

Port 4
(Cross)

Figure 21 1x2 Multimode Coupler

The 1x2 MMC separates the modal distribution at Port 1 into two different distributions
corresponding to higher order modes and the fundamental mode. The fundamental mode
propagates mainly on the Through Port while the higher order modes couple mainly into the Cross
Port. Hence the 1x2 MMC is used at the receiver end of a 2 mode multiplexed system to provide
modal diversity into the respective receivers. The 1x2 MMC can also be used in the reverse
orientation at the transmitting end of a 2x2 MIMO system as a 2x1 MMC. This provides the relevant
Tx modal diversity by multiplexing two independent mode distributions into the MMF channel.

3.2 1x2 Multimode Coupler Characterisation

Multimode coupler characterisation was required to show that the MMC could provide the
necessary modal diversity required for use in a 2x2 mode multiplexed MIMO system, Figure 22. The
transmitter and receiver MMCs are the same hardware just positioned in reverse orientation w.r.t
each other in the 2x2 system. The 2x1 MMC at the transmitter provides modal diversity by
multiplexing Tx1 and Tx2 into the MMF channel on different mode distributions. Tx1 mainly excites

39
the low order modes of the MMF by transmitting into the Through Port of the MMC. Tx2 (Cross Port)
excites the higher order modes in the MMF as a result of evanescent-wave coupling in the MMC.
Before the receiver a 1x2 MMC de-multiplexes the two transmitted distributions within the MMF
into the two separate MMC output arms; reversing the operation of the 2x1 MMC at the transmitter.
The lower order modes pass through the 1x2 MMC Through port and the higher order modes couple
mainly into the cross port output of the 1x2 MMC. This provides modal diversity at the receiver
allowing each modal distribution to carry a separate stream of data.

If the modal diversity provided by the transmitter and receiver is sufficient, two separate data
streams can be multiplexed into the MMF channel and demultiplexed before the receiver. Digital
signal processing (DSP) algorithms are then used to reverse the MMF modal dispersion and modal
coupling effects to recover the transmitted data [10]. The system modal diversity is optimised by
minimising the coupling between modes and maximising the desired power for each mode.

Data1 Tx1 MMF Rx1 Data1


MIMO
2x1 MMC 1x2 MMC Digital
Signal
Processing
Data2 Tx2 Rx2 Data2

Figure 22 2 Input/2 Output MMF MIMO System

Figure 23 shows a diagram of the 2x1 MMC.

Port 3
(Through)
Port 1
Tx Rx
Port 4
(Cross)

Figure 23 2x1 Multimode Coupler

The MMC was characterised by launching directly into the 62.5µm MMC core from an 8µm core
Single Mode Fibre (SMF) core. The SMF was scanned across the MMC core, thus exciting different
modes in the MMC. This was achieved by positioning the adjacent ends of the SMF and the MMC
port under test on a translation stage. This allowed the SMF to be varied radially with respect to the
MMC port to an accuracy of 1µm. The SMF was scanned in the X and Y direction across the core
diameter of each MMC port. The power output from the other two MMC ports was monitored using
two power meters. The power readings from the two MMC ports were plotted against X and Y offset

40
launch position, respectively. This showed the power characteristics for each MMC port as a function
of input launch position. The three tests performed to characterise the MMC are shown in Figure 24.

A 1550nm optical source provided the transmit signal from the launch SMF.

a) Port 1 Launch b) Port 3 Launch


Port 3 Port 3
(Through) (Through)
PM1 Laser
Port 1 Port 1
Laser PM2
PM2 PM1
Port 4 Port 4
(Cross) (Cross)

c) Port 4 Launch Key


Port 3
(Through) PM - Power Meter
PM1
Port 1 - Multimode Fibre
PM2
Laser - Single mode Fibre
Port 4
(Cross) - Translation Stage

Figure 24 MMC Characterisation Tests

The naming convention used for the tests in Figure 24 a), b) and c) is as follows:

Port 1/3/4_Xscan/Yscan

Where Port 1/3/4 indicates the MMC port which is being launched into. Xscan/Yscan indicates
whether the launching SMF is being scanned in the X or Y direction across the MMC port.

The centre point for each port was taken as the SMF launch position that resulted in the maximum
received power in the relevant opposite MMC port, namely:

- Port 1 centre point was taken as the SMF launch position into Port 1 which gave the
maximum received power at Port 3.
- Port 3 centre point was taken as the SMF launch position into Port 3 which gave the
maximum received power at Port 1.
- Port 4 centre point was taken as the SMF launch position into Port 4 which gave the
maximum received power at Port 1.

41
3.2.1 MMC Characterisation: Test a)

Figure 25 and Figure 26 show the X and Y scan test results from the setup shown in Figure 24 a). The
received power at ports 3 and 4 of the MMC was measured as a function of SMF launch position.
Both figures show that, by controlling the SMF launch position, the MMC output power distribution
could be controlled. The launch position that gave the maximum received power on port 4 (Cross
Port) resulted in the minimum received power on Port 3 (Through Port), and vice versa. This was
true for both the X and Y scans. This showed the modal diversity property of the 1x2 MMC.

As the SMF offset was increased into the MMF cladding the received power dropped to -50dBm.
These points were not plotted since the results of interest were those within the MMC core.

An optical power of 2.93dBm was transmitted from the SMF.

X Scan:

A maximum power of -0.25dBm was coupled into Port 4 (Cross Port) at an X axis launch offset of -
0.02µm from the centre. The power received at Port 3 (Through Port) at that offset launch position
was -0.93dBm.

The maximum received Port 3 (Through Port) power was 0.98dBm at an X axis offset position of 0µm.
The power coupled into Port 4 (Cross Port) at 0µm offset was –3.52dBm.

Giving:

 Through Port modal diversity = 0.98dBm - (-0.93dBm) = 1.91dB


 Cross Port modal diversity = -0.25dBm - (-3.52dBm) = 3.27dB

Y Scan:

A maximum power of 0.2dBm was coupled into Port 4 (Cross Port) at a Y axis launch offset of -
0.012µm from the centre. The power received at Port 3 (Through Port) at that offset launch position
was -0.90dBm.

The maximum received Port 3 (Through Port) power was 1.57dBm at an X axis offset position of 0µm.
The power coupled into Port 4 (Cross Port) at 0µm offset was –3.11dBm.

Giving:

 Through Port modal diversity = 1.57dBm - (-0.9dBm) = 2.47dB


 Cross Port modal diversity = -0.2dBm - (-3.11dBm) = 2.91dB.

42
Figure 27 shows the total power received by Ports 3 and 4. The total power is approximately flat in
the region where the SMF was scanned across the Port 1 core. This shows that the loss is constant
for all launch positions i.e. suggesting MMC mode dependent loss is negligible.

0
Power (dBm)

-1

-2

-3

-4

-5
-0.035 -0.025 -0.015 -0.005 0.005 0.015 0.025 0.035
X Distance (mm)

Figure 25 Port 1_Xscan: Port 3 (Solid) Port 4 (Dashed)

0
Power (dBm)

-1

-2

-3

-4
-0.042 -0.032 -0.022 -0.012 -0.002 0.008 0.018 0.028 0.038
Y Distance (mm)

Figure 26 Port 1_Yscan: Port 3 (Solid) Port 4 (Dashed)

43
10

0
Total Received Power (dBm)

-10

-20

-30

-40

-50

-60
-0.100 -0.050 0.000 0.050 0.100

Offset Distance (mm)

Figure 27 Port 1_Xscan (Solid), Port 1_Yscan (Dashed): Total received power on MMC Port 3 and 4

Test a) Summary: This test has shown how the MMC can provide modal diversity when used in a 1x2
orientation. This orientation would be used at the receiving end of a 2x2 MIMO system to
demultiplex the two modes in the MMF channel to two separate outputs. The X and Y scans
exhibited similar modal diversity values. This showed that the MMC modal diversity is relatively
independent of the mode group orientation into the MMC input.

3.2.2 MMC Characterisation: Test b)

Figure 28 shows the test results from the setup shown in Figure 24b). The SMF was scanned in the X
and Y directions across Port 3 (Through Port) of the MMC. The received power at Port 1 of the MMC
was measured as a function of SMF launch position. The power curve is relatively flat for the central
30µm scan across the MMC Port 3, displaying 0.3dB and 0.5dB variation for the X scan and Y scan
respectively. Thus the majority of the power was received at Port 1 when transmitting in the 30µm
central area of the Port 3 core. This showed that the lower order modes excited by launching into
the centre of the Port 3 propagated mainly on the through path to Port 1.

The power measured at Port 4 when launching into Port 3 was mainly the result of back reflections
from the end surface of the MMC. These results were omitted since they were not relevant to
understanding the coupling properties of the MMC.

44
0

-1
0.0, -0.10
-2

-3

-4
Power (dBm)

-5

-6

-7

-8

-9

-10
-0.05 -0.03 -0.01 0.01 0.03 0.05

Offset Position (mm)

Figure 28 Port 3_Xscan (Solid), Port 3_Yscan (Dashed): Showing Port 1 Received power

3.2.3 MMC Characterisation: Test c)

Figure 29 shows the test results from the setup shown in Figure 24c). The SMF was scanned in the X
and Y directions across Port 4 (Cross Port) of the MMC. The received power at Port 1 of the MMC
was measured as a function of SMF launch position. A clear drop in received power was observed as
the SMF was scanned across the centre of Port 4 (Cross Port). This was approximately within the
central 20µm core diameter area. Hence power coupled most efficiently from the Cross Port to the
Through Port when the SMF was launched more than 20µm from the core centre of Port 4 (Cross
Port) until the cladding where the received power dropped again. This showed that the higher order
modes, excited by offsetting the SMF from the centre, coupled more efficiently than the Lower
Order modes from Port 4 (Cross Port) to the Through path (Port 1).

45
2
0.027, 0.41
0

-2
Power (dBm)

-4

-6
0.0, -5.45

-8

-10
-0.055 -0.035 -0.015 0.005 0.025 0.045
Offset Position (mm)

Figure 29 Port 4_Xscan (Solid), Port 4_Yscan (Dashed): Showing Port 1 Received Power

Tests b) & c) Summary: The power distributions in Figure 29 is wider than that in Figure 28, in terms
of SMF offset. Showing that a SMF can launch into Port 4 (Cross Port) and excite separate modes to
those excited by the Port 3 (Through Port) launch. These two different modal groups can then be
multiplexed into the MMF channel therefore maintaining modal diversity. Tests b) and c) have
shown how the MMC can provide this modal diversity when used in a 2x1 orientation. This
orientation would be used at the transmitting end of a 2x2 MIMO system to multiplex two modes
into a MMF independently of each other.

Assuming that the cross port excites different modes to the through port then modal diversity is
achieved in the 2x1 orientation.

Tests a), b) and c) in Figure 24 have shown how using a 2x1 MMC and a 1x2 MMC can both provide
modal diversity when tested separately. Using the two MMCs to create a 2x2 MIMO system
employing 2 mode multiplexing will be investigated in the following section.

46
3.3 Insertion Loss and Modal Diversity Definitions

Insertion Loss:

The system insertion loss and modal diversity will be used to define the MIMO performance for the
remainder of the report. The Cross and Through Port insertion losses were defined as:

Tx Power (From Cross


Cross Port Insertion Loss Rx Cross Port Power
= or Through Port) - (51)
(dB) (dBm)
(dBm)

Tx Power (From Cross


Through Port Insertion Loss Rx Through Port Power (52)
= or Through Port) -
(dB) (dBm)
(dBm)

The above insertion losses were calculated for the following two cases:

1. Cross Port Tx ON/ Through Port Tx OFF


2. Cross Port Tx OFF/ Through Port Tx ON

These two cases allow the calculation of the Cross and Through Port Modal Diversity.

The Total insertion loss was defined as:

Tx Power (From Cross Total Rx Through Port


Total Insertion Loss (dB) = or Through Port) - and Cross Port Power (53)
(dBm) (dBm)

Modal Diversity:

The modal diversity is defined as the ratio of wanted to unwanted modal power for each received
mode. The unwanted received modal power is a result of mode coupling and non-ideal launch and
receiver coupling. The non-ideal MMF channel, MMCs and connectors scatter the transmit signal
causing unwanted modes to couple into the receiving modal path. This reduces the modal diversity
for that specific modal path.

The 2x2 MIMO system modal diversity was defined for the Cross and Through Ports by:

Insertion Loss due to Insertion Loss due to


Cross Port Modal Diversity unwanted Rx Cross desired Rx Cross Port
= - Power (54)
(dB) Port Power
(dB) (dB)

47
Insertion Loss due to Insertion Loss due to
Through Port Modal Diversity unwanted Rx Through desired Rx Through Port
= - (55)
(dB) Port Power Power
(dB) (dB)

The desired and unwanted powers are shown in Figure 30 by the solid and dashed arrowed lines,
respectively.

Desired Through Port Power

Tx Through Port Tx1 Unwanted Through Rx1


Port Power

2x1 MMC 1x2 MMC

Unwanted Cross
Tx Cross Port Tx2 Port Power Rx2

Desired Cross Port Power

Figure 30 2x2 MIMO System using 2 mode multiplexing

The desired and unwanted powers were measured as follows:

Cross Port Tx ON and Through Port Tx OFF:

 Rx Cross Port power = Desired Cross Port Power


 Rx Through Port power = Unwanted Through Port Power

Cross Port Tx OFF and Through Port Tx ON:

 Rx Cross Port power = Unwanted Cross Port Power


 Rx Through Port power = Desired Through Port Power

Note: The Through Port Tx was attenuated resulting in equal Cross and Through Port Rx Powers.
This reduced the unwanted coupling from the Through Port Tx to Cross Port Rx giving
approximately equal modal diversities for each port. This attenuation was applied in MATLAB once
the data had been captured.

48
3.4 2x1 – 1x2 Multimode Coupler: 2x2 MIMO System

Section 3.2 showed that modal diversity was provided by the 2x1 and 1x2 MMCs. With this, a 2
Input/2 Output optical MIMO system exploiting 2 mode multiplexing was setup using a 2x1 MMC Tx
followed by a Rx 1x2 MMC connected by 660m of 62.5µm MMF, Figure 31. The modal diversity and
mode multiplexing capability of the 2x2 system will be discussed in this section.

A 3dB coupler was used at the transmitter to split the laser source power equally into two SMFs.
These were then inputted to the 2x1 MMC. Two translation stages were used to offset the two SMF
transmitters to the 2x1 MMC input ports. This excited two different modal distributions which were
transmitted into the MMF channel from the 2x1 MMC output port.

Each SMF offset launch position was optimised to transmit maximum power into the MMF. This was
achieved by adjusting the translation stage of one SMF with the other SMF input turned off and
measuring the received power at point A, Figure 31, then repeating for the other SMF input. The
optimal offset position was taken as the position that resulted in maximum received power at point
A. Launching at the optimal offset position for each SMF ensured that maximum power was coupled
into the two modal distributions at the transmitter. It was assumed that each maximum transmit
value corresponded to minimum coupling into the other mode distribution, from the tests in Section
3.2.

At the receiver two SMFs were positioned using two translation stages to couple power out of the
two MMF outputs of the 1x2 MMC. The position of the SMFs was set to give maximum coupling
from each MMC output to each SMF. Coupling the signal from the MMC outputs using SMF allows
the signals to be detected using a SMF coherent receiver. Therefore the phase and polarisation
information of the modal distributions can be recovered. This allows complex modulation formats to
be transmitted, resulting in higher channel capacity.

49
Key
PM - Power Meter - Multimode Fibre - Single mode Fibre - Translation Stage

Through Port Through Port


A PM1

3dB SMF Tx Rx
Laser
Splitter 2x1 MMC 1x2 MMC

PM2
Cross Port Cross Port

Figure 31 2x2 Optical MIMO System Test Setup

Having both the Tx Cross and Through Ports on simultaneously produced fluctuations in received
power. This was due to the beat term arising from the direct detection of the two signals, falling
within the electrical bandwidth of the power meter. In the full system the coherent receiver is used.
Therefore the low frequency beat term is shifted up to an optical frequency due to the heterodyne
function of the local oscillator. This lies outside the electrical bandwidth of the receiver and is
filtered out. Because of this, the test where both the Tx Cross and Through Ports were turned on will
not be discussed as it is not relevant to the coherently detected system that will be used.

Tx Cross Port ON and Through Port OFF:

Figure 32 shows the Cross and Through Port insertion loss when the transmitter Cross Port was ON
and the Through Port was off. The received Cross Port insertion loss fluctuated by 11.53dB over the
114 second measured time window. This would cause a variation in system capacity or a time
changing system Q-Factor if the system capacity is fixed [11].

The power distributions of the LPl,m modes with non-zero ‘l’ values exhibit azimuth minima. This is
seen from the LP modal distributions in Figure 19. The MMC Rx Cross Port couples higher order
modes such as these. Since the power from the MMC Rx Cross Port output is coupled into a SMF,
only a proportion of the total Cross Port output power is detected. Mechanical perturbations with
time cause rotations in the received modal distributions. As the modal distributions rotate, the
modal azimuth minima pass the SMF detection area resulting in a fluctuating received power. This
explains the power fluctuations observed.

50
Figure 32 Tx Through Port OFF/Tx Cross Port ON:
Rx Cross Port Insertion Loss (+) Rx Through Port Insertion Loss (o) Total Insertion Loss (-)

Tx Cross Port OFF and Through Port ON:

Figure 33 shows the results for the case where the transmitter Cross Port was OFF and the Through
Port was ON. The Through Port insertion loss was approximately constant at 5dB. This suggested
that the majority of Through Port power was contained in the centre of the MMF i.e. in the
Fundamental/LP0,1 mode. Any Through Port modal rotations would have pivoted about the fibre
centre point therefore had negligible effect on the received Through Port receive power since the
SMF coupled the power from the core centre.

51
Figure 33 Tx Through Port ON/Tx Cross Port OFF:
(+) Rx Cross Port Insertion Loss (o) Rx Through Port Insertion Loss (-) Total Insertion Loss

2x1 – 1x2 MMC Test Summary: The 2x2 MIMO system setup in Figure 31 has shown to provide
Cross and Through Port modal diversity. However, the insertion loss fluctuations prevent the tested
2x2 MIMO system operating reliably.

Worst case Through Port Modal Diversity = 35.31dB IL – 15.23dB IL = 20.08dB, Equation (55):

Worst case Cross Port Modal Diversity = 52.96dB IL - 37.87dB IL = 15.09dB, Equation (54):

However the Cross Port IL fluctuated by 11.53dB causing a fluctuation in modal diversity. A method
of controlling these Rx Cross Port fluctuations is required to maintain a constant modal diversity
value.

52
3.5 Optimisation of System Tx/Rx Coupling

This section focuses on optimising the system Tx/Rx coupling aiming to minimise the fluctuations in
received power observed in the previous experiment, Section 3.4. The following test arrangements
will be tested and discussed:

1. 1x2 Tx MMC to 1x4 Rx MMC


2. 1x4 Tx MMC to 1x4 Rx MMC
3. 1x2 Tx MMC to 1x3 SMF Arrangement/Free Space
4. SLM Tx Coupling

Each test aims to improve on the performance of the previous one. This led to the optimal
transmitter and receiver setup. The different tests are compared in the summary, Section 3.7, and
the optimal setup is chosen based on system trade-offs.

3.5.1 2x1 – 1x4 Multimode Coupler 2x4 MIMO System

A 2x4 system was setup using a 2x1 MMC at the transmitter and a 1x4 MMC at the receiver, shown
in Figure 34.

Key
PM - Power Meter - Multimode Fibre - Single mode Fibre - Translation Stage

Through Port Cross Port


A PM1
Cross Port
PM2
3dB SMF Tx
Laser PM3
Splitter 2x1 MMC Through Port
1x4 MMC
Rx Cross Port
PM4
Cross Port

Figure 34 2x4 MIMO System using a 1x4 coupler at the Receiver

53
Figure 35 and Figure 36 show the insertion loss for the Rx MMC outputs for the two Cross/Through
Port Tx cases.

Figure 35 2x1 MMC to 1x4 MMC: Cross Port Tx ON/ Through Port Tx OFF:
(+) Green Rx Cross Port Insertion Loss, (o) Blue Rx Cross Port Insertion Loss, (*) Orange Rx Cross Port Insertion Loss,
(·) Rx Through Port Insertion Loss, (-) Total Insertion Loss

54
Figure 36 2x1 MMC to 1x4 MMC System: Cross Port Tx OFF/ Through Port Tx ON:
(+) Green Rx Cross Port Insertion Loss, (o) Blue Rx Cross Port Insertion Loss, (*) Orange Rx Cross Port Insertion Loss,
(·) Rx Through Port Insertion Loss, (-) Total Insertion Loss

The 1x4 MMC was used at the receiver in place of the 1x2 MMC to reduce the unwanted Cross Port
power fluctuations observed in the previous experiment, Section 3.4.

The 1x4 MMC provided three Rx Cross Ports. As the modal distribution of the higher order modes
rotated and coupled into the different Cross Ports, the Cross Port powers could be averaged to give
a more constant Cross Port Rx power. This averaging would be carried out in the receiver based
MIMO DSP.

Figure 37 shows a cross section through the 1x4 MMC where the 4 multimode cores are brought
close together to induce coupling. The diagram shows how the high order LP11 mode couples into
the 3 cross ports as it rotates with time. The blue and red colours denote a π phase shift between
the two spots.

55
Cross Port Cores

MMC Cladding

62.5µm

Through Port Core

Figure 37 Diagram of rotating LP11 mode in the 1x4 MMC coupling region

The variation in mean insertion loss for different cases is displayed in Table 2.

No. Of MMC Cross Ports Variation in Mean Insertion


Averaged Across Loss (dB)
All 3 6.63
2: Green and Blue 6.57
2: Blue and Orange 9.72
2: Orange and Green 6.28
1: Green 6.29

Table 2 Mean IL fluctuations for different 1x4 MMC Cross Port measurement combinations

Table 2 shows that using the 1x4 MMC at the receiver minimised the Cross Port Rx fluctuations
compared to the previous experiment in Section 3.4. Averaging across the Orange and Green Cross
Port Rx outputs provided the minimum IL variation. However, this averaging technique provided
minimal benefit since a similar IL variation was provided by monitoring the Green Rx Cross Port
output directly. Namely 6.28dB variation compared with 6.29dB variation, respectively. Therefore
measuring the Green Rx Cross Port directly provided the best solution for this test setup.

56
The Cross Port (Green) and Through Port modal diversity for each of the above tests was:

Tx Cross Port: ON Tx Cross Port: OFF


Through Port: OFF Through Port: ON
MMC Output Max Insertion Loss Min Insertion Loss (dB) Worst Case Modal
Monitored (dB) Diversity (dB)
Green Cross Port 17.34 28.14 10.8

Table 3 Cross Port IL and Modal Diversity

Tx Cross Port: ON Tx Cross Port: OFF


Through Port: OFF Through Port: ON
MMC Output Min Insertion Loss (dB) Max Insertion Loss (dB) Worst Case Modal
Monitored Diversity (dB)
Brown Through Port 39.85 36.92 2.93

Table 4 Through Port IL and Modal Diversity

The Modal Diversity was calculated using Equation (54) and Equation (55):

Test Summary: Using a 1x4 MMC at the receiver improved the Cross Port power fluctuations by
5.24dB in comparison with the previous system that used a 1x2 MMC at the receiver, section 3.4.
This resulted in a Cross Port variation of 6.36dB. This was achieved by measuring the Green Rx Cross
Port Output. It was shown that averaging across numerous Cross Ports provided no added benefit in
minimising the Rx Cross Port power fluctuations.

Although the variations reduced, the Through and Cross Port modal diversity reduced by 9.31 and
12.23dB, respectively, in comparison with the previous test.

3.5.2 4x1 – 1x4 Multimode Coupler 2x4 System

In order to improve the modal diversity in the previous test, Section 3.5.1, a 4x1 MMC was installed
at the transmitter. This aimed to maximise the system Cross Port coupling by mirroring the
transmitter and receiver arrangement. Such that the modes coupled at the transmitter better
matched those coupled at the receiver end therefore increasing the modal diversity. The system
setup is shown in Figure 38.

57
Key
PM - Power Meter - Multimode Fibre - Single mode Fibre - Translation Stage

Cross Port
PM1
Cross Port
PM2
3dB SMF
Laser PM3
Splitter Through Port Through Port
4x1 MMC 1x4 MMC
Tx Rx Cross Port
PM4
Cross Port

Figure 38 2x4 MIMO System using a 1x4 coupler at the Transmitter and Receiver

Figure 39 and Figure 40 show the IL at the Rx MMC outputs for the cases, as in the previous tests:

1. One Tx Cross Port ON/ Tx Through Port OFF


2. Tx Cross Ports OFF/ Tx Through Port ON

Table 5 and Table 6 show the insertion loss for the Cross and Through Ports and the modal diversity
measured for each. The Blue Rx Cross Port offered the best Cross Port modal diversity = 29.54dB.
The IL variation for this port = 3.57dB. This is an improvement on the previous system using the 2x1
MMC at the transmitter.
Tx Cross Port: ON Tx Cross Port: OFF
Through Port: OFF Through Port: ON
MMC Output Max Insertion Loss Min Insertion Loss (dB) Worst Case Modal
Monitored (dB) Diversity (dB)
Green Cross Port 43.61 65.93 22.32
Blue Cross Port 34.79 64.33 29.54
Orange Cross Port 42.56 57.81 15.25

Table 5 4x1 – 1x4 Setup: Modal Diversity for the Cross Ports

Tx Cross Port: ON Tx Cross Port: OFF


Through Port: OFF Through Port: ON
MMC Output Min Insertion Loss (dB) Max Insertion Loss (dB) Worst Case Modal
Monitored Diversity (dB)
Brown Through Port 43.4 22.17 21.23

Table 6 4x1 - 1x4 Setup: Through Port Modal Diversity

58
Figure 39 4x1 MMC to 1x4 MMC System: Cross Port Tx ON/ Through Port Tx OFF:
(+) Green Rx Cross Port Insertion Loss, (o) Blue Rx Cross Port Insertion Loss, (*) Orange Rx Cross Port Insertion Loss,
(·) Rx Through Port Insertion Loss, (-) Total Insertion Loss

Figure 40 4x1 MMC to 1x4 MMC System: Cross Port Tx OFF/ Through Port Tx ON:
(+) Green Rx Cross Port Insertion Loss, (o) Blue Rx Cross Port Insertion Loss, (*) Orange Rx Cross Port Insertion Loss,
(·) Rx Through Port Insertion Loss, (-) Total Insertion Loss

59
4x1 – 1x4 MMC Test Summary: Using a Tx 4x1 MMC and a Rx 1x4 MMC improved the Cross Port
power fluctuations by 6.29dB - 3.57dB = 2.72dB compared to the previous test using the 1x2 MMC at
the transmitter.
 Measuring the Blue Rx Cross directly provided 29.54dB modal diversity.
 The Through Port provided approximately 21.23dB of modal diversity.
 The Cross Port IL variation was 3.57dB.

3.5.3 Free Space Receiver Coupling Method

A 2x3 system was setup using a 2x1 MMC at the transmitter and 3 lensed SMFs at the receiver. The
transmitter and receiver ends were separated by 660m of MMF, a biconvex lens and an air gap of
0.45m, Figure 41. The aim of the free space method was to achieve improved coupling into the
individual SMF outputs thus reduce the coupling losses associated with the receiving MMC and
improve the modal diversity measured in the previous tests. A biconvex lens was positioned after
the MMF output. Ideally the light should have been collimated but this resulted in a spot size too
small to be detected by all 3 SMFs simultaneously. Therefore the lens was positioned at a distance
greater than the back focal length causing the optical beam to diverge to image the optical power
from the MMF onto the diameter of the 3 SMF lenses at the receiver. The 3 lensed SMFs were
placed in close proximity to minimise the beam divergence required therefore capturing the
maximum power possible.

Key
PM - Power Meter - Multimode Fibre - Single mode Fibre

- Lens - Optical Path - Lensed SMF

Through Port

PM1
3dB SMF Tx
Laser PM2
Splitter 2x1 MMC
Rx PM3

Cross Port

Figure 41 2x3 Optical MIMO System Test Setup

The Through Port and Cross Port transmit power was optimised in the same way as for the previous
tests. The three SMFs were separate from each other. Therefore they could be positioned in their
individual optimal positions. This was achieved by placing each SMF on a precision positioning stand
which gave control over the x, y, z and angular positions of each receiver. The lens was positioned at

60
the back focal length with the Through Port switched on, producing a collimated beam. The Through
Port SMF was aligned with the beam to give maximum received power, Figure 42. The remaining two
SMFs were placed above the Through Port receiver SMF creating a triangular arrangement. These
functioned as the two Cross Port SMF receivers across which the power was averaged to reduce the
Cross Port power fluctuations. With the Cross Port ON and Through Port OFF, the lens was brought
closer to the MMF output causing the beam to diverge until the received beam covered all three
SMF receivers, Figure 43. The lens position was then fine-tuned to maximise the received power at
the two Cross Port SMF receivers. This method resulted in the bottom SMF being aligned with the
centre of the beam and the top two SMFs collecting the maximum power from the outer core power
distribution, where mainly the higher order modal power was contained.

Top Left Cross Port SMF Top Right Cross Port


Receiver
SMF Receiver
Bottom Through Port SMF
Receiver
Figure 42 Direct View: SMF Receivers illuminated by the collimated Through Port beam

Top Left Cross Port SMF Top Right Cross Port


Receiver SMF Receiver
Bottom Through Port SMF
Receiver

Figure 43 Direct view into the SMF Receivers illuminated by the diverged beam

The power was monitored at the three SMF receivers over a 380 second time window. Figure 44 and
Figure 45 show the IL at the Rx SMF outputs for the cases, as in the previous tests:

1. One Tx Cross Port ON/ Tx Through Port OFF


2. Tx Cross Ports OFF/ Tx Through Port ON

61
For case 1, using the Cross Port average reduced the Cross Port IL variation. This is shown in Table 7
along with the individual Cross Port IL values:
MMC Output IL Variation (dB)
Monitored
Top Left Cross Port 8.11
Top Right Cross Port 4.8
Cross Port Average 3.3
Table 7 Cross Port IL Variation

Table 8 and Table 9 show the insertion loss for the Cross and Through Ports and the modal diversity
demonstrated by each:
Tx Cross Port: ON Tx Cross Port: OFF
Through Port: OFF Through Port: ON
MMC Output Max Insertion Loss Min Insertion Loss (dB) Worst Case Modal
Monitored (dB) Diversity (dB)
Cross Port Average 42.9 54.64 11.74

Table 8 Modal Diversity for the Cross Ports

Tx Cross Port: ON Tx Cross Port: OFF


Through Port: OFF Through Port: ON
MMC Output Min Insertion Loss (dB) Max Insertion Loss (dB) Worst Case Modal
Monitored Diversity (dB)
Through Port 31.06 16.84 14.22

Table 9 Through Port Modal Diversity

62
Figure 44 2x1 MMC to 1x3 Free Space System: Cross Port Tx ON/ Through Port Tx OFF:
(Light Blue: +) Top Left SMF Cross Port IL, (Dark Blue: O) Top Right SMF Cross Port IL,
(Red: ·) Bottom SMF Through Port IL, (Black: -) Total IL

Figure 45 2x1 MMC to 1x3 Free Space System: Cross Port Tx OFF/ Through Port Tx ON:
(Light Blue: +) Top Left SMF Cross Port IL, (Dark Blue: O) Top Right SMF Cross Port IL,
(Red: ·) Bottom SMF Through Port IL, (Black: -) Total IL

63
2x1 MMC to Free Space Rx Test Summary: The free space receiver method improved the Cross Port
IL variations seen in the previous experiments which used a Rx MMC to provide modal diversity.
Averaging the Cross Port values resulted in lower IL variations. This Cross Port IL variation was 3.3dB;
a reduction of 0.27dB compared to the previous experiment that used a Tx 4x1 MMC and a Rx 1x4
MMC. The measured modal diversity for this test was:
 Cross Port Modal Diversity = 11.74dB
 Through Port Modal Diversity = 14.22dB

64
3.6 Spatial Light Modulator Transmitter Coupling Technique

The following section will compare the previous experimental results with a technique using a
Spatial Light Modulator (SLM) in place of the Tx MMC. The SLM/Hologram controls the modal launch
profile in to the MMF channel, therefore controlling the mode(s) excited in the MMF. A SLM with a
256x256 pixel array was used. Each pixel caused a 0 or π binary phase shift on the incident light. By
setting the phase shift of each pixel, a custom power distribution was generated which excited the
desired mode(s) in the MMF. The SLM pixel matrix was controlled in Software.

The following experiments used a SLM at the transmitter to launch into 2km of MMF. The MMF
output was connected to a MMC. The 1x2 and 1x4 Multimode Couplers from the previous
experiments were used in turn. The SLM cycled through a library of predefined pixel matrix values
which resulted in launching the modes in Table 10. Each mode was cycled through 16 rotations.

Library of LPl,m modes in order from Lowest Order -> Highest Order
0,1 1,1 0,2 2,1 1,2 3,1 0,3 2,2 4,1 1,3
^ Lowest Order/ Fundamental Mode   
3,2 5,1 0,4 2,3 4,2 6,1 1,4 3,3 5,2 7,1
    
0,5 2,4 4,3 6,2 8,1 1,5 3,4 5,3 7,2 9,1
   Highest Order Mode in Library ^

Table 10 LPl,m modes in the order they were cycled through

The received power corresponding to each launch profile was measured and plotted. The results are
discussed in the following section.

3.6.1 1x2 MMC Characterisation using SLM Technique

The test setup for the SLM launch into the 1x2 MMC after 2km of MMF is shown in Figure 46. A SMF
was centre butt-coupled to the Through Port output. This provided modal filtering to suppress the
higher order modes giving a purer low order modal output. This resulted in greater modal diversity
at the Through Port receiver. The Cross Port output was measured directly at Power Meter 2.

65
Waveplate
PBS
256x256
SLM
Laser

Lens

Lens

SLM Driver
2km MMF
Software

Through Port Cross Port


Rx
SMF 1x2 MMC
PM2
PM1

Figure 46 1x2 MMC Characterisation using the SLM launch. PBS: Polarisation Beam Splitter

Figure 47 and Figure 48 show the Cross Port and Through Port insertion loss for each launched mode
profile.

Figure 47 SLM to 1x2 MMC System: IL for different launched modes


(Green: +) Cross Port IL, (Blue: ·) Through Port IL

66
Figure 48 shows a zoomed in version of the low order modes in Figure 47.

Figure 48 ZOOMED IN: SLM to 1x2 MMC System: IL for different launched modes
(Green: +) Cross Port IL, (Blue: ·) Through Port IL

Transmitting the lowest order mode LP01 resulted in a Through Port insertion loss of 3.23dB and a
Cross Port insertion loss of 9.26dB. The Cross Port insertion loss was lowest between the 5th and 8th
dashed Green lines in Figure 46. This region corresponds to the LP mode range from LP4,1 to LP7,1 in
Table 10. The minimum Cross Port insertion loss in the flat region was 2.048dB achieved by
transmitting a rotated version of mode LP1,4. This resulted in a Through Port insertion loss of 30.66dB.
Giving:

 Cross Port Modal Diversity = 9.26dB – 2.048dB = 7.21dB


 Through Port Modal Diversity = 30.66 – 3.23dB = 27.43dB

As the mode order of the launch profile was increased, the Through Port insertion loss increased.
This suggested that the higher order modal power was filtered out by the SMF at the Through Port
output, as desired. However, the Cross Port insertion loss decreased and remained relatively flat
until the higher order modes. This suggested that higher order modes, beyond a certain point, began
to couple into the continuum of radiation modes, resulting in a loss of signal power.

SLM Launch – 1x2 MMC Test Summary: Using the SLM method to launch into the 1x2 MMC showed
a worst case Cross Port and Through Port Modal diversity of 7.21dB and 27.43dB respectively. This
experiment aimed to show the effect of launching specific modes into the MMF channel to provide
spatial diversity. Insertion loss variation was not measured. To measure these fluctuations, the
optimal launch profiles for the Through and Cross Ports should be launched and measured over a

67
time window with a SMF coupling from the receiver MMC. This would provide a comparison
between the SLM and MMC Tx techniques with respect to the control of power fluctuations caused
by the Rx SMF output coupling.

Using a SLM at the receiver to provide spatial diversity could allow the two modal distributions to be
separated into their individual data streams. This would effectively allow for MMF coherent
detection thus eliminating the fluctuations observed when coupling from the MMF with a SMF [35].

Another advantage of using an SLM to provide spatial diversity is the ability to adapt the launch
profile into the MMF to deal with time varying MMF channel. This can alleviate the complexity
required in the MIMO DSP computation. This can potentially result in faster MIMO DSP convergence
and lower power consumption.

3.6.2 1x4 MMC Characterisation using SLM Launch

The Rx 1x2 MMC in the previous test was replaced with a 1x4 MMC and the output ports were
monitored as the library of SLM mode profiles were launched into the MMF. The results for all four
outputs are displayed in Figure 49 to Figure 52. The 4 MMC outputs were measured over two tests:

1. Through Port and Orange Cross Port


2. Blue Cross Port and Green Cross Port

These results were plotted on two graphs, below.

Through Port (Brown) and Cross Port (Orange):

Launching the lowest order mode LP01 resulted in a Through Port insertion loss of 6.45dB and a Cross
Port insertion loss of 24.22dB. The Cross Port insertion loss was lowest between the 4th and 8th
dashed Green lines in Figure 46. This region corresponds to the LP mode range from LP3,1 to LP7,1,
shown in Table 10. A Cross Port insertion loss of 10.79dB was observed when transmitting a rotated
version of mode LP1,4. This resulted in a Through Port insertion loss of 40.67dB.

Giving:

 (Orange) Cross Port Modal Diversity = 24.22dB – 10.79dB = 13.43dB


 (Brown) Through Port Modal Diversity = 40.67 – 6.45dB = 34.22dB

68
Figure 49 SLM to 1x4 MMC System: IL for different launched modes
(Orange: +) Cross Port IL, (Blue ·) Through Port IL

Figure 50 ZOOMED IN: SLM to 1x4 MMC System: IL for different launched modes
(Orange: +) Cross Port IL, (Blue: ·) Through Port IL

69
Blue Cross Port and Green Cross Port:

The Blue and Green Cross Port insertion losses were 29.97dB and 25.78dB for the LP0,1 mode
transmission, respectively. The maximum Cross Port insertion losses in the flat region, as in the
previous case, were 10.75dB and 10.73dB for the Blue and Green outputs respectively.

Giving:

 (Blue) Cross Port Modal Diversity = 24.22dB – 10.75dB = 13.47dB


 (Green) Cross Port Modal Diversity = 24.22dB – 10.73 dB = 13.49dB

(These results used the Through Port values from the previous test, Figure 49).

Figure 51 SLM to 1x4 MMC System: IL for different launched modes


(Green: +) Green Cross Port IL, (Blue ·) Blue Cross Port IL

70
Figure 52 ZOOMED IN: SLM to 1x4 MMC System: IL for different launched modes
(Green: +) Green Cross Port IL, (Blue: ·) Blue Cross Port IL

SLM Launch – 1x4 Test Summary: Using the SLM method to launch into the 1x4 MMC resulted in an
improved Cross and Through Port modal diversity in comparison to the Free Space method of modal
recovery, shown in Table 11, below.

71
3.7 Summary: Tx/Rx Coupling Optimisation

A number of test setups have been compared in the previous sections. They all aimed to optimise
the 2x2 MIMO system transmitter and receiver coupling. Cross Port IL fluctuations were observed
when coupling from the MMF output with a SMF to allow coherent detection. Each test setup aimed
to reduce this effect whilst increasing the modal diversity for both ports. The results summary for all
the test setups is shown in Table 11.

The smallest Insertion Loss fluctuation was observed with the 2x1 MMC to 1x3 SMF Free Space
arrangement. To achieve this, 2 Cross Ports were averaged across. This averaging function increases
the complexity of the MIMO DSP, requiring a 2x3 equaliser coefficient matrix.

However, it can be seen that for an IL fluctuation increase of 0.27dB, better system modal diversity
can be achieved using a 4x1 Tx MMC followed by a 1x4 Rx MMC system arrangement. This system
also provides less complex DSP than the setup using the 1x3 Free Space Rx. Therefore trading off the
IL fluctuation increase for improved modal diversity and DSP complexity resulted in the optimal
system setup being: 4x1 Tx MMC followed by the 1x4 Rx MMC (Highlighted in Table 11).

Worst Case Modal Cross Port Worst Case Insertion


Diversity (dB) Insertion Loss Loss (dB)
Experimental Setup
Fluctuations
Through Cross Through
Cross Port (dB)
Port Port Port
2x1 MMC -> 1x2 MMC 15.09 20.08 11.53 41.45 15.23
2x1 MMC -> 1x4 MMC 10.8 2.93 6.29 36.92 17.34
4x1 MMC -> 1x4 MMC 29.54 21.23 3.57 34.79 22.17
2x1 MMC -> 1x3 (Free Space) 11.74 14.22 3.3 44.52 16.84
SLM -> 1x2 MMC 7.21 27.43 N/A 2.23** 3.23**
SLM -> 1x4 MMC 13.43* 34.22 N/A 10.79** 6.454**
*
Observed for the Orange MMC Cross Port Output
**
IL values don’t account for SLM device loss. System input taken as the output from the SLM.

Table 11 Results Summary of all Test Setups

The worst case insertion loss values have also been presented. The 4x1 to 1x4 MMC setup provides
the lowest worst case IL of all the MMC setups = 34.79dB. For this value of IL, assuming the signal
remains above the receiver noise floor, the signal can be detected. Amplification may be required to
achieve this.

72
Improved insertion loss was observed using the SLM launch technique compared with the MMC at
the transmitter to excite the desired modes in the MMF channel. An improvement of approximately
20dB** and 30dB** was observed for the SLM Tx into the 1x4 MMC and 1x2 MMC, respectively. To
investigate the SLM launch method further, the transmitted modes resulting in the best Cross and
Through Port modal diversity should be transmitted. Running this test over time would show
whether the SLM launch method results in lower IL fluctuations compared with the other test setups.

73
4 Conclusion
Through MATLAB simulation this work has shown that a COMIMO system utilising Polarisation and 2
Mode Multiplexing over multimode fibre is achievable. The following unwanted channel effects
were simulated:

 Polarisation Rotation
 Polarisation Phase Shift
 Mode Coupling

It was shown that receiver based MIMO DSP was able to undo these channel effects and recover the
transmitted data. An adaptive blind equaliser updated by the Constant Modulus Algorithm was used
in the DSP to achieve this. Simulation results showed that a channel condition number <4.2dB was
required for the DSP to correctly recover the distorted data. The equaliser convergence rate was
optimised to provide optimal tracking of the unwanted channel effects.

This work also experimentally demonstrated the optimum transmitter and receiver coupling for a
2x2 COMIMO system exploiting 2 mode and polarisation multiplexing over MMF. This has provided
the preparation required to realise a state of the art polarisation multiplexed COMIMO system to
enhance the capacity of MMF, never before been published.

Initially, a 2x1 MMC was employed at the transmitter and a 1x2 MMC was employed at the receiver
to provide the system spatial diversity. It was shown that this setup caused unwanted fluctuations in
the received power as a result of coupling into the SMF coherent receiver. Therefore a number of
setups were tested to optimise the transmitter and receiver coupling to reduce this power
fluctuation. It was shown that a Tx 4x1 MMC followed by a 1x4 Rx MMC test setup provided the best
trade-off between modal diversity and receiver power fluctuations. This arrangement also provided
the minimum worst case system insertion loss of all the tested MMC arrangements.

Experiments using a SLM to provide controlled mode launching into a Rx MMC were also
investigated. This technique provided sufficient modal diversity, improved mode coupling efficiency
and system insertion loss. A benefit of using a SLM to provide spatial diversity is its ability to adapt
the mode launch profile, therefore adapt to channel variations. This could potentially alleviate the
MIMO DSP complexity; in turn reduce the system power consumption.

74
5 Further Work
Further work would use the 4x1 MMC to 1x4 MMC optimised system arrangement to transmit
complex data over a 2x2 Mode Multiplexed COMIMO System. Exploiting both polarisation states for
each mode would allow the transmission of DP-QPSK modulated data to further increase the
capacity. Applying offline MIMO DSP, discussed in this work, would recover the transmitted data.

Investigating the use of a SLM at the receiver end to provide spatial diversity could be beneficial. The
SLM is able to map the MMF mode power into a SMF to allow coherent detection. This provides
more efficient receiver coupling and could reduce the losses and power fluctuations observed when
using a SMF to couple out of the receiver MMC.

75
6 References
[1] (2011, June) Cisco Visual Networking Index: Forecast and Methodology, 2010-2015. [Online].
http://www.cisco.com/en/US/solutions/collateral/ns341/ns525/ns537/ns705/ns827/white_pa
per_c11-481360.pdf

[2] X Zhou et al., "64-Tb/s, 8b/s/Hz, PDM-36QAM Transmission Over 320 km Using Both Pre- and
Post-Transmission Digital Signal Processing," Journal of Lightwave Technology, vol. 29, no. 4,
pp. 571-577, February 2011.

[3] R J Essiambre, G Kramer, P Winzer, G Foschini, and B Goebel, "Capacity Limits of Optical Fiber
Networks," Journal of Lightwave Technology, vol. 28, no. 4, pp. 662-701, February 2010.

[4] B Zhu et al., "70-Gb/s Multicore Multimode Fiber Transmissions for Optical Data Links," IEEE
Photonics Technology Letters, vol. 22, no. 22, pp. 1647-1649, November 2010.

[5] T Hayashi, T Taru, O Shimakawa, T Sasaki, and E Sasoaka, "Low-Crosstalk and Low-Loss Multi-
Core Fiber Utilizing Fiber Bend," in OSA/OFC/NFOEC, 2011.

[6] R Hsu, A Tarighat, A Shah, A Sayed, and B Jalali, "Capacity Enhancement in Coherent Optical
MIMO (COMIMO) Multimode Fiber Links," IEEE Communications Letters, vol. 10, no. 3, pp. 195-
197, March 2006.

[7] L Raddatz, I White, D Cunningham, and M Nowell, "Increasing the bandwidth-distance product
of multimode fibre using offset launch," Electronics Letters, vol. 33, no. 3, pp. 232-233, January
1997.

[8] Y Ma, Y Tang, and W Shieh, "107 Gbit/s transmission over multimode fibre with coherent
optical OFDM using centre launching technique," Electronics Letters, vol. 45, no. 16, July 2009.

[9] H R Stuart, "Dispersive multiplexing in multimode fiber," Science, vol. 289, pp. 281-283, July
2000.

[10] B C Thomsen, "MIMO enabled 40 Gb/s transmission using mode division multiplexing in
multimode fiber," in OSA/OFC/NFOEC, 2010.

[11] B Franz, D Suikat, R Dischler, F Buchali, and H Buelow, "High Speed OFDM Data Transmission
Over 5 km GI-Multimode Fiber Using Spatial Multiplexing With 2x4 MIMO Processing," ECOC,

76
vol. Tu.3.C.4, September 2010.

[12] M Salsi et al., "Transmission at 2x100Gb/s, over Two Modes of 40km-long Prototype Few-Mode
Fiber, using LCOS-based Mode Multiplexer and Demultiplexer," in OSA/OFC/NFOEC, 2011.

[13] A Li, A Amin, X Chen, and W Shieh, "Reception of Mode and Polarisation Multiplexed 107-Gb/s
CO-OFDM Signal over a Two-Mode Fiber," in OSA/OFC/NFOEC, 2011.

[14] R Ryf et al., "Space-division multiplexing over 10km of three-mode fiber using coherent 6 x 6
MIMO processing," in OSA/OFC/NFOEC, 2011.

[15] A Shah, R Hsu, A Tarighat, A Sayed, and B Jalali, "Coherent Optical MIMO (COMIMO)," Journal
of Lightwave Technology, vol. 23, no. 8, August 2005.

[16] S Haykin, Adaptive Filter Theory, 4th ed., M J Horton, Ed. Upper Saddle River, NJ: Prentice Hall,
2002.

[17] C Johnson Jr et al., "Blind Equalisation Using the Constant Modulus Criterion: A Review,"
Proceedings of the IEEE, vol. 86, no. 10, pp. 1927-1950, October 1998.

[18] P Diniz, Adaptive Filtering Algorithms and Practical Implementation, 3rd ed. USA: Springer,
2008.

[19] D Godard, "Self-Recovering Equalization and Carrier Tracking in Two-Dimensional Data


Communication Systems," IEEE Transactions on Communications, vol. COM-28, no. 11, pp.
1867-1875, November 1980.

[20] S J Savory, "Digital Coherent Optical Receivers: Algorithms and Subsystems," IEEE Journal of
Selected Topics in Quantum Electronics, vol. 16, no. 5, pp. 1164-1179, September 2010.

[21] A Leven, N Kaneda, U Koc, and Y Chen, "Frequency Estimation in Intradyne Reception," IEEE
Photonics Technology Letters, vol. 19, no. 6, pp. 366-368, March 2007.

[22] S Haykin, Unsupervised Adaptive Filtering Volume II Blind Deconvolution, S Haykin, Ed. Canada:
John Wiley & Sons, 2000.

[23] I Fijalkow, F Lopez de Victoria, and C Johnson Jr, "Adaptive Fractionally Spaced Blind
Equalization," IEEE Signal Processing Workshop, pp. 257-260, October 1994.

77
[24] Y Li and Z Ding, "Global Convergence of Fractionally Spaced Godard (CMA) Adaptive Equalizers,"
IEEE Transactions on Signal Processing, vol. 44, no. 4, pp. 818-826, April 1996.

[25] S Kun and Z Xudong, "A SE-CMA based blind equalisation for MIMO systems," in ICSP, 2004, pp.
1674-1677.

[26] S Yao, "Polarization in Fiber Systems: Squeezing out More Bandwidth," The Photonics
Handbook, 2003.

[27] M B Shemirani, W Mao, R Panicker, and J Kahn, "Principal Modes in Graded-Index Multimode
Fiber in Presence of Spatial- and Polarisation-Mode Coupling," Journal of Lightwave
Technology, vol. 27, no. 10, pp. 1248-1261, May 2009.

[28] S Bottacchi, Multi-gigabit Transmission Over Multimode Optical Fibre. Chichester, England: John
Wiley & Sons, 2006.

[29] D Marcuse, Theory of Dielectric Optical Waveguides, 2nd ed., P Liao and P Kelley, Eds. San
Diego, CA: Academic Press, 1991.

[30] J M Senior, Optical fiber communications, 2nd ed. London: Prentice-Hall, 1985.

[31] A Snyder and J Love, Optical Waveguide Theory. London: Chapman and Hall, 1983.

[32] D Marcuse, "Coupled Mode Theory of Round Optical Fibers," The Bell System Technical Journal,
vol. 52, no. 6, pp. 817-842, July-August 1973.

[33] R Paschotta. (2009, April) Encyclopedia of Laser Physics and Technology. [Online].
http://www.rp-photonics.com/multimode_fibers.html

[34] L Raddatz, I White, D Cunningham, and M Nowell, "Influence of Restricted Mode Excitation on
Bandwidth of Multimode Fiber Links," IEEE Photonics Technology Letters, vol. 10, no. 4, pp. 534-
536, April 1998.

[35] J Carpenter and T D Wilkinson, "Holographics Mode-Group Division Multiplexing," in


OSA/OFC/NFOEC, 2011.

[36] P Winzer and G Foschini, "Outage calculations for spatially multiplexed fiber links," in
OSA/OFC/NFOEC, 2011.

78
[37] S J Savory, "Digital filters for coherent optical receivers," Optics Express, vol. 16, no. 2, pp. 804-
817, 2008.

[38] P L Neo, J P Freeman, and T D Wilkinson, "Modal Control of a 50µm core diameter Multimode
Fiber Using a Spatial Light Modulator," OFC/NFOEC, March 2007.

79
7 Appendix

7.1 Derivations

7.1.1 Mode solutions for Step Index Multimode Fibre

Derivation of the scalar wave equation from Maxwell’s equations:

and are the dielectric permittivity and the magnetic permeability of the medium, respectively.

Taking the curl of and gives:

Scalar Wave Equation:

Where is the or field solution. is the core refractive index. k is the propagation constant of
light in a vacuum. and are the fibre cylindrical coordinates. E(r) is the radially dependent field
amplitude described by a different Bessel function for each mode (in the fibre core region). Note the
periodic variation of around 2π which is dependent on corresponding to the number of field
maxima around the circumference of the core.

80
7.1.2 Mode Coupling in terms of Condition Number

Ideal mode coupling is represented by the identity matrix:

(56)

Introducing coupling terms into the zero elements gives:

The two singular values, Smax and Smin, of M are given by:

The linear condition number of M is:

(Note: This is the case for equal coupling ratios between the two spatial modes)

The condition number in dB, . Therefore:

81
7.2 MATLAB Code

7.2.1 Example Script to plot 3D Convergence Time Graphs


%% Example Code for Multichannel DP-QPSK Transmission

% DP-QPSK MIMO System:


% Models Channel Mode and Polarisation coupling effects

% Modified: W.Hughes July 2011

%% Transmitter and signal parameters


P.Fb=10; % Bit rate (Gbit/s)
P.Ns=16; % Number of samples per bit
P.PattLength=14; % PRBS pattern length

P.PdBm=0; % Signal average power (dBm)


P.ERdB=9; % Extinction ratio in dB (omit for ideal case)
P.n=5; % Filter order
P.GHzBW=0.7*P.Fb; % Filter 3dB bandwidth (GHz)
P.CenWave=1550e-9; % Central Wavelength (m)

P.Np=2; % Number of Polarisations


P.sm=2; % Number of spatial modes utilised
P.ModFormat=2; % Number of bits/symbol i.e. QPSK: P.ModFormat=2
P.Segments=1; % Number of coupling segments that make up the MMF

P.Nb=(2^P.PattLength)-rem((2^P.PattLength),(P.sm*P.Np*P.ModFormat));
% Number of bits approximately 2^14 depending on number of spatial modes utilised. Nb is a
multiple of (P.sm x P.Np x P.ModFormat). P.Np and P.Modformat are constant since assumed DP-
QPSK application in all cases.

[Sig data]=MIMO_IdealQPSKSignal(P);

Sig=BesselLowPassFilter(Sig,P);
Sig=TripleMZITransfer(Sig,P);
Sig=SetPower(Sig,P);
%figure(1),plot(Sig.Et(1,:),'.') % Display the constallation diagram

%% Booster EDFA
Booster.GdB=18; % amplifier gain (dB)
%Booster.NFdB=6;
OXC.AdB=Booster.GdB;

SigOut=MIMO_OpticalAmplifier(Attenuator(Sig,OXC),Booster);
[Npm,Nt] = size(SigOut.Et);

OSASetting.Resolution=12.5; % OSA Resolution (GHz)


OSASetting.Points=2^10; % Number of points to display
%figure(2),OSA(SigOut,OSASetting) % Display the optical spectrum (with a resolution of 0.1nm)

82
P.SigBW = 12.5; % Signal bandwidth (GHz)
P.NoiseBW = 12.5; % Noise bandwidth (GHz) For OSNR function
P.NoiseCentre = 50; % Noise centre (GHz) For OSNR function
% osnr=OSNR(SigOut,P);
%disp(['Signal Transmitter Power ' num2str(P.PdBm) 'dBm. OSNR ' num2str(osnr.OSNR_OSA) ' dB'])

%% Non-Ideal Channel
WhileBreak=0; % While loop initialisation
ThetaStep=3; % No. of increments between 0 and 90 degrees for theta and phi
ConvTime=zeros(2*ThetaStep+1,2*ThetaStep+1); % Initialise Convergence Time Matrix
% CoRx.RandPhase = randn(1,Nt); % Random LO Phase Noise generation

% for m=1:ModeStep;
% P.ModeCoupleStep=m;
P.ModeCoupleStep=100;

for i=-1*ThetaStep:ThetaStep;
for j=-1*ThetaStep:ThetaStep;
P.PolTheta=(pi/180)*(90/(ThetaStep))*(j)*ones(1,P.sm); % Polarisation Rotation
applied to both pols (degrees)
P.PhiShift=(pi/180)*(90/(ThetaStep))*(i)*ones(1,P.sm); % Polarisation Phase Shift
between X and Y pols w.r.t X pol (degrees)

count=0;

while WhileBreak==0
P.ModeCoupling=0; % 'yes'=1, 'no'=0 for mode coupling
for k=1:P.Segments;
[SigOut.Et]=MIMO_MMFChannel(SigOut,P);
end

%% Coherent Receiver
% Optical Hybrid
CoRx.Rxtype = 'OpticalHybridDifferential'; % 'AsymmetricCoupler' 'Ideal'
'OpticalHybridSingle' 'OpticalHybridDifferential'
% CoRx.Linewidth = 1; % Linewidth in MHz
CoRx.LOoffset = 0; % Local Oscillator Offset (GHz)
CoRx.SigLO_Ratio = 20; % Signal to LO ratio in dB
CoRx.sm = P.sm; % No. of Spatial Modes Utilised
CoRx.Np = P.Np; % No. of Polarisations

% Electrical filtering
CoRx.n=5; % Filter order
CoRx.BW=0.7*SigOut.Fb; % Filter 3dB bandwidth (Hz)

% CMA Equaliser
CoRx.mu = 1e-3;%[1e-3,2e-3,6e-3]; % Step Size
CoRx.N = 12; % Equaliser length M(2N+1)+1
CoRx.M=2; % Equaliser fractional spacing
CoRx.Loops=1; % Number of passes through data
sequence for convergence

83
% Carrier Phase Estimation
CoRx.NViterbi = 10; % Half width of phase estimation
window: integer of form 2N
CoRx.Figures=1;

if count==1
ConvTime(i+ThetaStep+1,j+ThetaStep+1)=NaN; % Convergence time of equaliser in
terms of number of symbols in to the data
count=0;
break
end

[SignalOut Eq]=MIMO_CoherentReceiverQPSK(SigOut,CoRx); % Input variables passed


are CoRx and SigOut, The CoherentReceiverQPSK function handles SignalIn,P
P.SDiscard = 1024;
P.EDiscard = 1024;

%% Data Analyser Results


XCor.Check=zeros(2*Npm,1);
XCor.ind=1;
% Mode 1
QPSKAnalyserIn.Et=SignalOut.Et(1:2,:);
[QResultsM1 XCor] = QPSKAnalyser(QPSKAnalyserIn,data,P,XCor);
M1BER=QResultsM1.BER;

XCor.ind=2;
% Mode 2
QPSKAnalyserIn.Et=SignalOut.Et(3:4,:);
[QResultsM2 XCor] = QPSKAnalyser(QPSKAnalyserIn,data,P,XCor);
M2BER=QResultsM2.BER;

% Checks that each stream has converged to a unique point -> 8 unique streams for
DP-QPSK on 2 modes
TotalBER=(QResultsM1.BER+QResultsM2.BER)/2;
XCor.Check=sort(XCor.Check);
diffcheck=0;
for k=1:2*Npm-1
if XCor.Check(k+1)-XCor.Check(k)==4096/P.sm
diffcheck=diffcheck+1;
elseif XCor.Check(k+1)-XCor.Check(k)==4096/P.sm-1
diffcheck=diffcheck+1;
else
break
end
end

if diffcheck==2*Npm-1

Filtered1=cumsum(Eq.Error(1,:).^2);
Threshold1=0.8*Filtered1(end);

84
symb1=find(Filtered1>Threshold1, 1); % Convergence time of equaliser in terms
of number of symbols in to the data

Filtered2=cumsum(Eq.Error(2,:).^2);
Threshold2=0.8*Filtered2(end);
symb2=find(Filtered2>Threshold2, 1); % Convergence time of equaliser in terms
of number of symbols in to the data

Filtered3=cumsum(Eq.Error(3,:).^2);
Threshold3=0.8*Filtered3(end);
symb3=find(Filtered3>Threshold3, 1); % Convergence time of equaliser in terms
of number of symbols in to the data

Filtered4=cumsum(Eq.Error(4,:).^2);
Threshold4=0.8*Filtered4(end);
symb4=find(Filtered4>Threshold4, 1); % Convergence time of equaliser in terms
of number of symbols in to the data

ConvTime(i+ThetaStep+1,j+ThetaStep+1)=(symb1+symb2+symb3+symb4)/4;
break
else
WhileBreak=0;
count=count+1;
end

end
end
end
% end
%% 3D Surface Plot
[x,y]=meshgrid(-90:90/(ThetaStep):90,-90:90/(ThetaStep):90);%[x,y]=meshgrid(0:90/(ThetaStep-
1):90,0:90/(ThetaStep-1):90);
figure
surf(x,y,ConvTime)

7.2.2 Multimode Fibre Channel Model


function [ MMFSigOut ] = MIMO_MMFChannel(SigOut,P)
% Models Polarisation rotation, phase shift and modal coupling in the 4x4
% MMF MIMO channel

% Author: W.Hughes July 2011

[Npm Nt]=size(SigOut.Et); % Npm = (No. of pols) x (No. of Spatial Modes Utilised), Nt


= No. of samples

phi=P.PhiShift; % Polarisation Rotation applied to both pols (degrees)


theta=P.PolTheta; % Polarisation Phase Shift between X and Y pols (degrees)

H=zeros(Npm,Npm);

85
%% 2x2 Polarisation Matrix - Polarisation Dependent Effects (Rotation, Phase shift)

for k=1:P.sm

PolMat=zeros(2,2);
PolMat(1,1)=(exp((1i*phi(k))/2))*cos(theta(k)); % 2x2 Pol effects matrix for each mode
PolMat(1,2)=(exp((-1i*phi(k))/2))*sin(theta(k));
PolMat(2,1)=(-exp((1i*phi(k))/2))*sin(theta(k));
PolMat(2,2)=(exp((-1i*phi(k))/2))*cos(theta(k));

% Channel Matrix for all modes (Npm x Npm):


if P.ModeCoupling==1;
%% Modal Coupling
H(1:2,2*k-1:2*k)=PolMat; % Use line for modal coupling case
for k=2:P.sm
H(2*k-1:2*k,:)=H(1:2,:);
end
% % Randomly Generated Unitary Modal Coupling Matrix P.sm x P.sm
% A=kron(0.8*rand(1),ones(2,2));
% B=(A+A')/2;
% M_couple=expm(1i*B); % Factor of 0.2 to make 0->50 dB range = 0->10 dB range
% AbsCouple=abs(M_couple(1,1))/abs(M_couple(2,1));

if P.ModeCoupleStep==0
AbsCouple=1;
else
AbsCouple=1/(10^((P.ModeCoupleStep*0.2)/10)) % Factor of 0.2 to make 0->50 dB
range = 0->10 dB range
end
M_couple=[1,AbsCouple;AbsCouple,1]

% Maintain unity power for each mode


Tot1=sum(M_couple(:,1));
Tot2=sum(M_couple(:,2));

M_couple(:,1)=M_couple(:,1)/Tot1;
M_couple(:,2)=M_couple(:,2)/Tot2;

M=kron(M_couple,ones(2,2));
H=M.*H;

elseif P.ModeCoupling==0;
H(2*k-1:2*k,2*k-1:2*k)=PolMat; % Use line for no modal coupling case
end
end

%% Multimode Fibre Output


MMFSigOut=H*SigOut.Et;

End

86
7.2.3 Code modelling Coherent Detection of the Optical Signal
function SignalOut=MIMO_CoherentRx(SignalIn,P)

% Converts the optical field to a 4 electric fields corresponding to the


% in-phase and quadrature compenents for each polarisation for each mode
%
%
% SignalOut=CoherentRx(SignalIn,P)
%
% Inputs:
% SignalIn - input signal structure
% P.LOoffset - offset frequency (GHz)
% P.LOpower - Local oscillator power (dBm)
% P.Rxtype - Ideal,AsymmetricCoupler,OpticalHybridSingle,OpticalHybridDifferential
% P.Linewidth - Linewidth of local oscillator (MHz)
% P.sm - Only for numerous spatial modes
%
% Returns:
% SignalOut - output signal structure
%
% Author: Seb Savory, July 2005
% Modified: Benn Thomsen, Oct 2010
% Modified: Will Hughes, May 2011

SignalOut=SignalIn;

[Npm,Nt] = size(SignalIn.Et); % Total number of fields and temporal points


dF = SignalIn.Fs/Nt; % Spectral resolution (Hz)
FF = [0:(Nt/2)-1,-Nt/2:-1] * dF; % Frequency array (Hz)
dT = 1/SignalIn.Fs; % Temporal resolution (s)
TT = (0:Nt-1) * dT; % Time array (s)

foffset=P.LOoffset*1e9; %(Hz)
foffset=dF*round(foffset/dF);
LOpower_W=10^((P.LOpower-30)/10);
E_lo=sqrt(LOpower_W)*exp(-1i*2*pi*TT*foffset);
if isfield (P,'Linewidth'),
fayz_variance=2*pi*dT*P.Linewidth*1e6;
SigPhase = sqrt(fayz_variance)*cumsum(P.RandPhase);
E_lo=E_lo.*exp(1i*SigPhase);
SignalOut.PhaseNoise = SigPhase;
end

E=zeros(8*P.sm,length(SignalIn.Et));

switch P.Rxtype

case 'OpticalHybridDifferential'
n=1;
for m=1:P.sm

87
E(n:n+1,:)=abs(0.5*[1 -1i; -1i 1]*[SignalIn.Et(m*2-1,:); E_lo]).^2; % Ex,inphase
E(n+2:n+3,:)=abs(0.5*[1 -1i; -1i 1]*[SignalIn.Et(m*2,:); E_lo]).^2; % Ey, inphase

E_lo=1i*E_lo; % Rotate LO polaristaion by pi/2 i.e. j

E(n+4:n+5,:)=abs(0.5*[1 -1i; -1i 1]*[SignalIn.Et(m*2-1,:); E_lo]).^2; %


Ex,quadrature
E(n+6:n+7,:)=abs(0.5*[1 -1i; -1i 1]*[SignalIn.Et(m*2,:); E_lo]).^2; % Ey,
Quadrature

E_lo=-1i*E_lo;

n=n+8;
end
%% SignalOut: Each mode gives [E(1) - E(2) +j(E(5) - E(6));E(3) - E(4) +j(E(7) - E(8))]
for k=1:P.sm;
SignalOut.Et((2*k-1):2*k,:) = [E(8*k-7,:)-E(8*k-6,:)+1i*(E(8*k-3,:)-E(8*k-
2,:));E(8*k-5,:)-E(8*k-4,:)+1i*(E(8*k-1,:)-E(8*k,:))];
end

SignalOut.Type ='Electrical';
end

7.2.4 Scaled-UP CMA Function


function [Signal Eq]=MIMO_QPSK_CMA(SignalCom,P)

%% Author: Benn Thomsen 1 Sept 2010


% Modified: Will Hughes June 2011

%% MIMO equalisation
Signal=SignalCom;
[Npm Nt]=size(Signal.Et); % Npm = (No. of pols) x (No. of Spatial Modes Utilised), Nt
= No. of samples

FFE.N=P.N; % P.N - Number of taps 2N+1


FFE.M=P.M; % P.M - Fractional spacing

L=Nt/2;
K=Npm;
FFE.L=FFE.M*(2*FFE.N+1)+1; % M(2N + 1)
FFE.u=P.mu; % Step size 1e-6 < u < 0.1

%% Use Gaussian window for tap weights (comment for uniform tap update)
FFE.u = P.mu.*gausswin(FFE.L)'; % Apply gaussian window to step size matrix of constant
value

u_blank=zeros(1,K*FFE.L);
for i=1:K;

88
u_blank((i-1)*FFE.L+1:i*FFE.L)=FFE.u;
end

FFE.u =ones(K,1)*u_blank; % Generates KxKL matrix: each row is gaussian distributed about
the central tap weight
s=[Signal.Et(:,Nt-FFE.M*(FFE.N+1)+1:Nt),Signal.Et,Signal.Et(:,1:FFE.M*FFE.N+1)]; %
pad input signal

% FFE Equaliser
if isfield(P,'Taps')
C = P.Taps;

else
C=zeros(K,K*FFE.L);
tapcenter=(FFE.L+1)/2;

for i=1:K;
C(i,(i-1)*FFE.L+tapcenter)=1; %% initial AFIR tap coefficients
end
end

y=zeros(K,L);
e=zeros(K,L);

for n=1:L, % L= 16384 and FFE.L = 51


x=zeros(1,K*FFE.L);
ind=FFE.M*(n-1);
for i=1:K;
x((i-1)*FFE.L+1:i*FFE.L)=s(i,ind+1:ind+FFE.L);
end
x=x.';
y(:,n)=C*x;
e(:,n)=1-abs(y(:,n)).^2; % CMA Error
C=C+FFE.u.*((e(:,n).*y(:,n))*x'); % CMA update Algorithm
end

Signal.Et=y;
Signal.Fs=Signal.Fs/FFE.M;

Eq.Ntaps=FFE.L;
Eq.Taps=C;
Eq.Error=e;

7.2.5 Subplots of the data


%% Plot Received Data

% Plots the Equaliser Tap Weights: Real and Imaginary curves

figure(100),title('Equaliser Tap Coefficient Values')

89
for k=1:(P.sm*P.Np)
for i=1:(P.sm*P.Np)
subplot(P.sm*P.Np,P.sm*P.Np,(k-1)*P.sm*P.Np+i),plot(1:Eq.Ntaps,real(Eq.Taps(k,(i-
1)*Eq.Ntaps+1:i*Eq.Ntaps)),...
1:Eq.Ntaps,imag(Eq.Taps(1,(i-1)*Eq.Ntaps+1:i*Eq.Ntaps)))
title(sprintf('H%d%d',k,i))
end
end

figure(101),subplot(231),plot(SignalSam.Et(1,4:2:end),'.')
title('Input')
subplot(234),plot(SignalSam.Et(2,4:2:end),'.')

subplot(232),plot(SignalEq.Et(1,1024:end-32),'.');
title('After CM Equaliser')
subplot(235),plot(SignalEq.Et(2,1024:end-32),'.');

subplot(233),plot(SignalOut.Et(1,1024:end-32),'.');
title('After Phase recovery')
subplot(236),plot(SignalOut.Et(2,1024:end-32),'.');

7.2.6 Power Meter Reader Code


%% Create the structure with instrument description
PM1.GPIBcard = 0;
PM1.GPIBadd = 24;
% Either 1, 2, or 3 (1 is the leftmost).
PM1.slot = 1;
% Optional: Set to 1 to get all the GPIB protocol debug.
verbose = 0;

SweepTime=120;
Ports=3;

PM2=PM1;
PM2.GPIBadd = 20;
PM2.Ports=Ports;
PM2.SweepTime = SweepTime;

%% Initialise and create a persistent connection


% PM1 = ongOMS150InitialisePowerMeter(PM1, verbose);
PM2 = ongOMS150InitialisePowerMeter(PM2, verbose);

%% Read power from power meter 1 and 2: All Slots


n=0;
ClockRef=clock;
ClockLog=zeros(1,3*SweepTime);
time=zeros(1,SweepTime);

Power=zeros(Ports+1,SweepTime); % The '+1' allows for total power storage as well

90
for n=1:SweepTime,
% Time Stamp
c=clock;
ClockLog(1,(3*(n-1)+1):3*n)=c(1,4:6)-ClockRef(1,4:6);
time(n)=3600*ClockLog(1,3*(n-1)+1)+60*ClockLog(1,3*(n-1)+2)+ClockLog(1,3*(n-1)+3);
% Through Port: Brown OP
% Power(1,n) = ongOMS150GetPower(PM1, verbose);
% X-Port: Blue OP (4x1_1x4 Setup)
[Power(1:Ports,n)] = ongOMS150GetPowerAll(PM2, verbose);
pause(0.5)
end
% Release instrument
% PM1 = ongInstrumentRelease(PM1, verbose);
% PM2 = ongInstrumentRelease(PM2, verbose);

%% Release instrument
%PM1 = ongInstrumentRelease(PM1, verbose);
PM2 = ongInstrumentRelease(PM2, verbose);

%% Calculate Total Power


AbsPower=zeros(1,SweepTime);
for i=1:Ports;
AbsPower=AbsPower+(10.^(Power(i,:)/10)); % Total Power (Abs)
end
Power(Ports+1,:)=10*log10(AbsPower); % Total Power (dBm)

%% Calculate Insertion Loss (dB):


CrossIn=-2.1;
ThroughIn=-2.4;
%% Cross ON - Through OFF
%InsLoss=CrossIn-Power;

%% Cross OFF - Through ON


InsLoss=ThroughIn-Power;

%% Cross ON - Through ON
%InsLoss=10*log10((10^ThroughIn/10)+(10^CrossIn/10))-Power;

XLIM=max(time);

%figure,hold;plot(time,InsLoss(1,:),'Color',[.7 .5
0]);plot(time,InsLoss(2,:),'b');plot(time,InsLoss(3,:),'Color',[1 0.7
0]);plot(time,InsLoss(5,:),'k');xlim([0 XLIM]);hold
figure,hold;plot(time,InsLoss(1,:),'b');plot(time,InsLoss(2,:),'r');plot(time,InsLoss(3,:),'c'
);plot(time,InsLoss(4,:),'k');xlim([0 XLIM]);hold
xlabel('Time (s)')
ylabel('Insertion Loss (dB)')
%legend('Cross Left','Cross: Blue OP','Cross: Orange OP','Cross: Green OP','Total Insertion
Loss')

91
7.3 Project Management

Regular project meetings and discussions were conducted throughout the project. The minutes are
given below:

7/3/11 – Project Aim/Discussion, Initial Literature Research, In conjunction with revision for final
exam on 26/3/11.
30/3/11 – Access to existing MATLAB toolbox. Understand the system code. Research adaptive
filtering and understand the equaliser functionality.
19/4/11 – Confirm project aim: Scale up MATLAB toolbox from 2x2 Polarisation MIMO system to 4x4
Polarisation/Mode MIMO system. Perform lab tests after the Easter Break. Next project meeting
planned. CMA, LMS and channel polarisation effects were discussed. Equaliser performance tested
under different conditions and documented.
4/5/11 – Discussed results of scaled up 4x4 MIMO MATLAB model. Tasks: Allow user entry to make
code fully scalable and implement non-ideal MMF channel to model polarisation and mode
dependent effects. Carrier recover was discussed.
16/5/11 – Discussed Mode Coupling software approach and software model assumptions.
19/5/11 – Confirmed simulation setup before capturing results
26/5/11 – Begin 2x2 MIMO system Lab Tests. Test equipment was setup and the test aims were
outlined. In the meantime, run MATLAB simulations.
10/6/11 – Discussed MMC characterisation test results.
13/6/11 – Cambridge University Lab Visit. 1x2 and 1x4 MMC characterisation was tested using the
SLM launch.
16/6/11 – Project Presentation: Update of results to date and future steps. Feedback was recorded
to improve for future presentations.
21/6/11 – Setup and coded MATLAB Power Meter measurement tool. Discussed results and future
tests required to optimise Tx and Rx coupling.
29/6/11 – Discussed lab results: provide lab report on results to date.
20/7/11 – Discussion of Lab Test Report. Use feedback to improve report
21/7/11 – Write-up simulation test results
7/8/11 – Finalise and retest any relevant lab tests
8/8/11 – Combine Simulation and Lab Test results to produce dissertation.

A Gantt chart was used to monitor project progress. Figure 53 highlights the project completion on
15/6/11.

92
Figure 53 Gantt chart: 15/6/11 snapshot

93

You might also like