AdaptiveEqualizer_MMA_RDE_v1 Class Reference

Radius directed equalization. More...

Inherits unit.

Public Member Functions
function	AdaptiveEqualizer_MMA_RDE_v1 (in params)
	Class constructor. More...
function	traverse (in obj, in in)
	Main routine.
function	checkSettings (in obj)
	Check user settings. More...
function	initialize (in obj)
	Initialize object. More...
function	train_equalizer (in obj, in in)
function	poldemux_cr_qam_v4_fast (in obj, in in)
	Fast equalization routine. More...
function	poldemux_cr_qam_v4_medium (in obj, in in)
	Medium-fast equalization routine. More...
function	poldemux_cr_qam_v4_slow (in obj, in in)
	Slow equalization routine. More...
function	plotConv (in obj)
	Plot equalizer convergence.
function	plotTaps (in obj)
	Plot filter taps.
Public Member Functions inherited from unit
function	traverseNode (in obj)
	Set unique ID when creating a unit. More...
function	connectOutput (in obj, in uobj, in unitOutput, in nextUnitInput)
	Specify where signal should go next. More...
function	connectOutputs (in obj, in units, in destInputs)
	Specify where signal should go next. More...
function	writeInputBuffer (in obj, in sig, in inputId)
	write input buffer
function	horzcat (in varargin)
	horizontal concatenation
function	vertcat (in varargin)
	vertical concatenation
function	setparams (in obj, in params, in REQUIRED_PARAMS, in QUIET_PARAMS)
	set parameters More...
function	view (in obj)
	Show interactive GUI through unit_view-class.

Public Attributes
Property	nInputs
	Number of inputs.
Property	nOutputs
	Number of outputs.
Property	type
	Type {'cma' | 'mma'}.
Property	h_ortho
	Reset taps to be orthogonal between training iterations (avoid CMA singularity) {true | false}.
Property	taps
	Number of taps.
Property	mu
	Tap update coefficient.
Property	cma_preconv
	CMA pre-convergence length.
Property	equalizer_conv
	Training period convergence length.
Property	iter
	Number of iterations for training.
Property	constellation
	Vector of symbols in constellation (see constref.m)
Property	operation
	Operation mode {'training' | 'equalization' | 'semiadaptive' | 'adaptive'}.
Property	trainingType
	Training type mode {'blind' | 'dataAided'}.
Property	trainingSequences
	N-by-2 training symbols matrix for dataAided mode.
Property	h_init
	Inital center tap value.
Property	hxx
	Store taps for h_xx.
Property	hxy
	Store taps for h_xy.
Property	hyx
	Store taps for h_yx.
Property	hyy
	Store taps for h_yy.
Property	R
	Store ring radii for MMA.
Public Attributes inherited from unit
Property	inputBuffer
	Buffer for storing inputs as we traverse the graph.
Property	nextNodes
	Children nodes.
Property	destInputs
	Destination inputs in children.
Property	results
	For storing results.
Property	label
Property	draw
	enable/disable plotting
Property	nInputs
	Number of signals traverse expects.
Property	nOutputs
	Number of outputs traverse expects.

Detailed Description

Radius directed equalization.

Runs blind adaptive equalization on dual-polarization coherent signals with M>=4. Equalization is split into a training part and a main filtering part. There are two ways to perform training and three ways to do the main equalization:

Training modes:

1. blind: Radius is chosen based on minimum Euclidean distance criteria
2. dataAided: Radius is chosen based on training sequence provide by user

In blind training mode, for M>4, we bootstrap initially using the constant modulus algorithm (CMA), then after pre-convergence we switch to the multiple-modulus algorithm (MMA). Training sequence length and pre-convergence length can be set using class properties.

Equalization modes:

1. training: train-only. In this case, there is still an output signal, and it is the estmated field during training.
2. equalization: Equalizer taps are calculated using the first part of the sequence for training, then applied without further adaptation to the rest of the sequence.
3. semiadaptive: Training is performed as usual to initialize, but taps are also updated over the whole sequence.
4. adaptive: Training is performed as usual to initialize, then both taps and the update step are also updated over the whole sequence.

Observations:

1. There are inherent limitations to blind equalization. This doesn't attempt to solve any of them.

Example:

paramDSP.eq.iter = 4;        %Run 4 iterations of CMA/MMA on training seq. (default 1)
paramDSP.eq.taps = 31;       %number of equalizer taps (default 7)
paramDSP.eq.mu = 1e-3;       %update coefficient (default 6e-4)
paramDSP.eq.type = 'cma';    %cma or mma
paramDSP.eq.h_ortho = true;  %Force-orthogonalize taps to avoid CMA singularity
paramDSP.eq.cma_preconv = 10000; %Run CMA this many samples
paramDSP.eq.equalizer_conv = 20000; %Train equalizer this many samples
paramDSP.eq.constellation = [constref('QAM',16)]; %constellation
paramDSP.eq.draw = false;    %disable convergence plotting
paramDSP.eq.operation = 'equalization'; %Run both training and equalization

DSP = AdaptiveEqualizer_MMA_RDE_v1(paramDSP.eq); %construct
Signal_after_eq = traverse(DSP, Signal_before_eq);   %run

Author

Edson Porto da Silva

Robert Borkowski

Version

1

Definition at line 62 of file AdaptiveEqualizer_MMA_RDE_v1.m.

Constructor & Destructor Documentation

function AdaptiveEqualizer_MMA_RDE_v1::AdaptiveEqualizer_MMA_RDE_v1

(

in

params

)

Class constructor.

Only constellation is required, everything else has a default value. Plotting is on by default, disable using params.draw=0.

Parameters

params

parameter structure (fields correspond to class properties).

Member Function Documentation

function AdaptiveEqualizer_MMA_RDE_v1::checkSettings

(

in

obj

)

Check user settings.

There are many options with this function, and it's easy to get confused. This checks for some more common mistakes and displays warnings when they arise

function AdaptiveEqualizer_MMA_RDE_v1::initialize

(

in

obj

)

Initialize object.

Set filter taps and check user-specified parameter values.

function AdaptiveEqualizer_MMA_RDE_v1::poldemux_cr_qam_v4_fast	(	in	obj,
		in	in
	)

Fast equalization routine.

Equalizer taps are calculated using the first part of the sequence for training, then applied without further adaptation to the rest of the sequence. For M>4, on the first training iteration, we bootstrap with CMA. Between iterations, the last value of the taps is retained. If obj.h_ortho is false, this is done for all filters; if it is true, we take h.yy = conj(h.xx) and h.yx = -conj(h.xy).

Parameters

in	input signal

Return values

out1	equalized output at 1 sample per symbol
outNss	equalized output at same oversampling rate as input

function AdaptiveEqualizer_MMA_RDE_v1::poldemux_cr_qam_v4_medium	(	in	obj,
		in	in
	)

Medium-fast equalization routine.

Equalizer taps are applied WITH further adaptation to the rest of the sequence. Tap update step size remains constant for the whole sequence.

Parameters

in	input signal

Return values

out1	equalized output at 1 sample per symbol

function AdaptiveEqualizer_MMA_RDE_v1::poldemux_cr_qam_v4_slow	(	in	obj,
		in	in
	)

Slow equalization routine.

Equalizer taps are applied WITH further adaptation to the rest of the sequence. Tap update step size is adapted at each step.

Parameters

in	input signal

Return values

out1	equalized output at 1 sample per symbol

---

The documentation for this class was generated from the following file:

• library/rx/DSP/AdaptiveEqualizer_MMA_RDE_v1.m