SIMULATION TOOLS: The DFEcator

Download the most recent version of the DFEcator (.tar.gz) (in UNIX, type "gunzip DFEcator.tar.gz" followed by "tar -xvf DFEcator.tar" to uncompress it. You must compile two MEX files by typing "mex adapt.c" and "mex ep.c" at the Matlab prompt.)

The most recent version of the DFEcator is dated Wednesday, 28-Nov-2001 15:20:41 EST.

General Information

The DFEcator is a MATLAB5-based interactive tool based on The BERGulator for design and simulation of minimum mean square error (MMSE) and adaptive (including blind) decision feedback equalizers (DFE). Both baud-spaced equalizers (BS) and T/2 fractionally-spaced (FS) DFEs may be designed to equalize complex, additive white gaussian noise channels (AWGN) for various choices of parameters. For MMSE designs some of these choices include:

channel (coefficients, source and SNR),
forward and feedback filter lengths OR total number of coefficients,
cursor delay (for non-blind equalization),
processing delay (delay required for processing decisions),
bias.

For adaptive designs, it includes choice of:

forward and feedback filter initializations,
adaptive algorithm (see below for details),
forward and feedback filter step-size.

Figures displaying information about the designs may be accessed through toggle buttons on the left side of the screen. Variables containing results of designs and experiments may be accessed by typing "dfec_global" at the Matlab prompt. There are four main pieces of The DFEcator: channel, equalizer and adaptation parameter boxes, and action buttons. A brief description of each follows:

Channel Parameters

Channel Type: demonstration channels, "Custom" channels by typing coefficients in editing box or typing name of a variable residing in Matlab workspace, or coefficients in a .mat "File".
Source: several real and complex valued sources, e.g. M-PAM, M-QAM.
dB SNR: signal to noise ratio in dBs, "Inf" for no noise. Can type range of SNRs, e.g. 10:2:20 for SER vs. SNR waterfall curves.
Normalize: normalizes channel to unit (l_2) norm when toggled on.
Display: displays channel impulse and frequency responses when toggled on.

Equalizer Parameters

Filter lengths: type [Nf Nd] for length of forward and feedback filters, e.g. [4 3] for a length Nf=4 forward filter and length Nd=3 feedback filter, or type single integer M for a total number of parameters. The DFEcator will compute all possible MMSE designs for each combination [Nf Nd] (over other parameters such as SNR and delay) such that Nf+Nd=M.
Cursor Delay: choice of transmission delay which is either matched to to maximum channel tap, optimized over every possible delay choice or over range of delays.
Spacing: choice of baud- ("BSE") or T/2 fractionally-spaced ("FSE") channels/equalizers.
Bias: design biased ("Biased") and unbiased ("Unbiased") MMSE designs.
Processing Delay: determine processing time required to make decisions on symbols.

Adaptation Parameters

Initializations:
- "Zero" sets initial forward and feedback filters to zero.
- "Center spike" sets initial forward equalizer to have a single, center spike and feedback filter to zero.
- "Double spike" sets initial forward filter to have double, center spikes and feedback filter to zero.
- "DFE-MMSE" sets filters to MMSE design DFE solution.
- "IIR-MMSE" sets filters to truncated Wiener linear IIR solution.
- "End spike" sets initial forward with a single spike to the right end of the tap-delay line, and feedback filter to zero.
- "Previous" sets filters to the final settings of the previous adaptation.
- "Custom" allows typing in the edit boxes "f(0)" and "d(0)" initial coefficients or variable names residing in the Matlab workspace.
- "From file" allows typing in a .mat file containing variables f0 for forward filter taps and d0 for feedback filter taps.
Algorithm:
- "DFE-LMS" uses the least mean squares algorithm with training sequence delayed by the choice of delay in the EQUALIZER PARAMETERS box to update the DFE structure.
- "DFE-DD" uses decision direction to update DFE structure with training and blind data.
- "DFE-CMA" uses the constant modulus algorithm to update DFE structure using training and blind data.
- "DFE-LMS+DD" uses LMS during training and decision-direction in blind mode.
- "DFE-LMS+CMA" uses LMS during training and CMA in blind mode.
- "IIR-LMS" adapts forward and linear AR feedback filters with training and LMS.
- "IIR-DD" adapts forward and linear AR feedback filters with decision direction.
- "IIR-CMA" adapts forward and linear AR feedback filters with CMA.
- "IIR-LMS+DD" uses LMS during training and decision-direction in blind mode.
- "IIR-LMS+CMA" uses LMS during training and CMA in blind mode.
For IIR algorithms either Feintuch or exact gradient updates.
Stepsize: separate forward and feedback filter step-sizes,

Actions

Design: update design once a set of parameters is selected.
MMSE Impulse response: impulse responses of filters and combinations of filters and channel.
MMSE Frequency response: frequency responses of filters and combinations of filters and channel.
MSE vs. Parameters: plot MSE vs. filter length and delay parameters. Three different plots may be displayed according to settings:
- If the filter length choice is set to Nf+Nd=M and a range of delays is selected with the "Range" or the "Opt" options the "MSE vs. Params" figure is a three-dimensional surface of MSEs over (Nf,delay) parameter space (1<=Nf<=M).
- If the format [Nf Nd] is used for filter lengths, and a range of delays is selected with the "Range" or the "Opt" options the "MSE vs. Params" figure is a curve of MSE for each possible delay.
- If the filter length choice is set to Nf+Nd=M but the cursor delay is set to "Max" (matching the channel cursor) then a curve of MSE is plotted for each Nf, 1<=Nf<=M.
MMSE cluster plot: cluster plot of MMSE design.
Error Propagation: Hard and soft decision errors as well as noise filtered through forward equalizer for 1000 transmitted symbols for MMSE design.
SER Curves: symbol error rate (SER) vs. SNR for MMSE design.
Adaptation: adapts filters using choice of algorithm, step-sizes, duration and initialization.
Trajectories: displays evolution of filter taps over time.
Adaptation Error: displays adaptation error associated with particular choice of algorithm.
Adapation cluster plot: evolution of cluster plot in time.
Final cluster plot: cluster plot associated with final equalizer.
Final Impulse response impulse responses of filters and combinations of filters and channel resulting from adaptation.
Final Frequency response: frequency responses of filters and combinations of filters and channel resulting from adaptation.
Help: helps.
Quit: quits.

DFEcator written by: Raúl Casas, September 1998

Send comments, bug reports, or cold hard cash to raulc@ee.cornell.edu.