Previous Project
 |
Signal Space Detection (SSD) Signal space detectors use one or more classifiers along with a logic rule
to partition an observation space into regions corresponding to different
symbols. In the case of a linear channel, the classifiers are constructed
using linear boundaries (hyperplanes) in an M-dimensional space, where M is
the number of samples used in the detection process. Although the technique
can use nonlinear boundaries as in a multilayer perceptron (MLP), the linear
case is important because it can be readily analyzed and related to existing
detection methods such as
FDTS/DF.
The advantage over FDTS/DF is that the
equivalent SSD can be implemented with less hardware than a direct hardware
mapping of the FDTS metric calculations.
One product of this research is an asyptotically optimal detector.
A sequence detector with a finite decision delay makes a decision on
a channel input symbol based on a finite number of observation samples.
To obtain a decision rule, a multi-dimensional signal space spanned by an observation sample sequence
is divided into non-overlapping decision regions according to the symbol decision values.
A symbol decision is then made according to the location of the observation samples
in the signal space using the architecture shown below. For correlated noise such as transition jitter, the decision boundary is
generally quadratic. We have developed a technique for obtaining a minimal set of
hyperplanes which approximates the quadratic decision boundary with a negligible
performance loss. In this process, a distance measure, which is consistent with the notion of
effective SNR, is defined and used as the design parameter to trade complexity for performance.
The technique leads to an asymptotically optimum detector when the target distance is set at an
effective SNR value associated with the maximum likelihood sequence detector (MLSD).
A related project was the development of the 3D-110 detector, which trades a small performance loss for a significant reduction in complexity. For more information see
|
|
Copyright @ University of Minnesota CDSLab |
