We consider the causal channel g always generates a Weyl-Heisenberg frame gna+b, n,m E Z, when a , b > 0, ab < 1. Here
g Z I Y ( t ) = exp (-nizy + 27riyt) g ( t -z), t E wWe truncate the channel response so that N , = 0 and AT, = 20.As before, the channel input is an i.i.d. sequence, and the noise is stationary and white with SNR,h,, =20 dB. We assume the number of feedforward taps is fixed at q = 10. Fig. 3 shows the optimal decision delay (defined as thlat which maximizes SNRDFE) versus the number of feedback taps. 'When the number of feedback taps is small, the optimal feedforward filter is two-sided; that is, it contains both causal and anticausal taps. As expected, however, the optimal A converges to q -1 as the number of feedback taps get large. Note that when A = q-1, the feedforward filter is anticausal; that is, {~k } = { w -~, . . . ,WO}.
REFERENCES
Some Counterexamples in the Theory of Weyl-Heisenberg Frames
A. J. E. M. JanssenAbstruct-We present an example of a positive function g with a positive Fourier transform g and reasonable smoothness and decay properties such that (-lInm exp (zitm) g(t -n ) , n, m E Z does not constitute a frame for L2(W). We also give counterexamples for the statement that one can tell (in)definiteness of a Weyl-Heisenberg frame operator from @)definiteness of its Weyl symbol.