Worst case equalizer for noncoherent HIPERLAN receivers

Coherent detection of HIPERLAN Gaussian minimum-shift keying signals calls for complex and expensive receivers. However, when the channel delay spread is limited to at the most 50% of the symbol time and a reliable line-of-sight component of the radiated signal exists (Rician fading model), noncoherent detectors are capable of achieving a good performance. Based on the above motivations, we compare four different demodulation techniques, namely the following: (1) one-bit differential detector; (2) discriminator detector; (3) limiter discriminator detector; and (4) limiter discriminator integrator detector (LDID). The intersymbol interference introduced by these demodulators is nonlinear (with respect to the data symbols) and a decision-feedback equalizer (DFE) based on a mean square-error criterion may not be appropriate. Moreover, at this high speed, a DFE may be very complex to implement. Hence, we propose a new DFE design method that increases the eye-diagram aperture by removing the worst case interference. Performance of the above demodulators in the presence of a simple nonlinear DFE (with feedback part only) is computed in terms of the bit-error rate (BER) by means of the saddle-point approximation. This procedure, for static channels, turns out to be a very general tool with a simple and robust implementation. The same method can be applied, for multipath fading channels, to the BER evaluation as part of a semianalytic approach. The main conclusion from this work is that for LDID demodulators and in the presence of Rician fading channels with an average normalized root mean square delay spread of 0.3 and dual antenna diversity, the new equalizer lowers the outage probability from 60% to 10% at a BER of 10/sup -4/.