In this paper, we consider the same general architecture as the second Gans model described above, but for a shorter
range (50 ft) and a lower bit rate of 100 Mb/s. We assume a certain configuration of four linear beamformers (only discrete beam-pointing choices) and a statistical (azimuth-only) propagation model that emphasizes clustering of paths in angle and delay [7]. This model has the significant delay spread in a narrow angular region that was mentioned in a caveat in [6]. In addition, we consider the use of tapering to reduce crossover loss and the use of two beams simultaneously (jointly equalized) for angle diversity. So, while our system requires no manual pointing of antennas, it does have the complexity of an equalizer.We assume TDMA/TDD QPSK, a very slowly varying channel (based on
pedestrian movement), no interference, and beam selection on the uplink based simply on power. In cellular or unlicensed
band applications consistent with high levels of interference, more robust beam-selection criteria are required to combat the “beam-falsing” problem [8]. Our link budget analysis is reported elsewhere [9]; in this paper,we consider onlySNRimprovements offered by various changes in BS receiver design. In particular, we determine the tradeoffs between the amount of tapering, the number of angle diversity beams (one or two), the total number of beams, and the number of forward taps in the DFE.