The markers in the figure indicate a certain SNR level the user's channel has to attain in order to satisfy a certain QoS requirement. It is seen that without any transmit processing (red curve), the fading level is always below the threshold. With perfect (coherent) channel knowledge, coherent beamforming would significantly increase this user's SNR (green curve) and the required threshold would be met. With ''random'' beamforming, the user's SNR is above the threshold every once in a while and the scheduler can make use of that information and schedule the user when its channel is ''good'', without the need for coherent channel knowledge. This is a completely new design paradigm, which is in contrast to conventional approaches where one attempts to exploit diversity in order to make a single link more reliable. Multiuser diversity is thus a promising way to improve the area or multiuser spectral efficiency, measured in datarate per bandwidth per area or user.
Several challenging problems arise if one is interested in exploiting multiuser diversity in a real system. One key assumption is always that the transmitter has knowledge about the channel conditions of each user, according to which the scheduler decides which user to schedule. This requires either channel estimates from a continuous uplink transmission in reversible channels, which are available only in TDD systems, or tight feedback from all users to the transmitter. Another point to keep in mind is that high rate wireless data will be transmitted with packet-based transmission protocols with quality of service (QoS) constraints, yielding additional scheduling constraints.
Our aim is thus to study the feasibility of exploiting multiuser diversity in a realistic multiantenna downlink scenario. We are interested in increasing the area spectral efficiency, taking into consideration all the challenges imposed by real system design such as channel estimation, feedback design, smart scheduling with QoS constraints and intelligent antenna processing at the transmitter. We will analyze single-cell and multicell scenarios, taking into account frequency selective fading for each user, different fading dynamics and various degrees of scattering around the transmitter.