Total Capacity of Multiaccess Vector Channels

The well known waterrlling power allocation policy maximizes the sum capacity of parallel Gaussian channels. We consider multiaccess vector channels with additive colored Gaussian noise and asymmetric user power constraints and completely characterize the sum capacity of this channel. We show that the sum capacity of the multiaccess vector channel is upper bounded by that of corresponding parallel Gaus-sian channels and that our solution (optimal powers and user signal directions) has a waterrlling structure. Two common examples in a wireless communication system that fall under this model are direct sequence code division multiaccess and multiac-cess channels with multiple antennas at the receiver. The multiaccess vector channel models communication from users within a cell to the base station and interference from those users communicating with neighboring base stations is modeled by additive colored noise. Our characterization of the sum capacity allows us to conclude a Schur-saddle property: the sum capacity is Schur convex in the additive noise covariance and Schur-concave in the user powers.