Linear receivers are often used to reduce the implementation complexity of MIMO
systems. Traditionally, the receive antennas are used to separate out the
codewords sent by each transmit antenna, which can then be decoded
individually.
Unfortunately, this zero-forcing approach can be highly suboptimal when the
channel matrix is near singular. We propose a new receiver architecture,
based
on compute-and-forward, that uses the receive antennas to create an effective
channel matrix with integer-valued entries, from which integer combinations of
the codewords can be decoded individually. This integer-forcing approach
closely
follows the performance of joint maximum likelihood decoding and, if there is
no
coding across transmit antennas, attains the optimal diversity-multiplexing
tradeoff.