Lecture 15 - EE 359: Wireless Communications - Fall 2011

MIMO Capacity, Beamforming, Diversity, RX Design

Lecture Outline
MIMO Parallel Channel Decomposition MIMO Channel Capacity Beamforming MIMO Diversity/Multiplexing Tradeos MIMO RX Design

1. MIMO Systems: Parallel Channel Decomposition With perfect channel estimates at the transmitter and receiver, the MIMO channel decomposes into RH independent parallel channels, where RH is the rank of the channel matrix (min(Mt, Mr ) for Mt transmit and Mr receive antennas under rich scattering). With this decomposition there is no need for vector signal processing. Decomposition is obtained by transmit precoding and receiver shaping. 2. MIMO Channel Capacity Capacity depends on whether the channel is static or fading, and what is known about the channel at the transmitter and receiver. For a static channel known at the transmitter and receiver capacity is given by C=
Pi :

max
i

Pi P

B log2 1 +
i

2 i Pi 2 n

=
Pi :

max
i

Pi P

B log2 1 +
i

Pi i . P

This leads to a water-lling power allocation in space. When channel is unknown at transmitter, uniform power allocation is optimal, but this leads to an outage probability since the transmitter doesnt know what rate to transmit at: Pout = p H : B log2 det IMr + HHH > C . Mt In fading, capacity with both transmitter and receiver knowledge is the average of the capacity for the static channel, with power allocated either by a short-term or long-term power constraint. Without transmitter knowledge, outage probability is the right metric for capacity. 3. MIMO Systems: Beamforming Beamforming sends the same symbol over each transmit antenna with a dierent scale factor. At the receiver, all received signals are coherently combined using a dierent scale factor. This produces a transmit/receiver diversity system, whose SNR can be maximized by optimizing the scale factors (MRC).

 Beamforming leads to a much higher SNR than on the individual channels in the parallel channel decomposition. Thus, there is a design tradeo in MIMO systems between capacity and diversity. 4. Diversity versus Multiplexing in MIMO Systems Can exchange data rate for probability of error. Dene rate scale factor r = R/ log(SN R). Dene diversity gain d = log)Pe )/ log(SN R)/. Can show (Zheng/Tse02) that in high SNR regime, the optimal tradeo is d (r ) = (Mt r )(Mr r ).. The optimal operating point on this traedeo curve depends on the application. 5. MIMO Receiver Design Optimal MIMO receiver is maximum-likelihood (ML) receiver. Finds input vector x that minimizes |y Hx|2 F for ||F the Frobenius (matrix) norm. Exponentially complex in the constellation size and number of transmit antennas. Can reduce complexity through linear processing of input vector Ax. Zero-forcing receiver forces all interference from other symbols to zero. This can result in signicant noise enhancement. MMSE receiver: trades o cancellation of interference from other symbols for noise enhancement. Reduces to zero forcing in the absence of noise. Decision-feedback receivers: uses matrix transformation of channel matrix into a triangular representation. Hence, the j th symbol can be decoded linearly once the j + 1th is decoded. Start with the symbol with the largest index, and work backwards, subtracting out previously decoded symbols. Prone to error propagation. Sphere Decoder: Approximates ML decoder, but only considers symbols that would result in an output, in the absence of noise, within a radius r of the received vector. Can trade performance for complexity via choice of r . Decoding reduces to dierent methods for pruning the tree of possible inputs to be within the given radius. Space-time processing: uses coding across time or across antennas. When coding across time, successive interference cancellation of interfering symbols is done, with the decoding order based on symbol SNR (highest to lowest).

Main Points
MIMO systems exploit multiple antennas at both TX and RX for capacity and/or diversity gain. With both TX and RX CSI, multiple antennas at both transmitter and receiver lead to independent parallel channels. With TX and RX CSI, capacity of MIMO channel uses waterlling in space or space/time - leads to min(Mt , Mr ) capacity gain. Without transmitter CSI, use outage as capacity metric. MIMO introduces diversity/multiplexing tradeo: Optimal use of antennas depends on application. MIMO RX design trades complexity for performance. ML detector optimal; exponentially complex. DF receivers prone to error propagation Sphere decoders allow performance tradeo via radius