2005 IEEE 16th International Symposium on Personal, Indoor and Mobile Radio Communications Capacity Enhancement via Multi-Mode Adaptation in Spatially Correlated MIMO Channels Antonio Forenzat, Matthew R. McKayt, Ashish Pandharipande*, Robert W. Heath Jr.t, and lain B. Collingst tWireless Networking and Communications Group (WNCG) ECE Department, The University of Texas at Austin, Austin, TX, USA {forenza, rheath}@ece.utexas.edu tTelecommunications Lab, Sch. of Elec. and Info Engineering, University of Sydney, Australia, and also the ICT Centre, CSIRO, Australia {mckay, i.collings}@ee.usyd.edu.au *Communications and Networking Lab Samsung Advanced Institute of Technology (SAIT), Suwon, Korea pashish@ieee.org Abstract-We consider a low-complexity adaptive MIMO transmission approach for spatially correlated channels. The pro- posed scheme adaptively switches between different transmission modes depending on the changing channel conditions, as a means to enhance system capacity. Each mode is a combination of a transmission technique (ie. statistical beamforming, double space- time transmit diversity and spatial multiplexing) and a modu- lation/coding scheme. We first motivate our adaptive algorithm by deriving new closed-form capacity expressions, and demon- strating significant information theoretic improvements over non- adaptive transmission. We then present a practical method to switch between different modes, based on the channel statistics. Our approach is shown to yield significant improvements in spectral efficiency for typical channel scenarios.' I. INTRODUCTION The performance of multiple-input multiple-output (MIMO) wireless communication systems can be improved by ex- ploiting partial or full channel knowledge at the transmitter. When instantaneous channel knowledge is available, adaptive schemes have been proposed to switch between transmit diver- sity and spatial multiplexing schemes as a means to improve error rate performance [1]. Other adaptive approaches have been designed based on time/frequency selectivity indicators [2]. Alternatively, the spatial selectivity of the channel, defined as in [3], can be exploited to switch between different MIMO schemes. It is now well-known that the capacity [4] and error rate performance [5] of MIMO systems depend on the spa- tial characteristics of the propagation environment (ie. angle spread, number of scatterers, angle of arrival/departure) [6,7]. This dependence is typically revealed through the eigenvalues of the transmit and receive spatial correlation matrices [8], which give an indication of the channel spatial selectivity. In this paper we exploit knowledge of the spatial selectivity in a low-complexity adaptive MIMO transmission scheme as a means to improve system performance. The proposed scheme switches between statistical beamforming (BF), double space- time transmit diversity (D-STTD) and spatial multiplexing 'The work of A. Forenza and R. W. Heath, Jr., is supported by Freescale Semiconductor, Inc., the Office of Naval Research, and by the National Science Foundation under grant numbers 0322957 and 0435307. (SM) depending on the estimated channel quality (partially re- vealed through the spatial selectivity information). To motivate the scheme, we first derive some new closed-form capacity results which show an explicit dependence on the eigenval- ues of the spatial correlation matrices. We then demonstrate the significant information theoretic improvements which are obtained by adapting based on this spatial selectivity infor- mation. We finally present a practical adaptive algorithm that switches between different transmission modes, by using the information of the spatial correlation matrices and the average signal to noise ratio (SNR). This practical approach yields significant spectral efficiency improvements over non-adaptive transmission in typical channel scenarios. II. SYSTEM AND CHANNEL MODELS We consider a narrowband MIMO system employing Nt transmit and N, receive antennas modelled (for each channel use) by y= JtHx+n (1) where y E CNr X 1 is the receive signal vector, x E CNt Xx1 is the transmit signal vector subject to the power constraint E{llxll} = Nt, and n E CNrXl is the zero-mean additive Gaussian noise vector with covariance matrix &{nnH} = NOINr. Also, H E (CNr x Nt is the spatially correlated Rayleigh MIMO channel matrix modelled as H = Rl/2ZSI/2 (2) where Z E CNr-XNt contains independent complex Gaussian entries with zero mean and unit variance, and where S and R denote the transmit and receive spatial correlation matrices respectively, with eigenvalue decompositions (3) We assume that R and S are normalized Hermitian positive definite matrices with Tr (R) = Nr and Tr (S) = Nt. Moreover, throughout this paper we assume that the receiver has perfect knowledge of the instantaneous channel H, and 978-3-8007-2909-8/05/$20.00 ©2005 IEEE U,A, UH , R = UrArU" 8 r 754