University of Southern California WiDeS - Wireless Devices and Systems Group The USC Andrew and Erna Viterbi School of Engineering USC

MIMO SYSTEM



BACKGROUND


Multiple-input – multiple-output (MIMO) systems are wireless systems with multiple antenna elements at both link ends.The multiple antenna elements of a MIMO system can be used for three different purposes: (i) beamforming (increase of the average SNR), (ii) diversity (reducing the possibility of deep fades), and (iii) spatial multiplexing (transmission of several data streams in parallel). Due to their possibility of increasing robustness and spectral efficiency of wireless systems, MIMO has been the major research topic of wireless engineers since the mid 1990s, and it is now being implemented in high-throughput cellular and WLAN systems.

CURRENT RESEARCH


Base Station Cooperation


While the spectral efficiency gains of MIMO systems are significant for point-to-point links [1], they are limited in multi-user cellular networks by inter-cell co-channel interference (CCI). In conventional cellular systems, CCI is reduced by careful radio resource management techniques such as power control, frequency reuse, and spreading code assignments. However, these techniques limit the achievable spectral efficiency gains and/or lead to insufficient suppression of CCI. Better results can be achieved whent base stations (BS) or access points (AP) can communicate with each other through a high speed reliable connection (backbone) possibly consisting of optical fiber links, and a signal processing central unit. This opens up the opportunity data and power cooperation among adjacent BS’s, which conceptually allows multiples BS’s to be treated like a single, giant, BS.

However, there is a fundamental difference between signals from a single (co-located) large BS, and several interconnected BSs: downlink interference (i.e., signals intended for other MSs) is inherently asynchronuous to the desired signal. Perfect timing-advance mechanisms can ensure that the signals from the BSs arrive at their intended recipients synchronously. However, the BSs cannot also align all the interfering signals at each MS because
of the different propagation times between the BSs and MSs. Thus, the simultaneous arrival of both the desired and interfering signals at all the MSs is fundamentally unrealizable. Consequently, existing precoding methods are not optimum anymore, and new algorithms have to be found. This problem, together with the problem of finding good training sequences for cooperative channel identification in such large MIMO systems, is at the center of our research.


Antenna Selection


Single receive antenna selection (AS) is a popular method for obtaining diversity benefits without additional costs of multiple radio receiver chains. Since only one antenna receives at any time, the transmitter sends a pilot multiple times to enable the receiver to estimate the channel gains of its N antennas to the transmitter and select an antenna. In time-varying channels, the channel estimates of different antennas are outdated to different extents, by the time the data is decoded. Hence the conventional selection rule of maximum channel gain is clearly sub-optimal in this case. We develop new affine/linear selection rules that account of the outdatedness in channel estimates and the additive noise.

Another significant problem in AS is the energy efficiency. Training for AS not only takes up more time but is also energy inefficient since a pilot must be transmitted multiple times. It is, therefore, of interest to determine how to optimally allocate energy among pilots and data so as to minimize the SEP. We proposed (N +1)-pilot training scheme for AS. In it, the transmitter sends an extra pilot symbol after the first N pilots. The inherent robustness of AS to selection errors enables the transmitter to significantly reduce the energy it allocates to the first N pilots, which are used for selecting the antenna. Instead, it boosts the energy of the extra pilot, which yields an accurate channel estimate for the selected antenna, which helps accurately demodulate the data symbols.

In the time-varying scenario, it is not obvious whether the (N + 1)-pilot scheme remains more energy efficient. One can even argue that sending an extra pilot is counter-productive because extending the training duration makes the estimates more outdated. Our current focus is on analyzing the performance of (N+1)-pilot scheme in time-varying channels and finding the optimal power allocation across the pilots and data.


 
 

Distributed MIMO


In order to investigate the implementation issues of next generation MIMO techniques, we have developed a Software Defined Radio (SDR) test bed consisting of WARP radios.  The advantage of using the WARP radio comes from its complete system-on-chip design: each base station contains a PowerPC processor, an FPGA, 2GB of RAM, and connections for up to four RF front ends.  We have combined these radios with a powerful multi-core server that can perform offline processing as well as control experiments.  Using SDRs, we have full control over PHY and MAC layer communications, and can also experiment with techniques that require real-time signal processing using the FPGA.

The RF front ends are designed to operate in the 2.4 GHz frequency ISM band, with selectable 20 MHz bandwidth channels.  The radios use a well established 802.11 compatible chip, the MAX2829 transceiver, for conversion to and from the baseband ADCs/DACs.  There are additional amplifiers for both the transmitter (30dB gain range) and the receiver (93 dB gain range) to allow for operation in a wide range of SNRs.  All of this is controlled through the FPGA, allowing for seamless operation via the low level software.

Given the complexity of developing and debugging FPGA designs in VHDL and Verilog, the WARP uses a Xilinx Virtex FPGA, allowing the use of a rapid prototyping tool called System Generator.  Using System Generator, we can design signal processing paths in MATLAB/Simulink using block diagrams that are translated into FPGA code.  Not only does this reduce development time, it also prevents the errors inherent in human-written HDL code.  In the event that more sophisticated blocks are necessary, System Generator allows for the definition of new blocks using a subset of MATLAB code.

With the WARPs, we have the ability to implement and experiment with a wide variety of new ideas in MIMO communications.  For example, the vast majority of MIMO precoding techniques require channel state information at the transmitter (CSIT), which may be difficult to obtain from a practical perspective.  Thus there is great interest in precoding without CSIT, of which Blind Interference Alignment is one promising new technique.  BIA uses "staggered antenna switching" to create a pre-defined spatial diversity pattern, from which multiple data streams can be transmitted concurrently.  The antenna switching, however, must be done on a packet to packet basis, which necessitates the use of a hardware-controlled antenna switch.  This makes the WARP platform a perfect candidate for testing the capabilities and limits of BIA in real world environments.
 

WARP FPGA and Radio Board

 

 

SPONSORS


Under Constructions


PUBLICATIONS

 

Book Chapters

Y. Tian, C. Yang, and A. F. Molisch, “Uplink Centralized Joint Detection”, in P. Marsch, G. Fettweis (ed.), “Coordinated Multi-Point in Mobile Communications”, Cambridge University Press (2011).
 

Journal Papers


C. Yang, S. Han, X. Hou, and A. F. Molisch, “From Theory to Practice: How to Design CoMP to Achieve its Promised Potential?”, IEEE Wireless Communications Magazine, in press

N. B.Mehta, S. Kashyap, and A. F. Molisch, ”Antenna Selection In LTE: From Motivation to Specification”, IEEE Comm. Mag., in press.

H. A. Saleh, A. F. Molisch, T. Zemen, S. Blostein, and N. B. Mehta, “Receive Antenna Selection For Time-Varying Channels Using Discrete Prolate Spheroidal Sequences”, IEEE Trans. Wireless Comm., 11, 2616 – 2627 (2012).



Conference Papers

V. Kristem, N. B. Mehta, A. F. Molisch, "Energy-efficient training for antenna selection in time-varying channels,”  to appear in Asilomar Conference on Signals, Systems, and Computers, Pacific Grove, CA, USA, Nov. 2011.

V. Kristem, N. B. Mehta, A. F. Molisch, "On Training and Training Voids for Receive Antenna Subset Selection in Time-Varying Channels," IEEE Global Communications Conf. (Globecom), Miami, FL, USA, Nov. 2010. 

V. Kristem, N. B. Mehta, A. F. Molisch, "A Novel Energy-Efficient Training Method for Receive Antenna Selection,"IEEE International Conference on Communications (ICC), Cape Town, South Africa, May 2010.  

V. Kristem, N. B. Mehta, A. F. Molisch,"Optimal Weighted Antenna Selection For Imperfect Channel Knowledge From Training,"  ICC, Dresden, Germany, Jun. 2009.

V. Kristem, N. B. Mehta, "Receive Antenna Selection with Imperfect Channel Knowledge from Training," National Conf. on Communications (NCC), Guwahati, India, Jan. 2009.
 
Q. Zhang, C. Yang, and A. F. Molisch, “Cooperative Downlink Transmission Mode Selection under Limited-Capacity Backhaul”, IEEE WCNC 2012.

H. A. Saleh, A. F. Molisch, T. Zemen, S. Blostein, and N. B. Mehta, “Antenna Selection For Time Varying Channels Based on Slepian Subspace Projections”, IEEE ICC 2012, in press.