Extreme MU-MIMO

---

The capacity of traditional single-user wireless systems is exclusively limited by the available spectrum and transmission power. Recent developments in information theory, however, have shown that such capacity limitations can be overcome by improving the spatial reuse efficiency through multi-user multiple-input multiple-output (MU-MIMO) technology, or its special form called multi-user beamforming (MUBF). With MUBF, a base station employs multiple antennas to simultaneously send independent data streams to multiple users, tremendously increasing the aggregated network capacity.
As MUBF theory dictates, the more antennas a base station has, the more users it can serve simultaneously. Encouragingly, this means that with enough base station antennas, the network capacity can be scalably grow to accomodate more users. This naturally motivates us to put as many antennas as possible on the base station to meet the increasing capacity demand from more and more users. However, to practically implement MUBF theory and realize its capacity benefit, one must address a set of challenges.
In the Argos project we develop new techniques that enable hundreds of antennas to be deployed on base stations, resulting in enormous network capcity gains. We use the WARP platform to build the first real-world prototype of a many-antenna MUBF base station and experimentally evaluate its performance. Our research demonstrates the feasibility of many-antenna base stations, and is designed to motivate industry adoption of this promising technology in the near future.

Hide

Challenges

---

We must address serious practical challenges in order to realize many-antenna base stations. The first, and also the most important step, is to build such a base station with hundreds of antennas. In particular, the large number of antennas leads to the following system challenges:
(1) MUBF, especially zero-forcing MUBF (ZF-MUBF), requires non-trivial baseband processing. How do we design a base station architecture that properly partitions computation in order to easily and flexibly scale with the number of antennas?
(2) MUBF is extremely sensitive to fluctuations in wireless channels; therefore the channel state information (CSI) must be updated both quickly and accurately. How do we efficiently collect full CSI for a large number of base station antennas and users?
(3) MUBF requires accurate clock and transmission synchronization to ensure correct phase and symbol alignment for the signals sent from the multiple antennas. How do we synchronize the clock and transmission for a large number of antennas?
Argos is our proposed base station architecture that solves the above challenges. Briefly speaking, Argos adopts a fat-tree structure with daisy-chained lead nodes to improve scalability and reliability. We propose a local conjugate beamforming method, which gracefully distributes the processing to each antenna with close-to-optimal performance. We also devise a novel internal relative channel calibration procedure which enables full CSI to be collected very quickly. Finally, we employ a a central clock distribution board to accurately synchronize the all antennas. These techniques collectively lead to a practical base-station which can scale to hundreds or even thousands of base station antennas.

Hide

Results

---

We use our WARP-based prototype to show the capacity and energy efficiency benefit of Argos. Specifically we have the following findings in our experiments (where M is the number of antennas and K is the number of terminals):
(1) With M = 64, the improvement for conjugate and zero-forcing MUBF over a single antenna are 5.7x and 12.7x for equal power, or 3x and 6.7x for 1/64 power, respectively.
(2) As M drops to K, i.e., M ≈ K = 15, the performance of zero-forcing drops steeply.
(3) When M is much larger than K, the capacity increases approximately linearly with K for both conjugate and zero-forcing MUBF.
(4) As K approaches M, the performance of zero-forcing drops sharply.
(5) When the transmission power is reduced, conjugate MUBF performs relatively better than zero-forcing. (6) In all cases, our local conjugate beamforming achieves near-optimal performance.

Hide