UC Berkeley

Future advancements in wireless communication standards will rely on two inter-related technologies. First, to address the saturation of traditional cellular spectrum in the $<6$ GHz bands, new and much-higher frequency mm-wave spectrum will be utilized. The mm-wave bands at 24, 28, 39, 60, and 72 GHz, among others, have tens of gigahertz of available and unused spectrum. Second, because modern coding and modulation techniques make near-optimal use of time and frequency resources, the spatial dimension of wireless channels must be exploited to send data streams in a spatially selective manner only to the desired users.

From an engineering perspective, both of these technology trends rely on the design of antenna arrays --- spatial selectivity is achieved by having a large number of radios while mm-wave operation needs directional transmissions to overcome the propagation loss. Designing large antenna arrays --- on the order of 64-256 or larger number of radios --- represents a huge departure from traditional radios used in wireless networks which have at most 4 independent radiators and transceiver chains.

This thesis explores system architecture, signal processing, and hardware design techniques which are suitable for massive antenna arrays which form many spatial beams simultaneously. First, beamforming is identified as the preferred spatial processing technique for the large array regime, and a beamforming-aware array architecture is proposed which is modular, scalable, and distributed. The core element of the proposed array architecture is a common module which integrates multiple transceivers, analog and digital signal processing, and interconnect. Additionally, a time-domain beamforming algorithm is proposed for channels with strong multipath components.

Next, synchronization architectures and algorithms are proposed for both carrier and baseband synchronization in massive antenna arrays. It is shown that uncorrelated phase noise between different transceivers causes decoherence between the front ends. Accordingly, a synchronization strategy consisting of co-optimized carrier recovery loops and distributed phase-locked loops (PLLs) is proposed to manage the total phase noise impact in antenna arrays. For baseband synchronization, a hierarchical timing alignment strategy is proposed which uses a background calibration loop to compensate for front-end and sampling skews while using distributed data timing recovery to compensate for true time delay effects in large arrays.

Finally, hardware implementations of these system- and signal-processing-level ideas are demonstrated. The first prototype consists of a system-on-chip for low-frequency massive MIMO. The proposed SoC integrates multiple full digital radios with on-chip DSP and serial links. The second prototype demonstrates the first massive, multi-user phased array operating at 72GHz. A 128-element phased array is implemented using off-the-shelf components which can form 16 simultaneous beams over 250MHz bandwidth, for an aggregate capacity of 20 Gbps.