Low-Latency Massive MIMO Design over Practical Channels
Doctor of Philosophy
With the emerging time-critical applications, low-latency communication is the focus of the next-generation wireless network, which is labeled as 5G. One of the key enabling technologies of 5G is massive MIMO or base-stations with tens to hundreds of antennas. In this thesis, I design cross-layer transmission control for massive MIMO and demonstrate that high throughput can be achieved at less than 1 ms latency. The latency minimizing transmission control design requires the correct understanding and abstraction of the physical radio propagation. To understand the unique propagation properties of massive MIMO, I measure the wireless channel in-the-wild with RiceArgos platform. Surprisingly, I find that geolocational far-away user channels can still be of high correlation even for the base-station with 64 antennas. More specifically, the mutual channel correlation between close-by users does not reduce with array size beyond 20. Using spatial signal processing and modeling, I pinpoint the root cause of the observed user correlation to be the overlapped angle-of-departures. Furthermore, I propose new channel models to capture the measured user channel correlation and then derive the achievable rates in closed-form for zero-forcing and conjugate based systems in the large-array asymptotic regime. The measured high user correlation is in direct conflict with the current point-to-point based channel models, where user channels are assumed to be independent. Therefore, my measurement findings and analysis collectively demonstrate the need for new multiuser channel modeling and open doors for better multiuser system designs. To understand the structure of the latency-optimal control, I perform cross-layer optimizations for single-user massive MIMO systems. I first present a Markov decision process based algorithm which computes the latency-optimal control at the cost of high computational complexity. To reduce complexity, I propose a low-complexity deterministic transmission control, Large-arraY Reliability and Rate Control (LYRRC). At the large array regime, LYRRC not only can achieve any throughput point inside the feasible rate region but also leads to optimal latency, which diminishes with the array size. Using simulations based on over-the-air channel traces, I demonstrate that LYRRC achieves 20X latency reduction over the state-of-the-art design. In addition, for users with near orthogonal channels, I demonstrate that LYYRC provides the latency-optimal transmission control for multiuser massive MIMO. However, LYRRC cannot be directly extended to the more challenging multiuser massive MIMO systems where high user correlation can exist. To guarantee low-latency over practical channels with user correlation, I propose a new low-complexity stochastic optimization framework that carefully balances queueing, re-transmission, user correlation, and throughput. The proposed cross-layer control consists of three key components: a). MAC that generates user selection by balancing queueing with user correlation; b). Transmission rate adaption that balances throughput with re-transmission; c). Congestion controller that balances queueing with throughput. Using the measured channel traces, I evaluate the proposed low-complexity cross-layer framework. My simulations show that the proposed scheme can achieve over 83% of the system throughput capacity with less than 1 ms latency. In addition, I find and demonstrate that LTE-type fixed rate adaption based design leads to protocol deadlock in real-world massive MIMO systems, which leads to almost no throughput and large latency.