Quantum and Quantum-Inspired Computation for NextG MIMO Wireless Communications

This extended abstract outlines our research on quantum and emerging computing systems for next-generation wireless networks. The research aims to leverage quantum and quantum-inspired computation to expedite baseband processing at base stations. We introduce our system design directions and prototype systems that are implemented on analog quantum processors. The prototypes are designed for quantum-accelerated near-optimal multi-user detection processing in MIMO systems that could drastically increase wireless performance for tomorrow's NextG cellular networking standards, as well as in NextG wireless local area networks (LAN).


INTRODUCTION
A central design challenge for future generations of wireless communications and networks is to meet users' ever-increasing demand for capacity, throughput, and connectivity.Recent advances in the design of wireless networks to this end, including the nextgeneration (NextG) efforts underway, call in particular for the use of Large and Massive multiple input multiple output (MIMO) antenna arrays to support many users near a base station (BS) in a cellular network or an access point (AP) in a wireless LAN.These techniques are coming to fruition, yielding significant performance gains, spatially multiplexing information streams concurrently.
In MIMO systems, the receiver requires signal processing to disentangle the mutually-interfering streams from each other, a technique called MIMO detection.Current MIMO systems use simple linear detection methods such as Zero-Forcing (ZF) and Minimum Mean Square Error (MMSE) filters under the Massive MIMO regime where a much smaller number of users are supported at a time than base station antennas.The number of concurrently serviced users needs to keep increasing for higher gains, but commonly used simple linear detection algorithms suffer from rapid degradation in detection performance.This is because when the number of user antennas approaches the number of base station antennas, MIMO detection becomes extremely difficult resulting in poor performance for conventional detection linear algorithms [2]: this is the Large MIMO regime that lies along the points where the number of users   equals the number of base station antennas   , as depicted in Figure 1.For Large MIMO, there exist maximum-likelihood (ML) exact optimal solvers such as Sphere Decoder (SD), that can achieve the lowest possible bit error rate and, therefore, restore a high throughput.Unfortunately, these optimal detection algorithms come at the expense of an exponential increase of the required computational resources as MIMO size increases, eventually becoming infeasible for many users because of the processing time limits in wireless systems.For example, at most four milliseconds of BS's computation are available for both the 5G uplink and downlink.
Over the last few years, there has been a surge of interest in alternative computation approaches to reduce the complexity of the optimal ML detectors by leveraging algorithms that relate optimization convergence to Physics principles.This interest is further accelerating in view of experimental initiatives featuring hardwarenative implementations of these approaches, using both quantum and classical physics-based computations.One common aspect of these algorithms is that they frame the computational problem as an energy minimization problem of a magnetic spin system, also known as the Ising spin model.In this regard, physics-inspired algorithms can be seen as parametric "black boxes" that accept an Ising spin problem as input, and output the configuration with the lowest associated energy.What distinguishes one algorithm from another is the underlying mechanism used to find the global minimum, which corresponds to the ML optimal solution in MIMO detection.
In this extended abstract, we overview our design directions of quantum and quantum-inspired computation-enabled MIMO wireless systems: (1) quantum optimization on specialized hardware, (2) hybrid classical-quantum computational structures, (3) quantum-inspired computing on classical computing platforms, and (4) scalable and flexible parallel quantum optimization.For the directions, the prototype systems are designed and implemented to investigate their feasibility.The prototypes aim to expedite the ML detection in MIMO systems to potentially enable Large MIMO.For envisioned scenarios, we imagine quantum-enabled Centralized Radio Access Network (C-RAN) architectures where quantum processors are co-located in a centralized baseband unit (BBU) pool along with classical processing resources [4,7].

PROTOTYPE SYSTEMS
We hypothesize that the growth trends in wireless network demand will continue and so Large MIMO with optimal ML methods will soon become essential.Towards this, the aforementioned nontraditional computing methods have been studied with a focus on enabling fast and efficient ML MIMO detection.In this section, we introduce our prototype systems undertaken to realize this vision.
(1) We presented QuAMax [4], the first prototype quantum MIMO system implemented on a real-world analog quantum processor, D-Wave 2000Q quantum annealer, to speed up the computation required for the ML MIMO detection by taking advantage of a quantum algorithm, quantum annealing (QA).The research introduced a novel way of transforming the problem into forms that quantum(inspired) computing algorithms can solve (i.e., ML-to-Ising reduction) and tutorial-like design guidance.QuAMax demonstrated that QA MIMO processing can potentially enable near-optimal detection performance even for some currently challenging Large MIMO regimes.The research also experimentally tested advanced QA techniques such as anneal pause and extended hardware constraint range with real-world applications for the first time.
(2) To take advantage of both classical and quantum processing, we presented IoT-ResQ [6], a hybrid detector system that leverages quantum reverse annealing (RA) algorithms, in the context of IoT technologies.Unlike standard QA algorithms used in QuAMax, IoT-ResQ's RA starts its optimization operation on a controllable candidate classical state, instead of a quantum superposition.This procedure (called a warm-started operation) provides an opportunity to utilize both classical-and quantum-based detectors together in a hybrid synergy [5].IoT-ResQ consists of an initial classical non-linear detector and RA.By limiting quantum fluctuations only around the initial state, IoT-ResQ allows refined local quantum search and thus improves quantum optimization.Hence, IoT-ResQ was able to scale up IoT connectivity significantly, compared to both fully quantum QuAMax and conventional classical detector systems.The research has paved the way for RA-based hybrid classical-quantum computing detector system designs.
(3) While quantum-based detectors were able to show great promise of accomplishing the next levels of wireless performance, we realized that quantum hardware may not always be available (e.g., in wireless LAN or Wi-Fi).Thus, our research question of interest was how can we make use of similar computing benefits on classical platforms?There exist generic quantum/physics-inspired algorithms that can be implemented on any computing platform.ParaMax [2] is a MIMO detector system that exploits physics-inspired simulated annealing (SA) and parallel tempering (PT) algorithmic techniques, implemented on general-purpose CPUs and GPUs.ParaMax operates flexibly in parallel for any number of available processors, supporting fixed latency and scalable parallelism.The evaluations have shown that ParaMax requires several orders of magnitude fewer processing elements to achieve near-optimal detection performance in some extreme MIMO regimes (e.g., over 100 × 100 MIMO), compared to a conventional parallel compute detection system.(4) Multi-core architecture-based detection parallelism is a promising direction in that the computations required for ML detection can be split into multiple parallel subtasks to reduce overall compute times at the expense of computation resources.We believe the same motivation holds for QA MIMO detectors [1], and thus designs of highly efficient and scalable parallelization strategies for QA MIMO detection need to be investigated in light of the expected largescale qubit processors.X-ResQ [3] is a QA-based detector system featuring flexible QA parallelism that is uniquely enabled by RA, a method called multi-seed parallel ensemble RA.Unlike conventional parallelization designs (e.g., decomposition approach in IoT-ResQ), X-ResQ's design has many desirable features for parallel quantum systems, elastically achieving gains with fine-grained parallelism and thus requiring fewer levels of parallelism (i.e., fewer qubits) to obtain near-optimal detection performance.We also implemented and evaluated X-ResQ in the non-quantum digital setting.This non-quantum X-ResQ showcased the potential to realize previously impossible ultra-large 1024 × 1024 MIMO, greatly outperforming even ParaMax for the same compute time and levels of parallelism.

CLOSING REMARKS
We take a long view that quantum and quantum-inspired computing can resolve the computational challenge in wireless networks and thus enable the next levels of wireless performance.As an initial guiding framework for experimental research on this new approach, our work takes major steps with a series of solutions toward accelerating ML MIMO processing via quantum-inspired classical computing and analog quantum computing.With the prototype systems, we have experimentally demonstrated their substantial achievable performance gains in many aspects of wireless networks.While the value of quantum and quantum-inspired computing in practical wireless systems scenarios remains largely unproven and many challenges unsolved, we believe the understanding of the methods in these early days could result in transformative impact.

Figure 1 :
Figure 1: Fundamental MIMO regimes and approximate feasibility of various detection approaches.