Project

Research Areas

1. THz MIMO

Emerging applications demanding for high data rate, such as metaverse, unleash the appetite for ever-higher frequency. The propagation mechanism in this fresh spectrum, the THz band ranging from 0.1 to 10 THz, is in stark contrast to that of the lower bands. Owing to a dearth of multipath component, strong LOS path chiefly determines the throughput. Interestingly, at this distant territory, spatial multiplexing is possible with pure LOS environment without an aid of multipath components, which has been largely overlooked. Focusing on this very property, we are working on the followings:
  • Quantifying the number of DOFs we can obtain
  • Designing wireless systems to efficiently exploit this property

2. FDD Massive MIMO

Massive MIMO is now a reality. This, however, is largely limited to the time division duplexing (TDD) systems. This is because, thanks to channel reciprocity, the pilot overhead scales proportionally to the number of users, not the number of base station antennas. This is not the case when employing the frequency division duplexing (FDD); pilot overhead scales proportionally to the number of base station antennas and, even worse, limited feedback from users to the base station is required. To this end, we are looking for some sort of reciprocity in the FDD systems, which can enable inferring the downlink channel with the uplink channel. Specifically, we are working on the followings:
  • Quantifying the amount of the downlink channel information which can be extracted from the uplink channel
  • Analyzing the behavior of prediction accuracy with respect to relevant parameters
  • Deriving information-theoretic footing on the achievable spectral efficiency

3-1. Federated Learning

We are using machine learning models in everyday life. The performance of these learning models is contingent on the size of the data set. Due to privacy concerns, in many cases, training data itself is not (and should not be) available to the learning models. Instead, one can learn the global model by aggregating updates from local models while leaving the data distributed, which is called federated learning. However, preserving privacy comes at the price of an increase in communication costs owing to an ever-increasing number of devices and size of learning models. To reduce such communication costs, we are working on the followings:
  • Designing a disruptive communication system dedicated to the federated learning to transcend the conventional communication systems
  • Finding a way to countervail link heterogeneity among the devices

3-2. Semantic Communications

Shannon’s landmark paper in 1948 established the fundamental limit of communications with the messages being modelled as a random process, neglecting their semantic aspects. It is unknown how the fundamental limit of communications would be reshaped when incorporating the semantic aspects—the “meaning” of messages. We are exploring this uncharted territory. Specifically, our interest is on the followings:
  • Defining a well-posed problem on semantic communications as the research community is rather diverging in views at this moment
  • Developing semantic communication framework

4. Joint Radar and Communication

Communication and radar technologies are simultaneously used in various next-generation systems such as autonomous vehicles, satellite communication networks, and cluster flight drones. Previously, each system achieved its own purpose by using its own resource. Since both systems use electromagnetic waves to transmit and extract information, we can efficiently utilize limited frequency resources by joint design of these systems. The fundamental question is then “what is the trade-off between these two functionalities?”. To answer this, we are working on the followings:
  • Defining an information-theoretic metric for the radar system, which can be added up with the metric of the communication system
  • Designing an optimal dual-functional radar-communication system, armed with combined performance measure

5. Block Orthogonal Sparse Superposition Codes for URLLC

Although ultra-reliable low-latency communication (URLLC) is increasingly required in various applications, performance degradation of state-of-the-art channel codes in the short-to-medium blocklength regime is a crippling problem. Block orthogonal sparse superposition (BOSS) codes, developed by our group, are well-suited for short-packet communications in that
  • The inherent orthogonality in codeword construction makes it possible to develop efficient decoding algorithms.
  • It outperforms polar codes under successive cancellation list decoders
  • It provides a flexible code rate and extremely low power consumption
Future directions include:
  • Development of a more efficient decoding algorithm
  • Designing a variant of BOSS codes for multiple access channel

6. Full-Duplex Radios

Full-duplex radios, simultaneously transceiving signals in the same frequency, are one of the promising technologies for next-generation wireless systems, which can theoretically double the spectral efficiency. The signal transmitted from the device is received by itself while the desired received signal is sent by another device. This undesired signal is called self-interference and is billions times stronger than the desired signal. Accordingly, to make such a system work, the self-interference signal should be sufficiently removed, which motivates us to work on the followings:
  • Designing self-interference cancellation algorithms based on adaptive filter theory
  • Making it into reality by implementation with software-defined radio

7. LEO Satellite Network Analysis

Recent industrial breakthroughs in space launch technologies, led by some key players such as SpaceX’s Starlink and Amazon’s Kuiper, make it more economical to provide internet service via low earth orbit (LEO) satellites than ever. Compared to satellites in the higher orbits, LEO satellites have their advantages both in capacity and latency, which comes at the cost of launching a large number of satellites to provide global coverage. As satellites cannot be retrieved once launched, we need to predict and optimize the network before launching. In this regard, we are working on the followings:
  • Developing an analytic framework for LEO networks through stochastic geometry
  • PENDING