My research focuses on characterizing the fundamental limits of low-latency communications over the Internet. I have investigated the maximum coding rates for low-latency communications over a wide range of wireless channel models including multiple access, broadcast, relay and multiple-input multiple-output (MIMO) channels. In addition, I have discovered a new class of optimal error-correcting codes for interactive real-time streaming over UDP. In recent years, I have investigated ultra-reliable low-latency strategies for industrial Internet-of-Things (IoT) in 5G/6G where the strategies are based on device-to-device (D2D) communications. From the system perspective, I have developed a novel and fast algorithm for obtaining an optimal 5G/6G network deployment involving base stations and repeaters.
Repeater technologies can be used for extending the coverage of a 5G/6G microcell at low cost. A repeater actively or passively forwards the signals between a user and a base station through the access link and backhaul link, where the access link is a wireless connection between the repeater and the UE and the backhaul link is another wireless connection between the repeater and the base station. A novel and fast algorithm based on MILP (mixed-integer linear programming) is developed in [J26] for obtaining an optimal wireless network deployment consisting of base stations and repeaters, which is useful for delivering high-speed fixed wireless access (FWA) and enhanced mobile broadband (eMBB).
In a typical industrial setup, there are several sensor/actuators (S/As) that are being controlled by Programmable Logic Controllers (PLCs) through command/response cycles in a deterministic and synchronous traffic pattern. The Fourth Industrial Revolution (Factory 4.0) will use 5G/6G technologies to enable wireless communication among the S/As and PLCs, which can achieve reconfigurable factories, reducing installation and maintenance of wires, and allowing deployment of innovative technologies that require mobility, e.g., automatic guided vehicles (AGV) and robotic arms. Original D2D designs for industrial IoT have been proposed [P2 - P31] to improve the efficiency, flexibility, and scalability of manufacturing automation in factories. In a particular setup [C26], the use of D2D communications can support 80% more IoT devices and consume 40% less resources.
The future wireless networks promise to make high-throughput at low-latency ubiquitous. This enables new applications such as high-quality video conferencing, virtual reality (VR) and Internet-of-things (IoT) applications including vehicle-to-vehicle communications (V2V) and mission-critical machine-type communications (MTC). Streaming codes, compared to Automatic Repeat Request (ARQ) schemes, are more amenable to low-latency communications for error control because they do not require retransmitting lost packets.
Point-to-point channel
Low-latency streaming codes over a packet erasure channel is investigated in [J18], which discovered low-latency convolutional codes with optimal tradeoff between the capability of correcting burst erasures and the capability of correcting isolated erasures. Mathematically speaking, the code construction achieves the maximum coding rate for convolutional codes with any given column distance, column span and decoding delay. The optimal convolutional codes outperform existing streaming codes over common statistical models for communication networks, including the Gilbert-Elliott channel and the Fritchman channel. Two explicit constructions of the optimal convolutional codes are given in [J20], [J23]. A novel network-adaptive streaming scheme based on an explicit construction of the optimal convolutional codes over GF(256) is proposed in [C25], [J24], which is shown to outperform traditional streaming schemes (based on MDS codes) in real-world experiments.
Relay network
Low-latency streaming codes over a three-node relay network consisting of two erasure channels is studied in [J21], which proposed a symbol-wise decode-forward (DF) scheme that outperforms traditional message-wise DF strategies. It is interesting to note that symbol-wise DF is not always optimal [J25] if the encoding at the relay can adapt to the erasure pattern of the first channel.
Two streaming messages with different deadlines
Multiplexed erasure codes for two streaming messages with different decoding delays are investigated in [J22]. If the erasure channel introduces only a burst erasure of a fixed length, then the capacity region is fully characterized.
The proposed streaming codes in the above works can be used in the transport layer to improve the performance of UDP (User Datagram Protocol).
The additive white Gaussian noise (AWGN) channel is an ubiquitous model for wireless communications. One common approach to measuring the performance of an AWGN channel is to study the tradeoff between the communication rate and delay under a fixed acceptance level of the probability of communication error ε. The 2nd-order coding rate directly impacts the aforementioned tradeoff. Bounds on the 2nd-order rate have been established for the following AWGN channels:
Parallel Gaussian channels (as well as MIMO channels) with feedback [J14]
AWGN channels with polar coding schemes [J13]
The new multiple-input polar code [J13] outperforms all existing binary-input polar codes for low-latency communication over the AWGN channel.
Wireless sensor networks (WSNs) consisting of energy-harvesting nodes will be used in 5G for providing ultra-reliable and low-latency communications. The simplest WSN can be modeled as an energy-harvesting (EH) channel. Rate-delay tradeoff analyses for EH channels have been conducted in the following studies:
In [J9], [J16] and [J19], the save-and-transmit and best-effort strategies for EH channels have been investigated and achievable rates subject to a fixed delay have been obtained.
In [J10], save-and-transmit polar codes for EH channels channels have been investigated, which shows that the additional EH constraints do not affect the scaling of the 2nd-order coding rate (scaling exponent).
Common practical multi-user communications include downlink, uplink and multi-hopping transmissions. The downlink, uplink and multi-hopping transmissions can be modeled by the multiple access channel (MAC), the broadcast channel (BC) and the relay channel respectively. One common approach to measuring the performance of a practical multi-user network is to study the tradeoff between communication rates and delay under a fixed acceptance level of the probability of communication error ε. The aforementioned tradeoff is closely related to the optimal achievable rates for infinite-delay communications under a fixed error probability ε (known as the ε-capacity regions in the literature), which have been characterized for the following networks:
Gaussian MACs [J6]
Gaussian MACs with feedback [J11]
Gaussian BCs [J15]
Gaussian degraded relay channels [J12]
Relay networks with tight cut-set bound [J7], [J17] including degraded relay channels, wireless erasure networks, linear deterministic networks, and networks consisting of independent DMCs.
The above results illuminate researchers to find effective coding schemes for low-latency communications over cellular networks.
Relays are important components in wireless networks and heterogeneous networks where simultaneous transmissions from different sources can interfere with each other. A low-complexity linear relaying strategy is proposed for the two-hop interference network with two sources, two relays and two destinations [J3, P1]. The linear relaying strategy significantly outperforms the traditional routing strategy and attains the optimal degrees of freedom (DoF) among all linear relaying strategies. The strategy can be applied to practical wireless relay networks with multiple independent information flows to increase their data rates.
In a general communication relay network, the amounts of delays incurred by the nodes/edges affect the achievable rates, which have been studied in the following works:
Cut-set bounds for relay networks with zero-delay nodes/edges have been established in [J2] and [J8].
Surprisingly, for some broad classes of relay networks identified in [J4] and [J8], the amounts of delays incurred by the nodes/edges do not affect the sets of achievable rates.
In a relay network where a source wants to multicast a stream of messages to different destinations through multiple relays, the coding rate of the source may change over time to adapt to varying network conditions. In order to facilitate efficient relaying, a linear network coding scheme is proposed to support variable-rate multicasting where the relays can keep the relaying strategies unchanged over time [J1][P1]. The strategy can be used for low-complexity low-latency relaying in content delivery networks.