It is a well-known fact that Shannon’s Communication system model is serving as the backbone of all modern communication systems including the latest fifth generation (5G) cellular networks. However, efficient implementation of a reliable beyond 5G (B5G) wireless networks for all three communication modalities, namely, Human-to-Human (H2H), Human-to-Machine (H2M), and Machine-to-Machine (M2M), has a larger set of requirements which is indicating the need of upgrading the Shannon’s Model itself. Introducing goal oriented design, utilizing the information residing in semantic contents of the message, creating and utilizing knowledge banks at the source and destination, unification of information generation, transmission and usage, emphasizing semantic error correction, etc. are some complimentary provisions envisioned over and above the Shannon’s model, which is collectively referred as ‘Semantic Communications’. The advancements in the technologies like Artificial Intelligence and Internet of Things, availability of significant storage and computational capabilities even in the handheld wireless devices, and many other similar developments can speed up the development of Semantic Communication. Observing the potential of the Semantic Communication system model further research is needed towards defining a robust mathematical framework for end to end semantic communication including the microscopic and macroscopic assessment of semantic value of the messages, identifying the characteristics of involved semantic noise along with the ways of nullifying its impact, efficient implementation models for goal oriented communication, upgrading the Machine Learning techniques from traditional pattern matching to the level of understanding and reasoning, Developing suitable Semantic Metrics, and Semantic Aware Multiple access, etc.

Despite the large available bandwidth and data rates provided, mmWave suffers from limited coverage and capacity due to the severe path loss and penetration loss experienced. One approach is living with the limited coverage via network densification by deploying more cellular base stations with smaller coverage in a given area. The large number of base stations require huge amount of CAPEX/OPEX cost for establishing their backhaul connections, especially if fiber backhauls are considered. Integrated access and backhaul (IAB) technology has emerged as a cost-effective alternative, where the base stations relay the backhaul traffic wirelessly. Sharing the radio resources between backhaul and access links orthogonally, time or frequency domain duplexing can be performed. In order to fully exploit the radio resources and double spectral efficiency, hence double the rates of backhaul and access links, full duplex (FD) communication can be applied via self-interference (SI) cancellation techniques, employing antenna suppression in propagation domain, analog cancellation at the RF level and digital SI cancellation at the baseband. In this talk, we present our digital SI cancellation solution for OFDM with a switched FD radio architecture along with neural network based frequency-domain non-linear estimation algorithm guiding time and/or frequency domain linear cancellation techniques. The proposed architecture can be implemented by only simple modification on an OFDM transceiver, allowing non-linear SI estimation to be performed separately from and prior to linear SI cancellation, so that the estimated non-linear SI signal is provided as a reference to linear cancellation. Experimental results obtained on OFDM based software defined radio set-up demonstrate that, with the proposed solution, non-linearity problem is alleviated and SI is reduced to the noise level for almost all transmit power levels. Moreover, in the proposed solution, the non-linear model needs to be trained once at power-up;  hence the performance is not affected by changes in the multi-path environment, unlike existing schemes, which require re-optimization of model parameters and re-estimation per setting. With the same reasoning, the estimation overhead is totally diminished to zero and complexity is significantly reduced, to that of linear estimation, making our solution promising for digital SIC for FD IAB basestation.

Wireless and mobile communication, sensor networks and Internet of Things (IoT) would form the backbone to create pervasive and ubiquitous environments that would profoundly influence society and thus are important to society. The wireless technologies such as IEEE standards, and Internet of Things (IoT) would encompass a wide range of domains such as hardware devices, routing algorithms, heterogeneous network elements, machine learning techniques, cloud computing and blockchain technology, etc. Although we have seen a significant deployment of IoT networks these days, there are many domains which require further research to reap the full potential of IoT based applications. There is an increasing demand in the industry to comply to Industry 4.0 (also known as "Industrie 4.0"), after the introduction of Industrial IoT (IIoT). Therefore, a lot of research is required at different layers of TCP/IP reference model. Further, existing IoT protocols, machine learning techniques, cloud computing, data storage and blockchain require newer standards and protocols for application specific implementations. Therefore, in this talk, I will discuss the pressing research issues in the IoT ecosystem and how we can overcome these problems for wide deployment of IoT networks in various application domains.

The report presents an account of combining machine learning and model-based development to design an algorithm for groups of targets classification, used in a radar system. The development of this component in an extremely short timespan was made possible by modeling an environment used for data synthesis and by using a graphical modeling tool to train a number of different neural networks. The report also lists other tasks, where machine learning has been tested in developing software for radar systems.