Digital Self-Interference Cancellation Methods in 5G and Beyond STAR Systems

Simultaneous transmit and receive (STAR) systems enable low delays, which is one key element in current 5G New Radio (NR) and future wireless technologies. However, operating the transmitter (TX) and receiver (RX) simultaneously leaves the transceiver susceptible to so-called self-interference (SI), which is the transmitted signal received by the transceivers own RX. SI is in general considerably higher in power than the relatively weak signals from other devices the RX is attempting to pick up, and therefore SI can be harmful for the transceiver. Hence, this thesis focuses on digital signal processing approaches to cancel the SI from the RX data in both frequency division duplex (FDD) and inband full-duplex (IBFD) systems. In addition to merely cancelling the SI, low computational complexity in the SI mitigation models is sought to facilitate real-time operation and low energy consumption.

In FDD, which operates the TX and RX on different frequencies, the focal point is on passive intermodulation (PIM), which generates the nonlinear SI from passive devices, which can thus leak to the RX band. Specifically, so-called air-induced PIM is considered, where a passive device outside the transceiver chain generates the PIM SI. Such SI is studied in systems, where multiple parallel TX chains transmit more than a single signal, through a technique called carrier aggregation (CA). In this framework, a wide variety of digital cancellation solutions are presented, first with single- block batch estimation models, and later with cascaded models employing gradient- based parameter adaptation. All of the models are tested with signals measured with commercial-grade base station (BS) equipment, where digital PIM suppression of up to 20 dB is evidenced, pushing the residual SI close to the noise floor. In addition, the complexity assessment reveals that the introduced canceller models require noticeably less computations to execute, compared to the basic single-block modeling. The emerging IBFD systems run the TX and RX on overlapping frequencies, which can double the spectral efficiency compared to traditional approaches. In this context, the cancellation of the SI is first done in real-time via a field programmable gate array (FPGA) implementation of a cascaded model canceller, and later the canceller model is extended to support CA, in a potential example of future flexible- duplex scenarios. It is shown that the real-time FPGA canceller can achieve more than 40 dB of SI cancellation, and facilitates a sum-rate increase of up to 90 %. The introduced canceller developed for the flexible-duplex scenario achieves some 35 dB of SI cancellation, with reduced complexity compared to a polynomial reference method.