Reza Farashahi

Reconfigurable intelligent surface at 5.2 GHz

A surface made of small tuned cells that acts like a steerable mirror for radio waves. Changing the bias voltage on each cell changes the phase of its reflection, so the reflected beam can be pointed where coverage is needed.

• Designed a new dual-polarised unit cell with a single varactor diode for quasi-continuous phase control.
• Wrote a Python tool that ran HFSS design sweeps in parallel across about 50 university lab computers, using a shared network drive as the job queue. The design search ran close to 50 times faster.
• Measured fabricated cells in a waveguide set-up on a vector network analyser and compared them against simulation for both polarisations.
• Optimised the surface with a genetic algorithm, using live S21 measurements between two horn antennas.

Over-the-air link test in an anechoic chamber

After the bench measurements, I tested the surface in a live radio link. A USRP software-defined radio transmitted a QPSK signal towards the RIS, and the receiver picked up what the surface reflected. The anechoic chamber keeps stray reflections out of the measurement.

• Generated and received the QPSK signal with the USRP, with the received constellation shown live on the laptop.
• Measured the bit error rate and signal-to-noise ratio with the surface in the environment, to see how the RIS changes the quality of the link.

FPGA control board for the surface

Each group of cells needs its own precise, stable bias voltage. I compared a microcontroller, a Raspberry Pi and an FPGA for the job, and chose the FPGA for its parallel I/O and room to scale to larger surfaces.

• FPGA drives a quad 12-bit DAC over I²C, implemented in VHDL.
• An op-amp stage boosts the DAC outputs to the varactor bias range.
• Brought up and verified on the bench with an oscilloscope before connecting it to the surface.

WiPy-RT: fast ray tracing for RIS channels

A ray-tracing engine written in Python that predicts signal strength across a room or a city, including surfaces like the RIS. It follows up to six reflections plus diffraction around building edges.

Benchmarked on the KAIST urban map against WiThRay, a MATLAB reference simulator, with 200,000 rays.

KoreroNET & EnviroNode LoRaWAN sensor nodes

Solar-powered field nodes for environmental monitoring. KoreroNET listens for birdsong and classifies it with on-device AI. EnviroNode, on the same platform, reports weather and soil data. Both send their results over LoRaWAN to a web dashboard.

• Evaluated and optimised the power consumption of the LoRaWAN link on solar power. I measured current and time in each state (sleep, sensing, transmit, receive windows), then cut energy per cycle with deep sleep, compact 32-byte frames, faster data rates where the link allows, and no unnecessary rejoins or retries.
• Set up sensor management over I²C: two BME280 air sensors on separate buses, an INA219 battery monitor, detection at start-up, graceful handling of failed sensors, and automatic bus recovery.
• Leading version 2: the first custom PCB, a leaner communication scheme and an improved web app.