More amazing talks from day 3 for GRCon20, Im still working on the CTF but got stuck fairly quickly… I would still encourage everyone to take a look at the signals in GnuRadio, hopefully they will release solutions, not just the flag values but how they got to them!
Schedule
Full Conference Schedule 9/14 – 9/18
Full Live Video Day 3
Cool Talks Day 3
How Strong is my SDR Signal? – Martin Braun
Ever want to know how strong your signal is in the real world when using an SDR? In this talk Martin goes through the math as well as the reality behind calculating this value. With the use of real SDR hardware we can see how transmitting from one USRP SDR into another affects the received signal power based on the receiver reference (associated with its gain) as well as artifacts that can effect the measurement. The other plot that he shows is how poor amplification of the same signal can distort the receivers ability to calculate signal power.
Teaching the Principles of Time Delay Spectrometry Ultrasound with GNU Radio – Paul Otto
GnuRadio is mostly used for communications but it can also be used in other domains for signal processing. This talk goes through how GnuRadio can be used for Ultrasound. Using GnuRadio to create chirp signals for finding distance he shows a couple of simulated demos to graphically display how the signal gets processed into range data.
Moving on from the simulated demos he finds some interesting quirks with trying to use a laptop speaker and mic as a substitute for low cost testing but then switches to HackRF One and the RTLSDR but is having issues connecting the impedance of the SDRs to the Ultrasound transducers.
Deep learning inference in GNU Radio with ONNX – Oscar Rodriguez; Alberto Dassatti
As Deep learning and Machine learning keeps giving momentum it is extending into almost every technology area. In order to support deep learning inference in GnuRadio we need processing blocks capable of running the inference portion of a DL design. This is where ONNX and the new OOT module gr-dnn. ONNX (open neural network exchange) is an open format, framework independent, widely supported ML format that is compatible with a bunch of the ML tools, such as Pytorch, tensor flow, Matlab and many more. ONNX defines basic operations to translate other ML formats for use common use. It also provides an abstraction layer between the model and the inference engine allowing for hardware acceleration on GPU, FPGA, etc.
The team developed gr-dnn OOT module for GnuRadio implemented in python3 so only GnuRadio 3.8 and above. The block creates the dimensional data format for the models input based on block input size parameter as data streams in. Then takes the models output data and streams it out. Learn more about it in their paper and their code.
