rfelten/rtl-spec

A Software-defined Sensor Architecture for Large-scale Wideband Spectrum Monitoring

RTL-Spec

rtl-spec is an open source implementation of the proposals done by D. Pfammatter, D. Giustiniano and V. Lenders in "A Software-defined Sensor Architecture for Large-scale Wideband Spectrum Monitoring" [IPSN15].

Installation

The following installation instructions are currently only tested for Debian-based Linux operating systems (such as Ubuntu or Raspbian).

Dependencies

libusb

libusb is a C library that gives applications easy access to USB devices on many different operating systems. rtl-spec uses the library to interface DVB-T USB dongles which are used in this project as cheap RF interfaces.

$ sudo apt-get install libusb-1.0-0-dev

librtlsdr

librtlsdr is a C library that turns RTL2832 based DVB-T dongles into cheap SDR receivers.

$ git clone git://git.osmocom.org/rtl-sdr.git
$ cd rtl-sdr/
$ mkdir build && cd build/
$ cmake ../ -DINSTALL_UDEV_RULES=ON
$ make
$ sudo su
$ make install
$ ldconfig
$ cat > /etc/modprobe.d/rtl-blacklist.conf << EOL
$ blacklist dvb_usb_rtl28xxu
$ blacklist rtl2832
$ blacklist rtl2830
$ EOL
$ rmmod dvb_usb_rtl28xxu rtl2832
$ exit
$ cd ../../

fftw

fftw is a C library for computing the discrete Fourier transform (DFT).

$ sudo apt-get install fftw-dev

RTL-Spec

Building

The software can be built as follows:

    make <TARGET> [CFLAGS="<CFLAGS>"]
    <TARGET> = sensor_cpu | sensor_gpu | collector
    <CFLAGS> = [-O2] [-ggdb] [-DVERBOSE] [...]

First, select the target you want to build. There are two options for building the sensor and a one for the collector. The target sensor_gpu compiles the sensor software for usage on dedicated hardware, i.e. the Raspberry Pi (RPi). This will lead to some CPU intensive tasks, such as the FFT, being rolled out to the RPi's VideoCore IV GPU, improving overall sensing performance. Note that for FFT computations on the VideoCore IV we rely on the library GPU_FFT, which typically comes preinstalled on Raspbian OS. In case you don't want to compile the sensor software for dedicated hardware, select the target sensor_cpu. The FFT will then be computed on general purpose CPUs using the FFTW library.

Second, you can choose gcc's compilation flags. Compiling any of the targets with flag -DVERBOSE will provide additional debugging information on stdout.

An example for building a collector and sensor instance on the same machine is given below:

$ cd rtl-spec/
$ make collector CFLAGS="-O2 -DVERBOSE"
$ make sensor_cpu CFLAGS="-O2 -DVERBOSE"

Running

Here, we provide the simplest example for running the collector and sensor instances built above. The collector listens on localhost port 5000 for incoming sensor data and dumps it to the local file system. The sensor monitors the frequency spectrum between 24 and 1766MHz and transmits the recorded samples to the afore mentioned collector.

$ ./run_collector 5000
$ ./run_cpu_sensor 24000000 1766000000

Run collector and sensor with option -h to learn more on the individual configuration options.

$ ./run_collector -h
$ ./run_cpu_sensor -h

Description

Processing Steps

Multiple spectrum sensing nodes can be served by a single (remote) collecting unit. The RF spectrum data recorded by sensors is transmitted over TCP to the associated collector which dumps the received data to the local file system. The following figure highlights the involved processing steps:

Data Format

The dumped data is stored in the following format:

Seconds since UNIX Epoch [secs]	Timestamp Extension [microsecs]	Frequency [Hz]	Squared Magnitude Value [dB]
1423490796	854275	23996876	-33.9
1423490796	854275	24006250	-20.6
...	...	...	...

Wideband Scanning Strategies

To overcome the sampling rate hardware limitations of low-cost radios (such as DVB-T USB dongles), the sensing software includes different wideband scanning strategies. Currently, you can choose from the following three options:

• Sequential: Sequentially sweep over the band of interest in steps limited by the sampling rate
• Random : Inspect frequency bands (limited by the sampling rate) in a random fashion
• Similarity: Visit frequency bands of particular interest more frequently than others

For more information, please refer to "A Software-defined Sensor Architecture for Large-scale Wideband Spectrum Monitoring" [IPSN15].