Okay, so ... FPGA designs aren't really software. Perhaps this should more appropriately be named a "Gateware Defined Radio" project--but that spoils the idea.
The goal of this project is to build a gateware design that can transfer audio from one radio to another. The design is built around the icebreaker FPGA board, the SX1257 Radio PMod, the Digilent MEMS microphone and the Digilent AMP2 PMod. You can see a picture of this setup here.
This repository actually contains several designs. There's an AM transmitter, an FM transmitter, a QPSK transmitter, an AM receiver, an FM receiver, and a QPSK receiver. There are also three composed designs, which compose the transmitter and receiver together for simulation (only) purposes. Which design gets built into the repository is controlled by which "RF" line in the AutoFPGA Makefile is uncommented.
To build, you'll first need AutoFPGA
installed and in your path. You can then run make autodata to build the
top level design, PCF,
simulation, and software data files. From
here, make rtl will build the design, make sim will build the simulator,
and make sw will build the host support software.
Beware, if you use any of the composed simulations, such as the AM transmitter
followed by the AM demodulator, you will run into a problem where the
microphone and the amplifier are both attempting to use the same pins. The
solution is not to build the composed (transmit/receive) design in rtl.
Instead, run make rtl-sim to build the full simulation.
To run the design in simulation, simply run the main_tb file in the sim/
directory. This isn't very useful at present, however, unless you add the
-d switch to turn on VCD file generation. You can then examine all the
traces from within the design.
Be aware, the simulation has no channel model. Outputs to the SX1257 are simply fed back into the simulation receiver.
To run the design on an icebreaker, you can load the design via
sudo iceprog rtl/sdr.bin. Once you do that, you'll want to then run
sw/netuart /dev/ttyUSB? (replace with your serial port device). At this
point, you can interact with your design using wbregs. Perhaps more
importantly, you can interact with the registers on the
SX1257
using the rfregs. In particular, txconfig.sh will turn the transmitter
on, and set it up at 915MHz.
This design offers two methods for debugging: capturing traces, and capturing histograms from within the data flow. Two bits of the GPIO register (three for the TX/RX composed simulations) control a tap point selection within the design. This tap point contains a CE signal, 32'bits of data, and 10 bits of data for a histogram.
To record a trace from wtihin the hardware, map the signals you wish to capture
to a Wishbone Scope C++ file--such as this one for the
microphone. You can then use wbregs rfscope 0 to reset
the scope any time you want to make a new capture, and
micscope to extract the captured dasta and then to turn
it into a VCD file you can then analyze with GTKWave.
Alternatively, you can capture histograms from within the design. These can be read with the histogram program. This program will also output a binary file of 32-bit integers containing your histogram. (The histogram is currently hardcoded at 1024 words, due to hardware limitations in the iCE40 up5k.)
With a little creativity, the histogram capture utility can be turned into a constellation capture utility. By sending 5-bits of I and 5-bits of Q into the histogram, and then interpreting the results accordingly, you can capture (and then plot) a constellation diagram. This is the purpose of the constellation program.
As of 20200204, ...
- The AM receiver works when composed with the transmitter in simulation. The simulation has no channel model, is only so good, etc., etc. It's a start.
- Both FM transmit/receive pair work when composed together in simulation.
- The AM transmitter works, and the results can be heard using a Lime SDR radio receiver. There's an annoying tone present which I haven't yet chased down.
As of 20200727, there's now a QPSK transmitter and receiver which both (roughly) work in simulation. I say roughly because the carrier tracking loop breaks lock even in simulation. Further, the filters used in these designs are horrible block average types of filters. Hence, while the design can recover the incoming audio, there's still lots of room for improvement.
This project and all of its files are licensed under the GNU General Public License, v3.