The following is a Python wrapper to control the Spectrum M4i6631-x8 arbitrary wave generator:
- PCI Express x8 Gen 2 interface
- 2 channels 16 bit 1.25 GS/s,
- +/-2.0 V into 50 ohm,
- 2 G Sample memory
The board comes with Windows and Linux drivers as well as a programming interface (API) that can be implemented in various languages. The instructions for downloading the software are located here.
Potentially useful information:
- The username/password for the computer where this is currently installed is Eli, S12D52qubit
- The drivers may require a specific version of Ubuntu
- The documentation for the API is located:
SPCM-418/Doc/english/hwmanuals/m4i_m4x_66xx
Overview: in order to generate pulse sequences, our python script must (1) establish a connection with the Spectrum AWG board, (2) calculate the array of voltages that the AWG DAC should output, (3) send this data to the AWG board, (4) enable triggering of the AWG.
All four steps are implemented and wrapped into a single class SPCM-418/Examples/python/spcm_m4i_6631-x8.py. While not incredibly well documented, the code should be understandable. Note: currently the default (and only) triggering behavior implemented is TTL, but more advanced option are available (see the API documentation).
Notably, a unique aspect of my implementation is during steps 2-3. In order to write data (i.e. the waveform) to the board, we must place those values into a specific buffer that is given to us by the API. Furthermore, the waveform is an extremely large array so it is not feasible (i.e. runtime) to calculate the waveform in python. Using numpy (which is written in C) would be fast enough, however, it is not possible in numpy to specify the particular memory location where an ndarray should be stored. Therefore, using numpy, we can quickly calculate the waveform, but the waveform will be stored in an area of the computer's memory separate the buffer the API wants. We could transfer the waveform into the API's buffer after it has been generated, but this takes too long.
The workaround implemented is that I have written so functions in C located SPCM-418/Examples/python/calc.c that can calculate the waveform and write the data to a specific buffer (the buffer the API specifies!). This is both fast and places the waveform in the right memory location. calc.c must be compiled into a librarry calc.so so that it can be invoked in Python spcm_m4i_6631-x8.py. Therefore, if you edit calc.c, you must recompile. To compile:
gcc -shared -o calc.so -fPIC calc.c