I've observed something weird running some USRP programs on Mac OS X: the gains in effect in usrp_siggen, usrp_oscope and usrp_fft seem unreasonably large. I only get reasonable looking output (from usrp_siggen) and input (with usrp_oscope and usrp_fft) with -a100 or less, otherwise the output (or input) saturates or wraps around in a very bizarre way.
It's as if -a100 were almost full scale when these programs run on Mac OS X. Running the same programs on the same USRP attached to a PC running Linux, -a32767 is full scale. It's as if the gain were about 200x greater on the Mac. This is with USRP Rev 2.0 and the Standard RX and TX daughterboards, and GNU Radio checked out from CVS on Nov. 28 2005. One possibly important detail: I have to run the programs with high interpolation and decimations (128 or 256) because the USRP+USB on the Mac can only manage 4 MBytes/sec. I have only seen this effect in the Python programs under gnuradio-examples/python/usrp, not the C programs under usrp/host/apps. When I run test_usrp_standard_tx (a C program) on the Mac with these arguments: test_usrp_standard_tx -I 128 -a 10000 A scope (a real one, not usrp_oscope) connected to TX-A shows a 15 KHz sine wave with a peak-to-peak amplitude of about 80 mV. -I 128 is the smallest interpolation that does not result in underruns on the Mac. The relatively low amplitude shows the effect of the AC coupling at this low frequency. The output of this command is similar on a PC running Linux (with GNU Radio 2.6 from the tarballs) Next I run usrp_siggen with -f15e3 to match the output frequency of test_usrp_standard_tx. On a PC running Linux usrp_siggen -I 256 -f15e3 -a 16000 (this is the default amplitude) results in about a ~125 mV peak to peak sine wave. On the Mac, this same command does not produce a sinewave at all, but a complicated waveform that looks amplitude-modulated. A clean sine wave appears only at small values for -a, for example usrp_siggen -I 256 -f15e3 -a75 results in about a ~130 mV peak-to-peak sinewave - about the same amplitude as with -a 16000 on Linux! There is a similar gain effect on input. Using usrp_oscope on the Mac, with this command usrp_oscope -g0 -f0 -d128 applying a 20 KHz input signal of only 20 mV peak-to-peak amplitude (from a signal generator, not from usrp_siggen) on RX-A results in a full-scale trace on the scope display, ranging over -30000..30000. Increasing the amplitude of the input signal results in the scope trace wrapping around (not just clipping). This is with -g0, the minimum gain; the scope panel shows Gain: 0. But on the PC running Linux, with the same command, a 500 mV (not 20 mV) peak-to-peak input signal causes the scope trace to range over only about -2400..2400. So again it looks like the gain on the Mac is about 200x the gain on Linux. I wonder if the versions of these programs built on the Mac are somehow setting the scale in the USRP's AD chip in some occult fashion. Or, I wonder if there is some interaction of interpolation/decimation and gain. An older Linux version of usrp_oscope (checked out from CVS on Oct 7) issued this message when invoked with -d 256: usrp_standard_rx::set_decim_rate: WARNING rates > 128 result in incorrect gain The current version does not issue this message, but it suggests there might be something tricky going on. (But programs running on Linux at large -i and -d do not show unusual gains). Has anyone else seen this? Can anyone suggest an explanation? Jon Jacky _______________________________________________ Discuss-gnuradio mailing list Discuss-gnuradio@gnu.org http://lists.gnu.org/mailman/listinfo/discuss-gnuradio
