Dropping TX samples?

[madengr]

In GRC I'm feeding the osmocom sink with a 1 MHz complex cosine. TX filters are set for 5 MHz cutoff, LO at 1 GHz. I'll start at a 10 MHz sample rate and move up to 20 MHz and 40 MHz. Looks like a sample is being dropped.

Here is the spectrum at 10 MHz sample rate, with the LO leakage in the middle at 1000 MHz and the tone at 999 MHz. Notice the modulation on the 999 MHz tone.




Spanning in reveals 1.22 kHz sidebands, so 10E6/1.22E3 = 8196. That's close to 8192, so is there a problem in a buffer somewhere?




Doubling the sample rate to 20 MHz doubles the rate to 2.44 kHz.




Then at 40 MHz it really craps.



If I drop the sample rate to 2 MHz, it's still there.

Also the tone should really be at 1001 MHz, not 999 MHz. Are the I and Q samples being swapped, inverting the spectrum? The inversion is random when I run the GRC diagram.

I know it's dropping RX samples too. It's very noticeable at 40 MHz with the spectrum jumping, but noticeable at lower rates too.

[madengr]

If I give it a constant source instead of the sinusoid, and it were replacing a TX sample with an arbitrary value, I would expect to see the modulation on the carrier. That's not the case, the carrier is clean.

[bpadalino]

I don't think I understand the observation? Are you saying that the constant value, which should just be CW at CF, is just CW at CF - and you were expecting that if it were dropping samples or doing something weird then you would get the modulation?

What is your CPU loading like when you're doing this?

[madengr]

88% CPU load.

Yes, constant results in just clean CW at CF. That means if the samples are being dropped, they are at least not being replaced with an arbitrary value, say full scale 0 or 1.

When I'm sending a sinusoid, a 1.2207 KHz modulation on that tone would mean it's skipping every 8192 sample at 10 MHz sample rate. Is there a FIFO somewhere that is 8k deep, that is perhaps not being fully emptied?

At 40 MHz sample rate that nice sinc spectrum may indicate the DACs are just toggling between two values.

Maybe the next step is to transmit a sinusoid file directly from the CLI, to see if it's a osmocom issue.

[madengr]

OK here I think you can actually see it. I set the baseband sinusoid frequency to 10E3/8192 Hz. This would provide a one second beat frequency; i.e. if the sample is dropping it should do so at a 1 Hz interval Then zero spanned the spectrum analyzer with a 1 MHz RBW and 1 Hz sweep rate, linear scale, and video trigger. Looks closely at 350 ms into the sweep and it looks like a DAC glitch. This is repeated maybe 1/2 sec later. There seems to be a third one near the bottom. Strange.... It's not the retrace, as they do not move unless I adjust the trigger level. I probably not right on center frequency either.

[madengr]

I also tried the experiment in reverse to see if RX samples were dropped, again at a 10 MHz sample rate. I set the RX LO to 1500 MHz and drove the unit with a 1501 MHz synthesized source. Then used a frequency translating FIR filter, decimating to 25 kHz sample rate, to center the 1501 MHz signal. There was no 1.22 kHz side tone. The noise floor however was slowly moving up and down 10 dB, almost like a gain drift. Anyway, the periodic sample drop seems to be contained to the TX side.