IQFLOAT isn't parsed corretly or not documented enough

[spaivaras]

Hello, I just can't get rpitx to work with IQFLOAT data (complex type of GNURadio) to work properly with rpitx. I have a simple FM transmitter under gnu radio and pipe complex floats via network with sampling rate of 190000 (doesn't really matter as no sampling rate is working)

If i use

 while true; do (nc -l -p 8011; dd if=/dev/zero bs=4096 count=30); done | sudo rpitx -i- -m IQFLOAT -f 99000 -c 1 -s 190000

and i get an to wide output with 2 peaks in them, the signal is unusable:

If I try the exact same IQ data, but first passing it via csdr

while true; do (nc -l -p 8011; dd if=/dev/zero bs=4096 count=30); done | csdr fmdemod_quadri_cf|csdr gain_ff 75000 |csdr convert_f_samplerf 5263 | sudo rpitx -i- -m RF -f 99000

I get a much better signal (actually quite good one) proper width:

It would be enough to use csdr, but it introduces additional latencies, adds requirement for gain multiplication and in general would be nice if rpitx would properly handle IQ data itself :)

In README.md about streaming from GNURadio its written to "shift" the signal by 5khz, and thats centers the signal on one of 2 peaks. But that doesn't fix the issue that signal itself is not a good representation of passed IQ values

[spaivaras]

Specs:
Raspberry Pi 2 B
Distro Minibian (jessie)
compiled from master

[F5OEO]

The IQFLOAT type is float : be sure to not have a double float output from gnuradio.
As IQ is convert to phase/frequency, value around zero are subject to round errors. That's why you should shift the signal.
Agree, IQFloat is not well documented : TODO !

[ha7ilm]

IQFLOAT means 32 bit interleaved floats, like:

I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) ...

Endianness depends on the system, it should be LE on both an RPi and a PC.

Also the signal should have the maximum amplitude possible (I mean that you should set the gain correctly in the appropriate GNU Radio blocks), but it should never exceed the [-1.0, 1.0] range. This is the line in the code where the data is converted to 16 bit signed imtegers:

https://github.com/F5OEO/rpitx/blob/master/src/RpiTx.c#L1308

However, if you want to generate an FSK or an FM signal, I suggest that you should always use rpitx in RF mode. Then the PWM will not be used and the signal will be cleaner.

@F5OEO Evariste, are these correct?

[F5OEO]

Yes , thx Andras for these nices explanations. Have some good news : works
on pizero and now surely on rpi3 ! Could go also upper in frequency .

2016-08-24 15:51 GMT+02:00 András Retzler notifications@github.com:

IQFLOAT means 32 bit interleaved floats, like:

I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) I(4 bytes) Q(4bytes) ...

Endianness depends on the system, it should be LE on both an RPi and a PC.

Also the signal should have the maximum amplitude possible (I mean that
you should set the gain correctly in the appropriate GNU Radio blocks), but
it should never exceed the [-1.0, -1.0] range. This is the line in the code
where the data is converted to 16 bit signed imtegers:

https://github.com/F5OEO/rpitx/blob/master/src/RpiTx.c#L1308

However, if you want to generate an FSK or an FM signal, I suggest that
you should always use rpitx in RF mode. Then the PWM will not be used and
the signal will be cleaner.

@F5OEO https://github.com/F5OEO Evariste, are these correct?

—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
#31 (comment), or mute
the thread
https://github.com/notifications/unsubscribe-auth/AF53Rl-1L-kiH7L0SKvVT3ttt0ThM6Rrks5qjExYgaJpZM4JrTw3
.

[ha7ilm]

works on pizero and now surely on rpi3 ! Could go also upper in frequency .

Great!

[spaivaras]

Thank you for the explanation, its exactly what I thought and what GNURadio complex type actually is, 2 floats each 32bits (no doubles). I am probably missing a single gear somewhere in my approach :) as it seems no one else had issues with IQFLOAT. I will poke more around with simple sine waves to get a better idea what actually happens

[F5OEO]

@spaivaras Do you succeed in using IQFLOAT with gnuradio ?

[spaivaras]

@F5OEO Nop, Tried everything, pass data thru network, read it directly from file. Lower sampling rates, higher sample rates, gains, swapping I and Q data, shifting from center. The signal always seems very distorted (like 2 carrier frequencies). Checked the data that is coming thru, everything seems fine with setup. RF modes works quite well, just can't generate a RDS signal with RF mode. I can even use IQ demodulation in gnu radio itself and pass resulting floats via csdr with convert_f_samplerf and a constant for timing. And it works. But no success with IQFLOAT. In any case I think this can be closed, as no one else seems to have any issues (or just didn't try it). I will sniff in code at some point, if I find something I will create a PR :)

[F5OEO]

Have to check this IQLOAT issue coming from external tool like gnuradio. Can you send me an example of file with IQFLOAT which is distorded to inspect and reproduce here ?

[spaivaras]

Sure, I generated 3 files, which can be played with no problem in GQRX or other sdr software
https://www.dropbox.com/sh/37bybfdsh2xa688/AACyNhk-O1mO6X_7rdfp-7zba?dl=0
File names say, what settings they have