Complex FIR and Stereo support for rtl_fm

historical

Thanks for testing, is it 32bit arch?

All memory buffers used in SSE should be 16bit aligned, please try no 
SSE version:

rtl_fm -X -F 2 -J 5 -f yourfreq

Also you can try modified rtl_fm version i send as attachment which use 
memalign to allocate buffers instead of malloc, but it is only for gcc.

On X86_64 platform all malloc are 16bit aligned so i missed this.

Those errors when compiling are not important, but also in SSE part.

Miroslav Slugeň
+420 724 825 885

Michał Morański napsal(a):
> W dniu 2013-09-16 11:38, Miroslav Slugeň pisze:
>> Hi, i used git from 14.9.2013, if you get errors from actual git it
>> could be because some functions might go into upstream, i also add as
>> attachment just rtl_fm.c so just replace it your version with this
>> version and it should work.
> Thanks for rapid answer. I just compiled your code.
> First, i got this message when compiling:
> /usr/lib/gcc/i486-linux-gnu/4.4.3/include/emmintrin.h:32:3: error:
> #error "SSE2 instruction set not enabled"
> Then i've added line to CMakeLists.txt:
> set(CMAKE_C_FLAGS ${CMAKE_C_FLAGS} "-msse -msse2 -msse3")
> After that i was able to compile the code, but with some warnings:
>
> /usr/src/rtl-sdr/src/rtl_fm.c: In function ‘low_pass_complex’:
> /usr/src/rtl-sdr/src/rtl_fm.c:451: warning: dereferencing pointer ‘v_r’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:436: note: initialized from here
> /usr/src/rtl-sdr/src/rtl_fm.c:452: warning: dereferencing pointer ‘v_i’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:436: note: initialized from here
> /usr/src/rtl-sdr/src/rtl_fm.c: In function ‘low_pass_real’:
> /usr/src/rtl-sdr/src/rtl_fm.c:769: warning: dereferencing pointer ‘v_m’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:755: note: initialized from here
> /usr/src/rtl-sdr/src/rtl_fm.c:850: warning: dereferencing pointer ‘v_m’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:872: warning: dereferencing pointer ‘v_m’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:834: note: initialized from here
> /usr/src/rtl-sdr/src/rtl_fm.c:851: warning: dereferencing pointer ‘v_p’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:834: note: initialized from here
> /usr/src/rtl-sdr/src/rtl_fm.c:852: warning: dereferencing pointer ‘v_s’
> does break strict-aliasing rules
> /usr/src/rtl-sdr/src/rtl_fm.c:873: warning: dereferencing pointer ‘v_s’
> does break strict-aliasing rules
>
> Now i got segfaults when i try to run rtl_fm -X -f 90.4e6 :
> root at dabplus:/usr/src/rtl-sdr# rtl_fm -X -f 90.4e6 -
> Found 1 device(s):
> 0: Realtek, RTL2838UHIDIR, SN: 00000001
>
> Using device 0: Lifeview LV5TDeluxe
> Found Fitipower FC0013 tuner
> Oversampling input by: 6x.
> Oversampling output by: 1x.
> Buffer size: 7.11ms
> Tuned to 90704000 Hz.
> Sampling at 1152000 Hz.
> Output at 48000 Hz.
> LP Complex: FIR hamming (SSE2), size: 32
> LP Real: FIR hamming stereo (SSE2), size: 64
> Tuner gain set to automatic.
> Tuner AGC ON.
> De-epmhasis IIR: 50.0 us
> Segmentation fault
>
> My processor:
> root at dabplus:/usr/src/rtl-sdr# cat /proc/cpuinfo
> processor : 0
> vendor_id : GenuineIntel
> cpu family : 6
> model : 28
> model name : Intel(R) Atom(TM) CPU D525 @ 1.80GHz
> .....
> root at dabplus:/usr/src/rtl-sdr# cat /etc/issue
> Ubuntu 10.04.4 LTS \n \l
>
> root at dabplus:/usr/src/rtl-sdr# uname -r
> 2.6.32-38-generic-pae
>
>
>
>>
>> I have RDS decoder also, but it is very ugly and not all issues are
>> solved.
> Wow :) If you searching for help with that you can count on me.
>
>>
>> Miroslav Slugeň
>> +420 724 825 885
>>
>> Michał Morański napsal(a):
>>> W dniu 2013-09-13 15:22, Miroslav Slugeň pisze:
>>>> Hi again,
>>>>
>>>> I am working on my own SDR project for Stereo FM radio support, but i
>>>> would like to also improve quality for rtl_fm application, i made
>>>> unoficial patch to add:
>>>>
>>>> Complex FIR - to filter strong signals close to wanted signal
>>>> Real FIR - to filter pilot from FM
>>>> Stereo FIR
>>>> Stereo Deemphasis
>>>> AGC support - it can give better resolution of IQ data
>>>>
>>>> Some other improvments in FM radio code.
>>>>
>>>> All FIR filters has 3 possible variants, simple, LUT, SSE2 instricts,
>>>> of course SSE is the fastest one and it should works on Intel Atoms,
>>>> but not on ARM.
>>>>
>>>> Feel free to use any part of code in any of you programs, I know that
>>>> this code is little to much to add it into rtl_fm, but maybe it could
>>>> somebody help to recieve HW stereo FM radio.
>>>>
>>>> Speed of SSE code is much better than anything you can get around
>>>> here, on Core i7 it consume only 5% of one CPU, so i could demodulate
>>>> at least 80 channels at the same time in stereo quality of course.
>>>>
>>>> I tried this code only on AMD64 and GCC Linux, so i am not sure if it
>>>> can be compiled under windows.
>>>>
>>> Hi! Very nice to hear, that someone working on Stereo FM reception. I'm
>>> building a small remote devices with rtl-dongles as fm stereo receivers.
>>> Now i'm using gnu-radio to decode fm-stereo, but as you all know,
>>> gnu-radio is a large and heavy project and it's wasting its
>>> capatibilities in that role.
>>> It would be very nice if native rtl-sdr software can decode fm-stereo.
>>>
>>> Which version of rtl-sdr was used as base version? I'm getting errors
>>> after applying the patch to last version:
>>>
>>> patching file rtl_fm.c
>>> Hunk #2 succeeded at 47 (offset 5 lines).
>>> Hunk #3 succeeded at 87 (offset 9 lines).
>>> Hunk #4 succeeded at 144 (offset 9 lines).
>>> Hunk #5 succeeded at 169 (offset 9 lines).
>>> Hunk #6 succeeded at 190 (offset 9 lines).
>>> Hunk #7 FAILED at 239.
>>> Hunk #8 FAILED at 258.
>>> Hunk #9 succeeded at 314 (offset 13 lines).
>>> Hunk #10 succeeded at 420 (offset 13 lines).
>>> Hunk #11 succeeded at 994 (offset 13 lines).
>>> Hunk #12 FAILED at 1106.
>>> Hunk #13 succeeded at 1148 (offset 15 lines).
>>> Hunk #14 succeeded at 1259 (offset 35 lines).
>>> Hunk #15 succeeded at 1268 (offset 35 lines).
>>> Hunk #16 succeeded at 1290 (offset 35 lines).
>>> Hunk #17 succeeded at 1301 with fuzz 2 (offset 36 lines).
>>> Hunk #18 succeeded at 1351 (offset 38 lines).
>>> Hunk #19 succeeded at 1378 (offset 38 lines).
>>> Hunk #20 succeeded at 1389 (offset 38 lines).
>>> Hunk #21 succeeded at 1508 (offset 50 lines).
>>> Hunk #22 succeeded at 1527 (offset 50 lines).
>>> 3 out of 22 hunks FAILED -- saving rejects to file rtl_fm.c.rej
>>>
>>> P.S. Are you planning to add support for RDS in the future?
>>>
>>> Regards, Michał.
>>>
>>>
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: rtl_fm.c
Type: text/x-csrc
Size: 48789 bytes
Desc: not available
URL: <http://lists.osmocom.org/pipermail/osmocom-sdr/attachments/20130916/63d11836/attachment.bin>