Re: [Discuss-gnuradio] Improve FSK4 decodes in OP25

[Andy Walls]

Hi Luke:

On Tue, 2017-01-17 at 12:50 -0500, Luke Berndt wrote:
> Andy - Thank you so much, this was extremely helpful. I switched to
> your RF chain and I think it helped a bit.

I'm glad it helped.

FWIW I tried an alternate FM demod approach - 4 PLLs one at each tone -
but the tones turned out to be not spaced far enough apart for that to
work well.  I'd didn't bother with the 4 bandpass filter, non-coherent
approach that Marcus mentioned, since 4 coherent PLL's didn't work out.

Stick with either the FM detector or Quad Demod.

> As you can probably guess, I am more of a programmer than an RF
> expert.

I'm not very good a great many things myself. :)

> 
> It does look like Max has a branch of OP25 which has a deemphasis
> filter that has been tuned for P25.
> https://git.osmocom.org/op25/tree/op25/gr-op25_repeater/apps/tx/op25_c4fm_mod.py?h=max
>  It might be tricky to port to C++ because it uses Numpy, but I am going to 
> give it a shot.

Don't bother.  That's just a moving average filter of approximately the
symbol length.  The same one I have in the flowgraph I gave you.  The
sinc(4800f) in the frequency domain, works out to a boxcar in the time
domain.

Now the APCO P25 transmitters' pre-emphasis filter is a little weird,
since it is 1/sinc(4800f) in the frequency domain and goes infinite.
However it is truncated at 2880 Hz, before it hits it's first infinity.
Luckily you don't need the pre-emphasis filter.

> 
> I do think you are right though, the underlying problem is that there
> is a lot of noise. This may sound crazy, but I think it is coming from
> the USB hub I am using. I know USB2 operates around 480MHz and I am
> trying to record at 490MHz. I am going to experiment with my setup a
> bit to see if this is it.

Not crazy; EMI is a real thing in UHF.

The source can also be PC DRAM clocks, PC monitors, etc.

Funny side note: my directional TV antenna gets lots of noise at least
once a day when it points at the galactic disc.  Apparently the Milky
Way is a large source of UHF noise.


> 
> Thanks again for taking the time to put this all together. I will
> report back on how things are going and if mange to get clean
> recordings.

It was fun.  Down where I live they've just upgraded to APCO P25, but
its all new and all Phase II, so no C4FM signals for me. :(

Good luck!

-Andy

> 
>  - Luke
> 
> On Sun, Jan 15, 2017 at 6:58 PM, Andy Walls
> <address@hidden> wrote:
>         Hi Luke,
>         
>         
>         See the attached GRC for what can be done in GNURadio.
>         
>         
>         1. You need the baseband noise filter.
>         
>         2. You may or may not need the deemphasis filter.  It appears
>         to make things better,
>         but it introduces receiver side ISI, since the filter doesn't
>         get dumped at every symbol
>         peak. (APCO 25 expects an integrate and dump deemphasis
>         filter, this one just
>         integrates).
>         
>         3. You may wish to use the FM detector PLL instead of the Quad
>         Demod block.  You
>         can limit symbol peak frequency deviation using the Max Freq
>         and Min Freq parameters.
>         Currently, I have them set at +/-3 * 600 Hz * 190%.  Leave the
>         loop bandwidth alone;
>         
>         it is set at symbol_rate/2 * 1.20 Hz = 2880 Hz, which is where
>         the transmitter Nyquist
>         
>         filter should cut off.
>         
>         
>         The signal you provided really does have a very poor SNR.  The
>         attached low SNR
>         
>         png shows the best I could do with it.  There are some of the
>         4 levels visible in the chaos,
>         
>         but it's not good.  I have attached a png from a high SNR test
>         signal for comparison.
>         
>         
>         So what you can do "outside" of GNURadio:
>         
>         
>         1. Attach an external receiver bandpass filter for the 470 -
>         512 MHz safety
>         
>         band to your radio.  The only small, reasonably priced, off
>         the shelf ones I could find
>         are ceramic packages that are expected to be soldered to a
>         board. :(
>         
>         
>         2. Set your gains on your SDR smartly to maximize your
>         receiver noise figure.
>         
>         Set the RF gain as high as you can without clipping the IF
>         stage, set the 
>         
>         IF gain as high as you can without clipping the Base Band
>         stage.  Set the Base
>         
>         Band gain as high as you can without clipping the ADC inputs.
>         
>         https://en.wikipedia.org/wiki/Friis_formulas_for_noise
>         "... the overall noise figure of a radio receiver is primarily
>         established by the noise figure of its first amplifying stage.
>         Subsequent stages have a diminishing effect on signal-to-noise
>         ratio. "
>         
>         
>         
>         Regards,
>         
>         Andy
>         
>         
>         
>         On Thu, Jan 12, 2017 at 12:00 PM,
>         <address@hidden> wrote:
>                 Message: 6
>                 Date: Wed, 11 Jan 2017 21:56:51 -0500
>                 From: Luke Berndt 
>                 To: Marcus M?ller ,   GNURadio Discussion List
>                         <address@hidden>
>                 Subject: Re: [Discuss-gnuradio] Improve FSK4 decodes
>                 in OP25
>                 
>                 Thanks Marcus! I will try putting in a low pass in
>                 first after the Quad demod. That BPF / Polyphase
>                 Channelizer approach sounds pretty cool too!
>                 
>                 I have attached a GRC file that I use for debugging
>                 and a link the raw dump from a File Sink. The sample
>                 rate at the sink is 96k/s, the system channel rate is
>                 48k/s, with 10 samples per symbol. The FSK Level has
>                 to be set to about 0.66.
>                 
>                 I am sure that some of the problem is that I don?t
>                 have a great signal coming in, but if there is any
>                 additional processing I could do to improve the
>                 reception that would be great.
>                 
>                 
>                 
>                 Wav file
>                 
> https://www.dropbox.com/s/ncpz6wt8tubm2v3/32784-1484187802_4.90762e%2B08.wav?dl=0
>  
> <https://www.dropbox.com/s/ncpz6wt8tubm2v3/32784-1484187802_4.90762e+08.wav?dl=0>
>                 
>                 Raw capture from a file sink
>                 
> https://www.dropbox.com/s/810zj7pudc66yo3/32784-1484187802_4.90762e%2B08.raw?dl=0
>  
> <https://www.dropbox.com/s/810zj7pudc66yo3/32784-1484187802_4.90762e+08.raw?dl=0>
>                 
>                 
>                 
>                 > On Jan 11, 2017, at 12:12 PM, Marcus M?ller
>                 <address@hidden> wrote:
>                 >
>                 > Hi Luke,
>                 >
>                 > you don't happen to have a waterfall plot or a raw
>                 IQ recording?
>                 > Quadrature demodulation isn't the most resilient to
>                 noise. So, a low-pass filter that makes sure that
>                 events that are sufficiently shorter than a symbol
>                 after the Quadrature demodulator would sound wise.
>                 Then again, I don't know the OP25 infrastructure
>                 overly well, so chances are the FSK4 demod already
>                 contains one.
>                 >
>                 > An alternative to your Quad FM demod would be a
>                 four-bandpasses-based decider. An easy way to build
>                 one is to either use four individual BPFs, or use one
>                 set of filter taps in the Polyphase Channelizer (which
>                 divides your f_sample_in  stream into four
>                 f_sample_in/4 -rate streams with channel centers).
>                 Just use a filter that would select the "zeroth"
>                 carrier. The magic of polyphase math will add three
>                 f_sample_in/4-shifted versions (you might need to use
>                 Rotator or similar to put the zeroth subcarrier at
>                 f=0). I have a "playing with channelizers" flow graph
>                 [1]. Advantage of the polyphase channelizer is that it
>                 uses very little more ressources than a single
>                 bandpass filter would ? basically, you get the three
>                 additional filters for the cost of a short FFT. Yay!
>                 >
>                 > Cheers,
>                 >
>                 > Marcus
>                 >
>                 > [1]
>                 https://gist.github.com/babffaa88f54a7fba9b52fc09bda0462 
> <https://gist.github.com/babffaa88f54a7fba9b52fc09bda0462>
>                 >
>                 > On 01/11/2017 04:44 PM, Luke Berndt wrote:
>                 >> I am using the C++ OP25 libs to capture and record
>                 the P25 transmissions from a trunked radio system
>                 ( https://github.com/robotastic/trunk-recorder
>                 <https://github.com/robotastic/trunk-recorder> ).
>                 >>
>                 >> Transmission can either use 4-Level FSK or CPSK. I
>                 am getting great decodes from a system using CPSK but
>                 I am having a lot of trouble with a system using 4FSK.
>                 >>
>                 >> At a high level, here is my 4FSK flow graph
>                 
> (https://github.com/robotastic/trunk-recorder/blob/master/trunk-recorder/recorders/p25_recorder.cc
>  
> <https://github.com/robotastic/trunk-recorder/blob/master/trunk-recorder/recorders/p25_recorder.cc>):
>                 >> Xlating + LPF to roughly capture the channel
>                 >> ARB Resampler to get the system sample rate
>                 >> Quad FM Demod
>                 >> Multiply Const - to get the right levels, -3 to 3
>                 >> FSK4 Demod - from OP25
>                 >> FSK4 Slicer - from OP25
>                 >> OP25 Decode
>                 >> it then converts it to what is needed to record it
>                 as a Wav file
>                 >>
>                 >> Are there some obvious blocks I should add in to
>                 help condition the input better? I think the incoming
>                 levels may not be constant. There also seems to be
>                 occasional interference. Is there a block that will
>                 discard values above a certain range? It would be
>                 great to throw-away samples above 3.5 or below -3.5.
>                 >>
>                 >> Attached is a picture with what I think is an
>                 example of the interference.
>                 >>
>                 >> Here is an example of a very clean recording:
>                 
> https://beta.openmhz.com/wmata?call-id=587652495833790100186198 
> <https://beta.openmhz.com/wmata?call-id=587652495833790100186198>
>                 >>
>                 >> Here is an example of a bad decode:
>                 
> https://beta.openmhz.com/wmata?call-id=58765250583379010018619c 
> <https://beta.openmhz.com/wmata?call-id=58765250583379010018619c>
>                 >>
>                 >> Any tips would be greatly appreciated!!
>                 >>
>                 >> - Luke
> 
>