Problems with decoding real RF signals

[sp5wwp]

I have set up a simple M17 receiver as follows:

The resulting eye diagram is pretty decent

however, I'm not able to get any decode. payload.bin is empty, the log is empty too. My transmitter is a USRP B210 with GNURadio running flowgraph shown below.

[jmfriedt]

In order to try and separate the faulty emitter or receiver: is this signal broadcast by the USRP properly decoded by the RTL-SDR fetching another reception software implementing M17?

[sp5wwp]

Yes. It works flawlessly with SDR++ software. I think the problem resides in the symbol sync block. The symbol stream at the output looks fine to me, so I have no idea what might be going wrong there. Can you take a look to confirm?

[argilo]

The general_work method does not preserve any state, so each time a (small) batch of symbols arrives it starts in the unsynchronized state, and there are never enough symbols to proceed to decoding.

The state variables (last, syncd, pld, pushed, etc.) need to be member variables of the m17_decoder_impl class, not local variables in the general_work method.

[jmfriedt]

Fantastic, thank you for identifying my mistakes and providing the means to correct.

I believe commit d8b2139411371265c3fc082621ed937b53667f1d allows for running the m17_loopback_noisychannel.grc (which includes the modulation and demodulation) example that had been resisting so far while keeping the m17_streamer.grc functional.

Thank you ! I am away from the lab and cannot test on actual hardware but am looking for some feedback on PlutoSDR and possibly RTL-SDR.

[sp5wwp]

It's actually a separate issue, but the 8.28 value for cross-correlation computations is wrong.

#define CODE_STD        8.78 // std(str_sync)*sqrt(length(str_sync))

when it should be

#define CODE_STD        8.21583836f //std(str_sync)*sqrt(length(str_sync)-1)

This gives value of 1.0 for ideal correlation, rather than 0.935 in your case.

[argilo]

And one more fix that needs to be pulled in: https://github.com/M17-Project/M17_Implementations/pull/6

[argilo]

last also needs to be changed to a member variable. Once that is done, the decoder block appears to work correctly.

[sp5wwp]

Fixed?

[argilo]

With this changeset, things are more or less working:

diff --git a/lib/m17_decoder_impl.cc b/lib/m17_decoder_impl.cc
index 7299c3300e..a7f5b083a6 100644
--- a/lib/m17_decoder_impl.cc
+++ b/lib/m17_decoder_impl.cc
@@ -42,7 +42,7 @@

 //#define XCORR_THRESHOLD 0.90 // arbitrary threshold between 0 and 1: might be tunable from GNU Radio Block for flexibility
 #define CODE_MEAN      -0.75 // mean(str_sync)
-#define CODE_STD        8.78 // std(str_sync)*sqrt(length(str_sync))
+#define CODE_STD        8.21583836f //std(str_sync)*sqrt(length(str_sync)-1)
 // see ../M17_Implementations/SP5WWP/inc/m17.h for const int8_t str_sync[8]={-3, -3, -3, -3, +3, +3, -3, +3};

@@ -105,6 +105,7 @@ void decode_callsign(uint8_t *outp, const uint8_t *inp)
                encoded/=40;
                i++;
        }
+       outp[i] = 0;
 }

     m17_decoder::sptr
@@ -167,7 +168,6 @@ void m17_decoder_impl::set_debug_ctrl(bool debug)
      int countout=0;

      float sample;                       //last raw sample from the stdin
-     float last[8];                      //look-back buffer for finding syncwords
      float xcorr;                        //cross correlation for finding syncwords
      float meanx;                        //mean value
      float normx;                        //cross correlation normalization
diff --git a/lib/m17_decoder_impl.h b/lib/m17_decoder_impl.h
index 84b0207fb2..c43fd027a8 100644
--- a/lib/m17_decoder_impl.h
+++ b/lib/m17_decoder_impl.h
@@ -25,6 +25,7 @@ private:
     bool _debug_ctrl=false;
     float _threshold=0.9;

+    float last[8] = {0};                //look-back buffer for finding syncwords
     float pld[SYM_PER_PLD];             //raw frame symbols
     uint16_t soft_bit[2*SYM_PER_PLD];   //raw frame soft bits
     uint16_t d_soft_bit[2*SYM_PER_PLD]; //deinterleaved soft bits

[sp5wwp]

Let's incorporate these changes?

[argilo]

OK, I committed those fixes to the main branch.

[argilo]

I'm still seeing broken packets printed sometimes during startup, so there could still be a bug or two.

Some other issues I noticed:

• lich_chunks_rcvd should probably be reset when an LSF is received
• decoding of the LICH for late joiners does not work because fl is not set to 1 until an LSF is received

[sp5wwp]

I'm going to close this issue as resolved. Other bugs can be reported with separate issues.