Re: [Discuss-gnuradio] Update 2: BPSK problem: intermittent inter-symbol interference--Answer about the burst shaper and SPS

[Edwin Li]

Hi Marcus,

To be honest with you, I have the burst shaper because the examples I found in /gr-digital/examples/packets have it.  I guess it would be helpful if my packets are short. But I don't really have to restrict my packets to be short. All I have in mind is compare the BPSK BER curve on USRP with the theoretical one. If long packets can solve the problem. I'll go with long. It seems that you and your colleague believe the intermittent ISI comes from the burst shaper? I'll try removing the block and see what happens.

I have the sample-per-symbol being 2 because I read it somewhere in the tutorial that 2 is practical.  Larger number would give higher SNR but at a higher computational cost. I tried 4. There would be a higher chance of getting the intermittent ISI.  Could it be the computational power that throttles the performance? But my CPU usage is only about 50% when running the simulation. However, I get quite a lot underruns if an USRP sink is activated. I use 1G Ethernet to connect the USRP. The sample rate is only 800K. 

Regards,

Edwin

Marcus Müller <address@hidden>于2017年9月1日周五 上午5:37写道：

Wow! Now, that is a flowgraph that's really pretty! Thanks! Will have a look at it!

On 08/31/2017 10:58 PM, Edwin Li wrote:

Hi Marcus,

I'm sorry. That was very inconsiderate of me. I made a new grc file, merging the tx and rx into one. You can find it in the attachment. 

I want to make sure the system works with an ideal channel. But I still have the intermittent ISI problem. 

                                                Figure 1 Down sampled signal

As you can see in Figure 1. There are 5 spots with severe ISI, while others are good. 

                                               Figure 2  Packet with severe ISI

Figure 2 shows the zoom in of Figure 1. How can one packet gets so much ISI while others are good? If my RRC filters have any error, all of the packets should suffer the same problem, right?

Or could it be that the parameters still need some tweaking? I'm guessing, in an ideal channel(in simulation), I can use an decimating FIR filter to replace the polyphase clock sync, while in experiment, I must use the polyphase clock sync as well as some kind of equalizer and tweak the parameters until I lower the ISI below an acceptable level. 

Regards,

Edwin