Jen,
Thats very good information. EVM values around 0.5 are pretty good, which explains why the constellation and eye diagrams are clean. It seems to me this might be a software issue then. I would try a couple of different things. First, try generating more symbols at a lower M-ary. In general, a certain degree of randomness is required in order to demodulate the signal correctly. Just to make sure, you should be using a pseudo random bistream to make sure that each symbol is generated ideally more than once. This is even more critical at higher M-aries such as 64 QAM.
Another thing you can try is to append a defined number of random bits onto the end of your actual bistream. For example, if you have 1000 bits of useful data, try appending 1000 random bits onto the end....and check to see if the first 1000 came through correctly. In one of the simulation examples that I wrote recently, I noticed an issue similar to what you're experiencing. In my example, I was generating the several symbols over and over again without hitting all of the symbols. By adding some random bits to the end of the bistream, I was able to ensure that all symbols were generated and recover all bits correctly.
Anyway, this is just a thought. Ideally, it seems like you should be getting a BER of 0 with EVM measurements like you mention. It sounds like the constellation plot is locked to the appropriate phase, which is good. At this point, I would experiment with your bistream. Do you get the same results at all M-aries? Or does the BER increase as the M-ary increases? I hope this helps and let me know if you have any more questions.
Regards,
David Hall | Applications Engineering | National Instruments
