# Help: QAM Constellation and PolyPhase Clock **URL:** \ **Category:** GNU Radio\ **Created:** [July 29, 2013, 6:21pm UTC](https://www.ruby-forum.com/t/help-qam-constellation-and-polyphase-clock/235998 "2013-07-29T18:21:43Z")\ **Posts on this page:** 2\ **Page:** 1
**Author:** ![alcina](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/a/a587f6/32.png) [@alcina](https://www.ruby-forum.com/u/alcina)\ **Post date:** [July 29, 2013, 6:21pm UTC](https://www.ruby-forum.com/t/help-qam-constellation-and-polyphase-clock/235998/1 "2013-07-29T18:21:43Z")
I have simple 16QAM Modulator block and am plotting the constellation with and without the a polyphase clock block . The constellation directly out of the modulator is not 16 “points” or even blobs but “random” looking points. I put it into a QAM Demod block and it demodulates it correctly. The constellation plot of the polyphase clock block shows 16 distinct “points” but when demodulated is quite noisy. Can anyone explain: 1. Why is the output of the QAM Modulator block not show as 16 points? 2. Why would the “corrected” QAM Modulation not give better demodulation output? I have attached the ConstellationQuestion.grc file. Thanks…Tom ---
**Author:** ![tom\_sutherland](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/t/85f322/32.png) [@tom\_sutherland](https://www.ruby-forum.com/u/tom_sutherland)\ **Post date:** [July 29, 2013, 6:39pm UTC](https://www.ruby-forum.com/t/help-qam-constellation-and-polyphase-clock/235998/2 "2013-07-29T18:39:50Z")
On Mon, Jul 29, 2013 at 12:14 PM, tom sutherland [removed\_email\_address@domain.invalid](mailto:removed_email_address@domain.invalid) wrote: > I have attached the ConstellationQuestion.grc file. > Thanks…Tom It’s what’s called ISI, or inter-symbol interference. The modulator applies a pulse shaping filter to the transmitted signal, which in this case is a root raised cosine filter. That has properties that actually “smear” the symbols in time and make them overlap in this way. That’s expected and what it’s supposed to do. Also, you have now increased the sampling rate to 4 samples / symbol, so you’ll get points in between the actual symbol points now. The clock recovery block in this case uses another root raised cosine filter. The convolution of an RRC with an RRC is a raised cosine filter, which is a Nyquist filter, and so you’ve removed the ISI you’ve introduced in the first RRC stage. This also finds the correct symbol points and outputs 1 sample per symbol. The PFB still has a little bit of noise added due to the quantization of the possible phases. If you increase nfilts, you’ll see this quantization noise go down. But the noise using 32 filters is about the same quantization noise you get for 16 bits of precision, so that limits your situation there. Besides, you’re going to quickly become dominated by thermal noise. Tom