GNU Radio - PSK Mod block - unexpected constellation diagram

Question

I am simulating a QPSK modulation in GNU radio. Here is the .

The source is a binary stream coming from the Random Source block. There are 4 PSK Mod blocks configured identically (4 constellation points - a QPSK modulation, Grey coding and Differential encoding turned on, excess bandwidth set to 0,35). The modulators differ in the parameter samples/symbol (sps), which is set to 2, 4, 8, and 16. The constellation sinks are configured identically. What is expected to be shown on the constellation diagram is a plot with 4 constellation points that could be the tips of a square or 4 clusters of points that are concentrated around those 4 points as a certain amount of noise or imperfection of the system is added. Instead, we have the following:

The constellation diagram when sps is set to 16 is .

The constellation diagrams for sps 2, 4 and 8 look the same but with less points on the sides of the rhombus and less points around tips of the rhombus that look like the extensions of the rhombus sides.

I understand that transition points at the edges of the rhombus appear because of the raised-cosine filter implemented inside the PSK Mod blocks used for pulse-shaping. By increasing the Excess BW parameter, number of points that appear at the edges decrease. We can increase Excess BW to 1. For Excess bandwidth > 1, the constellation diagram distorts, constellation points that are the tips of the rhombus are getting closer to the center. Why is that happening?

The other question is, how to get a 'neat', or at least 'cleaner' constellation diagram with 4 points, or at least 4 clusters of points?

Answer 1

For Excess bandwidth > 1, the constellation diagram distorts, constellation points that are the tips of the rhombus are getting closer to the center. Why is that happening?

Mathematical nonsense, probably, is what is happening here. If you do excess BW > 1, you probably should just increase your SPS.

The other question is, how to get a 'neat', or at least 'cleaner' constellation diagram with 4 points, or at least 4 clusters of points?

That is a curious question, considering that you

understand that transition points at the edges of the rhombus appear because of the raised-cosine filter implemented inside the PSK Mod blocks used for pulse-shaping.

So: either, omit the pulse shaping by not using the PSK mod block (which includes that), but just a constellation mapper, or reverse its effect by applying the matched filter, decimating your symbol down to symbol rate after timing recovery.

Generally, the question "how do I make the constellation diagram look more like the pure constellation" is only relevant if you don't want to transmit the signal -- pulse shaping is beneficial and should not (and usually can not) be omitted.

If you want to work with the (unmodulated/unfiltered) constellation points, you'd typically have a constellation object and a "chunks to symbols" mapper block, in which you use <constellation object>.points() as mapping, something like: