[Discuss-gnuradio] why changing decimation in benchmark_rx.py affects SNR so much?

[William Cox]

I'm running a link via benchmark_tx and benchmark_rx files. They are both running at 1e6 symbols/sec and a 5 Mhz carrier. The settings are default otherwise. I've noticed that when I change the decimation rate at the receiver, the SNR changes quite a bit while keeping BER the same. I would think with a 1 Msps BPSK, I'd need about 2 MHz of bandwidth at the receiver, but if I increase my bandwidth to 8 MHz the SNR will increase by 3-4 dB while the BER remains the same. Why is this? 

Also, what practical affects should I see from changing the RRC filter rolloff?

Thanks.

-William