DL-MMSE
This MR is based on MR !2185 (merged)
- Fix rxdataF_comp size in nr_rx_pdsch() function
- Improvements in nr_dlsch_channel_compensation() function
- Noise power estimation for DL at UE side
- MMSE computation for DL at UE side
Theoretically, MMSE is better than ZF for low SNR, and then its performance converges as SNR increases. The results on the phy-simulators confirmed this. The gain for a frequency selective channel was not as great as for an AWGN channel. This probably happened because of the DCI, as the following LOGs appeared many times: [PHY] DCI false positive. Dropping DCI index 0. Mismatched bits: 284/864. Current DCI threshold: 205
Some results obtained in the phy-simulators were:
AWGN channel
sudo LD_LIBRARY_PATH=. ./nr_dlsim -n10000 -s4 -x2 -y2 -z2 -e2
- ZF:
Eff Throughput 50.47
- MMSE:
Eff Throughput 92.84
sudo LD_LIBRARY_PATH=. ./nr_dlsim -n10000 -s9.7 -x2 -y2 -z2
- ZF:
Eff Throughput 53.63
- MMSE:
Eff Throughput 94.01
sudo LD_LIBRARY_PATH=. ./nr_dlsim -n10000 -s16.6 -x2 -y2 -z2 -e16
- ZF:
Eff Throughput 78.88
- MMSE:
Eff Throughput 81.37
Frequency selective channel
sudo LD_LIBRARY_PATH=. ./nr_dlsim -n10000 -s7 -x2 -y2 -z2 -gR -e2
- ZF:
Eff Throughput 70.78
- MMSE:
Eff Throughput 82.81