To add on to what Marcus and Brian have said, one of the ways of
slightly reducing this problem is Crest Factor Reduction. It would be
very useful to have some of the standard CFR algorithms added to GNU
Radio. Peak Cancellation is one that has looked promising to me. Peak
Windowing and Noise Shaping are two others.
If you do implement any of these methods please consider contributing
them back to GNU Radio so that we can all benefit.
Regards,
Derek
On 6/23/2020 3:49 PM, Brian Padalino wrote:
On Tue, Jun 23, 2020 at 10:38 AM Manav Kohli <mpk2...@columbia.edu
<mailto:mpk2...@columbia.edu>> wrote:
Hello,
This problem is visualized here:
https://www.dropbox.com/s/w7kdmf9dewwdomx/20M_2974_20_15_nocal_packet_time_tx.png?dl=0
This is an OFDM packet consisting of 6 symbols: the default sync
word 1&2, SIG field and three data symbols. The data symbols are
QPSK modulated and the sync words are BPSK.
Is there any way that I could reduce or eliminate these large
spikes that even with a moderate baseband scaling still go
considerably above a magnitude of 1? I have tried to use different
data, and a different number of packet data symbols, but to no
avail. I am definitely able to reduce the overall baseband scaling
and get those spikes within range, but this is not desirable as I
am trying to maximize transmit power.
The usage of the USRP-2974 and sampling rate is immaterial; this
happens using a variety of different radios (should not matter as
this is a GNU radio "issue") and sampling rates.
If anyone has seen this before or may have any advice, please let
me know.
It's the nature of OFDM to have a high Peak-to-Average-Power Radio
(PAPR). You can look at PAPR reduction techniques, but otherwise
you're in for around 10dB PAPR.
Good luck.
Brian
