NC-OFDM and Cross sub-carrier Interference

GNU Radio
1

Hi,

I’m currently developing a testbed for spectrum aggregation and
fragmentation using GNURadio and USRPs (B200s and B210s). My approach is
to
emulate aggregation and fragmentation using Non-Contiguous OFDM:
Theoretically, each node should be able to transmit in any subset of the
sub-carriers available.

Everything seems to be fine when tested in a loop back(the GRC OFDM Tx
example in gr_digital was used). And connected to USRPs, a single pair
transmission and reception is fine while transmitting in any subset of
subcarriers. The problem arises when I have more than one node
transmitting
in the same OFDM configuration. No matter how wide the OFDM span is (I
began from 64 and increased it to 1024), and no matter how far apart the
two transmitting channels are, the cross-interference is enough to
corrupt
both transmissions. I have played with the transmission power of both
transmitters to see if lowering it makes any difference, to no avail.

I have also thought of applying a bandpass filter before feeding OFDM
signals to USRPs to limit their out of band emissions, but as the system
is
supposed to handle fragmentation and requires complete spectrum agility,
a
single filter would not solve the problem.

I would greatly appreciate any advice on the matter.

Many Thanks,

Vahid

2

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi Vahid,

a few things:

On one hand, to be really orthogonal, all your nodes would need to
exhibit no frequency offset. As long as you don’t have a common clock,
be it through a clock distributor, GPSDOs or by doing estimations of
the own offset on the nodes and accounting for that, you’d get energy
of bins end up in others.

Then, it’s intuitive that one tends to drive the DAC with maximum
values (ie. using the whole -1,1 magnitude range in GNU Radio). Try
adding a factor of 0.7 before! OFDM especially is a bit sensitive to
the transmitter clipping, and that might increase the looks of your
spectrum. I don’t know your transmitter, so I’m going to point out
that OFDM can have quite an ugly PAPR, and although you feed in
symbols that themselves as time signal would never lead to complex
magnitude values greater than 1, the IFFT of a vector of these quite
likely can.

The same as for sample values goes for TX gain: tone it down a bit, if
possible. Does your time-domain signal on RX look clipped? Turn down
the RX gain.

Then: Yes, you can gain quite a bit by using a filterbank instead of
the good ole IFFT. Tom kind of has the blog post (you’d call it
article) to exactly your question:
http://gnuradio.squarespace.com/blog/2014/4/23/peer-review-of-a-dyspan-paper.html

And Tom wrote the polyphase filters in GNU Radio. Feed one big
tons-of-taps channel filter in, and apply a bit of elegant math,
smaller sub-filters and a bit of rotational magic done with an IFFT,
and you get nicely filtered channels. You should give it a try, it’s
fun :slight_smile: You might want to grab fred harris’ slides on polyphase
filters, if you’re interested in the background of how it works, which
should be around somewhere in the internets, but I can’t find that
bookmark :frowning:

Greetings,
Marcus

On 17.10.2014 20:29, Vahid B. wrote:

single pair transmission and reception is fine while transmitting
OFDM signals to USRPs to limit their out of band emissions, but as

_______________________________________________ Discuss-gnuradio
mailing list [email protected]
Discuss-gnuradio Info Page

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1

iQEcBAEBAgAGBQJUQjGaAAoJEAFxB7BbsDrL12cH/0fC244WhsK1TX8lvOFPogtQ
8y3xCo26gxVQ01UgmwgybsFE3BukVqG/i8Skk6u0HTgXdzD0oQmC/lJUStYVJwJ6
BwekwvJov7g+OoIpq7QltmHyXcK85SmqCkkRkk/3vShpVXz7CT2SPUdN9GsBSLiN
FtOUjD0qNFPgjjli9lEAzFp0KlPe3NzWGoY9OaXswvwMxyNUYZNrxLThVy+5DpPn
eeJT/07pIRJkJ4HmwVLXEcMV8AmF6L+lrjeprVZqC+D2tsFm6mpxNyRN6ZzvNU6N
Jbl4XEa53q1zw5nDjoIlENBLk7NXY9YZGvRC5UFiErChYxbrgTBy4sK56XPJSEU=
=sQIq
-----END PGP SIGNATURE-----