Re: [Discuss-gnuradio] OFDM PHY (802.11) ?

[Brian Padalino]

On 2/16/07, Tom Rondeau <address@hidden> wrote:
> Our biggest concern will probably be getting the signal to work on the RFX
> boards, which (I think) all have the same output amplifier stage. What we do
> with all of the modulators is scale it from +1 to -1 (even my QAM modulators
> are normalized amplitudes), then digitally amplify it to +-16384. When you
> look at the output power of the transmitter, at full power it only starts to
> go non-linear and you can start to see the third-order product (Matt's done
> a really good job with these guys). At around +-15000, I don't really see
> anything, so that's about the arbitrary max I use.

Ah, so you really want to go full scale with this - neat.  I am
worried about what type of transients you could see if you do some big
impulse signal with the interpolation that happens when transmitting.
I suppose your 1.0 normalized signal would really be the number of
active frequency bins since you could possibly hit a peak at each one
of those frequencies?  Does that sound right?

On the scale values - a complex signal going down to the USRP is
8-bits real / 8-bits imaginary if I remember correctly.  That should
give +/-127 for those values.

Assuming a real-only signal, the AD9862 has a dual 14-bit DAC which
gives +/-8191 values.  This probably doesn't mean anything since it
goes through interpolation, but the whole normalization of this stuff
would seem to get quite hairy after a while.

For a fixed-point model, maybe it would be worth while to choose a Q12
or Q15 number to be the standard normalized value - therefore any
bit-slicing done in the FPGA can be reliably done without the fear of
a loss of data?

> With the OFDM work, we still send modulate the maximum value to be 1 (for
> example, we send a "known symbol" at the start of a frame of all 1's), then
> we will digitally amplify it, so there shouldn't be a difference, and we
> shouldn't see any problems with the high peak-to-average ratio. I say
> shouldn't because I'm an experimentalist and don't like to make concrete
> claims until I see it.

You just use the one symbol for a preamble of sorts?  Is that reliable
enough to not false acquire enough to get data through?

> Is that what you're looking for?

Very much so.  Thank you very much!

Brian