Re: [Discuss-gnuradio] Channel Sounding @ 5 GHz

[Johnathan Corgan]

On Mon, Jul 13, 2009 at 08:48, Jordan J Riggs<address@hidden> wrote:

> I'm not an expert on the matter, but I suspect that the XCVR2450 board
> requires some sort of command to select the 5GHz band, which would need to
> be implemented in the sounder's FPGA code.

Actually, in gr-sounder the host is used to tune the daughterboard, so
nothing different is needed for the XCVR2450.

> At a chip rate of 32MHz, you have a resolution of 3e8/32e6 meters.

As the PN autocorrelation function is not a true delta, but
triangular, the effective resolution is about half this (twice the
distance).  This is about 20 meters per bin; really only useful for
outdoor propagation studies.

> The
> length of the PN sequence determines the maximum unambiguous distance at
> which multipaths can be detected. So with a degree-12 sequence, you have
> (2^12 - 1)*3e8 meters to work with.

You're missing the chip rate as a divisor in the above.

A full-length degree 12 code is 4095 chips, which at 32 Mcps, is about
12.8 us in length.  The maximum measurable path delay then is about 38
km.

> With long sequences, however, you are
> limited by the capabilities of the USRP. Depending on your DBoard, you have
> ~20mW of output.

Yes.  The SNR for each of the delay bins will decrease as the path
loss increases for each of the reflected signals, so you need to do
your transmitter power and receiver dynamic range calculations based
on the expected maximum path delay you wish to measure.

> In my experience (with a sounder of my own design), averaging multiple IR's
> appreciably reduces the noise.

You can increase the SNR against uncorrelated noise by doing so.  But
this is limited in practice (see below.)

> While I'm at it, can anyone explain why a complex correlation vector is
> coming out of the sounder FPGA? Doesn't a mathematical correlation return
> real values?

This is a correlation of a real reference PN code with a complex
baseband IQ signal, so the result is complex.  This preserves the
phase of the delay bin.  The received signal, in addition to
multipath, interference, and additive noise, will have frequency
offset, possible Doppler spread, and timing offset (the latter due to
the difference in sample frequency between TX and RX USRPs.)  These
all result in a complex channel impulse response.  If you are only
interested in the delay power, you can do I^2+Q^2 on the impulse
response vector.

Due to the fact that gr-sounder uses a simple O(N^2) serial correlator
without synchronization, the impulse response vectors are very
sensitive to timing offset between transmitter and receiver.  This
results in the correlation peaks being separated by more or less than
the PN code length number of samples, and makes it difficult to
coherently add them to reduce noise.  Using an external frequency
reference on both ends would make a dramatic difference.

I plan eventually (read: someday) to reimplement/republish gr-sounder
with synchronization (I have already done something similar for a
commercial customer), but it will be done on the USRP2.

Johnathan