Re: [Discuss-gnuradio] continuous data collection?

[Clark Pope]

----Original Message Follows----
From: Eric Blossom <address@hidden>
To: Clark Pope <address@hidden>
CC: address@hidden
Subject: Re: [Discuss-gnuradio] continuous data collection?
Date: Sun, 26 Feb 2006 17:00:40 -0800

On Sun, Feb 26, 2006 at 11:23:08AM -0500, Clark Pope wrote:
> I've been collecting snapshots through my DBS and USRP to do DBPSK and
> DQPSK bit error rate testing. I'm not able to get the BER below about 
1e-5.
> It seems that slips are occuring spontaneously in the data even though no
> over/underflow is reported during the collection.

What are you using as your input signal?  How's the level/gain?  What
is the avg peak signal level you are seeing from the USRP?  What
decimation rate are you using?  Are you running some kind of symbol
tracking algorithm?

> Questions:
> 1. Has anyone successfully done long term collections without any slips?

Yes.   I've run the gsmk2 code at 500k bit/sec with no FEC and no
errors for many (10's ?) seconds at a time.

> 2. Are the O/U flags sticky?, i.e. if I don't see an under or overflow
> reported then I can be sure the fifos were okay?

You're building from CVS, right?  There was a xrun reporting
problem that was fixed a couple of months ago.

Yes, they are sticky.  They are checked and cleared at about 10 Hz.

> I've done BERs from 1 Mbaud to 11 Mbaud all with the same frequency of
> slips so it does not appear to be speed related.

Eric


Signal is from agilent signal generator. I've set the level, frequency, 
code, etc. correctly. I know the source is fine because I use it on my 
digital radio all the time.

No symbol tracking. I'm just oversampling the 11 Mbaud signal by 4x and then 
looking for max magnitude to decide the best sample phase of the four 
possibilities. The SNR in the resulting constellation is very good (>20 dB) 
so I'm sure the bit errors are not theoretical.

I'm latest from CVS as of last week or so.

In my other post I mention that the current suspect is my matlab code...

Thanks,
Clark

_________________________________________________________________
Don’t just search. Find. Check out the new MSN Search! 
http://search.msn.click-url.com/go/onm00200636ave/direct/01/