Re: [Discuss-gnuradio] Equalization class help

[Tom Rondeau]

On Mon, May 12, 2014 at 10:43 AM, bob wole <address@hidden> wrote:

Thanks for your reply Tom. Could I have some  more explanation on it?  if a training sequence at start of each burst could help in fast convergence for burst modem?

--

Bob

Time-domain equalizers just generally take a bit of time to converge, and when there's no signal, it will try to equalize the noise floor, so every burst starts it fresh. It's certainly possible that the training sequence may be enough, and tuning the length of the equalizer and gain may help.

Also, if you know the training sequence, you could potentially use that to get calculate the channel. Look at the correlate_and_sync block, which uses a known training sequence to estimate the timing and phase of a burst. That could be modified to also estimate the channel and initialize your taps.

Tom

 

On Mon, May 12, 2014 at 6:01 AM, Tom Rondeau <address@hidden> wrote:

On Mon, May 12, 2014 at 5:59 AM, bob wole <address@hidden> wrote:

I want to work on equalization for ISI removal due to multipaths for burst frequency shift keying system. I found equalizer class in gnruadio, I want to know if somebody used cma_equalizer for the task with success or if I can use it for equalization of burst frequency shift keying system before continuing with it?

Thanks

--

Bob

It'll conceivably work, but the challenge with the CMA equalizer (and the LMS-DD equalizer as well) is the long convergence time. Between every burst, you're going to lose tracking of the taps, so you might want to think of a way to shut down the tap update loop when you're not receiving a packet (which involves knowing when that's happening in time to send the info to the equalizer). For starters, just use a few taps in the equalizer and a gain higher than you would expect to see if you can get it to converge to "close enough" early in the burst.

Tom