On Wed, Sep 24, 2014 at 10:53 PM, Tom Rondeau <[email protected]> wrote:
> On Wed, Sep 17, 2014 at 2:06 PM, Sam mite <[email protected]> wrote: > >> Hi, I am trying the CMA equalizer of gnuradio and for understanding I am >> studying the reference paper mentioned in the doc as well. I have a >> question after going through the code- Why the real and imag part of the >> error signals are being clipped at 1. >> >> float re = gr::clip(error.real(), 1.0); float im = gr::clip(error.imag(), >> 1.0); >> Shouldn't it be "d_modulus" instead of 1. >> >> Regards, >> >> Sam >> > > > The answer is no. The clipping is just to make sure that the error signal > doesn't walk away too far or go to some strange state if the incoming > signal is too ugly. There's nothing that I can find that ties that value to > the modulus at all, though I can see why you might think so. In fact, when > you make the change you suggested, the results get noticeably worse. > > I've attached an example that uses our dynamic fading model to explore the > results. You can select between different UMTS PDP models, both of which > are pretty bad. I found that restricting this error value even more makes > the algorithm a bit more robust, though it likely takes longer to lock. But > when using d_modulus here, the algorithm never actually converges. > > Tom > > > > I was playing around with the flowgrpah you attached and observed the same as you mentioned. I am studying further on CMA and I 'll also play more with your flowgraph. I'll get back on the same thread if I can find something different. Thanks. Regards, Sam
_______________________________________________ Discuss-gnuradio mailing list [email protected] https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
