Hello Robert, TKS for the correction. I returned to my books. LMS filters are in general linear (LMS-FIR), however they can also be in a recursive structure (LMS-IIR).
>We agree that they are no good for digital modes. Yes the a priori is not very favourable. 73 Patrick ----- Original Message ----- From: Robert McGwier To: digitalradio@yahoogroups.com Sent: Sunday, March 04, 2007 6:15 PM Subject: Re: [digitalradio] Noise Reduction and the digital modes Patrick Lindecker wrote: > Hello Andy and all, > > I don't think NR must be a so good idea for digimodes. Because, it can > be seen as non-linear filter. I disagree on the transfer function. It is an adaptive linear filter. Since it does not mix two tones in the passband, it can't be nonlinear. However, it does indeed introduce serious phase and amplitude distortion on the signals. This is not the way to better copy. These Widrow type/ LMS adaptive filters, in single sample update, or block adaptive form are intended TO AID THE HUMAN FATIGUE FACTOR in listening to noise or interfering tones. We agree that they are no good for digital modes. > In that type of filter, the next sample is calculated, knowing the > previous symbols and guessing what is the most probable symbol if > nothing change (a sort of "no more set of information" condition)... > > You are surely going to produce interference between symbols: the > decoding will be not so good and the necessary synchronization will be > more difficult because the difference between one symbol and the > following will be reduced (i.e the difference between two successive > symbols will be softened). > > But it would be interesting to experiment on calibrated signals and > different speeds (from the PSKAM10 to ALE or PSK220F). > > 73 > Patrick > > > Bob N4HY -- AMSAT Director and VP Engineering. Member: ARRL, AMSAT-DL, TAPR, Packrats, NJQRP, QRP ARCI, QCWA, FRC. ARRL SDR WG Chair "Taking fun as simply fun and earnestness in earnest shows how thoroughly thou none of the two discernest." - Piet Hine
