I’ll give a quick summary of that preso. The longest section is about noise blankers, not noise reduction. He discusses three types. First, the traditional blanker that takes the signal to zero during a detected impulse. Second, an interpolating blanker that replaces the impulse-affected samples with samples that have a linear slope between the samples before and after. Finally, a linear predictive coding (LPC) blanker that uses speech coding, detects large changes in the coefficients, then removes them. Nice for impulse noise, but it would probably fail very strangely for voice QRM because it would train on the QRM as well as the signal. He shows that it is superior to the other blankers for CW QRM. Of course, the other noise blankers are designed for impulse noise, not CW QRM.
For noise reduction, he gives a quick hand wave description of the Ephraim and Malah MMSE spectral noise reduction. That makes a short time base FFT, removes offending frequency sections, then converts it back to the time domain. He mentions recent work on that by some Germans. I’m guessing they are Gerkmann and Hendricks. He mentions, but doesn’t even handwave the LMS adaptive filter approach. I believe that is the algorithm used by Elecraft. Normally, an adaptive filter is used to optimize the filtering for a known waveform, picking it out of the noise. For noise reduction, we do the opposite and train the filter on the noise, then filter it out. The impulse noise blankers and the LMS adaptive filter both work better with a wider receive bandwidth. That gives them more information about the noise and makes it easier to remove. The LPC blanker and MMSE noise reduction characterize the signal, so they should work better with a narrower receive bandwidth. I might have gotten some of this wrong. I did quick listen to the preso and a short literature search. wunder K6WRU Walter Underwood CM87wj http://observer.wunderwood.org/ (my blog) > On Jan 14, 2019, at 6:53 PM, Wes Stewart <wes_n...@triconet.org> wrote: > > https://www.youtube.com/watch?v=FrVDL_-HOds starting at 41 minutes. > Particularly at 1 hour 3 minutes. > > On 1/12/2019 9:51 AM, David Woolley wrote: >> Do you have a reference for an algorithm that will do this? > ______________________________________________________________ > Elecraft mailing list > Home: http://mailman.qth.net/mailman/listinfo/elecraft > Help: http://mailman.qth.net/mmfaq.htm > Post: mailto:Elecraft@mailman.qth.net > > This list hosted by: http://www.qsl.net > Please help support this email list: http://www.qsl.net/donate.html > Message delivered to wun...@wunderwood.org ______________________________________________________________ Elecraft mailing list Home: http://mailman.qth.net/mailman/listinfo/elecraft Help: http://mailman.qth.net/mmfaq.htm Post: mailto:Elecraft@mailman.qth.net This list hosted by: http://www.qsl.net Please help support this email list: http://www.qsl.net/donate.html Message delivered to arch...@mail-archive.com
