[Elecraft] Auto Notch Revisited.

[David Woolley]

Whilst summing the inverted NR output doesn't sound logically any 
different from applying a custom filter, the tactic used by hearing aids 
(at least Oticon ones) is actually to determine the frequency and phase 
of the presumed interfering signal (howl round in that case) and add in 
an inverted version.

That does, have the disadvantage that if the tone is really external and 
than goes away, the cancelling tone remains for a while and you actually 
get an interfering signal as an artifact.  That would mean that it would 
not work well for modulated tones, e.g. morse signals, rather than pure 
carriers.

I don't really understand all the references to least squares.  Least 
squares is basically about the simplest way of matching parameters of an 
expected signal to an actual signal in noise.  What is interesting is 
the choice of parameters being fitted, and the expected form of the 
signa, not that there is a least squares fit.

If what is really being described here is to do a least squares fit of a 
sine wave and then add the inverted version of this, that would work 
better than an automated notch filter for continuous carriers 
(essentially because it would be extremely narrow bandwidth, much 
narrower than could be achieved with acceptable group delays with a 
filter), and would not impact on speech, but would not work at all well 
for modulated carriers (effectively its bandwidth is much lower than 
that of a morse signal).

Incidentally, the real research funding for noise reduction of speech is 
in the hearing aid industry - that's essentially what modern digital 
aids are about.  People with moderate to severe through profound loss 
often complain about the artefacts produced by the noise reduction and 
speech compression.  People with profound loss, who have previously used 
analogue aids, often have difficulty moving to digital aids because they 
can't get used to the artefacts.

-- 
David Woolley
Owner K2 06123


On 31/07/14 22:01, wa9fvp wrote:
> Back in the early 90's I worked with another Engineer on a project we called
> "The Hamblaster". I published two articles in QEX (October 1991 and January
> 1992). We developed a PC based sound card that connects to the receiver's
> audio and performed, among other things,  an auto-notch and noise reduction
> (NR).  In our DSP software, the auto-notch was generated by summing the
> inverted NR output to the input signal.  Whether it's a noise reduction or
> an auto-notch we still used a software algorithm called the LMS (Least Means
> Square).
>
> Most Ham transceiver use the same LMS algorithm and I never heard distorted
> audio when the notch was engaged.  That's especially true with Yaesu, Icom
> or Kenwood. I'm not sure about the other American brands but I bet they are
> OK as well.