April 17, 2008
Lyle Johnson, KK7P, designed the
DSP and Aux DSP boards and firmware used in the Elecraft K3. Here he writes his “philosophy” on how to get
the most out of Noise
Reduction performed in DSP…
I think there is a fundamental
misunderstanding of what NR is and how it works, at least in the context of the
1) NR is better named SE (signal enhancer).
2) It is a "short" FIR filter whose coefficients are continuously being recalculated.
3) It defaults to suppressing everything.
4) When it senses that there is something that correlates (i.e., has a pattern that doesn't seem to be entirely random), it attempts to build a filter around those frequencies that seem less random.
5) Because the FIR filter that is being implemented is short, the filter being built is less selective than the normal DSP filters in the radio.
6) NR is rarely useful if the bandwidth is narrow. If you set your CW width to 400 Hz or less, for example, there is no point in running NR *unless* you want to use it as a sort-of "smart squelch."
7) The narrower you set your WIDTH, the more that noise appears like a signal, and the worse the NR will perform. And the less noise there is anyway, assuming there is a signal present. NR cannot compete with a narrow filter, and was not designed to.
8) I find NR most useful during CW operation with the 2.8 kHz roofing filter NORM'ed so the Rx bandwidth is wide. Assuming band activity is low, the Rx is quiet. If a CW station comes on within the Rx passband, a filter will be built around the station and I can hear her.
9) Similarly, I find NR in SSB is mostly useful as a sort of squelch when tuning around, or monitoring a frequency you are expecting a call on (probably a net or a sked). I use NR1-1 or NR1-2 for this.
Mild suppression, not too much impact on fidelity, and lets me hear weak signals, too.
I hope this is helpful in better understanding how NR is implemented in the K3.