> This IS becoming a semi-annual thing isn't it. :-)
>
> It isn't just the topic, it's almost literally the same discussion. I'd
> call it déjà vu, but it's more than just the "spooky feeling" we've been
> here before. We can actually prove it ;-)
>
> Grant/NQ5T
Grant, you are soooo right. It raises the hair on the back of my neck ;)
Jerry said:
> While the effect on the output spectrum is that of a bandpass filter, I
> don't think NR works that way. I think it works on correlation of the
> pass band with a time delayed copy of the pass band. When the time delay
> is a integral number of cycles of the tone signal, that tone adds up
> while the noise doesn't because the noise isn't coherent. That can also
> work for multiple tones, it gets a lot more difficult to create multiple
> tracking filters for those multiple tones, yet NR should work for those.
You hit it on the head. Doug Smith's NR implementation was an "Adaptive
Predictor with Leaky LMS"; an autocorrelation filter. The "leaky LMS" part
allowed the coefficients to be adjusted via LMS-derived feedback to be
adaptable
to the current coherent signal(s), and to leak back to nominal in the absence
of
such signal.
From the source.....
On March 4, 2006, Doug Smith said:
>
> Hi Lin,
>
> No, the original Orion was also an adaptive
> predictor with leaky LMS. The user control
> changed the adaption rate by changing the
> value of mu in the equation and the amount
> of leakage.
> The adaption rate of that system is also
> dependent on the beginning SNR and the
> amplitude of the input signal. I think I explain
> that in "Signals, Samples...." When you increase
> mu, the adaption rate gets faster. Gary states
> on the Ten-Tec Web site that he's modified my
> original code but I don't know what he's done.
> I state that once the filter has adapted, no
> change in mu or the other parameters will change
> the shape of the adapted filter, assuming that the
> input signal hasn't changed. Gary's changes may
> have altered that.
> Thanks for your kind words about my scribblings.
> Just about everything I have is scattered among
> QEX, the Handbook (DSP chapter), my DSP
> book and the Web site.
>
> 73,
>
> Doug Smith, KF6DX
>
I took the NR to task using Spectrum Lab software.
My analysis of 3.037j NR:
The NR creates bandpasses of about 1000 hz wide with medium skirts around the
signals. At NR 9, it also provides downward expansion of 3 dB/dB; 3 dB drop in
output given a 1 dB drop in input once the input level is below the AGC
threshold). At NR 1, the ratio is more like 2 dB/dB.
So when in CW mode, with BW less than 1000hz, I watched a very weak CW signal
on
waterfall and spectrum plots with NR 9 and RF gain adjusted down a bit to
quiet the background noise, and a BW of 500Hz. Whenever a CW element appeared,
I
saw the signal at 700HZ, and with it, the noise across the entire selected BW
increases in step, which made it sound raspy. Closing up the BW removed the
raspiness.
And in conclusion...
Looks like it consists of an adaptive predictor or something similar, employing
many fewer coefficients compared to Doug's. The downward expansion process
helps
by quieting the noise in the absence of signal, even with the tightest BW, so
making CW, for example, more readable. The NR knob doesn't appear to change the
adaption rate as it did with Doug's version. It may be adjusting the leak rate.
Bill, W4ZV, said:
> In looking
> at the Timewave data sheet, it appears to me that it also
> simply builds filters. What you were describing before is
> much more complex, using auto-correlation techniques, etc.
> like JPL does to enhance images from space. The problem is we
> simply cannot do that in the near real time required by QSK...
Then you'll be surprised to hear that all DSP spectral filters are based on
autocorrelation!
Jerry said:
> Allow me to propose a test of the difference between autocorrelation and
> a narrow filter as the noise reduction process. Check a frequency with
> multiple weak signals, maybe a multiple tone HF data link. Note that if
> the NR works it enhances all the signals to noise while a narrow
> tracking filter would enhance only ONE of those signals. Tracking a
> dozen tones with filters would take a lot more DSP processing, and I
> think far more than it has.
Did this tonight, and indeed, the NR adapts to each signal.
Lin
WB1AIW
Grant Youngman wrote:
>> Grant, you old spoil-sport, don't take all the fun
>>out of our semi-annual NR discussions by confusing us with
>>facts and measurements!
>
>
> This IS becoming a semi-annual thing isn't it. :-)
>
> It isn't just the topic, it's almost literally the same discussion. I'd
> call it déjà vu, but it's more than just the "spooky feeling" we've been
> here before. We can actually prove it ;-)
>
> Grant/NQ5T
>
>
> _______________________________________________
> TenTec mailing list
> TenTec@contesting.com
> http://lists.contesting.com/mailman/listinfo/tentec
>
>
_______________________________________________
TenTec mailing list
TenTec@contesting.com
http://lists.contesting.com/mailman/listinfo/tentec
|