CW Filter Shape Factors

Paul Christensen, W9AC

Al,

Here's what I have between the filter center and the -6dB points of the 250 Hz filter.  I only plotted the upper frequencies from center.  The lower frequencies will be a mirror image with DSP.  I couldn't say that if this was a crystal filter.

14,175.000 = 0 dB

14,175.110 = -1 dB

14,175.116 = -2 dB

14,175.118 = - 3dB

14,175.120 = - 4dB

14,175.123 = -5dB

14, 175.125 = -6 dB

How does that look?  The measurement is super touchy.  As expected, just a few Hz means a couple dB change when we get close to the knee of curve.

Paul, W9AC







.

Barry N1EU

Yes Al, I believe you're correct and the steep filter is inherently superior.  However in your diagram, would it more accurately reflect real world cw signals if that blue signal had much more gradual/shallow slopes?  If so, those gradual slopes would translate into a smaller difference in performance between the rx filters.

Charles - K5UA

Barry has an interesting point about the bandwidth of a CW signal signal.  I don't know if there is a universally accepted number since I have read various opinions ranging from 100 hz to 500 hz.  I think using a screen capture of the panadapter zoomed in on a steady CW signal would give us an accurate picture of what the width and slope of a typical CW carrier would look like.  Then we could apply our filter curves superimposed on the CW curve to see the difference between effect the best crystal filters and our DSP filters would have on that CW signal.  I would imagine you could cozy up to an offending CW signal much closer with a 50 hz DSP filter with a 2.3 to 1 shape factor than you could with a 250 hz crystal filter with approximately the same shape factor. But as you mentioned Barry, it difference would depend upon the slope of the CW carrier.

So, could someone out there get a screen capture of a panadapter zoomed in on a CW carrier to determine its width and slope so it can be reproduced on a graph for evaluating how DSP filters and crystal filters differ?  With this last bit of information, we will be very close to really understanding and documenting the value of steep filter slopes for congested CW operation.  Once this is graphed out, I think it will vindicate the programming effort and the processor clock cycles dedicated to achieving these steep filter slopes.  Thanks to all who participated in this discussion.

Barry N1EU

My previous reply seems to have disappeared.  Using the last example spectral diagram that Al posted,  I would suggest the signal in blue would have much shallower slopes.  With the steep blue slopes shown, there is a 35-40dB improvement in rejection afforded by the steeply sloped rx filter.  But if the true real world tx slope was much shallower, that improvement might only be 5dB for example.

Al_NN4ZZ

Barry,
I think your previous comment is still there.....when there are a lot of comments, the community software collapses them.  You have to click on the "view more comments."  It's easy to miss.  




I agree that representing real world conditions are much more complex.  (wide CW, multiple signals close by, clicks, hash, etc).   The blue one I added was just to show how the narrower filter could help compared to the inrad.   Real world could be a lot different. 

Regards, Al / NN4ZZ  
al (at) nn4zz (dot) com

Steve-N5AC

The bandwidth of the CW keying signal is a function of the keying envelope of the signal.  Before the keying of the CW signal, there is no signal and after the keying (while the dot or dash are a continuous amplitude sine wave), the bandwidth is infinitely small (or really the width of the phase noise, which is very small).  So it is mostly about the duration of the rise in amplitude and the shape of that rise.  We looked at several keying waveforms for the FLEX-6000.  Each waveform distributes the energy of the rising amplitude in different ways -- some waveforms spread the energy across a very broad area, but at a very low level.  Some waveforms place most of the energy in sidebands of the signal itself making the signal look wider.  If you wanted to know, Matlab would give you very good details on the bandwidth of a keying waveform provided you know the keying function.

In the FLEX-6000, we use a single keying waveform function and we have only two rise-times that we adjust between (as of v1.1).  These two things set the bandwidth of the CW signal.  If you want to look at this with a spectrum analyzer, you have to be very aware of the window filter used in the analyzer because it will color the results -- what you see is NOT what you get.  Your brain does its own FFT on the CW signal in order for you to copy it.  If I was going to do the analysis with a spectrum analyzer, I would capture a CW signal, isolate the keying or unkeying waveform and do an FFT on it with very narrow bins (1Hz or less).  This will give you a less colored view of the actual spectrum occupancy of the signal.  Then your accuracy would be to 5Hz or so after bin leakage.  If you are using 5Hz bins, your accuracy is going to be in the 10-15Hz range which is probably more than you want for this kind of analysis.  

Al_NN4ZZ

 Ken - NM9P, Elmer

Thanks, Al.
I figured you would do it even without my challenge... ha.
Impressive performance.....
Now, can you superimpose a 50 or 24 Hz filter on top of the INRAD so we can REALLY brag?

Ken,
Paul provided data for the 100 HZ Flex filter so I superimposed it on the INRAD 250 HZ filter.  It should be a fair approximation and even at the 1.6 shape it looks brick wall in comparison.  At this scale the FLEX 50 HZ filter would look similar just half as wide.  It's easy to see why using a good 50 or 100 HZ filter (that doesn't ring) makes such a difference compared to the 250 HZ crystal filters.   

Regards, Al / NN4ZZ  
al (at) nn4zz (dot) com

Paul Christensen, W9AC

Al, Not sure if you missed it but I provided the detail of the 250 Hz filter in 1.0 dB steps between the center frequency and the -6 dB point. This will give you better detail of for the top graphical passband response. The message may be collapsed in between others. Paul

Al_NN4ZZ

Paul,
Yes overlooked it, but got it now.  I'll redo the 250 HZ plot...
Regards, Al / NN4ZZ  
al (at) nn4zz (dot) com

Al_NN4ZZ

Paul,
Here is the updated 250 HZ filter with your additional close in data points.   At the scale used for the Inrad filter graphic, it is hard to plot that level of detail very accurately.  But as you suspected the top is much flatter than my first picture. .

Regards, Al / NN4ZZ  
al (at) nn4zz (dot) com

Charles - K5UA

Here's a cheap and dirty research project protocol to determine the relative effectiveness of a 250 hz DSP filter with a 1.2 shape factor and an INRAD 250 hz filter wth a 2.2 shape factor: 1. Use a signal generator to insert a continuous carrier and use the 6500's 250 hz filter to determine the carrier's maximim S-meter reading. Then tune either above or below the carrier frequency until the 6500's S-meter is down 60 db. Write down that frequency. 2. Use a receiver with a good INRAD 250 hz filter, use the same signal generator set to the same frequency and signal level as above, and determine the maximum signal strength of the carrier on the S-meter. Now tune to the same frequency that the 6500 showed the carrier to be 60 db down from its maximum. Write down this S-meter reading and determine the difference in db from its maximum. 3. The difference between the INRAD filter differential S-meter reading, and the 6500's 60 db differential S-meter reading will illustrate how much more attenuation the DSP filter will provide in a real-world environment. Yes, the S-meter in the INRAD receiver will not be as linear as the one in the 6500, yes it will be harder to read, yes the steady state carrier is not as real-world as a keyed CW signal.........but this protocol eill give a fair assessment of how these two filter types will perform in a real RX environment. We'll need Rob Sherwood to do the definitive tests, but this will give us a good idea of what to expect. Thanks again to all that have made this such an interesting discussion.