CW Filter Shape Factors

Charles - K5UA

Has anyone with a signal generator measured the CW filter shape factors by measuring the -6/-60 dB points. It may be a little tough to do with the 10 hz frequency step and the S-meter not being calibrated in decibel level, but has anyone attempted this. I'm only asking because, after years and years of using 1.4:1 shape factor crystal filters, I'm not hearing what I expect to hear from "brick wall" digital filters. By using the term "brick wall" filters, I am assuming that this definition implies a much better approximation to the 1:1 shape factor of a "perfect" filter than is possible with analog filters. This is why I would like to know if anyone has measured this using the 6500 narrow CW filters.

Al_NN4ZZ

Charles, FYI, if you use DDUtil and a FlexControl knob to control the slice frequency you can set the frequency step down to 1HZ. See attached snapshot. By the way, the filters work better than any I've used so I'll be interested to see what you find out. Regards, Al / NN4ZZ

Charles - K5UA

Thanks for the tip about DDUtil Al. My interest, however, is not so much in the -6 dB filters width being created by SSDR or by DDUtil, but in the -60 dB filter width. Shape factor is the ratio of these two widths for any particular filter. A very good crystal filter would be one with a 1.4 to 1 ratio. For example, a 500 hz analog filter with a 1.4 to 1 shape factor would be 500 hz wide at the -6 dB points and 700 hz wide at the -60 dB points. So a 500 hz filter with a 3:1 shape factor would be a very poor filter for serious CW work. So why am I interested in this? Well, "brick wall filters" imply a filter shape factor of 1:1. I'm just not sure that the 6500 CW filters are ,at their current state, close to that very high standard. Tuning around CW signals with the 50 hz and 100 filters has given me reason to question the shape factors of these digital filters. That's why I would like someone with a signal generator to determine the actual shape factors of the 6500's CW filters. I'm not saying that they are not good or better than the competition, I just would like to know what the shape factors actually are since I am not hearing a "brick wall" response as I tune across a CW signal. I am not trying to be overly critical, I just would like to know what kind of filter performance we have in our arsenal.

George KF2T

Hi Charles, I agree with your assessment. I think they lost a little on the filters in this beta. They did seem sharper in the original release version, although I have not done any precision measuring. 1.1 should hit in the next couple of weeks; it will be interesting to see how it works out.

Charles - K5UA

I'll always been critical of analog filter shape factors after spending lots of money on poor shape factor filters over the years. Not that these filters in the 6000 series are poor, I don't have a way of accurately measuring them. I know the PSDR filter shape factors were highly dependent upon the sample rate and pan-adapter width selected, and there were plots given in the documentation to help us decide on the trade-offs to get decent shape factors without undue latency. As I mentioned earlier, I'm not hearing "brick wall" performance of the CW filters, especially at the 50 and 100 hz widths. A crowded CW band in a contest can have signals often less than 50 hz separation, so filter shape factors of 1.05 to 1, or 1.1 to 1, would be close enough to "brick wall" to make a tremendous difference in competitive CW work. If there is a penalty to making really good shape factors at these narrow bandwidths, and trade-offs are necessary,just let us know so we can accept the best compromise and we can move on. I've seen great posts by Steve that are concise about engineering decisions that had to made in such areas, so maybe he can nail this down for us.

Ned K1NJ

Charles,
               Did you reach any conclusions about the shape factors?  Did I miss
some info from Flex?  The vs. 1.1 filters seem fine, but are there some numbers
we can quote on the air?

Ned,   K1NJ

Paul Christensen, W9AC

As a quick test, I used the Flex's S-meter and an HP signal generator. CW mode, 20m, 250 Hz default BW. No AGC, but it does not matter in this architecture when observing SSDR. Here are my -6dB/-60dB numbers:

f1 = 14,175.340 (-60 dB)

f2 = 14,175.370 (- 6 dB)

f3 = 14,175.630 (-6 dB)

f4 = 14,175.660 (-60 dB)

For the shape factor, use the formula [f4 - f1] / [f3 - f2]

[14,175.660 - 14,175.340] / [14,175.630 - 14,175.370]

The resulting shape factor is about 1.2 for the 250 Hz filter. A few important points that affect computed shape factor accuracy: First, I am unsure of Flex's S-meter accuracy over a 60 dB level span; and secondly, small changes of a few Hz make a huge difference with digital filters. Better accuracy will be observed when moving in 1 Hz increments, rather than 10 Hz (e.g., using DDUtil).

Paul, W9AC

Charles - K5UA

Ned, look at Steve response to KA4B's post about Filter Quality vs Panadapter Bandwidth. Steve has a way of explaining why certain things are the way they are. As you noted, the 1.1 filters are improved over the 1.024, and I hope they continue to refine them with each new version. There is a time delay through a digital filter, the better the folter, the more the delay. Since I do not use QSK, delay is not that big a factor for me since the station listening to me can't reply until I finish anyway. Doubling the delay through the filter would not really affect CW ops like me who are non-qsk. However, they need to keep a balance for both qsk and non-qsk ops.

Steve-N5AC

In v1.1, the relative accuracy of the S-Meter should be better than 1/2dB.  The absolute accuracy in v1.1 is +/-3dB.  This should return to something like +/-1-2dB in a later release.

Also for CW the filter shape factor changes at 400, 1000 and 1500Hz to lower latency with a wider filter.

Charles - K5UA

Thanks Paul, 1.2 to 1 is excellent, considerably better than the best of the amateur crystal filters (1.4 to 1) that I have used in the past. Just for grins, would it be possible for you to run the test on the 100 hz filter? Steve commented above that the shape factor changes at 400 again to lower latency, although I must admit that I'm not sure if he is impling that the shape factor for the 400 and wider filters will have wider skirts to achieve less latency. Steve, if you see this post, could you amplify your comment about the wider filters? Are you saying that the 400 and wider filters will have shape factors higher (filter skirts are less steep) than the 250 filter tested by Paul?

Steve-N5AC

Off the top of my head I can't tell you if the shape factor is governed strictly by the number of taps in an FIR filter or if the chosen bandwidth plus the number of taps will set this.  What I can tell you is that the number of taps in the filter goes down by a factor of two at each of these crossings at the present time and I know that will affect the shape factor.  So there are two possibilities:

1) shape factor is independent of filter bandwidth chosen (I know how this sounds, but I believe it is a possibility since all that we are analyzing is one side of a skirt) in which case you will also see a 1.2 for the 100Hz filter.  If this is the case than all filters in each range will have the same shape factor (0-400, 401-1000, 1001-1500, etc).

2) The other possibility is that the shape factor is bandwidth dependent in which case there will be a progression from 0-400 and then a jump to 401, a progression from 401-1000 and a jump at 1000, etc.

Let me know which it is because I'm curious too!

Paul Christensen, W9AC

Same test conditions, but using the 100 Hz filter in CW mode and DDUtil V.3.0.5.00 for 1 Hz frequency steps.

f1 = 14,163.715 (-60 dB)

f2 = 14,163.748 (- 6 dB)

f3 = 14,163.848 (-6 dB)

f4 = 14,163.880 (-60 dB)

For the shape factor, use the formula [f4 - f1] / [f3 - f2]

[14,163.880 - 14,163.715] / [14,163.848 - 14,163.748] = 1.6

The resulting shape factor is about 1.6 for the 100 Hz filter.  Note the exact 100 Hz BW between the f2 and f3 - 6dB points. 

I went back and re-confirmed the 250 Hz CW filter setting, this time using DDUtil in 1 Hz steps.  Same S.F. result of 1.2.

Paul, W9AC

Steve-N5AC

OK makes sense ... I suspect if you do this at 399Hz it will be a shape factor of 1.1 and when you go to 401 it will drop down again.

Paul Christensen, W9AC

It was bound to be asked: "What's the shape factor of the Flex 6K's typical SSB filter."  Here it is:  same test conditions but in SSB mode, 2.7K filter and using DDUtil in 1 Hz steps:

f1 = 14,159.616 (-60 dB)

f2 = 14,163.649 (- 6 dB)

f3 = 14,162.348 (-6 dB)

f4 = 14,162.381 (-60 dB)

For the shape factor, use the formula [f4 - f1] / [f3 - f2]

[14,162.381 - 14,159.616] / [14,163.348 - 14,163.649] = 1.02. 

Again, note the exact bandwidth of 2.7kHz between the f2 and f3 - 6dB points.  That 1.02 shape factor is not a misprint  :-)

Paul, W9AC 

 

Steve-N5AC

Yes, we made the decision that a latency of 85ms doesn't bother a SSB operator and so we went for best filter for SSB all the time.  So the shape factor gradually increases the wider you make the filter.   It will have the same shape factor as the corresponding CW filter when your frequency is 0-400Hz, but it just gets better the wider you go unlike the CW filter.  Same goes for AM.  Only CW and DIGx are graduated.

Paul Christensen, W9AC

Steve, Maybe way down the roadmap, two DSP filter types could be selected for narrow positions, say...400 Hz and under. Icom does something similar in their '7700 and '7800 series. They use "hard" and "soft" terms to describe a choice of sharp v. wider shape factors. I no longer recall the S.F. differences. Wider S.F could be used for fast QSK, while a narrower factor could be used in instaces as George described where it's more important to dig out a signal from adjacent QRM. Paul, W9AC

Al_NN4ZZ

Paul,
Good idea......we frequently slow down a bit for the weak ones anyway.   Either an option as you suggested or just automatically sharpen up when the speed is below a certain value (e.g. 25 WPM?)

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

Al_NN4ZZ

Paul,
I just realized why the automatic option won't always work.   The speed is only known when using the internal keyer.  So your idea of the "sharp" option is better but I guess it also has a risk  (increased latency) if engaged when QSK at high speeds.   

Does that make sense?  

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

Paul Christensen, W9AC

Operationally, I think it makes sense to make it a manual change.  But for Flex's software engineers, the coding may not be trivial! 

Stan VA7NF

Don't get hung up on the ratios.  Both 100 and 250 wide have about 30hz from -6 to -60 on each side, only the initial width changes.  The math makes the ratio worse.

Do this on a 10hz bandwidth and the ratio would look horrible.

Besides, without a slight slope I couldn't demodulate a 2M FM signal on an AM filter slope.

Charles - K5UA

Thanks again Steve for pulling back the curtain a little so we know how it all fits together.  

If I can summarize, from 50 hz to 399 hz the filter shape factor gets better and better as it approaches 1.1 to 1 at the 400 hz boundry, then goes back to something like 1.6 to 1 at 401 hz and gets better and better until it approaches the next boundry at 1000 where it will be around 1.1 to 1 again.

The 1.6 to 1 shape factor at 100 hz is pretty impressive. And what can I say about 1.02 to 1 for SSB filters.  That is truly "brick wall". 

As Al and Paul said above, a non-QSK operator would not mind a little more latency through a 50 or 100 hz filter, if doing so could produce a 1.1 shape factor.  As you have probably witnessed on the recent DXpeditions, QRMers often get very close to the DX station's frequency.  It may be possible to attenuate the QRMers enough to copy the DX station with a 50 or 75 hz filter that has a 1.1 to 1 shape factor.  As for latency......"Latency is in the eyes of the beholder."

Good job with the Ver 1.1 filters Steve. 

Charles - K5UA

The comment by Stan,VA7NF, about the way shape factor is determined has stuck in my head ever since I read it. After a few calculations with the measured values by Paul on the 100, 250, and 2700 filters, it appears that Stan has a valid point. On Paul's three measurements, the distance between the -6 and -60 db points on all filers was about 33 hz. This is a slope of about 1.6 db per hz for all three filters. When applied to a 2700 hz filter, the shape factor comes out to be 1.02. Wow! Brick wall, right? Why can't we create a 100 hz filter with skirts like that? Well, they did! Paul's measurements show the same 33 hz distance between the -6 and -60 db points on the 100 hz filter. Only this time, the formula for shape factor gives a value of 1.6. So we hang our heads in sorrow that the 100 hz filter does not have as steep a skirt as the 2700 hz filter. BUT IT DOES!!! If you were a downhill skier looking over the edge of a 100 hz filter or a 2700 hz filter, you would see the same "steepness" of the slope...... a fall of 54 db in 33 hz, or 1.6 db per hz. When the edge of that 100 hz filter is applied against an unwanted signal, it is going to attenuate that unwanted signal just as effectively as that 2700 hz filter. From these three measurements by Paul, I am inclined to believe that all these digital filters are going to have the same slope factor of 1.6 db per hz. When drawn out to scale, it makes a 2700 hz filter appear to have virtually vertical sides. When drawn to scale, it makes a 50 hz filter look fair, and a 25 hz filter downright pathetic. But we must remember, it is the wall of the filter that is applied to the undesired signal, and a 54 db attenuation over a span of 33 hz is remarkable. For me, Stan's post is an epiphany because I finally understand that shape factor is a mathematical enigma. As filter width decreases, shape factor gets worse even though the slope of the skirts remain constant. The following list of filter width versus shape factor assumes a constant filter slope of 1.6 db per hz. 25 hz filter ----- 3.64 shape factor 50 hz filter ----- 2.30 shape factor 100 hz filter ----- 1.66 shape factor 250 hz filter ----- 1.26 shape factor 400 hz filter ----- 1.16 shape factor 600 hz filter ----- 1.11 shape factor 800 hz filter ----- 1.08 shape factor 1600 hz filter ----- 1.04 shape factor 3200 hz filter ----- 1.02 shape factor Take comfort in the fact that all these filters have the same "brick wall" slope of 54db per 33 hz (1.6 db per hz). Thanks Stan, you nailed it.

Paul Christensen, W9AC

Good analysis, Charles.  What you've shown also applies to crystal filters, but unlike the mathematical perfection attained in DSP, as the crystal filter's bandwidth decreases, even slight slope imperfections have an impact on shape factor and symmetry from center. 

Putting this into perspective, look at the shape factor of the best INRAD crystal filters.  Due to manufacturing cost and inability to attain repeatable tolerance, I don't believe they offer one under 250 Hz. Their typical 250 Hz filter (at 8 MHz) has a shape factor of 2.2. 

https://www.inrad.net/product.php?productid=203&cat=90&page=1

In addition to nearly perfect control of the slope, the DSP filters are providing much better ultimate filtering well beyond the -60dB point where crystal filters are subject to degradation due to I/O isolation and other ****-by effects that limit ultimate filtering performance.  Thru-loss also escalates as a crystal filter becomes narrow.  Look at the schemes developed by competing manufacturers to compensate for that loss.

It would be interesting to plot an overlay of the Flex's filters on top of a comparable INRAD crystal filter. 



 

Ken - NM9P

NN4ZZ will probably have that graphic by tomorrow!  He is really good at that stuff. [gauntlet dropped....grin]

Al_NN4ZZ

Ken,
Here you go!   Below is a graphic for the inrad 250 HZ filter in BLACK and the FLEX 250 Hz filter overlaid on it in RED using Paul's data.

6 Db = 260 HZ
60 Db = 320 Hz

The 1.2 shape looks very good to me!

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

Paul Christensen, W9AC

Nice work, Al.  Probably the top portion on the Flex plot is flatter than shown.  It would just require me to capture more data points on Hz-by-Hz basis between filter center and the -6 dB points. 

Paul, W9AC

Al_NN4ZZ

Paul,
This is probably more like it then....also moved the 60DB points slightly.  

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

Barry N1EU

I know this is a dumb question, but doesn't the shallow (wide) effective "shape factor" of a cw transmit signal mean that the steepness of the rx cw filtering isn't that critical as long as it's "reasonable"?  In other words, even if you have a 1:1 rx filter, isn't the adjacent cw signal within 100-200hz that you want to reject still going to bleed into your passband because all cw tx signals are a bit wide?

Al_NN4ZZ

Barry,
If there is a close by interfering signal, then the brick wall filter will be better.  Assume the BLUE signal is the offending one in the snapshot below.   The Yellow is the the overlap for the Inrad filter.  Much less for the Flex filter.   It probably makes a lot more difference when the signal you want to hear is a weak DX signal.  

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

Ken - NM9P

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 - NM9P

Of course, no filter can remove junk from a ****, phase noise ridden signal, CW, SSB or other, that is splattering the bands.  The best we can do is make sure that our OWN receive chain - oscillators, filters, mixers, etc - do not add to the problem. 

I demonstrated my 6500 to the local ham club last night.  I had it connected to my 20 Meter Hustler mobile antenna with a long jumper fed through a window into the meeting room. 

I started with the big pileup on 20 CW from one of the big DXpeditions.
I used the 3K wide filter and it was mayhem!  and said "This is what most of our rigs would hear without a good CW filter.
Then I switched in the 250 Hz filter and said "This is what it sounds like with a good CW filter."  There were still 3 or 4 signals in the passband, but several people said "Wow, pretty nice."

Then I said, "You haven't seen anything, yet." 
 I set the 50 Hz filter and the Audio Peak Filter and there was only ONE signal, weak, right beside a very strong signal.  And it stood out from a practically noiseless background.

That turned some heads!  And then I went A/B between the 3 Khz wide and 50 Hz filter several times.  Going split with the 50 Hz filter on one frequency and a 400 Hz on another freq about 20 Khz away with one in the left ear and the other in the right polished off that part of the demo.  "And that is how I worked the big DXpedition barefoot with a dipole!  I can't wait for my T-11 up 55 ft.!)
I had to brag a little bit.....  But as they say in Texas..."It ain't braggin' if ya can do it!"
And this FLEX can do it!

Ken - NM9P