Flex 6700 v. Elecraft K3 CW Bandwidth Measurements

Paul Christensen, W9AC

Transmitted CW bandwidth comparison between Flex 6700 and Elecraft K3 transceivers using latest versions of software/firmware in both transceivers.

TEST CONDITIONS:

Agilent N1996A Spectrum  Analyzer; 10 Hz RBW

40m, CW mode, swept from 7040 kHz to 7060 kHz, 2 kHz per horizontal division

External CW keyer used and set to 60 wpm for all measurements

Internal oscillators used for both TRX (not external Ref. input)

RF sampled from Bird variable RF coupler

Output power: 50W into Bird TermaLine load

--------------------------------------------------------------------------

Flex 6700 60 wpm slider setting (YELLOW TRACE)

Flex 6700 10 wpm slider setting (LAVENDER TRACE)

Elecraft K3 QRQ Mode (BLUE TRACE)

CONCLUSION:

At +/- 1 kHz spacing from Fc, occupied bandwidth is very similar at approximately -70 dBc.

Paul, W9AC

Steve N4LQ

Paul
I would like to see your results with some other rig for comparison. Something from Japan maybe. 
Steve

Paul Christensen, W9AC

Steve,

At the moment, I have nothing from JA in the shack.  However, I just measured the Expert Electronics SunSDR2 using the same test conditions but with two different CW rise/fall settings.  

The GREEN trace is with the CW waveform rise/fall set to 5 ms in software.  The YELLOW trace is with rise/fall set to 2 ms.  With the faster ramp time, there's more than a 20 dB difference at +/- 1 kHz from Fc.

Paul, W9AC

 

Steve N4LQ

Many of the JA rigs have CW rise time settings and some guys think they sound better with "hard" keying and set them to 2 ms. The TS-590 for example sounds really harsh and wide with the 2 ms setting. I've talked a few guys into increasing their setting to 6 ms and it sounds so much better. So this test certainly shows the difference! 

Sergey R5AU

Paul, superb, thank you for measurument results ! 

KY6LA_Howard

Paul

How do you reconcile your results with the results published in QST for the K3 which show a substantially lower noise

Tim - W4TME

I believe the ARRL was running v1.3.8

KY6LA_Howard

I understand that the new results from 1.4 are better. But the ARRL results show the K3 hovering around -90dB @ 2KHz out and Paul's results only show -80dB?

Paul Christensen, W9AC

Easily explained. I did not optimize analyzer dynamic range for noise teasing. The ARRL likely activated their internal analyzer preamp then reduced RF input level to reduce the noise floor. This is similar to us engaging the preamp on our Flex rigs and seeing the noise floor drop

A true apples-to-apples comparison requires knowing the ARRL's analyzer test parameters, including preamp, RBW and VBW settings. Optimizing would probably drop the noise floor by 10 dB although the occupied CW bandwidth at 1 kHz would be unaffected and that's what I was most interested in.

Paul, W9AC

DH2ID

Great, Paul. I will put my FT-897, IC-756PRO, TS-2000X and KX3 on the bench and use your settings. My spectrum analyzer can easily handle that. Only glitch could be that I have no Bird equipment, I'll use the RF coupler I built last year. Will write about the results here. 73, Alex DH2ID

N7AIG

It would really help if you could overlap the plot images, or else plot their dB difference vs frequency. Offhand, these look so similar to each other that it is difficult to see any conclusion between the FRS 6K and the K3. I'm very interested in these results.

Also, what exactly do you mean by K3 QRQ mode? What keying speed? It would be interesting to compare real-world speeds from 20 to 30 wpm. This has mostly to do with keying envelope shaping and rise / fall times. It would be nice to compare apples to apples here.

Assuming both rigs use reasonable rise / fall times, around 6-8 ms, and both use raised-cosine shaping, then the only difference should be in the phase noise of their LO's, eh?

73 de Dave, N7AIG

Paul Christensen, W9AC

They look similar because at the -70 to -80 dBc level the occupied bandwidth is essentially the same.  That's pretty far down the log scale.  Bringing the analyzer noise floor down to -90 dBc would show more detail about noise sources, but I'm satisfied with the result. 

> "Also, what exactly do you mean by K3 QRQ mode? What keying speed?"

As I previously mentioned, keying speed was set at 60 wpm for both transceivers.  You'll need to read-up on the K3's QRQ mode.  In a nutshell, Elecraft incorporates a fast T/R system with operational limits.  Until the latest synthesizer update, their QRQ mode did not allow one to use offsets, like split and RIT while in that mode.  That's mostly been resolved as of the latest K3 hardware update.

> "It would be interesting to compare real-world speeds from 20 to 30 wpm."

Bandwidth does not change between 20-30 wpm on either transceiver, nor should it.  Flex does not change the waveform shape until about 32 wpm and I've shown little difference at the break point, down to -70 dBc at 1 kHz.  Going below that would certainly show the effect of noise, but this far down, I think it's fair to say that keying bandwidth is essentially the same between the Flex and K3. 

This test was about keyed transmit bandwidth and not phase noise.  There's another test for that. I just recently acquired a phase noise test suite for another spectrum analyzer, an HP 8560E.  The phase noise testing module fully automates testing in about 45 seconds. I'll make comparative measurements as time permits.  Between then and now, let's look forward to DH2ID's results.

Paul, W9AC

N7AIG

Ah yes... on careful re-reading, you did state "external keyer at 60 wpm". What threw me off was your presentation of two graphs for the 6700, one at 10 wpm and another at 60 wpm. I own a Flex 1.5K and 3K, as well as a K3. So I don't know what the function of the wpm-slider is on the Flex 6700. Can you describe?

BTW... looked you up on QRZ.com. Very impressed by your handiwork!!

73 de Dave, N7AIG

George - AB4FH

Paul,
Thanks for sharing your test measurements. I own the KX3 for portable use such as backpacking and Field Day.  It and the Flex 6700 are excellent CW rigs.

Steve-N5AC

Just FYI under v1.3.8 we had the rise time set to 2-5ms and it is now 6-8ms, hence the lower bandwidth.  The bandwidth of the CW signal is directly proportional to the shape of the rise and the duration of the rise time.

Also when we did our testing, I found that there is no free lunch.  When you go from nothing to full power in 5ms, you get a "fixed amount of noise" generated.  You can place all of this right next to the carrier or you can spray it all around the band.  I could (if I wanted) design a ramp that will do the latter and will make the ARRL plot look better than any others.  The result will be that low-level noise is sprayed all over the band.  So be careful what you measure ...

KY6LA_Howard

Thanks Paul, Tim and Steve for sharing all this information.

Now I just need to get my K-3 buddies to read this and to stop their Flex has wide CW nonsense.

rfoust

I'm colorblind - I can't tell which one is lavender and which one is blue. (seriously)

KY6LA_Howard

Of the top 3 plots

#1 Top One  Flex 6700 60 wpm slider setting (YELLOW TRACE)

#2  Middle  Flex 6700 10 wpm slider setting (LAVENDER TRACE)

#3  Bottom iElecraft K3 QRQ Mode (BLUE TRACE)

Paul Christensen, W9AC

The Flex 6K and K3 are both extraordinary transceivers in different ways.  I'm fortunate to have both with no plans to sell either one.  Both transceivers have well-controlled CW ramp times and very well-managed ALC systems. 

Provided a transceiver's CW waveform shape follows a Gaussian curve, there should be very little difference among them, but there is.  The disparity is almost always related to how ALC is managed.  The Ten Tec Omni Six Plus, is such an example that goes back twenty years. 

The original Omni Six had a beautiful raised and decayed CW envelope.  Then, something bizarre happened when the "Plus" version was produced.  Ten Tec tampered with the ALC time constant and the result was a very asymmetrical waveform with a sharp 1-2 ms. leading edge that looked like the ramping of a triangle wave when expanded on a scope.  Back then, I didn't own a spectrum analyzer but by ear, I could detect who was using an Omni Six Plus (N4LQ could too). 

I devised a simple dual speed ALC that accomplished the original goal, yet left the leading edge intact.  That mod was done some 18 years ago, long before folks were talking about the impact of ALC on a keyed CW waveform.  The revised circuit changes are shown on N1EU's website.

Paul, W9AC 

Charles - K5UA

Hi Paul, Isn't there a formula about bandwidth and keying speed? Your plot at of the 6700 bandwidths at 60wpm and 10wpm look so similar that I wonder if that formula Would need to be revised ? Also, if Steve ever has a few minutes to burn, it would be interesting to know the details of how the waveform could be optimized by distributing the noise over the adjacent spectrum. Not that I recemmend that, inquiring minds just want to know. Anyway, good job Steve. I have no axe to grind concerning the K3, but it's nice to know the Flex 6X00 series is in the same league concerning CW waveform.

Steve-N5AC

Unfortunately, I don't believe we recorded all the data.  What we did was to take a few different functions to shape the waveform and we looked at the effects of each.  Once you get to any function that has relatively soft edges, the effects are very subtle.  I don't believe you could every distinguish them at a distance, but the do impact local noise.  So I would say it probably matters for a multi-2 or multi-multi, but not much else.  I'm guessing you could simulate it in Matlab fairly easily and duplicate what we did empirically/clinically..

N7AIG

what does that wpm slider do?

Steve-N5AC

The WPM slider adjusts the speed in Words Per Minute of the internal keyer.  Today, this also affects the rise time for both internally and externally keyed CW.  We'll probably move the rise time to a setup panel in the future.

Paul Christensen, W9AC

Adding to Steve's point, if the function of both transceivers' envelope is Gaussian in shape --  and if the rise/fall time is approximately the same -- and if the waveform is not distorted by aggressive ALC circuitry, nor passed through a limited bandwidth network like a crystal filter, shouldn't we expect similar keying bandwidth? 

No doubt there is a striking similarity between the K3 and Flex 6K with the latest Flex firmware/software.  But we should expect it.  In fact, it looked similar enough to me that I decided to exaggerate the keying envelope using a SunSDR2 transceiver and obtained the wide bandwidth mask you see in the image above, just by altering the SunSDR2's ramp time in software.  Why did I use the SunSDR2?  Because it has the flexibility to change the ramp time between 1ms and more than 10 ms.   I applaud their effort, but in the wrong hands, settings below about 4ms start becoming un-neighborly.  A more sane range is between 4-8 ms. 

If you look at changes in bandwidth as a function of SSDR's speed slider, it is minimal because the ramp time change is actually very small and the Gaussian shape is retained.  This is unlike past versions of SSDR where these changes had a more visible effect on displayed bandwidth - not a lot, but more noticeable than in my plots above. 

Kevin, W9CF has conducted a comprehensive Fourier analysis of keying rate and bandwidth.  Charles, this has all the math one can digest!   It's an interesting concept because the keying rate can have a substantial impact on bandwidth but not at normal keying speeds where the ramp time in ms dominates bandwidth.

http://fermi.la.asu.edu/w9cf/articles/click.pdf

Quoting from W9CF:

"The effect of the keying speed on the bandwidth as long as the rise time is small compared to pulse length is the change in shape of the central peak. It does get narrower for slower keying and wider for faster keying, however, the keying speed does not effect the overall bandwidth."

Mark, W8XR also examined keying speed versus bandwidth.  His work agrees with that of W9CF:

http://www.w8ji.com/cw_bandwidth_described.htm

W8XR's key bullet-points:

" If you take nothing else from this article, you should be able to say with confidence “The bandwidth of a typical CW signal depends on the shape of the keying envelope.

The hardness of the keying is greatly influenced by the rise and fall time of the keying envelope. A rise and fall time of 5mS or more will result in a readable signal at reasonable bandwidth.

Sidebands of a CW signal are caused by keying (an unmodulated carrier requires no bandwidth.)

You can estimate the bandwidth required by analyzing the shape of the keying envelope (“removing” the parts that are unchanging and calculating the frequency of the remaining portions.) This applies with “soft” keying.  Using “harder” keying increases the required bandwidth dramatically.

Lots of people like to argue about these points even if they don’t really understand them.  It’s best to just “walk away” from these arguments, unless you just like to argue.  Little exchange of knowledge is likely to occur, even if it is somewhat entertaining,"

Paul, W9AC

Charles - K5UA

Thanks so much Paul. I have a better understanding now about the rate vs bandwidth relationship. My hat is off to the Fourier analysis fellows. That was my last engineering math class in 1969 and I was happy to see that class come to an end. Now I know why some CW signals look like the nuclear explosion at Bikini Atoll on the panadapter.

Bill N5TU

Excellent!  The slower rise times have certainly improved the CW signal already.
Dit-dit!

Charles - K5UA

Steve, would the guys that write the CW generation code ever have time (once WAN is in the record books) to incorporate dot-dash ratio control as found in the Japanese rigs and the Micro-Ham keyers?  It really generates nice sounding CW at 3.6 to 1 ratio as opposed to the classical 3.0 to 1 ratio.

IW7DMH, Enzo

And I would like to see similar tests also on the more humble 6300.
Thank a lot for your work, dear Paul
73'

Steve-N5AC

Yes I think it's likely that we will do both the weighting and adjustable rise time in a future release.  These are both often-requested features.