[Elecraft] K3 pitch change with filter selection

[Brian Alsop]

Guys,

There has been quite a bit of discussion about use of very wide filters 
for HF narrow signal mode operation here recently.  If QRM gets to be a 
problem the fallback position is to narrow the filter.

Keep in mind the following.  The AF pitch you will get for a fixed 
frequency signal is a function of the filter.  The amount of pitch shift 
between filters can be 3 Hz at times.  How the decoder handles such a 
step function change in pitch, I don't know.

Here is some data (CW mode, pitch =500):

filter width    OFFSET      IF FREQ     pitch change
400 Hz 		0	 8.215 MHz	defined to be zero
200             -.83     8.214170          -3.6 Hz
250             -.02     8.214980          -2.5
1800              0      8.214450          -0.9
2.7             -.85     8.213200          -1.2

The IF frequencies are nominal. They reflect the filter offset and other 
considerations. In reality, actual frequency produced by the synthesizer 
board and its AD9834 synthesizer IC and may differ.  One reason may be 
the programming 32 bit word is inaccurate.  I ran into this with another 
synthesizer where it was impossible to calculate an precise 32 bit 
programming word with 32 bit IEEE floating point math.  One gets about 
25-26 bits of accuracy in the floating point operations results. There 
are certain lucky numbers which produce exact results. Like here, the 
observed differences were in the 0-5 Hz range.  In that case, I was able 
to augment the accuracy through error look up tables and reduce the 
error to under 0.5 Hz.  The math coprocessor used did not permit 64 bit 
math.  Numerical studies showed that 64 bit math would have completely 
solved the problem.

To see if the errors were similar to the numerical experiments, I used 
the 250 Hz filter and varied the filter offset between -0.050 and +0.050 
in 0.01 steps.  The result was a sawtooth pitch error curve ranging from 
-2.5 Hz to zero. (Numerical studies showed a sawtooth error form). There 
are also certain "magic" numbers which produce zero error. Based on 
this, I'm guessing the K3 math used for the programming word calculation 
is 32 bit floating point.  The resultant synthesizer 32 bit programming 
word error is most likely producing the observed effect.

The problem is the amount of the error is a function of your filter 
offset. One has no idea how much difference there may be as filters are 
switched.  If this is a problem for JT9 users, there is a solution for 
the wider filters-- simply tweak the offset in 0.01 steps till the pitch 
difference between filters disappears.

I then looked at the linearity of the 1Hz tuning steps suspecting the 
same sawtooth effect to show up.  It does.  2-3 Hz departures from 
linearity exist.

I suspect the small band to band differences in accuracy (calibrated at 
one frequency and off at another) are also due in part to synthesizer 
output frequencies.

All of this is irrelevant to most users.  I'm not sure if JT9 users are 
impacted or not.

How many other amateur rigs can say they do this well?  I am always 
amazed that the almost lab-grade performance of the K3.

Keep in mind of don't have super lab grade equipment here.  However the 
HP8675B generator is locked to Rb and has a very clean stable signal. 
SpectrumLab, as employed, is able to resolved pitch differences of 
<<.1Hz.  The K3 is using the high stability oscillator.  YMMV

I don't claim to be an expert in all of this stuff.  If the conclusions 
are off base, please correct me.

73 de Brian/K3KO





-----
No virus found in this message.
Checked by AVG - www.avg.com
Version: 2012.0.2242 / Virus Database: 3222/6235 - Release Date: 10/09/13