开云体育

?

I have previously described the impact of passband ripple response around the center of the crystal filter in my Developer Edition and V2 sBitx's (see: /g/BITX20/message/107621 ) with ~8dB difference between upper sideband and lower sideband transmitted power for same modulation. This shows up as low power on CW (versus CWR) and USB modes. A rough "fix" was applied by adding a coefficient of 1.4 to both real and imaginary values of the frequency values of upper sideband transmit computations in the subroutine tx_process(). A few other adjustments to various limits resulted in a reasonably usable transmitter for all modes.

?

Recently, I made some more careful measurements [WITHOUT the correction, so native response]? on my particular (V2 hardware) sBitx frequency response for both sidebands, recording lower sideband frequencies as "negative" and upper sideband as "positive," which matches their difference from the suppressed carrier and allows easy graphing. The results are shown in the figure as blue "before" values, and are graphed as relative dB from a peak response.

?

?

?

The upper sideband response appears to be reasonably modeled by a piece-wise linear response, approximately -4dB below 900 Hz and -8dB above that. I decided to apply a simple piece-wise linear correction for upper sideband below and above 900 Hz.

?

The correction is multiplied by both the real and imaginary values, in a statement like this inside tx_process():

?

for (i = 0; i < 19; i++){
__real__ fft_out[i] = __real__ fft_out[i] * 1.3;
__imag__ fft_out[i] = __imag__ fft_out[i] * 1.3;
}
for (i=19; i< 60; i++ ) {
__real__ fft_out[i] = __real__ fft_out[i] * 1.65;
__imag__ fft_out[i] = __imag__ fft_out[i] * 1.65;

}

Initially, I just looked at the square root of the magnitude of the power differences to calculate voltage correction coefficients, and thus I selected voltage multiplication coefficients that turned out to be much too large. This turned out to be because the total amplification of a coefficient A when applied to both real and imaginary portions of a complex number has the magnitude impact of not just A, but 1.414*A:

?

Magnitude of [ A*(real portion) + A*(imaginary portion) ]

?

= ( A²* (real portion)² + A² * (imaginary portion)² ) ^1/2

?

= (2 A²)^1/2 * (magnitude of original value)

?

= 1.414 * A * (magnitude of original value)

?

?

As a result, I ended up with multiplier of 1.3 for coefficients 0-18 and 1.65 for coefficients 19-65 to make the desired corrections. This ended up with a very good compensation that allows me to have nearly equivalent power on either sideband with CW, voice, or digital signals (supplied through the microphone input from external hardware).

?

Since different instances of the sBitx may have different crystal filter responses, the exact solution for each radio may be different. An automated system could be programmed to allow the sBitx to calibrate itself by programatically creating tones and measuring output powers, and then computing a correction. However, it was simpler for me to do it manually.

With these adjustments, I get very similar power levels on either CW or CWR and when using (externally generated) digital on either upper or lower sideband (e.g. external WINLINK or ALE or FLDGI).? ?Very nice!

Gordon KX4Z

?

I'm not a crystal filter guru.? ?Building a 25kHz wide filter with minimum crystals sorted from a pile of them is quite an achievement, so I'm not throwing stones, just seeing ways we can use computing power to improve hardware performance, the same way the varying power levels with frequency can also be addressed in frequency.

Here is a screen shot of background noise on the 10 meter band when it is getting no "real" signals:? ?You can see the ripples in the 25kHz-wide,? 40MHz filter by the alternating lighter and darker bands in the waterfall:



This may be different on every radio, and it is possible that mine is a little better or a little worse than the population average.? ?No way for me to tell.? You can even see that the upper side of suppressed carrier frequency is a little lower level than the lower side -- this may correlate with the difficulty I had getting USB power to match LSB power.? ?But by measuring this response and applying matched coefficients to the FFT, a filter can be constructed that makes the output "perfect."? ? All I did, was work on the 6kHz right around the center, to get the transmitter response "flatter".? ?The same effort should easily fix the receiver response (which was less important to me, but for INTELLIGIBILITY of upper side band signals, is probably worth working on also!)

AUTOMATED CORRECTIONS
EVENTUALLY, some really bright person will write some software that automatically observes this ripple in the filter response and automatically computes the corrections and stores them so the radio uses them to have a more perfect filter.??

This sort of adjustment is what we did to create our study of mitigation of high frequency hearing loss.? ?

It could easily also allow "equalization" or emphasis of transmitted audio on the radio, or emphases on the receiver response to help hams with hearing difficulties!? ?It is only a matter of time until these things are built, and the sBitx is the perfect radio for this!? ?It could be set up as an external program to set the coefficients, which would then be used from then on, and would replace $$$$ external equalization systems.

73
Gordon KX4Z

Gordon,

from the info I gues You are talking aboput the SBITX and its first (40 MHz) discrete Xtal ladder filter.

Your photos shows me that ghe filter must have a large ripple within the passband. This can be the result of

--wrong matching to the preceeding and following stages

-- wrong or not correct caps between the xtals to ground.

It may be helpfull to analyse the filter with a Nano VNA, hwo it behaves with 50 ohms source and load impedance.

73

Henning Weddig

DK5LV

Henning - Yes, you are correct on all points. At some point I may try to put it on the Siglent ?spectrum analyzer and plot its performance ? That will require disconnecting it from the circuitry however?

first I think I’m going to look at my developers edition more closely and see if it resembles the same ripple.
(when I get back home)


how about yours?

gordon Kx4z?