A New Line of Transceivers ~ DifX
Transceiver Architecture 2.20
6/10/2017~ More on Dishal Dyslexia Part 2
There always has to be a reason why something as fundamental as 1 + 1 does not equal 2 and so it is with the results arising from building a Dishal Crystal Filter. So lets us take a few steps back and examine why 1 + 1 may not equal 2.
The Dishal software in large part requires the user to make some initial measurements outside of the software itself. Starting with the G3URR oscillator and my SDR receiver (capable of measuring to 1 Hz and verified by first locking on to WWV) I measured the crystals in the loaded and unloaded condition. Again the receiver is calibrated to be accurate to 1 Hz and the G3URR component values are 470 PF (NPO COG) in the Colpitts circuit and the load cap was 30 PF NPO COG. So the data was taken and crystals were found that were very close in frequency and thus were the basis for entering the data into the Dishal software. It is pretty automatic once you enter the data.
The Dishal software can be tweaked by factors such as the number of crystals and the ripple factor. Once these are selected then you get a plot of the filter. But the software also spits out the Filter Center Frequency, the values of the coupling capacitors and the impedance that must be matched to 50 Ohms. So far so good.
Now the problem is one of what BFO frequencies do you select. Starting with the predicted Center Frequency I chose BFO frequencies about 1.5 KHz above and below that value. This choice was one based on the stock of 9 MHz crystal Filters I have and the matching BFO crystals are just that -- 1.5 KHz above and below. That isn't gospel but merely a place to start. Those values were input to the Arduino Sketch. It became pretty obvious that those were not the right values.
My next step was to disconnect CLK2 from the SBL-1 and in its place use my FeelTech signal generator to supply the BFO signals. Two cautions here with the first being that you crank down the output to 1.4 Volts PTP before connecting to the SBL-1 and second that you have a 10 NF cap between the FeelTech and the SBL-1. What I found was that the values that seemed to work were only about 1 KHz away from the Cf. There is something wrong here.
A 1 KHz BFO would place the signal on the top of the pass band and not on the slope of the curve! Thus 1 + 1 is not 2. This bothered me and that is when the light bulb went on. Dishal is lying to me!!!! Well not entirely true but close to the truth (I must sound like Donald Trump.) Is this being taped?
This is when I tried something and this may provide an explanation for some of the crappy results I am seeing using Dishal. The first thing I did is to connect the radio to a dummy load so that no signals were incoming. Then I listened to the rig as I tuned the BFO (FeelTech) from about 2.5 kHz below the supposed Cf to about 2.5 kHz above and just listened -- I noted the change in the background noise and at one point heard a "null". Was this in fact the Center Frequency? I think it may be --tell me I am wrong. This value was close but different that the Dishal value. So I think my evaluation may have some validity. Using this value I added/subtracted 1300 Hz to set the frequencies. In looking at the location of +/- 1300 Hz this puts the frequencies about 3 dB down on the curve and so that may add some additional validity to my selection.
I have changed the Arduino sketch with the 1300 Hz values and will now await the 40M band to perk up so I can run some listening tests and get some reports on the transmitted signal. I very likely missed this part in the Dishal Tutorial (who reads that stuff anyway) but the setting of the BFO frequencies does indeed affect the filter performance.
Lurking here is the theoretical (Dishal Predictive Curve) versus what really results and that may be a shift of the Center Frequency because of imperfect components, stray capacitance, lead length and whether you have dandruff or AFF (Athlete's Foot Fungus). So many things could impact if indeed 1 + 1 = 2.
My Dishal experience has consumed way too much time and the results so far do not justify the effort. Despite what you may have read in all of the testimonials and the touted results --did those users actually install the filter in a radio and "air test it" and match their real world experience with what the curves may be telling them? I can only share what I have seen and so far it has not been pretty.
I am disappointed with the results from using the Dishal method where quite often what you get does not match the predictions or expectations. There are those reading this blog who are now quickly preparing a flaming response. There will be the usual comments: you didn't do it properly; or you did not make the measurements correctly or you built it in a haphazard manner or on and on.
My 6 pole filter sucks and is real crap. I then decided to build a four pole filter using four different crystals but ones from the same batch. I was very careful (again) to follow all of the steps and to double check the build, the test setup and the measurement process. The outcome was not unlike the six pole but somewhat better. I did find that the skirt rejection was better than the six pole and closer to the predicted curve but installed in a radio sounded "not so good" when tuning across a signal. That to me is telling --how does it work in a rig.
So below are plots of the four pole filter responses both from the predicted value and the actual measured value. Using the FeelTech signal generator for the signal source and my Rigol scope for the detector. There was a 50 Ohm series resistor in the source end and a 50 Ohm termination resistor on the output side. The readings were taken as RMS readings.
Here is how I converted the Vrms data. First I multiplied the readings by 2.828 as this converts the readings to V Peak to Peak. I thought this would be useful to know and have as a matter of record the PTP readings. I then squared that PTP reading and divided that by 400. The 400 factor accounts for 50 Ohms and does in effect take the value back into RMS. The answer is in watts. If you multiply that by 1000 you now have milliwatts. Or you can simply take the squared value and multiply by 2.5 (1000/400). This answer is now in milliwatts. If you take 10 X LOG10 of this number you will have dBm. That is what I did. So find fault as you will; but the curve will represent the relative impact of the pass band and give you a prediction.
Here is my curve based on the measurement process. I took readings every 200 Hz. There might be some advantage at this point to repeat the test using 100 Hz data points. But this curve does look like the Dishal curve for the four pole filter. Installed in a radio I can say that on receive it does not sound that good.
The Dishal Predicted Curve!
So what do you do now? You have the filter, it looks like the prediction but does not sound too good. There are opportunities for changing things. One change: make the bandwidth less. I set it at 2.5 kHz so it would have that "hi fi" or enhanced SSB sound. A filter at 2.3 KHz night be a better choice, and doing so will change the cap values and the matching transformer. So before I do that I want to do some more testing with the current configuration.
This now begs the questions: Why did the 6 pole shoot craps and Why the 4 Pole while looking like the prediction not sound so good? Maybe the sounding not so good is that the bandwidth has been opened up and I am used to hearing like about 2.1 kHz. Stay tuned.