Thank you again for sharing your work, it has inspired me to work on my own SDR project. A few years ago I got the first generation esp32-a1s development board and have used it for SDR experiments. The older audio codec on it is a pain to work with and I should order the newer modules. More recently I've been working on an pure STM32F103 sdr, it has dual adc and I'm not using any codec, rx audio is PWM. I got very good results with the bluepill dev board, feeding the quadrature af from an usdx board to it. I then designed my own standalone board and it doesn't work correctly, sideband supression is horrible. I have pinpointed the problem to the 3253 detector and 5351 pll, since the DSP works fine injecting the af quadrature signals (I often use kiwi sdr in iq mode). My 5351 channel amplitudes are not even. I feel that the fst3253 performance varies batch to batch and it can be very touchy to get to perform correctly, of course circuit layout plays a big role. I'm making a new design with a johnson counter for the quadrature lo and either a 4066 detector or sa612 pair, hoping to improve the situation. I plan to feed the 5351 clock from the MCO pin on the STM32 to reduce artifacts by running both ics on 16 MHz. I think the best part of this experimentation is that there is constantly room for improvement, for example there still isn't a silver bullet for the quadrature detector chain in these hf dsp rigs. I have to also mention that I'm a recovering cheap component addict and sometimes those parts end up on my pcbs, making my troubleshooting even more painful :D Here's a link to my SDR project: gitlab.com/m0rph/pocket-rocket-sdr . I need to update the git page with my recent discoveries.
Thank you for the great writeup and I am glad the videos have inspired you. I moved away from STMs a little while back because I was intrigued by some of the onboard peripherals on the ESP32 (wifi and such). I did prefer the STM development environment though
Not sure it has anything to do with your problem but... why is the yellow trace "fatter" than the purple @ 14:00? Noise? Probe picking up noise? Just curious. Thank you for this video series, I'm enjoying it very much and learning from your experiences. K7SWC
I have been noticing that myself Steve - as you might have seen I sent that scope off for a service a little while back. I am wondering if there are some lingering problems...
Just a guess but I think it could be something to do with the SI5351 clock generator generating noise, what is voltage are you are supplying it? Also I have found filtering the clock outputs really cleans up the waveforms. As I say it's just a guess as I am sure that you will get to the bottom of the issue. Another interesting video.
Thank you John - I was wondering that. The si5351 is very close to the Audio switch. I should really have placed it elsewhere on the board. Its supplied with 3.3V
First of all, I admire your work on this project. I have another suspicious thing as the reason for the described problem. Although you only turned on linein in the software, the microphonein inputs are not turned off. The ESP32-A1 now on the header board has longer lines on the mic inputs and this can result in additional signals to be trapped on those lines. Try cutting the mic lines right at the module. Good luck and Best Regards from S52UV op. Stojan
@@na5y Just one more option. While testing the ESP32-A1S module, I realized that it is very sensitive to ground traces also. The noise was significantly improved if I grounded module in only one place (one pin on module) and connected this and ground pins of all other circuits (Si5351, Opamp supply AF output amp...) to one and only ground point in the radio. This point was the output terminal of the power supply for the entire radio. So a correct PCB layout is sometimes the biggest challenge. I really wish you a success. Stojan
@@stojanmarkic1033 Grounding has definitely been a problem for me in the past with these ESP32-A1S boards so you might be onto something. Its strange that the artifacts are so clean though
Wow - that's super observant Martin. Yes I tried powering the opamps from 5V as I really didn't need a 12V swing. Mid rail is at 2.5V and the signal shouldn't get anywhere near that. I ended up suspecting it as a problem and changed it back to 12V - unfortunately it didn't make things any better. Thank you for the comment!
I also build several Tayloe IQ mixers and the result is very depending on the phase and gain match. 0.2 - 0.5 db gain and 1 degree phase difference is enough to reduce it to 30 db. Maybe adding a pot trimmer on one of the opamps to adjust the gain a little bit is worth a try. Or compensate in software is also a possibility.
Thanks Paul - I actually do have amplitude compensation in software controlled by the pot on the right. That strange signal at +8KHz is something else though I think. I am going to have to think about this a bit more...
@@na5y An other problem I had was groundloop, noise from the CPU injected into the ad converter. In that case the 8khz signal should be there without rf input
You mean does the output after the splitter look good (equal amplitude and 180 phase) - I am pretty sure I checked - but will definitely check again. Thank you for the suggestion!
I was not clear I understood you used a trifilar rather than a bifilar and then a seperate winding on the input side, if the turns ratios are different you may have suboptimal impedance. Like I said clutching at straws. One more thing do you have a splatter of 0.1uF caps across the DC near all of the chips?
@@chrisherd991 I do have some .1uF caps sprinkled here and there on the supply of the ICs. On the splitter I ended up going with trifilar rather than the bifilar + single winding. That is exactly the same as the other radio I showed
Ok I have been looking at other circuits that use the same technique as you.They use a 2.2uF cap between the center point of the voltage dividers and ground. If you have done that I will keep pondering.
I'd first try another brand of 3253. Perhaps swap the2 you have in the better performing rig and the new one. They appear to be 'fussy' parts when used in this application. Meaning some chips are simply better than others by a few dB. You might also look carefully at the drive signals toggling the 3253. Slightly slower rise or fall times could make a difference between the 3253 outputs of the 2 boards. Second, I'd try a magnetic isolation divider plate of some sort between those 2 toroids that are in line and so close together, (the splitter toroid and the toroid in the bandpass filter.)
