Re: QCX-SSB: SSB with your QCX transceiver
toggle quoted messageShow quoted text
This is a bit embarrassing. But I just found out that there was an overflow in the Hilbert transform filter for the louder speech; the upper bits simply did not fit anymore in the 16 bits register. Not that fixing this has resolved all the audible artifacts: https://youtu.be/Q6_BCqBZjZU But anyway thanks for giving a critical remark on the video.
Karlis, thank you very much for posting the link to the PhD thesis, it is pretty much relevant for what we are discussing and trying to achieve here. Very interesting to hear about your simulation finding that due to the rapid phase discontinuities the bandwidth of the phase and amplitude signal are much higher than the actual modulation, and thus should not be band limited otherwise distortion artifact occur. I also think that chapter 2.2 describes the phenomena quite well, and the effect of band limiting (figure 3.8).
I can confirm this finding experimentally; after removing C31 the IMD performance gets hurt when C32 gets too big, it is currently passing about four times the SSB bandwidth. Look at the SI5351 VCXO and Spread Spectrum Modulation Bandwidth, my best guess is that the loop bandwidth of the PLL is something like 20 kHz, which means that the PLL probably will follow quite instant a frequency register change. I do not think there is a need to over-sample the envelope and phase signals (they are currently sampling at the same rate as the ADC input), and already exhibit the huge bandwidth components due to the discontinuities, but maybe I am wrong here.
In the video you can still hear some noise, especially when the speech is "CQ". I guess because of the low sample-rate and the low number of effective bits (8 or 9). I had quite a bit of a challenge with the ATMEGA328P chip to get a noise free ADC conversion, and it seems to me that the first two LSBs of the 10 bits ADC conversion are quite noisy and crackling especially when there is some RF around. These tiny bit changes cause huge phase changes (when the audio is at a low) and are causing terrible noise when the envelope shaping is not done property.
Your suggestion for a C-P approach with 2 class-D amplifiers sounds to me like a great idea and probably resolves the bandwidth issue completely; I am also intrigued by the idea mentioned in chapter 6.1, where (if I understand it correctly) two amplifiers are used with opposite phase and polarity (a bidirectional polar amplifier) driven by a single phase.
On Mon, Apr 29, 2019 at 7:09 PM Kārlis Goba <karlis.goba@...> wrote:
Guido, I was so intrigued by this method of SSB modulation that I made a script to simulate the modulation. Of course, it is an idealisation of real-world components and their speed/linearity, but it gives at least the upper bound of quality for such method. Unfortunately I found that the phase signal exhibits very rapid discontinuities, and as such has a very high bandwidth. Limiting this bandwidth (e.g. by a finite update rate of Si5351 VFO) produces the same kind of artifacts that can be heard in your recording. I found that even 2-3 times the SSB bandwidth could still be insufficient. The envelope signal is somewhat more tame but I guess the quality still suffers a bit from the low-bit quantization in QCX hardware.