I think I need to explain my proposal a little bit further to avoid
misunderstanding. I attach a PDF outlining the present design and my
alternative.
In the present design a BANDPASS filter is used, nominally with a center
around the sidetone frequency of 700 Hz. In my sketch it also includes
the Hilbert transform. I guess that the present filter is of order 10 or
so according to the graphs in the manual.
My proposal is to use ZERO BEAT of the received signal and use a LOWPASS
filter. To get similar performance as above, two filters (for I and Q)
of order 5 or so will be needed. The side tone is mixed after the filter
just in front of the DAC feeding the speaker/phones. An optional offset
(OF) can be added to adjust the center of the CW filter up and down.
This will mean two settings: bandwidth and center. These can be adjusted
independently.
The Goertzel algorithm can be replaced by an moving average to feed the
CW decoder.
Yours is a proposal for a radical redesign of the CW filtering, not just a change to the way the UI works.?
I don't think your proposal would work well, though I find myself lacking a deep enough understanding to explain why clearly. Perhaps someone else could contribute here also. But I'll try.
You say that we can get rid of the Hilbert transform - I don't see how that is the case. You need the Hilbert Transform, for image rejection at the IF.?
But if you consider abolishing the IF completely and doing everything at baseband, then I think what you are really proposing is Weaver method SSB reception - see my old analog page about this here (written 2014). This does indeed avoid the Hilbert Transform (at least explicitly, though the same is effectively achieved mathematically by the second mixing operation at the sub-carrier), and now you implement two low pass filters, each with half the desired bandwidth, and then finally a sub-carrier oscillator and mixer, which in this case would occur at the CW center frequency.?
Weaver demodulation is used in some SDRs. A disadvantage is that there is a hole in the response at the sub-carrier frequency (the CW sidetone frequency, in your nomenclature). In the analog world it is hard to make the hole small but in DSP it could be made a lot smaller, such that it is unobtrusive on an SSB receiver. However for CW, I wonder how that would work... the "hole" would be exactly at the center of the passband, and the CW transmission could well have energy either side of it, and given that the actual bandwidth of a CW transmission?could be only 10 or 20 Hz, even a small "hole" would have a marked effect.?
An additional problem is that you lose the advantage of doing the SDR at a 12 kHz IF. The 12 kHz IF gives immunity to hum and noise which can often be found on a direct conversion receiver near to (or within a few kHz of) 0 Hz. Operating at a 12 kHz IF produces significant advantages. The conversion to 12 kHz IF is quite trivial digitally and computationally inexpensive because the sampling is at 4x this (48 ksps).?
Finally another problem. To obtain a 50 Hz bandwidth CW filter, you would need to have a 25 Hz LPF. To do such a low pass filter in DSP will require a very long (large number of taps) DSP filter. To avoid this, it will be necessary to decimate down to a much lower sample rate. Which is fine. But you cannot avoid the fact that there will be a large delay through the filter. Intuitively and/or approximately speaking the filter must be 40 ms long for a 25 Hz LPF. That is a very large latency for a CW operator and prevents the use of QSK (full break-in). In reality this 40 ms is just the delay of the selectivity filtering, the actual receive latency would be even longer than that because you still have to do the conversion from analog to digital (ADC) and back to analog (DAC), as well as the decimation to lower sample rate which also involves a DSP transform. Then you will want to put in some AGC which is another delay line etc etc.?
In contrast if a filter is made by the superposition of two 500 Hz bandwidth filters, the bandwidth will still be 50 Hz but now the processing delay due to the?bandpass filter is 10x smaller, only 4 ms.?
So for a number of reasons I think that your proposal would work (it is just Weaver method SSB demodulation), but I think for CW it would be an inferior performance solution.
73 Hans G0UPL
