Re: [PIC] FSK demodulation

On Thu, July 24, 2025 5:07 pm, Dave Tweed wrote:
> Harold --
>
>
>> I've been playing with AFSK demodulation (project at
>> https://w6iwi.org/rtty/DspTU3/ ). I'm currently using biquad bandpass
>> filters for mark and space, then comparing the level out of each to
>> determine whether I am receiving mark or space. I also adjust the
>> mark/space threshold based on the last received mark or space level to
>> allow for selective fading.
>>
>> I've also heard of using the Goertzel algorithm to detect the mark and
>> space tones.
>>
>> Is there an advantage to one or the other? The biquad filters seem
>> pretty simple and operate on a sample by sample basis (instead of a
>> block of samples).
>
> Goertzel is basically a one-channel DFT converted to recursive form,
> which makes it pretty much equivalent to any IIR filter as far as
> implementation. It will generally be significantly more selective (narrow
> passband) compared to the biquads, which may or may not be an advantage in
> your application.
>
> But to avoid the limitation of working only on blocks of samples, there's
>  no reason not to do the sliding-window DFT calculation directly, which
> gives you a result on every sample. If you only care about two or three
> frequencies, it's quite doable.
>
> I was messing around with this stuff myself some 30 years ago, trying to
> decode the time signals from radio station CHU in Canada (their equivalent
>  of WWV). One element of their signal is a Bell-103 compatible burst of
> data that I wanted to capture. I wrote up some of my efforts for Circuit
> Cellar -- issues 83 and 84, June and July 1997. I was using an ADSP-2101
> evaluation board and was focusing mainly on FIR-based techniques.
>
> -- Dave Tweed

Thanks! I'm working with 170 Hz and 850 Hz shift at 45 bps. I'm using a
pair of cascaded biquads for mark and another for space. The bandwidth of
each is 1.7 times the baud rate which results in close to 0 dB attenuation
at the tone frequency +/- BR/2. The result is an eye diagram that goes to
full amplitude on alternating bits. A narrower bandwidth attenuates that
sideband resulting in the eye not going to full amplitude.

Harold
https://w6iwi.org


-- 
Not sent from an iPhone.