[hpsdr] FFT latency

[G3XJP]

Great.  Many thanks all for that collective wisdom.  All understood and 
all internalised. 

I only operate SSB with full conversational VOX using my h/b PICaSTAR.  
Usually in QSOs where nobody says more than a few words in one go.  
Often with interruptions - which give a "double" of no more than one 
word.  Exactly as per a conversation in the local (British) pub.  Other 
PICaSTAR users run full break-in CW - and some just have long overs.

All in the time domain, it has very small latency.  But I'm just 
finishing off some serious Noise Reduction capability in the freq domain 
and even with 40Hz bins, I'm worried it will cause exactly the problem I 
asked about.   You have confirmed my worst fears - namely that the 
delays can be minimised but are inherent. 

Very easily measured. I connected up my CW ditter to Rx ANT and put one 
'scope lead on those 20wpm dots. The other 'scope lead on the Rx audio 
out. The difference between them is 23 millisecs. That is longer than 
you might dream of - but exactly on the edge of acceptable.   The 
redeeming feature is that the instant you go from Tx to Rx you do hear 
some residual of the Rx audio that was there in the pipeline at the 
moment when you went to Tx. With little experience, you just KNOW you 
were briefly doubling.

But you have to conclude that more narrower bins - giving better 
resolution for noise and tone removal - would be operationally 
unacceptable.  My last vestigial thought is that it might be possible to 
dynamically change the bin width at the moment of T/R or even bypass Mr 
Fourier briefly.  But there are a finite number of MIPs in the real-time 
path even with an ADSP 2181 and either of those approaches feel as if 
they would be both tricky and clicky.  As ever, the pleasure is in the 
trying.

cheers and many thanks,  Peter G3XJP

 1428307505.0