[hpsdr] Mercury sampling rate

[Alberto I2PHD]

Alex, VE3NEA wrote:

> Alberto,
> 
> The clock generator in the soundcards is crystal controlled and typically 
> has an accuracy of 10^-4 or better. Consider a program that inputs audio 
> data at a rate of 48 kHz in blocks of 1K samples, and suppose that the 
> output rate is lower than the input rate by 10^-4. When the input device 
> produces 10,000 blocks, the output device will consume only 9,999 blocks, so 
> one extra block is produced every 200 seconds. This results in one crackle 
> in the output audio every 3 minutes or so - which is totally acceptable and 
> is very difficult to detect by the ear in the background noise. Note that 
> the crackle occurs only in the data sent to the speakers but not in the data 
> used to compute the spectra, decode digital signals, etc.

  Alex,

    I agree of course on the numerical part of your analysis, but not on the conclusions.
Suppose that the output rate is 11025 Hz, a commonly used sampling rate for output. If you do the processing in blocks 
of 1k samples, then each block correspond to a duration of about 92.9 ms. If you receive CW at a speed of 30 wpm, the 
dot length is about 40 ms. So, when you have to throw away one block, because you are in a condition of buffer overflow, 
you throw away more than two elements of the incoming CW signal. Not nice.

The same for buffer starvation/underflow. In such a circumstance you have to send to the sound card either a replica of 
the last buffer sent, or a buffer of silence. In both cases the effects are very negative for CW. I agree that for SSB 
reception, especially in conditions of high QRM, maybe the fact can go unnoticed, but not for CW.

So I don't see alternatives to a continuously adjustable fractional resampler, if what is sought is the best audio 
quality. But I would like to be wrong, as this choice as other negative implications, which, if possible, I would like 
to avoid altogether.

73  Alberto  I2PHD


 1208190620.0