Cnf

[Craig Maier]

Hi,

I have found that it largely depends on the nature of the music as to what value ultimately works the best. But, I too start off with 2048 and then experiment up or down from there.

Craig

[ric29]

Usualy for typical music, as a starting point I use 2048. Is it ok or better to start from bigger value?

[Craig Maier]

Hi Ric,

No - downsampling to 22.05 kHz will cut the top end off and you will only have a frequency response to ~ 11 kHz. Not a good idea. To imrove selectivity, consider raising the FFT Size.

Craig

[ric29]

Hi -

When one uses a higher sample rate with the CNF, the frequency bin width is raised. So, for example, if you are using 1024 ffts at 44.1 kHz, the bin width will be ~ 86 hz. That same fft size at 96 kHz will become ~188 Hz. So, put another way, the selectivity of the CNF is reduced with higher sample rates at a given fft size.

So it implies another question-isn't better to make downsampling to 22 kHz?

[Dan McDonald]

Great thanks Craig.

[Craig Maier]

Hi,

Bin (frequency) Width will be:

F(bw) = (Sample Rate) / (FFT Size / 2)

where F(bw) is in Hertz

&

where Sample Rate is given in Samples per second.

Craig

[Dan McDonald]

That's clear, Craig and Marc. So I guess if you are working with a 96kHz file and you're not ready to convert to 16 bits yet, you could raise the fftsize to compensate. Is there a basic formula for estimating how the sample rate and fft size change the bin size?

[Craig Maier]

Hi -

When one uses a higher sample rate with the CNF, the frequency bin width is raised. So, for example, if you are using 1024 ffts at 44.1 kHz, the bin width will be ~ 86 hz. That same fft size at 96 kHz will become ~188 Hz. So, put another way, the selectivity of the CNF is reduced with higher sample rates at a given fft size. In audible terms, it may have a harder time discriminating adjacent noise frequencies from one another, but will have an improved time domain transient response. So, I guess one should consider raising the fft size for 96 kHz sampled files to get similar results compared to 44.1 kHz sampled files.

I do not know if that answer is clear - - - probably not. Let me know if not and I will try to clarify it some more.

Craig

[Marc Hildebrant]

Dan,

I asked a similar question recently, but I didn't ask it in a very clear manor. The post was on 5/15/12.

The answer that I received was that the CNF operation should be used at 44.1 kHz.

My limited testing showed that the 96 setting didn't work as well as 44.1

Marc

[Dan McDonald]

Just checking in the 'collective knowledge' here. As far as we know, there's no advantage to higher sampling rates when using the CNF is there? I know it helps with impulse noise, but I wondered about the CNF.

Dan