From Newsgroup: comp.dsp

Am 07.08.19 um 14:18 schrieb Richard:

I am processing an audio signal with CS4244 codec at 24 bits and 48kHz. Every 2048 samples the FFT is calculated and its amplitude.
Without input signal I get FFT values rCirCiclose to 30dB. With maximum input values rCirCi(overflow) I get values rCirCiof 90dB, with a scale of 0 to 120dB.
Why not get a dynamic range of 100-110dB as is logical in a 24bit process?

Calculate the RMS energy from the raw ADC output and do the same for the
FFT of the signal. Normally they should be nearly equal (as long as your
FFT is normalized).
If the energy significantly increases by the FFT you have calculation
problems in the FFT algorithm. E.g. quantisation noise. The latter
happens if the FFT uses no more bits than 24 bits internally.
If the RMS level of the ADC output is already that high you have
problems in the analog part.


Marcel
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On 08/07/2019 02:18 PM, Richard wrote:

Hello,
I am processing an audio signal with CS4244 codec at 24 bits and 48kHz. Every 2048 samples the FFT is calculated and its amplitude.
Without input signal I get FFT values rCirCiclose to 30dB. With maximum input values rCirCi(overflow) I get values rCirCiof 90dB, with a scale of 0 to 120dB.
Why not get a dynamic range of 100-110dB as is logical in a 24bit process? Any ideas?
Thanks.

maybe you are missing a reconstruction filter somewhere...
--- Synchronet 3.22a-Linux NewsLink 1.2

From Newsgroup: comp.dsp

On Wednesday, August 7, 2019 at 10:31:25 PM UTC+8, Tauno Voipio wrote:

The number of codec bits says nothing of the purity of
the analog audio signal coming to your codec.

Are you sure that the no-signal is quiet enough?

If you are talking about audio (20 kHz) then you will only have 90 dB SNR
in the recording. If you go to seismic (128 Hz filter) then you would not
get 120 dB, due to various noise from transducer and pre-amp.
--- Synchronet 3.22a-Linux NewsLink 1.2