Hi STeve
For your 8000 to 7500 conversion (and I wonder WHY the hell 7500 was
chosen and not some sub multiple of
8000/11025,16000,22050,32000,44100,48000 etc) ? there must be a good
reason, David understands the practicalities.
I'd change the FDV freq to 8000.... (or 6000 if it will fit, 6000 is
anothe rrate easily acheived)
However, if 7500 must be used, suggest something a bit different :
The result does not need to be the performance of a Hi-Fi audio sample
rate converter. aliasing artifacts and production does not need to be
down more than maybe 10dB above the SNR requirements per bin.
Use a FARROW FILTER. basically, the interpolation ocefficients are
calculated on a sample by sample basis.
But the passband cannot be too much of the nyquist to be easily realisable,
8000>>16000 with a half band FIR filter. (a few taps) . then use a
FARROW to go to 15000, then do a 2:1 downsample with (the same) halfband
filter). The farrow though is more commonly used to sample rate match
two ends that have very slightly different sample rates . like say a
transmitter with a fix sample rate trying to match perfectly the source
information rate.
if the farrow is too heavy to stomach :
8000>16000 with a half band FIR interpolator,
then use a CIC (just adds , no MULS) up to 240000,
and then another CIC down from 240000 to 15000
then a half band FIR downsampler.
the reason for the 2:1 with the FIRs each end is that the CICs will tend
to introduced some drrop into the top of the passband and if the nyquist
is lifted the droop wont be as troublesome (if at all for this
application). this way you get to use a reduced number of CIC stages to
do the job , and only have a cheap FIR at each end .. and maybe I dunno
11 taps would do it.
or just change the system to use 8k, not 7.5k
On 10/12/2018 2:59 AM, Steve wrote:
A couple investigations. First I graphed the current FIR filter used
in Coherent FDM, but since I have no idea where these coefficients
came from, I quickly threw those out and used TFilter to give me some
new ones. The results were pretty much the same, only now I had total
control. Basically the FIR filter is a Low Pass filter that passes
everything up to 2700 Hz, and then rolls-off to a stopband of 3730 Hz.
The design is for 100 dB
_______________________________________________
Freetel-codec2 mailing list
Freetel-codec2@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/freetel-codec2
