Re: QMF banks in G.722

On Mar 13, 11:37 pm, "Jessecw" <jessechen...@xxxxxxxxx> wrote:
Thanks, Steve.

Please allow me to specify my simulation here.

signal source --> H0, H1 ---> Downsample by 2
--> Upsample by 2--> G0, G1--> FFT analysis

H0, H1 are analysis filters, and G0, G1 are synthesis filters. In
this
model ADPCM coding is not considered.

Now, the signal source is 3k with 16k Hz sampling rate. But I can see
the output signal has an aliasing frequency 5kHz.

On 3月14日, 上午10时28分, Steve Underwood <ste...@xxxxxxx> wrote:



Jessecw wrote:
Hi all,

     I have studied G.722 recently. And have done a simulation on QMF
banks in G.722.

     I find QMF banks in G.722 are not perfect reconstruction. Am I
right?

Do you know any filter which is perfect? If you want to eliminate 100%
of the low band from the high band, and vice versa, you need a perfect
brick wall filter. Those aren't available at my local DSP store. :-)
What QMF gives you is a reasonable, and well balanced, compromise about
how much of each band spills into the other.

You should find that for any single input tone the sum of the two output
energies will match the input energy. However, around the middle of the
input band, that energy will be split between the two output bands.

The QMF in G.722 isn't bad as a band splitter. If you want to feed 16k
samples/second G.722 audio into an 8k samples/second telephony network
you can simply take the low band from the G.722 stream, and everything
sounds fine.

Regard,
Steve- Hide quoted text -

- Show quoted text -

Some general comments related to QMF decimation/interpolation filter
banks:

An ideal (no numerical accuracy issues) single QMF HP/LP split with
decimation and then reconstruction without any quantization of the two
band signals will remove any aliasing that occured during the
decimation when the signals are recombined (interpolated). With
quanitzation of the two band signals that statement is not true. Note
that the two-band decimate-interpolate pass-through frequency spectrum
is not quite flat, so additional splitting that does not take this
into account correctly will result in aliasing (possibly too small to
care, possibly not) even if the band signals are not quantized. Also
note that the delays for all bands (which may have been split
different numbers of times) must be adjusted to have the same delay
for all of the cancellation to work out.

Dirk

Dirk Bell
DSP Consultant

.

Relevant Pages

  • Re: AM synchrodyne radio with PLLexplained.
    ... but at BC band frequencies maybe ... one of the problems with balanced tube mixers is balancing the ... modulate a 38kHz carrier and the the 38kHz carrier is removed, ... signals with AM added with a carrier. ...
    (rec.audio.tubes)
  • Re: GPS tracking device allowed in commercial flights?
    ... what does this tell you about possible interference of a cell phone? ... > A cell phone is only supposed to emit signals within its band. ... > Well, I think a VHF radio, beit a ham radio or an aircraft radio, are of ...
    (sci.geo.satellite-nav)
  • Tropical Bands from a Tropical Land - Puerto Vallarta, Mexico
    ... to Puerto Vallarta. ... conditions each morning seemed to skip 41 and 49 meters; only a few signals ... printout of the DSWCI's Tropical Band Monitor, ...
    (rec.radio.shortwave)
  • Re: Bit off topic ...
    ... analogue signal going in, and the same seems to be true now of film stock ... Yes, you're right, the plasmas, at least, do consume a lot of power. ... We are heading back for the old days of band III TV in terms of ironwork on ... in order to receive the crap digital signals - which let's ...
    (sci.electronics.repair)
  • Re: Walters and Stanton
    ... When the band is open I wouldn't even need 50W but it is nice to ... The reason for the slope is that all aerials (in the context of ... This is important because getting your signals to the distant station ... angle you can arrange. ...
    (uk.radio.amateur)