Hi,
On Fri, Apr 8, 2011 at 5:02 AM, Benjamin Larsson <ba...@ludd.ltu.se> wrote:
> On 04/08/2011 03:58 AM, Ronald S. Bultje wrote:
>> 2011/4/7 Benjamin Larsson <ba...@ludd.ltu.se>:
>>> +static void qmf_decompose(DCAContext *c, int32_t in[32], int32_t out[32],
>>> int channel)
>>> +{
>>> + int band, i, j, k;
>>> + int32_t resp;
>>> + int32_t accum[DCA_SUBBANDS_32];
>>> +
>>> + add_new_samples(c, in, DCA_SUBBANDS_32, channel);
>>> +
>>> + /* Calculate the dot product of the signal with the (possibly inverted)
>>> + reference decoder's response to this vector:
>>> + (0.0, 0.0, ..., 0.0, -1.0, 1.0, 0.0, ..., 0.0)
>>> + so that -1.0 cancels 1.0 from the previous step */
>>> +
>>> + memset(accum,0,sizeof(accum));
>>> +
>>> + for (k = 48, j = 0, i = c->start[channel]; i < 512; k++, j++, i++)
>>> + accum[(k & 32) ? (31 - (k & 31)) : (k & 31)] +=
>>> mul32(c->history[channel][i], UnQMF[j]);
>>> + for (i = 0; i < c->start[channel]; k++, j++, i++)
>>> + accum[(k & 32) ? (31 - (k & 31)) : (k & 31)] +=
>>> mul32(c->history[channel][i], UnQMF[j]);
>>> +
>>> + resp = 0;
>>> + /* TODO: implement FFT instead of this naive calculation */
>>> + for (band = 0; band < DCA_SUBBANDS_32; band++) {
>>> + for (j = 0; j < 32; j++)
>>> + resp += mul32(accum[j], band_delta_factor(band, j));
>>> +
>>> + out[band] = (band & 2) ? (-resp) : resp;
>>> + }
>>> +}
>>
>> I'll probably sound very naive, but what does this piece of the code do?
>
> Decomposes the signal into bands. Time domain -> frequency domain.
Ah, that's what the fixme is for I guess. Sounds ok to me then.
(Didn't Mans just implement fixed-point FFT or something like that? Or
was that "just" MDCT?)
Ronald
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
