"After the 32-band PQMF filter, blocks of 36 sub-band samples are overlapped by 50%, multiplied by a sine window and then processed by the MDCT transform...."

I read this from a book, it is about Layer III Hybrid Filter Bank. My query is, how can we ensure there are 36 sub-band samples in each of the 32-band? Aren't the number of samples proportional to the length of the data?

At first step audio data is divided in frames of exactly 1152 samples each. This gives us 1152 / 32 = 36 samples is each subband after PQF.

-Eugene

No.. The audio data is divided into frames of 576 samples.. So 576 / 32 gives you 18 samples per subband.. You have to combine the 18 samples of the previous frame to get 32 samples per subband for the MDCT calculation.