When I have a picture in JPEG, I decode it to raw, and then reencode it to JPEG using the same settings, the image quality should not decrease (by my understanding anyway). The high frequency content is already lost, so there is nothing more to be lost. However, if I modify the raw version by displacing all pixels a distance that’s not a multiple of 8, there should be a quality loss even with the same settings, because we are distributing the quantization noice to neighboring blocks. Now here’s my question: Does the same happen with mp3? I think I remember mp3 also being a DCT-based technique. So reencoding with the same settings should work (I think… unless it’s VBR maybe?). What happens when I shift all the samples back in time?