Meridian Unplugged - Powered by MrTechGuy.co.uk

Careful VK ....some of the folks on the slim forum can get quite upset if you try and "discuss" with them. I recall George had a torrid time from a few of them when he was pushing the "don't do volume control digitally" story.

Looking at the replies to VK's linked post, I think it's only a matter of time...

I didn't read the thread, but I was always curious what dither is required going from 16 to 24bit. It makes sense going from 24 to 16bit. But if you have iterpolated the new low 8 bits adding dither say to the low 2-3 bits is not really audible and I doubt Meridian does dither any more significant than 2-3 bits (out the 24bits).

I'm beginning to suspect that the dither applied when going from 16 *to* 24 bits is purely for resolution enhancement purposes (as per the Wadia paper that I've linked to a lot.

When going from 16 to 24 bits, sox just adds zeros to the 8 LSBs which may not "technically" be altering the sample (i.e. it's a trivial operation which does not modify the sample per se).

On the other hand, re/up-sampling most certainly alters the samples so dither is going to be, IMO, critical to maintaining linearity.

Didn't Meridian use the image of the dog to explain dither at HH2? Or was the dog used for something else? Its been a (too) long time. If it was the dog, I thought at the time that it explained dither perfectly, not that I can remember it now!

No, the dog (a dalmatian) was an illustration of the memory effect: once you know something is there, you can always see it (or, more pertinently, hear it).

Unfortunately, the 518 and its manual pre-date upsampling. I'd love a Bob Stuart white paper on recent digital audio innovations.

But they did come up with a very good explanation of dither didn't they? I recall coming away with a good understanding of what dither (in audio) was all about - now forgotten.

They used the sound of a gong to demo dither IIRC. The tail of the ringing went on for quiteb a while when dithered. Or am I now seeing a poodle not a dalmation ?

Yes, the gong demo was about dither and maintaining infinite resolution. I'd forgotten about that.

Right! Here is a 4 millisecond mono 16/44.1 WAV file with a single maximum sample at 1.5 milliseconds. Use it with your preferred sox parameters and the output is the impulse response. And here's what the impulse looks like:




Here are the impulse responses for three different sox command-lines. I've used sox's version of the 518 "High pass" TPDF dither for all three as, unfortunately, the "modified-e-weighted" dither, which looks very much like the 518 "Shape A", isn't available at the 88.2kHz sample rate. This isn't a big deal though.


1) First off, this is the "default" very high quality minimum phase upsampling with a 20.95kHz passband and aliasing/imaging allowed above the passband (rate -v -M -a 88200 dither -S). There's about 2 milliseconds of visible ringing.




2) Next is a more relaxed minimum phase filter with a reduced passband of 20kHz and *without* aliasing (rate -v -M -b 90.7 88200 dither -S). There's about 1.75 milliseconds of visible ringing. Clearly, reducing the passband and not allowing aliasing have cancelled each other out in terms of the amount of post-ringing.




3) Last, but not least is the more relaxed minimum phase filter from (2), but *with* aliasing/imaging allowed above the passband (rate -v -M -a -b 90.7 88200 dither -S). There's just 1 millisecond of visible ringing.




For comparison, here is the impulse response of the 808.2, as published by Stereophile, with the same overall horizontal scale (now you understand why I chose a 4ms sample with an impulse at 1.5ms!):




To my eyes, the last of the three impulse responses that I generated looks very similar to that of the 808.2. This wasn't entirely a fluke as I remembered that, in his AES talk, Bob Stuart showed an apodizing filter frequency response curve that seemed to be quite gentle.

One final point: Impulse response is not the whole story. Although sox is generating minimum phase filters, I don't believe these are apodizing. Meridian's apodizing filters are shaped (frequency response) to have a null at the ringing frequency and thus remove ringing even from existing recordings. It might be possible to achieve something similar using sox's FIR filter engine, but first I need to have a good listen, and to number (3) in particular (when my ears stop ringing - pun intended!).


Edit: I did a very rough measurement of the ringing frequency for (3) and it's 18.75kHz. That seems "low" as theoretically a brick wall filter would produce ringing at the Nyquist frequency (22.05kHz in the case of 44.1kHz audio). As John Atkinson mentioned in Stereophile in the 808.2 review that the problem with linear phase filters is that they introduce pre-ringing which the brain can detect, even if you can't actually "hear" it. Given that most of us will be deaf at 18.75kHz, I think that post-ringing at this frequency probably isn't a problem. I'm pretty sure that lowering the bandpass of the filter also lowers the ringing frequency. Everything is a compromise; hence the need for listening tests.

Awesome work. Keep it going !!!

+1 on that V.K.

Regards,
Carl