Dolby's TrueHD 96K Upsampling To Improve Sound On Blu-Rays 255
Stowie101 writes in with a story about your Blu-ray audio getting better. "The audio on most Blu-ray discs is sampled at 48kHz. Even the original movie tracks are usually only recorded at 48kHz, so once a movie migrates to disc, there isn't much that can be done. Dolby's new system upsamples that audio signal to 96kHz at the master stage prior to the Dolby TrueHD encoding, so you get lossless audio with fewer digital artifacts. The 'fewer digital artifacts' part comes from a feature of Dolby's upsampling process called de-apodizing, which corrects a prevalent digital artifact known as pre-ringing. Pre-ringing is often introduced in the capture and creation process and adds a digital harshness to the audio. The apodizing filter masks the effect of pre-ringing by placing it behind the source tone — the listener can't hear the pre-ringing because it's behind the more prevalent original signal."
You cant hear it anyway. (Score:5, Informative)
44.1khz audio is already transparent to the human ear. Blind studies have been done where a 16 bit 44.1khz ADC-DAC pair was inserted into a high resolution analog audio source. No significant difference was observed.
Don't waste money on the placebo effect.
Re:You cant hear it anyway. (Score:5, Informative)
The hearing limit is actually about 20kHz. You need more than 40kHz sampling if you want to capture a sine wave at 20kHz.
The purpose of capturing at higher than 48kHz is to prevent sounds at frequencies above 20kHz being captured at a too-low sampling frequency, and appearing as audible frequencies. These can be removed by analog filtering, but only about one octave above the cutoff frequency. Analog filters are not ideal brick-wall filters, so 96kHz sampling is useful.
However, once the audio is acquired and digitized, software can provide a true brick-wall digital filter. This is impossible to do in analog hardware. After applying the brick wall filter, it can be sampled down to 48kHz or 44kHz with no loss. So, there is absolutely no reason to put 96kHz on disc.
The article isn't clear whether it's 96kHz on just the master, or the disc also.
Apodizing Filter (Score:5, Informative)
The title is misleading if the actual goal of this is to apply an apodizing filter. I suspect the reason it's called "Advanced 96K Upsampling" is because that's much easier to get people to buy into that than a "Apodizing Filter" sticker.
The article explains how the audible benefit comes from the application of the Meridian apodizing filter, which changes the analog signal reproduced from digital data by reducing the pre-ringing. IIRC the trade-off is that post-ringing increases. The claimed benefit is that since the ringing now occurs after the "real" music of larger amplitude and as a result the ringing is masked or could be considered like an acoustic echo that naturally occurs.
The 96K upsampling is just a side-effect of wanting the extra samples when you are applying the filter.
Here's a decent summary of what is supposed to happen to the analog audio signal as a result of the filter application: Technical analysis of the Meridian Apodizing filter [wordpress.com].
That being said, from what I've read over the past few years I think people are kind of mixed on whether or not the filter makes things better, worse, or just different but not better.
Re:Apodizing Filter (Score:4, Informative)
A fancy name, but it seems to me that this mostly just a DSP textbook minimum phase filter [wikipedia.org]. A minimum phase filter is just a causal filter which has minimal group delay means it can be made to sound more "analog" (since analog filters are usually mostly causal meaning there is no filter contribution from the "future"). This is of course basically tipping the hat to the sound those vinyl/tube-amp purists claim is the "best" sound (not that vinyl/tube-amp purists would actually like this as it is still an soul-less digital approximation, although perhaps listening through oxygen-free copper speaker wire might help ease them to this "approximation")...
I guess having "minimum" in the name isn't a good marketing technique thus "apodizing"...
Re:You cant hear it anyway. (Score:5, Informative)
I'm a sound engineer and you are totally right.
Going back in history. 44.1kHz was chosen because it syncs with PAL video frames, 48kHz syncs with NTSC. If you were doing linear editing, you can dub and cut the audio perfectly to the half frame.
44.1kHz stuck because Umatic, an analogue videotape that you could buy a PCM head as an optional extra, was chosen to create the master copies for CDs to be sent to duplication in to pressed CDs.
Re:You cant hear it anyway. (Score:4, Informative)
Here is a link to the original paper by Dr. Peter Craven where he mathematically proves that an apodizing filter can make audible improvements in sound reproduction.
http://www.aes.org/e-lib/browse.cfm?elib=12992 [aes.org]
Unfortunately it's behind a paywall. but take my word for it's a pretty impressive piece of work.
Most people, whose sensibilities are not trained to the point where they are discriminating enough won't likely notice the difference. However the fact of the matter is the differences are measurable, the principle is based on sound math, and the results are in the perceptible audio spectrum.
Peter Craven made several important contributions to digital recording. He and Michael Gerzon did a lot to push forward the early development of surround sound technology, and made other significant contributions to the process of digital recording. In particular their work on dithering has had a big impact in improving the quality of CD recordings.
http://en.wikipedia.org/wiki/Michael_Gerzon [wikipedia.org]
http://www.aes.org/e-lib/browse.cfm?elib=5872 [aes.org]
http://www.aes.org/e-lib/browse.cfm?elib=6777 [aes.org]
http://www.aes.org/e-lib/browse.cfm?elib=6647 [aes.org]
Re:You cant hear it anyway. (Score:2, Informative)
The Nyquist Shannon theorem makes some assumptions that are not necessarily valid for digital recording of music.
http://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem [wikipedia.org]
"The theorem assumes an idealization of any real-world situation, as it only applies to signals that are sampled for infinite time; any time-limited x(t) cannot be perfectly bandlimited. Perfect reconstruction is mathematically possible for the idealized model but only an approximation for real-world signals and sampling techniques, albeit in practice often a very good one.
The theorem also leads to a formula for reconstruction of the original signal. The constructive proof of the theorem leads to an understanding of the aliasing that can occur when a sampling system does not satisfy the conditions of the theorem."
Re:You cant hear it anyway. (Score:5, Informative)
Actually, polystyrene caps can make a huge difference over electrolytic or tantalum caps in certain parts of some circuits. For example, in condenser microphones, the coupling capacitor between a microphone element and the first FET stage is a critical part of the circuit in which the signal level is very, very weak. Thus even tiny amounts of noise from cheap capacitors can have a significant effect on the final result. A fair number of cheap Chinese microphones sound dramatically better if you replace the cheap dipped tantalum caps they use with a film cap or poly cap.
We're not talking about a small difference here, either. We're talking night and day. A deaf person could just about hear the difference. :-) Replacing just the handful of tantalum capacitors in those microphones can make the difference between a muddy sound with a harsh, brittle top end and a fairly clean, accurate representation of what is being recorded... all for about five bucks and a few minutes of soldering. (Even better, the most important one—the FET coupling cap—is usually direct-wired between the capsule mount and the FET's lead, so you don't have to worry about lifting traces....)
Capacitors within the feedback path of an amplifier circuit can also degrade the sound noticeably. Admittedly, this isn't as much of an issue these days with the rise of modern, chip-based amplifier circuits, but it is still worth keeping in mind, particularly given that most condenser microphones still use transistor-based amplifier circuits.
Just to be clear, though, it doesn't have to be polystyrene film. The difference between a polystyrene cap and a traditional metal (polyester) film cap is negligible compared with the difference between film caps and electrolytic or (*shiver*) tantalum caps. Tantalum caps simply should not be within a city block of any trace that carries an audio signal.... Okay, slight exaggeration, but you get my point.
And, of course, it doesn't make sense to replace every capacitor. If it isn't in the signal path, it usually won't make much difference (though the absence of capacitors in the right places on power supply rails can cause some fun problems), and even if it is, it may or may not make much of a difference, depending on where the capacitor is in the signal path.
Re:Worthless gimmick with no audible benefits (Score:2, Informative)
Mathematically you've got 5 choices when you want to sample and have sound almost all the way up to nyquest:
1 a brick wall, phase linear filter. Mathematically the best, and yes it has pre and post ringing - the less roll off you allow the more ringing
2 you can do less filtering and allow aliasing instead. In that case you'll get a mirror image of the spectrum above the nyquest rate that wont matter much because only dogs and small children can hear that high. And less ringing
3. You can let the treble roll off a bit. In fact 48k sampling rate is more than cd just so that the roll off from 20 to 24 is longer than that from 20 to 22 and you'll get less ringing. A little roll off never killed anyone
4 you can use an old style filter with some phase shift. It just trades off preringing for postringing and delays some frequencies more than
others and is overall less efficient. Frankly the frequencies being discussed are so high no one will notice the delays. In theory you can mess up the imaging and sound a little that way. There's a reason that the industry has preferred linear phase digital filters to older style analog filters, but no doubt in the digital domain you can optimize a filter with phase delays just like you can optimize one without.
5. You can have an adaptive filter that decides between options 1 2 3 and 4 depending on some unimportant critera like masking. It's unimportant because only children can hear high enough to detect even a hint of ringing or aliasing above 20khz and as far as I know they're not the market.
Re:You can prefectly represent anything up to Fs/2 (Score:5, Informative)
Try a high but more audible frequency.
It may be less confusing if I put it this way: If you can't hear a sine wave beyond, say, 20 kHz, then you are not going to be able to tell the difference between a sine wave at 7 kHz and a square wave whose fundamental frequency is 7 kHz. That's because the lowest harmonic in the square-wave signal will be at 21 kHz. Your ears will filter it out, just as the antialiasing filter in the recording system would need to do.
Now, that being said, the argument has been made that intermodulation effects in the human ear can allow us to perceive sounds beyond the usual 20 kHz limit when they mix with each other. To the extent these effects occur when listening to the source material at a given level, you could argue that the ultrasonic parts of a performance should be captured and reproduced along with everything else, and that would require a higher sampling rate.
The showstopper for this argument is that any desirable sonic content resulting from IMD at ultrasonic frequencies could only be reproduced "properly" at a specific volume level, because distortion products by definition are generated by nonlinear processes.
And Harry Nyquist is rolling around in his grave (Score:5, Informative)
Not that this whole thing isn't absurd for the reasons already discussed above, but what no one bloody well seems to understand it that an audio stream is not a godamn bitmap picture. You can't improve audio quality of *audible frequencies* by increasing resolution of the horizontal axis (sampling frequency) beyond a rate which surpasses the Nyquist frequency for human hearing. Assuming a high quality anti-aliasing filter is used and excellent quality recording and playback equipment, audio sampled at 48kHz can be unambiguously represented up to about 24kHz. 96kHz is a waste of bits.
Vertical resolution (# of bits) is the only theoretical way to improve actual audio quality further... and beyond about 16-18 bits, it's also beyond the ability of even the most diehard audiophiles to discern (in properly conducted experiments.)
Re:Worthless gimmick with no audible benefits (Score:4, Informative)
The signal is only "stair stepped" because they chose to graph it that way. The audio signal coming out of the DAC does not look like that. Those stair-steps are happening at frequencies more than half the sampling rate -- they are eliminated from the analog output by a low-pass filter. This is essentially performing this "splining" you are talking about.
Re:And Harry Nyquist is rolling around in his grav (Score:4, Informative)
That said, doubling the sampling rate isn't going to do anything for a digital signal. At best, the new signal will simply play each of the old signal's samples twice.
Actually upsampling can be useful when you apply digital filters. There is no such thing as an ideal filter, so if you modify one frequency band (e.g. in a equalizer) you end up modifying all others. The higher is the sample rate the lower is this sideband interference.
Re:lol wut (Score:4, Informative)
Re:You cant hear it anyway. (Score:3, Informative)
> NTSC is 59.97Hz, not 60Hz
*Color* NTSC is 59.97, but black and white NTSC is an even 60Hz.