Cramped QuartersThis month, I'm forgoing speculative fiction to focus on some practicalities. Let's look at the numbers associated with a product's specifications and 3/01/2002 7:00 AM Eastern
This month, I'm forgoing speculative fiction to focus on some practicalities. Let's look at the numbers associated with a product's specifications and performance, also known as “speeds and feeds.” Specifically, we'll examine data rate vs. word length, how much space high-density audio really consumes and why uncompressed audio just won't fit down a typical ADSL pipe.
Back in the first months of 1998, I worked up some numbers for “high-density” audio production. That spreadsheet has evolved over time, but it pales in comparison to Bob Stuart's MLP worksheet (www.meridian-audio.com/mlp/t_mlp.htm). Along with Bruce Nazarian's DVD-V bit-budget spread (found at www.recipe4dvd.com), I can get a sense of what's possible — or impossible — in terms of estimating data availability when planning most any CD or DVD title.
Why should you give a rodent's posterior about all this? Because, as my editor said, “we live in a hybrid world,” where consumers have widely varying acquisition and playback capabilities. If you want your client's or your stuff to be available to the biggest market, then you need to consider storage and distribution requirements for that data.
To start with a simple case, take the CD format: 44.1 kHz times 16 bits times 2 channels results in a data rate of 1.4112 Mbit/second (Mbps) or 0.1764 MBytes/second. A full 60 minutes of that stuff requires 635.04 MB of raw storage space, excluding overhead. Nothing too scary there.
Now let's take a look at some slightly more involved numbers:
The table on the following page represents raw storage requirements for 20-bit audio, in Megabytes, again excluding overhead. A 20-bit word is a good compromise for music distribution because redithering a 24-bit final signal down to 20 bits saves a bunch of space — 17% to be exact. Not to mention the data-rate reduction, which I'll get to in a minute. Playback at 20 bits also preserves sufficient dynamic range for even the quietest home listening environments, which are often much quieter than your studio.
I didn't include more than two channels of 192kHz, because only DVD-A supports that file type, and only two channels' worth at best. Score one for SACD. The same table with 24-bit files is a tad fatter, which, you may think, is why we have DVD: all that extra room to store long words. But no, my friend. HDCD works just fine to encode extra dynamic range on a CD, yet it failed in the marketplace. DVD was created to provide an all-singing, all-dancing replacement for the aging CD format, and most members of the consortium couldn't care less about fidelity. Fidelity doesn't sell.
The mighty DVD format labors under many serious limitations, one of which is the sustained data-throughput rate of the transport. To play your basic 96/24 5-channel material, you'll need a 11.52Mbps pipe, way too big for DVD and pushing your luck for 100BaseT, I might add. To the rescue comes lossless data compression in the form of Dolby's Meridian Lossless Packing. An MLP-packed version of worst-case material, what Meridian classifies as “Metal,” would be only 7.18 Mbps on average.
|44.1 kHz||48 kHz||96 kHz||192 kHz|
|2||794 MB||864 MB||1,728 MB||3,456 M|
|3||1,191 MB||1,296 MB||2,592 MB||—|
|5||1,984 MB||2,160 MB||4,320 MB||—|
MLP inspects multiple parameters when performing its magic, including dynamic range and spectral spread. So, a 96/24 5-channel solo piano track would, due to its wider dynamic range and statistically limited spectral composition, pack down to an average data rate of only 4.4 Mbps. That's within most any modern pipe's capacity.
Of course, DVD has other tricks to reduce data rate and storage space, like half-rate groups for rear channels and LFE, along with better downmix abilities than DVD-V. Because listeners who have stereo playback would apparently be less interested in spatialization, one could safely assume that, for multichannel material played back in a 2-channel environment, tonal balance is the important issue when downmixing. Given that, MLP's downmix abilities should suffice for the majority of cases, eliminating the need for a separate Lt/Rt mix — in the audio zone, anyway. MLP also eliminates the need to redither 24-bit material to 16 or 20 bits, because it “knows” how to eliminate any unused dynamic range in the data.
Earlier, I threw out some numbers with the caveat, “…excluding overhead.” This brings me back to storage issues and that overhead. If you dig around in DVD-Audio bit-budget calculations, then you'll find a better than 50% discrepancy in the storage overhead required between PCM and packed files. The difference turns out to be related to DVD sector formatting. Bob Stuart at Meridian Audio helped to create MLP and was kind enough to explain that, “…because MLP carries decoder instructions very efficiently in the stream, it is possible to use less overhead in the headers of each disc sector than is possible in the LPCM case…In fact, MLP is the most efficient stream in the DVD family. Moving video is less efficient than LPCM, and the lossy-compressed stream options (e.g., MPEG, DTS, Dolby Digital) are also packed less tightly on the disc.” MLP has better fidelity as well, I might add.
Now, this is all fine and good for physical media: Format your disc, burn it and watch 'em march out the door into the open arms of Jane Q. Public. But what I think is more interesting is the data rate required to push these files through a pipe. Maybe you're sending a mix out for approval via the public network. Or, maybe you work in a broadcast environment, where folks think nothing of playing audio over a LAN. I know…as Marvin might say in this situation, “Don't talk to me about jitter.” Ethernet, 1394 and most other LAN-like mechanisms exhibit absurd amounts of jitter, but that's not important in this discussion, and, yes, jitter reduction can be achieved if needed. The point is that the aforementioned 1.4MB/second rate for CD-quality material can easily travel uninterrupted down a 10BaseT connection if you have the right hardware. Increasing the word length to 24 bits brings that rate up to 2.1 MB/second, making things more difficult for low-speed Ethernet. Of course, fast Ethernet would ensure uninterrupted service, but what about trying to move that data over a “broadband” connection?
I don't know about you, but my ADSL service is the cheap version, typically providing 384 Kbps, topping out at about 512 Kbps, and only in the download direction at that. The alternative is a download-to-disk approach, which works great as long as the number of channels and duration remains low. As my colleagues and I have found, it's painful even with ADSL or cable to move multichannel, high-density files around on the public network. An alternative is SDSL service, which should come with a Service Level Agreement, or SLA, that guarantees some minimum QoS and monetary compensation for outages, if you can afford it.
Another desirable feature for which you pay extra is a fixed IP address. This means that your local host can be addressed from a remote location, say, your client's place. This assumes that you've configured your machine for insecure but convenient FTP access ahead of time. Consumer broadband accounts sometimes offer fixed IP addresses as a value-add, but usually your router, gateway or modem is assigned an IP address dynamically, making it tough to FTP into your local network.
So, there you have it. My 512Kbps “broadband” pipe can really only carry about 5 bits of 44.1 stereo audio, not too hi-fi by anyone's standard. So, I continue to groove to the dulcet strains of lossy codecs, the subject of next month's “Bitstream.” Until then, stay warm!
Thanks to reader George “The Red” Blood for suggesting this column's subject. Got a topic? Give a shout and I may write it up. For links, back issues and occasional commentary, visit http://sene schal.net.
OMas has just returned from a long weekend of winter camping on the brawny shoulders of majestic Mt. Whitney. He took his iPod, so this column was created while under the influence of 192Kbps VBR stereo MP3s.