Downsampling

September 30, 2013
 by

In yesterday’s post 192kHz?I explained how we wound up with the different sample rates. A convoluted story at best. Today I thought it might be interesting if I touched on a little known problem with pretty much all CD’s. Downsampling.

Nearly all master digital studio recordings are made at higher sample rates and bit depths than a CD can handle, even if it’s only 48kHz. For example, Keith Johnson at Reference records all their materials at 176.4kHz/24 bits and nearly every recording studio is making masters at 96kHz/24 or higher. When a recording engineer is making a master recording of a group, a copy of a master analog tape etc. he is naturally going to make that master at the best possible resolution he can. Not at the lowest needed.

So what happens when we want to go make a CD version of that master tape? The original master is downsampled to 44.1kHz/16 bits through a sample rate converter. Depending on what the engineer started with, downsampling clearly loses some of what was in the original recording in the process.

How much of what we recognize as inferior playback of the CD, relative to high rez media, is due to this downsampling process?

This practice, I am convinced, is one reason many CD’s sound worse than their high resolution counterparts. If you read any of the thoughts by recording engineers on this practice, there are many that believe recording at higher rates and downsampling sounds better than keeping everything at the intended use rates. And then there are those that prefer upsampling. Clearly, there’s not much in the way of consensus.

In a conversation with Keith Johnson (whose ears I trust completely) he told me the single worst thing that ever happened to any of his work was the downsampling of the master tapes to CD’s. He hated the results and I am sure this must have prompted their subsequent release of their HRx series with exact copies of the masters. I’ll bet our friend Cookie Marenco of Blue Coast might have similar misgivings about downsampling.

The difference between the CD and HRx version played on a PWT is obvious and immediate. But I don’t believe it has as much to do with the increase in dynamic range and frequency response afforded by the higher resolution format as it does by the lack of downsampling.

Downsampling may be one of the worst things to ever happen to the CD and one of the central reasons we judge it so harshly.

It may also partially answer why we can make a digital copy of an analog event without much in the way of loss. We don’t have to mess with the final result of the conversion, just play it back at whatever sample and bit rate we recorded it at.

We never downsample.

Subscribe to Paul's Posts

5 comments on “Downsampling”

  1. Thank you for the mention, Paul. I agree that downsampling doesn’t help the sound of the CD. For that reason, we always mix simultaneously to DSD and 44.1. In our commercial studio, we use the 44.1 to create the masters for disc replication.

    We began doing the comparison tests about 20 years ago… comparing simultaneous mixes of 48k and 44.1k, then converting the 48k to 44.1k and comparing that to the 44.1k. There was enough of a difference that we felt it better to record at 44.1k. Many audio engineers prefer taking the analog outputs of 48 and re-recording into 44.1 over the conversion.

    When making a decision as to which audio to release for downloads, it can become quite complex. Many audio engineers, who I have deep respect for in our community, choose to mix/master the CD completely different than what might come out for SACD or high resolution. They feel they are optimizing the CD to compete with other CDs in the market (like loudness, for instance). Another reason that comparison tests without knowledge of the source of the masters can be deceiving.

    Regardless of anyone’s opinion of DSD vs PCM, we feel that releasing the audio from the mixes, provides the music listener a sonic experience the way the artist and engineer intended without conversion. Of course, the music listener’s playback system makes a tremendous difference. For that reason, we continually test our processes for all formats released.

    I’m working on a how-to paper for “Comparison Listening Tests” for our studio use. I’ll let you know when it’s done. 🙂

    1. Which is probably one of the reasons your Blue Coast releases sound so darned good, even the CD versions. Thanks Cookie, I’d love to publish your paper if you want it open to the public.

      Your work is well respected in our community.

  2. Whatever data the recording was made at, native that is, is the best for playback. We need to return to the assumption that everything is audible (even if we can’t hear it right now). Also, you might want to consider if once downsampled, can the losses be recovered by upsampling? I don’t think so. I do not own any of Cookie’s downloads, so the next is from that place:

    With due respect to all in the DSD camp (an I am one), there are PCM recordings, ie, AIX’s, that perform very well, arguably just as well. While I admit to questions still left, one has to look at remastering or mixing a digital recording in analog and why this “shortcut” for want of a better word is being used. Just as digital has it’s dark side, analog too, has limitations inherent in its technology, and those distortions and noises will be added back into the final digital product, no?

    As long as the music transports me, I don’t really care what the media is. I do care why though…and I do care to move the art forward.

  3. Yes, downsampling causes an audible degradation. My friend Barry Diament of Soundkeeper Recordings has written that downsampling from 24/192 to 24/96 causes the biggest degradation; going down to 16/44.1 is only a little worse. My own listening comparisons with sample tracks made from pristine, purist 24/192 or DSD masters and then downsampled to various rates has me agreeing with Barry.

    But there is no question that mix studios — by recording and mixing at 24/96 or higher and then downsampling for CD release — can create better sounding CDs than they did when 16/44.1 or 48k was the professional digital standard, all else equal. Of course, it’s not exactly equal, because now they’re using much more dynamic compression. 16/44.1 is not so bad unless you add processing (EQ, reverb, compression, various effects, level riding) and mix multiple tracks together. The digital artifacts are pretty bad. Working at 24/96 results in much, much cleaner-sounding mixes. Downsampling is audibly trivial compared to mixing with 16/44.1 tracks, given all the processing that typically goes on in a mixing session.

    Ironically, one of the reasons that modern mixing and mastering sometimes sounds awful (even on the 24/96 master) is that engineers can push the compression to ridiculous levels without introducing obvious distortions of one type or another. The “credit” goes to better hardware and software working at higher bit and sample rates. The engineers couldn’t get away with that sort of compression in the days of tube compressors or 16/44.1 ProTools projects.

  4. Paul,

    Perhaps it would be helpful for folks to learn why SRC might damage sound quality (ignoring changes in resolution). Ultimately, how the conversion is done can make a big difference in my experience. I suspect that this will have to lead to a discussion of digital filters.
    Oversampling, for playback, is not always bad. For example, many computer audio playback softwares now allow for oversampling in the computer. Some DACs now accept 352.8 and 384 rates (via async USB). Most DAC chips usually have a first OSF step that takes incoming data up to these 8x rates. Sometimes, one can do a better job with that first 8x OSF in the computer, with a custom filter setting, than what the 8x OSF does internally in the DAC. The playback software Audirvana for example, has very, vert sophisticated user configurable DFs where one can experiment with all kinds of filter configurations.
    In any case, seems like a discussion on digital filters, and the tradeoffs inherent in their designs would be appropriate here. Might be nice to use this as also a comparison between the differences between DSD/PCM.

Leave a Reply

Stop by for a tour:
Mon-Fri, 8:30am-5pm MST

4865 Sterling Dr.
Boulder, CO 80301
1-800-PSAUDIO

Join the hi-fi family

Stop by for a tour:
4865 Sterling Dr.
Boulder, CO 80301

Join the hi-fi family

linkedin facebook pinterest youtube rss twitter instagram facebook-blank rss-blank linkedin-blank pinterest youtube twitter instagram