Upsampling CDs

May 16, 2021
 by Paul McGowan

9 comments on “Upsampling CDs”

  1. Let’s see if I’ve understood correctly. Upsampling creates intermediate samples between the samples in the original data (in this case, a CD). If the upsampled rate is (for example) 4 x 44.1 = 176.4 kHz then upsampling creates three added samples between each adjacent pair of samples in the 44.1 kHz data. The intermediate samples are created by interpolation using a digital low pass filter at about 22 kHz.

    One reason why we might use upsampling is because there are more options in the design of a digital filter, allowing for a final analog reconstruction filter with a gentler characteristic. The overall filter shape is the combination of the digital interpolating filter and the analog filter. The benefit is entirely dependent on the design of the digital filter.

    Given that the DAC might well use upsampling as part of it’s implementation, why would we use the interpolating filter in a games console in preference to the filter in a highly regarded DAC?

      1. Good afternoon Mark!
        Perhaps you mite have an answer to this question.
        If what Paul said is true, then how come a 32 to 64 bit PCM wave file sounds better to me then an 8 to 24 bit PCM wave file?
        I rip CDs on my computer all the time.
        But however, my computer spits them out in 24 bit 48 KHZ.
        But I can use either Audacity or Wave Pad to up sample them all the way up to either 32 or 64 bit float in to 398KHZ.
        But when I get ready to put them on my digital talking book player, I have to reconvert them to 16 bit, 44.4 or 44.8 KHZ.
        The player won’t play anything higher then that.
        I guess that’s the way that NLS programed it.

  2. This has me wondering: how are the intermediate samples created when upsampling? Are they linear interpolations of the two adjacent incoming data samples? Or is a more sophisticated algorithm used (an interpolation based on Fast Fourier Transforms and a much wider set of data values, say)? Or do approaches vary? Might inartful interpolations add ‘information’ that is very subtly counter to what our brains would expect? In that case, I can imagine upsampled music introducing a subtle discordance that our brains have to work to discount – at least in that one respect.

    I presume that there must be *some* form of interpolation in order to raise the frequency at which the bypass filter operates: simply reiterating the data value of a sample multiple times in the interval before the next sample wouldn’t get you anywhere.

    Aside from the above mentioned effects, it seems that upsampling would not introduce jitter. That is, assuming perfect interpolation for the intermediate data values, there would be no jitter introduced as a result of clock timings. Is this also correct?

    1. Sirdodo: I believe that the intermediate samples are normally interpolated using a low pass filter with a corner frequency around 20 kHz. There are many different alignments of low pass filters, and the designer can optimize for flattest frequency response, most linear phase, lowest overshoot in the time domain, or some compromise between. As you point out, linear interpolation is a type of LPF, but probably not a realistic choice in this application.

  3. Ultimately, & practically, if you already have a 1,000+ Redbook CD library & you are
    using a Marantz CD player that constantly receives five star reviews, does it really behoove
    you to spend the extra to upsample everything for such a small audio improvement?
    I guess it’s a matter of personal priorities, especially if you are satisfied with your current audio rig.

  4. There is no improvement with interpolation. It is all in your mind. You can’t “invent” more music. 16/44 is all you need for reproduction.

    1. Oh goodness, this is unfortunately far from the truth. Listen, for example to a DSD recording vs. a 44.1kHz PCM recording from the same microphone feed and if you can’t hear that difference as plain as night and day you might want to rethink the resolution of your HiFi system.

    2. I think CtA makes two different points here.

      First, interpolation can’t invent detail. It can only fill in the gap between two CD samples by making the assumption that the analog signal does not contain any frequency components above half the sampling rate. However, we do have to consider if there might be an audible advantage in using a digital filter at 20 kHz as opposed to a brick wall analogue filter.

      Second, CtA suggests that there is little benefit in recording with resolution higher than 16-bit PCM sampled at 44.1 kHz. Now, that’s a more interesting subject for discussion.

Leave a Reply

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

4865 Sterling Dr.
Boulder, CO 80301

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