24-bit/192kHz Is Pointless?

I wish to highlight certain misconceptions on this 16-bit / 24-bit word length modifications (I am not an IT guy) from what I read below:
All digital filters work by mutiplying data by coefficients. So a 16-bit word turns to 24-bit word by multiplying by a 8-bit coefficient. But usually the coefficients are much longer than 8 bits and for example, if you multiply 16-bit data by 24-bit coefficient, you end up with a 40-bit word. Then this must be rounded off or truncated to get back to 24-bit data at the output of the digital filter. Any modern digital filter will "interpolate" 16-bit input data to 24 bits at the output. But no matter what you do, WHEN YOU START WITH 16-BIT DATA, YOU WILL ONLY HAVE 16 BITS OF PRECISION IN THE OUTPUT. Even if you multiply by 1000-bit coefficients, increasing the actual resolution of the original signal is impossible.

> murali

I am not sure if I am qualified enough to talk about this. However, let me just try to present a bit of my understanding on this.

Digital audio is a discretized form of analog audio which is, in some way of saying, continuous in space and time. Analog audio is given by a wave in a medium. A wave is actually nothing but periodic undulations of the points (in space) in a continuous medium.

When this wave is discretized, one has to do it obviously both in space and time. The redbook (CD) format is described as 16 bits/44.1 kHz. In some sense a wavefrom in space is discretized by 65536 (=16th power of 2) points instead of infinite number of points as in a continuous medium. Obviously this waveform is changing with time and one has to take snapshots of the waveform many times in a second and in CD format this is done 44100 times in a second (and not infinite times per second).

Obviously it does not take a genius to appreciate that we are approximating the real thing by this digital discretization. This obviously takes a lot away from the music and explains why a lot of people still do not get nearly the same musicality with all its nuances in CD and still today prefer LP records.

It also naturally follows that more the no of bits or the sampling rate, the better the digital discretization. I have a portable LPCM recorder from Sony (called the Sony PCM-D50, details only available in the Sony Professional equipments website) which can record upto 24 bit and 96 kHz sampling rate. I have done a lot of recording of live music with it and one can easily see the advantage of having a better resolution both in no of bits and sampling. Even a 24/96 recording downsampled to the redbook CD format sounds a lot better than originally recording with the CD format.

Now upsampling has two sides, as I see it. One is a purely engineering reason of not having to make an almost perfect anti-aliasing filter which is used in CD players right after the signal goes thru the DA converter. Without going into anymore details of a rather technical subject, this is needed to get rid of certain undesirable and harmful artifacts of the digital discretization process.

The other is actually an attempt to 'improve' the digital signal by putting in data that is not there in the first place. In scientific terms this is called 'interpolation', the most naive example of which would be to join two points by a straight line (for example). This is an attempt to make the data less discrete by increasing both bits and sampling freq. Obviously a lot of mathematics would have to go into a good interpolation. It is also very clear that if the added (interpolated) points in the middle are arrived at in a bad way, the result would be disastrous. It can also happen that for one kind of music it may work nicely, but for another kind it may fail to improve, it may even prove detrimental. This is costly research and that's why upsampling CD players have generally been quite expensive, until the recent CA products.

This sort of interpolation of data (that is adding more data points by interpolation) is always happening in a LCD TV whenever the source picture has less resolution than that of the LCD TV, for example from the set top box or cable or from DVD. This is why when I bought my LCD TV, one of the major criterion for me was to see which brand gave the best picture from a SD (standard Definition) picture source, that is, which brand had a better mathematics and software put into it for this upscaling. We ended up buying Sony Bravia (despite lots of suggestions on Samsung in this forum, we also very much liked Pana) and after a coulple of months of burn-in the picture from my BigTV source is really really good (BTW LCD TVs also do have a burn-in time, after which black levels really improve).

At the end I like to add, I generally found nothing against upsampling in the interview of Dan Lavry (although he emphasizes only one aspect of the upsampling) or in the other forum as mentioned above.

My CA azur 740c is a wonderful CD player for the price. It upsamples to 24 bits/ 384 kHz. I have had it for slightly over a month now. According to user reviews, it's got a burn-in time of 150 hours. So far I guess I have put in about 100 hours, but the sound is improving and so far has stood the test of all kinds of music from folk to Indian classical, to Piano concertos to symphonies, to African music etc.
Get the Award Winning Diamond 12.3 Floorstanding Speakers on Special Offer