ben@moncol.UUCP (Bennett Broder) (01/24/85)
With all the talk on the net recently about CD sampling rates, I have
wondered why nobody has brought up the question of sampling accuracy.
I have read in audio magazines that there are 2 basic types of CD
players: those following the design of Sony, and those following the
design of Phillips. From what I understand, the Sony uses a 16 bit
sample, but the Phillips only uses 14 bits. It seems to me logical
that the Sony system, which quantizes the signal into one of 65536
different amplitudes would have far greater resolution than the
Phillips, which quantizes into one of 16384. But I notice that many
of the high end CD players (like my Revox) use the 14 bit
system. Are there advantages to the Phillips system that justify
the use of fewer bits? Do they use some kind of 'trick' to achieve
the same amount of resolution? I would appreciate a discussion of
the merits of these competing systems.
Ben Broder
..vax135!petsd!moncol!ben
..princeton!moncol!benkarn@petrus.UUCP (01/25/85)
> ...But I notice that many > of the high end CD players (like my Revox) use the 14 bit > system. Are there advantages to the Phillips system that justify > the use of fewer bits? Do they use some kind of 'trick' to achieve > the same amount of resolution? Yes, they do: the 4x conversion rate. The 14-bit D/A converters run at four times the rate of the 16 bit units. The resulting noise power WITHIN THE AUDIO BAND (0-20 khz) is the same with either scheme. There are a couple of intuitive ways to see this: 1. The 14-bit D/A converter generates 4 times as much quantizing noise. But it is spread over 4 times the frequency range, so when you filter off at the same point (20 khz) you end up with the same noise level. (This is the explanation given in the Phillips Tech journal on the CD player.) 2. The 14-bit D/A converter has four times as many "opportunities" to control the final output voltage during a CD sample interval as a 16-bit D/A converter running at only 1x the sample rate. By averaging the discrete values over four D/A clock intervals (in the low pass filter) the 14-bit D/A converter can achieve "intermediate" voltages that it cannot create directly in a single sample. Suppose I had a D/A converter that could only produce the integral voltages 0, 1, 2, 3... volts. If I could average the four voltage samples 1, 2, 2, 2 (which the converter CAN produce) I get the result 7/4 = 1.75 v, a finer resolution result than I could get with just one sample. With a little thought you can see that such a scheme would yield 0.25 volt resolution, four times the resolution of the basic D/A converter. This assumes of course that the "coarse" steps out of the D/A converter are very precise, much better than the usual +/- 0.5 bits. There are speech encoding schemes that take this oversampling technique to the extreme, e.g., ADPCM which takes 1 bit samples at 32 khz. Clearly, if you try to get a 16 khz bandwidth it would sound terrible (only 6 db S/N) but if you filter it down to a voice bandwidth it sounds quite alright. It is interesting to consider the possibility of oversampling and digital filtering on the RECORDING side of a digital audio system, but it doesn't make as much sense because the speed of a successive-approximation A/D converter is generally inversely proportional to the number of bits. (Remember that such an A/D converter INCLUDES a D/A converter that must run at N times the sample rate, where N is the number of bits/sample.) A 14-bit A/D converter running at 4x44.1 khz would have to have a basic clock rate of 2,469,600 hz while a regular 16-bit A/D running at 1x44.1 khz only needs 706,600 hz. I suppose that if you got the resolution down enough you could use a flash A/D converter, but it probably isn't worth it. Phil
jj@alice.UUCP (01/25/85)
Well, you've left half of it out. The Phillips method, which uses a 14 bit D/A, also does some digital processing that's been discussed at least 15 times here that ups the sampling rate by 4, thus spreading the noise out over four times the bandwidth. The sigal, however, is NOT spread out by this process. A little preemphasis/deemphaiss built into the various filters, and you have the same SNR, except that the Phillips is linear phase, and the Sony isn't. -- TEDDY BEARS HUG PENGUINS "When I first landed here, I didn't like the beer, but the attitude seemed good enough, and the way ahead seemed clear. The beer still tastes like glue, the road seem rougher too, and the ..." since my heart's home to me. ..." (allegra,harpo,ulysses)!alice!jj
hrs@homxb.UUCP (H.SILBIGER) (01/25/85)
Its is PhiLips, not PhiLLips! ////////////////////////////////////// What do you call milk of magnesia and vodka? A Phillips screwdriver! Now will you remember?