Sony CD 's, Variable Co-A-Wha??? (long)

[Digital Bob]

Hey guys,

Maybe one of you all can help me out with this one. I cannot find a definitive answer for why Sony uses a variable coeffecient digital filter in their better CD players. I don't want to get too techy mostly because I'll invite someone who is way smarter than me to give an answer I won't understand But let me lay out the details as I know them. Please point out where my understanding is incomplete.

I understand (at least I think I do) that this feature provides different filter "curves" for ultrasonic frequncies. Now, the nay-sayers will say "Who cares? It can't effect the sound because all the changes are above 20k".

Many believers will say "Well, some studies have shown that we may be sensitive to harmonic relationships well above 20k. Even if we can't detect the pure fundamentals up there"

Now, if we agree that there may be validity (there is, IMHO) to the high frecuency harmonics theory, there is still a problem. None of those frequencies can possibly exist on the disc. As we all know the maximum response on the disc is limited to 22.05k by the sampling rate. So where does that leave us?

Some true believers may say "Even though the ORIGINAL hf harmonics are not on the disc, one natural byproduct of reproducing multiple tones simutaneously is overtones. These overtones might act to *fill-in* the missing hf harmonics of the original recording"

OK, so if we agree that this is possible. I'm still confused. Here is what no one I've spoken to (even at Sony) can explain to me... Why a *varible* filter?
If you believe those hf tones are good (no matter how they got there) then you would want to preserve as much of them as possible by using a very gentle slope. Since the sampling noise is moved WELL into the ultrasonic via oversampling, there is no reason not to go with a low order roll off to preserve the "air". The only reason to use a steeper slope is to stop intermodulation of ultrasonic and audible frequencies which can produce tone multiples. But since we agreed that these "created" ultrasonic overtones "replace" the missing natural hf harmonics.... why try and remove them with steep filters?

Again the question is why do we need *varible* filter curves. If the "recreated" overtones are audible (in some way) then use a gentle slope to preserve them. Even if they are not audible then use a gentle curve anyway (the sampling noise is a non-factor) and prevent the possible "phase ringing" when using steeper slopes.

I don't get the need to vary the slope. And my sometimes cynical side comes out when no one at Sony I've spoken to has a cogent answer. Please don't misunderstand the point of my lengthy post, I'm not debating the validity of ultrasonic perceptions, audibilty of filter slopes, or anything else. Heck I wouldn't have the technical credentials to even begin staking out a position on those. It's simply a question why the filter is varialbe instead of fixed at the best position.

Thanks for making it all the way through this novella and hopefully for shedding some light on this for us.




dB

[ThomasW]

db

I can't begin to answer you question. But I can tell you that there is an subtle, but audible difference between the filters on my SCD-777ES.

First the filters aren't "variable" they are fixed. And it's necessary to switch between them.

I'm not sure that I subscribe to Sony's "characteristics" of what filter is best for a specific type of music though.

Here's a quote from the 777ES operating instructions that may explain what's going on.

"On the other hand, slow roll-off filters cut noise generated during sampling, and are able to hold pre-ringing and post ringing (a kind of sound smearing) in the impulse response to a minimum"

They then "qualify" their entire "Digital Filter Function" section of the manual with the following statement.

"Note: The Digital filter function primarily changes the characteristics outside the audible frequency. They cannot affect changes within the audible frequency such as those provided by the tone controls of the amplifier. Therefore, with certain combinations of hardware and software, there may be no noticable effect after switching the filter"

Talk about hedging their bets

It's also highly possible that given the original retail price of these units, Sony felt that the customer needed a little "extra" for their money. :B

P.S. I'll see if I can get our resident "digital expert" JonMarsh to "chime-in" on this. He too has a SCD-777ES.




theAudioWorx
Klone-Audio

[Digital Bob]

Thomas,

You know I actually don't doubt that here is an audible difference on a high-res system. I thought I detected one too.

I've been around enough hifi over the years to appreciate that there may some subtle things we can barely/if at all explain. {Oh God, I'm a tweaker! h: } I just wish someone could explain why they included *variable* option instead of simply selecting the one that sounds best.

I guess a gross analogy would be... Why include the ability to vary the bass and treble (tone controls)if they sound best in one position? In this case "off". Instead you would not go through the expense and sonic degradation to include them in your design, you would simply "voice" the product the way you want it to sound and delete the controls.

This is not a perfect analogy but you get the idea. BTW- the cynic in me might agree with your statement about "the extra for their money" marketing spin. This IS Sony we are talking about here afterall.




dB

[JonMarsh]

First, let's take a more accurate look at what Sony did, than what has been described above. When in doubt, go to the horse's mouth, or any other convenient and functional orifice in the anatomy


From Sony:
• Std. The standard setting meets the classic
digital filter objectives of the steepest possible
roll off above 20 kHz, with the greatest possi-ble
suppression of ultrasonic noise.
• df-1. This setting incorporates smoothing inter-polation.
Conventional 8x oversampling uses
three cascaded stages of 2x oversampling. In
contrast, df-1 performs the full 8x oversam-pling
process in a single pass. The result is
smoother processing and clearer sound.
• df-2. This setting maximizes the digital word
length. In each stage, it adjusts the word
length of the input number to match the noise
shaper in the next stage. This eliminates all
non-linear portions in the computation process
and suppresses quantization noise.
• df-3. This is an all-new 224-order design with
an even number of filters. Unlike conventional,
odd-number filters, the original data samples
are retained as-is, without alteration by the
oversampling process. Frequency response is
flat to 17 kHz, while noise suppression at 26
kHz is 80 dB.
• df-4. An enhanced version of df-2, this setting
increases the computation word length by four
bits, increases the filter degree twofold and
employs an even-number filter. Frequency
response is flat to 18 kHz, with a gain at 20 kHz.

Why does Sony do this, and offer several different digital filter algorithms? You may well ask, why do Wadia (RIP), Pioneer Legato Link, Levinson, Krell, CAL Audio, Meridian, et. al., offer different filter algorithms (they do, just check their literature and the test reports- different from each other, anyway).
You'd kind of figure that all of these bright people would have figured out by now the one CORRECT way to do digital filtering for a CD player, huh?
Well, no, they haven't, because people don't agree on what is audibly most important or measurably most important, or where those two meet. So, Sony is the first manufacturer that has implemented in one player several approaches to the implementation of the digital portion of the output filter, which is selectable by the user.

Do you believe that the absolute flatest response through the 20-20KHz passband, with the steepest possible attenuation of out of band noiseshaping products is the only way to go? (the basis on which conventional DAC designs have been done for years by most companies)- then go with the Std. Filter.

Do you think a slow roll off with minimum phase impact, minium FIR time artifacts, but some upper passband attenuation and very poor supersonic attenuation is the only way to go? (Wadia did) Then DF1 is your choice.

Or maybe your'e inclined to an approach more equivalent to up sampling, where the orignal sample levels are NOT thrown away as most all oversampling FIR filters do, and instead the original sample points are all retained, with high bit rate interpolation, with a gentle roll off filter with minimal phase impact that starts rolling at a lowish 17 kHz, but has 80 dB or more attenuation of ultrasonic noise products? (Sign me up for this) Then DF3 is your choice.

With these options, you have choices to select for reproduction based upon your own biases and values. IMO the true impact is not in the presence or absense of frequencies above our nominal hearing range, but the effect in the audible range on transient response, phase shift, and retention of original sampling points that these filters offer, as well as the variation in ultrasonic noise, which some systems will be more sensitive to than others (particularly amplifiers and tweeters, in my experience).

Regards,

Jon




Earth First!
_______________________________
We'll screw up the other planets later....