What's the PMC "house sound?"

guest2521 · « **Reply #20 on:** 28 Feb 2005, 08:05 am »

Accurate should mean that the system reproduces the signal on the disc adding nothing and taking nothing away (ideally). So far the three way pmc models come the closest to this that I have heard.

Obviously no speaker can sound sound "live" but on female voice say my MB2 seems to come closer than any other to making you beleive that the singer is in front of you. I think when a lot a of people say "live" they really mean the beleivability factor.

James Tanner · « **Reply #21 on:** 28 Feb 2005, 01:44 pm »

Hi All,

I think the old phrase - "suspension of disbelief" is what a great audio system should do.

james

nicolasb · « **Reply #22 on:** 28 Feb 2005, 01:49 pm »

Quote from: biovizier

Accurate should mean that the system reproduces the signal on the disc adding nothing and taking nothing away (ideally).

If that were the case then there would be no point in upsampling (or even in oversampling). What you want to reproduce is not what's on the disc but what the original sound was like before it was compressed down to what can be fitted into any given audio format.

guest2521 · « **Reply #23 on:** 28 Feb 2005, 01:59 pm »

Once information from the signal is lost it is gone forever and cannot be regained. That is scientific fact. Any "information" that any process is adding back in is completely arbitrary though it may well be pleasant. With upsampling etc there is no addition of information - all you are doing is reducing the artifacts of the DAC process due to filtering and the conversion process itself by conducting them further away from the audio band (i.e. improving the accuracy of the reproduction of the signal). There is no contradiction with my statement, in fact it supports it. Also the sampling process is not compression - cd is an lossless digital process. What you are in fact referrring to is limitations on quantisation and sampling of the original recording. None of which over sampling and upsampling addresses anyway - though manufacturers and retailers will be happy to sell you snake oil that claims otherwise. All that can be done is to reproduce exactly what is on the disc - no more no less.

James Tanner · « **Reply #24 on:** 28 Feb 2005, 03:42 pm »

From my point of view we are dealing with "Production" vs "Reproduction" in our audio playback systems.
The goal of the recording engineer is to "Produce" as best he can the instruments and voices relative to the venue he is dealing with. If that venue is a full orchestra then that is the goal - if a single instrument in a small room - then that is the goal.
In playback our goal as a manufacturer is to build product that 'Reproduces' the input as accurately as possible. That means we strive to change the input as little as possible.
Some equipment will 'produce' a specific result - which may be pleasant -but will not be accurate to the input.

james

guest2521 · « **Reply #25 on:** 28 Feb 2005, 03:50 pm »

Exactly - the rest is voodoo and belongs with valves, cables costing more than $100 etc etc. However even manufacturers that still have legitimate products that do improve the accuracy of the reproduction do still sometimes create a mythology to help sell their products. Upsamplers and over samplers are a case in question. The manufacturers know they will sell more if they make outrageous claims about how the device achieves it ends - after all a resampler that merely carries out DAC and anti-aliasing filters at higher frequencies to avoid effects in the audio band sounds a lot less impressive than a mystical audio equivalent of a crystal ball that can somehow recreate missing parts of the signal or the original production.

The best products and best companies stand on their own without this based on good quality engineering and first rate results. Thats why I picked Bryston - no bullshit and class leading products. PMC likewise.

nicolasb · « **Reply #26 on:** 28 Feb 2005, 05:08 pm »

Quote from: biovizer

Once information from the signal is lost it is gone forever and cannot be regained. That is scientific fact.

That is simultaneously true, and totally irrelevant.

Imagine I give you a black and white photograph of my garden, taken in summer. If you know something about the biology of grass, you will easily be able to tell from that picture whether the grass was green or brown at the point the photograph was taken, despite the fact that there is no colour information in the picture at all. You can tell from the comparative brightness of the grass compared to other leaves, by whether or not there are bare patches of earth, lots of other things.

Starting from that black and white image you could therefore (given sufficient time, knowledge and computing power) reconstruct a colour image which accurately reflects what the original scene would have looked like in colour, despite the fact that your starting point is in black and white. If you didn't know anything at all about what colour grass normally is, then it would be accurate to say that the colour information would be unrecoverable. But of course you do know.

Now imagine capturing the sound of a violin as an arbitrarily good analogue signal, and then converting that into a red-book CD track. Again, information is lost - but, again, the recording itself is not our only source of data. It's possible to study what happens when a very high-grade analogue signal is condensed into a CD. And using the combination of that knowledge and the recording, you can recreate information that was lost during the recording process.

The aim should not be to construct a new analogue that exactly tracks the digital version. The aim should be to take the recording and extrapolate backwards to what the original sound must have been to produce that digital recording.

guest2521 · « **Reply #27 on:** 28 Feb 2005, 05:41 pm »

If the information is gone and cannot be regained then it is entirely relevent. The facts:

a) It is not the original picture but a distorted version of it. Even if you make all the grass green it will not look the same as real grass with regards to it's colour structure. It will not look as accurate as the B&W picture (some of the relationships within the original will have been distorted and extra false information has been added) but some people will find it more pleasing - which is why people do it, in accordance with my point. An analogy would be valve amps adding audible 2nd harmonic distortion or "colour" to "B&W" low distortion transistor amps without that added colouration. I work in nonlinear image and signal processing so I have some experience in these matters directly. But make no mistake it is still a distortion, even if people like it. I have never seen a recoloured movie or photo that looked even remotely authentic - and to the eye sins of omission are lesser than sins of addition due to the evolutionary development of the sensory parts of the brain and the way the mind processes information and focuses its attention on specific areas. I prefer the original black and white photo every time - it is more accurate and it is what the photographer intended!

b) Upsamplers and over samplers do not even do this anyway. They act in accordance with my priniciple of accuracy rather than your principle of reconstruction of missing information. The manufacturers may well claim they have somehow captured the information that was lost but this is solely a ploy so that they can charge an order of magnitude higher prices. certianly these kind of device can be built to current limits of device technology for $500. However people are charging $10,000 in some cases.

Your comment about the B&W photograph is a misleading analogy. There you are substituting into the B&W photo large scale changes to which the eye does not really care - using prior high level knowledge. Like replacing a violin on a recording mix with a bassoon, both of which have been accurately recorded, i.e. it is a large scale "remix". That is not the same as making compisite signal changes at a complex level to which the ear is very sensitive. The audio signal is a complex signal and the ear is a complex and nonlinear measurement instrument which works differently on a qualatative level to the eye. These adjustments are therefore intrusive - if upsampling worked as you beleive.

I suggest you do some more research on sampling theory, information theory and also the concept of entropy. Particularly nyquist. The problem with CD is playback - information is not lost by sampling at a frequency double that of the analog bandwidth. Upsampling improves sound by shifting CD's problems further away from the audio band. Simple as that - nothing else is required and it most definately does not reconstruct missing information or build in prior knowledge. If the sampling frequency is lower then that information is lost and cannot be reconstructed. The signal is the only source of DATA - the rest is a priori assumptions that you are using to introduce an arbitrary distortion.

nicolasb · « **Reply #28 on:** 28 Feb 2005, 07:41 pm »

Quote

The signal is the only source of DATA

It doesn't matter how many times you repeat that assertion: it's still wrong.

guest2521 · « **Reply #29 on:** 28 Feb 2005, 09:29 pm »

In your OPINION. Mine differs, and is backed up by science and testable hypotheses not belief. You cannot extrapolate back to recreate missing parts of the production that were not captured in the recording - that breaks the laws of the physical world as it is currently understood. If you know better then please state how information can be created from nothing. It doesnt matter how many times you state I am wrong - the universal laws exist regardless of your opinion. The signal is the only source of data concerning the original production - anything else is, in scientific terms, a belief prior and subjective. Apply Occams razor - a perfectly reasonable and simple explanation exists for the bebefits of upsampling. There is no need to invoke pseudo-science to explain it, except for marketing purposes.

thomaspf · « **Reply #30 on:** 28 Feb 2005, 09:54 pm »

"suspension of disbelief"

this might be an old phrase but I never have heard this summed up so well. My point is that in order to create a recording that achieves this you might needs tools (speakers) that are optimized differently from the speakers you would use to actually experience this.

For the heated discussion on upsampling I can only point to the endless stream of thoughts on this on some other well known audio forums. The consensus seems to be that you obviously can not invent additonal data that is not there but that is only half of the story. For playing back 16/44.1Khz data you need to employ a high degree of digital filtering to avoid aliasing. This filtering is impacting the sound in the audible range but you might be able to get away without any such filtering if you upsample. Most dacs do internal oversampling anyhow and upsampling just moves part of that process outside the chip.

If you stay away from statements like bicubic splines for interpolation (Wadia?) sound more musical than other forms of interpolation and have a DAC that avoids the filtering of the upsampled signal then there might be a benefit after all.

Cheers

Thomas

guest2521 · « **Reply #31 on:** 28 Feb 2005, 10:43 pm »

Thats exactly the point I was making - a simple technical explanation without any need for mysticism. As I said earlier, by upsampling you move the "brick wall" filtering to a much higher frequency that does not interfere with the audio band quite so much. Anti-aliasing filtering is still required however. The interpolation is just a way of further improving upon pure upsampling - however this is just another way to reduce the artifacts of digital to analog conversion - no additional information is added it is just a way to improve the DAC and filter process and produce a more accurate signal, using the information in the digital stream (and only this information!). Manufacturers play upon this interpolation as a means of suggesting that they are somehow adding all the information that was "compressed" out of the original analog when converted to digital. This is completely misleading as no compression is used and the capture of information is total within the limits of nyquist and the jitter of the ADC (which could be reduced with further processing). Compression is a misdirection as it suggests somehow that this misisng information can be "uncompressed" back if you are some sort of hifi shaman.

Nick.

dan_lo · « **Reply #32 on:** 28 Feb 2005, 10:49 pm »

About the debate about digital data:
nicolasb - you are dealing with voodoo stuff here. Trust me, there aren't many people who knows more about data modeling. What you are saying cannot be achieved in a *general* data model. There are mathematical proofs that show it. If you are dealing with a specialized model it can theoretically be done, but it's very, very complicated, and not neccesarily correct. It does not suit all recordings. No one even tries to do it.

Oversampling and upsampling, "interpolation" and all that are not meant to add data upon the digital bits written on the disc, but to improve the accuracy of the conversion within the audio band. That's all. If you want further explanation I can post it.

guest2521 · « **Reply #33 on:** 28 Feb 2005, 11:06 pm »

Dan_lo,

Yes thats exactly correct - what nicolasb suggested only works in a very limited subset of situations where you have very specific aims. One example I have worked on myself is nonlinear interpolation schemes to minimise error in a specific nonlinear function - which is not the same as recreating the signal. Such schemes are extremely sensitive to a priori knowledge and may well bear little resemblence to the actual underlying system. It works well in medical tomography but in underdetermined systems you get out exactly what you get in - apply tomographic techniques to ionospheric electron density and it's a whole different ball game. All you get out is the model you put in and the output has no scientific value - its just a mirror to the modelers preconceptions. Such techniques are not applicable AT ALL to reconstruction of audio signals. Moreover it has never been done - despite the claims of the manufacturers. A simple explanation exists for the efficacy of upsampling and interpolation and nothing further is required. In any case the effects are not profound - just expensive, hence the need for mumbo jumbo to sell these products. DCS? Wadia anyone?

Nick.

dan_lo · « **Reply #34 on:** 28 Feb 2005, 11:06 pm »

Quote

If you stay away from statements like bicubic splines for interpolation

However ridiculous it sounds - it is possible . I'm not saying it's true , but it's possible. You can look at the process as a sort of equalizer. There is more than one way to build it and more than one windowing method. It is certainly possible that one option have different attributes that can be desribed as more "musical" than others.

It can also be rubbish.

guest2521 · « **Reply #35 on:** 28 Feb 2005, 11:10 pm »

Splines do have certian desirable statistical properties especially for interpolation and prediction. Certainly they should be the first port of call. I suspect the exact means of interpolation isn't too critical and is more likely to be dictated by what's avaialble or easily implemented in DSP.

thomaspf · « **Reply #36 on:** 28 Feb 2005, 11:45 pm »

If we talk about the different ways to connect two points with a non linear interpolation function there are many ways to do this.

I have not heard of anyone taking a large body of material recorded at 96Khz, stripping out every second sample, and then upsampling the resulting 48Khz back to 96Khz.

This seems the only sane way to decide whether smoothing function X or X gives results that are actually closer to the original you started with. Of course you will be able to always find a counter example that works particularly bad with one function.

This seems a typical case that one smooting function introduces artefacts that are more pleasing to the ear than others. Like a digital tube filter that is adding uneven harmonics to create that warm tube sound.

Cheers

Thomas

guest2521 · « **Reply #37 on:** 1 Mar 2005, 08:00 am »

The interpolation is working in a very constrained way with the sole intention of reducing DAC and filter artifacts. As you say - there is no way in a general case you can suddenly invent half the data (even if there are auto-correlative and other nonlinear constraints). Some smoothing functions will produce less or less obtrusive artifacts than others.