[Marbles]

I just got a P1a, and I understand upsampling just fine, but don't know what interpolation is.  Maybe I know it by some other word?

Thanks for any definition.

[Marbles]

Just found this:

Interpolation is the process of using known data values to estimate unknown data values.

I suspect that you use interpolation in your upsampling in the P1a?

[dallasstarsfan]

Marbles,

Check on these threads:

quick explanation


and


long explanation

and

really long explanation

[JeffB]

Well, I only read the quick explanation, but I have questions.
Interpolate means to guess values in between known values.
If you have a sample at time(t0) of 65531 and a sample at time(t1) of 65535.
Well you could guess that half way in between t0 and t1 you would have a sample of 65533.   In effect, you're making up samples.  Information theory tells us that 44K samples is enough, so there is no point making up more.

However, if you could guess the value of a sample to more precision, then that might be an improvement, but I don't think you would call that process interpolation.  That would be more like creating a best fit line through a set of points.

And it is a little hard to believe that you could use an algorithm to predict a more accurate sample.  It would be interesting to see the same source event recorded in 16 bit and 24 bit.  Run the 16 bit through this algorithm, and then statistically quantify whether the 16 bit sample is more or less accurate than a modified 16 bit.

However, after reading about jitter earlier today, I really have my doubts about drawing any accurate conclusions.

[gary]

Maybe I was missing something, but when I had my P1-A I could hear absolutely no difference whether interpolation was on or off.

Gary

[eico1]

Interpolation is the process of using known data values to estimate unknown data values.

That is a very good explanation.

All oversampling/upsampling/re-sampling uses interpolation, though there are endless variations in the implementation!

steve

[ghersh]

Maybe I was missing something, but when I had my P1-A I could hear absolutely no difference whether interpolation was on or off.

Gary

Well, it, most likely, won't make any difference. The point about interpolation is that it doesn't add any information. Whatever was missed during the original sampling is gone. The audio and video signals are of impulsive nature, not the smooth curves. So think of the situation where you've missed an impulse. It's gone, and no interpolation will restore it because it just doesn't know it was there in a first place.

A marketing gimmick, for all I care.

[ctviggen]

When they oversample, they have to use some interpolation.  Oversampling is simply taking two samples and making, for instance, three from the two.  That third sample has to have a value.  An easy to calculate value is simply the average of the two samples.  It should be noted that oversampling is more complex than this, as there are frequency domain issues that need to be addressed.

[woodsyi]

I don't think it's all marketing gimmick.   As I understand it, over sampling and upsampling to 24/192 doesn't add any to the original signal, but it gives better working data for processing.  Here is an example of one way over and up sampling (using interpolation) theoretically can be beneficial.  Of course, everyone claims to have the most "musical" and psychoacoustic proprietary algorithm developed for DSP.  That may be some marketing gimmick!

First of all, let me point out that there's no kind of magic here, as everything can be explained technically.
First of all, let's say the upsampling method CAN'T improve anything. The sound of a digitally upsampled DAC is better because it is the non-upsampled one to be worse.
For, let's see what happens to a standard 44.1 kHz digital signal when it is converted directly by a DAC. Before going into analogue, the digital signal crosses a digital filter that oversamples it (normally 8 times, 8x oversampling, as usually called) and a second digital filter with very high slope that cuts off all the garbage above a certain frequency, quite close to the audio band.
Once the signal has been converted into analogue, it crosses another filter, an analogue one, normally of the 2nd or 3rd kind, that introduces phase rotations into the audible spectrum.
Now, how can we consider the effect of a phase rotation in the time domain?
Let's suppose to have a musical instrument that plays its fundamental tone and its harmonics. The first ones normally are reproduced fine...but the higher order ones are delivered to your ears with a phase rotation (with respect to the first ones) and hence with a time delay that can be heard as distorion.

What happens with upsampling? The standard 44.1 kHz digital stream is interpolated and the samples are calculated as the original signal had a 192 kHz sampling rate. BUT!!!! This process adds NOTHING to the original signal!!!! Even at 192 kHz the signal is still extended till 20 kHz!
The difference now is that the signal crosses digital filters centered at 96 kHz and the following analogue filter will be centered far from the upper limit of the audio band (actually, near 96 kHz!!!).
This means the analogue signal coming out of the DAC will be more faithful to the original one in the time domain (less phase rotations, that is).

[/b]

[PhilNYC]

Interpolation isn't meant to add any kind of resolution to a recording.  It can't.  If it could, record companies could record things at a sample rate of 100hz, interpolate/upsample/oversample, and fit 1000 hours of music on a single CD.  Unfortunately, it doesn't work that way.

The purpose of interpolation (or upsampling/oversampling) is to make the DAC chip behave a little better and to move digital artifacts to a higher frequency (making filtering them out easier).  Arguments against upsampling are that it introduces more artifacts than it removes, but if its implemented well, it should do the job well.

[ghersh]

I don't think it's all marketing gimmick.  

OK, let's separate different issues. You're talking about the benefits of increasing the sampling rate because of the brick wall filters. Can't argue with that. But then simply doubling the sampling rate, from 44.1Khz to 88.2Khz and performing simple interpolation is the way to go. BUT - consider the situation when sampling rate is changed from 44.1 to 96 or any other value which is not multiple of 44.1. We simple can't resample in some clean and unambigious way, interpolation becomes more complicated, in general we introduce some noise. The martetting gimmick is to claim that changing from 44.1 to some higher rate not multiple of 44.1 is somehow improves the sound. Changes, I would say, and possibly degrades.

Changing from 16bit to 24bit is primarily making it compatible with the numerious processors designed to deal with 24 bit words. Other than that, this doesn't add any resolution, of course, and in general, completely unnecessary step.

[woodsyi]

[quote="ghersh
OK, let's separate different issues. You're talking about the benefits of increasing the sampling rate because of the brick wall filters. Can't argue with that.[/quote]

Given that, the real question is whether the implementation of upsampling and DSP will introduce more errors than what it corrects for.  This is where different hardware, data transfer protocol and software will make a difference.  I don't know which is ultimately better, but some oversamplers are better than non-oversamplers and the same is true vice versa.  I am using an upsampler (it gets to 24/192 in three stages which I don't like) that uses I2S transfer protocol which sends separate clock and data informations to DAC.  I like the resulting sound.  It sounds more like my TT than my other CDPs including a SACD player.  It deciphers more real, non-spurious details in a redbook CD without losing harmonic balance.  But I am sure there are some non-oversampling system with possibly tubed output stages that are better then this one in somebody's system.  I am happy with mine for now but I am looking into tweaking somethings in it as I learn more about the IC chips and other parts in it.

[ctviggen]

That could be because of better timing, not necessarily because of more samples.  It may also be because of the algorithm it uses to determine what the sample values should be.

[ghersh]

That could be because of better timing, not necessarily because of more samples.  It may also be because of the algorithm it uses to determine what the sample values should be.

Well, I2S may be a factor.

The best way to compare is to have the same CD player which allows to tuggle between upsamping say to 88Khz and leaving 16 bits, or upsampiing (oversampling?) to 24bit/192Khz, before going thorugh D/A conversion.

Here is by the way an excellent example of marketoid crap, taken from the web page of unnamed vendor. Here is their new 24/192 CD player:

[starts]
The technology comprises of two main parts - interpolation and up-sampling.  Interpolation in essence is the re-description of 16 bits of data into 24 bits - 8 bits are 'added' to the combination.  Up-sampling is effectively a massive increase in the number of times a particular data 'word' is described per second - this increase goes from 44,100 times per second to 192,000 times per second (or from 44.1kHz to 192kHz).

As a result each data 'word' has approximately five times the amount of information or detail and therefore results in much greater accuracy than a normal CD player relying on conventional technology.
[ends]

What can I say? You don't add a single bit of information when you convert from 16/44.1 to 24/192. You distort some, that's true.

[eico1]

Don't forget asynchronous src also has the benefit of reducing input jitter as you are using the local clock in the dac and not a derived one from the input stream. That could be a big difference in some systems.

This benefit would be lost if the src is not done in that dac though.

steve

[woodsyi]

With I2S (this one uses RJ45 which means 8 conductors) sending separate data, I found that cable does matter.  Revelation Audio's prophecy makes a big difference over standard network cable.  Since I2s only requires 3 conductors, I am not sure why this product uses RJ45.  My guess is that they are being redundant or it's for future upgrade path for DVD-A or something else.  I like the data transfer scheme but it probably could use a clock and output stage upgrade down the line. I just have to find time away from this work thing.  If only I could be independently rich .........

[kallsop]

Asynchronous SRC's do not "reduce input jitter" but rather they very efficiently convert it. Time domain jitter to the input side of the ASRC is converted to amplitude jitter on the output side. Reclocking with a very low jitter local clock is the correct solution. If ASRC is needed, reclock first to reduce input side jitter and then use the ASRC.

I believe the discussion on 'interpolation' is talking about such techniques as used in the old Audio Alchemy DTIPro and DTIPro32, and the new Perpetual Technologies P-1A, and copycats. The act of taking a signal and changing to a higher data rate (typically higher but could also be lower) is handled in the traditional way and gains nothing in precision. The output may indeed have more bits of resolution than the input, but those bits are merely necessary to provide for the accurate positioning of the amplitude for the new samples, located at points inbetween the incoming samples.

The 'interpolation', which is called Resolution Enhancement in the P-1A, can operate on the data even when there is no rate conversion. So it is not a result of upsampling (or downsampling) and filtering. So what is it? Without giving away the store, I can tell you that it runs a real time analysis on the incoming audio signal and estimates what is missing. For example, a low level analog sine wave when sampled at 16 bits will tend to be "squarer". This is a direct consequence of the low resolution of the quantized sine wave. Dither and noise shaping will tend to mask the squareness but those are long term statistical remedies and looking over a short sequence will appear to be a noisy square wave. If you knew, a priori, that the signal was a sine wave, it is easy to reconstruct to as many bits of precision as you need. That's where the heuristics come in and the algorithm has to 'best guess' what the missing resolution is, or give up and let pass as is.

Fortunately, music is full of patterns and those are in use everyday in ways that we are all familiar with. Reducing to mp3 or Dolby Digital format is, in part, finding those patterns and using the information to decide what is important and what is not when encoding the signal. Resolution Enhancement works on the decode side. It takes the music, finds the patterns, and reconstructs what was left out by the quantization process. It's not magic, but it is in a sense, creating something from nothing.

In the video arena, line doublers and quadruplers and scalers and other enhancement techniques are used to create something from nothing, and the results are very satisfying to the vast majority of users. So it goes with Resolution Enhancement.  Most listeners perceive improvements in low level detail and imaging. It's something from nothing and it's based on decades of research and techniques commonly used and accepted in many other audio domains.

Bottom line - if you don't perceive a difference and prefer the certainty of the source 16 bits, turn off the processing and enjoy the music.

[Todd Krieger]

I just got a P1a, and I understand upsampling just fine, but don't know what interpolation is.  Maybe I know it by some other word?

Interpolation is something upsampling and oversampling do, in order to execute digital filtering.  An upsampler/oversampler wouldn't be doing its job properly if it didn't interpolate.

If you check out Audio Asylum, I explain why that Perpetual article is technically corrupt , and will only confuse anyone who tries to decipher what it says.

Interpolation is a mathematical calculation of points between the samples.  It does not "guess" the values or plot "bogus" values, as some suggest.

[Todd Krieger]

I don't think it's all marketing gimmick.  

OK, let's separate different issues. You're talking about the benefits of increasing the sampling rate because of the brick wall filters. Can't argue with that. But then simply doubling the sampling rate, from 44.1Khz to 88.2Khz and performing simple interpolation is the way to go. BUT - consider the situation when sampling rate is changed from 44.1 to 96 or any other value which is not multiple of 44.1. We simple can't resample in some clean and unambigious way, interpolation becomes more complicated, in general we introduce some noise. The martetting gimmick is to claim that changing from 44.1 to some higher rate not multiple of 44.1 is somehow improves the sound. Changes, I would say, and possibly degrades.

GIVE THIS MAN A CIGAR!!

Too bad the marketing gimmickry won over sound technical thinking...  For the trend has been asynchronous sample-rate conversion, and personally, I think it is the biggest step backwards in audio since the mainstream demise of the LP.  The positive press for certain DACs that utilize ASRC (like the Benchmark DAC1) notwithstanding.

[Greg Marberry]

Hey Keith,

Thanks so much for chiming in to help clear some of this up and explain the technology from our perspective and our implementation.