I was watching a youtube video about the PS Audio Directstream DAC and how it improves the sound quality of CD
by converting to DSD. I think the idea is that when a CD is created its sampled at 44.1khz (highest frequency
captured is 22khz) and this causes problems because presumably any higher frequencies leave unresolved 'traces' at 44khz
(distortion). By interpolating the signal we can upsample to 176khz (DSD) and resolve out what would otherwise be noise at these higher frequencies. Does anyone know if this is about right or completely wrong?
Wrong, all modern DACs on playback internally up-sampling to remove the aliasing issue. Most are delta sigma which are 1 bit dacs, so nothing the world has not been doing for the last 30 years...
Does anyone know if this is about right or completely wrong?
It is nonsense.
If you're looking for such content, this Youtube channel (https://www.youtube.com/comedycentral) is much better.
I was under the impression that the Directstream DAC was 'high end' (some $5000) and converting the PCM to DSD was unique.
So you saying that most DAC's ($500) will be converting PCM to a form of DSD (1 bit delta sigma) and then to analogue.
Re: It is nonsense:
I thought it was common knowledge that PCM 44.1khz cut off causes problems - the analogy is echoing off a wall. Presumably
there would'nt be any point upsampling unless it allows you to undo the echoing.
Just checked the youtube channel - its not available to the UK.
I thought it was common knowledge that PCM 44.1khz cut off causes problems - the analogy is echoing off a wall.
I think you are referring to aliasing, the audio signal is mirrored at half fs (not 44 but 22).
That is indeed the case and to filter it out you need a very steep filter (brickwall) because you want to preserve the audible range at one hand but all signal must be gone before you hit ½ fs.
I believe it was the second gen of CD player (late 80’s) that applied 4 times oversampling and later 8 times oversampling. Of course we still keep the alias (in digital everything remains the same) but
8x22 = 176 kHz for the first alias to appear. No need to brickwall anymore.
Please observe that this has nothing to do with DSD
Maybe this link to my website is of use: http://www.thewelltemperedcomputer.com/KB/Sampling.htm
PS Audio Directstream DAC
This US $6,899 DAC can be found here: https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-ps-audio-perfectwave-directstream-dac.9100/
You might compare to e.g. https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-rme-adi-2-dac.2582/
Gives you a taste of the performance of a PS audio product and a $999 RME
I think this video by Monty Montgomery is very helpful about how digital signal processing works. https://www.youtube.com/watch?v=cIQ9IXSUzuM
Aliasing happens during sampling/digitizing (recording) or when downsampling, not during digital-to-analog conversion (playback).
I once had a soundcard with no filtering. For some reason (I don't remember why) I hooked-up an oscilloscope and I was surprised to see a "clean" stair-stepped" waveform! I hadn't noticed anything "wrong", but the harmonics are at the sampling frequency or higher so of course I couldn't hear them. (Plus, they were probably filtered-out mechanically by the speakers and maybe the amplifier was band-limited too.)
With modern DAC chips it's cheap and easy to build a DAC that's better than human hearing. If you're not getting excessive noise from your soundcard, your regular soundcard/soundchip is probably better than human hearing.
Aliasing happens during sampling/digitizing (recording) or when downsampling, not during digital-to-analog conversion (playback).
It also happens during digital-to-analog conversion, but it's typically called imaging rather than aliasing. You still need a low-pass filter after the DAC to remove all signal above the Nyquist limit (half the sampling frequency).
Re: Monty Montgomery video:
Had a scan through didn't seem to find anything on AtoD aliasing or as above DtoA imaging.
I think the original PS Audio video was referring to DtoA 'imaging' since it also mentioned the low pass filter.
I think the original PS Audio video was referring to DtoA 'imaging' since it also mentioned the low pass filter.
As spoon said in the 2nd message,
all modern DACs on playback internally up-sampling to remove the aliasing issue.
I was under the impression that the Directstream DAC was 'high end' (some $5000) and converting the PCM to DSD was unique.
So you saying that most DAC's ($500) will be converting PCM to a form of DSD (1 bit delta sigma) and then to analogue.
20 or so years ago, yes that was very common. Most are actually multibit the last 15 years, since fabrication improved and it got a lot easier to make better converters. I think most (all?) dedicated DSD devices no longer use 1 bit in practice anymore either. If you search on this forum, there were many topics about this in the early 2000s.
Re: It is nonsense:
I thought it was common knowledge that PCM 44.1khz cut off causes problems - the analogy is echoing off a wall. Presumably
there would'nt be any point upsampling unless it allows you to undo the echoing.
What are you asking?
I thought it was common knowledge that PCM 44.1khz cut off causes problems - the analogy is echoing off a wall.
Indeed it is common fodder for trolling technical forums. Now remember to feign all insulted as a "newb" with 8 posts, didn't read TOS rules, etc, etc.
We've seen it all before, many times, no worries. Have fun.
After reading Roseval's "An introduction to computer audio" - this makes what was in the
PS Audio upsampling and directstream DAC videos a bit clearer.
From Roseval's article:
Inherent in the mathematics involved in DtoA conversion higher frequency multiples are created.
At 44.1khz these aliases start at 22.05khz
You can remove these with a low pass filter. A filter at 20khz will remove everything before the alias start.
But this introduces phase distortion and pre-ringing artifacts.
By oversampling, say 8 times, the mathematics in the DtoA conversion creates the aliases at
330khz. I was expecting this to be half the new sampling rate - but anyway the idea is that you have changed
the maths.
The "echoing off a wall" analogy I referred to must be referring to the "pre-ringing artifacts" you
get from a low pass filter cutting everything off at 20khz. I can see the analogy but it could do with more
detail.
The Fear Of Ringing is the bigfoot for audio peddlers for maybe 20 years now. It also can be shown in funny pictures. Terrific!
Paul,
maybe have a look here: https://www.audiosciencereview.com/forum/index.php?threads/review-and-measurements-of-ps-audio-perfectwave-directstream-dac.9100/
Probably learns you more about the peformance of PS audio DACs than a Youtube video (a promo by PS audio?)
The "echoing off a wall" analogy I referred to must be referring to the "pre-ringing artifacts" you
get
There will be such "referral" ambiguity when making stuff up along the trip.
Thanks for the link Roseval I've had a look but its quite technical so will take
a bit more reading.
On the "echoing off a wall" reference - the source of this was the following
comment on the video "How does upsampling increase information?"
"A good way to explain the filter problem is to consider a square ledge in a tank of water - a wave will reflect off the sharp edge.
In music you get the artifacts as ring harmonics. So if you have a soft filter the wave does not reflect and you don't get the wave
"bouncing" of the band pass filter."
Why is that a problem?
On the "echoing off a wall" reference - the source of this was the following
comment on the video "How does upsampling increase information?"
"A good way to explain the filter problem is to consider a square ledge in a tank of water - a wave will reflect off the sharp edge.
In music you get the artifacts as ring harmonics. So if you have a soft filter the wave does not reflect and you don't get the wave
"bouncing" of the band pass filter."
That is a bad explanation because it does not make sense and doesn't explain anything. It is just someone telling you their opinion that a "soft filter" is better for some reason.
It's common knowledge.
Overheard some fellas talking about it at the bait shop
Does anyone know if this is about right or completely wrong?
With the right signal it is equally as easy to demonstrate ringing at 176-bit DSD. This is the wrong line of thinking.
Is the result audible?
https://en.m.wikipedia.org/wiki/Law_of_excluded_middle
The analogy "a square ledge in a tank of water - a wave will reflect off the sharp edge" I'd assumed this was a good analogy
to start with for explaining "pre-ringing artifacts" you get from a low pass filter (cutting everything off at 20khz). But, in practice,
not sure how soundwaves generated in this way will bounce off something. The quote then refers to "the wave "bouncing" off the band pass filter" - don't get this as it sounds like something that is happening to the generated analogue signal inside the DAC rather the soundwaves in the room.
So if someone has a model/explaination for what the soundwaves leaving the loudspeaker are doing to generate pre-ring artifacts
please post.
For clarity this is of interest rather 'baiting' as suggested by ajinfla.
So if someone has a model/explaination for what the soundwaves leaving the loudspeaker are doing to generate pre-ring artifacts please post.
Soundwaves leaving the loudspeaker do not generate pre-ring artifacts.
@Paul77 you are mixing too many things.
* Ringing: Some information on wikipedia, although it might be a bit dense https://en.wikipedia.org/wiki/Ringing_artifacts
The basics: Ringing means to oscillate. Instead of having a steady signal, the signal oscillates at sharp changes.
pre-ringing and post-ringing: Ringing is seen before and after the cutoff place. Having both is just natural, but depending on the type of windowing and filter response type, one can avoid pre-ringing.
Aliasing: A generic article also on wikipedia. The textual explanations are a bit complicated, but the image can be understood: https://en.wikipedia.org/wiki/Aliasing
The basics: Aliasing means getting a modified signal that contains something mirrored from another place.
Here is where your example of a tank of water could work, but only if you understand what we are talking about:
Lets think about a tank of water ( or a swimming pool, if you're more famirialized) where the water is calm.
You drop something into it, and water waves get generated from that point moving away from this point.
When one of this waves reaches the wall, it cannot continue advancing and bounces back.
Now, with that idea in mind, let's take a frequency spectrum of an audio signal.
You would have some frequencies in there, let's think about a "two-mountains and a lake" graphic = ^o^
If we change the frequency of this, so that the second mountain is after the cutoff frequency, something like this: ^o|^ and no filter is done when doing this operation, the signal will actually become this: ^ô (i.e. the second mountain will mirror and mix with the part of the signal that was there. in this case, our "lake").
Also, aliasing can happen on the opposite direction, which is what this "CD to analog" thread is about.
A sampled signal when converted from digital to analog, needs a lowpass filter (actually, this is called a reconstruction filter) in order to remove the mirroring caused by straight conversion from discrete values to continuous values.
This is what the image called "spectral folding, caused by sampling" tries to show in that Wikipedia Aliasing article.
As you were told, this might have been a problem with cheap computer equipment on the 90s, but even then, it was perfectly possible to not have such a problem.
Problem happens when the output jumps from one value to another, and stays at that value until the next sample come in. A reconstruction filter properly interpolates from the first value to the next, providing a continuous signal.
To think about this problem, one can think about a square wave.
A square wave is a signal that has a fundamental frequency, and then it has many harmonics that are multiples of that frequency. If you filter all harmonics, you get a sine wave. If you have just the fundamental and the first harmonic, then the sine wave starts to deformate, but still resembles a sine wave.
The more harmonics that you add, the more it changes from sine wave to square wave.
Converting a digitized (sampled) signal to an analog (continuous) signal resembles this sine and square problem.
If you don't use the reconstruction filter, the analog signal between the digitized samples resemble a square wave, and as such, has harmonics.
But in this case, you don't have a fundamental frequency, but instead a full band signal.
Oh... and indeed, this has nothing to do with speakers. Speakers might change the shape of the spectrum (have more bass or miss some mids or whatever..), but do not generate frequencies, although depending on the materials, they can ressonate and cause distortions.
I don’t know that lakes and mountains is a good analogy, in fact I consider it to be a very bad one, ignoring the fact that they are coming on the heels of a giant misconception.
Well, to be fair, if an audiophile sees a CD and soundwaves are generated, pass through the ear canal and into the cranial cavity, there will certainly be some wall bounce echo, which could then manifest as "ringing" artifacts.
So not too far fetched, nor any iffy analogies needed.
That cavity would need to be empty.
I believe that would be the point.
@greynol : I just used the image of two mountains and a lake in order to easily write with text what I wanted to represent as a frequency plot.
I am not sure if you mean that it is difficult to see it, or you mean that the whole "bounce back" or mirror concept is not applicable to what we are talking about.
Let's better get an image from our well-known resampler tests page (I've specifically chosen images that help see the effect. I'm not making audio judgements):
Here we can see aliasing: https://src.infinitewave.ca/images/Sweep/XRECODE3_swr.png
Here there isn't: https://src.infinitewave.ca/images/Sweep/XRECODE3_sox.png
On both cases, the signal would increment above the top if it could.
In the first image, we can see the bouncing (mirroring) of the signal that should have gone above, but instead goes down and even bounces again when reaching the bottom.
That's what I was talking about.
Sure, but there absolutely is no bouncing of the sort going on.
Remember this guy is talking about waves in water that meet the edge of a surface. This is nothing like ringing, aliasing or mirroring. Trying to shoe-horn in a water analogy is only going to reinforce the idea of there being a relationship between what are two very different phenomena.
The standing wave in an empty cranial cavity isn’t a good analogy, but the quip was never meant to be. ;)
If the user is confusing sound waves with the effects that ringing or aliasing are having on the frequency domain then yes, any kind of explanation about bouncing will be misleading. I expected to correct that when I explicitly talked about frequency domain.
If we talk on the amplitude domain (i.e.. sound waves), then there's no good explanation to be told.
Ringing/resonance is pretty easy to understand. A bell rings or a guitar string rings/resonates, or you'll hear the effects of ringing/resonance if you speak into a drinking glass or into the sound hole of a guitar. Those lousy-sounding "one note" subwoofers you sometimes hear in cars are resonating/ringing. I guess you might get ringing if you "play around" with a parametric equalizer. But of course you can't hear ultrasonic ringing.
For aliasing I can't think of an analogy and you'd probably have to make an unfiltered down-sampler or build an unfiltered ADC to demonstrate it. I've never actually heard aliasing so I can only imagine what it might sound like .
https://community.sw.siemens.com/s/article/the-gibbs-phenomenon
The example is a bit cherry picked here as the sound of wind chimes themselves end up sounding like something else entirely, so it's extreme in that regard. It's very difficult to explain what aliasing is in an analogy but it's effects are quite demonstrable. There isn't typically much outside of the audible spectrum other than noise and properly designed ADC will filter it all out.