advertisement


Upsampling

tuga said:
Some comments from Grimm's engineers:

Since almost all DA converters run at a different internal sample rate and bit depth than the original audio format (such as 44.1/16 for CD), digital signals need to be converted from the original format to the ‘native’ format of the actual DA element. This conversion process is called oversampling.
It includes a phase linear filter that preserves the audio band, but removes its ultrasonic ‘mirrors’.
All converter chips contain such filters.
Our investigation has revealed that these filters pose a threat to sound quality when they have insufficient ‘processing power’ to calculate the filter and its requantization at the required precision.
For instance, the filtering is usually executed in several simplified, cascaded steps, which results in a loss of audio quality details.

https://www.grimmaudio.com/publications/the-pure-nyquist-filters-of-the-mu1/
Click to expand...

This is why it surprises me that few DAC manufacturers dont support 44.1k and 48k rates and auto switch as need.

This means if you want to upsample, its integer based.

Bryston's series of DAC's (BDA1,2,3,3.14) all did/do this with defeatable integer upsampling (i.e. no re-rating) which is why I have owned several of these but I cant think of any other company that provides this.

Now of course many will say upsampling from 44.1 to a single 48x internal rate (96,192 etc) is a solved problem but why not give the consumer, as Bryston has, the best of both world.

Peter
 
ricko01 said:
This is why it surprises me that few DAC manufacturers dont support 44.1k and 48k rates and auto switch as need.

This means if you want to upsample, its integer based.

Bryston's series of DAC's (BDA1,2,3,3.14) all did/do this with defeatable integer upsampling (i.e. no re-rating) which is why I have owned several of these but I cant think of any other company that provides this.

Now of course many will say upsampling from 44.1 to a single 48x internal rate (96,192 etc) is a solved problem but why not give the consumer, as Bryston has, the best of both world.
Click to expand...

I don't understand what you mean.
According to Stereophile's measurements of the Bryston DAC 3 the unit performs better in the HF intermodulation test when upsampling is active.
Said DAC uses AKM4490 chips which perform better when fed DSD256, or PCM705.6 although the latter will still need to go through the SDM. Best world is upsampling to DSD256; if you look at the chip diagram feeding it DSD will bypass internal DSP and SDM (the signal goes straight into the D/A stage):

jLpm5yQ.gif
 
gez said:
Has anybody actually ever done blind ABX testing to prove aliasing is audible?
Click to expand...
But of course - and You will have heard it too - sidebands from intermodulation are the most of the 'why' people find differences in, say, the sound of amplifiers.
And that's just 'linear' electronics.
Start hosing such with far out-of-audio-band HF energy on the input -and depending on really, really simple choices about input filtering (- often idiotically-'wide-open', in the world of tweaky hifi), the results cn be from 'forget about it' subtle, to really ...quite gross. For amps that 'measure' 'well'.

It's the whole chain, that matters.
 
tuga said:
I don't understand what you mean.
According to Stereophile's measurements of the Bryston DAC 3 the unit performs better in the HF intermodulation test when upsampling is active.
Said DAC uses AKM4490 chips which perform better when fed DSD256, or PCM705.6 although the latter will still need to go through the SDM. Best world is upsampling to DSD256; if you look at the chip diagram feeding it DSD will bypass internal DSP and SDM (the signal goes straight into the D/A stage):
Click to expand...
In that case upsampling removed the aliases of the 19+20kHz tones because more filtering can fit in the transition band. Upsampling did not reduce the 'pure' intermod products (2f1-f2, 2f2-f1 etc).
 
I have experimented with driving a dome tweeter with 20 KHz and higher at very moderate levels and it could make squeals that my 65 year old ears can hear when I hit the oil can modes
 
davidsrsb said:
I have experimented with driving a dome tweeter with 20 KHz and higher at very moderate levels and it could make squeals that my 65 year old ears can hear when I hit the oil can modes
Click to expand...

Not sure if this is a joke or sarcasm but I really, really strongly doubt that 65yr old ears can hear anywhere near 20 Khz let alone above that
 
davidsrsb said:
I have experimented with driving a dome tweeter with 20 KHz and higher at very moderate levels and it could make squeals that my 65 year old ears can hear when I hit the oil can modes
Click to expand...

This is something I'm quote curious about as it looks like an exampls of what mathematicians call 'period doubling'. Would be interesting to know the mechanism. From the maths if the system has *three* (or more) 'state variables' that can also mean reapeated doublings (i.e. 4, 8, or more subharmonics) and, indeed, the old favourite 'chaotic' behaviour that becomes more like a quasi-noise pattern. Something I wrote about for the old Electronics World many decades ago. Hence potentially a serious flaw that simple tests might not show.
 
vacuum_tubes said:
Not sure if this is a joke or sarcasm but I really, really strongly doubt that 65yr old ears can hear anywhere near 20 Khz let alone above that
Click to expand...

IIUC The point is that the > 20 kHz drive may generate distortion products at *lower* frequencies. e.g. 20kHz drive also outputs a generated output at 10kHz... or 5kHz, or 2.5kHz...

Back when people wore 'Mandelbrot Set' T-shirts that would have been a topic mathematicans were investigating. But people stopped wearing the shirts. :)

Some of the early SACD modulators (encoders) and demodulators (decoders) also did this. One reason Philips quietly changed the design and specs shortly after it was released. FWIW I had some conversations with their engineers about it at the time. The snag being that they refused to release the details of their *actual* DSD modulators. Only had 'examples' in the printed papers that weren't the ones they ended up with.
 
Jim Audiomisc said:
This is something I'm quote curious about as it looks like an exampls of what mathematicians call 'period doubling'. Would be interesting to know the mechanism. From the maths if the system has *three* (or more) 'state variables' that can also mean reapeated doublings (i.e. 4, 8, or more subharmonics) and, indeed, the old favourite 'chaotic' behaviour that becomes more like a quasi-noise pattern. Something I wrote about for the old Electronics World many decades ago. Hence potentially a serious flaw that simple tests might not show.
Click to expand...

Does the mechanism suggested below make sense to you:

Whoa, cool new tweeter from Peerless!!

Even best motors in the business aren't good enough to hide a peak that big. Not that it is a problem - by definition any distortion raised by it is still ultrasonic.
www.diyaudio.com www.diyaudio.com
 
Jim Audiomisc said:
IIUC The point is that the > 20 kHz drive may generate distortion products at *lower* frequencies. e.g. 20kHz drive also outputs a generated output at 10kHz... or 5kHz, or 2.5kHz...

Back when people wore 'Mandelbrot Set' T-shirts that would have been a topic mathematicans were investigating. But people stopped wearing the shirts. :)

Some of the early SACD modulators (encoders) and demodulators (decoders) also did this. One reason Philips quietly changed the design and specs shortly after it was released. FWIW I had some conversations with their engineers about it at the time. The snag being that they refused to release the details of their *actual* DSD modulators. Only had 'examples' in the printed papers that weren't the ones they ended up with.
Click to expand...

The Grimm link I posted earlier above had measurements of their filter showing how the noise shaping pushed the noise higher in frequency than 'typical' filters.
 
Moderate low pass filtering above the audio band is easy to do and can be done without introducing serious group delay. Abusing the amplifier and tweeters with long wave RF is not a sensible move
 
tuga said:
Does the mechanism suggested below make sense to you:

Whoa, cool new tweeter from Peerless!!

Even best motors in the business aren't good enough to hide a peak that big. Not that it is a problem - by definition any distortion raised by it is still ultrasonic.
www.diyaudio.com www.diyaudio.com
Click to expand...

Not yet looked at the reference. But given two input components even 'normal' nonlinearity can generate something at the difference frequency. However this may be academic when the *amount* at or above c20k is 3/4 of SFA"
 
It might be very low energy at 20kHz as an average over a whole track but we hear more like a set of "events" - e.g. cymbal crashes. The 20kHz content can be significant considering just a perceptual "event".
 
Last edited:
davidsrsb said:
You would hope so, but tape bias leakage, aggressive noise shaping and accidental recording of ultrasonic interference in the recording studio do happen
Click to expand...

And of course, unfiltered DSD has an ultrasonic 'noise hill' at HF that hets *bigger* when the recorder audio level is lower.

Yes. Engineering 101 => There are an infinite number of ways for 'engineers' who are just knob-twiddlers to foul things up.

What I confess I hadn't twigged via this thread is that the HF ringing of a tweeter might have the kind of nonlinearity that can period double. That makes it a possible source of problems for the unwary.
 
I'll see if I can find the copy of EW where I did a demo of period doubling effects in simply electronic circuits. Alas, the 'wordprocessor' I used at the time can't run on any modern system as it was so long ago. But I can scan the pages of the issue if I can find it.
 
Jim Audiomisc said:
I'll see if I can find the copy of EW where I did a demo of period doubling effects in simply electronic circuits. Alas, the 'wordprocessor' I used at the time can't run on any modern system as it was so long ago. But I can scan the pages of the issue if I can find it.
Click to expand...
There is quite a lot out there, often very old eg
and
The key seems to be nonlinearity and low damping, which a hard dome tweeter is a good example of
 
There are two significant points here.

The driven system needs a suitable 'state variable' that 'remembers' what has happened in the recent past. That then allows period doublings. However you need three or more state variables to enable a cascade to chaos - i.e. an infinite number of period doublings.

In some (analog) practical cases more state variables exist than shown on the circuit diagram as a result of parasitics and/or imperfections in generator or probes.

The above stems from thinking of the state in terms of a 'location' in a 'space'. A 2D space tends to limit you to ending up having to pass though the same state again later on. 3D or more means you can go on moving around whilst *never* coming back to a previous place.

Hence some reports/papers may confuse a large number of period doublings with chaos. Not the same thing, though, mathematically.

We'd need to know the physical and electronic 'state variables' for the tweeters to decide which case we have.
 
FWIW I just put this up. Been meaning to do it for years but never got the round tuit...
https://jcgl.orpheusweb.co.uk/Chaos/Chaos.html
It shows the article I wrote at the time on period doubling and 'Chaos'. At the time I had in mind creating Gunn or similar oscillators as a way of using them to communicate covertly via 'noise modulations'. i.e. set up a system that used the cascade of period doublings as 'noise' that could use the specific oscillation states as info symbols. But would seem like wideband noise to people who lacked the exact type of oscillator. No idea if it even got used. But worked OK in tests.
 
You must log in or register to reply here.


advertisement


Back
Top