Ciunas Audio
Trade: Ciunas Audio
Jim, are your two PDFs complete articles - the first seems to be curtailed & the second ends in mid sentence?
Noise Shaping
- Thread starter Jim Audiomisc
- Start date
Jim Audiomisc
pfm Member
No, they are partial drafts. I never got further because I wanted to find out if the DSD modulators actually used behaved in similar ways or not. But I could never find out their details to model them. Details buried under a layer of 'secret sauce'.
So what I did was useful as a demo that oddly 'non random' behaviours will crop up when using DSD. But by the time you get to high-orders it is almost impossible to predict them all by simple analysis. You have to 'search for them' by trying out running the system to encode or decode DSD. Which is a long task and will be easy to miss something.
In practice I moved onto something more interesting. The use of simple electronic devices like diodes, etc, to both generate and detect sequences that mimic noise but are deterministic. i.e. The source looks like a 'noise diode' but it isn't,and the output will 'lock' another similar diode's behaviour to it. Useful for covert comms, etc...
But not audio. Day job work when I was no longer in audio.
So what I did was useful as a demo that oddly 'non random' behaviours will crop up when using DSD. But by the time you get to high-orders it is almost impossible to predict them all by simple analysis. You have to 'search for them' by trying out running the system to encode or decode DSD. Which is a long task and will be easy to miss something.
In practice I moved onto something more interesting. The use of simple electronic devices like diodes, etc, to both generate and detect sequences that mimic noise but are deterministic. i.e. The source looks like a 'noise diode' but it isn't,and the output will 'lock' another similar diode's behaviour to it. Useful for covert comms, etc...
But not audio. Day job work when I was no longer in audio.
Ciunas Audio
Trade: Ciunas Audio
That explains it, thanks.
Very interesting articles, particularly the second article in which you computed the 'hash/noise' inherent in the 5th order DSD modulator published by Sony & Phillips
Please explain Fig 3 some more, "Hash/noise levels versus length of duration examined"
I don't quite follow the explanatory text regarding the meaning of X & Y axes "The vertical scale shows power relative to the SACD 0dB level in terms of what total 'noise' would be implied across the 0 - 20 kHz range if the computed hash level was replicated across that band. The horizontal scale represents the length (duration) of the waveform's DSD stream over which the hash level was computed."
Does it mean, for instance that all the frequencies simulated, 0.5KHz, 1.2KHz, 6.5KHz, 25 & 100KHz are, for some time duration, showing artifacts at -60dB approx down from the main 1KHz signal - except for 100KHz which is only -20dB to -40dB approx down?
I'm not sure what the time duration signifies?
Very interesting articles, particularly the second article in which you computed the 'hash/noise' inherent in the 5th order DSD modulator published by Sony & Phillips
Please explain Fig 3 some more, "Hash/noise levels versus length of duration examined"
I don't quite follow the explanatory text regarding the meaning of X & Y axes "The vertical scale shows power relative to the SACD 0dB level in terms of what total 'noise' would be implied across the 0 - 20 kHz range if the computed hash level was replicated across that band. The horizontal scale represents the length (duration) of the waveform's DSD stream over which the hash level was computed."
Does it mean, for instance that all the frequencies simulated, 0.5KHz, 1.2KHz, 6.5KHz, 25 & 100KHz are, for some time duration, showing artifacts at -60dB approx down from the main 1KHz signal - except for 100KHz which is only -20dB to -40dB approx down?
I'm not sure what the time duration signifies?
Jim Audiomisc
pfm Member
In that case I didn't do FFTs but plain single-frequency transforms and extended their durations. This was to get coherent addition of the result. In effect, I let the FTs run for longer and longer, noting the result every now and then as I did this.
Thus the scale shows the various lengths of the time-spans of the transforms used to determine the level at a given frequency. Incoherent addition of powers might have missed the behaviour which implies that some sections of the 'noise' tend to 'cancel out' the previous noise by having a spectrum which is of the same power level but opposite phase. Thus showing up deterministic patterns a standard FFT power observation, chunk by chunk. wouldn't reveal.
The *actual* power over a shorter time may not dip. So not be obvious unless you do this to show the coherent relationship between different time-periods. A high resolution (i.e. long time span) FFT might show it as a 'comb' of components though. That can be a characteristic of 'semi-chaotic' oscillations or 'orbits'. I'll see if I can find an example of that to make it clearer, but from a different context.
Thus the scale shows the various lengths of the time-spans of the transforms used to determine the level at a given frequency. Incoherent addition of powers might have missed the behaviour which implies that some sections of the 'noise' tend to 'cancel out' the previous noise by having a spectrum which is of the same power level but opposite phase. Thus showing up deterministic patterns a standard FFT power observation, chunk by chunk. wouldn't reveal.
The *actual* power over a shorter time may not dip. So not be obvious unless you do this to show the coherent relationship between different time-periods. A high resolution (i.e. long time span) FFT might show it as a 'comb' of components though. That can be a characteristic of 'semi-chaotic' oscillations or 'orbits'. I'll see if I can find an example of that to make it clearer, but from a different context.
Jim Audiomisc
pfm Member
OK, this *isn't* audio, but the output of a diode in the 95GHz region. Taken sometime around 1990 IIRC.
http://jcgl.orpheusweb.co.uk/temp/spectsc.jpeg
Photo off the screen of an HP specan having used a mixer to downconvert to a band the specan can see.
The diode in question is changing frequency in a complicated but repetitive way with a bandwidth of the order of a GHz. I can't recall if it was a Gunn diode or a multiple-barrier one, but the principle is the same. In some ways the result looks like 'noise' and can be used as a noise source. But when examined appropriately shows repetitive patterning.
http://jcgl.orpheusweb.co.uk/temp/spectsc.jpeg
Photo off the screen of an HP specan having used a mixer to downconvert to a band the specan can see.
The diode in question is changing frequency in a complicated but repetitive way with a bandwidth of the order of a GHz. I can't recall if it was a Gunn diode or a multiple-barrier one, but the principle is the same. In some ways the result looks like 'noise' and can be used as a noise source. But when examined appropriately shows repetitive patterning.
darrenyeats
pfm Member
Linear phase is the only way to preserve phase in a multiple ADC/DAC loop. BUT ...
If we take filters in the audible band, as used in music production, the comments I could find previously from engineers say they prefer intermediate phase. Apparently pre-ringing in the transition band doesn't sound good. Since in this application everyone agrees the transition band is audible, there was no religious argument when they wrote their preference. So I don't take it as given that linear phase is "better", it depends what you mean.
The down-side of intermediate or minimum phase is that these filters can combine with each other or gear in the chain leading to variable outcomes, just as you wrote. But that isn't always a problem - it wasn't a problem for these engineers when they formed their preference using their gear.
Arguments occur for CD because for the skeptical camp the anti-imaging filter is (conventionally) ultrasonic, and for the hi-res believer camp the CD filter is necessarily very steep i.e. ringing and in area often with energy present.
But then images are ultrasonic too! Yet DACs use anti-imaging filters (because we don't have zero distortion tweeters or amps, so it can leak down).
Last edited:
Jim Audiomisc
pfm Member
In principle the questions of pre-ringing level, filter (amplitude/frequency) shape, and aliasing levels are all adjustable with some independence. The snag is that in reality all digitial filters have a finite number of states/stages/etc. So some trade-offs become necessary. All a question of where you push them and what seems 'best'.
Personally I'm quite happy with reasonably made sinc-type filter for replay and leave it to the people making recordings to decide what they prefer for a given recording. Beyond that, I regard replay filter choice as a form of 'tone control adjustment'. Simply because what you get comes from the combination of everything in the chain - including variable parts you weren't involved in choosing that differ from one recording to the next.
Personally I'm quite happy with reasonably made sinc-type filter for replay and leave it to the people making recordings to decide what they prefer for a given recording. Beyond that, I regard replay filter choice as a form of 'tone control adjustment'. Simply because what you get comes from the combination of everything in the chain - including variable parts you weren't involved in choosing that differ from one recording to the next.
darrenyeats
pfm Member
That's true of the final mix, but if the final mix is hi res (or analogue for an old recording) then what I'm talking about - the choice of CD rate anti-aliasing filter (or for analogue ADC filter) and its alleged impact upon the sound - comes later, in reality often done even by a separate organisation i.e. the mastering engineers.
Jim Audiomisc
pfm Member
True enough. And also IMHO the reason so many 'remasters' are fouled up by clipping and/or insane amounts of level compression, etc.Problem is that the end-user is then stuck with this. You can use your replay filters to try and correct for variations of the inband linear response. But aliases, clipping, etc, added by a 'remastering guru' aren't generally correctable or removable on the end-product. Best you can do is apply a 'tone control' to try and make some of the damage less distracting if you're lucky.
darrenyeats
pfm Member
For sure. I'd say most "remasters" I've heard have been a step backward!
PS: Most aspects of a filter could technically be called a tone control (though changing these would change more than frequency response) ... but phase type is different, since it does not change the frequency response.
PS: Most aspects of a filter could technically be called a tone control (though changing these would change more than frequency response) ... but phase type is different, since it does not change the frequency response.
Last edited: