[jeremy205100]

 
The megaburrito filter runs at 32 bits -- the input 16 bit signals have zeros added at the input and are all output as original - that is the same 16 bits followed by zeros.  The megaburrito filter also does 32 bit computations on the original samples and outputs 7 interpolated samples at 32 bit resolution, rounded to 20 bits (Yggy), and 18 bits (Gumby).  For the Bimby, there are the same 32 bit calculations resulting in three interpolated samples at 32 bit original resolution, rounded to 16 bits.
 
So for the Bimby only, there are only 16 bits of data output.  The advantage is the extra sample interpolation with the frequency and time domain optimization.

Would it be possible to explain what the megaburrito filter does in the most basic, easy to understand way possible? I have read tons about it on here but don't fully understand what it does. If it is only on the interpolations why not just use the original samples for it? I don't know if I'm understanding that correctly.
 
I was able to read an insightful article somewhere to get the difference between delta sigma and multibit. 

 

[Peaceofmind]

If you imagine music as a sine wave that we're trying to reproduce, the mega burrito filter upsamples (adds data points) to the original data points that exactly traces the sine wave, all other upsample filters can only mathematical approximate place those extra data points as well as approximate the original data points.

 

[madwolfa]

In the graphical terms it's a method of antialiasing called super sampling. When the whole scene is rendered at a higher resolution than the output device's, so you naturally get rid of "jaggies" and the resulting image is smooth, but still sharp (detailed) and not blurred.

Only in case of superburrito filtering, all the original samples are still there, just interpolated (and frequency/time domain optimized) with additional samples. No approximation.

 

[ScottFree]

In the graphical terms it's a method of antialiasing called super sampling. When the whole scene is rendered at a higher resolution than the output device's, so you naturally get rid of "jaggies" and the resulting image is smooth, but still sharp (detailed) and not blurred.

Only in case of superburrito filtering, all the original samples are still there, just interpolated (and frequency/time domain optimized) with additional samples. No approximation.

Personally I think it would be a lot simpler to say "It's Magic". I remember the old saying.

"Any sufficiently advanced technology is indistinguishable from magic"

 

[Ableza]

"If you imagine music as a sine wave" then you are oversimplifying music to be the same as a test tone.  It is nothing like a sine wave, actually. 

 

[wink]

If I imagine music as a sine wave, I get seasick.... 

 

[wink]

 
Would it be possible to explain what the megaburrito filter does in the most basic, easy to understand way possible? I have read tons about it on hear but don't fully understand what it does. If it is only on the interpolations why not just use the original samples for it? I don't know if I'm understanding that correctly.
 
I was able to read an insightful article somewhere to get the difference between delta sigma and multibit. 

There's an old saying:- "Quit while you're behind - when you're in a hole, don't dig "

 

[Ableza]

A Megaburrito is just liked a regular burrito but larger, with extra meat, beans and cheese.

 

[jfoxvol]

Fourier and LaPlace would tend to disagree. It is precisely sine waves. Not just one though.

 

[Ableza]

Fourier and LaPlace would tend to disagree. It is precisely sine waves. Not just one though.

And thus my comment about complexity, with the corollary that all supposed depictions of sine wave sampling are inaccurate and oversimplified to the point of being useless. 

 

[ScottFree]

A Megaburrito is just liked a regular burrito but larger, with extra meat, beans and cheese.

and real big chunks of cheese. None of that processed crap

 

[Ableza]

 
Would it be possible to explain what the megaburrito filter does in the most basic, easy to understand way possible?

I think he just did, in the quote you posted. 

 

[kstuart]

 

 
The megaburrito filter runs at 32 bits -- the input 16 bit signals have zeros added at the input and are all output as original - that is the same 16 bits followed by zeros.  The megaburrito filter also does 32 bit computations on the original samples and outputs 7 interpolated samples at 32 bit resolution, rounded to 20 bits (Yggy), and 18 bits (Gumby).  For the Bimby, there are the same 32 bit calculations resulting in three interpolated samples at 32 bit original resolution, rounded to 16 bits.
 
So for the Bimby only, there are only 16 bits of data output.  The advantage is the extra sample interpolation with the frequency and time domain optimization.

Would it be possible to explain what the megaburrito filter does in the most basic, easy to understand way possible? I have read tons about it on hear but don't fully understand what it does. If it is only on the interpolations why not just use the original samples for it? I don't know if I'm understanding that correctly.
 
I was able to read an insightful article somewhere to get the difference between delta sigma and multibit. 

The process of digital-to-analog conversion creates spurious high frequency noise that must be filtered out.
 
If you convert a 16-bit 44.1khz CD to analog, then to keep the spurious high frequency noise out of the analog output, you need a really steep low-pass filter (if you do not understand "steep low-pass filter" then you need to read Wikipedia on EQ (equalization).)
 
A really steep analog low-pass filter is either really expensive or else it sounds like crap.
 
So, early in the 1980s, "oversampling" was developed.  This means that the audio signal is converted to 88.2khz (" 2X "), 176.4khz (" 4x "), or 352.8khz ( " 8x " ).  This is done, as Mike explained above, by "interpolating", i.e. creating data points in between the existing data points.   For example, if there are two data points that are 17 and 21, then the oversampling filter calculates that halfway in between, it was probably 19 (oversimplified example).
 
Anyway, when the sample rate is 176.4khz, then the analog low-pass filter can have a much gentler curve, and can be done inexpensively in a way that is relatively transparent.
 
But, all the new data that was added by the oversampling filter - is also part of the analog output of the DAC.   So, the fact that Mike's oversampling filter does a better job of predicting the missing data, means that what is produced by the DAC, will sound more like the original analog signal.
 
( The reason that both the multibit DAC and the high quality oversampling are innovations, is that DACs are used in so many new devices - Smartphones, laptops, TVs, Chromecast sticks, etc - that all of the major development has been in the opposite direction - simple and cheap approximations rather than complex, more expensive, but more exact output. )

 

[jcx]

There is another point about realizing “greater resolution”, using high wordlength calculations: dither is theoretically required when reducing bit depths. As in going from Mike's 32 bit megaburritto calculated points to the wordlength of the DACs which is 20 bits in the Yggy's AD5791 or even 16 bit in the cheaper multibits.
 
So did Mike just not mention dither that he uses or is there a way to reconcile megaburitto filter added complexity, theoretical advantages with not using dither to squeeze all the audio resolution possible out of the shorter wordlengths of multibit DACs?

 

[JohnnyCanuck]

  There is another point about realizing “greater resolution”, using high wordlength calculations: dither is theoretically required when reducing bit depths. As in going from Mike's 32 bit megaburritto calculated points to the wordlength of the DACs which is 20 bits in the Yggy's AD5791 or even 16 bit in the cheaper multibits.
 
So did Mike just not mention dither that he uses or is there a way to reconcile megaburitto filter added complexity, theoretical advantages with not using dither to squeeze all the audio resolution possible out of the shorter wordlengths of multibit DACs?

 
Isn't dither added when the analogue signal is digitized?  ie, at the ADC stage?