Forums

cool paper on dithering

Started by John Larkin December 14, 2016
https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja

or maybe

http://tinyurl.com/zhlgolz


This has some really interesting ideas about extending DAC resolution
with scrambled thermometer codes and such. Some class-D audio stuff,
too.

I've got to take time to study this; there are a lot of ideas here.

I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform
generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in
stock, DAC2904s. They have speed to burn, so I should be able to
dither them up to 16 bits. I can't do simple (wiggle the LSB)
delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither
has to be big enough to smooth out DNL errors too. This will involve
noise shaping of the dither pattern, sort of like a high-order
delta-sigma, where you push the dither spectrum up so the final analog
5 MHz lowpass filter kills the noise you add.

I'm playing with that in LT Spice, which is admittedly a clumsy tool
for this sort of thing.

I have a huge respect, and some envy, for people who are really good
with this signals+systems stuff. I do have a copy of "Signals and
Systems for Dummies."





-- 

John Larkin         Highland Technology, Inc
picosecond timing   precision measurement 

jlarkin att highlandtechnology dott com
http://www.highlandtechnology.com

On 12/14/2016 01:23 PM, John Larkin wrote:
> https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja > > or maybe > > http://tinyurl.com/zhlgolz > > > This has some really interesting ideas about extending DAC resolution > with scrambled thermometer codes and such. Some class-D audio stuff, > too. > > I've got to take time to study this; there are a lot of ideas here. > > I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > stock, DAC2904s. They have speed to burn, so I should be able to > dither them up to 16 bits. I can't do simple (wiggle the LSB) > delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither > has to be big enough to smooth out DNL errors too. This will involve > noise shaping of the dither pattern, sort of like a high-order > delta-sigma, where you push the dither spectrum up so the final analog > 5 MHz lowpass filter kills the noise you add. > > I'm playing with that in LT Spice, which is admittedly a clumsy tool > for this sort of thing.
It sounds like one of those problems one could apply math to or somethin'!
> I have a huge respect, and some envy, for people who are really good > with this signals+systems stuff. I do have a copy of "Signals and > Systems for Dummies."
I once implemented a digital sine wave generator based off the feedback state-variable filter topology on an 8 bit microcontroller. I stole some 6502 assembly and ported it. I was very proud of myself!
On Wed, 14 Dec 2016 13:51:38 -0500, bitrex
<bitrex@de.lete.earthlink.net> wrote:

>On 12/14/2016 01:23 PM, John Larkin wrote: >> https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja >> >> or maybe >> >> http://tinyurl.com/zhlgolz >> >> >> This has some really interesting ideas about extending DAC resolution >> with scrambled thermometer codes and such. Some class-D audio stuff, >> too. >> >> I've got to take time to study this; there are a lot of ideas here. >> >> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform >> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in >> stock, DAC2904s. They have speed to burn, so I should be able to >> dither them up to 16 bits. I can't do simple (wiggle the LSB) >> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither >> has to be big enough to smooth out DNL errors too. This will involve >> noise shaping of the dither pattern, sort of like a high-order >> delta-sigma, where you push the dither spectrum up so the final analog >> 5 MHz lowpass filter kills the noise you add. >> >> I'm playing with that in LT Spice, which is admittedly a clumsy tool >> for this sort of thing. > >It sounds like one of those problems one could apply math to or somethin'!
Sure, but covering pages with math doesn't train my instincts much. I think a person can push a lot of symbolic math around and not actually understand what's happening. (And can make a mistake, and get crazy results without noticing. I've seen that happen.) And playing with Spice is more fun. This is fun: make a 4-bit binary counter and dump that into a DAC. Then reverse the bits into the DAC and repeat. The amplitude histograms are identical, but the spectra are very different. The "noise" gets pushed up. There must be an optimum remap of the 4 bits to push the spectrum up. Bit reversal may be it. -- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On Wed, 14 Dec 2016 10:23:37 -0800, John Larkin wrote:

> https://www.google.com/url?
sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https %3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams% 2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja
> > or maybe > > http://tinyurl.com/zhlgolz > > > This has some really interesting ideas about extending DAC resolution > with scrambled thermometer codes and such. Some class-D audio stuff, > too. > > I've got to take time to study this; there are a lot of ideas here. > > I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > stock, DAC2904s. They have speed to burn, so I should be able to dither > them up to 16 bits. I can't do simple (wiggle the LSB) > delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither > has to be big enough to smooth out DNL errors too.
At second blush (first blush saying "cool!") I wonder if that's going to help -- AFAIK DNL comes from mismatches between legs of the R-2R ladder, and any given leg can be mismatched from another. I'd start by figuring out how to model that, then test it (or work it out on paper), and then finally I'd make sure to test the snot out of it in hardware before I decided that I'd actually gotten something good to 16 bits.
> This will involve > noise shaping of the dither pattern, sort of like a high-order > delta-sigma, where you push the dither spectrum up so the final analog 5 > MHz lowpass filter kills the noise you add. > > I'm playing with that in LT Spice, which is admittedly a clumsy tool for > this sort of thing.
Scilab (or Matlab) would probably be better, just because there's no ceiling on the degree of hifalutin' math you can do.
> I have a huge respect, and some envy, for people who are really good > with this signals+systems stuff. I do have a copy of "Signals and > Systems for Dummies."
It's just math, with enough reality thrown in that if you're a head-in- the-clouds egghead you'll never amount to anything. -- Tim Wescott Wescott Design Services http://www.wescottdesign.com I'm looking for work -- see my website!
On 12/14/2016 02:23 PM, John Larkin wrote:
> On Wed, 14 Dec 2016 13:51:38 -0500, bitrex > <bitrex@de.lete.earthlink.net> wrote: > >> On 12/14/2016 01:23 PM, John Larkin wrote: >>> https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja >>> >>> or maybe >>> >>> http://tinyurl.com/zhlgolz >>> >>> >>> This has some really interesting ideas about extending DAC resolution >>> with scrambled thermometer codes and such. Some class-D audio stuff, >>> too. >>> >>> I've got to take time to study this; there are a lot of ideas here. >>> >>> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform >>> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in >>> stock, DAC2904s. They have speed to burn, so I should be able to >>> dither them up to 16 bits. I can't do simple (wiggle the LSB) >>> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither >>> has to be big enough to smooth out DNL errors too. This will involve >>> noise shaping of the dither pattern, sort of like a high-order >>> delta-sigma, where you push the dither spectrum up so the final analog >>> 5 MHz lowpass filter kills the noise you add. >>> >>> I'm playing with that in LT Spice, which is admittedly a clumsy tool >>> for this sort of thing. >> >> It sounds like one of those problems one could apply math to or somethin'! > > Sure, but covering pages with math doesn't train my instincts much. I > think a person can push a lot of symbolic math around and not actually > understand what's happening. (And can make a mistake, and get crazy > results without noticing. I've seen that happen.) > > And playing with Spice is more fun. > > This is fun: make a 4-bit binary counter and dump that into a DAC. > Then reverse the bits into the DAC and repeat. The amplitude > histograms are identical, but the spectra are very different. The > "noise" gets pushed up. There must be an optimum remap of the 4 bits > to push the spectrum up. Bit reversal may be it. >
Is the "sampling period" an integer, so that the aliasing from a non-band limited sawtooth wave only folds back onto harmonic components? The sawtooth wave has uniformly-distributed quantization error, so shuffling the bits in that way might be optimal for that particular waveform and bit depth. The quantization error for most other waves, including sines, is signal-dependent, so it probably wouldn't be for any arbitrary waveform, though clearly it approaches uniform as the bit depth is increased.
On Wed, 14 Dec 2016 15:06:31 -0600, Tim Wescott
<seemywebsite@myfooter.really> wrote:

>On Wed, 14 Dec 2016 10:23:37 -0800, John Larkin wrote: > >> https://www.google.com/url? >sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https >%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams% >2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja >> >> or maybe >> >> http://tinyurl.com/zhlgolz >> >> >> This has some really interesting ideas about extending DAC resolution >> with scrambled thermometer codes and such. Some class-D audio stuff, >> too. >> >> I've got to take time to study this; there are a lot of ideas here. >> >> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform >> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in >> stock, DAC2904s. They have speed to burn, so I should be able to dither >> them up to 16 bits. I can't do simple (wiggle the LSB) >> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither >> has to be big enough to smooth out DNL errors too. > >At second blush (first blush saying "cool!") I wonder if that's going to >help -- AFAIK DNL comes from mismatches between legs of the R-2R ladder, >and any given leg can be mismatched from another.
I figure that I need to dither 2 or 3 bits at the 14-bit DAC level, to smear out the DNL, mostly. And some sub-bits to get from 14 to 16 bit resolution. Our spec will claim 16 bit resolution, but certainly not 16 bit accuracy. The real benefit to 16 bits is to avoid gross steps when making small signals. Dithering can really help there. There's no reason we shouldn't try to dither out to 18 bits... it's easy in an FPGA. These fast dacs tend to be hybrid architectures, some mix of R-2R and thermometer codes at the low end. So the dithering will be a little weird.
> >I'd start by figuring out how to model that, then test it (or work it out >on paper), and then finally I'd make sure to test the snot out of it in >hardware before I decided that I'd actually gotten something good to 16 >bits. > >> This will involve >> noise shaping of the dither pattern, sort of like a high-order >> delta-sigma, where you push the dither spectrum up so the final analog 5 >> MHz lowpass filter kills the noise you add. >> >> I'm playing with that in LT Spice, which is admittedly a clumsy tool for >> this sort of thing. > >Scilab (or Matlab) would probably be better, just because there's no >ceiling on the degree of hifalutin' math you can do. > >> I have a huge respect, and some envy, for people who are really good >> with this signals+systems stuff. I do have a copy of "Signals and >> Systems for Dummies." > >It's just math, with enough reality thrown in that if you're a head-in- >the-clouds egghead you'll never amount to anything.
-- John Larkin Highland Technology, Inc picosecond timing precision measurement jlarkin att highlandtechnology dott com http://www.highlandtechnology.com
On Wednesday, December 14, 2016 at 4:35:28 PM UTC-5, John Larkin wrote:
> On Wed, 14 Dec 2016 15:06:31 -0600, Tim Wescott > <seemywebsite@myfooter.really> wrote: > > >On Wed, 14 Dec 2016 10:23:37 -0800, John Larkin wrote: > > > >> https://www.google.com/url? > >sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https > >%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams% > >2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja > >> > >> or maybe > >> > >> http://tinyurl.com/zhlgolz > >> > >> > >> This has some really interesting ideas about extending DAC resolution > >> with scrambled thermometer codes and such. Some class-D audio stuff, > >> too. > >> > >> I've got to take time to study this; there are a lot of ideas here. > >> > >> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > >> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > >> stock, DAC2904s. They have speed to burn, so I should be able to dither > >> them up to 16 bits. I can't do simple (wiggle the LSB) > >> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither > >> has to be big enough to smooth out DNL errors too. > > > >At second blush (first blush saying "cool!") I wonder if that's going to > >help -- AFAIK DNL comes from mismatches between legs of the R-2R ladder, > >and any given leg can be mismatched from another. > > I figure that I need to dither 2 or 3 bits at the 14-bit DAC level, to > smear out the DNL, mostly. And some sub-bits to get from 14 to 16 bit > resolution. > > Our spec will claim 16 bit resolution, but certainly not 16 bit > accuracy. The real benefit to 16 bits is to avoid gross steps when > making small signals. Dithering can really help there. There's no > reason we shouldn't try to dither out to 18 bits... it's easy in an > FPGA. > > These fast dacs tend to be hybrid architectures, some mix of R-2R and > thermometer codes at the low end. So the dithering will be a little > weird. > > > > > >I'd start by figuring out how to model that, then test it (or work it out > >on paper), and then finally I'd make sure to test the snot out of it in > >hardware before I decided that I'd actually gotten something good to 16 > >bits. > > > >> This will involve > >> noise shaping of the dither pattern, sort of like a high-order > >> delta-sigma, where you push the dither spectrum up so the final analog 5 > >> MHz lowpass filter kills the noise you add. > >> >
nice discussion re dither here INTRODUCTION TO DIGITAL AUDIO by John Watkinson Focal Press and The Art of Digital Audio By John Watkinson m
On Wednesday, December 14, 2016 at 2:24:08 PM UTC-5, John Larkin wrote:
> On Wed, 14 Dec 2016 13:51:38 -0500, bitrex > <bitrex@de.lete.earthlink.net> wrote: > > >On 12/14/2016 01:23 PM, John Larkin wrote: > >> https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja > >> > >> or maybe > >> > >> http://tinyurl.com/zhlgolz > >> > >> > >> This has some really interesting ideas about extending DAC resolution > >> with scrambled thermometer codes and such. Some class-D audio stuff, > >> too. > >> > >> I've got to take time to study this; there are a lot of ideas here. > >> > >> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > >> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > >> stock, DAC2904s. They have speed to burn, so I should be able to > >> dither them up to 16 bits. I can't do simple (wiggle the LSB) > >> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither > >> has to be big enough to smooth out DNL errors too. This will involve > >> noise shaping of the dither pattern, sort of like a high-order > >> delta-sigma, where you push the dither spectrum up so the final analog > >> 5 MHz lowpass filter kills the noise you add. > >> > >> I'm playing with that in LT Spice, which is admittedly a clumsy tool > >> for this sort of thing. > > > >It sounds like one of those problems one could apply math to or somethin'! > > Sure, but covering pages with math doesn't train my instincts much. I > think a person can push a lot of symbolic math around and not actually > understand what's happening. (And can make a mistake, and get crazy > results without noticing. I've seen that happen.)
Yeah I think everyone's brain works a bit different, (some more different than others. :^) For me I've got to have a picture of what's going on, then I can try and put/match the equations to my pictures. When things don't match, there's most often a problem with my pictures.... For years I imagined E-M waves being 90 deg out of phase, I was stuck in the near field or something... George H.
> > And playing with Spice is more fun. > > This is fun: make a 4-bit binary counter and dump that into a DAC. > Then reverse the bits into the DAC and repeat. The amplitude > histograms are identical, but the spectra are very different. The > "noise" gets pushed up. There must be an optimum remap of the 4 bits > to push the spectrum up. Bit reversal may be it. > > > > -- > > John Larkin Highland Technology, Inc > picosecond timing precision measurement > > jlarkin att highlandtechnology dott com > http://www.highlandtechnology.com
On Thursday, December 15, 2016 at 8:48:19 AM UTC+11, mako...@yahoo.com wrote:
> On Wednesday, December 14, 2016 at 4:35:28 PM UTC-5, John Larkin wrote: > > On Wed, 14 Dec 2016 15:06:31 -0600, Tim Wescott > > <seemywebsite@myfooter.really> wrote: > > > > >On Wed, 14 Dec 2016 10:23:37 -0800, John Larkin wrote: > > > > > >> https://www.google.com/url? > > >sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https > > >%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams% > > >2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja > > >> > > >> or maybe > > >> > > >> http://tinyurl.com/zhlgolz > > >> > > >> > > >> This has some really interesting ideas about extending DAC resolution > > >> with scrambled thermometer codes and such. Some class-D audio stuff, > > >> too. > > >> > > >> I've got to take time to study this; there are a lot of ideas here. > > >> > > >> I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > > >> generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > > >> stock, DAC2904s. They have speed to burn, so I should be able to dither > > >> them up to 16 bits. I can't do simple (wiggle the LSB) > > >> delta-sigma, because the DNL of the DACs is 4 LSBs p-p, so the dither > > >> has to be big enough to smooth out DNL errors too. > > > > > >At second blush (first blush saying "cool!") I wonder if that's going to > > >help -- AFAIK DNL comes from mismatches between legs of the R-2R ladder, > > >and any given leg can be mismatched from another. > > > > I figure that I need to dither 2 or 3 bits at the 14-bit DAC level, to > > smear out the DNL, mostly. And some sub-bits to get from 14 to 16 bit > > resolution. > > > > Our spec will claim 16 bit resolution, but certainly not 16 bit > > accuracy. The real benefit to 16 bits is to avoid gross steps when > > making small signals. Dithering can really help there. There's no > > reason we shouldn't try to dither out to 18 bits... it's easy in an > > FPGA. > > > > These fast dacs tend to be hybrid architectures, some mix of R-2R and > > thermometer codes at the low end. So the dithering will be a little > > weird. > > > > > > > > > >I'd start by figuring out how to model that, then test it (or work it out > > >on paper), and then finally I'd make sure to test the snot out of it in > > >hardware before I decided that I'd actually gotten something good to 16 > > >bits. > > > > > >> This will involve > > >> noise shaping of the dither pattern, sort of like a high-order > > >> delta-sigma, where you push the dither spectrum up so the final analog 5 > > >> MHz lowpass filter kills the noise you add. > > >> > > > > nice discussion re dither here > > INTRODUCTION TO DIGITAL AUDIO by John Watkinson Focal Press > > and > > The Art of Digital Audio By John Watkinson
First published in 1988. Americans don't seem to know about it. Sloman A.W. "Comment on 'Noise averaging and measurement resolution" Review of Scientific Instruments, 70 4734 (1999) was a comment on a paper by somebody who had just re-invented dithering. My commnet did cite Watkinson, and some of the people he'd cited back in 1988. -- Bill Sloman, Sydney
On 15/12/2016 05:23, John Larkin wrote:
> https://www.google.com/url?sa=t&rct=j&q=&esrc=s&source=web&cd=38&ved=0ahUKEwjG19u1n_TQAhVnrlQKHRUMAEc4HhAWCEgwBw&url=https%3A%2F%2Fwww.cscamm.umd.edu%2Fprograms%2Focq05%2Fadams%2Fadams_ocq05.pdf&usg=AFQjCNFFz_m97QqzWywq87nBJJ5uONIk0A&bvm=bv.141320020,d.cGc&cad=rja > > or maybe > > http://tinyurl.com/zhlgolz > > > This has some really interesting ideas about extending DAC resolution > with scrambled thermometer codes and such. Some class-D audio stuff, > too. > > I've got to take time to study this; there are a lot of ideas here. > > I want to make a 16-bit multi-channel, maybe 5 MHz, arbitrary waveform > generator, and I have some nice, fast (120 MHz) 14-bit dual DACs in > stock, DAC2904s. They have speed to burn, so I should be able to > dither them up to 16 bits.
You might have a chance, given that the converter is a lot faster than you are aiming for, but it would be a good idea to plot out the actual INL first, as you won't fix e.g. the major code transition with a few bits of dither. Also, beware glitch energy. Your scheme might be perfect near DC (with sample rates near DC) but if code transitions have a lot (or differ by a lot) of glitch energy, that could still mess you up. It's funny that everyone who buys ADC or DAC chips thinks that they are the first ones to discover dithering, interleaving and averaging etc., and decides that they can make a cheap converter as good as a more expensive one. Sometimes the chip company has already used all of those techniques inside the chip, to the maximum extent that they could, without telling you. (Not necessarily in the part that you are using - I have not looked into that one.) When polishing a shiny turd, you have to stop before you start to rub off the polish that the last guy added.