Hi,
Many audio codec uses sigma-delta technology. I do not understand why there are so many bits,
which are far more than the SNR spec. For example, ADI has ADAU1761 which is 24-bit, 98 dB SNR.
Could you tell me that?

Thanks.

In a sigma delta ADC, there are several different noise sources;

1) the thermal noise of the analog electronics
2) the shaped quantization noise of the sigma-delta loop
3) the output truncation noise of the digital filter

In a well-designed decimation filter, the internal word lengths are made large to avoid any
truncation effects, so you have a lot of bits available internally, and the question is how many
should you send to the output serial port.  Since the total noise is the sum of all the
components mentioned above, it makes sense to minimize the contribution of the output truncation
noise in order to achieve the best snr.  Since most digital systems don't like bit-widths that
are not a multiple of 8, it makes sense to go to 24 bits since 16 bits would not be enough.

Bob

Robert Adams <r...@analog.com> writes:

> In a sigma delta ADC, there are several different noise sources;
>
> 1) the thermal noise of the analog electronics
> 2) the shaped quantization noise of the sigma-delta loop
> 3) the output truncation noise of the digital filter
>
> In a well-designed decimation filter, the internal word lengths are
> made large to avoid any truncation effects, so you have a lot of bits
> available internally, and the question is how many should you send to
> the output serial port. Since the total noise is the sum of all the
> components mentioned above, it makes sense to minimize the
> contribution of the output truncation noise in order to achieve the
> best snr. Since most digital systems don't like bit-widths that are
> not a multiple of 8, it makes sense to go to 24 bits since 16 bits
> would not be enough.

Hi Bob,

Yes, I agree. If it were truly a 24-bit converter, I'd say take the
internal bit width up another few bits past 24, but the thermal noise
probably swamps any quantization noise at the 24 bit level.
-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

Robert Adams <r...@analog.com> writes:

> In a sigma delta ADC, 

and by the way, it's "delta sigma"!
-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

Robert Adams wrote:
> In a sigma delta ADC, there are several different noise sources;
>
> 1) the thermal noise of the analog electronics 2) the shaped
> quantization noise of the sigma-delta loop 3) the output truncation
> noise of the digital filter
>
> In a well-designed decimation filter, the internal word lengths are
> made large to avoid any truncation effects, so you have a lot of bits
> available internally, and the question is how many should you send to
> the output serial port.  Since the total noise is the sum of all the
> components mentioned above, it makes sense to minimize the
> contribution of the output truncation noise in order to achieve the
> best snr.  Since most digital systems don't like bit-widths that are
> not a multiple of 8, it makes sense to go to 24 bits since 16 bits
> would not be enough.
>
> Bob
>

It's nice to have the option of 24 bits, but there are vanishingly
few cases where you actually get even 96 full dB (16 bits ) of clean
input into an ADC.

SFAIK, fab process and the general ... economics of electronics
dictate a standardization to 24 bit ADCs but being in a place
to exploit them fully is rare. 2^24 is one big ole
number... 16,777,216, aka 144 dB.


--
Les Cargill

Robert Adams <r...@analog.com> wrote:
> In a sigma delta ADC, there are several different noise sources;

> 1) the thermal noise of the analog electronics
> 2) the shaped quantization noise of the sigma-delta loop
> 3) the output truncation noise of the digital filter

(snip)
>  Since the total noise is the sum of all the components mentioned 
> above, it makes sense to minimize the contribution of the output 
> truncation noise in order to achieve the best snr.  
> Since most digital systems don't like bit-widths that are not 
> a multiple of 8, it makes sense to go to 24 bits since 16 bits 
> would not be enough.

It seems to me that one could do 20 bit stereo. That is, 
five bytes for two samples. It doesn't seem that anyone
does that, though.

-- glen

On 10/18/12 7:14 AM, Robert Adams wrote:
> In a sigma delta ADC, there are several different noise sources;
>
> 1) the thermal noise of the analog electronics
> 2) the shaped quantization noise of the sigma-delta loop
> 3) the output truncation noise of the digital filter
>
> In a well-designed decimation filter, the internal word lengths are made large to avoid any
truncation effects, so you have a lot of bits available internally, and the question is how many
should you send to the output serial port.  Since the total noise is the sum of all the
components mentioned above, it makes sense to minimize the contribution of the output truncation
noise in order to achieve the best snr.  Since most digital systems don't like bit-widths that
are not a multiple of 8, it makes sense to go to 24 bits since 16 bits would not be enough.

i seem to remember, back in the early '90s, when i first got introduced 
to you, Bob, i remember me complaining about the unnecessary and 
unfortunate truncation of noisy or "meaningless" bits in the ADCs of the 
day.  like we were getting only 16 or 18 bits from a sigma-delta that 
had about 98 dB. i remember telling you and some guy from Crystal that 
we audio DSPers *wanted* those noisy bits.  at least 24 bits total. 
truncating, essentially replacing the lowest-order bits coming out of 
the digital decimation filter, simply adds additional noise and removed 
a handy dither in the signal that sometimes makes the audio algorithm 
work better and sound better as signals slide into the noise floor. 
there is still some signal left buried in the noise of those lower bits.

it looks like our complaints might have been heard.

considering that single-precision floating point is really 25 bits in 
the mantissa, that some DSPs (like your sigma) have 28 bits (or more) 
below the guard bits on the left, it seems to me that nowadays, *even* 
if the ADC is 96, 108, or maybe 120 dB dynamic range, that we should be 
able to get up to 32 bits out of the down-sampling filter in the ADC.


-- 

r b-j                  r...@audioimagination.com

"Imagination is more important than knowledge."

"Les Cargill" <l...@comcast.com> wrote in message 
news:k5osd0$abh$1...@dont-email.me...

> It's nice to have the option of 24 bits, but there are vanishingly
> few cases where you actually get even 96 full dB (16 bits ) of clean
> input into an ADC.

Good quality audio ADC have dynamic range over 120dB. Instrumental ADCs 
reach over 140dB.
Despite of those high parameters, ADCs are always a bottleneck and you have 
to switch gears and do all kinds of analog signal conditioning before the 
ADC.

> SFAIK, fab process and the general ... economics of electronics
> dictate a standardization to 24 bit ADCs but being in a place
> to exploit them fully is rare. 2^24 is one big ole
> number... 16,777,216, aka 144 dB.

If you want to measure something to a resolution of 1%, then 24 bits is mere 
100dB of dynamic range at best.

Vladimir Vassilevsky
DSP and Mixed Signal Consultant
www.abvolt.com

>Robert Adams wrote:
>> In a sigma delta ADC, there are several different noise sources;
>>
>> 1) the thermal noise of the analog electronics 2) the shaped
>> quantization noise of the sigma-delta loop 3) the output truncation
>> noise of the digital filter
>>
>> In a well-designed decimation filter, the internal word lengths are
>> made large to avoid any truncation effects, so you have a lot of bits
>> available internally, and the question is how many should you send to
>> the output serial port.  Since the total noise is the sum of all the
>> components mentioned above, it makes sense to minimize the
>> contribution of the output truncation noise in order to achieve the
>> best snr.  Since most digital systems don't like bit-widths that are
>> not a multiple of 8, it makes sense to go to 24 bits since 16 bits
>> would not be enough.
>>
>> Bob
>>
>
>It's nice to have the option of 24 bits, but there are vanishingly
>few cases where you actually get even 96 full dB (16 bits ) of clean
>input into an ADC.
>
>SFAIK, fab process and the general ... economics of electronics
>dictate a standardization to 24 bit ADCs but being in a place
>to exploit them fully is rare. 2^24 is one big ole
>number... 16,777,216, aka 144 dB.

There are audio converters that achieve 120dB SNR. If you want to be able
to easily swap between a 98dB one and a 120dB one it pays to use a
standardised number of bits on the interface, Of course, the marketing
department just loves this. :-)

I know of one commercial converter chip which achieves a genuine 23.5 ENOB,
but at only 10 samples per second. So, we can achieve an impress dynamic
range under the right circumstances.

Steve

Steve

On Thu, 18 Oct 2012 10:57:18 -0500
"steveu" <31473@dsprelated> wrote:

> [snip]
> I know of one commercial converter chip which achieves a genuine 23.5 ENOB,
> but at only 10 samples per second. So, we can achieve an impress dynamic
> range under the right circumstances.
> 
> Steve
> 

You know, even at 10 sps that's fairly impressive.


-- 
Rob Gaddi, Highland Technology -- www.highlandtechnology.com
Email address domain is currently out of order.  See above to fix.