R/E/P Community

Please login or register.

Login with username, password and session length
Advanced search  

Pages: [1]   Go Down

Author Topic: Noise limits in modern oversampled codecs  (Read 2539 times)

John Roberts {JR}

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 94
  • Real Full Name: John Roberts
  • 10,000+
Noise limits in modern oversampled codecs
« on: September 18, 2011, 01:43:18 pm »

I did a search and didn't find a specific discussion so i though I would ask this knowledgeable group.

What is the real world story with modern codecs that appear to to have quantization levels a few bits below apparent noise floors?

I think I can visualize input related noise in active input analog circuitry and the initial oversampling comparator that will directly create uncertainty in LSB or near LSB level capture.

I guess my question is, does the first conversion actually capture digital data below the input circuitry's noise floor and fix the final relationship, or is the spread between noise floor and quantization level changed by the decimation? 

TIA.

JR
Logged
Circular Science     http://RESOTUNE.COM

"tune it or don't play it..."

Bruno Putzeys

  • Moderator
  • Newbie
  • *****
  • Offline Offline
  • Posts: 22
  • A contrarian for no reason at all.
Re: Noise limits in modern oversampled codecs
« Reply #1 on: September 23, 2011, 05:22:48 am »

To make a long story short, the AES3 interface limits word length to 24 bits. This means the noise contribution of the wordlength is at -146dB. AD/DA converter chips with SNR hovering around 125dB can be had these days from companies like ESS and ARDA. This means that the 24 bits word length is not yet even close to being an important contributor.

Noise performance is limited by a variety of factors, most are purely analogue in nature (thermal noise, settling time errors, matching errors, jitter). It's a matter of economics that an optimum is sought between all noise sources, so there's never a situation where one noise source dominates or another one is wholly negligible. The only exception is quantisation noise where noise shapers are easily designed to contribute far less noise than anything else in the circuit and internal word lengths are chosen 4 or more bits longer than necessary.

This means there isn't a platonic "perfect conversion" sitting somewhere in an otherwise noisy circuit. Everything adds its own flavour of error.
Logged
Not necessarily expressing the opinion of www.grimmaudio.com or www.hypex.nl

John Roberts {JR}

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 94
  • Real Full Name: John Roberts
  • 10,000+
Re: Noise limits in modern oversampled codecs
« Reply #2 on: September 23, 2011, 12:03:18 pm »

Thank you, that is kind of what I though, but it's nice to hear from one who actually knows.

Just to be sure I understand. 24b is the practical word length for storage and interface (more bits may be used for crunching within sundry platforms).

Practical (analog sounding) noise floor for modern convertors is down around a 20-21b equivalent.

Dithering can reduce the LSB quantization energy, below the nominal -146dB. Is this practical or even useful in the context of analog like noise floor 20+ dB louder?
 

Thanks again..     

JR
Logged
Circular Science     http://RESOTUNE.COM

"tune it or don't play it..."

Bruno Putzeys

  • Moderator
  • Newbie
  • *****
  • Offline Offline
  • Posts: 22
  • A contrarian for no reason at all.
Re: Noise limits in modern oversampled codecs
« Reply #3 on: September 26, 2011, 09:32:04 am »

Dithering doesn't reduce the quantisation energy (if anything it adds to it) but what it does is decorrelate it. Undithered quantisation noise sounds nasty while dithered quantisation noise is sonically indistinguishable from analogue noise.

Strictly speaking dither noise has to fulfil certain statistical criteria which analogue noise doesn't, so you should be a bit careful about assuming that analogue noise will be enough to dither a converter. Practically though when the analogue noise is as much as 20dB over the quantisation noise floor, this would be academical nitpicking. I wouldn't be too embarassed about not dithering the 24-bit output of a 120dB SNR converter.

The important thing is not to rely on the same analogue noise to dither subsequent requantisations because that is clearly incorrect.
Logged
Not necessarily expressing the opinion of www.grimmaudio.com or www.hypex.nl

John Roberts {JR}

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 94
  • Real Full Name: John Roberts
  • 10,000+
Re: Noise limits in modern oversampled codecs
« Reply #4 on: September 26, 2011, 10:42:20 am »

Thank you... 

JR
Logged
Circular Science     http://RESOTUNE.COM

"tune it or don't play it..."
Pages: [1]   Go Up