dCS 904 User Manual

Manual for Software Version 1.5x and 1.36

dCS Ltd

June 2000

 

 

Noise Shaping

The dCS 904 uses noise shaping6 that is optimised to the F weighting curve7. It does not affect signal frequency or transient response, but shapes the frequency response of errors (Q noise, or truncation errors) so that they fall as much as possible in the less sensitive part of the spectrum. The architecture used also shapes dither, where this is added. For all the major sample rates (32 kS/s,

44.1kS/s, 48 kS/s, 88.2 kS/s, 96 kS/s) the noise shapers have been individually optimised and the first 10 orders are offered. The 1st, 3rd, and 9th shapes for

44.1kS/s agree well with Wannamaker’s published results8.

Noise Shaping adds more noise power, but because of the shaping it is perceived as lower noise. There is a compromise to be drawn – as more aggressive shaping is used, more noise is added, and less perceived improvement occurs. In practice, things stop improving much above the 9th order. The increased real noise power can cause (small) clicks in editing, if this is carried out after the shaping. For this reason, noise shaping should be used as late as possible in the mastering process – we recommend recording at the very highest possible sample rate and resolution, and only reducing either at the latest possible minute.

If, however, you have to reduce word length, the perceived noise gain (taking into account the ear’s response) and the actual increase in noise (mainly out of band) in given in the table below.

Sample

Perceived

Actual

Perceived

Actual

Perceived

Actual

Rate

Gain, F

Increase

Gain, F

Increase

Gain, F

Increase

(kS/s)

weighted,

in Noise,

weighted,

in Noise,

weighted,

in Noise,

 

1st Order

1st Order

3rd Order

3rd Order

9th Order

9th Order

 

(dB)

(dB)

(dB)

(dB)

(dB)

(dB)

32

-3.3

1.9

-7.5

4.2

-8.1

6.1

44.1

-5.5

2.4

-10.5

6.9

-17.9

23.4

48

-6.2

2.5

-11.7

7.6

-21.0

23.8

88.2

-11.1

2.8

-23.8

11.3

-42.2

24.0

96

-11.8

2.9

-25.7

11.3

-45.3

22.5

176.4

-17.0

3.0

-40.6

12.6

-63.0

21.8

192

-17.7

3.0

-42.8

12.6

-65.9

21.8

Table 5 – Noise Shaper Gain by Order and Sample Rate

The 3rd order shaping tends to follow the E weighting curve, by chance. The 9th order is very aggressive, and can give very large gains at the higher sample rates. For example, 176.4 kS/s or 192 kS/s material truncated to 16 bits (so it can be stored on a DA-88 or ADAT) loses nothing in the audio band in terms of perceived noise, with either 3rd or 9th order shaping. For more information on this topic, either see section “Word Length Reduction” on page 64 or read the references below.

6

7

8

It actually uses an Error Shaping architecture, but the name is now being used for entirely other things and is less well known, so we call it, erroneously, Noise Shaping

“Minimally Audible Noise Shaping”, S.P.Lipshitz and R.A.Wannamaker, J AES vol 39 no 11, p836-852 “Psychacoustically Optimal Noise Shaping”, R.A.Wannamaker, J AES vol 40 no 7/8, p611-620

Manual part no: DOC135904 iss 2B2

Page 41

135904ma2b2.pdf file available from website

Contact dCS on + 44 1799 531 999

email to: more@dcsltd.co.uk

(inside the UK replace + 44 with 0)

web site: www.dcsltd.co.uk