<< Chapter < Page Chapter >> Page >

Noise-shaping

Noise shaping is a procedure which allows to put some color on an additive white noise. While this adds to the overall noise,one can shape the resulting noise such that its prevalent “color” lies in the high frequencies and at the same time reduce thepresence of noise in the low frequencies.

Usual quantization noise.
Usual quantization noise.
When Oversampling, only the part in the band
When Oversampling, only the part in the band [ - f 0 / 2 , f 0 / 2 ] will be relevant after down-sampling.
Noise Shaping moves the noise away from the smaller frequencies.
Noise Shaping moves the noise away from the smallerfrequencies.

In the context of oversampling, noise shaping achieves an improvement if it is done so that it reduces noise in the band [ - f 0 / 2 , f 0 / 2 ] and pushes it to the “colors” with frequencies in [ f 0 / 2 , f e / 2 ] and [ - f e / 2 , - f 0 / 2 ] .

The most simple version of noise shaping results in two noise terms, the one produced by the system, one added with a delay on purpose.Set ε 0 = 0 , then:

x k - ε k - 1 ( x k - ε k - 1 ) Q u a n t . y k output ε k = y k - ( x k - ε k - 1 ) feedback

This assumes that we know or can know the error. Such is the case with quantization, where we can compute ε k simply as the difference between output and input of the quantization (see [link] ). This is not feasible in the example of a wireless channel with inference.

To study the effect of noise shaping we compute the Fourier transform of the overall error n k = y k - x k . We find

n k = y k - x k = ε k - ε k - 1

or

n ( t ) = y ( t ) - x ( t ) = e ( t ) - e ( t - τ e )

with Fourier transform (use that f e = 1 / τ e )

N ( f ) = E ( f ) - E ( f ) e - j 2 π τ e f = E ( f ) ( 1 - e - j 2 π f / f e ) = E ( f ) 2 sin ( π f / f e ) e - j π f / f e

and power spectrum, (use that 2 sin 2 ( x ) = 1 - cos ( 2 x ) )

| N ( f ) | 2 = | E ( f ) | 2 2 1 - cos 2 π f f e

Note that the spectrum is no longer flat; in other words, the noise n k is colored. We note that the colored spectrum is small for small f , and it is still spread over a period of length f e . Thus, noise shaping results in a reduction of noise inthe small frequencies (see [link] ).

Now we continue as with simple oversampling: since the signal has been sampled at f e = β f 0 , with f 0 2 B and β > 1 we may low-pass filter after noise shaping at cutoff frequency f 0 / 2 . Thus, we take advantage of the fact that most of the power of the“shaped” or “colored” noise is in the `high'-frequency bands [ f 0 / 2 , f e / 2 ] and [ - f e / 2 , - f 0 / 2 ] with very little noise at frequencies around 0.

To assess the gain we compute the power of the noise after this low-pass filter. Note that the low-pass filter sets N ( f ) = 0 for f e / 2 > | f | > f 0 / 2 : this removes much of the noise as demonstrated in the following computation; it also guarantees that power won't be changedwhen downsampling after filtering to f 0 . Using again that E ( f ) = Δ 2 / 12 we get:

P shaping = 1 f e - f 0 / 2 f 0 / 2 | N ( f ) | 2 d f = 1 f e 2 0 f 0 / 2 | E ( f ) | 2 2 1 - cos 2 π f f e = 4 Δ 2 12 f e f - f e 2 π sin 2 π f f e | 0 f 0 / 2 = Δ 2 12 4 1 2 β - 1 2 π sin π β

Using the approximation sin ( x ) x - x 3 / 6 + x 5 / 5 ! - + . . . we find

P shaping P noise = 4 1 2 β - 1 2 π sin π β 4 1 2 β - 1 2 π π β - π 3 6 · β 3 = π 2 3 β - 3

Noise reduction under oversampling with noise-shaping

In conclusion, the SNR improves (as compared to no oversampling) under noise shaping by the inverse of the factor [link] , thus roughly by 3 π 2 β 3 . In decibel, this corresponds to approximatively 10 log 10 ( 3 ) - 20 log 10 ( π ) + 30 log 10 ( β ) , thus, roughly 3 times more than with oversampling alone.

Numerical Examples:

With an OSR of β = 128 the SNR improves

  • under oversampling alone by 10 log 10 ( 128 ) 21 dB
  • under oversampling coupled with basic noise-shaping as above by approximatively 10 log 10 ( 3 ) - 20 log 10 ( π ) + 30 log 10 ( 128 ) = 58 . 0445 dB, or roughly 3 times more.

When doubling the OSR the SNR improves

  • under oversampling alone by 10 log 10 ( 2 ) 3 . 01 dB
  • under oversampling with noise-shaping by approximatively 30 log 10 ( 2 ) = 9 . 03 dB, or approximatively 3 times more.

Get Jobilize Job Search Mobile App in your pocket Now!

Get it on Google Play Download on the App Store Now




Source:  OpenStax, Sampling rate conversion. OpenStax CNX. Sep 05, 2013 Download for free at http://legacy.cnx.org/content/col11529/1.2
Google Play and the Google Play logo are trademarks of Google Inc.

Notification Switch

Would you like to follow the 'Sampling rate conversion' conversation and receive update notifications?

Ask