Sign in

Not a member? | Forgot your Password?

Search Online Books

Search tips

Free Online Books

Free PDF Downloads

A Quadrature Signals Tutorial: Complex, But Not Complicated

Understanding the 'Phasing Method' of Single Sideband Demodulation

Complex Digital Signal Processing in Telecommunications

Introduction to Sound Processing

C++ Tutorial

Introduction of C Programming for DSP Applications

Fixed-Point Arithmetic: An Introduction

Cascaded Integrator-Comb (CIC) Filter Introduction


IIR Filter Design Software

See Also

Embedded SystemsFPGA

Chapter Contents:

Search Spectral Audio Signal Processing


Book Index | Global Index

Would you like to be notified by email when Julius Orion Smith III publishes a new entry into his blog?


Hamming Window

The Hamming window is determined by choosing $ \alpha $ in (3.1) (with $ \beta=1-\alpha$) to cancel the largest side lobe [97].4.1 Doing this results in the values

\alpha &=& \frac{25}{46} \approx 0.54\\
\beta &=& \frac{1-\alpha}{2} \approx 0.23.

The peak side-lobe level is approximately $ -42.76$ dB for the Hamming window [97].4.2 It happens that this choice is very close to that which minimizes peak side-lobe level (down to $ -43.19$ dB--the lowest possible within the generalized Hamming family) [185]:

$\displaystyle \alpha = 0.53836\ldots\,.

Since rounding the optimal $ \alpha $ to two significant digits gives $ 0.54$, the Hamming window can be considered the ``Chebyshev Generalized Hamming Window'' (approximately). Chebyshev-type designs normally exhibit equiripple error behavior, because the worst-case error (sidelobe level in this case) is minimized. However, within the generalized Hamming family, the asymptotic spectral roll-off is constrained to be at least $ -6$ dB per octave due to the form (3.1) of all windows in the family. We'll discuss the true Chebyshev window in §3.10 below; we'll see that it is not monotonic from its midpoint to an endpoint, and that is is in fact impulsive at its endpoints. (To peek ahead at a Chebyshev window and transform, see Fig.E.26.) Generalized Hamming windows can have a step discontinuity at their endpoints, but no impulsive points.

Figure 3.4: A Hamming window and its transform.

The Hamming window and its DTFT magnitude are shown in Fig.3.4. Like the Hann window, the Hamming window is also one period of a raised cosine. However, the cosine is raised so high that its negative peaks are above zero, and the window has a discontinuity in amplitude leaving the window (stepping discontinuously from 0.08 to 0). This makes the side-lobe roll-off rate very slow. On the other hand, the worst-case side lobe plummets to $ -41$ dB,4.3 which is the purpose of the Hamming window. This is 10 dB better than the Hann case of Fig.3.3 and 28 dB better than the rectangular window. The main lobe is approximately $ 4\Omega_M$ wide, as is the case for all members of the generalized Hamming family ( $ \beta\neq 0$).

Due to the step discontinuity at the window boundaries, we expect a spectral envelope which is an aliased version of a $ -6$ dB per octave (i.e., a $ 1/\omega$ roll-off is converted to a ``cosecant roll-off'' by aliasing, as derived in §4.5 and illustrated in Fig.4.12). However, for the Hamming window, the side-lobes nearest the main lobe have been strongly shaped by the optimization. As a result, the nearly $ -6$ dB per octave roll-off occurs only over an interior interval of the spectrum, well between the main lobe and half the sampling rate. This is easier to see for a larger $ M$, as shown in Fig.3.5, since then the optimized side-lobes nearest the main lobe occupy a smaller frequency interval about the main lobe.

Figure 3.5: A longer Hamming window and its transform.

Since the Hamming window side-lobe level is more than 40 dB down, it is often a good choice for ``1% accurate systems,'' such as 8-bit audio signal processing systems. This is because there is rarely any reason to require the window side lobes to lie far below the signal quantization noise floor. The Hamming window has been extensively used in telephone communications signal processing wherein 8-bit CODECs have been standard for many years (albeit $ \mu$-law encoded). For higher quality audio signal processing, higher quality windows may be required, particularly when those windows act as lowpass filters (as developed in Chapter 8).

Previous: Matlab for the Hann Window
Next: Matlab for the Hamming Window

Order a Hardcopy of Spectral Audio Signal Processing

About the Author: Julius Orion Smith III
Julius Smith's background is in electrical engineering (BS Rice 1975, PhD Stanford 1983). He is presently Professor of Music and (by courtesy) of Electrical Engineering at Stanford's Center for Computer Research in Music and Acoustics (CCRMA), teaching courses and pursuing research related to signal processing applied to music and audio systems. See for details.


No comments yet for this page

Add a Comment
You need to login before you can post a comment (best way to prevent spam). ( Not a member? )