22.09.2015 Views

of Microprocessors

Musical-Applications-of-Microprocessors-2ed-Chamberlin-H-1987

Musical-Applications-of-Microprocessors-2ed-Chamberlin-H-1987

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

436 MUSICAL ApPLICATIONS OF MICROPROCESSORS<br />

As we shall see, a block <strong>of</strong> samples representing a segment <strong>of</strong> sound can<br />

be transformed into the frequency domain via Fourier transform and appear<br />

as a bunch <strong>of</strong> harmonic amplitudes and phases. This data can then be transformed<br />

back into a block <strong>of</strong> samples unscathed and indistinguishable (within<br />

round-<strong>of</strong>f error) from the original block via an inverse Fourier transform!<br />

Thus, Fourier transformation is a reversible operation. However, while in the<br />

frequency domain, we can manipulate the spectrum directly by altering individual<br />

frequency component amplitudes as desired; no filrers to design and<br />

tune are necessary. After the manipulation has been accomplished, the inverse<br />

transform returns the data to the time domain. This is one way to<br />

implement a filter with a completely arbitrary amplitude response.<br />

One can also synthesize a spectrum directly and convert it to the time<br />

domain for output. This can be valuable, since most sounds are more easily<br />

described in terms <strong>of</strong> their spectrum rather than in terms <strong>of</strong> their waveforms.<br />

Although a method was just discussed for Fourier series synthesis, the Fourier<br />

transform can require far less computation if the required spectral detail is<br />

great.<br />

In source-signal analysis using digital techniques, the first step is<br />

nearly always Fourier transformation <strong>of</strong> the input into spectral form. Since<br />

the ear hears in the frequency domain, it is logical that audible features <strong>of</strong> the<br />

source signal will be much more apparent than when in the time domain. In<br />

most cases, transformation greatly reduces the quantity <strong>of</strong> data to be processed<br />

as well.<br />

Characteristics <strong>of</strong> the<br />

Discrete Fourier Transform<br />

Fourier transforms applied to sampled data are usually called discrete<br />

Fourier transforms, since the time domain data are at discrete instants <strong>of</strong> time<br />

and the frequency data are likewise at discrete frequencies. One very important<br />

property <strong>of</strong> the discrete transform is that the waveform data are specificsized<br />

chunks called records, each consisting <strong>of</strong> a specified number <strong>of</strong> samples.<br />

The discrete transform assumes that the samples in the record represent<br />

exactly one cycle <strong>of</strong> a periodic waveform. This assumption must be made<br />

regardless <strong>of</strong> whether or not it is actually true, as in Fig. 13-9. The transform<br />

then gives all <strong>of</strong> the harmonic amplitudes and phases <strong>of</strong> the assumed<br />

periodic waveform.<br />

This record-oriented property has several important ramifications when<br />

constantly changing arbitrary sounds are to be synthesized or analyzed. For<br />

most applications, the record size is fixed and <strong>of</strong>ten is a power <strong>of</strong> two. Thus,<br />

even if a periodic waveform were being analyzed, it is unlikely that a tecord<br />

would exactly span a single cycle. In order to reduce the resulting error, the<br />

record size is chosen to be great enough to span se!Jeral cycles <strong>of</strong> the lowest<br />

frequency expected. Then the partial cycle at the beginning and end <strong>of</strong> the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!