Page 359 - Analog and Digital Filter Design
P. 359

356 Analog and Digital Filter Design




                       Interpolation
                       Interpolation  is  the  opposite  of  decimation.  Suppose  our  process described
                       above must produce an output at 48 kHz sample rate, to be compatible with the
                       rate of  the input. We  have  reduced the internal data rate to 24kHz,  so addi-
                       tional samples must be inserted between the data samples. These samples are
                       usually the average value of  the previous sample and the following sample. In
                       some cases the same data may be in pairs of samples, but this does not add value
                       to the signal. By inserting average data values the signal output from the digital-
                       to-analog  converter is  smoother.  It also contains a lower  signal power  at the
                       sampling frequency.

                       Decimation  and  interpolation  are usually arranged to reduce or multiply the
                       data rate by a power of two. So we may have decimation rates of  ‘h? ‘I4,  and
                       so on. This means  that  there  is  an equal spacing between  samples, which is
                       important for reconstructing the signal.

                       Decimation takes place ”naturally” in sigma-delta analog-to-digital converters.
                       These devices sample the signal by  typically 64 times the output data rate. A
                       converter having a 40 kHz output data rate may sample a signal with a 10 kHz
                       bandwidth, so the signal is sampled at 2.56MHz in this case. The digitized signal
                       is decimated in the conversion process because each sample is used to produce
                       one bit of  data. The output is logic 1 if  the signal is higher than the previous
                       sample. The output  is  logic 0 if  the  signal voltage  is  lower  than  before. The
                       binary word is thus built up by  adding or subtracting bits until the data word
                       represents the signal level.



                 Digital Filtering

                       This book has described analog filtering in some depth, both in terms of  the
                       frequency response and in terms of  the pole and zero locations. It is somehow
                       easy to imagine (for me, at least) signals flowing in a circuit. I can imagine the
                       potential  divider action as the  impedance of  an inductor increases with  fre-
                       quency while the impedance of a capacitor reduces. In a filter diagram the signal
                       path is usually through a single wire.

                       Digital filtering is a completely different concept from analog filtering. Digital
                       filtering processes signals in the time domain. Therefore, if  a certain frequency
                       domain response is required, it is necessary to convert this response into the
                       equivalent  time  domain.  It  is  not  always intuitively obvious  what  this  time
                       domain signal looks like. What is more, the signals are usually in parallel digital
                       form, in other words, they are a binary-coded version of  an analog signal on,
                       perhaps,  a  16-bit-wide bus. In  some digital filter diagrams  it  is  not  obvious
                       that the signal path is a data bus (unless you are familiar with microprocessor
   354   355   356   357   358   359   360   361   362   363   364