DSP filters essential for comms apps

DSP filters essential for comms apps
We've come to expect a great deal from our digital filters. DSL-a miraculous imposition of broadband technology on the installed base of moldy twisted-pair cabling, barely suitable for truncated voice-is but one example. Today, we expect more from our cell phone RF tuners, our digital-camera image processors and the new digital audio players, which get startling audio fidelity from an otherwise questionable serial data stream.

The delta-sigma D/A converters used for compact disk playback are actually special-purpose DSP processors. They rebuild the serial digital audio stream to resemble the one you might have gotten if you sampled the original audio signal at, say, 256 times the 44-kHz Nyquist frequency (at roughly 12.5 MHz). This generates many more ones-and-zeros in the data stream than you might have had by sampling at 44 kHz. The new data stream positions the quantization noise at a lower frequency from the rest of the audio signal, which makes it very easy to filter.

In analog terms, we pass a high-speed signal through a filter bank in order to elevate certain frequencies of interest and truncate those frequencies of less interest to us. In digital terms, we multiply a digital word representing a sample (a discrete time slice) of the analog signal by another digital word representing a filter function, and then smooth the result.

If you do this several million times a second, you can not only get your cell phone to make smooth cell-to-cell handoffs as you barrel down the highway, but you might even be able to stream video images across your little LCD screen.

As Texas Instruments Inc.'s Alan Gatherer, a distinguished member of the technical staff, points out in his contribution on Universal Mobile Telecommunications System (UMTS) transmission, 3G cellular is less dependent on increased RF data transmission frequencies than on the ability to extract more data symbols from the digitally modulated RF carrier. The use of 16QAM means that there are 16 data symbols mapped by every subtle shift the carrier makes in phase and amplitude. It is up to the DSP filter to extract these shifts.

In several online exclusives, meanwhile, contributors to this week's In Focus identify the DSP filter operations that control the noise and harmonics in a range of high-speed communications circuits. Analog Devices Inc. applications engineers David Katz and Rick Gentile describe the operations necessary to alter pixels-video images-on a CRT or flat-panel screen. A complex convolution, they explain, will alter the numerical value of a pixel relative to the ones around it. Such mathematically intensive convolutions are necessary for changing colors or sharpening contrasts.

STMicroelectronics' Stephane Barbu explains how DSP filter operations (primarily decimation) can partition telephone lines into the 256 4-kHz slices that characterize DSL. One slice, sharply filtered, is dedicated to voice, while each of the other slices-though sensitive to line conditions and distances from the source-can transmit up to 64 kbits/second.

Bit precision impact

TI audio applications manager David Zaucha describes the effects of word width-"bit precision"-on the infinite-impulse-response filters used in professional audio applications. Below 20-bit precision, Zaucha writes, quantization noise will overlap the lowest-level audio signals, making it obvious why 24 bits or more is becoming standard for audio processors.

Finally, Patrick Lejoly, application manager for Philips Semiconductor's data converter product line, describes the bursty nature of charge-coupled-device outputs in his online piece. The filter that processes such signals for digital cameras must subtract the reference pulses from bursts that contain both color and light intensity for millions of pixels. Lejoly describes the machine that performs the signal conditioning as an "analog front end" but notes that it performs quantization and filtering before sending the captured image to storage.

DSP filtering is very math-intensive. On a microprocessor, a multiplication operation must be programmed as a repeated sequence of additions-millions of them. The DSP has an advantage in its ability to perform a multiply operation directly. The programmer however, must inject the filter parameters as coefficients and polynomials. Even so, many filter operations require lengthy repeated loops.

http://www.eet.com