DSP chips for IF processing lower the cost of wireless communications As DSP chips get faster, they make possible precise IF functions that analog circuits couldn't manage BY DAVID B. CHESTER and GEOFF PHILLIPS Harris Semiconductor Melbourne, FL Digital intermediate frequency processing is becoming the technology of choice in nonmobile applications like cellular base stations. As IF processing matures, it will migrate into mobile applications as well. About four companies provide products in this area, and more should be coming along as applications become more widespread. Digital IF processing had its beginnings in large, low-volume applications in which performance was the only concern. As system requirements became more complex, it became clear that digital IF processing provided not only the most economical system solution, but often the only viable solution. The first use of DSP technology at intermediate frequencies was direct digital synthesis (DDS). As A/D converter technology matured, making it possible to quantize signals at IF, DDS was replaced by totally digital IF processing. The future for digital IF processing looks toward higher integration, lower power consumption, and higher speeds. The integration will expand in both directions. As analog-digital interface technology matures, digital IF frequencies will continue to increase. In the other direction, road maps are already laid out to integrate programmable baseband processors with function-specific IF processing to offer complete digital solutions. As system designers become more familiar with digital IF processing, they will be able to work more closely with IC designers to specify total digital solutions that will simplify overall system designs. IF DSP functions Digital IF processing is dominated by two DSP functions: digital frequency modulation (DDS and digital mixing) and filtering. If data rates are too high for general-purpose DSP microprocessors, traditional baseband processes like equalization, demodulation, or modulation can be considered digital IF processes as well. In most digital IF-processing applications, the DDS or local oscillator function produces a quadrature sinusoid. The sinusoid has a high degree of amplitude, phase, and frequency accuracy. Digital mixing is usually either a complex multiplication or the real part of a complex multiplication. IF filtering in a receiver usually consists of filtering out a narrow band of interest from a wide-band input and decreasing the sampling rate appropriately. In a transmitter (see Fig. 1a), IF filtering usually consists of expanding the sampling rate of a baseband signal so it can be frequency-division multiplexed with other signals. Then it is sent to a D/A converter to be converted to analog for transmission. Equalization filtering is necessary when multipath effects or channel bandwidth compression produce intersymbol interference. The process employs a digital filter with either fixed or time-varying (adaptive equalization) coefficients. Advantages of digital IF processing Digital IF-processing chains give the designer tight control over the IF local oscillator, precise mixing, programmable filter bandwidths, linear phase characteristics, computation of adaptive coefficients, and control over modulation and demodulation parameters. Specifically, today's digital down-converters can control center frequencies with a precision of better than 0.01 Hz. They can mix signals with over 100 dB of spurious-free dynamic range (SFDR). In contrast to analog implementations, which require separate filters for different bandwidths, a digital filter can simply be programmed for any desired bandwidth, while phase characteristics remain linear. Digital IF processing can produce a quadrature data stream with a quadrature match equal to the mixing precision, and all component accuracies are reproducible and do not degrade over time or temperature. In a typical digital IF-processing receiver, the analog IF is conditioned by the RF front end and sampled by the A/D converter (see Fig. 1b). Then the IF signal is mixed with a quadrature sinusoid to center the band of interest at dc. The resulting complex data stream goes through a low-pass filter to produce a quadrature baseband signal. (Quadrature data streams enhance computational efficiency in the IF process itself and in subsequent baseband operations like equalization and demodulation.) If the resulting baseband data is too wide-band for general-purpose DSP processors, it can be further processed by digital IF-processing components. The quadrature sinusoid is synthesized in the digital local oscillator, which consists of a phase generator and a sine/cosine generator. Tuning resolution is determined by the width of the frequency control word. A phase offset in the local oscillator can be generated by adding a constant to the phase accumulator. Tuning speed is governed by how much time it takes to load a new frequency control word, usually one clock cycle. Because of the inherent characteristics of the phase accumulator, all frequency changes are phase-synchronous. If a digital IF processor is correctly designed, the SFDR of the entire processing chain is determined by the SFDR of the local oscillator, which is a function of the phase word width, the word width of the output sinusoid, and the arithmetic accuracy of the sine and cosine calculation. As a guide, the number of bits used in the phase word controlling the sine/cosine generator in the digital local oscillator is 1 bit greater than the number of bits, M, used to represent each of the two sinusoidal values output by the local oscillator. In that case, the worst-case signal-to-noise ratio (SNR) and SFDR will both be approximately 6 dB times M. Filter selectivity can be made arbitrarily sharp using digital filters. A digital filter is simply a discrete-time, discrete-amplitude convolver. Application characteristics that favor digital filters are linear phase, high stop-band attenuation, low passband ripple, low shape factors (the ratio of the filter's passband width plus the filter's transition bandwidth to the filter's passband width), programmable or adaptive filter response, and phase manipulation. All-digital IF processing has advantages in terms of flexibility of design and reliability of manufacture. Such processing is flexible because most of the parameters, such as local oscillator center frequency, filter bandwidth, filter shape and output format, can be made programmable. Even if some parameters are not programmable, once a basic architecture is derived, it is easily modified. Manufacturing really benefits because digital components are relatively easy to build and test. They require no component trimming and they hold their original component characteristics for life. High-volume commercial wireless Before wireless technology can come into widespread use, some issues remain to be addressed. For one thing, lower cost requires more integration. The near future will see the integration of down-conversion, equalization, and demodulation on chips or chip families in digital receivers and the integration of modulation, up-conversion, and multichannel combining on chips or chip families in transmitters using digital IF processing. Higher integration forces semiconductor manufactures to become more system-capable. Their designers must gain a detailed understanding of the target systems. One of the biggest stumbling blocks to moving digital IF processing into the mainstream is simply that many RF engineers are not yet comfortable with DSP technology. The system block diagram may be the same, but the implementation details are far different. Until the transformation is complete, digital IF-processing vendors must not offer just devices, but solutions and design partnerships. No panacea Although digital IF processing is powerful, it is still relatively expensive and more power-hungry than analog. It is still true that if a function can easily be implemented in an analog technology, it will be smaller, cheaper, and consume less power than the digital alternative. Among the overall system considerations are the performance and cost of the analog-digital interface. While digital IF-processing ICs provide better than 100 dB of SFDR and sampling rates approaching 100 Msamples/s, A/D and D/A converters cannot yet approach these rates. Therefore, digital IF-processing devices can be used at lower sampling rates with higher dynamic ranges or at higher sampling rates with lower dynamic ranges. Even in receiver applications where digital provides a large SNR gain, the SFDR of the A/D converter is still limiting unless the designer can count on the spurs falling outside the final band of interest. Even when suitable converters are available, or where external sample-and-hold circuitry enhances performance, cost remains a driving factor. One way to reduce the impact of interface cost is to design the system so multiple digital IF-processing channels use the same conversion and analog hardware. This is exploited in multichannel applications like cellular base stations. In some cases, certain elements of digital IF-processing chains can be shared between receiver and transmitter. The impact on the analog front end must also be considered. Analog gain, noise floor, and intercept point should match the interface and digital IF-processing characteristics. Because the rejection filters in the analog front end now act as the anti-aliasing filter for the A/D converter, they must have a relatively sharp rolloff characteristic and well-behaved group delay characteristics. Otherwise, much of the digital bandwidth is wasted on the transition region. This effect is amplified by the anti-alias filtering being typically a bandpass process for digital IF. Power dissipation is also a concern, in both the digital IF-processing circuitry itself and in the interface circuitry. As a rule, analog processing at a given frequency consumes much less power than digital processing of the same frequency. Thus, the digital advantage comes when performance is only possible digitally. Available products The major sources of digital IF-processing products are Harris Semiconductor, Stanford Telecommunications, Zilog, and Graychip. Harris Semiconductor offers standard products and high volume custom products. Harris' standard product offerings have concentrated on up/down converters, digital modulators, and IF digital filters. Its high volume custom capabilities include standard product derivatives, application-specific equalization products, and application-specific modulators and demodulators. Harris plans to expand its filter and down converter lines and expand into programmable demodulator standard products. Stanford Telecommunications offers a broad range of communications products. These include frequency synthesis products, demodulators, forward error correction products, and pseudo-random noise (PRN) coding products. Zilog is now a second source for some of Stanford Telecommunication's spread-spectrum products. Zilog is targeting high-volume, low cost applications. Graychip specializes in low-volume, high-end applications with both standard and custom products. Current standard product offerings from Graychip include digital receivers, digital filters and equalizers, and a resampler. cc CAPTIONS Fig. 1. In a wireless transmitter (a) and receiver (b), digital IF processing is handled by hard-wired special-purpose ICs. Baseband functions like equalization and modulation/demodulation (the dashed boxes), are often handled by software-controlled general-purpose DSP microprocessors. For more information from Graychip, Inc. Harris Semiconductor xxx Standford Telecommunications xxx Zilog xxx OVERLINE: DSP chips for IF processing
Advertisement
