Editor's Note: To view a PDF version of this article, Click Here.
With the popularity of wireless-LAN connectivity has come the demand for smaller form factors, lower power consumption, fewer parts and lower cost, while still meeting ever-tighter time-to-market requirements. Thanks to advances in the area of zero-intermediate-frequency (zero-IF or ZIF) radio architectures, many of these requirements can now be met.
The ZIF concept of translating information directly to the transmitted frequency band (vice versa on the receive side) was developed even before the superheterodyne architecture for receivers. Also called the direct-down-conversion (or homodyne) architecture, it is attractive for its overall simplicity and reduced parts count, though it has historically been difficult to work with. Challenges awaiting the designer include the very difficult direct-current (dc) offset issue. However, by leveraging advanced technology along with the application of good design practice, many of these problems can be solved, to the extent that ZIF can be used in both single- and dual-band wireless-LAN (WLAN) radios.
While ZIF solves the issue of many spurious frequencies with which a radio has to contend, it trades many small spurious problems for one very big one, namely direct current. To compound this problem, the dc issue has thermal and external considerations that must be addressed. Since most of the amplification in the ZIF radio is at baseband, it needs matched, balanced automatic gain control (AGC) amplification. Another problem is that local-oscillator leakage on the transmitter is at band center and can jam nearby receivers.
ZIF Caveats
When ZIF technology is used, the signal being demodulated is centered at dc and must contend with extraneous dc introduced by practical circuits. The signal to be processed cannot have a significant dc component, because it would be impossible to distinguish which part of the dc belonged to the signal. There are waveforms that meet this criterion such as direct-sequence spread spectrum (DSSS) that uses double sideband suppressed carrier phase-shift keying modulation. Another is orthogonal frequency-division multiplexing (OFDM) with the center carrier not used.
The signal will vary in strength over a range of 75 dB or so. It can be 30 to 40 dB weaker than the dc introduced by the mixer's inherent dc offsets. However, the biggest problem is that the direct current is not stationary, but can vary with external influences, such as time, supply voltage, temperature and, worst of all, with changes in gain needed to accommodate the signal level. This means that dynamic dc compensation is needed to make sure the dc will not interfere with the signal.
One external factor that must be compensated for is the reflection of the signal off a nearby surface back into the antenna. Depending on the strength and phase angle of the reflection, the dc that is created this way can vary considerably. If the surface is moving, there is both a Doppler component and a rapidly varying fading component. Another impact is the effects of nearby surfaces on the voltage standing-wave ratio of the antenna itself. If the VSWR changes due to near-field loading, the signal is reflected back into the mixer stage, creating dc. Since this dc will be time varying, the ac-coupling or dc-feedback loop must have a corner frequency response that is faster than the change rate of the reflection dc. This is usually about 100 kHz. The response time of this ac coupling must be taken into account in the acquisition time line, especially with 802.11a, which has only a 16-microsecond preamble.
Dc compensation
There are several ways to combat dc and gain-balance issues:
- Avoid using ZIF. Consider using superheterodyne techniques where dc is out of band and the signal can be amplified by a single amp.
- Use ac coupling at all stages.
- Use dc coupling with dc feedback (works somewhat like ac coupling).
There are numerous trade-offs necessary to make the zero-IF concept practical. Basically, the zero-IF receiver takes the whole spectrum in through the antenna and single-sideband downconverts it such that the desired signal is at baseband, or zero-IF, frequency. It then uses low-pass filters (LPFs) to remove everything but the signal of interest before it amplifies and detects it. The synthesizer is at the same frequency as the desired signal and the downconverted signal is centered at dc. Normally, the desired signal must be processed as a dc signal and must contend with a large (relatively) dc offset in the mixer.
Additionally, the baseband signal is complex and has a real and an imaginary part (I and Q). These two signals may range from microvolts to volts and must be amplified linearly while maintaining the same relative amplitude and phase. So, the automatic gain control circuit must deal with large gain changes in two very well-matched AGC amplifiers.
ZIF Coupling Options
There are two ZIF coupling options-ac or dc. If the RF signals are DSSS or OFDM, the designer can ac-couple the baseband signals and can also ac-amplify them (see Figure 1). This removes one of the problems of the ZIF receiver. But one difficulty with this approach is the large number of discrete capacitors needed, and the need to move on and off chip at each stage. Integrating the capacitors is not feasible due to the large size of the capacitors needed. If the preamp, mixer, low-pass filter and AGC amplifier are all on one chip and the signal is handled as balanced differential, then eight pins and four capacitors are required for each ac-coupling stage. With this approach, at least two ac-coupling stages are needed.
Dc-coupled receiver: Dc nulling can be accomplished either by capacitors as shown previously, or by dc-feedback techniques that implement the same functionality (see Figure 2). In either case, you have a high-pass frequency characteristic. The advantages of dc-feedback techniques are that the signal never needs to go off-chip and no discrete capacitors are needed.
A typical spectrum has many signals of highly varying amplitudes. This makes it necessary to have a wide dynamic range in the input stages of the receiver. The desired signal may be very low or it may be very high in this spectrum. All signal processing must handle the full range of the signal environment until the bandwidth is reduced to encompass just the signal of interest. Then it only has to handle that signal's dynamic range. If that signal can be ac-coupled and hard-limited, it is very easy to handle. However, hard-limiting the I and Q components of a signal independently will destroy the phase and amplitude characteristics of the signal. Since the baseband signals need linear amplification, this must be accomplished with a matched set of I and Q amplifiers with tracking-gain control.
Other Considerations
Since the ZIF receiver does not have an intermediate-frequency stage, there is no basic problem in making it cover a very wide range of signal frequencies. The main limitation is the bandwidth of the preselect filter and the RF amplification stages, and the tuning range of the local oscillator. With this in mind, ZIF can be made to cover multiple bands easily.
Integrated-circuit technology allows the close matching of amplifiers that makes it practical to do the AGC job. These amplifiers must match in gain and phase over a wide gain control range. Adding to the concern is the need to make the gain control very rapid for packet communications. For example, IEEE 802.11a has a preamble length of only 16 ms, covering both AGC and dc-compensation settling times.
Low-Pass Filter Concerns
The LPF is the only line of defense for channel selectivity for the zero-IF concept. All in-band selectivity of the receiver is achieved in this filter. This does not mean that the preselect filter is not needed, but only that it is much too wide to do any good on adjacent-channel signals. The LPF must also handle leakage of local-oscillator and RF signals from the mixer. For example, the LPF passband for the 802.11a signals would be 8 MHz and the leakage, 5.3 GHz. This is almost three decades higher, so if the self-resonant frequency of the parts is below this frequency, the theoretical rejection of a five-pole Butterworth filter is not met. If the filter is active, it will have a gain element that has some gain-bandwidth product. If the leakage signals are at or above the gain bandwidth of the amplifier, the filter will, again, not have the desired response.
Also, if high-level signals are to be rejected and they are above the slew-rate limits of the amplifier, the result will be high distortion and intermodulation. So, a passive lumped-element filter or a mix of passive and active may be needed.
The output signal from the LPF will have the desired signal and noise as well as any residual out-of-channel signals. In the pure-ZIF case, the desired signal dominates and no additional filtering is necessary. This desired signal can have any level from -95 dBm to -20 dBm (referred to the antenna) and therefore a 75 plus/minus dB dynamic range has to be accommodated. It can be done by either using an 18-bit A/D with no preamplification or by having 60 dB of AGC amplification and an 8-bit A/D. A trade-off is required between the difficulty of making a high number of bits in the A/D and of meeting the needs of the balanced AGC amplifiers. If some additional filtering is needed after the A/D conversion, the additional dynamic range must be achieved with additional A/D resolution. A compromise can be reached between these positions where more A/D resolution can be used to relieve the AGC amplifier from the full burden and also allow some relaxation in the pre-A/D filtering.
The automatic gain-controlled amps are at baseband, so they are in the I and Q paths and must have identical characteristics. That is, they are controlled together and must track over the full range to within 1 dB or so. The phase shift of these amps must be matched over the whole control range, but that's not too stringent a requirement, as they can be much wider in bandwidth than the signal.
In some regards, the ZIF radio can outperform conventional superheterodyne radios. It does not have the SAW filter with its group delay to distort the signal, but on the other hand, this means that ZIF radios have slightly less adjacent-channel rejection without the high-performance SAW filter. Additionally, ZIF radios have many fewer spurious influences except for dc. Power consumption is a little higher as advanced circuit techniques are needed to combat ZIF's problems. However, the overwhelming advantage of ZIF architectures in terms of parts count is shown in Figure 3.
With respect to the proposed IEEE 802.11g standard, getting the necessary I/Q gain and phase balance in ZIF technology to support the high error vector magnitude and signal-to-noise ratio requirements of OFDM at 54 Mbits/second requires advanced circuit designs. This will include internal calibrations to handle the impairments.
Dual-Band ZIF Receivers
With a dual-band radio, as shown in Figure 4, the 2.4-GHz ISM and 5.2-GHz U-NII bands complement each other. With this radio, the best general receiver technique is to have as narrow a filter as can be achieved as close to the antenna (before any gain) as is practical. If the receiver must cover a band of frequencies, then this is the bandwidth needed in the preselect filter. If the receiver must cover two separate bands as in ISM plus U-NII, then the receiver can either be switched between two filters covering each band or the filter can be split into two passbands. The two-band problem also requires that the synthesizer switch between these two bands. The U-NII-to-ISM band switching would require an excessive tuning range in the synthesizer without some switching. This can be accomplished with a voltage-controlled oscillator element switch that controls some capacitance or inductance.
A zero-IF receiver does not have image-frequency problems, so it can theoretically handle multi-octave-input filter bandwidths. Practically, this makes the dynamic-range requirement of the receiver harder to achieve since it allows in more signal energy and harmonics in-band. Thus, very low-order intermodulation products can exist in the input band. Dual-band also requires innovative antenna techniques on the crowded end of a PCMCIA card.
Related Article
For a related article, check out "Direct Conversion: No Pain, No Gain," at www.commsdesign.com/story/OEG20020402S0032.
About the Author
Carl Andren (candren@intersil.com) is a senior systems engineer at Intersil Corp. He holds a BSEE from Stevens Institute of Technology, an MSEE from Johns Hopkins University, and an MBA from the Florida Institute of Technology.
