Direct-conversion transmitters appeal to designers of wireless systems for their simplicity and low cost. Unfortunately, the simple architecture does not allow the filtering of broadband noise, images and spurious components typically executed at intermediate frequencies (IFs) in a more complex superheterodyne transmitter. For designers to migrate single-carrier base stations to multicarrier architectures using direct-conversion approaches, they must use components with high output compression and low noise. What follows is an examination of the direct-conversion approach for multicarrier WCDMA and CDMA2000, with particular attention on that critical component, the in-phase/quadrature (I/Q) modulator.
In a single-carrier WCDMA system (Fig. 1), a base station typically transmits at carrier power levels to +46 dBm (40 W). The 3GPP standard requires that the power in the adjacent and alternate channels be no greater than −45 and −50 dBc, respectively (the carrier power and adjacent/alternate channel power are both measured in a 3.84-MHz bandwidth).1 To achieve such performance, components are backed off from their maximum power levels, and predistortion techniques are typically used in the power amplifier for improved linearity. Further from the carrier, performance requirements are dominated by the noise-floor specifications (or spurious emissions).
For example, at carrier offsets to 50 MHz, noise or spurious components, measured in a 1-MHz bandwidth, can be no greater than −15 dBm. At greater offsets from the carrier, requirements are more stringent, with the worst case at 60 MHz offset from the carrier (or at the edge of the band, whichever comes first); at the 60 MHz offset, the noise floor must be no greater than −30 dBm (1-MHz bandwidth).
Figure 2 compares single-carrier (left) and multiple-carrier (right) transmitter spectra. If the same model power amplifier is used in both systems, the per-carrier must be reduced in the four-carrier system to maintain a total output-power level of +46 dBm, resulting in multiple carriers with transmit powers of +40 dBm. For both approaches, the 3GPP standard still requires adjacent- and alternate-channel power ratios of −45 and −50 dBc, respectively, and a noise floor of −30 dBm.
As the power of each carrier is reduced by 6 dB, the resulting intermodulation distortion (IMD) in adjacent channels is also reduced as the distance to the system's third-order intercept point and compression point increases. This suggests that the adjacent-channel leakage ratio (ACLR) should improve. However, because the noise floor remains relatively constant, the signal-to-noise ratio (SNR) degrades and begins to affect the ACLR (in the single-carrier case, the ACLR is dominated by distortion). Also, even though the per-carrier power is lower than for single-carrier case, the four carriers modulate each other and contribute to ACLR. The net result is that the ACLR will degrade as a particular hardware configuration is alternately driven by a single-carrier signal and by a multi-carrier signal with the same total power. For optimum performance, multicarrier systems require signal chains that have the highest possible signal-to-noise ratio.
Figure 3 shows a block diagram of a direct-conversion transmitter, with representations of the signal spectrum along the designated points in the signal chain. The dual DACs create a baseband I/Q spectrum (A), while lowpass filters following the DACs eliminate Nyquist images and noise (B). Although this noise filtering has traditionally been less critical, the emergence of low-noise I/Q modulators (with noise floors rivaling even 14- or 16-b DACs) has made noise filtering more meaningful. This suggests that the corner frequency of the filter be as close as possible to the edge of the spectrum (this will help to improve in-band noise at the antenna). However, a trade-off must be made as placing the 3-dB corner of the filter too close to the edge of the spectrum will give in-channel group delay variations and will degrade error vector magnitude (EVM).
The filtered baseband signals drive the I and Q inputs of a quadrature modulator which is also driven by a local oscillator (LO) with frequency centered at the desired output frequency. The LO is applied to the modulator's internal limiter and split into quadrature signal components. Multiplying these quadrature components together with the baseband I and Q components creates a modulated carrier centered on the LO frequency (C). Unfortunately, any unwanted DC components in the baseband I and Q signals will also be multiplied with the LO and generate LO leakage (the arrow in the center of the spectrum). The presence of this LO leakage will degrade the quality of the modulated carrrier's EVM. Because this signal component falls within the desired channel, it cannot be filtered without removing the desired signals.
The problem can be avoided by using an I/Q modulator with low LO leakage (low input offset voltages on the I and Q input ports). If the DAC and the modulator have the same DC-bias level, allowing a DC-coupled connection, it is possible to use the DAC to apply compensating offset voltages to eliminate LO leakage. However, this is only effective if the I/Q modulator's input offset voltages are stable over temperature.
Nonideal quadrature splitting of the LO and/or gain mismatch between the I and Q channels will also degrade EVM (but will not add out-of-channel spurious signals). Like LO leakage, this effect can be reduced by varying the relative amplitudes and phases of the baseband I and Q signals, although such control must also be maintained over temperature.
In a frequency-agile system, the signal chain must be designed so that carrier frequencies can be synthesized over a defined range. For example, a WCDMA base station might be designed to operate anywhere from 1930 to 1990 MHz or 2110 to 2170 MHz. The LO must tune over this range, but the modulator output cannot be filtered inside this range. Thus, post-modulator filtering can at best reduce out-of-band noise rather than in-band noise.
In Fig. 3, once the signal from the modulator has been filtered and subjected to some gain control, it is boosted in amplitude by a power-amplifier (PA) predriver and then a high-power amplifier (HPA) before being transmitted (point E). The amount of amplifier gain depends on the output power provided by the modulator. Since the PA gain also increases noise, less gain is better. Ideally, the modulator should provide the highest possible output with the lowest possible noise.
Direct-conversion transmitters are susceptible to an effect known as LO pulling which occurs if some of an HPA's output signal leaks back to the LO and causes phase modulation. The problem is potentially severe when the PA is located close the transceiver PCB, although careful layout and effective grounding can minimize the problem.
Figure 4 shows a block diagram of a superheterodyne transmitter, obviously more complex than the direct-conversion transmitter. The baseband section is similar to the direct-conversion modulator, with a filtered baseband spectrum being driven into an I/Q modulator (points A and B). However, the output of the I/Q modulator is now translated to an IF (C). At this point in the signal chain, the signal can be narrowband filtered with a selective IF filter, such as a surface-acoustic-wave (SAW) filter (point D). This filtering option is the key advantage of the superheterodyne architecture over a direct-conversion approach. Still, the LO leakage that occurs in direct-conversion transmitters can also plague superheterodyne designs, and the IF filter will not help. The superheterodyne approach allows gain control at IF, with typically better performance and lower cost than variable-gain amplifiers (VGAs) at RF.
After the signal has been narrowband filtered, it is translated to the final carrier frequency, requiring a relatively high-frequency LO offset from the final carrier by the IF. The mixing process produces sum and difference components, one of which will fall in-band (E). The operation will also result in some LO leakage and will produce a family of spurious signal products from the intermodulation of LO and IF harmonics. Careful frequency planning is required to ensure that none of these spurious signals falls within the transmission band. There is no chance to filter out spurious components that fall in-band. Also, the IF must be selected high enough so that the LO leakage signal falls well out of band. Following the signal flow in Fig. 4, the signals from the mixer are filtered (point F) and then amplified in a manner similar to the direct-conversion transmitter (point G).
Comparing the two transmitter architectures emphasizes the importance of a high-dynamic-range I/Q modulator for an effective direct-conversion transmitter. Figure 5 shows a plot of a the spectrum of a single-carrier WCDMA signal spectrum at 2140 MHz synthesized using AD9777, a 16-b dual-DAC (model AD9777) and an 700-to-2700-MHz I/Q modulator (model AD8349) with a compression point of +5 dBm and noise floor of −156 dBm/Hz. The plot indicates an ACLR of approximately −69 dBc at a carrier output power of −17 dBm (the measurement is slightly degraded by the noise floor of the spectrum analyzer). The power in the adjacent channels is dominated by spectral leakage from the modulated carrier and not from the device's noise floor. The alternate-channel power ratio is flat across the channel, indicating that it is dominated by noise.
Figure 6 shows how the ACLR of a single-carrier WCDMA signal at 1960 MHz and 2140 MHz varies with output power. An ACLR of −68 dBc occurs at a power level of approximately −15 dBm. Above this level, the ACLR degrades because of increasing distortion; below this level, the device's noise floor degrades the ACLR. The noise floor measured 20 MHz from the carrier is relatively flat with channel power, indicating degradation of SNR with decreasing carrier power.
The goal of the transmitter signal-chain designer is to select a modulator output level that provides acceptable ACLR while also satisfying in-band system noise requirements. For example, at 1960 MHz, if the output power level is set at −10 dBm, the ACLR is equal to −64 dBc. Choosing a modulator output power level of −10 dBm calls for 56 dB post-modulator gain for the desired +46 dBm base-station output power. But the modulator's noise will also be boosted by the gain. Measured in a 1-MHz bandwidth, the noise at the antenna is:
Noise (dBm/1 MHz) = −155 dBm/Hz + 10 log10(1 MHz) + 56 dB = −39 dBm
which is well within the worst-case requirements of the 3GPP spurious emissions specification of −30 dBm and even suggests that the modulator could be run at a slightly lower output power with improved ACLR.
Figure 7 shows a plot of four WCDMA carriers at 1960 MHz, synthesized by the AD8349. The per-carrier output power must be reduced to maintain the overall output power at a reasonable level. In this case, optimum ACLR of −61 dBc was achieved at a per-carrier power of −23 dBm. In Fig. 7, side-skirts are less apparent suggesting that the ACLR is dominated by the device's noise floor.
If the total output power is chosen to be −17 dBm (i.e., four carriers each at −23 dBm), 63 dB of post-modulator gain will be required to achieve total power of +46 dBm at the antenna. Since the noise floor is −155 dBm/Hz, the noise power at the antenna will be:
Noise (dBm/1 MHz) = −155 dBm/Hz + 10 log10(1 MHz) + 63 dB = −32 dBm
While this noise power level is closer to the −30 dBm limit in the 3GPP standard, it still represents reasonable margin. However, the calculation suggests that it may be prudent to run the modulator at a slightly higher output power and accept degraded ACLR.
- Standard 3GPP TS 25.104, Universal Mobile Telecommunications System (UMTS); UTRA (BS) FDD; Radio transmission and reception. Section 18.104.22.168.