With the help of wide-dynamic-range components, direct-conversion architectures offer significant advantages in simplicity and cost compared to superheterodyne approaches.
Base-station receivers for next-generation wireless systems must deliver higher performance at lower cost than their predecessors. The direct-conversion receiver architecture is a good candidate to satisfy these conflicting requirements. Although the approach has been applied to different designs in the past, performance has been compromised by limitations in available hardware, including demodulators. Fortunately, the improved performance of commercial integrated-circuit (IC) quadrature demodulators make direct-conversion receiver designs viable alternatives to traditional superheterodyne receiver architectures.
To better understand the benefits of direct conversion, it might make sense to compare the receiver approach to a superheterodyne system (Fig. 1), commonly used because of its high selectivity and sensitivity. In a superheterodyne receiver, the received RF signal is filtered by the first RF preselection filter to remove out-of-band signals, and then boosted by the low-noise amplifier (LNA). The second RF preselection filter at the LNA output provides additional filtering to attenuate undesirable signals at the image frequency. The resulting signal is then translated to a lower, intermediate frequency (IF) by a downconversion mixer in conjunction with a local oscillator (LO). The IF must be sufficiently high that the image channel falls is within the filter's stopband. Such image-rejection considerations usually dictate that the IF should be on the order of 10 percent of the carrier frequency. The RF preselection filters serve to remove out-of-band energy and reject the image-band signals. Since a superheterodyne receiver performs the channel-filtering function in the IF and baseband stages, aggressive dynamic-range requirements are imposed on the components in these stages.
For superhetero-dyne receivers in base stations, a fixed-gain LNA is typically used for initial amplification of the received signals. The entire passband, including noise, is translated in frequency to a fixed IF. For the frequency downconversion, a passive (diode) mixer is most often utilized in order to meet the dynamic-range requirements of high linearity and low noise, although high LO power (greater than +10 dBm) is needed to drive such a mixer. Poor LO-to-IF isolation, typical of passive mixers, complicates the LO filtering in the receiver's IF section. At the IF output of the mixer, the desired signal channel always resides at the center of the IF channel-select filter, which is used to remove unwanted adjacent or alternate channels.
Following the IF channel-select filter, the desired channel is boosted by a variable-gain amplifier (VGA) and then demodulated to baseband for further signal processing. The high-quality-factor (Q) IF channel-select filter passes desired signals and rejects unwanted signals, including larger-amplitude alternate-channel signals. Unfortunately, such selective filters are expensive and add undue cost to the superheterodyne receiver. Moreover, high-Q filters are typically accompanied by high insertion loss requiring additional gain in the LNA and mixer stages to offset the filter loss and lower noise figure in the VGA.
Since the LNA gain is fixed in the base-station receiver, the mixer in particular must achieve very high linearity to meet the system's strict dynamic-range requirements. Moreover, the IF channel-select filter has a frequency response precisely tuned to the required channel bandwidth. The inflexibility of the IF channel-select filter limits the receiver hardware to a single RF standard. Because of the proliferation of standards for wireless communications, however, new receiver systems must support a variety of different standards seamlessly and cost-effectively, with limited cost budget for any one standard.
The direct-conversion receiver architecture can achieve the goals of a superheterodyne design, but with considerably less complexity (Fig. 2). In this system, the received signals are amplified with a fixed-gain LNA after the first RF preselection filter. Subsequently, the RF signals are directly downconverted to in-phase (I) and quadrature (Q) baseband signals without an intervening IF stage. The requirements for the second RF preselection filter are less stringent than for the first, because there is no image channel. In practice, an inexpensive RF bandpass filter can prevent strong out-of-band signals from overloading the I/Q demodulator. . After the RF signals are demodulated to baseband, individual channel selection is performed using a baseband channel-select filter. The baseband filter is more compact and less expensive than the superheterodyne receiver's IF channel-select filter. In addition, the baseband channel-select filter can be designed with variable bandwidth, facilitating multi-mode or multi-standard operations.
Although baseband channel-select filters offer a great deal of flexibility, the composite baseband signals contain all of the adjacent-channel blocking signals that are normally filtered before they reach the I/Q demodulator (see Fig. 1). As a result, the direct-conversion-receiver's I/Q demodulator must provide a dynamic range as wide as 80 dB.
Fortunately, the LT5515 and LT5516 I/Q demodulators from Linear Technology (Milpitas, CA) are among a handful of commercial products that provide this kind of performance. The two ICs each integrate the functionality of an RF signal splitter, a precision quadrature LO signal splitter and two high linearity downconverting mixers. The chips directly downconvert an RF signal to baseband, and demodulate the in-phase (I) and quadrature (Q) signal components. The devices' matched I and Q channels ensure precise gain and phase matching, so that significantly less calibration is required. The LT5515 operates from 1.5 to 2.5 GHz while the LT5516 handles an RF input-signal range from 0.8 to 1.5 GHz. The chips also integrate single-pole, lowpass filters with 260-MHz bandwidth on each of the I and Q channels (see table).
Both direct-conversion quadrature modulators achieve impressive amplitude and phase balance between signal channels. The gain levels between I and Q arms of the LT5515, for example, are maintained within 0.3 dB of each other, while the phase balance is within 1 deg. For the lower-frequency LT5516 quadrature demodulator, the gain matching between the I and Q channels is within 0.2 dB while the phase matching is within 1 deg. or less. The devices are rated for RF and LO differential voltages of ±2 V (an equivalent level of +10 dBm) and maximum power-supply voltage of +5.5 VDC. They are rated for operating temperatures from −40 to +85°C and are supplied in 16-lead plastic QFN packages with exposed leads; the square packages measure just 4 mm on a side.
The efficient demodulators provide a shutdown mode in which only 20 µm current is consumed. The turn-off and turn-on times for the shutdown mode are typically 120 and 650 ns, respectively.
The LT5515 and LT5516 demodulators are ideal for receivers requiring good linearity and wide dynamic range, such as wireless base stations (for GSM, CDMA, WCDMA, etc.) and wireless infrastructure equipment, as well as instrumentation applications. Direct-conversion receiver ICs such as the quadrature demodulators eliminate the need for additional IF stages and relax the demands on high-frequency filters, especially by eliminating the IF channel-select filter. With their +20 dBm input third-order intercept (IIP3) and +50 dBm input second-order intercept (IIP2), the quadrature demodulators meet the strict dynamic-range requirements of base-station receivers.
One concern with direct-conversion receiver architectures is spurious LO leakage. This problem arises when a small amount of LO energy is coupled to the I/Q demodulator input, either from the antenna or by means of another path. The LO leakage can mix with LO itself to generate DC offset. Depending upon the LO leakage path, the carrier feedthrough may superimpose large, possibly time-varying DC errors on the desired baseband signals. In the base-station infrastructure, however, because the receiver systems are typically stationary, the DC offset caused by LO self-mixing is likely static, rather than time varying. Because the LT5515 and LT5516 employ active rather than passive mixers, only −5 dBm LO power level is required rather than the +10 dBm typically needed with passive mixers. Thanks to good isolation between the LO and RF ports, the LO leakage is minimized to a mere −46 dBm for the LT5515 and −65 dBm for the LT5516. Consequently, only a few millivolts of static DC offset may result from the LO self-mixing.
Another concern of the direct-conversion approach is DC offsets caused by device mismatches. Mismatch-induced DC offsets can originate from the quadrature demodulator and/or the VGA. DC offsets at the quadrature demodulator's outputs will not in themselves cause receiver malfunctions or performance degradation. However, due to the limited VGA voltage headroom, a few millivolts of DC offset may be enough to significantly reduce the signal swing or possibly saturate the VGA when it operates in high-gain mode with the gain as high as 60 dB, thus degrading the receiver's effective dynamic range. To handle a large blocking signal, the LNA gain is usually limited to the 20-dB range, so that the desired signal level reaching the mixer under weak-signal conditions may be on the order of a few hundred microvolts. Thus, the accumulated DC offset with reference to the VGA input must be controlled to less than that level. DC offset cancellation or AC input coupling is required to properly operate the VGA for further baseband signal processing.
Most infrastructure base stations work in full-duplex mode, albeit with receiver and transmitter at different frequencies. The settling time of the DC voltages is less of a concern in this type of receiver system. In many modern wireless receiver systems, the baseband signals contain little low-frequency information. This allows the I- and Q-channel outputs of the LT5515 and LT5516 demodulators to be AC-coupled to the baseband filters or VGA through a blocking capacitor, effectively eliminating DC offsets. Since each of the I-channel and Q-channel outputs of the LT5515/LT5516 is internally connected to the supply voltage through a 60-Ω resistor, the resulting output highpass filter's −3-dB rolloff frequency is defined by the RC constant of the blocking capacitor and the output resistive load Rload, for Rload sufficiently large (much greater than 60 Ω).
When DC coupling of the LT5515/LT5516 to the baseband circuits is required, a digital offset removal method can be employed at the baseband VGA inputs. The DC offsets may be estimated and removed using the baseband processor at each VGA setting. Although the DC offset will not impact the RF performance of the receiver, it must be cancelled in order for the VGA to operate properly. The spectrum loss around DC can be as low as a few Hertz. For a half duplex system, the DC offsets can be separated using an adaptive approach combining carrier recovery, symbol timing recovery, automatic gain control and data detection in the baseband. Typically, in the receiver system, the preamble in the frame structure has a known DC content that allows adaptive, frame-by-frame DC offset removal. With the LO running at −5 dBm, the output DC offsets of LT5516 and LT5515 are as low as 1 mV and 4 mV, respectively. These low offset voltages allow the receiver to implement the offset cancellation with a low-cost analog to digital converter.
Another concern with direct-conversion receiver architectures is even-order distortion products. In a conventional superheterodyne receiver, second-order distortion terms usually fall out of band and can be easily filtered. However, in a direct-conversion receiver, even-order distortion, particularly second-order products, will cause in-band interference. For example, when two strong interferers with frequency spacing close to the channel bandwidth are present at the input of the quadrature demodulator, the second-order nonlinearity of the demodulator will produce a low-frequency intermodulation product. This distortion product falls in the baseband spectrum and cannot be filtered out in the subsequent baseband signal processing. Consequently, excellent IIP2 is a prerequisite for good performance in a direct-conversion receiver. The presence of mismatches in the mixers of the demodulator and LO signal paths may result in in-band second-order intermodulation products. The second-order harmonic of the input RF signals (from second-order distortion of the RF amplifier) may also be mixed with the second harmonic of the LO signal to produce a similar effect. The high IIP2 of the LT5515 and LT5516 (+51 and +52 dBm, respectively), therefore, are important to prevent even-order intermodulation from corrupting the baseband signals. This performance can be further enhanced by properly filtering the unwanted high-frequency mixing products at the I and Q outputs. This effectively prevents the unwanted mixing products from coupling back to the demodulator to generate in-band second-order intermodulation. A convenient approach is to terminate each output with a shunt capacitor. The capacitor value can be optimized depending upon the operating frequency and the specific printed-circuit-board (PCB) layout.
The design of high-performance direct-conversion receiver systems is at the leading edge of modern base-station receiver development. Although direct-conversion receiver approaches have been studied for decades, only recently have available high-performance components made the direct-conversion architecture practical for a wide range of wireless applications.