Direct Conversion: No Pain, No Gain

Direct Conversion: No Pain, No Gain
The design of UHF radio receivers has received great attention over the past few years, fueled by consumers' demand for smaller, cheaper and lower-power cellular handsets. As a result, designers have had to revisit receiver architectures, available components, and integration and cost-reduction possibilities. In particular, a resurgence of interest in the direct-conversion receiver architecture is forcing designers to rethink their options.

The superheterodyne receiver, invented in 1917, has enjoyed a long run of popularity. The basic principle is to convert a band of RF signals to an intermediate frequency by mixing a tunable local oscillator (LO) with the incoming signals (Figure 1). A fixed-frequency narrow-channel filter is applied at the mixer output, followed by most of the receiver's gain and demodulation. Usually, RF filters, which pass all channels in the desired band, are placed both before and after the LNA stage.

Without the prefiltering, an image signal would also be mixed down to the passband of the IF filter, thereby interfering with the desired signal. The width of the desired band and the image-rejection characteristics available in the RF filters are traded off against the cost and size of the IF filter to determine the IF frequency.

The benefits of the superhet architecture are enormous: Most of the filtering and gain takes place at one fixed frequency, rather than requiring tunable high-Q bandpass filters or stabilized wideband gain stages. In some systems, multiple IFs are used to distribute the gain and selectivity for better linearity.

Low-IF receivers
However, the cost and size of every component matters, and getting rid of the IF filters (usually bulky and expensive surface-acoustic-wave devices) is attractive. With cellular handsets' trending toward multimode operation, superhet receivers would require a separate IF SAW filter for the channel bandwidth of each mode, increasing size, component count and cost.

The IF in a superhet receiver is usually chosen to be high enough that out-of-band image signals can be removed by prefiltering at RF. Under some conditions, a lower IF can be used. A special type of mixer, called an image-reject mixer, uses a pair of quadrature phase shifts to cause the image signal to be canceled by ultimately adding out-of-phase signals (Figure 2). The frequency plan can be chosen such that the images are actually adjacent channels. The technique is well-known and has been used in numerous receiver designs over the years. The two quadrature phase shifts can be both done in the analog domain or split into a hybrid approach.

The advantage is that the gain and filtering are done at a lower frequency than the conventional high-IF superhet. That reduces the power and opens up the possibility of integrating the filter components on-chip, thus reducing the total number of components. If the gain stage is ac-coupled, any issues relating to dc offsets should be eliminated.

One disadvantage of the near-zero-IF approach is that the receiver's polyphase filters require more components than an equivalent low-pass filter used in a direct-conversion receiver. Another is that the image cancellation is dependent on the LO quadrature accuracy (matching of the quadrature phase-shift networks and the polyphase filters), which may vary with process variations and temperature changes.

In hybrid implementations, where the image-reject function is divided into analog and digital phase-shift stages, the A/D conversion process occurs at the IF frequency. That generally requires higher power than baseband converters and more stringent control of the sampling clock, since clock jitter will degrade the conversion of an IF signal.

Direct-conversion option
Direct conversion (also called zero-IF or homodyne) is a special case of the superhet receiver. In this case, the LO is set to the same frequency as the desired RF channel. That means that the IF is zero, or dc. Now the filtering and gain can take place at dc, where gain is easier to achieve with low power, and filtering can be accomplished with on-chip resistors and capacitors instead of the expensive and bulky SAW filters.

The placement of the filter at baseband (usually split between the analog and digital domains) permits multiple filter bandwidths to be included at no penalty in board area, since the filtering is accomplished on-chip. Thus, direct conversion is the key to multimode receivers for the future.

In the ideal world, the direct-conversion receiver looks perfect (Table). In the real world, however, the benefits of direct conversion require the designer to carefully consider the non-ideal world, or receiver design, and to conquer the issues one at a time.

Most receiver designers are aware of the problem of third-order distortion products, quantified in the specification called third-order intercept point (IP3). Third-order distortion creates spurious interferers from two nearby signals at frequencies defined as M*f1 +/- N*f2, where the sum of M + N is 3.

For example, consider a receiver tuned to a desired channel f0, with two nearby input signals at f1 = f0 + Δ and f2 = f0 + 2Δ. Nonlinearity in the RF stages, for example, will create four third-order products at frequencies equal to:

2f1 + f2 = 2f0 + 2Δ + f0 + Δ= 3f0 + 3Δ. That's well away from the desired channel, f0. So it is is easily removed by filtering. Also:
2f2 + f1 = 2f0 + 4Δ + f0+ Δ= 3f0 + 5Δ, which is also well away from f0. Then:
2f2 - f1 = 2f0 + 4Δ - f0 - Δ= f0 + 3Δ, which is removed by a narrow channel filter. Finally:
2f1 - f2 = 2f0 + 2Δ - f0 - 2Δ= f0, or the same frequency as the desired channel.

This last intermodulation product cannot be removed by filtering and must be minimized by sufficient linearity in the LNA and mixer stages, regardless of receiver architecture.

Direct-conversion issues
Several well-known issues that have historically plagued direct-conversion receivers are self-detection due to LO-RF leakage, dc offset and AM detection. It's worth noting that some "direct-conversion problems" can also affect low-IF superhets in the real world.

The first problem is caused by the fact that the local oscillator in a direct-conversion receiver is at the exact same frequency as the desired signal. Hence, any leakage of the LO signal to the RF input or the antenna will pass through the entire signal chain, which appears as a dc offset. The basic operation of a direct-conversion receiver can be described as mixing an input signal frequency of (fC + Δ), where (Δ) is the bandwidth of the modulation, with a local oscillator at fLO, yielding an output at:

f_MIXOUT = (f_C + Δ- f_LO) and (f_C + Δ + f_LO)

The second term is at a frequency twice that of the carrier frequency and can be filtered out easily. The first term is much more interesting, since f_LO = f_C, and substitution yields:

f_MIXOUT = f_LO + Δ- f_LO =
Δ

That means the modulation has been converted to a band from dc to the modulation bandwidth, where gain, filtering and A/D conversion are readily accomplished. The dc-offset problem occurs when some of the on-channel LO (at f_C) leaks to the mixer RF port, creating the effect that: f_LO - f_LO = 0 (or dc)!

Shielding and other layout techniques are often used to reduce this effect. Another approach is to convert an off-channel (or even out-of-band) LO signal to an on-channel LO inside the chip, reducing leakage paths. Operating the LO at half (or twice) the necessary injection frequency is a good solution for single-band applications; a regenerative divider simplifies multiband designs.

Once the dc offset due to LO-RF leakage has been reduced, the second problem arises: inherent dc offset in the baseband amplifier stages, and its drift over temperature. Here, the best solution is to employ extreme care in the design of the gain stages and to make sure that enough gain-but not too much-is provided. Excessive gain in the baseband section can cause offsets that can be corrected momentarily but that may drift excessively and require additional temperature compensation.

There are three possible methods by which offsets may be handled in the receiver: continuous feedback, track-and-hold and open-loop.

The continuous-feedback scheme (in software or hardware) attempts to null dc error at the mixer output. That generally requires tight coupling between the baseband processor and software and makes it difficult to mate an RF IC from one vendor with a baseband controller and software from another vendor.

In the "track and hold" method, the dc offset is estimated just prior to the active burst (track mode) and then stored (hold mode) during the active burst. Such schemes are generally completely integrated with the radio IC and can be made transparent to the user by locally generating all the necessary timing signals. Practical issues with the scheme include dealing with multislot data (GPRS) where the baseband gain may be changing on a slot-by-slot basis (without adequate time to recalibrate) and also ensuring that the dc estimate obtained during the track mode is accurate. Such schemes can be implemented in either digital or the analog domains.

Latest-generation radios using the open-loop approach have substantially lower dc offsets and can operate with lower-performance A/D converters (typically 60 to 65 dB of available dynamic range), without any special software requirements.

The third problem in direct-conversion receivers, AM detection, is also a problem in near-zero-IF receivers, especially in cellular handset applications, and is worth examining in detail.

Any mixer will detect the modulation envelope of a large-enough AM input signal and reproduce it at its output. That effect arises from a number of sources, such as RF-to-LO coupling and second-order distortion (quantified and expressed as the second-order intercept point, or IP2). In a cellular system, it can affect real-world performance.

For example, in the GSM system, it is possible for basestations in the same geographic area to be unsynchronized in their burst timing. GSM systems have a means for calibrating the frequency and phase of the handset receiver relative to the basestation by transmitting a known bit sequence at a fixed time in the middle of each transmission burst. This "midamble" is critical for the receiver to perform equalization of the channel as well as for fine-tuning the receiver's automatic frequency-control loop.

Consider what happens when a handset is tuned to its basestation and a burst from another nearby basestation occurs in the middle of the desired signal. A direct-conversion receiver with poor rejection of second-order distortion products will detect the envelope of this burst, producing a step in the dc output of the receiver.

The dc error will serve to consume dynamic range in the receiver and confuse the equalizer and detection algorithms, leading to bit errors and possibly to dropped calls. A superhet receiver (either low-IF or high-IF) is immune to the problem, since dc steps do not pass through the IF filter (which passes only IF signals), while a direct-conversion receiver for GSM must pass signal content down to dc. The GMSK (Gaussian minimum-shift keying) modulation used in GSM is a form of phase modulation and does not contain significant amplitude modulation. The envelope is constant after the initial turn-on.

GSM Type Approval testing tests for AM suppression to ensure that GSM/GPRS terminals can continue to function properly when nearby base-stations transmit unsynchronized bursts. The GSM AM-suppression test allows 3-dB reduction in sensitivity for a unsynchronized burst of -31 dBm. This can be replicated by a classic two-tone IP2 test, where each tone has a value of -34 dBm. The required signal level is -99 dBm, and the required C/I level to meet a 3-dB reduction in sensitivity is about 9 dB. Therefore, the required two-tone IP2 would need to be -34 + (99 + 9 - 34) dBm or +40 dBm referred to the antenna. Most well-designed receivers meet that specification.

Effects of evolving GSM
However, the GSM world is evolving (Figure 3). Some operators are deploying the Edge (Enhanced Data Rates for GSM Evolution) system, which uses the same 200-kHz channels as GSM but adopts 8-PSK modulation to increase the available bit rate in the channel to 384 kbits/s (the original GSM/GPRS system uses GMSK modulation).

Of the various schemes, only asynchronous GSM burst case is currently covered in type approval testing, but of course consideration needs to be given to real-world performance as Edge and IS-95 CDMA signals enter the picture.

As the number and types of interfering signals in the band increase, the impact of those signals on the GSM system performance-specifically the receiver performance-must be assessed. Both IS-95 CDMA (QPSK) and Edge (8-PSK) signals have significant AM content in their envelopes. Since those envelopes will be detected by any second-order distortion in the receiver, it's important to assess the impact of the AM detection through careful testing and signal analysis.

Remember, AM detection an issue requiring attention in direct-conversion receivers, since the dc step caused by an unsynchronized GSM burst is within the passband of the baseband channel filters. But AM detection of other types of signals will affect other receivers as well.

The IS-95 spread-spectrum signal has wideband AM content. Since the pseudo-noise code used to spread an IS-95 signal creates approximately constant power spectral density across the 1.25-MHz channel, first principles suggest that approximately 1/6 of the AM signal power will fall within a GSM receiver's 200-kHz bandwidth, and measurements have confirmed that estimate. The received CDMA interferer can be as high as -25 dBm, since there is no prefiltering possible in the front end. From that, the required receiver IP2 can be calculated to be approximately +44 dBm for the GSM system, which is 4 dB more severe than in GSM-to-GSM AM detection.

The IS-95 signal is sufficiently wideband that this requirement applies to receivers using a "low-IF" architecture as well as to direct-conversion receivers. That is because the detected envelope of the IS-95 signal contains significant energy in a 200-kHz band centered at 100 kHz (an IF used in some low-IF receivers). Therefore the direct-conversion receiver (and, as shown, the low-IF receiver as well) must achieve IP2 in excess of those calculated above.

Jon Strange is a director at Analog Devices' Kent design center. He has a BSc in Physics from Durham University and an MSc in Electrical Engineering from Edinburgh University. Jon can be reached at jon.strange@analog.com.

Doug Grant is business development director of the RF and wireless systems business unit of Analog Devices. He has a BSEE from the University of Lowell and attended graduate school at Northeastern University.Soug can be reached at doug.grant@analog.com.