Direct Conversion Receivers

In this section we consider an alternative architecture to the superhet�"the direct conversion
receiver (DCR). Direct conversion receivers, also referred to as zero IF (ZIF),
were first used in amateur radio in the 1950s, then HF receivers in the 1960s and 1970s,
VHF pagers in the 1980s, 900 MHz/1800 MHz cordless and (some) cellular phones in
the 1990s, and in GPRS and 3G designs today.
The superhet is a well-tried and -tested approach to receiver implementation, having
good performance for most applications. However, it does have some disadvantages:
 It requires either additional front-end filters or a complex image reject mixer to
prevent it from receiving two frequencies simultaneously�"the wanted frequency
and an unwanted frequency (the image frequency).
 If multiple bandwidths are to be received, multiple IF filters may be required.
 The digital sampling and conversion is performed at IF and so will require
functions to work at these frequencies�"this can require considerable current as
the design frequency increases.
The DCR is directed at overcoming these problems. The principle is to inject the LO
at a frequency equal to the received signal frequency. For example, if a channel at 920
MHz was to be received, the LO would be injected into the mixer at 920 MHz.
The mixer would perform the same function as in the superhet and output the difference
of the signal and the LO. The output of the mixer, therefore, is a signal centered
on 0 Hz (DC) with a bandwidth equal to the original modulation bandwidth. This
brings in the concept of negative frequency.

If the signal is obtained simply as the output of the mixer, it is seen as a conventional
positive-only frequency where the lower sideband has been folded onto the upper
sideband—the energies of the two sidebands are inseparable. To maintain and recover
the total signal content (upper and lower sidebands), the signal must be represented in
terms of its phase components.
To represent the signal by its phase components, it is necessary to perform a transform
(Hilbert) on the incoming signal. This is achieved by splitting the signal and feeding
it to two mixers that are fed with sine and cosine LO signals. In this way an
in-phase (I) and quadrature phase (Q) representation of the signal (at baseband) is constructed.
The accuracy or quality of the signal representation is dependent on the I and
Q arm balance and the linearity of the total front end processing (see Figure 2.5).
Linearity of the receiver and spurious free generation of the LO is important, since
intermodulation and distortion products will fall at DC, in the center of the recovered
signal, unlike the superhet where such products will fall outside the IF. Second-order
distortion will rectify the envelope of an amplitude modulated signal—for example,
QPSK, π/4DQPSK, and so on to produce spurious baseband spectral energy centered
at DC. This then adds to the desired downconverted signal.
It is particularly serious if the energy is that of a large unwanted signal lying in the
receiver passband. The solution is to use balanced circuits in the RF front end, particularly
the mixer, although a balanced LNA configuration will also assist.

If the balancing is optimum, even order products will be suppressed and only odd
products created. However, even in a balanced circuit, the third harmonic of the
desired signal may downconvert the third LO overtone to create spurious DC energy,
adding to the fundamental downconverted signal. In the superhet, this downconverted
component lies in the stopband of the IF filter.
Although the even and odd order terms may themselves be small, if the same intermodulation
performance as the superhet is to be achieved, the linearities must be superior.
As circuit balance improves, the most severe problem remaining is that of DC offsets
in the stages following the mixer. DC offsets will occur in the middle of the downconverted
spectrum, and if the baseband signal contains energy at DC (or near DC) distortions/
offsets will degrade the signal quality, and SNR will be unacceptably low.
The problem can have several causes:
 Transistor mismatch in the signal path between the mixer and the I and Q
inputs to the detector.
 The LO, passing back through the front end circuits (as it is on-frequency) and
radiating from the antenna then reflects from a local object and reenters the
receiver. The re-entrant signal then mixes with the LO, and DC terms are produced
in the mixer (since sin2 and cos2 functions yield DC terms).
 A large incoming signal may leak into the LO port of the mixer and as in the
previous condition self convert to DC.
The second and third problems can be particularly challenging, since their magnitude
changes with receiver position and orientation.

DCRs were originally applied to pagers using two-tone FSK modulation. In this
application DC offsets were not a problem because no energy existed around DC—the
tones were + and -4.5 kHz either side of the carrier. The I and Q outputs could be AC
coupled to lose the DC offsets without removing significant signal energy.
In the case of GSM/GPRS and QPSK, the problem is much more acute, as signal
energy peaks to DC. After downconversion of the received signal to zero IF, these offsets
will directly add to the peak of the spectrum. It is no longer possible to null offsets
by capacitive coupling of the baseband signal path, because energy will be lost from
the spectral peak. In a 200 kHz bandwidth channel with a bit error rate (BER) requirement
of 10-3, a 5 Hz notch causes approximately 0.2 dB loss of sensitivity. A20 Hz notch
will stop the receiver working altogether.
It is necessary to measure or estimate the DC offsets and to remove (subtract) them.
This can be done as a production test step for the fixed or nonvariable offsets, with
compensating levels programmed into the digital baseband processing. Removing the
signal-induced variable offsets is more complex. An example approach would be to
average the signal level of the digitized baseband signal over a programmable time
window. The time averaging is a critical parameter to be controlled in order to differentiate
dynamic amplitude changes that result from propagation effects and changes
caused by network effects, power control, traffic content, and so on.
Analog (RF) performance depends primarily on circuit linearity usually achieved at
device level; however, this is a demanding approach both in power and complexity,
and compensation at system level should be attempted. Baseband compensation is
generally achieved as part of the digital signal processing and hence more easily
achieved. Using a basic DCR configuration, control and compensation options may be
considered (see Figure 2.7).
Receiver gain must be set to feed the received signal linearly to the ADC over an 80-
to 90-dB range. Saturation in the LPF, as well as other stages, will unacceptably
degrade a linear modulation signal—for example, QPSK, QAM, and π/4DQPSK. To
avoid this problem, gain control in the RF and baseband linear front-end stages is
employed, including the amplifier, mixer, and baseband amplifiers.
The front-end filter, or preselector, is still used to limit the RF bandwidth energy to
the LNA and mixers, although since there is now no image, no other RF filters are
required. Selectivity is achieved by use of lowpass filters in the I and Q arms, and these
may be analog (prior to digital conversion) or digital (post digital conversion). The
principle receive signal gain is now in the IQ arms at baseband, which can create the
difficulty of high low-frequency noise, caused by flicker effects or 1/f.
Another increasingly popular method of addressing the classic DCR problems is to
use a low IF or near-zero IF configuration. Instead of injecting the LO on a channel, it
is set to a small offset frequency. The offset is design-dependent, but a one- or twochannel
offset can be used or even a noninteger offset. The low IF receiver has a frequency
offset on the I and Q and so requires the baseband filter to have the positive
frequency characteristic different from the negative frequency characteristic (note that
conventional filter forms are symmetrical). This is the polyphase filter. As energy is
shifted away from 0 Hz, AC coupling may again be used, thus removing or blocking
DC offsets and low-frequency flicker noise.
This solution works well if the adjacent channel levels are not too much higher than
the wanted signal, since polyphase filter rejection is typically 30 to 40 dB.