Вы находитесь на странице: 1из 23

Chapter 4 Modulation and Demodulation

4.1 Introduction
There is intense activity to drive down the cost, size and power drain of wireless terminal units. Most of the effort is focused on Frequency Conversion and the associated modulation and demodulation schemes. The major cellular radio systems are distinguished by the type of modulation used and how the channels are partitioned. Modulation schemes in wireless can be categorized as being either analog or digital. In analog modulation the RF signal can have a continuous range of values but in digital modulation the output has a number of prescribed discrete states. The major modulation schemes are FM Frequency modulation PM Phase modulation MSK Minimum shift keying GMSK Minimum shift keying using Gaussian filtered data. /4 DQPSK pi on 4 Differential Encoded Quadrature (or Quadraphase) Phase Shift Keying. OQPSK Offset quadrature (or quadraphase) phase shift keying FM and the similar PM modulation schemes are used in Analog Radio. The other schemes are used in Digital Radio. The FM, GMSK and PM modulation techniques produce constant RF envelopes and so no information is contained in the amplitude of the signal. Therefore errors introduced into the amplitude of the system are of no significance and so efficient saturating mode amplifiers such as class C can be used. In contrast the MSK and /4 DQPSK modulation techniques do not result in constant RF envelopes and so information is contained in the amplitude of the RF signal. The primary tradeoff in the design of modulation and demodulation schemes is the tradeoff between image rejection and adjacent channel suppression. In the nearly universal heterodyne conversion architecture and with a low IF (the RF and LO are close in frequency) the image corrupts the desired converted signal. However it is much easier to suppress the adjacent channels as the fractional bandwidth required of the bandpass filters is relatively large. With a high IF system (higher in frequency than the bandwidth of the preselected RF signal) the IF filter is smaller but requires greater control on the tolerances of the IF components. It is also easier to implement subsequent stages, either digitally or using analog ASIC technology. The first IF filter, e.g. a surface acoustic wave (SAW) filter, is bulky, and as of 1998, is one of the limiting factors in determining the aspect ratio and size of cellular phones. There is great interest in

34

developing alternative filter technologies or developing schemes such as direct conversion that eliminate the IF altogether. As well, the first mixing stage also generates image signals that must be isolated from the RF LNA. This requires an image rejection filter which now must be placed off-chip. This introduces constraints on the LNA as it must be designed to drive a 50 load. There are several different forms of what has come to be called direct conversion (or sometimes just dicon). The central attribute is that the multiple conversion stages of superheterodyne conversion are replaced by a single conversion stage which converts the RF directly to baseband. Superheterodyne conversion involves RF frequency amplification in a tuned stage, followed by a mixing stage with an offset local oscillator to a lower (but still much higher than baseband) signal. In what is called the IF Strip, this IF signal is amplified and bandpass filtered to select the main channel and to a lesser extent its neighbors. Then a second mixing stage delivers the desired baseband signal. The first IF (after the first mixer) is generally around 90 MHz but higher frequency IFs of 150 MHz are also used with a reduction in the cost and size of the components operating on this signal. Amplification and filtering at these high frequencies increases power dissipation as the IF amplifiers must be biased at high currents to achieve the necessary linearity but also to drive the many large passive filter components, many of which must be off-chip. The hope is that by using a single stage much of the cost can be avoided through elimination of components. Also it is only possible to use CMOS circuitry throughout the receive path if direct conversion is used. Thus, as well as eliminating the number of components, a very low cost technology would be used. The potential cost reduction is the driving force behind the intense activity. Since the desired signal may be very weak compared to other potentially interfering signals, a large dynamic range is required. This implies the use of higher power levels in amplifiers, and passive filters instead of active filters. Hence the great activity in direct conversion receivers. The challenge is formidable as superheterodyne mixing has excellent performance as measured by noise levels and power consumption. There are a number of significant shortcomings in all of the direct conversion schemes but the performance of the best of these systems is very close to being acceptable. There is considerable commercial interest so it is possible that a cellular radio product using direct conversion could be announced at any time. It is also possible that it may take many years for the deleterious effects to be overcome. The review that follows describes the various direct conversion techniques and their limitations.

4.2 Frequency Conversion


Frequency conversion is the process of converting information at one frequency (present in the form of a modulated carrier) to another frequency. The second frequency is either higher, in the case of frequency upconversion, where it is more easily transmitted, or lower, called frequency downconversion, where it is more easily captured. Capture of the downconverted signal is nearly always by an A/D converter. Frequency conversion occurs with any nonlinear element. In Figure 1(a) a nonlinear device is driven by two signals at S and P. The larger signal, here P, is called the pump and the other the signal or RF. The pump is also referred to as the local oscillator (LO). The spectrum of the signals present in the circuit is shown in Figure 1(b). The aim here is to produce a signal at the difference frequency (or intermediate frequency IF) with the same modulation, and hence the same information, as the original RF signal. In order to suppress the pump and the other mixing tones the balanced mixer circuits shown Figure 2 are 35

used. The symmetrical (or balanced) nature of these circuits means that only differential mode signals at the input of the common source differential pairs can appear at the output. Thus the largest signal present, the LO, is suppressed. An exception is the circuit in (c) where RF and IF filters are used to suppress the LO. In addition the circuits in (b), (c) and (d) also suppress the image signal (2fP-fS). The circuit shown in Figure 2b) is the basic form of the Gilbert mixer and is the preferred form for monolithic integration. Other Gilbert mixer topologies are shown in Figure 2(c) and Figure 2(d).

RS
SIGNAL PUMP

VM cos(M t) VCcos(Ct)

NONLINEAR DEVICE

LOAD

2C

3C

C - M

2 C - M 3 C - M C+ M 3 C+ M 2 C+ M

FREQUENCY

Figure 1: schematic and spectrum of a diode mixer.


IF BALUN IF

RF

RF FILTER

IF FILTER

IF

LO

LO BALUN

LO

LO FILTER

GATE BIAS

RF

RF BALUN

(a)

(b)

36

(c)

(d)

Figure 2: Transistor-based mixer circuits: (a) Single ended MESFET mixer with LO, RF and IF bandpass filters; (b) Gilbert MESFET mixer with baluns producing differential sLO and RF signals; (c) MOSFET Gilbert mixer configuration with shared RF stage; and (d) MOSFET Gilbert mixer with dual gate mixer.

37

4.3 Modulation Methods


4.3.1 Quadrature Modulation
The generalized quadrature modulation equation:
c

describes nearly all modulation processes. Here i(t) and q(t) embody the particular modulation i q(t) embody the particular modulation rule for phase, and c is the carrier radian frequency. The spectrum produced by this modulation formula could include a large component at frequency c as well as sidebands on either side of this frequency. It is desirable not to transmit the carrier (suppress the carrier) and to transmit only one of the sidebands. This type of modulation is called suppressed carrier single sideband SCSS modulation. This is achieved by choosing particular forms for i(t), q i q(t). Consider the quadrature modulator shown in Figure 3 where the carrier frequency c produced by the voltage controlled oscillator (VCO). Actually two versions of the carrier are produced and these are ideally exactly 90o out of phase, that is they are in quadrature. Let i(t) and q(t) be finite bandwidth signals centered at radian frequency m i q(t) so that
i (t ) = cos( m t ) and q (t ) = sin( m t )

Then the intermediate signals resulting after the multiplication of i (t ) and q (t ) are 1 a (t ) = cos[( c + m )t ] + cos[( c m )t ] 2 and 1 b(t ) = cos[( c m )t ] + cos[( c + m )t ] 2 So that the combined signal at the output is
s(t ) = a (t ) + b(t ) = cos[( c + m )t ]

and so as well as the carrier being suppressed so is the lower sideband. This lower sideband is also referred to as the image. In modulators it is important to suppress this image and in demodulators it is important that undesired signals at the image frequency are not converted along with signals in the desired spectrum.

P IH 6G

FE DCB

g f e dc #96b

x(t) = i(t) cos[c t

i(t)]

+ q(t

a Y X R VT S R `WU#Q

q(t)]

A @ 8 75 #964

2  0  & $ " !    31)('%# 

38

c-m c+ m
i(t) a(t)

c FREQUENCY
s(t)

c+ m c FREQUENCY

VCO

90 b(t)

q(t)

c+ m c FREQUENCY

c-m

Figure 3 Quadrature modulator showing intermediate spectra.

4.3.2 BPSK -- Binary Phase Shift Keying


Binary phase shift keying (BPSK) is a spectrally inefficient scheme with a binary data rate of 512 bps or 1.2 kbps modulating the carrier at 900 MHz (for one paging band) by 4.5 kHz. This is a high modulation index and results in a symmetrical spectrum with two peaks offset from the carrier with almost a null at the carrier, see Figure 4. Each peak corresponds to a one or zero and if quadrature mixing was not used it would not be possible to discriminate between them. Quadrature mixing is used in heterodyne receivers for image suppression and in some sense the two peaks are different channels and appear as images channels. In zero-IF conversion, quadrature mixing is used to discriminate between the data ones and zeros. This is spectrally inefficient but overcomes limitations of zero IF direct conversion as little information is lost by using high pass filtering. It is ideally suited to low power applications and single chip implementations with a reference crystal and off-chip inductors for the RF amplifier and quadrature phase shift. The typical signal flow is from an antenna, through an RF tuned amplifier, quadrature mixing to produce I and Q channels which are then low pass filtered with a cutoff of 10 kHz or so. The filtered I and Q channels are then either high pass filtered, or more commonly, integrated over the duration of a bit. In the most sensitive scheme the I and Q channels are over sampled at a multiple of the bit rate and the signal correlated with the expected zero-crossing of the 4.5 kHz data rate. The binary phase shift keyed spectrum is a relatively inefficient use of spectrum. A far more efficient spectrum is shown in Figure 5(b) for the types of modulation schemes used in a cellular radio. The price for increased spectral efficiency is a much more complicated circuit with higher power consumption. The spectral efficiency of paging signals is so low that it will not be considered for wireless phone communication. Thus schemes that are based on zero-IF must be concerned with distortion at the very low baseband frequencies. Zero-IF down conversion and separation of the signal into quadrature components reduces the demand placed on A/D 39

converters as each quadrature signal, the I and Q signal, occupies half of the bandwidth of the original RF signal.

Figure 4 Direct conversion of an RF signal.

(a)

(b) Figure 5 Two common RF spectra: (a) a binary FSK signal as used in a pager; and (b) a far more spectrally efficient cellular wireless signal.

40

4.3.3 Quadraphase Phase Shift Keying


Binary phase shift keying has a maximum spectral efficiency of 1 bit per second per hertz (1 bit s 1 Hz 1). In wireless systems good spectral efficiency is desirable with the immediate effect that fewer basestations are required. The candidate schemes for increasing spectral efficiency include 64-QAM with a spectral efficiency of around 4 bit s 1 Hz 1 and 256-QAM with a spectral efficiency of around 6 bit s 1 Hz 1. QAM stands for quadrature amplitude modulation and there are multiple phase and amplitude states each of which specifies a particular symbol. Each symbol represents two or more bits of information. These high order schemes work well in a telephone line environment where the channel is unchanging and equalization can be performed at startup. The dominant characteristic of the wireless channel is the existence of deep fades. Fades can be viewed as deep amplitude modulation and so it is difficult to transfer information in the amplitude of a carrier. Because of this, phase modulation schemes falling in the class of Mary Phase Shift Keying (MPSK) is most appropriate. In mobile environments there are just a few modulation formats that have been found acceptable. These all fall in the class of quadraphase phase shift keying (QPSK) (also called quadrature phase shift keying). The characteristic of QPSK is that there are four allowable phase states per symbol period and so two bits of information are transmitted per change in the characteristic of the modulated signal. There are many other four-state PSK schemes. In current wireless systems it suffices to study just a few of these: / 4DQPSK and OQPSK both derived from QPSK; and GMSK derived from MSK. The modulation schemes are characterized in terms of a constellation diagram. This characterization is integrally related to quadrature modulation as described in Figure 6. The constellation is the result of plotting the i (t ) and q (t ) in the generalized modulation circuit of Figure 6. More commonly we refer to these quantities as I and Q . However I and Q can take on several meanings depending on what we view of the modulation process that we are referring to. In the QPSK the modulation circuit schematic of Figure 6, the input bit stream is converted into two parallel bit streams. Thus a two bit sequence in the serial bit stream becomes one I K bit and one QK bit. The ( I K , QK ) pair constitutes the K th symbol. I and Q are plotted on an axes at right angles. A modulation scheme with four allowable states is shown in Figure 7(b). In the absence of wave shaping circuits i (t ) and q (t ) have very sharp transitions and the paths shown in Figure 7 are almost instantaneous. This however leads to large spectral spreads in the modulated waveform s(t ) . So to limit the spectrum of s(t ) , the shape of i (t ) and q (t ) is controlled usually by filtering. So a pulse shaping circuit changes binary information into a more smoothly varying signal. Each transition or path in Figure 7 represents the transfer of a symbol or minimum piece of information. In the modulation scheme shown here, there are three possible transitions from each point in the constellation in addition to the possibility of no transition. Thus each symbol contains two bits. Thus the maximum efficiency of this type of modulation scheme is 2 bits s 1 Hz 1. What is actually achieved depends on the pulse shaping circuits and on what criteria is used to establish the bandwidth of s(t ) . Modulation schemes have relative merits in terms of spectral efficiency, tolerance to fading, carrier recovery, spectral spreading in nonlinear circuitry and many other issues that are the real of communication system theorists.

41

The waveforms corresponding to the state transitions shown in Figure 7(b) most immediately affect the bandwidth of s(t ) and the ability to demodulate the signal. The signal trajectories through the origin are particularly troublesome and it is difficult to achieve a functional wireless communication system if this is allowed. The ability to demodulate the signals is equivalent to being able to reconstruct the original constellation diagram of the modulation signal. The transitions through the origin implies that there is significant amplitude variation in the RF signal. Also, the 180o instantaneous phase changes leads to considerable spectral regrowth.

IK
SERIAL BIT STREAM

WAVEFORM SHAPING

i(t)

a(t)

SERIAL TO PARALLEL

s(t) VCO
WAVEFORM SHAPING

90 q(t) b(t)

QK

Figure 6: Quadrature modulation circuit indicating the role of pulse shaping.

Q 1 START I FINISH 3

Figure 7: The constellation diagram of (a) a binary modulation scheme; and (b) a four-state phase modulation scheme.

4.3.4 Differential Quadraphase Phase Shift Keying


Reconstruction of a QPSK signal is difficult as the phase of the modulated signal must be precisely extracted. This is possible with carrier recovery but this is particular troublesome in a multipath environment where constructive and destructive interference of multiple paths can result in rapid additional phase rotations. The solution is to use differential coding so that

42

information in the modulated signal is contained in changes in phase rather than in the absolute phase. In addition, ensuring that the phase transitions avoid a signal trajectory through the origin led to the development of / 4DQPSK modulation. So we must distinguish between the input bit stream now denoted by pairs of bits (OK , E K ) and the modulation symbol denoted by the pair of bits ( I K , QK ) . The input bits specifies a phase change as given in

Table 1 and the ( I K , QK ) pair is given by


I K = I K 1 cos K QK 1 sin K QK = I K 1 sin K QK 1 cos K

These equations describe the constellation diagram shown in Figure 8(a). Where the allowable transitions rotates according to the last transistions. For example the transitions shown in Figure 8(b) for 6 successive time intervals describes the input bit sequence 101100011111 . The schematic of the modulator is shown in Figure 9. Incorporating the modulation as changes in phase rather than absolute phase makes the modulation scheme relatively easy to demodulate. The spectral efficiency of / 4DQPSK as implemented in the DAMPS system (it depends slightly on the filter parameters) is 1.62 bit s 1 Hz 1. Table 1: Phase changes in a / 4DQPSK modulation scheme.
OK 1 0 0 1 EK 1 1 0 0 3 / 4 3 / 4 /4 / 4

10
2

1 5

START

00 11 01
I

4 3 6

I FINISH

(a)

(b)

43

Figure 8: Constellation diagram of / 4DQPSK modulation: (a) state transition for one time interval; and (b) state transitions for 6 symbol intervals.

IK
SERIAL BIT STREAM DIFFERENTIAL SERIAL TO PARALLEL

WAVEFORM SHAPING

i(t)

a(t)

(OK, EK)

s(t) VCO
WAVEFORM SHAPING

90 q(t) b(t)

QK

Figure 9: / 4DQPSK modulator.

4.3.5 Offset Quadraphase Phase Shift Keying


The Offset Quadraphase Phase Shift Keying (OQPSK) modulation scheme is used in the CDMA system. This is also a scheme to avoid IQ transistions that do not pass through the origin on the constellation diagram, see Figure. As in all QPSK scheme there are two bits per symbol but now one bit is used to directly modulated the RF signal whereas the other bit is delayed by half a symbol period as shown in Figure 10. The maximum phase change for a bit transition is 90o and as the I K and QK are delayed a total phase change of 180o is possible during one symbol. The constellation diagram is shown in Figure 11.

IK
SERIAL BIT STREAM

WAVEFORM SHAPING

i(t)

a(t)

SERIAL TO PARALLEL

s(t) VCO
HALF BIT DELAY
WAVEFORM SHAPING

90

QK

q(t) Q

b(t)

Figure 10: OQPSK modulator.

44

Figure 11: The constellation diagram of OQPSK.

4.3.6 Guassian Minimum Shift Keying


Guassian minimum Shift Keying (GMSK) is the modulation scheme used in the GSM cellular wireless system and is a variant of MSK with tightly specified guassian low pass filter requirements. GMSK modulation is also used in the Digital European Cordless Telephone (DECT) standard. The spectral efficiency of GMSK is implemented in the GSM system (it depends slightly on the guassian filter parameters) is 1.35 bit s 1 Hz 1. Unfiltered MSK as well as unfiltered GMSK have constant RF envelopes and so the linear amplification requirement are reduced. Filtering required to limit spectral spreading result in amplitude variations of around 30%. However this is still very good so that one of the fundamental advantages of this modulation scheme is that nonlinear, power efficient amplification can be used. GMSK is essentially a digital implementation of FM with a binary change in the frequency of modulation. The switch from one modulation frequency to the other is timed to occur at zero phase. Put another way, the input bit stream is shaped to form half sinusiods for each bit of the input stream. The phase of the modulating signal is always continuous but at the zero crossings the half sinusoid continuous as a poitive or negative half sinusoid depending on the next bit in the input stream. The constellation diagram for GMSK is similar to that for OQPSK shown in Figure 11. The MSK and GMSK modulator is shown in Figure 12.

45

I
cos((t))
SERIAL BIT STREAM

i(t)

a(t)

WAVEFORM SHAPING

(t) s(t) VCO sin((t)) 90

q(t) Q

b(t)

Figure 12: MSK modulator.

4.3.7 Digital Modulation Summary


Digital transmission requires greater bandwidth than does analog modulation for transmission of the same amount of information. The appeal of digital modulation is directly related to the reduction of bit rate accomplished with modern speech coding algorithms. Acceptable speech is achieved with bit rates of 3.8 kbps and higher. (The measure of speech quality is purely subjective.) The speech coding algorithms achieve bit rate reduction by utilizing the substantial redundancy is speech. In a typical speech coding algorithm, units of 160 samples are characterized by just a few autocorrelation and related parameters. These parameters generally do not require many bits to describe so that a factor of 8 or even 16 reduction in the bit rate is achieved.

4.4 Homodyne Frequency Conversion


Homodyne mixing and detection is one of the earliest of wireless receiver technologies and is used in AM radio. Homodyne mixing can be used for detecting modulation formats other than AM. In homodyne mixing the carrier of a modulated signal is regenerated and synchronized in phase with the incoming carrier frequency. Mixing the carrier with the RF signal results in an IF signal centered around zero frequency. The only simply way to ensure that the pump is in phase is to transmit the carrier with the RF signal. AM transmission does just this but at the cost of transmitting large carrier power with the additional prospects of interference that go along with this. Homodyne mixing can be used with digitally modulated signals with I and Q. Signal spectra that result in homodyne mixing are shown in Figure 13. In Figure 13(a) the RF signals are shown on the right hand side and the baseband signals are shown on the left hand side. It is usual to show both positive and negative frequencies at the lower frequencies so that the conversion process is more easily illustrated. The characteristic of homodyne mixing is that the pump corresponds to the carrier and is in the middle of the desired RF channel. RF signal

46

components mix with the pump and it appears that the entire RF spectrum is down shifted around DC. Of course the actual baseband spectrum (the term for the lowest frequency signals) is only defined for positive frequencies so the negative frequency baseband signals and the positive frequency baseband signals add to yield the baseband spectrum shown in Figure 13. With other modulation schemes this loss of information is avoided using quadrature demodulation discussed later in the discussion of zero IF conversion. An amplitude modulated signal has identical modulation sidebands and so the collapsing of positive and negative frequencies at baseband is not a problem. Then a simple amplitude detection circuitry, such as a rectifier, is used and the rectified signal is passed directly to a speaker.
DOWN CONVERTED CHANNEL (a) PUMP

MAIN CHANNEL

DC FREQUENCY

FILTER (b)

DC

FREQUENCY

Figure 13: Frequency conversion using homodyne mixing: (a) The spectrum with a large local oscillator or pump; and (b) the baseband spectrum showing only positive frequencies.

4.5 Heterodyne Frequency Conversion


The main difference introduced by heterodyne mixing is that the pump and the main RF channel are not collocated as shown in Figure 14. In this figure the RF signals (shown as three discrete channels on the right hand side of the spectrum) mix with the pump signal to produce signals at a lower frequency. This lower frequency is usually not the final baseband frequency desired and so is called an intermediate frequency. The intermediate frequency of the main channel is at the difference frequency of the RF signal and the pump. The pump must be locally generated and so is called the local oscillator signal. There are several important refinements to this. The first of these is concerned with limiting the number of signals which can mix with the pump signal. This is done using an RF preselect filter as shown in Figure 15. This figure identifies a particularly troublesome set of frequencies called the image frequencies. In a regular RF spectrum with many channels the channel of concern is called the image channel. More about this latter. After the RF preselect filter the RF spectrum is as illustrated in Figure 15(b). The important characteristic is that the signals at frequencies below the pump have been suppressed. However 47

the image channel is still of concern as the image could be 100 dB higher than that of the desired channel. So even after substantial filtering it could still be significant. To see the difficulties introduced by the image channel consider the frequency conversion-to-baseband process described in Figure 16. The RF spectrum after RF preselect filtering is shown in Figure 16(a) and the baseband (or IF) spectrum is shown in Figure 16(b). Again positive and negative frequencies are used to better illustrate the down conversion process. Note the downconverted image and the main channel are equidistant from DC. So referring the signals to a positive frequency only spectrum, Figure 16(c), we see that the image channel interferes with the main channel. In the worst scenario the IF image could be larger than that of the desired channel. Fortunately there is a circuit fix which compensates for this.
DOWN CONVERTED CHANNEL PUMP FILTER

MAIN CHANNEL

DC FREQUENCY INTERMEDIATE FREQUENCY

Figure 14: Frequency conversion using heterodyne mixing.

IMAGE (a)

RF PRESELECT FILTER MAIN PUMP CHANNEL

DC FREQUENCY RF PRESELECT FILTER MAIN PUMP CHANNEL

IMAGE (b)

DC FREQUENCY

Figure 15: Frequency conversion using heterodyne mixing showing the use of an RF preselect filter to reduce the image signal.

48

IMAGE (a)

RF PRESELECT FILTER MAIN PUMP CHANNEL

DC FREQUENCY IF FILTER IMAGE (b)

DOWN-CONVERTED MAIN CHANNEL

IF

DC

+IF

FREQUENCY

MAIN CHANNEL WITH IMAGE (c)

DC

+IF

FREQUENCY

Figure 16: Frequency conversion using heterodyne mixing showing the effect of image distortion: (a) The RF spectrum following filtering using an RF preselect filter; (b) the baseband downconverted signal showing positive and negative frequencies; and (c) the single sided baseband spectrum following IF filtering showing the contamination of the final signal by the image signal. The block diagram of the circuit that corrects this image problem is shown in Figure 17 for both transmit and receive functions. There are several desirable attributes from this type of circuit. For now consider the receiver shown in Figure 17(b). Shown here is an antenna that takes in the RF signal and the RF preselect function is performed by a bandpass filter (BPF). This signal is initially amplified (usually) and then fed equally to two mixers. Each mixer has a pump signal which are 90o out of phase (that is, in quadrature) with each other. This is usually indicated by writing sin for one of the local oscillators and cos for the other. It is important that these two signals are very closely in quadrature and with identical amplitudes. The phase relationships at the outputs of the two mixers have very particular relationships with each other as illustrated in Figure 18.

49

(a)

(b)

Figure 17: Quadrature mixing: (a) transmit modulator; and (b) receive modulator. In Figure 18(a) the baseband spectrum at the I output (the top in-phase output) of the heterodyne receiver is shown. In Figure 18(b) the baseband spectrum at the Q output (the bottom and quadrature output) of the heterodyne receiver is shown. The second figure shows the negative frequency components inverted. This is a short hand way of saying that the negative (image) frequency components are 180o out of phase with the down converted image components of the in phase signal. However the down converted main channel and neighbors are in phase in both spectra. Thus the combination of the I and Q outputs suppresses the image signals from baseband. Figure 18(c) illustrates the positive spectrum following the summation of the I and Q channels at the output of the heterodyne receiver. The image rejection is key to all commonly used wireless communications systems today with the notable exception of paging systems which use a zero-IF scheme discussed latter on.

50

IF FILTER IMAGE (a)

DOWN-CONVERTED MAIN CHANNEL

IF

DC IF FILTER

+IF

FREQUENCY

IMAGE (b)

DOWN-CONVERTED MAIN CHANNEL

IF

DC

+IF

FREQUENCY

MAIN CHANNEL WITH IMAGE CANCELLATION (c)

DC

+IF

FREQUENCY

Figure 18: Frequency conversion using heterodyne mixing and quadrature mixing: (a) the baseband spectrum at the I output of the heterodyne receiver; (b) the baseband spectrum at the Q output of the heterodyne receiver; and (c) the positive spectrum following the summation of the I and Q channels at the output of the heterodyne receiver. Heterodyne systems implement signal processing such as filtering, modulation and demodulation, and image rejection at RF and IF frequencies using hard to integrate discrete components leading to expense and limitations on size reductions. Heterodyne architectures are regarded as approaching their limit in size, integration and fabrication cost. The primary issue in mixer design is limited image rejection resulting from gain and phase mismatches of the I (inphase) and Q (quadrature) paths.

4.6 Zero-IF Frequency Conversion


Zero-IF direct conversion receivers are very similar to quadrature homodyne receivers with a local oscillator signal placed at the center of the RF channel. The difference is that in homodyne receivers the phase of the carrier (i.e. is the phase of the local oscillator) is precisely known as the carrier is transmitted with the signal. In virtually all RF transmission schemes (above a few megahertz) the carrier is not transmitted. Thus in zero-IF direct conversion schemes the local

51

oscillator signal has inherent phase error with the original carrier. The important characteristic is that there is only one level of mixing. The conversion process is described in Figure 19.
DOWN CONVERTED CHANNEL (a) PUMP

MAIN CHANNEL

DC FREQUENCY FILTER (b) FILTER

I
DC FREQUENCY DC

Q
FREQUENCY

FILTER (c)

FILTER

I
DC FREQUENCY DC

Q
FREQUENCY

Figure 19: Spectra in the quadrature demodulation using homodyne or zero IF conversion. Thus the I and Q mixer outputs are necessary as the two sides of the RF spectrum contain different information and there would be irreversible corruption if a scheme was not available to extract the information in each of the sidebands. The main nonideality of this design is the DC offset in the down converted spectrum. DC offset results mostly from self-mixing of the local oscillator. The local oscillator mixes with itself to produce a DC signal. This DC offset can be much larger than the down converted signal itself and because of the nonlinearities of baseband amplification stages, either severely limits the dynamic range of the receiver or places limitations on the modulation format that can be used. One way of coping with the DC offset is to high pass filter the down converted signal but high pass filtering requires a large passive component (e.g. a series capacitor) at least to avoid dynamic range problems with active filters. High pass filtering the down converted signal necessarily throws away information in the signal spectrum and it is only satisfactory to do this if there is very little information around DC to begin with. Zero-IF direct conversion is most widely used in pagers with binary phase shift keying.

52

The primary effort in zero-IF converters is overcoming the DC offset problem and to a lesser extent coping with jitter of the local oscillator. The primary local oscillator noise of concern is close in phase noise which can be at appreciable levels 100 kHz from the carrier. This noise is commonly referred to as flicker noise and increases rapidly as the offset from the carrier reduces. This is of concern in all conversion processes. However one of the properties of heterodyne mixing is that the RF signal is considerably offset from the large phase noise region. Consequently the local oscillator phase noise at the frequency of the RF, has reduced impact on the resulting offset IF signal. For these reasons heterodyne mixing is sometimes used in paging receivers for higher performance. In summary, direct conversion receivers must deal with flicker noise and thermal noise whereas high-IF heterodyne systems must only be concerned with thermal noise. Proposed standards for increased bandwidth paging receivers are using a more spectrally efficient modulation scheme. The FLEX paging receiver has a proposed 6400 bps data rate and 4 level FSK modulation. As of 1998 it has not been possible to use direct conversion with this modulation format and heterodyne conversion is being employed. In cellular wireless the radio signals are spectrally efficient and the spectrum is fairly constant across the channel. So the near-DC distortion signals that result from direct conversion have appreciable information content and cannot be discarded so easily without significant distortion. Cellular wireless appears to compose random binary data and the spectrum of the data peaks at DC. It has been estimated that in the absence of noise and frequency offset that the baseband HPF, must introduce a distortion of less than 0.1% implying a corner frequency of 50 Hz in North American TDMA wireless systems. Much would be gained by developing a modulation format with little or no information content near DC but still with acceptable spectral efficiency The main problems of zero-IF conversion in cellular radio applications are 1. Spurious LO leakage. Retransmission of the LO is possible because the LO is tuned precisely to the RF signal frequency and reverse leakage through the RF path will radiate from the antenna. Spurious LO transmission is severely regulated. The limits on this inband LO radiation is between 50 and 80 dBm. The problem is reduced by using differential local oscillators and using multiple RF amplifier stages to increase the reverse isolation between the mixer and the antenna. 2. Interferer Leakage. A large RF interferer can leak through the RF amplifier and enter the mixer through both the local oscillator port as well as the RF port. The mixing of these components results in DC offset. 3. Distortion. Direct conversion receivers are more sensitive to undesired signals than are heterodyne receivers. The nonlinearities of the input mixers will rectify strong spurious RF signals to produce output components around DC. This is the result of second order nonlinearities and so this effect can be suppressed through the use of balanced RF circuits which will have only odd order nonlinearities. This reduces the baseband signals that can result from large RF interferers but it is still possible to produce baseband distortions if the RF interferer is large enough to produce third harmonics in the mixer. This effect of this distortion can be largely eliminated through the use of high pass filtering at the baseband but this is not acceptable for cellular radio signals. Heterodyne conversion is susceptible to

53

4.

5. 6.

7.

distortions resulting from odd-order nonlinearities but in zero-IF converters second order distortion is also a problem. LO self mixing. Mixing of the LO with itself will produce a DC signal in the mixer output. This DC level may be many orders of magnitude larger than the baseband signal itself so it can desensitize or saturate the baseband amplifier. DC offsets can also result from circuit mismatch problems. The DC offset that results primarily from LO self mixing is the most significant problem in the use of zero-IF architectures in cellular wireless. The DC offset can be reduced through the use of balanced designs but circuit mismatch errors still result in very large DC offsets. LO frequency error. A difference between the LO and the carrier will cause the RF signal to be asymmetrically converted around DC. Second order distortion. Because of second order distortion second harmonics of the signal can appear in the baseband. This is a problem if the RF signal is large to begin with. This problem can be circumvented by using designs that utilize differential signals. I/Q mismatches. Mismatches of the I and Q paths also results in DC offsets. These offsets however vary negligibly with time and analog or digital calibration techniques can be used to remove their effect.

The problem of DC offset is made worse as the DC level can vary with time as the amplitude of the interferer varies, or the LO that leaks from the antenna reflects off moving objects and is received as a time-varying interferer itself. A general problem with DC offset cancellation schemes is that interferers can be stored along with offsets.

4.7 Subsampling Analog-to-Digital Conversion


Subsampling analog-to-digital converters are also known as harmonic mixer direct conversion receivers. Subsampling receivers overcome the DC offset problem. The idea is to sample the modulated RF signal using an exact subharmonic of the carrier of the RF channel to be converted. The sampling rate must be at least twice the bandwidth of the RF signal and the track-mode bandwidth must be greater than the carrier frequency. Thus the sampling aperture is the critical parameter and must be several times smaller than the period of the carrier. Fortunately the aperture times of CMOS tracking circuits are adequate. It is critical that an RF preselect filter be used to eliminate unwanted interferes and noise outside the communication band. Aliasing of signals outside the Nyquist bandwidth onto the baseband signal is a consequence of subsampling. Adjacent channel signals are converted without aliasing but these will lie outside the bandwidth of the baseband signal. Flicker noise on the sampling clock is multiplied by the subsampling ratio and appears as additional noise in the baseband. Also broadband noise at the output of the RF tuned amplifier also is converted to baseband unless a second RF preselect filter is used after the amplifier.

54

4.8 Direct Conversion Band-Pass Sigma Delta Receivers


Direct conversion band-pass sigma delta receivers utilize a bandpass delta-sigma modulator to digitize a low IF signal resulting from typically only one stage of heterodyne conversion.. Band pass delta sigma modulators have been reported with bandwidths of 200 kHz achieving 8 bits of linearity for IF signals up to 50 MHz while being clocked at 500 MHz.

4.9 First IF to Baseband Conversion


In a superheterodyne conversion architecture there are two heterodyne stages with the IF of the first stage in the range of 20 to 150 MHz but with 90 MHz being the most common. (The assignment of frequencies is known as frequency planning and this is treated as proprietary by the major cellular phone vendors.) This IF is then converted to a much lower IF frequency typically around 455 kHz. This frequency is generally called baseband but strictly it is not as the signal is still offset in frequency from DC. Some direct conversion architectures leave the first heterodyne mixing stage in place and use direct conversion of the first IF to baseband (true baseband -- around DC). The zero IF and subsampling schemes outlined above have been tried. But now it is possible to use additional schemes which use analog ASIC technology including analog sampled architectures such as data switched capacitor circuits. One of the attractions for doing this as the normal second heterodyne stage utilizes Gilbert mixers which are not suitable for low voltage operation. Since in this architecture the first LO can be adjusted to as to a fixed offset of the carrier of the desired RF channel, a selection filter can be used to control the spectral content of the IF to just the desired channel and a few channels on either side. With a better defined signal to the second conversion stage, direct conversion can be to produce a slightly offset baseband signal and so get around the DC offset problems.

4.10 Direct Conversion Modulation


Direct conversion modulation is also an important aspect in the digitization of wireless communications but it is seen as less urgent than the digitization and re-architecture of the receiver system. Also the transmitted signal has a much simpler and controlled spectral content being just a single channel and much simplification of the RF transmit path has already been achieved. Issues such as image rejection and interferers are not problems. The overwhelming majority of wireless transmitters are based on the heterodyne principle where baseband I and Q signals are used in a vector modulator to produce a modulated IF which is upconverted in a mixer to an RF signal. Generally two upconversion stages are used with the IF at a relatively low frequency where a filter can easily remove the out of channel leakage. The final RF signal is then amplified by a power amplifier. A direct conversion transmitter generates the IF signal directly without the IF stage or with only a single traditional mixer-based upconversion stage. The architecture of a direct conversion transmitter is shown in Figure 20.

55

I
DATA TRANSMIT MODEM PREDISTORTER

QUADRATURE MODULATION CORRECTOR

QUADRATURE MODULATOR

PA

Figure 20: Architecture of a direct conversion transmitter. Here the transmit modem produces I and Q baseband signals from the data. This is then translated directly to RF via a quadrature modulator and then amplified by a power amplifier (PA). Practically the nonlinearities of the PA must be linearized using a predistorter and quadrature modulation errors must be accounted for in a quadrature modulation corrector. The transmit modem, the predistorter and the quadrature modulation corrector can be combined in a digital signal processor. Then following the quadrature modulation corrector A/D converters are used to produce a baseband signal or perhaps convert the signal to a low intermediate frequency. So considerable complexity is shifted to the DSP chip. One of the major problems in this architecture is the noise introduced by errors in step-size mismatch. Noise-shaping techniques implemented in a DSP have been developed to shift this noise outside the bandwidth of the generated signal. Similar errors are associated with mismatches of the A/D converters and of the analog circuit paths of the separate I and Q paths. This distortion is also pushed out-of-band by the noise-shaping algorithm.

56